Clinical effectiveness and cost-effectiveness of parenting interventions for children with severe attachment problems: a systematic review and meta-analysis

Stability

The term natural history here refers to the progression, evolution and stability of early patterns of attachment. As a rule of thumb, providing there is no change in caregiving pattern (by either the same or different caregivers) and with secure attachment, there is evidence of overall stability of the pattern. 18 Insecure and, more so, disorganised attachment are associated with caregiver and caregiving difficulties, which are more likely to undergo change over a child’s development, both because of their likely inherent instability and because they are more liable to interventions which may influence them. 19 These factors are likely to be associated with a change in the child’s attachment pattern. However, Bowlby20 referred to ‘defensive exclusion’, by which he meant the child excluding new information about relationships which did not accord with his or her existing internal working models. This would suggest that there would need to be a sustained and perceptible change in caregiving to exert a meaningful effect on the child’s attachment pattern.

Evolution within disorganised pattern

There is some evidence21,22 that the behavioural pattern described as disorganised in infancy and early childhood evolves into coercive controlling or compulsive caregiving patterns in preschool and middle childhood, even in low-risk settings. 23 However, there may be continuing disorganisation at the representational level, as shown in narrative stem completion tasks19,24 and family drawings. 25

Change of assessed manifestation of attachment

With development, presumed manifestations of attachment, and therefore ways of assessing attachment, change. Thus, in infancy and early childhood, attention is given to the distressed child’s behaviour in relation to his or her caregiver, classically in separation and reunion procedures. In middle childhood, it becomes increasingly difficult to create sufficiently stressful situations in order to activate and then assess attachment behaviour. The solution has been to devise assessments of representation of attachment26 using narrative completions and pictures. In adolescence, there has been a further progression using linguistic representation of state of mind with respect to attachment, that is, coherence of accounts, by ‘surprising the unconscious’ (Ammaniti M, Candelori C, Dazzi N, De Coro A, Muscetta S, Ortu F, et al. University of Rome, 1990, unpublished protocol). The question then arises regarding how closely related the putative age-related manifestations or expressions of attachment are and how well they are measured by various proposed instruments used at different ages. This suggests that it might be preferable to refer to predictability rather than stability.

The significance of attachment and its relationship to psychopathology

In studying associations between attachment patterns and impaired functioning or psychopathology, the question arises about the nature of the association. If the impairment can be causally explained by prior or concurrent attachment difficulties, then the impairment can be properly considered as an aspect of the natural history. However, it is also possible that the antecedents of attachment difficulties – specifically, harmful parent–child interactions and their associated risk factors – could, independently of attachment, contribute to the functional impairment and psychopathology. In practice, it is difficult to disentangle these two mechanisms. 27 For this reason, discussion of the significance of attachment and its relationship to psychopathology is placed in its entirety under natural history.

There are various examples of studies that have attempted to link attachment patterns with subsequent disorders or outcomes. Studies have sought to show that behaviour problems in children can be predicted by attachment patterns. 28–30 These include both emotional and conduct problems. 30 For example, Speltz and colleagues31 found that only 20% of a sample of clinic-referred children with early-onset conduct problems were securely attached to their parents, whereas 72% of children in the control group were securely attached. Futh and colleagues32 examined how attachment representation related to social functioning and psychopathology in a sample of 113 children, 50% of whom were defined as high risk and 50% as low risk. Behaviour problems rated by teachers were linked to disorganised attachment patterns. Disorganised attachment was also predictive of poorer social functioning32 and poor school attendance, conduct disorder and academic underachievement. 33 Offenders are also more likely to report disturbed or insecure attachments, and separation from attachment figures in childhood is suggested as being associated with personality disorder in offenders. 34 Insecure attachment is also purportedly linked to increased reactivity to stress,35 notably in increased cortisol reactivity, which has itself been associated with a range of psychopathologies, including psychotic illness. 36 Longitudinal studies have linked disorganised attachment with hostility and hyperactivity, aggression and oppositional defiant disorder in children37 and with dissociative symptoms in 17- to 19-year-olds. 27 Furthermore, attachment disorders, as distinct from insecure attachment patterns, are purported to have increased comorbidity with conduct disorders, developmental delay, attention deficit hyperactivity disorder and post-traumatic stress disorder. 38

