Journals LibraryNHS NIHR - National Institute for Health Research
Predictors of Dementia in Primary Care: A cohort study using The Health Improvement Network (THIN) database
Predictors of Dementia in Primary Care: A cohort study using The Health Improvement Network (THIN) database
28 May 2015
01 June 2012
30 June 2014
Dementia; Memory; Risk Assessment; Primary Care; Routinely Collected Data
- Dr Kate Walters, Research Department of Primary Care and Population Health, UCL
- Dr Greta Rait, Research Department of Primary Care and Population Health, UCL
- Professor Steve Iliffe, Research Department of Primary Care and Population Health, UCL
- Dr Irene Petersen, Research Department of Primary Care and Population Health, UCL
- Professor Irwin Nazareth, Research Department of Primary Care and Population Health, UCL
- Sarah Hardoon, Research Department of Primary Care and Population Health, UCL (Statistician, research associate)
- Professor Rumana Omar, Department of Statistical Science, UCL (Professor of Medical Statistics, expert in risk algorithms)
- To explore the relationship between recording of potential presentations of dementia and dementia diagnosis in primary care.
- To identify factors recorded in GP case notes predictive of a dementia diagnosis.
- To develop and validate a dementia risk prediction tool incorporating these factors to predict five-year risk of dementia.
Changes to project objectives
Following consultation with a risk algorithm expert (RO, as above) we modified the design and statistical analysis plan, changing the ratio of practices randomly selected for development and validation cohorts from 50:50 to 80:20 to increase the number of events and power in our development cohort.
This study was conducted using routinely collected healthcare data from The Health Improvement Network (THIN) primary care database (http://www.epic-uk.org/), an extensively used and validated data resource. At the time of the study, the THIN database included around 10 million patients with geographical coverage broadly representative of the UK population. Medical diagnoses, symptoms, procedures, prescriptions, health promotion activity and referrals to secondary care are recorded in a longitudinal healthcare record for each patient, and identifying personal information is removed. We only used data that had met quality standards for an acceptable level of data recording.
We extracted data on 1,359,729 patients aged 60 to 100 years from 472 participating general practices across the United Kingdom from the THIN database. They were free from dementia, memory symptoms, confusion/delirium and cognitive impairment at baseline, and followed for recorded incidence of Alzheimer’s, vascular or mixed dementia, memory symptoms, cognitive impairment and confusion between 1995 and 2011.
Objectives 2 & 3. Development cohort study
We randomly selected 377 General Practices in the THIN database identifying 930,395 patients aged 60 – 95 years without dementia at baseline. We excluded people with less than a year of follow-up data. Our outcome was newly recorded dementia diagnoses including Alzheimer’s disease, vascular dementia, and mixed dementia, but excluding dementia diagnoses associated with Parkinson’s disease, Lewy body dementia, Huntingdon’s, Picks, HIV, drug-induced and alcoholic dementia.
We identified socio-demographic factors, health measurements, lifestyle, diseases and medication potentially associated with an increased risk of dementia from the literature, and determined their association with dementia within 5 years of follow-up in our cohort. We used this information on potential predictive factors to develop five-year dementia risk scores for two age groups: those aged 60-79 years and 80-95 years. Separate model development was carried out for the two age groups in the development cohort. We used the two-fold Fully Conditional Specification (FCS) algorithm method for multiple imputation of longitudinal clinical datasets to impute missing data. For each age group we derived the dementia risk score using a Cox proportional hazards regression model, with robust standard errors to account for clustering of individuals within general practices. Backwards elimination was used to determine which of the predictive factors considered were retained in the final model.
Objective 3. Validation cohort study
We externally validated the risk algorithms in a cohort of 264,224 patients from 95 separate practices contributing to the THIN database. For each age group, the model developed using the development cohort was applied to individuals in that age group in the validation cohort, to assess performance. We assessed the discriminative performance of the dementia risk models by computing the Uno’s C and Royston’s D statistics for the validation cohort. We assessed calibration by comparing the observed and predicted dementia risk in the validation cohort per decile of predicted risk, and computing the calibration slope. We calculated the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) using a range of potential risk thresholds, to explore the clinical utility of the risk algorithms.
