EarlyCDT Lung blood test for risk classification of solid pulmonary nodules: systematic review and economic evaluation

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number NIHR133496. The contractual start date was in April 2021. The draft report began editorial review in September 2021 and was accepted for publication in June 2022. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2022 Duarte et al. This work was produced by Duarte et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2022 Duarte et al.

Lung cancer

Lung cancer is one of the most common types of cancer, with around 47,000 diagnoses per year in the UK. 1 Lung cancer is commonly associated with smoking, being responsible for > 70% of cases. Other causes of cancer include passive smoking and exposure to asbestos or other carcinogenic chemicals.

Population and relevant subgroups

The population of interest was all persons with solid non-calcified pulmonary nodules identified by CT, whether undertaken for conditions unrelated to lung cancer, as part of a cancer diagnosis procedure for people with possible lung cancer symptoms or as part of a lung cancer screening programme. Specifically, the assessment examined the following:

• people with a nodule of 5–8 mm in diameter or 80–300 mm³ in volume

• people with a < 10% risk of malignancy using the Brock model after initial CT or using the Herder model after PET-CT

• people with a 10–70% risk of malignancy using the Brock model, or the Herder model (after PET-CT).

People with other cancers, or who had had a cancer diagnosis in the previous 5 years, were excluded from consideration: EarlyCDT Lung (Oncimmune Holdings plc, Nottingham, UK) is not recommended for such persons.

Table 1 shows the possible distribution of patients at different parts of the BTS pathway, taken from a study by Al-Ameri et al. 6 of 186 individuals who match the population of interest for this assessment. This study suggested that the majority of patients presenting with incidentally detected nodules are classed as small (nodules between 5–8 mm diameter) or low risk. These are assigned to CT surveillance, which suggests a large burden on the health system from the multiple follow-up CT scans. This evidence also suggests that there is a meaningful proportion of cancers detected at the metastatic disease stage across all risk groups.

In all populations, patients would receive an EarlyCDT Lung test and proceed to excision or surgery if deemed to be at high risk of malignancy (> 70%). At lower risk of malignancy (< 70%), patients would go on to CT surveillance, or possibly biopsy or excision for patients at intermediate risk (10–70%).

The protocol specified that the key subgroups of interest were the different reasons for receiving an initial CT scan (patients with symptoms, incidental finding when scanning for other conditions, or as part of a cancer screening programme). However, no data on these subgroups were identified, so they could not be investigated.

Diagnostic technologies under assessment

EarlyCDT Lung

EarlyCDT Lung is a blood test that can be used to assess the malignancy risk of people at risk of lung cancer. The test can, in principle, be used on any at-risk person; this assessment will consider its use among persons with solid pulmonary nodules found by chest CT or radiography. 7–9 Incidental finding of pulmonary nodules in asymptomatic individuals, when performing CT for other medical purposes, or during lung cancer screening, is an increasingly common clinical dilemma encountered by lung cancer clinicians. EarlyCDT Lung could be used as part of the standard diagnostic pathway for early detection of lung cancer, where it might result in treatment being offered earlier, giving improved patient outcomes.

EarlyCDT Lung uses a standard enzyme-linked immunosorbent assay (ELISA) method. It is manufactured by Oncimmune and is available as a Conformité Européenne (CE)-marked in vitro diagnostic kit. It was launched commercially in November 2010, with physicians in routine practice across the USA ordering the test on behalf of their patients. 10 The test measures the presence of autoantibodies to a panel of seven lung cancer-associated antigens (p53, NY-ESO-1, CAGE, GBU4-5, HuD, MAGE A4 and SOX2). 7 A blood sample is considered positive when at least one of the seven autoantibodies is elevated above a predetermined cut-off point (Table 2). 11 Elevated levels of these autoantibodies may indicate current (or past) malignant disease. The thresholds were set to give a high test specificity with the aim of reducing false-positive results that would lead to unnecessary, and potentially invasive, diagnostic procedures. The EarlyCDT Lung test results are interpreted by skilled medical professionals in combination with other clinical information. In particular, it is suggested that its results be used to modify the risk of malignancy estimated by existing nodule risk calculators, including the Brock model and the Herder model. 12,13

TABLE 2 - Recommended cut-off points for autoantibodies measured using EarlyCDT Lung

Taken from NICE Final Scope for this assessment. 11 © NICE 2021 EarlyCDT Lung for cancer risk classification of indeterminate pulmonary nodules: Final Scope. Available from www.nice.org.uk/guidance/dg46/history. All rights reserved. Subject to Notice of rights. NICE guidance is prepared for the National Health Service in England. All NICE guidance is subject to regular review and may be updated or withdrawn. NICE accepts no responsibility for the use of its content in this publication.

Autoantibody		Low cut-off value		High cut-off value
CAGE	No significant level of autoantibodies detected ↓	4.25	Moderate-level result ↓	5.27	High-level result ↓
GBU4-5		4.36		5.92
NY-ESO-1		3.02		4.27
p53		5.79		6.47
SOX2		5.48		5.58
MAGE-A4		6.19		7.94
HuD		7.31		8.15

Oncimmune has described EarlyCDT Lung as a ‘rule-in’ test to help identify pulmonary nodules that may benefit from earlier diagnosis and treatment. Results of EarlyCDT Lung tests are reported as one of three options:

1. no significant levels of autoantibodies detected (if no autoantibody is above the low cut-off level)

2. positive moderate (if at least one autoantibody is above the low cut-off level, but below the high-cut-off level)

3. positive high (if at least one autoantibody is above the high cut-off level).

A patient will have a pre-test risk of lung cancer predicted by their sex, age, smoking history and other risk factors alone, calculated by the Brock (or Swensen/Mayo) nodule malignancy risk calculator. If a person is being assessed after PET-CT, their risk may be assessed using the Herder malignancy risk tool.

Oncimmune proposes that the EarlyCDT Lung test result is used to update these estimated risks of malignancy. For people who test negative with EarlyCDT Lung, Oncimmune recommends that the estimated risk is left unchanged from the pre-test risk, in this way defining this test as a ‘rule-in’ test. Statistically, a patient with a negative test result should see their risk scores downgraded, but this is not proposed for this assessment. Clinical management for these individuals would then proceed in line with the pre-test risk.

A positive-moderate result would lead to a moderate increase in the chance of malignancy from the pre-test risk. If the increase in risk is large enough, it might suggest that further diagnostic testing is needed, such as image-guided biopsy. A positive-high result would lead to a larger increase in the chance of malignancy from the pre-test risk. This might suggest that further diagnostic testing is needed, or, if the new risk estimate is sufficiently high, that the person should proceed directly to surgical resection of the nodules.

Oncimmune has produced a graph detailing how the pre-test risk could be modified given a positive-moderate or positive-high EarlyCDT Lung test result (Figure 1). The calculation of post-test mortality risk from the baseline risk obtained from the Swensen/Mayo calculator and the EarlyCDT Lung test result is described in Healey et al. 10 Oncimmune proposes applying this calculation to pre-test risks derived with both Brock and Herder models.

FIGURE 1.

Impact of EarlyCDT Lung test result on lung cancer risk assessment.

The EarlyCDT Lung test should not be used among people with a previous history of cancer of any type, except for basal cell carcinoma, as other cancers may lead to elevated levels of autoantibodies, and hence to false-positive results. It should also not be used among people known to have diseases that result in an elevated level of serum total protein, for example myeloma, amyloidosis and monoclonal gammopathy of undetermined significance.

As far as the Evidence Assessment Group (EAG) can determine, the EarlyCDT Lung test is not currently in regular use in the UK for the assessment of pulmonary nodules. It has been used in a large-scale trial in Scotland to investigate whether or not it can be used as part of a lung cancer screening programme. 14

Cost of EarlyCDT Lung testing

The cost of EarlyCDT Lung testing to the NHS includes the costs of (1) the test, (2) consumables required to process the test, (3) test administration, (4) training needed to process/administer the test and (5) costs of delivering test results to individuals.

In response to a request for information by NICE, Oncimmune provided an estimate of the cost per EarlyCDT Lung kit of £600 (excluding value-added tax). According to the company, each kit can run up to 10 patient samples. Therefore, assuming that patient throughput is sufficient to always ensure simultaneous processing of 10 patient samples, and that there are no test failures, the cost of the test per patient will be £60. When determining this element of cost in future assessments, patient throughput will have to be considered, as throughput may vary across NHS trusts and will be a determinant of cost. Furthermore, the need to ‘batch’ tests for processing may lead to delays to diagnosis in trusts with low throughput, which may have to wait for a sufficient number of samples before processing a full batch. The extent of this delay is unknown, but it may reduce some of the benefit from early diagnosis with EarlyCDT Lung if too prolonged. The company also provided an estimate of the costs of consumables required to process each EarlyCDT Lung kit in their laboratories, but notes that costs may differ in an NHS setting (the company suggests that these costs may be lower to the NHS without supporting evidence). The costs of consumables to process one EarlyCDT Lung as provided by the company are shown in Table 3; it suggests a cost per kit of £7.13 and a cost per test of £0.71.

TABLE 3 - Cost of consumables for processing one EarlyCDT Lung kit (up to 10 patient samples)

Note

Table adapted from company’s response to a request for information in 2021 (all company-supplied documentation and correspondence has been deleted as required by NICE).

Item	Quantity used	Cost per case (£)	Pack size (n)	Unit cost (£)	Cost per kit (£)	Cost per test (£)
Two-plate format for running 10 tests
5-ml pipette tips	2	63.21	1000	0.06	0.13	0.01
Serum dilution tubes	10	99.76	700	0.14	1.43	0.14
20-µl pipette tips	10	77	960	0.08	0.80	0.08
1000-µl pipette tips	13	37.78	768	0.05	0.64	0.06
1200-µl pipette tips	24	84	960	0.09	2.10	0.21
dH20 (l)	1	7.32	5	1.46	1.46	0.15
Falcon tube (50 ml)	1	56.04	500	0.11	0.11	0.01
Reagent troughs	3	15.26	100	0.15	0.46	0.05
Total cost					7.13	0.71

The EAG notes that these costs are likely to vary across NHS trusts, as local procurement arrangements may result in variation of the cost per item for each consumable. The patient throughput is less likely to affect the calculations of cost per patient, assuming that most of the listed consumables are routinely used materials in NHS laboratories.

The company did not report any estimates for the remaining elements of costs (i.e. the costs of test administration, of training needed to process/administer the test and of delivering test results to individuals).

According to EarlyCDT Lung’s instructions for use (as submitted by Oncimmune), the test requires a blood sample (serum or plasma) to be collected. Therefore, the test administration cost should reflect the NHS staff time required to collect the blood sample. This cost may vary depending on local protocols, and whether or not these impose additional contacts with health-care professionals. For example, if the test is to be administered only to patients at a risk level of < 70% on the Herder score, the patient may require one additional contact with the health service after PET-CT (see Place of the intervention in the care pathway) to have their blood sample collected, and this cost will be incurred only by patients with a risk score of < 70%. Alternatively, local protocols might require a sample to be collected for all patients when they undergo PET-CT; this may allow for some efficiency gains if blood collection can be fitted within the workup. However, it means that the cost of collecting the blood is incurred for all patients, regardless of their Herder score, and that the blood sample needs to be stored until the Herder score is available for all patients. Another option, to avoid storing unnecessary blood samples, is to test all samples with EarlyCDT Lung, regardless of a patient’s risk score; this would imply that all patients incur the cost of the test, as well as the cost of collecting a blood sample.

The company did not provide information on the training requirements needed to process and interpret the test, but these also need to be considered. These costs would need to reflect the cost of laboratory staff time to learn how to run the test and use the associated software to obtain a result. Similarly, the cost of clinician time to learn how to interpret the results should also be included.

Finally, a cost may have to be included to reflect additional time for the clinician to interpret the test results (which may be negligible in the context of the diagnostic workup) and, more importantly, any additional contacts between the patient and the health-care system to deliver the results of the test. The BTS guidelines recommend offering ‘patients the choice of seeing a lung cancer nurse specialist where the probability of malignancy is high or when patients are anxious about the possibility of having lung cancer’. 3 So it may be appropriate to include the cost of an appointment with a specialist nurse for some of the patients. Although this cost may also be incurred in strategies without EarlyCDT Lung, the number of patients incurring the cost will vary across strategies and it should, therefore, be considered.

Comparators

The overall comparator was the current BTS-recommended diagnostic pathway for pulmonary nodules without EarlyCDT Lung. Specifically, this included diagnosis and management of nodules using:

• the Brock model

• the Herder model (after PET-CT)

• no risk assessment (for nodules between 5–8 mm in diameter or 80–300 mm³ in volume).

To fully interpret the clinical and economic impacts of using EarlyCDT Lung, the diagnostic accuracy and clinical effectiveness of the following specific parts of the diagnostic pathway were also investigated:

• CT surveillance (for small or low-risk nodules)

• PET-CT (for intermediate-risk nodules)

• biopsy of suspicious nodules (for high-risk nodules).

Place of the intervention in the care pathway

Lung cancer is often diagnosed at a more advanced stage than other common cancers. National Cancer Registration and Analysis Service data show that almost half of all lung cancers are diagnosed at stage 4. Late diagnosis, when curative treatment is not possible, is a contributing factor to poor survival rates for people with lung cancer. Early detection is key to improving outcomes.

Oncimmune has produced a flow chart showing the proposed position of the EarlyCDT Lung test within the current BTS pathway for solid pulmonary nodules (BTS guidelines3). This is shown in Figure 2. This pathway includes an option in which PET-CT is not available. Clinical opinions received at scoping, and during the project, suggested that lack of access to PET-CT is not of concern for the NHS. This assessment will therefore consider only the part of the pathway where PET-CT is available.

FIGURE 2.

Proposed position of EarlyCDT Lung within the current BTS pathway for lung cancer. a, For example hamartoma, typical perifissural nodule; b, consider PET-CT for larger nodules in young patients with low risk by Brock score, as this score was developed in a screening cohort aged 50–75 years, so performance among younger patients is unproven.

The position of EarlyCDT Lung has been stated to be after the first CT scan, or post PET-CT when the result suggests intermediate risk. EarlyCDT Lung could be used to assess people with nodules of < 8 mm in diameter or 300 mm³ volume and those with < 10% risk of malignancy after using the Brock model. The test could also be used for people with 10–70% risk of malignancy after using either the Brock or the Herder model. If the EarlyCDT Lung test is positive, the malignancy risk is increased and people with a post-test risk of > 70% could then be moved into the intervention pathway immediately, without the delay caused by CT surveillance, or further diagnostic testing.

This assessment will consider the following specific locations in the diagnostic pathway where EarlyCDT Lung could be used; the feasibility and relevance of the proposed placements will be established based on clinical advice:

• for people with nodules of 5–8 mm in diameter or 80–300 mm³ in volume

• in combination with CT and the Brock model, for people with nodules of > 8 mm in diameter who have < 10% risk of malignancy using the Brock model after initial CT

• in combination with PET-CT and the Herder model, for people with nodules of > 8 mm in diameter who have < 10% risk of malignancy using the Herder model after PET-CT

• in combination with CT and the Brock model, for people with nodules of > 8 mm in diameter who have 10–70% risk of malignancy using the Brock model (with EarlyCDT Lung preceding PET-CT)

• in combination with PET-CT and the Herder model, for people with nodules of > 8 mm in diameter who have 10–70% risk of malignancy using the Herder model, after PET-CT.

Outcomes

Below is a list of all key outcomes judged to be relevant to the assessment of the clinical effectiveness and cost-effectiveness of EarlyCDT Lung, and the general diagnostic pathway for pulmonary nodules. This represents a comprehensive list of outcomes listed in the protocol. Owing to the limited nature of the published literature, particularly for EarlyCDT Lung, many of these outcomes could be evaluated using only indirect evidence (such as data from lung cancer screening studies), or could not be formally assessed. The key outcomes are as follows:

• diagnostic accuracy

  • sensitivity, specificity, positive and negative predictive values, diagnostic likelihood ratios, area under the receiver operating characteristic (AUROC) curves

  • for EarlyCDT Lung in isolation and in combination with the Brock and Herder models

• short-term clinical outcomes

  • impact of test on risk classification

  • impact on clinical decisions relating to diagnostic or treatment pathway

  • further tests used

    • including PET-CT and image-guided or excision biopsy

  • adverse events during or after testing

• longer-term clinical outcomes

  • lung cancer mortality

  • lung cancer-related morbidity

  • morbidity associated with other diagnostic tests or procedures

  • overall and disease-free survival

• patient-focused outcomes

  • health-related quality of life (HRQoL)

    • Short Form questionnaire-36 items, EuroQol-5 Dimensions (EQ-5D)

  • impact on anxiety and cancer concern

    • false-positive tests

    • unnecessary biopsies or other procedures

    • overdiagnosis of tumours not requiring immediate treatment

    • delay in diagnosing treatable cancers

    • understanding and communication of test results

• implementation of test

  • time to obtain results

  • laboratory capacity

  • training requirements

  • clinical variation in interpreting and using results.

Objectives

The aim of the project was to appraise existing evidence on the potential clinical effectiveness and cost-effectiveness of the EarlyCDT Lung test for lung cancer risk classification of solid pulmonary nodules, and to develop a conceptual economic model to provide a common understanding of the evidence requirements and evidence linkages required to undertake a robust cost-effectiveness analysis. To achieve this, the following objectives were set.

Clinical effectiveness

• To perform a systematic review and, if feasible, a meta-analysis of the diagnostic accuracy of EarlyCDT Lung for lung cancer risk classification of solid pulmonary nodules.

• To perform a narrative systematic review of the clinical impact and practical implementation of using the EarlyCDT Lung test.

• To perform a scoping review of the evidence on EarlyCDT Lung for uses outside the specified diagnostic pathway (e.g. as a lung cancer screening tool), where this will inform the overall review.

Cost-effectiveness

• To perform a systematic review of published cost-effectiveness studies of EarlyCDT Lung for lung cancer risk classification of solid pulmonary nodules.

• To review cost-effectiveness models for other surveillance and diagnostic strategies for the identification of malignancy in solid pulmonary nodules, and cost-effectiveness models of screening strategies for lung cancer.

• To conceptualise a decision model structure to provide a common understanding of how the cost-effectiveness of EarlyCDT Lung for lung cancer risk classification of solid pulmonary nodules in the different positions of the diagnostic pathway proposed for the technology can be quantified.

• To scope existing evidence that could support the implementation of the conceptualised decision model, highlighting key evidential and structural uncertainties.

The objectives for this assessment were set out in the development of the protocol, which acknowledged that the existing published evidence base on EarlyCDT Lung was too small to allow a full assessment of the clinical and economic value of the test. This assessment was therefore restricted to review the extent of the existing evidence and provide a common understanding of the evidence requirements and evidence linkages required for a full assessment of the value of EarlyCDT Lung to the NHS. The EAG was therefore not requested to develop and implement a de novo decision analytic model.

Systematic review of diagnostic accuracy and clinical effectiveness

The systematic review was conducted following the general principles recommended in the guidance from the Centre for Reviews and Dissemination (CRD) and is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. 15,16

Cost-effectiveness reviews

Conceptualisation of the decision model and identification of evidence requirements for future assessments

This component of work will focus on the conceptualisation of a decision model, structured according to good-practice recommendations,23,24 to quantify the broader consequences to health and overall NHS and Personal Social Services costs associated with the use of EarlyCDT Lung (i.e. its cost-effectiveness). The recommendations and model conceptualisation will comply with the NICE reference case. 25 The key outputs of this element of work will be as follows:

• an outline of key considerations for the development of an appropriate model structure, considering key structural assumptions and identifying the nature of the evidence linkages required

• an outline of key parameter inputs required, including an assessment of the possible data gaps that would need to be addressed in future research.

The conceptualisation process combined problem-oriented and design-oriented elements identified in Kaltehthaler et al. 24 The problem-oriented element of the conceptual modelling will describe (1) current clinical understanding of the clinical condition and important events and (2) clinical pathways through which patients are detected, diagnosed, treated and followed up. The design-led element of conceptual modelling will identify potentially feasible and credible model choices to represent the events and pathways deemed relevant in the problem-oriented element, considering the availability of existing evidence.