One of the problems, however, is that much of this often-quoted research uses a range of methodologies, often in small or selected samples and often using bespoke or unvalidated instruments for measuring attachment. For us to be confident in these associations, this research needs to be carefully scrutinised using high-quality standards. Although insecure attachment patterns may represent risk factors for some future problems, approximately one-third of infants in normal populations show some form of insecure attachment. Thus, insecure patterns of attachment should not be considered as indicators of pathology, but rather, may be considered as potential risk factors for the child’s future functioning. 39 In this sense, although many people with psychopathology may be more likely to have had insecure attachments, many infants with attachment pattern difficulties may not go on to develop psychopathology. Indeed, some argue that measurements from the SSP are poor predictors of psychopathology in longitudinal studies. 40

Work that has sought to quantify these issues suggests that genetic influences for prosocial behaviours are strong and independent of attachment pattern. 41,42 The interaction between environment and genetics is complex, with different children varying in susceptibility to environmental influences on their subsequent attachment pattern. However, once a particular attachment pattern has developed, genetic influences appear to take a significantly less part in the development of those behaviours for which attachment patterns are seen as risk factors.

In summary, while, there is some evidence that disorganised attachment patterns are related to psychopathology, the link between insecure patterns and subsequent problems is not so clear. 39 This lends itself urgently for review, given that many clinicians use the paradigm of attachment in assessment and intervention, and there is a need to better understand the evidence that informs clinical practice. We have enough literature to consider that disorganised attachment is the most promising candidate. It is associated with poor outcomes and is a group to follow up, exploring systematically whether or not parental interventions are effective or cost-effective. Attachment disorders, to be discussed below (see Attachment disorders), could also be included in the overall term ‘severe attachment problems’.

Assessment of attachment behaviour

Representations of attachment

The two main procedures by which to assess the older child’s representations of attachment are narrative stem completion and the use of pictures, commencing from the age of 4 years. Variants include the MacArthur Story Stem Battery (MSSB),49 the Story Stem Assessment Profile (SSAP) (Hodges J, Steele M, Hillman S, Henderson K, 2002, unpublished data) and the Manchester Child Attachment Story Task (MCAST). 50 Drawings are used in the Separation Anxiety Test (SAT) and the School-age Assessment of Attachment. 51

Coherence of accounts

These assessments are based on semistructured interviews with the child, and what is rated is the linguistic representation of the child’s state of mind with respect to attachment. The two main tools are the Child Attachment Interview (CAI) for 7- to 11-year-olds, adapted from the Adult Attachment Interview (AAI),52 and the Friends and Family Interview. 53

Self-report attachment pattern questionnaires have also been used in 4- to 12-year-olds. 54

Meta-analysis evidence23 shows numerous subcatergorisations of attachment patterns, but does suggest that the measurement of disorganised attachment can be reliable.

How do we identify those who will benefit from interventions?

The first step in providing clarity as to who should be prioritised for treatment is to clarify how we identify the children who will benefit from the available treatments in a valid and reliable way. As discussed in Chapter 1, a central problem facing this review on attachment is large differences between attachment patterns and attachment disorders. Furthermore, how do we identify severe attachment problems in infants or children when stability over time may vary? For example, a meta-analysis of 840 infants in nine samples, where assessments took place between 2 and 60 months apart, found a stability of r = 0.34 for disorganised attachment. 23 The concept of attachment may also be used by clinicians in many different ways, with some straying far from Bowlby’s original construct1 by using it to describe broad aspects of the quality of relationships between parent and child. These all lead to misunderstandings in interpreting the literature.

Many attachment instruments have not been well validated. There is currently no biological measure of attachment patterns. In regard to attachment disorders, the construct is under scrutiny and subject to revision, as is the case with both the APA (DSM-IV to DSM-V56,58) and WHO [ICD-1055 to International Classification of Diseases, Eleventh Edition (ICD-11)88] definitions of attachment disorders. We can, however, review extant research diagnostic systems for attachment disorders from both groups.