Objective 1: Presentations of dementia and subsequent dementia diagnosis
A total of 43,217 new cases of recorded dementia were identified during our study period over 8,788,310 person-years of follow-up, corresponding to an overall incidence rate of 4.92 events per 1000 person years (95% CI 4.87 to 4.96). Incidence of recorded new memory symptoms, cognitive impairment and dementia diagnoses all rose over the period 1995 – 2011 by 10.5% (95% CI 10.2 to 10.8) per annum, 16.2% (95% CI 15.3 to 17.0) and 3.2% (95% CI 3.0 to 3.5) respectively, while incidence of recorded confusion/delirium fell by 1.91% per annum.
Overall 15,895 (37%) of 43,217 people with new dementia diagnoses had a prior record of memory loss, 1,868 (4%) had a prior record of cognitive decline and 9,000 (21%) had a prior record of confusion. The median time from first recorded presentation to a first dementia diagnosis was 8-9 months, with similar time gaps for all three presentations. Incidence of dementia after initial presentation was 279.0/1000PYAR (95%CI 252.8 to 308.0) for memory symptoms, 337.3/1000PYAR (95%CI 254.9 to 446.3) for cognitive impairment and 187.2/1000PYAR (95%CI 168.5 to 208.0) for confusion.
Objectives 2 & 3: Predictors of dementia and a new dementia risk score
There was a sharp increase in incidence of dementia at age 80 years onwards, and some differences in associations of risk factors with subsequent dementia diagnosis, supporting development of separate risk algorithms for those aged under and over 80 years. For those aged 60 – 79 years newly recorded dementia diagnoses were positively associated with increasing age, female gender, calendar year, living in a deprived area, current smoking, hazardous/harmful alcohol drinking, history of stroke/Transient Ischaemic Attack (TIA), diabetes, Coronary Heart Disease (CHD), Atrial Fibrillation, statin prescriptions and current depression/anti-depressant drugs, anxiety/anxiolytic drugs, insomnia/hypnotic drugs, and aspirin use. There was a small negative association with both Body Mass Index (BMI) and systolic blood pressure. There were no significant associations with Non-Steroidal Anti-Inflammatory Drugs (NSAIDs, excluding aspirin), and anti-hypertensive drugs. For those aged 80 – 95 years at baseline new dementia diagnoses were positively associated with increasing age, female gender, history of stroke/TIA, diabetes, AF, statin prescriptions, hazardous/harmful alcohol drinking, current depression/anti-depressant drugs, anxiety/anxiolytic drugs, hypnotic drugs and aspirin use. There was a small negative association with current smoking, BMI, systolic blood pressure, anti-hypertensive drugs and NSAIDs (excluding aspirin). There were no significant associations with living in a deprived area, CHD and total cholesterol/HDL ratio.
The discrimination/calibration of the risk score were good for the 60-79years model (D statistic 2.03 (95%CI 1.95 to 2.11), C index 0.84 (95%CI 0.81 to 0.87), calibration slope 0.98 (95%CI 0.93 to 1.02). The 80-95 years model performed poorly in terms of discrimination (Royston’s D statistic 0.86, 95%CI 0.76 to 0.95 and Uno’s C index 0.56, 95%CI 0.55 to 0.58) and calibration (calibration slope 1.04, 95%CI 0.89 to 1.18), when applied to the validation cohort.
Risk classification for those aged 60-79 years
Utilizing a range of possible cut-offs to indicate “high risk” for dementia the specificity of the risk algorithm is high but with lower sensitivity, and there was a high Negative Predictive Value (NPV), but a low Positive Predictive Value (PPV). These are reported in full in our paper (publication pending). We did not examine the specificity and sensitivity of the 80-95 years algorithm due to its poor performance in our validation study.