Explicit processes were used for the conceptualisation process, including interviews with a clinical expert, and supported by the learnings from the set of reviews conducted within this project. The results of the conceptualisation were recorded using influence diagrams,26,27 which are reported in Chapter 6. Influence diagrams are compact representations of decision problems focusing on illustrating relationships between parameters in a model. These can be parameterised and implemented as decision-analytic models (because of the probability-based representation of influence diagrams, these are typically translated into decision trees). However, we here use the influence diagrams to, more generally, reflect on relationships that need to be considered in a future assessment of EarlyCDT Lung. These diagrams, therefore, are not to be used to convert the problem conceptualisation into an appropriate model structure, but to support further attempts to do so as further evidence emerges updating knowledge of the disease, the technology and the process to be modelled.

The technology of interest is diagnostic, presenting a value proposition that is complex, including indirect effects from changes in management decisions. The conceptualisation process was therefore structured to first identify value drivers and value components that could be of relevance for establishing the cost-effectiveness of EarlyCDT Lung in the diagnostic pathway for solid pulmonary nodules. Value drivers are here defined as factors, such as disease prevalence, that are expected to have a significant impact on cost-effectiveness. Value components are here defined as different mechanisms for clinical and economic impacts of this technology (including any potential consequences of suboptimal treatment decisions among those misclassified) in this decision problem. These impacts may include,28 for example, direct effects of the technology, effects derived indirectly by altering clinical decision on further tests or treatments, effects on the timing of decisions and actions, or influence on patient and clinician perspectives. These will include implications for resource use and for processes of health-care service provision of the use of the test in relation to its alternative(s).

The conceptualisation then focused on identifying possible mechanisms for evidence linkage for each of the components of value identified, for example reflecting the consequences of diagnostic test accuracy as final cost and health outcomes.

EarlyCDT Lung studies

Quantity of research available

Figure 3 presents a summary of the EarlyCDT Lung study identification and selection process. The searches identified a total of 3233 unique records. After title and abstract screening, 115 references were retrieved, and 47 references were included in the review. Over half the included references were reported as conference abstracts.

FIGURE 3.

The PRISMA flow diagram for the systematic review of EarlyCDT Lung studies.

Many references were excluded because the study populations consisted of patients with already diagnosed lung cancer (i.e. they studied validation cohorts of patients who would not receive the EarlyCDT Lung test in practice); seven published papers formed part of this group of references. 29–35 Although 47 references were identified as being eligible for inclusion in the review, they covered only six distinct patient cohorts, with some references reporting on subgroups within a cohort. See Appendix 4, Tables 32 and 33, for full details.

Synthesis of diagnostic accuracy

For four of the five identified cohorts of patients with pulmonary nodules, diagnostic accuracy data were reported in one paper or abstract for each cohort. For the HIPAA cohort, diagnostic accuracy data on patients with pulmonary nodules were reported in three papers. In this analysis, we used data reported in Massion et al. ,47 as that was the most recently published and most comprehensive paper for that cohort. Full diagnostic accuracy data from all papers are presented in Appendix 3, Tables 30 and 31.

The summary sensitivity and specificity data for the five cohorts are presented in Figure 4. The results of most cohorts are broadly consistent, with high specificity of > 90%, but low sensitivity of < 30%. The HIPAA cohort showed higher sensitivity for lower specificity, but this may be because different test thresholds were used. The HIPAA cohort used the cut-off values for the commercial form of EarlyCDT Lung at that date. The González Maldonado et al. 54 cohort used a ‘high-specificity’ cut-off value reported in another HIPAA paper (Healey et al. 10); this threshold is presented in Figure 4 and in the meta-analyses. Diagnostic accuracy data of patients with nodules were not reported for this ‘high-specificity’ cut-off value in any HIPAA paper.

FIGURE 4.

Diagnostic accuracy of EarlyCDT Lung from the five included cohorts. LCS, Lung Cancer Screening.

The González Maldonado et al. 54 paper also reported the diagnostic accuracy of using the combination of the ‘high-specificity’ and ‘moderate-specificity’ thresholds from Healey et al. ,10 which is the approach suggested by Oncimmune (see Figure 1). This found no change in sensitivity from using only the high-specificity threshold [13%, 95% confidence interval (CI) 4.9% to 26.3%], but a reduced specificity of 91.1% (95% CI 83.2% to 96.1%), compared with 95.6% (95% CI 89.0% to 98.8%).

The summary sensitivity from univariate meta-analysis (Figure 5) was 22% (95% CI 11% to 37%). The summary specificity (Figure 6) was 92% (95% CI 86% to 95%). It should be noted that these estimates are based on different EarlyCDT Lung test cut-off values, and so may not represent test accuracy at any specified cut-off value. As the test cut-off value used was unclear for the three cohorts reported only as conference abstracts, a meta-analysis at specific test cut-off values was not possible.

FIGURE 5.

EarlyCDT Lung: meta-analysis of sensitivity. FN, false negative; LCS, Lung Cancer Screening; TP, true positive.

FIGURE 6.

EarlyCDT Lung: meta-analysis of specificity. FP, false positive; LCS, Lung Cancer Screening; TN, true negative.

The summary positive predictive value (PPV) was 32% (95% CI 11% to 64%). The summary negative predictive value was 85% (95% CI 63% to 95%). It should be noted that these summary results do not adjust for possible variation in prevalence across studies. The summary diagnostic odds ratio was 3.32 (95% CI 1.75 to 6.31). No study reported data on the AUROC curve.

The results of a full bivariate meta-analysis of the EarlyCDT Lung cohorts, including a summary HSROC curve, are shown in Figure 7. The summary sensitivity in the bivariate model was 20.2% (95% CI 10.5% to 35.5%) and the specificity was 92.2% (95% CI 86.2% to 95.8%). However, as this includes cohorts using different EarlyCDT Lung cut-off values, this may not be a reliable summary. Instead, from the HSROC curve, we predict that EarlyCDT Lung has around 26% sensitivity at 90% specificity, or 12% sensitivity at 95% specificity. The area under the HSROC curve was 0.694, suggesting poor to moderate overall diagnostic accuracy.

FIGURE 7.

EarlyCDT Lung: bivariate meta-analysis and HSROC curve. LCS, Lung Cancer Screening.

An analysis restricted to the three cohorts with > 100 patients gave almost identical results: sensitivity 23.3% (95% CI 11.9% to 40.5%), specificity 91.2% (95% CI 84.3% to 92.5%) and area under the curve (AUC) 0.694].

The EAG notes that the diagnostic accuracy of EarlyCDT Lung among people with pulmonary nodules therefore appears to be poor and is lower than that predicted by Oncimmune (see Case–control studies of EarlyCDT Lung among patients without confirmed pulmonary nodules). This may be because the risk models developed by Oncimmune were based on case–control studies of patients without pulmonary nodules.

Diagnostic accuracy by nodule size

Diagnostic accuracy of EarlyCDT Lung by nodule size was reported for the HIPAA cohort only; here we consider results from the Massion et al. 47 paper, presented in Table 7. Results were also presented in Healey et al. 10 (see Table 9).

TABLE 7 - Diagnostic accuracy by nodule size in Massion et al.47

Nodule diameter (mm)	Sensitivity (95% CI) (%)	Specificity (95% CI) (%)
< 4	No malignant nodules	72.2 (46.5 to 90.3)
4–20	40 (16.3 to 67.7)	83.9 (74.4 to 90.9)
> 20	36.4 (17.2 to 59.2)	91.7 (73.0 to 98.8)

These results show no clear evidence of variation in diagnostic accuracy by nodule size, although it is possible that sensitivity declines, but specificity increases, with increasing nodule size. There were no malignant nodules with diameters of < 4 mm, so sensitivity could not be estimated for these smallest nodules. It does suggest that nodules are rare in this group, so most positive EarlyCDT Lung results will be false positives (27.8% false-positive rate).

Combining EarlyCDT Lung with other risk scores

No studies reported any diagnostic accuracy data for the combination of EarlyCDT Lung with either the Brock or Herder risk assessment tool. Massion et al. 47 reported data from the HIPAA cohort when combining EarlyCDT Lung with the Mayo risk tool. This compared the Mayo risk alone with both Mayo and EarlyCDT Lung being positive, at both 30% Mayo risk and an overall 97% specificity. The results are presented in Figure 8.

FIGURE 8.

Diagnostic accuracy of combining EarlyCDT Lung with Mayo risk.

At 30% risk, adding EarlyCDT Lung to Mayo substantially increased the specificity, but also decreased the sensitivity. At 97% specificity, there is evidence that adding EarlyCDT Lung to Mayo risk can increase sensitivity. The paper47 does not state what risk level a specificity of 97% will equate to. Given that the specificity is much higher than at 30% risk, it is likely to correspond to a high risk of malignancy.

It is not clear whether or not these results from using Mayo risk would be similar if Brock or Herder risk were used. Furthermore, the ‘both-positive’ approach analysed here is not what is currently proposed for EarlyCDT Lung: risk will be recalculated if EarlyCDT Lung is positive (see Figure 1).

Evidence on EarlyCDT Lung outside the diagnostic pathway

The Early Detection of Cancer of the Lung Scotland trial

The ECLS study was a randomised trial that addressed the question, ‘Does using the EarlyCDT Lung test to identify those at high risk of lung cancer and any subsequent CT scanning reduce the incidence of patients with late-stage lung cancer (III and IV) or unclassified presentation at diagnosis, compared with standard clinical practice?’. 14 This trial was undertaken in 12,208 individuals at increased risk of lung cancer (based on smoking history), with the intervention arm receiving the EarlyCDT Lung test and, if test-positive, low-dose 6-monthly CT scans for up to 2 years. EarlyCDT Lung test-negative and control arm participants received standard clinical care (symptomatic presentation). The trial was therefore not designed to assess the incremental contribution of the EarlyCDT Lung test and, although it did report sensitivity and specificity as outcomes, the lack of a focus on a population, or subgroup, with pulmonary nodules means that the diagnostic accuracy results are of very limited applicability to this assessment.

The ECLS study results have been reported across many conference abstracts and published papers (summarised in Table 8). The main trial paper by Sullivan et al. 14 reported that 127 lung cancers were detected in the study population (1.0%) at 2 years. For the trial’s primary outcome, in the EarlyCDT Lung test arm, 33 out of 56 (58.9%) lung cancers were diagnosed at stage III/IV, compared with 52 out of 71 (73.2%) in the control arm (hazard ratio for stage III/IV presentation: 0.64, 95% CI 0.41 to 0.99). There were no statistically significant differences between groups in lung cancer mortality and all-cause mortality after 2 years. Five intervention-related adverse events (related to blood sample collection) were reported; all were considered to be minor.

TABLE 8 - Summary of references reporting on the ECLS trial

IES, Impact of Event Scale; LCWS, Lung Cancer Worry Scale; N/A, not applicable; NR, not reported; PANAS, Positive and Negative Affect Schedule.

Reference and sample size	Reference type	Population or subgroup reported	Period of recruitment	Outcomes reported
Sullivan 2021;14 n = 12,208	Paper	Screening: adults aged 50–75 years at increased risk of developing lung cancer	April 2013 to July 2016	Sensitivity, specificity, mortality, adverse events, anxiety, depression, worry outcomes. Uptake of subsequent investigations such as CT or bronchoscopy
Bedford 2017;60 n = 1096	PhD thesis	All EarlyCDT positives and random selection of negatives and control participants	January 2014 to May 2016	PANAS, LCWS, IES
Clark 2017;61 n = 1032	Conference abstract	Sample of EarlyCDT positives, EarlyCDT negatives and control participants	NR	PANAS, LCWS, IES
Dorward 2016;62 n = 12,018	Conference abstract	Adults aged 50–75 years at high risk of developing lung cancer	Completed in June 2016	EarlyCDT Lung test results and number of cancers
Sullivan 2015;64 n = 12,000	Conference abstract	Adults aged 50–75 years at high risk of developing lung cancer	NR	EarlyCDT Lung test results and number of cancers. Current PPV of test
Sullivan 2014;63 n = 10,000	Conference abstract	Adults aged 50–75 years at high risk of developing lung cancer	NR	EarlyCDT Lung test results and number of cancers
Sullivan 2017;65 n = 12,208	Conference abstract	Adults aged 50–75 years at high risk of developing lung cancer	Completed by June 2016	EarlyCDT Lung test results and number of cancers
Sullivan 2017;66 n = 12,210	Conference abstract	Adults aged 50–75 years at high risk of developing lung cancer	NR	EarlyCDT Lung test results and number of cancers
Sullivan 2017;58 n = N/A	Paper (protocol)	Adults aged 50–75 years at high risk of developing lung cancer	N/A	Protocol (list of outcomes)
Young 2017;68 n = 1032	Conference abstract	Subsamples of EarlyCDT positives, EarlyCDT negatives and control participants	NR	Smoking point prevalence, attempts to quit, number of cigarettes smoked per day and the Heaviness of Smoking Index
Young 2017;69 31 interviews	Conference abstract	Sample of people with positive and people with negative screening test results, either successfully or unsuccessfully attempted to stop smoking or no attempt since screening	NR	Qualitative interviews on facilitators of smoking cessation/cessation support; facilitators included emotional responses to test results
Young 2017;70 n = 31	Conference abstract	Aged 51–74 years screened with the EarlyCDT Lung test (13 positive, 18 negative) and long-term smokers at screening	NR	Qualitative interviews: looking at how screening affected decisions about smoking, including interpretation of test results and emotional responses to results
Clark 2018;56 n = 338	Paper	Subsample of EarlyCDT Lung-positive participants (split between presence of nodules on first CT scan vs. those without)	December 2013 to April 2015	PANAS, LCWS, Health Anxiety Subscale of the Health Orientation Scale, IES. Revised Illness Perception Questionnaire-adapted for lung cancer and lung cancer risk perception
Young 2018;59 n = 31	Paper	Subsample of EarlyCDT Lung test participants (13 positive, 18 negative)	NR	Qualitative interviews: looking at how screening affected decisions about smoking, including interpretation of test results and emotional responses to results
Clark 2019;57 n = 338	Paper	Subsample of EarlyCDT Lung-positive participants (split between presence of nodules vs. those without)	December 2013 to April 2015	Smoking behaviour
Sullivan 2019;67 n = 12,210	Conference abstract	Adults aged 50–75 years at high risk of developing lung cancer		EarlyCDT Lung test results and number of cancers

Another ECLS paper looked at psychological outcomes among 338 patients who tested positive with the EarlyCDT Lung test. 56 The responses of patients with pulmonary nodules on their first CT scan were compared with those of patients without pulmonary nodules on their first CT scan at 3 and 6 months. The paper reported no statistically significant differences between the groups in affect, lung cancer worry, health anxiety, illness perceptions, lung cancer risk perception or intrusive thoughts. Two papers57,59 reported on smoking behaviour outcomes following lung cancer screening.

Danish cohort of Borg et al.78

Borg et al. 78 performed the EarlyCDT Lung test on a cohort of 246 patients suspected of having lung cancer by their physician. This paper was published after our searches were completed. As patients did not have identified pulmonary nodules, and no data on patients with nodules were reported, the study is not eligible for the main analysis, and so is considered here.

All 246 patients received EarlyCDT Lung, with levels above either the ‘high’ or ‘moderate’ thresholds described in Healey et al. 10 being considered a positive result. Patients then had CT and cancer diagnosis. All patients were followed up for 1 year to confirm or exclude cancer. The mean age was 65 years, with approximately equal numbers of men and women; 76% of patients were current or former smokers. There were 75 diagnosed lung cancer cases (11 stage I, 17 stage II, 22 stage III and 25 stage IV).

The overall estimated diagnostic accuracy of EarlyCDT Lung was a sensitivity of 33% (95% CI 23% to 45%) and specificity of 88% (95% CI 82% to 92%). The paper78 reported diagnostic accuracy for several patient subgroups. The paper noted poor diagnostic accuracy for stage I and II cancers (21% sensitivity for 88% specificity) and for patients aged ≤ 60 years (11% sensitivity at 94% specificity).

The paper78 concluded that EarlyCDT Lung has insufficient sensitivity to be recommended as part of a low-dose CT lung cancer screening programme. The EAG notes that the paper does not report results for patients with pulmonary nodules, and inclusion was based on physician suspicion of cancer alone, so the study is not directly applicable to diagnosing pulmonary nodules. However, the low diagnostic accuracy in the study is consistent with that seen in the meta-analysis in Synthesis of diagnostic accuracy.

Case–control studies of EarlyCDT Lung among patients without confirmed pulmonary nodules

A series of case–control studies were performed to assess the potential diagnostic accuracy of EarlyCDT Lung. These were reported in 2011 in papers by Boyle et al. 29 and Lam et al. 32 All these initial case–control studies were of a different panel of autoantibodies to the current version of EarlyCDT Lung: they consisted of a different set of six autoantibodies, rather than the current seven. Moreover, as these were case–control studies, EarlyCDT Lung was performed after cancer diagnosis, and not after identification of pulmonary nodules. For these reasons, the EAG considers these studies to be ineligible for inclusion in the main synthesis.

One of the case–control groups (235 cases and 236 controls, from the UK, the USA, Ukraine and Russia) was subsequently retested using the current seven-panel version of EarlyCDT Lung, using stored serum samples. The results of this reanalysis were reported in Chapman et al. ,38 alongside some analysis of the included HIPAA cohort. The diagnostic accuracy for this re-evaluated sample was 41% sensitivity (95% CI 35% to 48%) at a fixed 91% specificity. As this analysis was not of patients with diagnosed pulmonary nodules, it was also not included in the main synthesis.

Case–control studies may have substantial risk of bias when assessing diagnostic accuracy. This is because the test is performed after lung cancer diagnosis rather than before, and it is uncertain whether or not the test results (i.e. the levels of autoantibodies) would change over time, altering accuracy. The patients with cancer are unlikely to be representative of patients who would be included in a prospectively recruited cohort. The case–control study may be missing early-stage tumours, which may be harder to diagnose with EarlyCDT Lung. Similarly, the control sample may not represent typical patients with benign nodules, particularly as patients were not matched on nodule characteristics, and control patients may not have had pulmonary nodules at all. The EAG therefore considers the case–control studies to be at high risk of bias for assessing diagnostic accuracy.

This risk of bias is particularly concerning as the case–control group assessed using the seven-panel version of EarlyCDT Lung was analysed again as part of the paper by Healey et al. 10 in 2017. In that paper, the case–control group (called the ‘optimisation cohort’) was reanalysed alongside data from HIPAA, including the subset of patients with pulmonary nodules that was included in our main synthesis. Diagnostic accuracy results were presented and are summarised in Table 9 (based on table 1 of Healey et al. 10).

TABLE 9 - Diagnostic accuracy as reported in Healey et al.10

Group	Sensitivity (95% CI) (%)	Specificity (95% CI) (%)	Positive likelihood ratio (95% CI)
Case–control	41.3 (35.0 to 47.6)	90.6 (87.1 to 94.1)	4.4 (2.9 to 6.6)
HIPAA (all patients)	47.4 (24.9 to 69.8)	90.5 (88.4 to 92.5)	5.0 (3.0 to 8.3)
HIPAA (with nodules)	37.8 (22.2 to 53.5)	85.6 (79.1 to 92.1)	2.6 (1.4 to 4.8)
Small nodules (4–20 mm)	40.0 (15.2 to 64.8)	83.9 (76.2 to 91.6)	2.5 (1.1 to 5.4)
Larger nodules (> 20 mm)	36.4 (16.3 to 56.5)	91.7 (80.6 to 100)	4.4 (1.0 to 18.4)

These results show that diagnostic accuracy from the case–control group was similar to accuracy in the overall HIPAA cohort. However, diagnostic accuracy among patients with pulmonary nodules was notably worse than for the case–control group for sensitivity, specificity and likelihood ratio. This appears to be driven mainly by poorer diagnostic accuracy among smaller nodules, which are both more common than larger nodules in the HIPAA cohort and more likely to be absent from the case–control group (because a cancer has to be diagnosed to be included). The EAG therefore considers that there is reasonable evidence that the diagnostic accuracy estimates from the case–control group may overestimate accuracy among patients with nodules.