Clinicians do not currently know which is the best assessment tool to use to identify severe attachment problems. Therefore, the aim of the first supplementary systematic review of the literature is to identify the valid and reliable assessments of attachment patterns and disorders. As this review has its focus on parenting interventions, we are interested in identifying severe attachment problems early in life. In order to better understand the relationship between attachment patterns and attachment disorders, we will also explore how they relate to one another by looking for any studies that have compared their use in the same children at the same time.

Who is at risk and who is it that we should be treating?

Once we have identified clear ways of measuring severe attachment problems that are reliable and valid, we need to know what this means in terms of outcomes for the child, whether they receive the intervention or not. We need to understand more clearly what it means to have different attachment patterns in infancy1,6,8 or attachment disorders57 in older infants and children. What happens to those children in the longer term? For health economic reasons, we are particularly interested in studies that look at follow-up that takes infants or children at least to the end of primary school education, and hopefully considerably beyond, to inform any health economic modelling work. Very short- or short-term studies, although important for many reasons, are less useful for this purpose.

Using the findings from the first systematic review on assessment and measurement, we seek to evaluate the longer-term outcomes for those identified that have been left untreated. This forms the second supplementary systematic review (see Chapter 5). This will look at the evidence from longitudinal studies that follow children up for 10 years or more. We will explore attachment outcomes and, where possible, whether or not other outcome information is of value in its current form. This is a small supplementary review to inform the main focus of this work, which centres on the clinical effectiveness and cost-effectiveness of parental interventions.

Which parenting interventions work and are they cost-effective?

Clinicians are often unsure about the best intervention or treatment options for the children (and their families) identified as having severe attachment problems. Resources for these interventions are limited. There are ambiguities surrounding the clinical effectiveness and cost-effectiveness of interventions provided to families, including whether or not any improvement in attachment would be associated with a change in other outcomes (educational attainment, psychological well-being, quality of life, future criminality, etc.), and how acceptable these interventions would be in terms of the practicalities of delivering them in busy services and their acceptability to service users.

The attachment literature investigating the efficacy of parenting interventions consists of a variety of research designs, from single case-study designs to randomised controlled trials (RCTs). We seek to systematically review this literature, selecting only RCT designs which illustrate the highest level of evidence for the clinical efficacy of treatment. Do the interventions work (see Chapter 6) and are they cost-effective (see Chapter 7) in a current environment where funding is tight?

By liaising with experts and service users in patient and public involvement (PPI) groups as we conduct our reviews, we hope to identify any gaps in the literature and the acceptability of interventions that are found to be clinically effective.

Inclusion criteria

All study designs were eligible in this stage of the review. For inclusion, studies had to provide sufficient data for extraction. Sensitivity and specificity analysis data were not a requirement, although this analysis was undertaken where possible (only where complete raw data were available).

The PICOS criteria were as follows:

• Population and setting Children being assessed for attachment patterns or disorders where the research reports an average age of 13 years or below (we chose this in discussion with PPI and experts in the light of the overall aim of the review on early parental interventions). As discussed in Chapter 1, we refer to attachment patterns to mean any paper that explored attachment patterns, attachment styles or attachment organisation, recognising that different authors in the field use different terminology. We felt that it was important not to exclude any papers that were relevant but used different terminology.

• Intervention Screening, assessment and/or diagnostic tools evaluating attachment patterns or disorders. The instrument must have been under development or evaluation, and must have been a completed tool or subscale on attachment rather than an individual item. Attachment pattern requires a primary caregiver (NB a member of staff in a child care institution is not considered a fair test).

• Reference A comparison tool assessing attachment patterns or disorders identified by ICD-1055 or DSM criteria. 56

• Outcomes Studies reporting on the psychometric properties and validity of the tools.

• Study design Cross-sectional studies, case–control studies or prospective cohort studies incorporating any method of assessment (for example observation, semistructured interviews and questionnaires).

Data extraction

A data extraction form was developed, piloted and adapted on the basis of this piloting. Subsequently, all studies were dual extracted and reviewers met to agree and discuss discrepancies in data items. Where studies had multiple publications, data were extracted as a single study. The following items were extracted from each study: study characteristics, population details, index and reference tool details, data for sensitivity and specificity analysis, economic resource information and psychometric properties of index and reference tools.

Diagnostic accuracy

Where possible, a sensitivity and specificity analysis was calculated.