Incidence of recorded dementia, memory symptoms and cognitive decline have been rising, whilst recorded acute confusion/delirium is declining in primary care. Around half of people newly diagnosed with dementia in GP records have records of earlier memory symptoms, cognitive impairment or confusion/delirium. In those with earlier presentations there is a median delay of 8-9 months before dementia diagnosis is recorded, which may be reflective of time taken for referral for memory clinic/specialist assessment, investigations and the information to be conveyed back to the GP. Around a quarter of people recorded as presenting with memory symptoms to primary care will be diagnosed with dementia within a year, suggesting that this is an important group to consider for further investigation.
In our risk algorithm analysis we demonstrated that clinical data can predict five year risk of recorded dementia diagnosis for those aged 60-79 years, but not those 80-95 years. The risk algorithm for those aged 60-79 years has a high negative predictive value, and has potential applications to rule out those at very low risk for dementia in the next five years, avoiding further testing or intensive preventative activities.
Plain English summary
There are increasing numbers of people affected by dementia. It is a condition that has enormous costs to the person with dementia, their families and carers and also to health and social care. UK government policy promotes early diagnosis of dementia and wants GPs to be more proactive in making the diagnosis. There is also increasing research into how we can prevent dementia, and for this identifying people at higher risk for dementia is potentially important.
Our study used information from the anonymous general practice (GP) records of 1.3 million people over 60 years old in the UK. We investigated the relationship between recorded memory symptoms, confusion and mild cognitive impairment (mild problems on memory testing that do not meet criteria for dementia) and the future development of dementia. We identified a range of factors including lifestyle factors, health measurements, medical conditions and drug treatments that were associated with an increased risk of dementia. We used this to develop and test a new risk score, to predict people’s individual risk of being newly recorded with dementia within the next five years.
We found around half of people newly diagnosed with dementia have records of an earlier potentially related condition/symptom (memory symptoms, confusion or cognitive impairment). The average gap between this first recorded presentation and a recorded dementia diagnosis was 8-9 months. Around a quarter of people recorded as presenting with memory symptoms in general practice will be diagnosed with dementia within a year. We were able to develop a five-year risk score for dementia for those aged 60-79 years using this general practice health data. The risk score for those aged 80 and over did not work well. Some of the people identified as ‘high risk’ in our study might already be developing early dementia.
Our work suggests that GP data could be used to provide information to people aged 60-79 years about their risk of dementia. This risk score needs to be tested in other populations. It could be used to improve timely diagnosis of dementia in general practice, and to rule out those at very low risk of dementia from unnecessary testing.
- Kate Walters, Sarah Hardoon, Irene Petersen, Steve Iliffe, Rumana Z Omar, Irwin Nazareth, Greta Rait. Predicting dementia risk in primary care: development and validation of the Dementia Risk Score (DRS) using routinely collected data.
- Sarah Hardoon, Greta Rait, Irene Petersen, Steve Iliffe, Rumana Omar, Irwin Nazareth, Kate Walters. Memory symptoms, cognitive impairment and confusion and subsequent dementia diagnosis in primary care: cohort study using routinely collected healthcare data.
Public and participant involvement has been helpful in considering the implications of the research, and how the risk score could be used in practice. Discussions with public contributors and clinicians in practice suggested that further work is needed to understand the role of dementia risk assessment and case-finding in primary care, and how this can be best supported. This is being taken forward with a two linked qualitative studies exploring the role of dementia risk assessment and case-finding in primary care from the perspectives of patients and family care-givers and health care professionals.
Our new Dementia Risk Score (DRS) algorithm for 60-79 year olds can be added to clinical software systems. A practice could, for example, run this risk model on all eligible people and offer those at risk more detailed testing. Using a range of thresholds, there was good specificity but lower sensitivity, and a very high Negative Predictive Value, but a low Positive Predictive Value. This risk algorithm may be most helpful to ‘rule out’ those at low risk from dementia case finding programs. This might avoid unnecessary investigations and anxiety for those at very low risk and make these programs more cost-effective. The risk algorithm may also enable the identification of ‘at risk’ groups to approach for future research studies.
This project was funded by the National Institute for Health Research School for Primary Care Research (project number 79)
Department of Health Disclaimer
The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the NIHR School for Primary Care Research, NIHR, NHS or the Department of Health.