In Healey et al. ,10 the diagnostic accuracies for nodule and ‘optimisation’ groups were claimed to be similar because Fisher’s exact tests found no evidence of difference (e.g. Fisher’s exact test for specificity: p = 0.28). However, the number of patients in the nodule group was small (111 patients), and so the EAG considers that lack of evidence of a difference cannot be equated to no difference. The EAG therefore considers that it may be inappropriate to assume that diagnostic accuracy in the case–control group applies to patients with nodules.

The Healey et al.10 risk model

The case–control group in Healey et al. 10 was then used to construct two new sets of EarlyCDT Lung test thresholds: the ‘high-specificity’ threshold (98% specificity, with 28% sensitivity) and the ‘low-specificity’ threshold (49% specificity for 80% sensitivity). The risk model proposed for general use (see Figure 1) was constructed assuming the stated diagnostic accuracy for these two new thresholds is valid. If it is, in fact, an overestimate of the diagnostic accuracy, then the post-test risk estimated by these models will be too high and decisions made using the rule may be invalid.

As the EAG meta-analysis does not support these estimates of diagnostic accuracy, we compare the risk model from Healey et al. 10 with an ‘EAG model’, which has sensitivity estimates taken from the bivariate meta-analysis (see Figure 7) at the same specificity thresholds (sensitivity of 5.1% at 98% specificity; sensitivity of 46% at 80% specificity). We note that this analysis does not account for uncertainty in diagnostic accuracy, in either the EAG analysis or that of Healey et al. 10

For people who test negative with EarlyCDT Lung, the post-test risk was assumed to be unchanged from the pre-test risk. For people who test positive with EarlyCDT Lung, the post-test risk was calculated from pre-test risk and diagnostic accuracy as set out in Healey et al. 10 Briefly, the pre-test risk and sensitivity/specificity of EarlyCDT Lung were combined to estimate the true-positive and false-positive rates, and these were used to calculate the PPV, which was taken to be the post-test risk.

The pre-and post-test risks for the model using the ‘high-specificity’ and ‘low-specificity’ thresholds from Healey et al. 10 and from the EAG model are presented in Figure 9. The increase in risk if EarlyCDT Lung is positive is much smaller for the EAG model, for the ‘high-specificity’ threshold, because of the much lower predicted sensitivity. Consequently, a positive EarlyCDT Lung test is less likely to change a patient’s risk classification in the EAG model [e.g. from low (< 10%) to intermediate risk (10–70%)].

FIGURE 9.

Post-test risk using the Healey et al. 10 and EAG models.

Comparators

Database searches for all comparators described in Chapter 1, Comparators, were performed. These searches produced 3647 potentially relevant publications. Given the size of this database, and the limited evidence on EarlyCDT Lung, it was decided to perform targeted screening of these identified publications using keyword searches in EndNote to identify relevant papers.

Keyword searching was used to identify all likely systematic reviews or meta-analyses (91 papers), and these were screened for inclusion. For comparators for which systematic reviews were not identified, further keyword searches were performed to identify relevant individual studies.

The Brock risk model

We identified no systematic reviews or meta-analyses of the Brock risk assessment model. Targeted keyword searching for ‘Brock’ or ‘PanCan’ (an alternative name for the test) within our EndNote database of comparator studies identified 28 possibly eligible papers, of which nine reported data on the diagnostic accuracy or clinical effectiveness of the Brock model. Studies conducted among Asian populations were excluded because several showed evidence that the Brock model has inferior accuracy in East Asian countries, and so were deemed not relevant to the UK context. 80 Other studies were excluded as no full text was available, no relevant accuracy data were reported or they were multiple publications of the same cohorts.

A summary of the nine included publications is given in Table 10. Three of these papers reported data from the National Lung Screening Trial (NLST) cohort. It was unclear whether the papers analysed the same, or different, patients within the wider cohort. For completeness, we report the results of all papers.

TABLE 10 - Summary of papers reporting diagnostic accuracy data for the Brock model

BCCA, British Columbia Cancer Agency; LUSI, German Lung cancer Screening Intervention; NR, not reported; PanCan, Pan-Canadian Early Detection of Lung Cancer Study; PLuSS, Pittsburgh Lung Screening Study.

Study	Cohort	Location	Design	Participants (n)	Cancers (n)	Mean age (years)	Male (%)	Recent or current smokers (%)
Al-Ameri 20156	Independent	UK	Retrospective cohort	244	99	69	50	76.2
Baldwin 202081	Independent	UK	Retrospective cohort	1187	229	65 (approximately)	51.3	NR
Chung 201882	Independent	Netherlands	Retrospective case–control	1786	381	63 (approximately)	47.4	NR
González Maldonado 202083	LUSI (trial)	Germany	Clinical trial	1159	62	57.6	65.8	100 (61.9 current)
McWilliams 201384	PanCan/BCCA	Canada	Prospective cohort	2961	144	NR	NR	NR
Raghu 201985	PLuSS	USA	Prospective nested case–control	50	42	64	57	55 (current)
Tammemagi 201986	PanCan/NLST	Canada/USA	Prospective cohort	1711	111	62.5	53.1	NR
White 201787	NLST	USA	Prospective cohort	2819	116	62	61	NR
Winter 201988	NLST	USA	Prospective cohort	5018	194	63.7	61.3	NR

Studies were a mix of prospective and retrospective cohorts, with two case–control studies and one clinical trial. All appeared to use a reasonable reference standard of biopsy or surgery or clinical follow-up to confirm the presence or absence of cancer. In all studies, CT was performed before diagnosis, but in retrospective cohorts the Brock risk calculation will have been performed after diagnosis. This is unlikely to lead to substantial bias, given that Brock risk is based on the CT results.

Meta-analysis

Studies generally reported the AUROC curve to summarise diagnostic accuracy, rather than using sensitivity and specificity. This accounts for the fact that the Brock model might be assessed at different risk cut-off points (e.g. 5%, 10%). The forest plot for the meta-analysis of reported AUC values is given in Figure 10. This suggests that the Brock model has very good diagnostic accuracy (AUC 92%, 95% CI 90% to 95%), but with some evidence of heterogeneity across studies (I² = 90%), with estimated AUCs varying from 79% to 96%. We note that AUC does not provide evidence of the diagnostic accuracy at specific cut-off points of interest (such as the 10% risk cut-off point). A sensitivity meta-analysis excluding two of the three papers reporting data on the NLST cohort, and retaining only the most recent (Winter et al. ,88 from 2019) had a very similar result (AUC 91%, 95% CI 87% to 95%).

FIGURE 10.

Forest plot of AUC values for studies assessing the Brock risk model. The Chung et al. 82 study did not report a CI for the AUC, and so was not included in the meta-analysis.

Five of the included studies reported sensitivity and specificity estimates for the Brock model at various thresholds. These are plotted in Figure 11. There was some heterogeneity across studies, even when using the same threshold of risk (e.g. 10%, the squares in Figure 11), but all studies suggest high diagnostic accuracy, with 80% sensitivity at 90% specificity appearing to be achievable. This contrasts with the estimated 25% sensitivity at 90% specificity for EarlyCDT Lung.

FIGURE 11.

Sensitivity and specificity estimates by test threshold for the Brock model.

As there were only two cohorts [i.e. the NLST and the German Lung cancer Screening Intervention (LUSI)] reporting sensitivity and specificity at most risk thresholds, and as these had heterogeneous results (see Figure 11), no meta-analysis of sensitivity or specificity is presented here. Consequently, the diagnostic accuracy of the Brock model at any particular risk cut-off point (e.g. the 10% cut-off point to distinguish low-risk and intermediate-risk nodules) is uncertain.

The Herder risk model

We did not identify any systematic reviews or meta-analyses of the Herder risk assessment model. Targeted keyword searching for ‘Herder’ or ‘Mayo’ in our EndNote database of comparator studies identified seven possibly eligible studies, of which four reported data on the diagnostic accuracy or clinical effectiveness of the Herder model explicitly. Given this limited number of studies, we also included two studies that reported diagnostic accuracy when combining PET-CT with the Mayo risk tool, which is functionally very similar to the Herder model. As for the Brock model, studies in Asian populations were excluded. One study was excluded as it reported no relevant data.

A summary of the six included cohorts is given in Table 11. All six papers reported data from different cohorts. All studies were retrospective cohort studies. All appeared to use a reasonable reference standard of biopsy and surgery or clinical follow-up to confirm the presence or absence of cancer. In all studies, PET-CT was performed before diagnosis, but the Herder or Mayo risk calculation will have been performed after diagnosis. This is unlikely to lead to substantial bias, given that Herder risk is based on the results of PET-CT.

TABLE 11 - Summary of papers reporting diagnostic accuracy data for the Herder risk model

Study	Test	Location	Size (n)	Cancers (n)	Mean age (years)	Male (%)	Recent/current smokers (%)
Al-Ameri 20156	Herder	UK	244	99	69	50	76.2
Herder 200512	Herder	Netherlands	106	61	63	33	75
Murphy 201989	Herder	UK	97	75	69	52	84
Perandini 201790	Herder	Italy	259	153	66	64	90
Evangelista 201491	PET-CT + Mayo	Italy	59	31	70	54	54
Isbell 20119	PET-CT + Mayo	USA	189	138	63	50	74

Meta-analysis

For the Brock model, most studies presented results as summary AUCs. A forest plot of these results is shown in Figure 12. These results suggest good diagnostic accuracy for the Herder model overall, with an AUC of 84% (95% CI 77% to 92%). There was substantial heterogeneity. The much lower accuracy seen in the Perandini et al. 90 study than in earlier studies was notable. That paper acknowledged this difference, but could not explain the heterogeneity.

FIGURE 12.

Forest plot of AUCs for the Herder risk model.

Only three studies reported any sensitivity or specificity estimates for the Herder model. These are summarised in Appendix 5, Figure 24. Again, these suggest only moderate to good diagnostic accuracy, of approximately 50–60% sensitivity at 90% specificity. Data were too limited to perform meta-analyses at any specific risk cut-off points.

Further analyses of clinical effectiveness

As noted in Synthesis of other clinical effectiveness outcomes, no study presented any evidence on the clinical impact of using EarlyCDT Lung. Similarly, no study has reported any evidence on the diagnostic accuracy or clinical impact of using EarlyCDT Lung within the diagnostic pathway, in combination with the Brock and Herder risk assessments. To address these issues, we performed a simulation study to examine the potential impact of using EarlyCDT Lung within the diagnostic pathway, and in accordance with BTS guidance.

Results

Figure 13 presents a bar chart of the extracted data from the Al-Ameri et al. 6 and Perandini et al. 90 publications showing the distribution of risks. People with benign nodules typically have risks of < 20% for both Brock and Herder. Results for people with malignant nodules are more variable, with a mix of both low and high risks, although risks with Herder model are skewed towards higher values.

FIGURE 13.

Distribution of risks in the Al-Ameri et al. 6 and Perandini et al. 90 studies. (a) Brock: Al-Ameri et al. 6 (benign); (b) Herder: Al-Ameri et al. 6 (benign); (c) Herder: Perandini et al. 90 (benign); (d) Brock: Al-Ameri et al. 6 (malignant); (e) Herder: Al-Ameri et al. 6 (malignant); and (f) Herder: Perandini et al. 90 (malignant).

Diagnostic accuracy of Brock risk

Table 15 summarises the diagnostic accuracy of Brock risk, and combining Brock with EarlyCDT Lung at a risk threshold of 10%, which is the cut-off point to distinguish between low-risk nodules to go to CT surveillance and higher-risk nodule requiring further investigation. Adding EarlyCDT Lung may slightly improve sensitivity, while reducing specificity, as would be expected, because more patients will be ‘test-positive’ after EarlyCDT Lung assessment. However, changes are small. Similarly, changes in positive or negative predictive values are small. This means that people with a post-test risk of > 10% after EarlyCDT Lung are no more likely to have malignant nodules than when using Brock risk alone. Differences between the Healey et al. 10 and EAG model are small, with no clear evidence of difference. This may be because few patients are test-positive at the ‘high-specificity’ threshold for the Healey et al. 10 model.

TABLE 15 - Diagnostic accuracy of combining Brock risk with EarlyCDT Lung at the 10% risk threshold

NPV, negative predictive value.

Method	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
Brock model only	92.9	62.1	62.6	92.8
With EarlyCDT Lung (Healey et al.10 model)	95.6	58.5	61.2	95.1
With EarlyCDT Lung (EAG model)	95.8	57.1	60.4	95.2

Figure 14 shows the full summary ROC curves at all risk thresholds. Results for thresholds of 10%, 20% and 70% are shown. These results also show no clear benefit of adding EarlyCDT Lung to Brock risk assessment, with ROC curves being close together. There is a possible, but small, improvement in sensitivity, matched by a decline in specificity, at each risk threshold. Improvements in sensitivity appear to occur at higher-risk thresholds (e.g. 70%), but the Brock risk is not generally used at higher levels of risk. The improvement in sensitivity is notably smaller using the EAG model than the Healey et al. 10 model.

FIGURE 14.

Summary ROC curve when combining EarlyCDT Lung with Brock risk.

Diagnostic accuracy of Herder risk

Table 16 summarises the diagnostic accuracy of Herder risk, and combining Herder with EarlyCDT Lung at risk thresholds of 10% and 70%, which are the cut-off points to distinguish between low-risk, intermediate-risk and high-risk nodules. As for Brock risk, at the 10% threshold, adding EarlyCDT Lung to the diagnostic pathway leads to no clear improvement in sensitivity, but with a possible small drop in specificity. Differences in diagnostic accuracy are too small to be conclusive.

TABLE 16 - Diagnostic accuracy of combining Herder risk with EarlyCDT Lung at the 10% and 70% risk thresholds

NPV, negative predictive value.

Method	Data	Risk threshold (%)	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
Herder only	Al-Ameri et al.6	10	97.7	66.0	82.4	94.6
With EarlyCDT (Healey et al.10 model)			98.3	63.7	81.5	95.8
With EarlyCDT (EAG model)			97.7	65.3	82.0	94.5
Herder only		70	72.1	90.6	92.5	66.7
With EarlyCDT (Healey et al.10 model)			81.4	87.9	91.7	74.6
With EarlyCDT (EAG model)			78.9	88.0	91.4	72.0
Herder only	Perandini et al.90	10	91.8	52.4	69.8	84.3
With EarlyCDT (Healey et al.10 model)			95.1	47.9	68.6	89.2
With EarlyCDT (EAG model)			94.6	49.2	69.0	88.4
Herder only		70	49.0	89.0	84.2	59.3
With EarlyCDT (Healey et al.10 model)			59.9	87.3	85.0	64.6
With EarlyCDT (EAG model)			55.6	87.8	84.4	62.3

At the 70% risk threshold, there is a possibility that using EarlyCDT Lung will increase sensitivity substantially while reducing specificity by around 1–2%. This increase in sensitivity is smaller when using the EAG model. This translates into some possible improvement in negative predictive value, but no change in PPV. However, some differences between the Al-Ameri et al. 6 and Perandini et al. 90 data sets makes drawing firm conclusions difficult.

Figure 15 shows the full summary ROC curves at all risk thresholds. Results for thresholds of 10%, 20% and 70% are shown. When using the Healey et al. 10 data and model, these results show a possible increase in sensitivity when adding EarlyCDT Lung to Herder risk assessment, which is most prominent at higher risk thresholds. When using the EAG preferred estimates of diagnostic accuracy, however, this apparent benefit of adding EarlyCDT Lung is substantially reduced.

FIGURE 15.

Summary ROC curve when combining EarlyCDT Lung with Herder risk: (a) Al-Ameri et al. ;6 and (b) Perandini et al. 90

Impact on clinical decision-making

To assess the impact of adding EarlyCDT Lung to Brock and Herder risk assessments, Table 17 shows how many patients would be reclassified into a higher-risk group after using EarlyCDT Lung for four risk categories (0–10% after Brock risk assessment, and 10–20%, 20–50% and 50–70% after Herder risk assessment). Numbers are shown as a percentage of the pre-test risk group. Fuller results, including 95% bootstrap CIs, are given in Appendix 5, Table 34.

TABLE 17 - Summary of patient risk reclassification when using EarlyCDT Lung in combination with Brock and Herder risk assessments

Test	Data	Model	Risk group (%)	As proportion of risk group
				Correctly upgraded	Incorrectly upgraded	Correctly upgraded to > 70% risk	Incorrectly upgraded to > 70% risk
Brock	Al-Ameri et al.6	Healey et al.10 model	0–10	3.0	7.5	0.0	0.0
		EAG model	0–10	2.8	5.1	0.0	0.0
Herder	Al-Ameri et al.6	Healey et al.10 model	0–10	1.5	3.2	0.0	0.0
			10–20	16.0	12.5	3.4	0.2
			20–50	34.9	0.3	29.0	0.3
			50–70	27.9	6.8	27.9	6.8
		EAG model	0–10	0.0	1.2	0.0	0.0
			10–20	16.4	12.6	0.0	0.0
			20–50	31.8	0.0	1.4	0.0
			50–70	28.6	6.9	28.6	6.9
Herder	Perandini et al.90	Healey et al.10 model	0–10	6.2	7.3	0.0	0.0
			10–20	20.8	10.1	3.2	0.7
			20–50	20.6	4.2	16.4	0.9
			50–70	32.3	4.9	32.3	4.9
		EAG model	0–10	5.2	5.1	0.0	0.0
			10–20	21.1	10.2	0.0	0.0
			20–50	8.6	2.5	0.0	0.0
			50–70	32.8	5.0	32.8	5.0

In the 0–10% risk group, assessed using Brock risk, the number of people with a malignant nodule correctly reclassified as having > 10% risk (and so meriting PET-CT or biopsy, rather than CT surveillance) is fairly small, at around 3% for both the Healey et al. 10 and EAG models. This is because of the low diagnostic accuracy of EarlyCDT Lung in this risk group, and the small number of malignant nodules. A small number of people with benign nodules will be incorrectly reclassified as having > 10% risk (either 5.1% or 7.5% of the group). The number of incorrect reclassifications exceeds the number of correct reclassifications by approximately 2 : 1. The increase in risk with a positive EarlyCDT Lung is never sufficient to increase risk to > 70%.

Therefore, combing Brock with EarlyCDT Lung may mean that more people are wrongly reclassified, and will have an unnecessary biopsy, than are correctly reclassified. It is unclear whether or not any benefits of correctly identifying some malignant nodules will outweigh the harms of these unnecessary biopsies, or what the clinical benefits of EarlyCDT Lung might be in smaller nodules for which biopsy would not be feasible.

The results among intermediate-risk patients, after Herder risk assessment, appear more favourable. Using the model proposed by Healey et al. ,10 at ≥ 20% pre-test risk, 20–35% of patients will be correctly reclassified to a higher risk, generally to > 70% risk. By comparison, few patients with benign nodules are wrongly reclassified (at most 7%). Results with the EAG model were broadly similar. This suggests that a positive EarlyCDT Lung test in this risk range may be a good indicator of a malignant nodule.

There is some variation in results between the Al-Ameri et al. 6 and Perandini et al. 90 data sets, suggesting some uncertainty in the exact proportions of patients who will have risk reclassified after EarlyCDT Lung.

It is currently unclear what the clinical impact of such a reclassification would be, as it is not clear that there is any clinical benefit from proceeding directly to surgery, rather than first receiving a biopsy.

Discussion

Search and studies identified

The search detailed in Chapter 2, Literature search, identified 3233 records. The first stage of screening identified two potentially relevant records, based on their title and/or abstract. The corresponding full-text articles were retrieved and assessed for inclusion. The two studies120,121 met the inclusion criteria (see Chapter 2, Cost-effectiveness reviews, Study selection) and were included in this review.

Methods and key assumptions of the identified studies

The two identified studies are summarised in Table 18. The quality assessment of these studies followed a checklist specific to model-based economic evaluations of diagnostic tests,21 which is reported in Appendix 7 (see Tables 35 and 36 for Edelsberg et al. 120 and Sutton et al. ,121 respectively).

Both studies assess the cost-effectiveness of EarlyCDT Lung, compared with routine CT surveillance, for the diagnosis of lung cancer among patients with solid pulmonary nodules using a decision-modelling approach. The two studies took a health-care payer perspective; Edelsberg et al. 120 is set in the US health-care system, whereas Sutton et al. 121 is set in the UK NHS. Edelsberg et al. 120 assumed a higher cost of EarlyCDT Lung than Sutton et al. 121 (cost per test: US$575 vs. £70).