Quality assessment strategy

Each study was assessed for methodological quality by two reviewers using the quality assessment of diagnostic accuracy studies – version 2 (QUADAS-2). 92 Discrepancies in quality assessment were discussed and resolved between reviewers. QUADAS-292 is a validated quality assessment tool for diagnostic studies. It consists of four key domains: domain 1, patient selection; domain 2, index test(s); domain 3, reference standard(s); and domain 4, flow and timing [flow of patients through the study and the timing of the index and reference test(s)]. To help reach a judgement on the risk of bias, signalling questions were included. These flagged aspects of study design related to the potential for bias and aimed to help reviewers make risk-of-bias judgements. A further three questions in the tool consider the applicability of the patient selection, index tool and reference tool. Each item was rated ‘yes’, ‘no’ or ‘unclear’ according to the guidance provided.

Following quality assessment of the first few studies, it became apparent that the range of study designs made two questions irrelevant to some studies, as follows:

• Domain 2. Question 2: if a threshold was used, was it prespecified?

• Domain 3. Question 1: is the reference standard likely to correctly classify the target condition?

In order to avoid penalising studies where these aspects were not relevant, we agreed to enter a response of ‘not applicable’. The following circumstances led to an opinion of ‘not applicable’: in cases where screening was performed using observational opinion, question 2 is not applicable; in studies where diagnosis is not the objective and, therefore, a ‘cut-off’ is not specified, question 2 is not applicable; and finally, in case–control studies where only one tool is assessed, question 1 is not applicable.

Data synthesis

A meta-analysis was not conducted because it was not appropriate. A wide range of instruments were compared with the reference standards, most of which were not repeated in further study to enable comparison between studies. A descriptive summary of results is presented.

Study characteristics

An overview of study characteristics is detailed in Table 1 and a taxonomy of the tools identified is presented in Tables 2 and 3. Thirty-three studies were published between 1988 and 2011, of which the majority were undertaken in the USA (n = 1824,47,93,94,96–100,102–104,106,111,113,115,116,118,121), with the rest spread across the UK (n = 450,52,107–109), Canada (n = 425,105,110,119), Germany (n = 2101,122), the Netherlands (n = 2117,120), Romania (n = 194) and Spain (n = 1114).

A range of screening and/or diagnostic tools were identified, and the variation in the classification systems used with procedures was large. In relation to the SSP alone, 16 studies reportedly used the original Ainsworth (1969,2 19788) procedure,25,93,94,98,102,103,105,106,113,115,116,118–122 using 12 different variations on the classification system. Confusingly, many different authors use a variety of nomenclature to describe various classifications of attachment patterns. This is summarised in Table 1 and Box 1. Some authors describing subcategories of insecure attachment use different names. For example, anxious–avoidant attachment is sometimes simply called ‘avoidant’, and sometimes ‘defended’ or ‘dismissing’. Similarly, anxious–ambivalent attachment pattern is sometimes simply called ‘ambivalent’, and sometimes ‘anxious–resistant’ or ‘preoccupied’ (see Box 1). In some papers it is not clear whether authors are creating new categories with subtle differences or simply renaming existing categories. Nonetheless, this practice makes the literature extremely confusing to new trainees coming into the field, or indeed any practitioners or researchers wishing to better understand the field of attachment work.

There are two main schools of hierarchicalisation for attachment patterns. Table 1 and Box 1 describe classifications used by Ainsworth8 and extended by Main and Solomon6 whereby organised attachments may be secure or insecure, with insecure attachments having subcategories. A separate disorganised category exists in this system. By contrast, another body of work led by Crittenden127 suggests that the disorganised category is actually a subgroup of insecure attachment that is unpredictable or changing in nature. Crittenden’s A/C category, sometimes called avoidant/resistant, might be regarded as mapping on to the disorganised attachment described above, but it is understood in different ways. Nevertheless, it is a category of interest in terms of long-term psychopathology and long-term outcomes.

The identified comparison tools employed various techniques to assess children, utilising observational techniques, questionnaires, interviews, stories with child responses and interpretation of a child’s drawings. Procedures involved the child, caregiver, teacher, researchers and clinicians depending on the child’s age and the tool used. Descriptions of these assessments in full are included in the taxonomy (see Tables 2 and 3).