The proposed positioning of EarlyCDT Lung and the target patient population are defined differently in the two studies. The target population in Edelsberg et al. 120 is defined as patients with incidentally detected nodules of 8–30 mm and intermediate risk (5–60%) of lung cancer. This population was considered relevant by the authors because there was some evidence that this group of patients might be followed up with routine CT surveillance instead of PET-CT, as recommended by the ACCP for nodules of this size. Sutton et al. 121 considered a patient population with nodules of 4–20 mm, and a risk of malignancy (lung cancer) of 10–65%. The population choice was not explicitly justified in this study. The authors state only that the BTS guidelines consider that some nodules with a 10–70% risk of malignancy have too low a risk to be considered for biopsy and can instead be followed up with CT surveillance or ‘watchful waiting’.

Both studies assumed the same prevalence of malignant nodules (9.5%), sourced from a study on the diagnostic follow-up and management of nodules of 8–30 mm by a pulmonologist and/or a thoracic surgeon in a North American (USA and Canada) setting. 122 In what concerns the cancer stage distribution at baseline, Edelsberg et al. 120 assumed that all patients with malignant nodules had local disease (sourced from Tanner et al. 122), whereas Sutton et al. 121 assumed that only 87.5% had local disease and the rest had regional disease (sourced from Gould et al. 123).

Both studies compare EarlyCDT Lung in addition to a CT surveillance schedule with CT surveillance alone. Patients in the EarlyCDT Lung + CT surveillance strategy receive a one-off EarlyCDT Lung test at the start of the model, which produces a dichotomous test result (positive/negative). Neither of the studies reports the criteria for positivity (i.e. the diagnostic cut-off point for each of the seven autoantibodies in the test panel and whether one or more autoantibody levels need to be elevated for the test to be positive). Patients who test positive with EarlyCDT Lung receive a diagnostic biopsy (some patients in Edelsberg et al. 120 receive wedge resection instead). Biopsy is assumed to be 100% accurate and patients with benign nodules (false-positive result on EarlyCDT Lung) proceed to CT surveillance whereas those with malignant nodules proceed to excision. Patients who test negative with EarlyCDT Lung enter CT surveillance to assess tumour growth, following the schedules described in Table 18.

All patients under CT surveillance remain under CT surveillance until they test positive (i.e. until nodule volume doubles in Edelsberg et al. ;120 unclear in Sutton et al. ,121 but also based on change in nodule volume) or reach the end of the surveillance interval.

Patients who test positive during CT surveillance in Sutton et al. 121 receive a biopsy to excise the nodule (type of surgery not specified in the manuscript). It is unclear how patients who test positive during CT surveillance in Edelsberg et al. 120 are managed.

The modelling approach taken by Edelsberg et al. 120 is insufficiently described in the manuscript, but it appears to quantify long-term outcomes using life-expectancy projections (and HRQoL gains) conditional on nodule malignancy, cancer histology [non-small cell lung cancer (NSCLC) or small cell lung cancer (SCLC)], cancer stage (local, regional or distant) and patient characteristics (see Evidence linkage). The model tracks VDT and cancer stage progression over 2 years, with VDT and progression probabilities informed by Gould et al. 123 The stage distribution for each strategy is calculated at 2 years based on this. It is unclear from which time point in the model the life-expectancy projections were applied. This model also considers overdiagnosis of indolent malignant nodules (i.e. nodules that are not aggressive, despite being malignant); this was implemented as a reduction in the malignancy prevalence of 18% (based on data from a lung cancer screening population). The authors state that because all lung cancers are diagnosed by both strategies under comparison, overdiagnosis is equally frequent for these strategies and it affects only the de facto prevalence.

The modelling structure in Sutton et al. 121 comprises a decision tree and a Markov model with monthly cycles and half-cycle correction. The Markov model component is stated to have the same structure as the model used by Gould et al. 123 At the first cycle in the model, the decision tree dichotomises patients in each strategy according to the disease status (positive/negative) and then applies test diagnostic accuracy estimates to classify patients according to their test results as true positive, false negative, true negative and false positive. Patients then enter the Markov component of the model according to whether they have been correctly diagnosed (diagnosed health states: benign or malign) or not (undiagnosed health states), and their disease stage for patients who have malignant nodules.

In the Markov model, only patients in undiagnosed cancer health states (local or regional) seem to be able to progress between disease stages (local to regional and regional to distant). Progression between cancer stages for undiagnosed people is dependent on nodule growth over time. Patients in the diagnosed benign state remain in the state, but may undergo surgical biopsy in future cycles if CT surveillance detects nodule growth. Patients with diagnosed (and treated with either surgery alone or with chemotherapy or radiotherapy) malignant local and regional nodules have a time-dependent mortality risk because of cancer (due to recurrence or regional cancer for diagnosed local and regional cancer, respectively) for 5 years, after which they transition to the disease-free state. Patients in the diagnosed distant states have a lifetime constant risk of cancer-related mortality, as do patients with undiagnosed malignant nodules. Patients in all health states are subject to age-adjusted general population mortality. Cancer-related mortality, probability of benign nodule growth and disease progression probability were sourced from Gould et al. 123

Two scenarios are analysed separately in each study, considering alternative values for the diagnostic accuracy of EarlyCDT Lung: scenario A considers a sensitivity and specificity for EarlyCDT Lung of 41% and 93%, respectively, and scenario B considers a sensitivity and specificity of 98% and 28%, respectively (both value sets are sourced from Healey et al. ;10 see details in Chapter 3, EarlyCDT Lung studies). These scenarios were each evaluated as a pairwise comparison against CT surveillance alone. One of the key differences between the two studies is that CT surveillance is assumed to detect all malignant nodules over the 2 years of follow-up in Edelsberg et al. , whereas, in Sutton et al. ,121 there is misclassification under CT surveillance, leading to a proportion of undiagnosed malignant nodules at the end of the surveillance schedule. Sutton et al. 121 synthesised diagnostic accuracy data from studies identified in Gould et al. 123 to inform the sensitivity and specificity of CT (92.3% and 72.3%, respectively). Both Edelsberg et al. 120 and Sutton et al. 121 assumed that biopsy was 100% accurate.

Results of the identified studies

Table 19 summarises the cost-effectiveness results from the two studies. Both studies conclude that EarlyCDT Lung is a cost-effective use of health-payer resources, compared with CT surveillance alone, as the incremental cost-effectiveness ratios are below the cost-effectiveness thresholds in the studies’ jurisdiction. Despite the two models relying on similar data sources and assumptions, the incremental cost-effectiveness ratios of EarlyCDT Lung + CT surveillance, compared with CT surveillance alone, differ substantially across the two studies.

Beyond the difference in the per-patient costs of the test itself, it is difficult to understand which parameters are driving these differences in cost-effectiveness, given the lack of detail and clarity on important analytical choices in both models. Furthermore, Sutton et al. 121 do not report life-years gained in the model or conduct a thorough exploration of parameter and structural uncertainty, which would have aided interpretation of differences between models. The main differences between the two models in terms of parameterisation and structural assumptions are as follows:

• higher cost of EarlyCDT Lung in Edelsberg et al. ,120 compared with Sutton et al. 121 (cost per test: US$575 vs. £70)

• higher costs of health care in Edelsberg et al. 120

• baseline distribution of malignant nodules across disease stages

• actual malignancy prevalence estimate applied in the model: approximately 2% lower in Edelsberg et al. 120

• assumption of no misclassification at the end of surveillance in Edelsberg et al. 120

• Edelsberg et al. 120 only explicitly model disease progression during the 2 years of CT surveillance with a 55.3% probability of progression over 2 years (or an annual probability of 33.1%)

• modelling of long-term outcomes as life-expectancy (HRQoL adjusted and unadjusted) pay-offs,120 versus through a Markov model. 121

We do not focus here on examining the studies’ results, as these are unlikely to be appropriate to inform the decision problem defined by the NICE scope (see Decision problem and relevance to National Institute for Health and Care Excellence Diagnostics Assessment Report scope). However, the differences in results between studies suggest that there are differences in terms of how each study modelled the value of EarlyCDT Lung and the evidence linkage used to translate this into impact on health and cost differences. In the next section, we first critique the two studies in terms of their relevance to the scope of this assessment. We then examine in more detail the drivers and components of value of EarlyCDT Lung (vs. CT surveillance) and the evidence linkage approach taken, to better understand the modelling of the mechanisms of value accrual and support the development of a conceptual model to assess the cost-effectiveness EarlyCDT Lung.

Critique

Conclusions of cost-effectiveness review of EarlyCDT Lung studies

There is limited evidence on the cost-effectiveness of EarlyCDT Lung on the diagnostic pathway for pulmonary nodules, with neither of the two studies identified being considered suitable to inform the current decision problem because of important differences in, for example, the patient population, the position and use of EarlyCDT Lung within the diagnostic pathway and exclusion of relevant diagnostic comparators, and the diagnostic accuracy evidence used to inform it.

The existing evaluations consider and quantify a number of components of clinical and economic value for EarlyCDT Lung among patients otherwise referred to CT surveillance, including (1) the increased cost of testing with EarlyCDT Lung, (2) the cost and adverse event trade-offs of replacing CT surveillance with further investigations for those testing positive on EarlyCDT Lung, (3) the early detection of lung cancer (and potential stage shift) among those with true-positive test results, and (4) the potential for increased detection of lung cancer if some of the true positives would have been missed by CT surveillance. The mechanism of value from increased detection is also expressed as early detection and assumes that cancers missed by CT surveillance would present clinically later in time. Despite overtreatment of indolent malignant nodules being of unclear relevance, neither model reflected the potential for increased overtreatment with the introduction of EarlyCDT Lung.

The evidence used to inform the population, the diagnostic outcomes, and the health and cost outcomes is sparse in many key aspects that will drive value such as the prevalence of malignancy and disease progression under CT surveillance. Modelling relies on unclear structural assumptions, without the support of relevant evidence. Therefore, the EAG considers that the evidence supporting the modelled effect on stage distribution (stage shift) is very limited.

The use of EarlyCDT Lung as part of a screening strategy for lung cancer has been evaluated in a large trial conducted in Scotland. 14 For the reasons described in Chapter 3, The Early Detection of Cancer of the Lung Scotland trial, there is little relevance of this evidence to inform the clinical effectiveness of EarlyCDT Lung in the diagnostic pathway. As a consequence, the within-trial cost-effectiveness evidence is also of little relevance to this assessment.

Given the limited evidence on the cost-effectiveness of EarlyCDT Lung in the diagnostic pathway for pulmonary nodules, and to allow a fuller critical assessment of the assumptions and data sources used in the existing cost-effectiveness studies and assist in the conceptualisation of a new decision model, further targeted literature searches for cost-effectiveness studies were undertaken. The review of the identified studies is reported in Chapter 5.

Searches and studies identified

The searches retrieved 615 records, of which 546 were excluded on the basis of title and or abstract. Full-text publications were retrieved for 77 records and these were screened for potential inclusion in the reviews of diagnostic (28 titles) or screening studies (49 titles). The full-text publications of two records identified at the first stage of screening for potential inclusion in the screening review were not retrievable, and were, therefore, excluded from the review.

Forty-five studies met the inclusion/exclusion criteria for inclusion in the reviews. Ten of these studies were cost-effectiveness studies of diagnostic tests. 120,121,123,126–132 Because two of the studies120,121 had already been reviewed (see Chapter 4), only the remaining eight studies were included in the review of diagnostic studies. Of the 36 screening studies,107,133–167 one study134 did not report sufficient information to characterise the evidence linkage, and was excluded from the screening review. The remaining 35 studies107,133,135–167 were included in the screening reviews.

Details on both the diagnostic and screening reviews are reported in Appendix 8. A summary of the reviews is presented over the next sections.

Summary of the review of cost-effectiveness studies on diagnostics for lung cancer diagnosis

Table 23 identifies and briefly summarises the eight diagnostic studies123,126–132 in terms of the population and important features of the sequences of diagnostic tests considered. It also summarises the three key components of each evaluation: the final classification (i.e. how the nodules were classified at the end of the diagnostic strategy), the treatment choices (determined by the classification) and whether or not long-term health outcomes were linked to disease staging.

All studies appear to condition long-term health outcomes on disease staging.

The indirect value components (i.e. those relating to classification) identified across studies (see detail by study in Appendix 8, Table 37) were as follows:

• earlier diagnosis/increased detection of lung cancer

• management of those with false-positive results with unnecessary follow-up tests and/or treatment, and decisions to treat benign nodules

• regression of benign nodules leading to early discharge from CT surveillance.

All the diagnostic studies modelled earlier diagnosis,123,126–132 and all but one127 considered increased detection. The increased detection, compared with surveillance, was imposed variably and relied mostly on assumptions on the specificity of CT surveillance or the uptake of CT surveillance rather than robust evidence (see Appendix 8, Table 38). As in the cost-effectiveness studies on EarlyCDT Lung (see Chapter 4, Critique), both earlier diagnosis and increased detection were modelled via stage shift.

The delay to diagnosis with CT surveillance was modelled either by assuming that diagnosis occurred at a single specific point in time in the future or across multiple future time points (see details in Appendix 8); this was informed either by assumptions or by explicit modelling of nodule growth. The evidence used to inform models of nodule growth was not robust or appropriate (e.g. one study123 used the same VDT that was used to inform the EarlyCDT Lung cost-effectiveness studies;120,121 see Chapter 4, Evidence linkage). Another study modelled nodule growth and disease progression using an existing natural history model developed to simulate the outcomes of patients identified by screening,129 but insufficient detail is provided to characterise the evidence linkage and its appropriateness.

The delay to diagnosis was linked to disease staging by either assuming fixed stage shift for tumours with non-immediate diagnosis (e.g. all tumours diagnosed by CT surveillance progress from stage 1 to stage 2) or using a preclinical (i.e. before diagnosis) progression model. The assumptions in models reflecting a fixed stage shift from the delay to diagnosis were not justified. The models that included a preclinical progression component were informed by (1) lung cancer screening trial data,168 (2) VDT data collected with pre-CT imaging technology124 or (3) elicited evidence from public health policies to promote early diagnosis of lung cancer. 169

Overdiagnosis of indolent malignant nodules is not modelled in any of the diagnostic studies. Some studies consider treatment for a proportion of benign nodules (i.e. those that show growth during CT surveillance; see Appendix 8), and reflect this on short-term mortality and morbidity in the health outcomes and costs considered in the models, but evidence supporting malignant growth rates for benign nodules is not robust. In addition, some studies consider the possibility of nodules presenting a negative biopsy being referred to treatment, reflecting that biopsy results may have limited bearing on treatment decisions.

The handling of false-positive results is detailed in Appendix 8. False-positive results at the end of the overall diagnostic strategy are handled in the identified studies by applying to the patients who undergo unnecessary surgical treatment the procedural mortality, HRQoL loss and costs associated with surgery.

To establish the link to final outcomes, the models conditioned outcomes on disease status and to disease stage for patients with lung cancer (see Appendix 8). Survival outcomes of lung cancer patients were also conditioned on age. One study129 included a competing mortality risk for coronary artery disease (CAD), thus reflecting comorbidity in the study population that was composed of patients with incidentally detected solid pulmonary nodules who underwent investigations for CAD. The HRQoL of patients with lung cancer was conditioned on staging, histology, recurrence of cancer, type of treatment and response, and time post treatment. Only one study132 considered that lung cancer costs vary by disease stage; other studies seemed to reflect mostly the costs of immediate cancer treatment with surgery. The health outcomes of patients with benign nodules were conditioned on age and sex, and generally reflect those of the general population. The models assumed that these patients did not accrue costs beyond those determined by the diagnostic pathway (procedural costs with or without complications).

Some studies considered the possibility of a proportion of benign nodules regressing during CT surveillance, but do not provide details on how this component of value was modelled (see Appendix 8). Regression of benign nodules may lead to early discharge from surveillance for a proportion of patients who will no longer incur the costs of CT surveillance and potentially assuage anxiety due to surveillance.

The review identified a single UK study,132 which used UK-relevant evidence on long-term survival, costs and HRQoL, all by disease stage at diagnosis. The study sourced other-cause mortality from UK life tables.

Summary of the review of cost-effectiveness studies on screening for lung cancer

As stated in Searches and studies identified, 34 studies on the cost-effectiveness of screening for lung cancer were identified by the searches. Given that the aim of the review was to have a general (but not comprehensive) understanding of how value components relevant to EarlyCDT Lung were modelled in the screening literature, and given the high volume of studies identified, we selected a sample of publications for review. This sample of screening studies aimed to include a sufficient range of modelling approaches. We also included in this sample all identified UK model-based cost-effectiveness studies, as the evidence used in these studies is more likely to be relevant to the UK context. The fully reviewed studies are identified and briefly summarised in Table 24 in terms of the type screening strategies considered (no screening vs. one-off screening and/or repeat screening); key features of the disease model, including the modelling approach; the sources of effectiveness data; and whether or not survival outcomes were linked to disease staging.

There is one key common mechanism by which screening strategies derive value, compared with no screening, and this relates to earlier diagnosis arising from identification of preclinical cancer that would have otherwise only been clinically detected (this is commonly denominated ‘lead time’ in the screening literature). The link between early diagnosis and outcomes is mediated via a disease stage shift in almost all models. This is similar to the mechanism modelled in the diagnostic studies reviewed in Chapter 4 and in Summary of the review of cost-effectiveness studies on diagnostics for lung cancer diagnosis. The studies differ in terms of the clinical evidence used to inform the lead time estimates and stage shift, and in how this evidence is used. For example, some studies used (experimental) comparative effectiveness evidence of lung cancer screening by low-dose computerised tomography to infer preclinical to clinical progression. 154,155,158,160,170 One model estimated the probabilities of preclinical to clinical progression using cancer registry data. 143 Other models did not model preclinical to clinical progression, and used clinical effectiveness evidence differently. For example, some studies directly applied non-randomised evidence for stage distributions of screened versus clinically detected lung cancer, combined with assumptions on lead time and survival conditional on stage. 21,107,139,142,151,152,163

All studies except one164 condition survival outcomes on stage at detection.

The value components relating to classification identified across studies (see detail by study in Appendix 8, Table 40) were as follows:

• earlier diagnosis (increased) detection of lung cancer

• earlier recalls resulting in some patients undergoing additional screening after a suspect result and incurring delays to diagnosis

• overdiagnosis of malignant indolent tumours

• management of those with false-positive results with unnecessary follow-up tests and/or treatment

• radiation exposure with increased cancer risk.

The studies established the evidence linkage required to model early diagnosis in screening models (see Appendix 8) in two main ways: (1) by modelling preclinical to clinical progression or (2) by linking effectiveness data on stage distribution, combined with assumptions on lead time, to survival outcomes.

Where disease progression is explicitly modelled (see Appendix 8), the lead time and stage shift for screened versus unscreened patients with lung cancer is quantified by tracking patient flow in the natural history model until detection (clinical or via screening). Overdiagnosis (i.e. the proportion of tumours that are detected with screening in excess of those clinically presenting with a no-screening strategy) is also a model output. The probabilities of preclinical to clinical progression were inferred using calibration methods and (mostly) comparative evidence from RCTs on lung cancer screening [e.g. the NLST and the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial in Ten Haaf et al. 159]. Preclinical to clinical progression probabilities are stage specific in these models, and two models further conditioned these probabilities on tumour histology. 158–160 All assume that preclinical progression is sequential across disease stages. One study160 explicitly modelled the relation between tumour size, tumour growth and metastatic spread, and linked it to disease progression (and probability of cure). Most of these studies do not model disease progression after lung cancer detection; the exception is Hofer et al. ,143 which models progression across three stages of ‘aftercare’ and further treatments (chemotherapy + radiotherapy or palliative treatment).

The models without a preclinical to clinical progression component (see Appendix 8) rely more heavily on assumptions and are more likely to be affected by bias. For example, two studies107,139,142 used evidence on stage distribution for screened patients from screening studies (UKLS or Manchester lung screening pilot), but used data from national cancer statistics for unscreened patients: this implicitly assumes comparability between lung cancer patients participating in screening pilots and those clinically detected. Another issue with these models is that they require assumptions to model lead time, and these assumptions are not always robustly supported by evidence (see Appendix 8). Failure to appropriately model lead time risks biasing survival estimates, which may be overestimated for patients with screened detected cancers. Lead time bias arises from screening prolonging the interval between diagnosis and death (even if early treatment had no effect on patient survival), as diagnosis occurs earlier with screening than with clinical detection. Thus, it is important that estimated survival benefits do not unduly incorporate lead time. The handling of lead time bias in models without a preclinical to clinical progression component varied, either by a direct adjustment on survival estimates (relying on assumptions) or a differences-in-differences methodology was applied to age-adjusted survival differences between screened and unscreened patients with lung cancer (see Appendix 8).

Although models with a preclinical to clinical progression component do not rely solely on assumptions to estimate lead time, lead time bias can still arise in these models if additional constraints are not placed on survival. For example, one of the UK-based models154,155 imposed the same lung cancer survival in each disease stage regardless of the type of detection (screening vs. clinical).

The survival of lung cancer patients (see Appendix 8, Table 42) was conditioned across most models on staging, histology and age. Some studies also conditioned the survival of these patients on detection type. One study160 explicitly links survival to the probability of cure, which is conditional on tumour size and metastatic burden.

A common assumption across studies that modelled preclinical to clinical progression was that of no or negligible lung cancer mortality in preclinical stages (i.e. patients could die only of other causes). One study154,155 explicitly allowed for early diagnosis within the same disease stage (comparing screening with no screening), so that, for a proportion of patients, there was no stage shift with early diagnosis. However, the model did not assume any survival benefit for early diagnosis in the absence of a stage shift, because the authors considered that evidence suggesting improved survival for screen-detected cancers versus non-screen-detected cancer (when detected at the same stage) was at high risk of bias.

The HRQoL of patients with lung cancer was conditioned across models on staging, histology, detection type (clinical or screening), treatment and/or treatment type, time post successful treatment, post-detection/treatment (clinical) health state, proximity to end of life, age and sex; the majority of studies conditioned HRQoL on staging, age and sex (see Appendix 8, Table 42). HRQoL was assumed to be constant over time (post detection) or to vary with time (1) with age or (2) assuming general population utility after being disease free for 5 years. One study154,155 assumed a temporary disutility from screening for both individuals with and those without lung cancer to reflect anxiety associated with undergoing the intervention.

The costs incurred by patients with lung cancer (see Appendix 8, Table 42) were also conditioned on staging across a number of studies. Costs were either assumed to be constant over time or to vary with time, dependent on time elapsed post diagnosis/treatment and/or phase of treatment (initial vs. later treatment).

In the majority of models, the survival and HRQoL of individuals without lung cancer was conditioned on age/birth year and sex, with some models further adjusting estimates to reflect the characteristics of the population eligible for screening in terms of smoking status, exposure or history. The costs incurred by individuals without lung cancer are not included in any of the models (other than the costs of screening and any further investigations).

Overdiagnosed lung cancers (see Appendix 8) in models with a preclinical to clinical component appear to have the same outcomes as those of other true positives. Only one study explicitly states that constraints were placed on survival (e.g. the stage-specific survival of lung cancer did not differ between screen-detected and clinically detected tumours) to mitigate overdiagnosis (and other) bias(es). In models without a preclinical to clinical progression, overdiagnosis was handled in scenario analyses in which the survival benefit across the overall screened population was assumed to be smaller, or by assuming an adjustment to prevalence with impact on costs and survival of those with overdiagnosed tumours. None of these scenario analyses was informed by evidence on the proportion of overdiagnosed tumours or their outcomes (see Appendix 8).

The majority of studies modelled the impact on outcomes of false-positive results to screening as additional costs due to further unnecessary investigations (see Appendix 8). Only two models143,154,155 explicitly linked false-positive results to survival to reflect the disutility associated with subsequent diagnostic follow-up, and another model linked false-positive results to the associated mortality. 160

Two components of value not considered in the diagnostic studies, but modelled in the screening studies, relate to (1) early recalls and (2) radiation exposure (see Appendix 8). One study143 considered early-recall CT for a proportion of patients who screened positive, instead of proceeding directly to the diagnostic pathway. This was modelled as an additional cost and not linked to a delay to diagnosis. As mentioned previously, Yang et al. 166 applied a lifetime cost to reflect the impact of radiation exposure due to screening on patients who die from radiation-induced cancer. It is unclear to whom this impact applies and how radiation exposure differed across strategies.

Two sources of bias associated with early diagnosis, namely lead time and length time bias, were considered in some screening studies, but not in diagnostic studies. Screening models handled lead time bias in three ways (see Appendix 8):

1. constraining stage-specific survival of patients with screen-detected cancers, so it did not exceed that of patients with clinically detected cancers

2. reducing the survival benefit of patients with screen-detected cancers by an arbitrary amount of survival time (not supported by evidence)

3. applying a differences-in-differences methodology to age adjust the survival differences between screened patients and unscreened patients with lung cancer.

Length time bias was explicitly discussed in only one model. 154,155 It was handled in the same way as lead time bias (i.e. by constraining the stage-specific survival of patients with screen-detected cancers, so it did not exceed that of patients with clinically detected cancers). It is worth noting that, as length time bias arises from slow-growing tumours being more likely to be detected by screening (given the interval between screening appointments; see Appendix 8), length time bias also relates to overdiagnosis of indolent malignant tumours (an extreme case of slow growth).

A few studies (see Appendix 8) use UK-relevant data sources to inform survival and costs by cancer stage. No UK-specific HRQoL evidence was used to inform the outcomes of patients with lung cancer. UK-relevant life tables were used to estimate the survival of individuals without cancer. Survival and HRQoL adjustments to reflect the outcomes of smokers were also informed by UK-relevant data.

Conclusions of the additional reviews

The additional reviews highlight that cost-effectiveness evaluations conducted within the diagnostic pathway for solid pulmonary nodules are generally based on sparse evidence. Despite the lack of evidence, these studies rely on a common (assumed) value mechanism: that diagnostic technologies displacing CT surveillance may lead to diagnosis of lung cancer at an earlier stage. Screening cost-effectiveness studies also use such a value mechanism. The reviews identified a number of additional value components that could be of relevance for EarlyCDT Lung, for example the potential for increased detection (i.e. the potential for the introduction of EarlyCDT Lung leading to a higher number of lung cancers detected). Finally, these broader reviews have helped identify structural assumptions and parameter estimates that could be used in alternative to those implemented in the EarlyCDT Lung cost-effectiveness studies. Many important gaps, however, still remain. These will be further systematised and explored in the following section.

Core components of the decision problem

For most diagnostic technologies, such as EarlyCDT Lung, patient and health-system benefit arises from the information the test provides, which is used to tailor subsequent patient management decisions; value is therefore accrued indirectly.

In the context of this assessment, EarlyCDT Lung is being considered to be included in the diagnostic pathway for solid pulmonary nodules. The BTS pathway (see Chapter 1, Diagnostic pathway for pulmonary nodules), commonly used in the UK, grounds management decisions, which range from CT surveillance (less interventional) to excision (more interventional), on numerical assessments of malignancy risk. EarlyCDT Lung test results are being proposed to update these malignancy risk scores.

Clinical decisions are explicitly grounded in two risk thresholds: one determining referral to CT surveillance (< 10% risk) and another determining referral to excision (> 70%) (Figure 16). The guidelines are less prescriptive for the intermediate-risk group, recommending image-guided biopsy, but also allowing the use of CT surveillance and excisional biopsy. Clinical decisions for this risk group are determined on a case-by-case basis and depend on risk of malignancy, determined by the net trade-offs of further interventions for individual patients (including the patient’s fitness to undergo invasive diagnostic follow-up and subsequent treatment), patient preference and nodule characteristics (e.g. nodule location: peripheral nodules will be easier to access than central ones).

FIGURE 16.

The BTS-recommended actions according to malignancy risk.

To support conceptualisation of a future decision model, and to illustrate some of the considerations arising in subsequent sections, we will use influence diagrams,26,27 which provide a simplified representation of the decision problem. These diagrams use shapes to represent important aspects of the evaluation: rectangles represent deterministic events (such as decisions), ovals represent probabilistic events (events that are uncertain) and diamonds represent the outputs of interest. Arrows between shapes reflect dependencies, which matter only if they directly or indirectly affect outcomes.

The influence diagram in Figure 17 represents the core components of the decision problem for EarlyCDT Lung. In the diagram, disease status (Disease) is represented as a chance node, reflecting the probability of malignant (+) or benign (–) disease. The malignancy risk score (Risk) is probabilistic (represented by a distribution), and, because it is a continuous variable (between 0% and 100%), the shape is represented using a double line. The arrow from Disease to Risk indicates that the risk score is determined by malignancy status, that is the risk score distribution is expected to differ between benign and malignant nodules. Options for management decisions within the BTS pathway (Decision) are surveillance (surv), biopsy (biop) or treatment (treat), and these are determined by the risk score. The diagram represents treatments as deterministic decisions from risk scores. This means that, for a given risk score, a single decision is taken (in later sections this assumption is relaxed). The risk score here is shown as continuous, but the score could also be categorised (e.g. 0–10, 10–20, 20–50, 50–70 and 70–100, as done in Chapter 3, Impact on clinical decision-making) to simplify the representation of how risk scores determine management decision. These management options, alongside disease status, will impact on outcomes (O), an output of the model. These would include both the short-term impacts of the management decisions and the long-term effects of treating malignant nodules.

FIGURE 17.

Influence diagram: core components of the decision problem. biop, biopsy; Decision, management decision within BTS pathway; Disease, disease status; n, no; O, outcomes; Risk, malignancy risk score; surv, surveillance; Test, include new test in pathway; treat, treatment; y, yes.

A decision node is used to reflect the decision to include a new test (Test), such as EarlyCDT Lung, in the diagnostic pathway. The direct arrow from Test to Risk illustrates the case in which the test is not used. When the test is used, its results (Test results) update the quantitative risk score. The diagram reflects that the test is assumed to affect further management decisions only by changing the risk score. The test can itself have a direct impact on outcomes (represented by the arrow between Test and O), reflecting its costs and any adverse events.

This core conceptualisation diagram identifies important aspects of this evaluation, which will be looked at in further detail in the next sections. These include the following:

• population, particularly in what concerns value drivers such as prevalence of disease (see Population)

• subsequent clinical management decisions and how EarlyCDT Lung affects these (see Clinical decisions under current pathway and clinical impact of EarlyCDT Lung)

• how changes in subsequent clinical decisions affect outcomes (see Components of the clinical and economic value of EarlyCDT Lung arising from changes in management decisions).

Population

In this section, we summarise the evidence available on the characteristics of the populations and subpopulations of interest (described in full in Chapter 1, Population and relevant subgroups, and Place of the intervention in the care pathway, and listed in Table 25), and highlight important issues around subsequent actions determined by test results, which are fundamental in determining the clinical and economic value of EarlyCDT Lung.

Evidence on the population with pulmonary nodules is sparse, of unclear representativeness and is heterogeneous. 3 This includes evidence on characteristics that drive value for a new diagnostic test, such as prevalence of disease (as shown in Edelsberg et al. ;120 see Chapter 4, Results of the identified studies). This is reflected in existing cost-effectiveness studies, in which value drivers have been informed by either evidence of limited relevance (e.g. the use of Tanner et al. 122 to inform prevalence in Sutton et al. ,121 as critiqued in Chapter 4, Decision problem and relevance to National Institute for Health and Care Excellence Diagnostics Assessment Report scope), or unsubstantiated assumptions (such as stage distribution; see Chapter 4, Evidence linkage, and Chapter 5, Summary of the review of cost-effectiveness studies on diagnostics for lung cancer diagnosis). A single small UK study by Al-Ameri et al. 79 described the flow of patients through the BTS pathway (described in Chapter 3, Computerised tomography surveillance). This study suggests that more than half of patients with incidentally detected nodules present with small or low-risk nodules with a low prevalence of malignancy, and that approximately one-third present with intermediate-risk nodules and a higher prevalence of malignancy. A non-negligible proportion of cancers detected at metastatic disease stage were observed across both risk groups.

Clinical decisions under current pathway and clinical impact of EarlyCDT Lung

Clinical evidence on EarlyCDT Lung that would be required for an economic model is discussed in Chapter 3, Main gaps and limitations in the clinical evidence, and includes the following:

• robust diagnostic accuracy evidence on the population and subpopulations of interest

• validation of pre- and post-test risk scores

• evidence on clinical impact of EarlyCDT Lung in changing subsequent management decisions.

Core components of the decision problem identified that important impacts on patient outcomes from the use of EarlyCDT Lung arise from the changes in management that EarlyCDT Lung can lead to. The range of possible actions after risk assessment with EarlyCDT Lung are listed in Chapter 1, Action after risk assessment. The evidence reviewed in Chapter 3 and clinical advice indicated a number of further relevant considerations:

• Management decisions in the intermediate-risk group are heterogeneous, with the proportions referred to CT surveillance, biopsy or excision being largely unknown.

• Some nodules are difficult to biopsy, such as subcentimetre nodules and nodules centrally located in the lung. This restricts management options to either CT surveillance or excision.

• The value of EarlyCDT Lung in determining malignancy risk is unclear. The EAG analysis (see Chapter 3, Synthesis of diagnostic accuracy) found poor diagnostic accuracy of EarlyCDT Lung, and, consequently, based on EAG modelling (see Chapter 3, Further analyses of clinical effectiveness), a limited impact on risk of malignancy. For example, an individual with a pre-test risk of 10% would obtain a maximum post-test risk score of 22%; a post-test risk of 70% can be achieved only for individuals with a pre-test risk of > 48% (see Figure 9).

• The widespread availability of PET-CT means that all patients in the UK are expected to have access to this technology. When Brock risk is reclassified to > 10% after EarlyCDT Lung, patients are expected to undergo PET-CT to inform further management decisions.

Based on these considerations, the potential for changes in management in the proposed positionings of EarlyCDT Lung are as follows (further detail is presented in Table 25):

• EarlyCDT Lung is unlikely to change referrals to CT surveillance in a number of subgroups, including small and low-risk nodules that cannot be biopsied. EarlyCDT Lung is therefore unlikely to present clinical or economic value to these groups.

• Computerised tomography surveillance → biopsy: low- or intermediate-risk nodules that would have been referred to CT surveillance, but that can be biopsied. Note that the intermediate-risk nodules considered here are likely to show a lower pre-test risk score (close to 10%). At the range of 10–48% pre-test risk, EarlyCDT Lung cannot lead to post-test risk scores of > 70% (under the EAG’s analyses; see Table 17); therefore, it is unlikely that these nodules will see their management change from CT surveillance to excision.

• Computerised tomography surveillance → treatment: intermediate-risk nodules with a pre-test risk score of > 48% and that cannot be biopsied.

• Biopsy → treatment: intermediate-risk nodules with a pre-test risk score of > 48% and that would have been biopsied.

Evidence required and modelling considerations

It is important that further research allows a better understanding of how the Brock or Herder risk of malignancy scores are used to inform clinical management decisions. In recognising that there is variability in management decisions, particularly in the 10–70% range, future evidence should explore the relationship between risk of malignancy and the likelihood of referral to surveillance and excision. In addition, evidence discerning how factors such as patient preference and fitness to receive more invasive tests contribute to these decisions is currently unavailable. It is important to consider the potential impact of such variation in clinical practice in decision modelling to accurately predict outcomes and obtain unbiased results from the economic modelling. The influence diagram in Figure 18 modifies the diagram in Figure 17 to include a probabilistic relationship between risk score and management decisions: management options are no longer represented by a rectangular (deterministic) node, as in the previous diagram (see Figure 17), but by an oval chance node, reflecting that, for each value of the risk score, there is a probability of referral to surveillance, biopsy or excision.

FIGURE 18.

Influence diagram: expanded diagram to reflect expected variation in management decisions and to consider heterogeneity. biop, biopsy; Decision, management decision; Disease, disease status; Het, heterogeneity; n, no; O, outcomes; Risk, malignancy risk score; surv, surveillance; Test, include new test in pathway; treat, treatment; y, yes.

The EAG’s analysis in Chapter 3 shows that the extent to which EarlyCDT Lung leads to changes in management depends on the test’s accuracy. Further evidence emerging on the accuracy of EarlyCDT Lung should therefore be carefully considered in future modelling attempts, and interpreted in the context of the test’s ability to affect subsequent management choices. Direct evidence on how EarlyCDT Lung test results affect subsequent management decisions would also be important to support assumptions over its clinical utility.

In this section, we have listed important considerations on management decisions relating to the different subpopulations/positionings for the test, and future assessments should explicitly consider these. Note that none of the published cost-effectiveness studies on EarlyCDT Lung has considered these (see Chapter 4). Future modelling efforts should reflect subgroups with restricted versus unrestricted management options (e.g. D’Andrea et al. 126 restricted options for diagnostic follow-up with biopsy of central nodules to only those with a diagnostic bronchoscopy result), which will include people with nodules that can, or cannot, be biopsied, and people at higher/lower risk of serious adverse events from biopsy.

The previous section Population identifies important sources relating to patient and nodule characteristics that are linked to prevalence of disease. The reasons determining variation in management decisions considered here may also be related to prevalence of disease, particularly those related to nodule characteristics (e.g. small nodules). Therefore, in the influence diagram in Figure 18, heterogeneity (Het) is broadly considered. The diagram illustrates that sources of heterogeneity can determine prevalence of disease (arrow from Het to Disease) and subsequent management decisions (arrow from Het to Decision). It also represents the possibility of sources of heterogeneity affecting outcomes directly (arrow from Het to O), which is also important to consider in further decision modelling (e.g. histology of malignant tumours; see Long-term outcomes).

Components of the clinical and economic value of EarlyCDT Lung arising from changes in management decisions

In this section, we focus on the link between changes in subsequent management decisions arising from EarlyCDT Lung’s clinical utility in the diagnostic pathway for solid pulmonary nodules and outcomes. These highlight key trade-offs (components of value), arising indirectly via changes in management decisions, that are relevant to consider against the cost of introducing the test itself and any adverse events or anxiety introduced by the test (which affect all individuals tested, and have been previously detailed in Chapter 1, Cost of EarlyCDT Lung testing, and Chapter 3, The Early Detection of Cancer of the Lung Scotland trial) when determining the clinical and economic value of EarlyCDT Lung. These have been identified by bringing together the issues/limitations from the different reviews (see Chapters 3–5), and are as follows:

• The short-term impacts (costs and adverse events) of escalating the current pathway to more interventional diagnostic investigations/treatments for those who test positive for whom management is changed. These include (1) the costs and harms imposed by unnecessary invasive diagnostics or treatments on benign nodules (false positives) and indolent nodules (true positives that would not have shown significant growth on CT surveillance) and (2) the implications of radiation exposure from increased referral to PET-CT.

• Longer-term health benefits and cost implications of earlier detection (and treatment) of lung cancer for those with true-positive results for whom management is changed, and/or increased detection from the overall diagnostic strategy that includes the test (i.e. a higher proportion of true-positive results in relation to the current pathway).

These key components are further linked to the clinical utility of EarlyCDT Lung in Table 26, highlighting that the trade-offs arise as a consequences of changes in management: short-term impacts arise for both true- and false-positive patients who see management change (facing the risk of overtreatment of benign and indolent nodules and the potential for increased radiation exposure), and that the long-term effects will be realised only for the true-positive patients who see management escalated, leading to early or increased detection (and consequent treatment) of malignant lung cancer.

Evidence required and modelling considerations

Future cost-effectiveness models for EarlyCDT Lung should clearly justify the trade-offs quantified and include consideration for each of the value components in Table 26. The influence diagram in Figure 19, expanded from that in Figure 18, highlights the two key components of value from changes in management. First, the direct effects of these choices over outcomes (represented in Figure 19 by the direct arrow from Decision to O), which includes their costs and adverse events. Note that, for example, a surveillance strategy (current management for many of the subpopulations here considered) will include further diagnostic workup for malignant nodules showing quick growth. Therefore, the net impacts from escalation will arise only from higher rates of detection of clinically significant disease and from differences between the strategies in detecting indolent disease.

FIGURE 19.