A total of seven papers included tools assessing children with attachment disorders. 95,98,101,104,108,110,114 Tools included DSM-IV, Disturbances of Attachment Interview (DAI), Preschool Age Psychiatric Assessment (PAPA), May-Nichols Child Behaviour Rating Scale (CBRS), Randolph Attachment Disorder Questionnaire (RADQ), Revised Classification of Mental Health and Developmental Disorders of Infancy and Early Childhood (DC: 0–3R), RDC for ICD-10, Behavioural and Emotional Rating Scale (BERS) with Biopsychosocial Attachment Types (BAT), Relationships Problems Questionnaire (RPQ) used in combination with the Child and Adolescent Psychiatric Assessment (CAPA), Waiting Room Observation (WRO) and the RPQ. Relationships Problems Questionnaire (RPQ) used in combination with the Child and Adolescent Psychiatric Assessment (CAPA) and Waiting Room Observation, (WRO) and the RPQ alone.

Population characteristics

Figure 3 summarises the profile of the population studied. Table 1 includes detail on the participant age, gender and ethnicity.

FIGURE 3.

Summary of study populations from studies included in supplementary review 1.

The age of the children assessed for attachment problems ranged from 12 months to 20 years. We included studies where the mean age was 13 years or below, and therefore in some cases the upper end of the age range exceeded 13 years, as the sample included children older than this but with the average age still below 13 years. Many studies did not report on the child’s ethnicity. In studies where ethnicity was recorded, participants were predominantly white or Caucasian. Other ethnic groups were generally under-represented.

Parental/caregiver information was less likely to be described in detail, with 29 studies reporting incomplete data or no detail. 24,25,47,50,52,93–96,99–102,104–108,111–122 In those studies providing demographic detail, parental age ranged from 12 to 70 years and ethnicity was predominantly white or Caucasian.

Information on the population description was collected and classified into 11 categories (see Figure 3). These groupings can be further categorized according to four general dimensions:

• socioeconomic status (SES)

• risk populations (such as fostered, adopted or maltreated children or those in institutional care)

• mental health status of child (e.g. RAD or mental health user)

• groups deemed as low risk or population samples (e.g. playgroup children).

Four studies did not adequately describe the population,50,117,118,120 for example classifying the population as ‘low risk’ without further clarification.

No papers reported on child literacy level; however, this is not surprising given that many studies were assessing infants with instruments completed by researchers or parents. A handful of studies did report on other child ability measures such as picture vocabulary and verbal ability.

Quality assessment

Table 4 summarises the results of the quality assessment. High risk of bias was the most frequently reported classification in domain 4 (flow and timing). The lowest risks were in domain 2 (index test) and domain 3 (reference standard). An unclear risk of bias was the most frequently reported classification in domain 1 (patient selection) and in domains 2 and 3.

All studies achieved a low risk of bias on all three applicability questions (Table 5) in QUADAS-2. 92 The screening criteria were such that this would inevitably be the outcome.

The three applicability questions in QUADAS-2 are:

• Is there concern that the included patients do not match the review question?

• Is there concern that the index test, its conduct or its interpretation differ from the review question?

• Is there concern that the target condition as defined by the reference standard does not match the review question?

Taxonomy of instruments assessing attachment patterns

The findings of the review are split into tools assessing (1) the construct of child attachment to the caregiver, described as an attachment pattern or an attachment security score, and (2) the symptomatology of attachment disorder as defined by the DSM-IV56 or ICD-10. 55

Taxonomy of assessment for attachment disorder

Psychometric properties

Table 6 summarises the psychometric properties of the index and reference tests and reports the outcome of screening for content and construct evidence.

With the exception of two studies,94,119 all publications reported at least partial reliability data.

Test–retest data were the least reported information in the included studies, with only five studies reporting these data for either the index or reference test. 47,52,110,111,120

Inter-rater reliability data were the most frequently reported validity data. A total of 26 studies reported these data for the index test24,25,47,52,93,95–108,111,113–118,121,122 and 23 studies for the reference test. 24,25,47,93,96,98–100,102,105–108,111,113,115–122 Of these 26, 24 had good inter-rater reliability as defined by a level of 0.7 or above. 25,47,52,93,95–103,106–108,111,113–117,121,122

Cronbach’s alphas were reported in 12 studies for the index tests (12 studies α > 0.7)47,52,95–97,104,108,110,111,113,114,116 and in four studies for the reference tests (four studies α > 0.7). 97,110,110,114

Sensitivity and specificity analysis

Nine studies were found that compared an attachment assessment procedure with a reference standard. 25,93–95,102,108,115,118,122 Only three of these reported data that enabled calculations of concurrent validity sensitivity and specificity scores: two for attachment patterns93,94 (Table 7) and one for attachment disorders95 (Table 8).