Influence diagram: expanded diagram to reflect key components of value for EarlyCDT Lung arising from changes in management decisions. biop, biopsy; Decision, management decision; Disease, disease status; infty, infinity; n, no; O, outcomes; Risk, malignancy risk score; surv, surveillance; Test, include new test in pathway; treat, treatment; y, yes.

The second component of value reflects the longer-term health benefits and cost savings arising indirectly from earlier detection (and treatment) of malignant disease (represented by the addition of the event ‘Time to detection’, which links management decisions to outcomes).

The following subsections summarise existing evidence, and identify further evidence requirements and the evidence linkages necessary to support economic modelling on these two components: the following subsection focuses the on short-term impacts of escalating subsequent diagnostic/treatment decisions, and the subsection after that focuses on the longer-term impacts from increased/earlier detection of lung cancer (see Longer-term impacts from increased/earlier detection of lung cancer). The balance of each of the components of value will differ for each of the proposed placements for EarlyCDT Lung; this is discussed in the concluding subsection (see Conclusions).

Longer-term impacts from increased/earlier detection of lung cancer

By facilitating earlier treatment, the earlier detection of lung cancer provides an opportunity for improvements in overall survival. Earlier detection has been demonstrated to be linked to detection at earlier stages of disease, that is to stage shifts, in RCTs of screening (in relation to no screening, i.e. clinical detection). Some of these RCTs have also shown reductions in lung cancer mortality. The most recent (2020) study demonstrating stage shift and mortality benefit is NELSON. 173 Stage shift and mortality benefits have been further modelled to establish the long-term clinical and economic value of alternative screening strategies (see Chapter 5, Summary of the review of cost-effectiveness studies on screening for lung cancer).

To the EAG’s knowledge, there is no experimental evidence of early detection or stage shift from alternative diagnostic strategies for incidentally detected nodules (see Chapter 4, and Chapter 5, Summary of the review of cost-effectiveness studies on diagnostics for lung cancer diagnosis). However, this has been the key mechanism of value for EarlyCDT Lung in existing cost-effectiveness studies. It is therefore important that further research generates evidence to support this mechanism of value for diagnostic strategies in general, and EarlyCDT Lung in particular, ideally using an experimental design.

In the cost-effectiveness analyses reviewed, one of the models for EarlyCDT Lung and a couple of models on diagnostic strategies also assumed increased detection, attributing additional value to the diagnostic technologies evaluated. Increased detection relies on assuming that comparator strategies (e.g. CT surveillance) fail to detect a proportion of cancers (usually small), which would present clinically at a later point in time. The new strategy, introducing an additional test with relevant sensitivity (such as EarlyCDT Lung), would therefore have the potential to detect these lung cancers earlier. The value of increased detection also relies on the earlier detection mechanism. However, there is no empirical evidence supporting this value component for EarlyCDT Lung.

The sensitivity of the overall CT surveillance strategy determines the potential for increased detection with EarlyCDT Lung. Clinically, it is considered that there is no growth rate threshold beneath which, nor duration of radiological stability beyond which, malignancy is definitely excluded. 3 There is, however, uncertainty concerning the proportion of clinically significant cancers missed by the BTS CT surveillance schedule. Further evidence on the likelihood of a malignant cancer being missed by surveillance is therefore required to support such an assumption.

To establish the value of early detection (including increased detection) in the absence of empirical evidence directly on the magnitude of stage shift attained, evidence linkage is required. The following mechanism (illustrated schematically in Figure 20) has been used across the diagnostic and (most) screening studies reviewed in Chapters 4 and 5, encompassing the following:

• the identification of differences in the time to diagnosis between current and proposed identification strategies, and mapping of these differences against likelihood or time to preclinical stage progression, to define the level of stage shift

• the linking of the stage distributions, with and without stage shift, to expected long-term outcomes conditional on disease stage.

FIGURE 20.

Schematic representation of the evidence linkage required to establish value from early detection.

One possible representation of such a mechanism of value is shown in the influence diagram in Figure 21. In this diagram, the stage distribution at baseline (stage.b) (i.e. at the time of the first CT scan) is represented to include the absence of malignancy (using the value zero) alongside the categorisation in disease stages (S1 to Sx). Management decisions determine time to detection (which may assume values between zero, reflecting immediate detection, and infinity, reflecting no detection). This should be parameterised to consider that CT surveillance imposes a longer time to detection than biopsy or treatment; in this way, the shift from surveillance to biopsy/treatment that EarlyCDT Lung may facilitate will reduce time to detection. As explained previously, by exposing time to detection, both of the following can be explored: (1) the value of earlier detection from immediate diagnosis with biopsy or immediate treatment in relation to CT surveillance and (2) the value of increased detection of cases that would not have been diagnosed with current care, and would therefore have only presented clinically, much later. Time to detection determines the stage distribution at detection (stage.d), with consideration for the speed of preclinical progression.

FIGURE 21.

Influence diagram: expanded diagram to include the stage shift evidence linkage mechanism. biop, biopsy; Decision, management decision; Disease, disease status; infty, infinity; n, no; O, outcomes; Risk, malignancy risk score; surv, surveillance; Test, include new test in pathway; treat, treatment; y, yes.

It is worth noting that disease staging classifications use discrete categories. However, there also needs to be some consideration over whether or not earlier detection within the same disease stage can also be associated with long-term benefits. Stage classifications are mostly based on criteria such as size of nodule, location, lymph node involvement and metastatic spread. The latest tumour–node–metastasis (TNM) staging system emphasises the difference in prognosis between stages T1a and T1b that differ only in the size of the tumour. This implies a relationship between size of the tumour and health outcomes, which suggests that earlier detection within the same stage (i.e. not allowing nodule growth) may be associated with improvements in the outcomes of treatment. Only one of the screening cost-effectiveness studies reviewed explicitly modelled the relationship between growth and stage progression (see Chapter 5, Summary of the review of cost-effectiveness studies on screening for lung cancer), but neither this study nor others modelled within-stage benefits, despite the potential for such an effect having been discussed. 154,155 This is an area where further evidence is required. If evidence emerges supporting the benefits of within-stage detection, the association between nodule size and health outcomes from treatment could be used, alongside metastatic burden, location and histology.

The influence diagram in Figure 22 illustrates a possibility for tracking the size of the nodule at detection (using VDT and time to detection) in future decision models, to allow quantifying within-stage benefits. The tracking of the size of the nodule is represented in a way similar to the tracking of disease stage. The size of the nodule at detection is determined by the size of the nodule at baseline, VDT and time to detection. Given that size is one of the dimensions considered in most staging of disease classifications, in the influence diagram, size of the nodule is represented to determine stage of disease. To allow within-stage growth to affect outcomes, the diagram reflects that both stage and size determine outcomes.

FIGURE 22.

Influence diagram: expanded diagram to reflect within-stage benefits (via tracking of nodule size). biop, biopsy; Decision, management decision; Disease, disease status; infty, infinity; n, no; O, outcomes; Risk, malignancy risk score; surv, surveillance; Test, include new test in pathway; treat, treatment; y, yes.

Further details on evidence and modelling requirements in relation to this mechanism are discussed next.

Stage shift

Time to diagnosis

In the diagnostic pathway of detection of solid pulmonary nodules, the most significant source of delay is the possibility of referring patients to a surveillance strategy. Surveillance refers to a schedule of regular imaging screening aimed at measuring nodule growth to identify VDT, as significant growth (below a certain threshold of VDT) is commonly associated with malignancy. Given the need for time to establish VDT, surveillance imposes a delay to diagnosis. Delay is determined by the probability of surveillance detecting clinically significant cancer at the different scheduled screening points.

Chapter 3 highlights the absence of evidence on the overall sensitivity and timing of diagnoses with CT surveillance. The models identified in our review of diagnostic technologies/strategies (see Chapter 5, Summary of the review of cost-effectiveness studies on diagnostics for lung cancer diagnosis) either rely on unsubstantiated assumptions or on limited evidence of questionable relevance. Some of the models, including the EarlyCDT Lung cost-effectiveness studies (see Chapter 4 and Appendix 8 for further details), infer probability of detection from further modelling of tumour size and growth. The evidence underlying these VDT ‘submodels’ is limited, lacking relevance and robustness, and failing to characterise heterogeneity. Heterogeneity in VDT can be associated with patient and nodule characteristics, such as nodule size, probability of disease spread and histology, among others. 3 This would need to be considered explicitly to appropriately determine the probability of detection at different time points.

A key source of variation is histological subtype, which is likely to be related to size, VDT (progressively longer VDTs were identified for small cell carcinoma, squamous cell carcinoma, adenocarcinoma and bronchioalveolar carcinoma/adenocarcinoma in situ3) and outcomes. It is therefore important to reflect on whether or not the distribution of histologies detected may differ over the different timings of CT surveillance imaging. This has not been explored in previous models (see Chapters 4 and 5), but could be considered in future modelling.

Stage progression

Despite tumour size being one of the features defining disease stage, with tumour growth therefore inherently determining progression, there is no evidence that stage progression happens within the time frame of CT surveillance (see Chapter 3, Computerised tomography surveillance). Most of the models reviewed of diagnostic technologies relied on unsubstantiated assumptions to define the likelihood of progression during surveillance (e.g. Dietlein et al. 128 and Lejeune et al. ;131 see Chapter 5, Summary of the review of cost-effectiveness studies on diagnostics for lung cancer diagnosis, for details). Evidence on the likelihood of stage progression with CT surveillance is therefore required to support a future assessment of EarlyCDT Lung.

It is worth noting that the likelihood of stage progression should depend on the stage classification used [most diagnostic cost-effectiveness studies reviewed in Chapter 5, Summary of the review of cost-effectiveness studies on diagnostics for lung cancer diagnosis, use three-stage (local, regional, distant) or four-stage (I–IV) classifications]. The use of more granular categories, that is of a more disaggregated level of staging categories (e.g. T1a distinct from T1b, or stage IA1 distinct from stage IA2), could allow stage-shift-based evidence linkage approaches to capture additional benefits that are currently not captured, reducing the impact of ignoring potential within-stage benefits.

Given the absence of evidence on the likelihood of stage progression for incidentally detected nodules followed up by CT surveillance, wider evidence on the speed of preclinical stage progression is valuable. Screening RCTs provide a particularly robust foundation for evaluating the speed of preclinical stage progression. These studies typically compare clinical detection (no screening) with early detection from screening and observe the stage distributions at detection across the groups (which differ in time to detection). Further modelling uses these data to infer time to preclinical progression, based on the assumptions imposed in a natural history structural model and by calibrating such a model to a variety of data sources (e.g. Tomonaga et al. ,158 Ten Haaf et al. ,159 Toumazis et al. ,160 Griffin et al. ,154 Snowsill et al. 155 and Hofer et al. 143). [31]The reporting is not sufficiently detailed to allow comparison of estimates of mean time to progression across studies, or the influence of structural assumptions on these estimates (see Appendix 8); further research on this would be welcomed.

Heterogeneity in stage progression is clinically acknowledged, and has been considered in a few of the cost-effectiveness models reviewed (see Toumazis et al. 160 as an example). The influence diagram in Figure 23 exemplifies a number of different ways in which heterogeneity can affect aspects related to stage progression. The diagram already includes heterogeneity determining the prevalence of malignancy, restricting management options and determining outcomes, which have been previously discussed (see Population and Clinical decisions under current pathway and clinical impact of EarlyCDT Lung). Heterogeneity affecting disease progression is also exemplified in the diagram. For example, tumour histology may affect likelihood of progression (two of the screening models identified in Chapter 5, Summary of the review of cost-effectiveness studies on screening for lung cancer, have conditioned preclinical progression probabilities on tumour histology158–160). Another case is where heterogeneity, for example in histological subtype, is associated with the stage distribution at baseline. Slow-growing nodules will be more likely to be picked up incidentally (or by screening strategies), and faster-growing nodules are more likely to be identified at an advanced stage of disease, with potentially limited capacity to benefit from the level of earlier diagnosis expected from averting surveillance (this is associated with length time bias in screening studies). The final example presented reflects the possibility of heterogeneity, for example in nodule location, affecting the likelihood of, and time to, detection. Further evidence should consider characterising heterogeneity in stage progression, linked to size (or growth), and could also include histology or other patient and nodule characteristics. Future decision models should reflect such heterogeneity explicitly.

FIGURE 23.

Influence diagram: expanded diagram reflecting the potential impact of heterogeneity. biop, biopsy; Decision, management decision; Disease, disease status; Het, heterogeneity; infty, infinity; n, no; O, outcomes; Risk, malignancy risk score; surv, surveillance; Test, include new test in pathway; treat, treatment; y, yes.

Long-term outcomes

Outcomes of detected malignant nodules, by disease stage

As highlighted in previous sections, to link stage shift to health outcomes, the majority of the studies reviewed conditioned survival and HRQoL on disease stage at diagnosis (see Chapters 4 and 5 for further details). Other common health outcome determinants modelled in the screening and diagnostic studies include age, sex and tumour histology. It is important that future modelling carefully considers these and other possible determinants, identifying which may be more broadly relevant across the value mechanism proposed. For example, histology is related to health outcomes, but also to tumour growth and, via this, to the probability of the tumour being identified via CT surveillance. It is therefore important to reflect this source of heterogeneity across the entire evidence linkage mechanism.

Another consideration that arises when modelling survival outcomes of patients with malignant tumours is that of competing mortality risks, as these risks may limit the ability to benefit from early diagnosis of patients. Some diagnostic studies of patients with pulmonary nodules incidentally identified in the context of coronary disease investigations have modelled the impact on mortality of CAD. As noted previously (see Population), there is considerable heterogeneity in the population defined by the decision problem, and comorbidities (with impact on survival) are likely to vary by identification route. For example, patients identified via screening, usually aged ≥ 50 years and with high smoking exposure, may have cardiovascular and respiratory comorbidities with increased mortality risk, compared with the general population. For patients whose nodules were identified incidentally, the profile of competing mortality risks may also vary according to the reason for referral for the original CT imaging. Patients referred to CT imaging after a trauma event are likely to be younger and with fewer comorbidities than patients undergoing CT in cardiovascular diagnostic pathways. As the proportion of patients with nodules identified via the different routes is uncertain, as are the patient characteristics within each subgroup, it will be challenging to accurately reflect the competing risks in the overall population. However, the decision model should be flexible enough to incorporate competing risks by population subgroup, so the impact can be explored in subgroup and/or sensitivity analyses.

Across the majority of the cost-effectiveness diagnostic123,126–132 and screening studies,107,139,142,147,148,151,152,154,155,158–160,163,164,166 the evidence linkage between stage shift and survival was established directly via disease stage without being mediated by treatment. The evidence used to inform survival outcomes is typically sourced from observational data such as cancer registries (e.g. the SEER programme), which reported the survival outcomes of patients treated for lung cancer by disease stage at diagnosis (as well as age, sex and histology). The use of registry data is usually driven by the need to have sufficiently long follow-up to capture impacts on mortality. However, the evidence does not reflect survival outcomes of patient treated with more contemporaneous lung cancer treatment (primary tumour treatment and subsequent treatments), but rather with the treatments available when the data were collected. In recent years, surgical techniques have advanced and a range of new treatments for lung cancer have become available, including a number of mutation-specific targeted therapies and immunotherapies. 174 Therefore, the use of registry data that may not reflect the outcomes and costs of patients treated for lung cancer over more recent years introduces additional uncertainty concerning the magnitude of survival benefits linked to earlier diagnosis of lung cancer. It is worth noting that the effectiveness of newer lung cancer treatments will also have associated uncertainties, as the evidence will be less mature than for earlier treatments.

An alternative approach to establish the link between stage shift and survival would be to further condition survival on treatment. Modelling the effect of lung cancer treatment could better characterise the survival of patients with earlier lung cancer diagnosis, compared with current practice, but would present additional practical and evidential challenges. For the stage shift link to outcomes to also be mediated by treatment, additional evidence would be required:

• treatment allocation conditional on stage, histology and presence of treatment-relevant biomarkers (e.g. anaplastic lymphoma kinase or programmed death-ligand 1)

• distribution of treatment-relevant biomarkers in the population (and its correlation with histology)

• characterisation of subsequent treatment sequences and their health outcomes.

The biases introduced in the evidence linkage necessary to quantify the impact on outcomes of the early diagnosis component of value, namely lead time and length time bias, are highlighted in the screening studies (e.g. in Griffin et al. 154 and Snowsill et al. 155). The selection of an approach to handle lead time and length time bias (three alternatives have been identified in Chapter 5, Summary of the review of cost-effectiveness studies on screening for lung cancer) in a future assessment should take into account the adequacy of the method to the model structure (e.g. does the model explicitly track disease progression from preclinical to clinical?) and make use of good-quality evidence on survival gains with early detection of lung cancer.

Outcomes of undiagnosed lung cancer

In the majority of cost-effectiveness studies reviewed, the evidence linkage mechanism explicitly includes the diagnosis of lung cancer at clinical presentation if these are undetected by other means (incidental or through screening). Most models assume that lung cancer, while undiagnosed, has similar outcomes to the general population (i.e. its clinical significance is limited); examples that assume differential outcomes are Sutton et al. 121 and Hofer et al. 143 A future assessment should consider evidence on the clinical significance of undiagnosed lung cancer.

Outcomes of benign nodules

The long-term health outcomes of patients with benign nodules have been implicitly considered equivalent to those of the general population in previous diagnostics models, and individuals were assumed to not accrue costs beyond those determined by the diagnostic pathway (see Chapter 5, Summary of the review of cost-effectiveness studies on screening for lung cancer). No robust evidence was identified to support this assumption. The prevalence of malignancy may differ across positionings for EarlyCDT Lung (and other potential factors, such as route of presentation), and the rate of benign resection is expected to differ across strategies. Because of this, if there are differences in the longer-term outcomes of benign nodules (such as those resulting from long-term morbidity caused by benign resection), these should be explicitly considered in future modelling.

Conclusions

There is currently insufficient evidence to support an explicit quantification of the value of EarlyCDT Lung in the diagnostic pathway of solid nodules. Our reviews identified that, to justify the additional costs and health system implications of introducing EarlyCDT Lung in the BTS diagnostic pathway, the short-term trade-offs of escalating diagnostics/treatment (including overtreatment of indolent lesions and benign resection) should be considered against the long-term benefits that may arise from earlier identification of lung cancer. A number of important uncertainties arise, but, based on current evidence and clinical judgement, it can be established that EarlyCDT Lung is unlikely to present value for small nodules (between 5 mm and 8 mm), for low-risk nodules that are not eligible for biopsy, and for intermediate-risk nodules with a (pre-test) risk score of < 48% that would undertake biopsy in the current pathway, as EarlyCDT Lung has limited ability to change management decisions for these groups.

Whether or not EarlyCDT Lung presents clinical and economic value for the remaining subpopulations will be determined by explicit assessments of the following:

• For low-risk nodules eligible for biopsy.

  • The likelihood of EarlyCDT Lung changing management decisions (likely to be from surveillance to biopsy).

  • The prevalence of malignancy (expected to be < 6%) and the accuracy of EarlyCDT Lung followed by biopsy, which will determine the probability of detection. This is to be compared with the accuracy and timing of detection with CT surveillance (and subsequent investigations) to determine the potential for early detection.

  • The stage distribution of the nodules at the time of initial identification (noting that a proportion may already be at advanced stages) and the likelihood of disease progression under surveillance, which determines the potential benefits of early diagnosis with EarlyCDT Lung.

  • The prevalence of malignancy and specificity of EarlyCDT Lung, which will determine the likelihood and consequences of escalating management of patients with false-positive results to the test. Because pre-test risk is low and EarlyCDT Lung has limited ability to increase risk score, benign resection is unlikely.

• For intermediate-risk nodules that would be assigned to CT surveillance in the current BTS pathway (these are, therefore, likely to be at the lower end of the risk spectrum), but on which biopsy can be undertaken.

  • The likelihood of EarlyCDT Lung changing management decisions (likely to be from surveillance to biopsy). This is more likely than in the low-risk population because of the higher pre-test risk score.

  • The prevalence of malignancy is expected to be low (although higher than in the low-risk population); therefore, the net benefits of early detection may be low.

  • Given the low prevalence, the likelihood of increased intervention (biopsy) on benign nodules is of concern. As with the previous group, benign resection is unlikely.

• For intermediate-risk nodules presenting risk scores of > 48% that would be assigned to biopsy in the current BTS pathway.

  • Likelihood that EarlyCDT Lung changes management decisions (likely to be from biopsy to excision).

  • The potential for early detection may be limited, as surveillance is not on the current pathway for these nodules and time to treatment may not be not significantly changed. Potential benefits for patients with true-positive results may amount only to avoiding biopsy: its cost, mortality and morbidity.

  • The higher prevalence of malignancy in this group determines a lower likelihood of increased intervention of benign nodules; however, here intervention is likely to mean resection given that preoperative confirmation of malignancy is not obtained. Resection has important morbidity, mortality and cost implications.

• Intermediate-risk nodules presenting risk scores of > 48% that would be assigned to CT surveillance in the current BTS pathway for not being eligible for biopsy.

  • Higher likelihood that EarlyCDT Lung changes management decisions (likely to be from CT surveillance to excision).

  • Given the higher prevalence in this group, there is a higher potential for early detection and stage shift.

  • Owing to the higher prevalence of malignancy, a lower likelihood of increased intervention of benign nodules is expected in this group, but this is likely to be benign resection.

The potential for EarlyCDT Lung to lead to overtreatment of indolent lesions that would otherwise not be detected by surveillance is unclear, as the prevalence of solid slow-growing nodules is unknown, but likely to be very small. The potential for EarlyCDT Lung to lead to increased detection is also unclear. Clinically, the presence of malignancy is not ruled out after little or no growth being observed within a CT surveillance schedule; however, its probability is thought to be extremely low.

Statement of principal findings

Strengths and limitations of the assessment

This review performed comprehensive searches for EarlyCDT Lung studies. It is likely to have identified all evidence on EarlyCDT Lung currently published, including all studies reported only as conference abstracts. This appears to be the first attempt to synthesise all the evidence on EarlyCDT Lung, including the first meta-analysis for this technology. This review also appears to be the first to attempt to investigate the clinical impact of using EarlyCDT within the BTS diagnostic pathway, although this was limited to a simulation study rather than real data.

The key limitations of the review are a result of the lack of relevant data, the potential for bias in the data that have been published, and its uncertain generalisability to the diagnosis of pulmonary nodules. Consequently, there was little scope for thorough statistical analysis and meta-analysis, and considerable uncertainty as to the robustness of the results.

No direct evidence on the clinical impact of EarlyCDT Lung was identified, severely limiting the ability to investigate how useful or effective EarlyCDT might be in practice. This could be investigated only in a simulation study, which required strong assumptions of uncertain validity. These included strong assumptions on how diagnostic accuracy estimates will translate into post-test risk, and assuming that EarlyCDT Lung is entirely independent of other factors, including nodule size.

Uncertainties

Uncertainties remain, largely because of the limitations of the data. There appear to be no cohort studies in which EarlyCDT Lung is used explicitly within the BTS guidelines (i.e. EarlyCDT Lung being performed after identification of nodules, and in combination with Brock or Herder risk assessment). There are limited data on EarlyCDT Lung among people with pulmonary nodules (only five studies, with only two fully published). All studies are at risk of bias. Consequently, there are too few data from patients with pulmonary nodules to be confident of the diagnostic accuracy of EarlyCDT Lung.

The EAG identified no evidence on the clinical impact of using EarlyCDT Lung, including how patients might be reclassified in terms of risk, or changes of clinical management, and this could be assessed only by simulation. Therefore, the clinical impact of using EarlyCDT within the BTS diagnostic pathway is largely unknown. The EAG identified no relevant evidence on the cost-effectiveness of using EarlyCDT Lung. Therefore, the economic impact of using EarlyCDT within the BTS diagnostic pathway is also largely unknown

The EAG identified comparatively limited evidence on other parts of the BTS diagnostic pathway, including the diagnostic accuracy and clinical impact of Brock and Herder risk assessments, and the clinical impact of CT surveillance. This increases uncertainties as to how using EarlyCDT Lung in the BTS diagnostic pathway might affect patients and health systems.

There is no evidence to support the idea that potential early diagnosis with EarlyCDT Lung will result in stage shift and, importantly, in improved patient outcomes, compared with current practice, at any of the proposed positionings in the diagnostic pathway. This, combined with the limitations of the diagnostic accuracy data on this technology and its limited scope to change clinical decisions on patient management, makes the clinical and economic value of EarlyCDT Lung highly uncertain.

Implications for service provision

The EAG concludes that the current evidence on EarlyCDT Lung is insufficient to determine its value in the diagnosis of people with suspect solid pulmonary nodules. This is because of the limited size of the relevant evidence base, uncertainties as to whether or not current evidence generalises to the UK diagnostic pathway, and a lack of evidence on the diagnostic accuracy and clinical impact of using EarlyCDT Lung.

Based on the limited data available, it appears that EarlyCDT Lung has poor diagnostic accuracy when used in isolation to diagnose pulmonary nodules, with low sensitivity to detect malignancy. It is therefore unclear what it can add to existing diagnostic methods, such as Brock and Herder risk assessments and the use of CT surveillance.

Based on results from the EAG’s simulation study, EarlyCDT Lung may have little clinical benefit when diagnosing low-risk or smaller nodules, as it appears unlikely to appropriately change clinical management decisions. EarlyCDT Lung may possibly have clinical value when identifying malignancy in intermediate-risk nodules (10–70% risk after PET-CT and Herder risk assessment), by correctly identifying high-risk nodules that are malignant, and so might benefit from prompt excision. However, these results are from a simulation study, based on limited data, and requiring various strong assumptions. These conclusions are therefore only suggestions that would require further research.

The uncertainty over EarlyCDT Lung’s clinical utility means that its cost-effectiveness is also unclear. In addition, the main mechanism of value proposed in existing cost-effectiveness studies of EarlyCDT Lung is earlier detection, but there is no evidence that a meaningful stage shift can happen within the diagnostic pathway for pulmonary nodules. It is, however, clear that a future assessment of EarlyCDT Lung should explore its cost-effectiveness in each of the alternative positionings proposed and in additional subgroups of relevance (such as eligibility for biopsy). This is because the potential for EarlyCDT Lung to alter subsequent clinical management decisions will differ across these groups, and the balance of trade-offs that determine value will also differ (the extent of false-positive results will determine harm from use of further invasive investigations and the extent of true-positive results will determine the long-term benefits of early detection and treatment). It is also clear that there will need to be appropriate consideration for heterogeneity, as factors such as nodule size or histology not only determine the prevalence of disease, but also may restrict management decisions; be associated with the likelihood of, and time to, detection; and determine long-term outcomes. Any future cost-effectiveness analyses need to appropriately justify the value components included in quantifications, and conduct extensive sensitivity analyses to explore the impact of the assumptions that are likely to be required given the limitations in the current evidence base,

The EAG notes that these conclusions relate only to the use of EarlyCDT Lung within the BTS-recommended pathway for the diagnosis of pulmonary nodules. The EAG has not considered how EarlyCDT Lung might be used in other areas, such as lung cancer screening.

Suggested research needs

The EAG’s main concern is the general lack of data on EarlyCDT Lung among patients with pulmonary nodules, who will be assessed using BTS guidance. Two studies on EarlyCDT Lung have yet to be fully published. One study, set in China, which aims to recruit 1000 patients, is still ongoing,77 and the other study, the EarlyCDT Lung Cancer Screening study of US patients at high risk of developing lung cancer, has been completed, but not yet fully published. 8,49–51 It is unclear how many of the patients in these studies have taken the test in a pathway position relevant to where EarlyCDT Lung is most likely to be used in the NHS. It is also unclear whether the test has been used as currently recommended (i.e. to update a risk score) or as a simpler positive/negative result.

Large, independent, prospective cohort studies are therefore needed, in which patients with identified pulmonary nodules receive the EarlyCDT Lung test. Patients should be diagnosed and managed in line with the BTS diagnostic pathway, with sufficient follow-up to confirm malignancy by biopsy or surgery, or to confirm its absence with at least 2 years’ follow-up without nodule growth. This will permit the estimation of the diagnostic accuracy of EarlyCDT Lung:

• in isolation

• in combination with Brock risk

• in combination with PET-CT and Herder risk (among patients undergoing PET-CT).

These cohort studies should also assess the clinical impact of EarlyCDT Lung by reporting outcomes including the following:

• impact on risk classification [e.g. moving from low risk (< 10%) to intermediate risk (10–70%), or from intermediate risk to high risk (> 70%)]

• change in clinical management (e.g. moving from CT surveillance to biopsy, or biopsy to immediate excision)

• timing and tumour stage at detection and treatment of malignant nodules

• avoidance of unnecessary CT or PET-CT imaging

• promotion of unnecessary PET-CT imaging, biopsies or surgical excisions (as a consequence of false-positive EarlyCDT Lung test results) and their consequent risk of adverse events.

The EAG has concerns that the proposed risk model for EarlyCDT Lung (see Figure 1) may be based on biased estimates of diagnostic accuracy (see Chapter 3, Case–control studies of EarlyCDT Lung among patients without confirmed pulmonary nodules). This risk model requires appropriate validation in independent cohorts. The cohort studies described previously could be used for this. If the model is found not to be valid (i.e. its estimated risks do not match observed risks), a new model will be required, based on robust diagnostic accuracy data from new cohorts. Further cohort studies would then be required to validate the new model.

Diagnostic accuracy studies do not tell us whether or not differences in accuracy result in clinically important effects on patient health outcomes. These effects may occur as a result of changes to further therapeutic or diagnostic interventions, based on test results. Therefore, the optimal approach to determining the clinical value of EarlyCDT Lung would be to conduct a RCT, in which patients with identified pulmonary nodules are randomised either to standard BTS management or to BTS management with EarlyCDT Lung included (as in Figure 2). A trial may be beneficial if cohort studies suggest potential, but inconclusive, benefits of EarlyCDT Lung. A randomised trial may not be required if evidence from high-quality cohort studies is sufficient to support the use of EarlyCDT Lung.

Currently, the broader evidence base on the whole BTS diagnostic pathway is insufficient to allow explicit and formal quantifications of the clinical end economic value of EarlyCDT Lung (or any other future test in this area). Although the EAG has not conducted a full systematic review of the entire pathway, our limited review, and the reviews conducted to support the BTS guidance, both suggest that large well-designed and UK-based prospective cohort studies are needed to investigate the following:

• the diagnostic accuracy and clinical impact of using the Brock risk model, in the context of UK clinical practice

• the diagnostic accuracy and clinical impact of using PET-CT and the Herder risk model, in the context of UK clinical practice

• the clinical consequences of CT surveillance (e.g. numbers of cancers identified and missed, delay in diagnosis and the possibility of tumour progression)

• how patient and nodule characteristics determine malignancy prevalence; eligibility for alternative clinical management options; likelihood of, and time to, detection under CT surveillance; and patient outcomes.

To further support the evidence linkage approaches likely to be required to support a future cost-effectiveness study (in the absence of an outcomes study), a number of additional studies could be important. These could include a comparative analysis of the preclinical progression models developed from screening studies, which would provide a broader understanding of the speed of preclinical progression of lung cancer. In addition, evidence allowing a better understanding of current variation in management decisions in the intermediate-risk group and of its determinants (malignancy risk, patient preference and fitness to undergo invasive procedures) would also be valuable to allow appropriate reflection of this variation in a future decision model. Finally, evidence on the current extent of benign resection would be important.

Contributions of authors

Ana Duarte (https://orcid.org/0000-0002-0528-4773) performed cost-effectiveness reviews and wrote the economic sections of the report.

Mark Corbett (https://orcid.org/0000-0002-5937-1493) performed the systematic review and wrote the associated sections of the report.

Hollie Melton (https://orcid.org/0000-0003-3837-510X) performed the systematic review and contributed to the associated sections of the report.

Melissa Harden (https://orcid.org/0000-0003-2338-6869) performed all database searches.

Stephen Palmer (https://orcid.org/0000-0002-7268-2560) provided economic advice and commentary for economic sections of the report.

Marta Soares (https://orcid.org/0000-0003-1579-8513) led the cost-effectiveness components of the project and wrote the economic sections of the report.

Mark Simmonds (https://orcid.org/0000-0002-1999-8515) led the clinical components of the project, performed the statistical analyses and wrote the associated sections of the report.

Data-sharing statement

No primary data were used in this report and diagnostic accuracy data are presented in Appendix 3, Tables 29–31. Further data can be obtained from the corresponding author.

Disclaimers

This report presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care.

MEDLINE ALL via Ovid

Includes Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Ovid MEDLINE Daily and Ovid MEDLINE.

Date range searched: 1946 to 5 March 2021.

Date searched: 8 March 2021.

Records retrieved: 1323.

1. EarlyCDT.af. (18)

2. Early CDT.af. (2)

3. Early-CDT.af. (2)

4. Early cancer detection test.af. (6)

5. or/1-4 (24)

6. ECLS trial$.af. (4)

7. 5 or 6 (26)

8. Oncimmune.af. (31)

9. 7 or 8 (48)

10. Autoantibodies/ (68,737)

11. (autoantibod$ or auto-antibod$ or AABT or AAb or AAbs or TAAb or TAAbs).ti,ab. (57,395)

12. 10 or 11 (94,716)

13. exp Lung Neoplasms/ (239,412)

14. Solitary Pulmonary Nodule/ (4170)

15. ((lung$ or pulmonary or bronchial or bronchogenic) adj3 (neoplas$ or carcinoma$ or cancer$ or nodule$ or tumor$ or tumour$ or malign$ or adenocarcinoma$ or blastoma$)).ti,ab. (251,893)

16. NSCLC.ti,ab. (45,747)

17. SCLC.ti,ab. (8139)

18. ((lung$ or pulmonary) adj2 (lesion$ or mass or masses)).ti,ab. (16,824)

19. ((noncalcified or non calcified) adj2 (nodule$ or lesion$ or mass or masses)).ti,ab. (454)

20. NCPN.ti,ab. (3)

21. ((ground-glass or solid or part-solid or subsolid or sub-solid) adj2 (nodule$ or lesion$ or mass or masses)).ti,ab. (9645)

22. ground glass opacit$.ti,ab. (4208)

23. (GGN or GGNs or GGO or GGOs).ti,ab. (1546)

24. ((benign or malignant or indeterminate) adj2 nodule$).ti,ab. (5427)

25. coin lesion$.ti,ab. (484)

26. (IPN or IPNs).ti,ab. (1722)

27. or/13-26 (356,910)

28. 12 and 27 (1318)

29. 9 or 28 (1345)

30. exp animals/not humans.sh. (4,796,559)

31. 29 not 30 (1323)

Key:

• / = subject heading [medical subject heading (MeSH)]

• sh = subject heading (MeSH)

• exp = exploded subject heading (MeSH)

• $ = truncation

• af = search of all fields

• ti,ab = terms in title or abstract fields

• adj3 = terms within three words of each other (any order).

EMBASE via Ovid

Date range searched: 1974 to 5 March 2021.

Date searched: 8 March 2021.

Records retrieved: 1973.

1. EarlyCDT.af. (56)

2. Early CDT.af. (17)

3. Early-CDT.af. (17)

4. Early cancer detection test.af. (14)

5. or/1-4 (71)

6. ECLS trial$.af. (9)

7. Oncimmune.af. (81)

8. 5 or 6 or 7 (121)

9. autoantibody/ (73,725)

10. (autoantibod$ or auto-antibod$ or AABT or AAb or AAbs or TAAb or TAAbs).ti,ab. (83,322)

11. 9 or 10 (104,669)

12. exp lung tumor/ (391,077)

13. lung nodule/ (22,179)

14. lung coin lesion/ (560)

15. ((lung$ or pulmonary or bronchial or bronchogenic) adj3 (neoplas$ or carcinoma$ or cancer$ or nodule$ or tumor$ or tumour$ or malign$ or adenocarcinoma$ or blastoma$)).ti,ab. (364,778)

16. NSCLC.ti,ab. (86,396)

17. SCLC.ti,ab. (13,679)

18. ((lung$ or pulmonary) adj2 (lesion$ or mass or masses)).ti,ab. (24,742)

19. ((noncalcified or non calcified) adj2 (nodule$ or lesion$ or mass or masses)).ti,ab. (715)

20. NCPN.ti,ab. (4)

21. ((ground-glass or solid or part-solid or subsolid or sub-solid) adj2 (nodule$ or lesion$ or mass or masses)).ti,ab. (15,482)

22. ground glass opacit$.ti,ab. (7717)

23. (GGN or GGNs or GGO or GGOs).ti,ab. (2499)

24. ((benign or malignant or indeterminate) adj2 nodule$).ti,ab. (8324)

25. coin lesion$.ti,ab. (460)

26. (IPN or IPNs).ti,ab. (2268)

27. or/12-24 (540,557)

28. 11 and 27 (1968)

29. 8 or 28 (2021)

30. (animal/or animal experiment/or animal model/or animal tissue/or nonhuman/) not exp human/ (6,204,998)

31. 29 not 30 (1973)

Key:

• / = subject heading (Emtree heading)

• sh = subject heading (Emtree heading)

• exp = exploded subject heading (Emtree heading)

• $ = truncation

• af = search of all fields

• ti,ab = terms in title or abstract fields

• adj3 = terms within three words of each other (any order).

Cochrane Central Register of Controlled Trials via Wiley Online Library

Date range searched: issue 3 of 12, March 2021.

Date searched: 8 March 2021.

Records retrieved: 29.

The following strategy was used to search both CENTRAL and CDSR.

#1 EarlyCDT:ti,ab,kw (5)

#2 “Early CDT”:ti,ab,kw (13)

#3 Early-CDT:ti,ab,kw (13)

#4 “Early Cancer Detection Test”:ti,ab,kw (4)

#5 (OR #1-#4) (14)

#6 ECLS next trial*:ti,ab,kw (3)

#7 Oncimmune:ti,ab,kw (0)

#8 #5 or #6 or #7 (14)

#9 MeSH descriptor: [Autoantibodies] this term only (686)

#10 (autoantibod* or auto next antibod* or AABT or AAb or AAbs or TAAb or TAAbs):ti,ab,kw (1932)

#11 #9 or #10 (1932)

#12 MeSH descriptor: [Lung Neoplasms] explode all trees (7828)

#13 MeSH descriptor: [Solitary Pulmonary Nodule] this term only (81)

#14 ((lung* or pulmonary or bronchial or bronchogenic) near/3 (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*)):ti,ab,kw (23,544)

#15 NSCLC:ti,ab,kw (9385)

#16 SCLC:ti,ab,kw (1351)

#17 ((lung* or pulmonary) near/2 (lesion* or mass or masses)):ti,ab,kw (600)

#18 ((noncalcified or non calcified) near/2 (nodule* or lesion* or mass or masses)):ti,ab,kw (1478)

#19 NCPN:ti,ab,kw (0)

#20 ((ground next glass or solid or part next solid or subsolid or sub next solid) near/2 (nodule* or lesion* or mass or masses)):ti,ab,kw (469)

#21 ground next glass next opacit*:ti,ab,kw (120)

#22 (GGN or GGNs or GGO or GGOs):ti,ab,kw (67)

#23 ((benign or malignant or indeterminate) near/2 nodule*):ti,ab,kw (226)

#24 coin next lesion*:ti,ab,kw (3)

#25 (IPN or IPNs):ti,ab,kw (38)

#26 (OR #12-#25) (26,367)

#27 #11 AND #26 (24)

#28 #8 or #27 (29)

Key:

• MeSH descriptor = subject heading (MeSH)

• * = truncation

• ti,ab,kw = terms in title, abstract or keyword fields

• near/3 = terms within three words of each other (any order)

• next = terms are next to each other.

Science Citation Index via Web of Science

Date range searched: 1900 to 5 March 2021.

Date searched: 8 March 2021.

Records retrieved: 1536.

#22 #21 OR #7 (1536)

#21 #20 AND #8 (1513)

#20 #19 OR #18 OR #17 OR #16 OR #15 OR #14 OR #13 OR #12 OR #11 OR #10 OR #9 (387,884)

#19 TS=(IPN or IPNs) (5619)

#18 TS=(“coin lesion” or “coin lesions”) (246)

#17 TS=((benign or malignant or indeterminate) NEAR/2 nodule*) (5559)

#16 TS=(GGN or GGNs or GGO or GGOs) (1516)

#15 TS=(“ground glass opacity” or “ground glass opacities”) (3427)

#14 TS=((ground-glass or solid or part-solid or subsolid or sub-solid) NEAR/2 (nodule* or lesion* or mass or masses)) (12,363)

#13 TS=NCPN (7)

#12 TS=((noncalcified or non-calcified) NEAR/2 (nodule* or lesion* or mass or masses)) (439)

#11 TS=((lung* or pulmonary) NEAR/2 (lesion* or mass or masses)) (16,328)

#10 TS=(NSCLC or SCLC) (68,356)

#9 TS=((lung* or pulmonary or bronchial or bronchogenic) NEAR/3 (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*)) (344,322)

#8 TS=(autoantibod* or auto-antibod* or AABT or AAb or AAbs or TAAb or TAAbs) (77,236)

#7 #6 OR #5 (45)

#6 TS=Oncimmune (2)

#5 #4 OR #3 (45)

#4 TS=(“ECLS trial” or “ECLS trials”) (5)

#3 #2 OR #1 (41)

#2 TS=(“Early cancer detection test”) (8)

#1 TS=(EarlyCDT or “Early CDT” or Early-CDT) (35)

Key:

• TS = topic tag; searches in title, abstract, author keywords and keywords plus fields

• * = truncation

• NEAR/3 = terms within three words of each other (any order).

EconLit via Ovid

Date range searched: 1886 to 18 February 2021.

Date searched: 8 March 2021.

Records retrieved: 3.

1. EarlyCDT.af. (0)

2. Early CDT.af. (0)

3. Early-CDT.af. (0)

4. Early cancer detection test.af. (0)

5. 1 or 2 or 3 or 4 (0)

6. ECLS trial$.af. (0)

7. Oncimmune.af. (0)

8. 5 or 6 or 7 (0)

9. (autoantibod$ or auto-antibod$ or AABT or AAb or AAbs or TAAb or TAAbs).mp. (3)

Key:

• $ = truncation

• af = search of all fields

• mp = terms in title, abstract, keywords, subject heading fields.

Cochrane Database of Systematic Reviews via Wiley Online Library

Date range searched: issue 3 of 12, March 2021.

Date searched: 8 March 2021.

Records retrieved: 0.

See above under CENTRAL for search strategy used.

Database of Abstracts of Reviews of Effects via Centre for Reviews and Dissemination

Date range searched: inception to 31 March 2015.

Date searched: 8 March 2021.

Records retrieved: 1.

The following strategy was used to search all CRD databases: DARE, HTA Database and NHS EED.

1. (EarlyCDT) OR (Early-CDT) OR (“Early CDT”) (1)

2. (“Early Cancer Detection Test”) (0)

3. (“ECLS trial”) OR (“ECLS trials”) (0)

4. (Oncimmune) (1)

5. #1 OR #2 OR #3 OR #4 (1)

6. (MeSH DESCRIPTOR Autoantibodies) (46)

7. (autoantibod* or auto-antibod* or AABT or AAb or AAbs or TAAb or TAAbs) (67)

8. #6 OR #7 (67)

9. (MeSH DESCRIPTOR Lung Neoplasms EXPLODE ALL TREES) (1151)

10. (MeSH DESCRIPTOR Solitary Pulmonary Nodule) (27)

• ((lung* or pulmonary or bronchial or bronchogenic) adj3 (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*)) (1449)

• ((neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*) adj3 (lung* or pulmonary or bronchial or bronchogenic)) (891)

• (NSCLC or SCLC) (284)

• ((lung* or pulmonary) adj2 (lesion* or mass or masses)) OR ((lesion* or mass or masses) adj2 (lung* or pulmonary)) (64)

1. ((noncalcified or non-calcified) adj2 (nodule* or lesion* or mass or masses)) OR ((nodule* or lesion* or mass or masses) adj2 (noncalcified or non-calcified)) (6)

2. (NCPN) (0)

3. ((ground-glass or solid or part-solid or subsolid or sub-solid) adj2 (nodule* or lesion* or mass or masses)) (24)

4. ((nodule* or lesion* or mass or masses) adj2 (ground-glass or solid or part-solid or subsolid or sub-solid)) (1)

5. (“ground glass opacity” OR “ground glass opacities”) (2)

6. (GGN or GGNs or GGO or GGOs) (0)

7. ((benign or malignant or indeterminate) adj2 nodule*) OR (nodule* adj2 (benign or malignant or indeterminate)) (53)

8. (“coin lesion”) OR (“coin lesions”) (1)

9. (IPN or IPNs) (0)

10. #9 OR #10 OR #11 OR #12 OR #13 OR #14 OR #15 OR #16 OR #17 OR #18 OR #19 OR #20 OR #21 OR #22 OR #23 (1565)

11. #8 AND #24 (1)

12. #5 OR #25 (2)

Health Technology Assessment database via Centre for Reviews and Dissemination

Date range searched: inception to 31 March 2018.

Date searched: 8 March 2021.

Records retrieved: 1.

See above under DARE for search strategy used.

NHS Economic Evaluation Database via Centre for Reviews and Dissemination

Date range searched: inception to 31 March 2015.

Date searched: 8 March 2021.

Records retrieved: 0.

See above under DARE for search strategy used.

International Health Technology Assessment database

Date range searched: inception to 9 March 2021.

Date searched: 10 March 2021.

Records retrieved: 16.

((((IPN or IPNs)[Title] OR (IPN or IPNs)[abs] OR (IPN or IPNs)[Keywords]) OR ((“coin lesions”)[Title] OR (“coin lesions”)[abs] OR (“coin lesions”)[Keywords]) OR ((“coin lesion”)[Title] OR (“coin lesion”)[abs] OR (“coin lesion”)[Keywords]) OR (((nodule*)[Title] OR (nodule*)[abs] OR (nodule*)[Keywords]) AND ((benign or malignant or indeterminate)[Title] OR (benign or malignant or indeterminate)[abs] OR (benign or malignant or indeterminate)[Keywords])) OR ((GGN or GGNs or GGO or GGOs)[Title] OR (GGN or GGNs or GGO or GGOs)[abs] OR (GGN or GGNs or GGO or GGOs)[Keywords]) OR ((“ground glass opacities”)[Title] OR (“ground glass opacities”)[abs] OR (“ground glass opacities”)[Keywords]) OR ((“ground glass opacity”)[Title] OR (“ground glass opacity”)[abs] OR (“ground glass opacity”)[Keywords]) OR (((“ground-glass” or “ground glass” or solid or “part-solid” or “part solid” or subsolid or “sub-solid” or “sub solid”)[Title] OR (“ground-glass” or “ground glass” or solid or “part-solid” or “part solid” or subsolid or “sub-solid” or “sub solid”)[abs] OR (“ground-glass” or “ground glass” or solid or “part-solid” or “part solid” or subsolid or “sub-solid” or “sub solid”)[Keywords]) AND ((nodule* or lesion* or mass or masses)[Title] OR (nodule* or lesion* or mass or masses)[abs] OR (nodule* or lesion* or mass or masses)[Keywords])) OR ((NCPN)[Title] OR (NCPN)[abs] OR (NCPN)[Keywords]) OR (((nodule* or lesion* or mass or masses)[Title] OR (nodule* or lesion* or mass or masses)[abs] OR (nodule* or lesion* or mass or masses)[Keywords]) AND ((noncalcified or “non calcified” or “non-calcified”)[Title] OR (noncalcified or “non calcified” or “non-calcified”)[abs] OR (noncalcified or “non calcified” or “non-calcified”)[Keywords])) OR (((lesion* or mass or masses)[Title] OR (lesion* or mass or masses)[abs] OR (lesion* or mass or masses)[Keywords]) AND ((lung* or pulmonary)[Title] OR (lung* or pulmonary)[abs] OR (lung* or pulmonary)[Keywords])) OR ((NSCLC or SCLC)[Title] OR (NSCLC or SCLC)[abs] OR (NSCLC or SCLC)[Keywords]) OR (((neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*)[Title] OR (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*)[abs] OR (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*)[Keywords]) AND ((lung* or pulmonary or bronchial or bronchogenic)[Title] OR (lung* or pulmonary or bronchial or bronchogenic)[abs] OR (lung* or pulmonary or bronchial or bronchogenic)[Keywords])) OR (“Solitary Pulmonary Nodule”[mh]) OR (“Lung Neoplasms”[mhe])) AND (((autoantibod* or auto-antibod* or AABT or AAb or AAbs or TAAb or TAAbs)[Title] OR (autoantibod* or auto-antibod* or AABT or AAb or AAbs or TAAb or TAAbs)[abs] OR (autoantibod* or auto-antibod* or AABT or AAb or AAbs or TAAb or TAAbs)[Keywords]) OR (“Autoantibodies”[mh]))) OR (((Oncimmune)[Title] OR (Oncimmune)[abs] OR (Oncimmune)[Keywords]) OR ((“ECLS trials”)[Title] OR (“ECLS trials”)[abs] OR (“ECLS trials”)[Keywords]) OR ((“ECLS trial”)[Title] OR (“ECLS trial”)[abs] OR (“ECLS trial”)[Keywords]) OR (((“Early Cancer Detection Test”)[Title] OR (“Early Cancer Detection Test”)[abs] OR (“Early Cancer Detection Test”)[Keywords]) AND ((lung* or pulmonary)[Title] OR (lung* or pulmonary)[abs] OR (lung* or pulmonary)[Keywords])) OR ((“Early-CDT”)[Title] OR (“Early-CDT”)[abs] OR (“Early-CDT”)[Keywords]) OR ((“Early CDT”)[Title] OR (“Early CDT”)[abs] OR (“Early CDT”)[Keywords]) OR ((EarlyCDT)[Title] OR (EarlyCDT)[abs] OR (EarlyCDT)[Keywords]))

Key:

• [Keywords] = search of keywords field

• [abs] = search of abstract field

• [Title] = search of title field

• [mh] = subject heading search

• [mhe] = exploded subject heading search

• * = truncation.

ClinicalTrials.gov

Date searched: 9 March 2021.

Records retrieved: 27.

Advanced search used.

1. 4 Studies found for: EarlyCDT OR Early-CDT OR “Early CDT”

2. 3 Studies found for: “Early Cancer Detection Test”

3. 11 Studies found for: (autoantibody OR auto-antibody OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) | lung cancer

4. 3 Studies found for: (autoantibody OR auto-antibody OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) | pulmonary nodule

5. 3 Studies found for: (autoantibody OR auto-antibody OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) | NSCLC OR SCLC

6. 1 Study found for: (autoantibody OR auto-antibody OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) | coin lesion

7. 1 Study found for: (autoantibody OR auto-antibody OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) | (indeterminate nodule OR IPN OR IPNs)

8. No Studies found for: (autoantibody OR auto-antibody OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) | (ground glass OR GGN OR GGNs OR GGO OR GGOs)

9. No Studies found for: (autoantibody OR auto-antibody OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) | NCPN OR noncalcified OR non-calcified)

10. 1 Study found for: Oncimmune

Key: | = combine with AND.

EU Clinical Trials Register

Date searched: 9 March 2021.

Records retrieved: 8.

1. EarlyCDT OR Early-CDT OR “Early CDT” (0)

2. “Early Cancer Detection Test” (0)

3. “ECLS trial OR “ECLS trials” (0)

4. Oncimmune (0)

5. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (Lung OR pulmonary) AND (neoplasm OR carcinoma OR cancer OR nodule OR tumor OR tumour) (3)

6. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (NSCLC OR SCLC) (2)

7. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (Lung OR pulmonary) AND (lesion OR lesions OR mass OR masses) (3)

8. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (noncalcified or “non calcified” OR non-calcified) AND (nodule OR lesion OR lesions OR mass OR masses) (0)

9. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND NCPN (0)

10. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (ground-glass or “ground glass” OR solid OR part-solid OR “part solid” OR subsolid OR sub-solid OR “sub solid”) (0)

11. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (GGN OR GGNs OR GGO OR GGOs) (0)

12. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (benign OR malignant OR indeterminate) AND (nodule OR nodules) (0)

13. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (benign OR malignant OR indeterminate) AND (“coin lesion” OR “coin lesions”) (0)

14. (autoantibody OR auto-antibody OR autoantibodies OR auto-antibodies) AND (IPN OR IPNs) (0)

Conference Proceedings Citation Index – Science via Web of Science

Date range searched: 1990–5 March 2021.

Date searched: 8 March 2021.

Records retrieved: 75.

# 22 #21 OR #7 (75)

# 21 #20 AND #8 (69)

# 20 #19 OR #18 OR #17 OR #16 OR #15 OR #14 OR #13 OR #12 OR #11 OR #10 OR #9 (51,033)

# 19 TS=(IPN or IPNs) (832)

# 18 TS=(“coin lesion” or “coin lesions”) (12)

# 17 TS=((benign or malignant or indeterminate) NEAR/2 nodule*) (657)

# 16 TS=(GGN or GGNs or GGO or GGOs) (182)

# 15 TS=(“ground glass opacity” or “ground glass opacities”) (208)

# 14 TS=((ground-glass or solid or part-solid or subsolid or sub-solid) NEAR/2 (nodule* or lesion* or mass or masses)) (1403)

# 13 TS=NCPN (1)

# 12 TS=((noncalcified or non-calcified) NEAR/2 (nodule* or lesion* or mass or masses)) (39)

# 11 TS=((lung* or pulmonary) NEAR/2 (lesion* or mass or masses)) (1461)

# 10 TS=(NSCLC or SCLC) (10,680)

# 9 TS=((lung* or pulmonary or bronchial or bronchogenic) NEAR/3 (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*)) (43,692)

# 8 TS=(autoantibod* or auto-antibod* or AABT or AAb or AAbs or TAAb or TAAbs) (6780)

# 7 #6 OR #5 (7)

# 6 TS=Oncimmune (0)

# 5 #4 OR #3 (7)

# 4 TS=(“ECLS trial” or “ECLS trials”) (0)

# 3 #2 OR #1 (7)

# 2 TS=(“Early cancer detection test”) (3)

# 1 TS=(EarlyCDT or “Early CDT” or Early-CDT) (4)

Key:

• TS = topic tag; searches in title, abstract, author keywords and keywords plus fields

• * = truncation

• NEAR/3 = terms within three words of each other (any order).

Proquest Dissertations & Theses A&I

Date searched: 9 March 2021.

Records retrieved: 28.

(TI,AB,SU,IF(EarlyCDT OR “Early CDT” OR Early-CDT) OR TI,AB,SU,IF(“Early Cancer Detection Test”) OR TI,AB,SU,IF(“ECLS trial” OR “ECLS trials”) OR TI,AB,SU,IF(Oncimmune)) OR (TI,AB,SU,IF(autoantibod* OR auto-antibod* OR AABT OR AAb OR AAbs OR TAAb OR TAAbs) AND (TI,AB,SU,IF((lung* OR pulmonary OR bronchial OR bronchogenic) NEAR/3 (neoplas* OR carcinoma* OR cancer* OR nodule* OR tumor* OR tumour* OR malign* OR adenocarcinoma* OR blastoma*)) OR TI,AB,SU,IF(NSCLC OR SCLC) OR TI,AB,SU,IF((lung* OR pulmonary) NEAR/2 (lesion* OR mass OR masses)) OR TI,AB,SU,IF((noncalcified OR non-calcified) NEAR/2 (nodule* OR lesion* OR mass OR masses)) OR TI,AB,SU,IF(NCPN) OR TI,AB,SU,IF((ground-glass OR solid OR part-solid OR subsolid OR sub-solid) NEAR/2 (nodule* OR lesion* OR mass OR masses)) OR TI,AB,SU,IF(“ground glass opacity” OR “ground glass opacities”) OR TI,AB,SU,IF(GGN OR GGNs OR GGO OR GGOs) OR TI,AB,SU,IF((benign OR malignant OR indeterminate) NEAR/2 nodule*) OR TI,AB,SU,IF(“coin lesion” OR “coin lesions”) OR TI,AB,SU,IF(IPN OR IPNs)))

Key:

• TI,AB,SU,IF = search of title, abstract, subject headings, keyword fields

• * = truncation.

Open Access Theses and Dissertations

Date searched: 9 March 2021.

Records retrieved: 52.

1. “EarlyCDT Lung” OR “Early-CDT Lung” OR “Early CDT Lung” (1)

2. Oncimmune – 1)

3. (autoantibod* OR auto-antibod* OR “auto antibody” OR “auto antibodies”) AND (lung* OR pulmonary) AND (neoplas* OR carcinoma* OR cancer* OR nodule* OR tumor* OR tumour* OR lesion OR mass OR masses OR indeterminate OR “coin lesion” OR “coin lesions” OR “ground glass” OR ground-glass) (35)

4. (autoantibod* OR auto-antibod* OR “auto antibody” OR “auto antibodies”) AND (NSCLC OR SCLC) (8)

5. (autoantibod* OR auto-antibod* OR “auto antibody” OR “auto antibodies”) AND (noncalcified OR “non calcified” OR non-calcified) AND (nodule* or lesion* or mass or masses) AND (lung OR pulmonary) (7)

Key: * = truncation.

International Prospective Register of Systematic Reviews (PROSPERO) via Centre for Reviews and Dissemination

Date range searched: inception to 8 March 2021.

Date searched: 9 March 2021.

Records retrieved: 0.

#1 EarlyCDT or “Early CDT” or Early-CDT (0)

#2 “Early Cancer Detection Test” (0)

#3 “ECLS” trial or “ECLS trials” (0)

#4 Oncimmune (0)

#5 MeSH DESCRIPTOR Autoantibodies (20)

#6 autoantibod* or auto-antibod* or AABT or AAb or AAbs or TAAb or TAAbs (195)

#7 #5 OR #6 (199)

#8 MeSH DESCRIPTOR Lung Neoplasms EXPLODE ALL TREES (475)

#9 MeSH DESCRIPTOR Solitary Pulmonary Nodule (5)

#10 (lung* or pulmonary or bronchial or bronchogenic) adj3 (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*) (1337)

#11 (neoplas* or carcinoma* or cancer* or nodule* or tumor* or tumour* or malign* or adenocarcinoma* or blastoma*) adj3 (lung* or pulmonary or bronchial or bronchogenic) (606)

#12 NSCLC or SCLC (502)

#13 (lung* or pulmonary) adj2 (lesion* or mass or masses) (47)

#14 (lesion* or mass or masses) adj2 (lung* or pulmonary) (31)

#15 (noncalcified or “non calcified” or non-calcified) adj2 (nodule* or lesion* or mass or masses) (1)

#16 (nodule* or lesion* or mass or masses) adj2 (noncalcified or “non calcified” or non-calcified) (0)

#17 NCPN (0)

#18 (ground-glass or solid or part-solid or subsolid or sub-solid) adj2 (nodule* or lesion* or mass or masses) (41)

#19 (“ground glass” or “part solid” or “sub solid”) adj2 (nodule* or lesion* or mass or masses) (10)

#20 (nodule* or lesion* or mass or masses) adj2 (ground-glass or solid or part-solid or subsolid or sub-solid) (13)

#21 (nodule* or lesion* or mass or masses) adj2 (“ground glass” or “part solid” or “sub solid”) (2)

#22 “ground glass opacity” or “ground glass opacities” (40)

#23 GGN or GGNs or GGO or GGOs (22)

#24 (benign or malignant or indeterminate) adj2 nodule* (44)

#25 nodule* adj2 (benign or malignant or indeterminate) (16)

#26 “coin lesion” or “coin lesions” (1)

#27 IPN or IPNs (8)

#28 #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14 OR #15 OR #16 OR #17 OR #18 OR #19 OR #20 OR #21 OR #22 OR #23 OR #24 OR #25 OR #26 OR #27 (1636)

#29 #7 AND #28 (0)

Key:

• MeSH DESCRIPTOR = subject heading (MeSH)

• * = truncation

• adj3 = terms within 3 words of each other (order specified).

Domain 1: risk of bias arising from the randomisation process

Domain 2: risk of bias due to deviations from the intended interventions (effect of assignment to intervention)

Domain 3: missing outcome data

Domain 4: risk of bias in measurement of the outcome

Domain 5: risk of bias in selection of the reported result

Overall risk of bias