The characteristics of these studies were summarised in Table 1. The majority validated an attachment assessment procedure under evaluation against the SSP. Most studies (six) were conducted in the USA25,93,94,102,115,118 and the remainder (three) were conducted in European countries (Germany,122 Romania95 and Scotland108). The size of the samples ranged from 33122 to just below 300. 118 There was no significant discrepancy in the proportions of boys and girls in any of the studies. Two studies did not report sufficient data to calculate 2 × 2 tables. 108,122

Most studies attempting to compare two attachment instruments did not include raw data. Two studies reported data that allowed concurrent validity calculation of sensitivity and specificity of a new procedure for assessing infant–mother attachment pattern. 93,94 Both studies were conducted in the USA and the samples sizes were small (60 toddlers93 and 16 twin pairs94 respectively).

Inclusion criteria

The PICOS criteria were as follows:

• Population and exposure Studies that measured attachment disorder or disorganised attachment pattern in children under the age of 13 years at inception. If infants were too young to measure attachment at inception, studies were included if attachment was measured at, or before, the age of 12 months.

• Instruments Studies were only included where attachment problems had been measured using the diagnostic criteria for attachment disorder or disorganised attachment patterns using the SSP8 with the disorganised pattern described by Main and Solomon. 6 If any measures had been validated against the reference standard in supplementary review 1, studies including those measures would have been included here, but none was found.

• Comparator Those without a disorganised attachment pattern or attachment disorder, at baseline or earliest time point, served as a comparator against those with a severe attachment problem.

• Outcomes Studies had to contain any of the following:

  • Data on the prevalence of severe attachment problems (defined as disorganised attachment pattern6 or the diagnosis of attachment disorder). 55,56

  • Epidemiological data including outcome data. The long-term outcomes searched for were stability of severe attachment problems, measured by an adolescent or adult measure of attachment; rates of subsequent mental ill health; psychosocial development; educational attainment; entry into care; or involvement in the criminal justice system. Only validated outcome measures or objectively measurable full-population outcomes, such as school grades or criminal convictions, were examined in each of these domains.

• Study design Studies were prospective longitudinal cohorts with a follow-up period of 10 years or more.

There are many studies on short-term outcomes of severe attachment problems which have been the subject of systematic reviews23 and selective reviews by other authors. 147 Our initial scoping demonstrated a vast literature, and as the main focus of this review was on parental interventions, the steering group decided after lengthy discussion to limit this supplementary review to long-term sequelae (10 years or more) of severe attachment problems to explore any important long-term outcomes that might provide useful information for health economics analysis.

Data extraction

Data were extracted by two independent reviewers, who met to discuss any discrepancies in order to reach an agreement. Studies with multiple papers were examined and extracted separately. The following items were extracted for each paper: study characteristics, population details, prevalence and incidence of RAD and attachment patterns, stability of attachment and the specified long-term outcomes of children with severe attachment problems. The data extracted on participant characteristics and prevalence of attachment classifications relate to the sample that was followed up longitudinally, rather than the full, original sample. Where data were not reported in full, calculations were made from the reported data included. If this was not possible, the paper was excluded.

Quality assessment

For each study reporting the prevalence of severe attachment problems and/or the long-term outcomes for this population, we applied a bespoke 13-item quality assessment tool (Table 9). The tool was developed using the CRD’s suggested quality criteria for assessment and recommendations from a systematic review of tools to assess the quality of cohort studies,148 and by reference to previously administered cohort quality scales, such as the Newcastle–Ottawa Quality Assessment Scale for cohort studies. 149 We specified an attrition rate of ≤ 20% as adequate for long-term cohort studies to receive a rating of ‘low bias’, based on previous quality assessment guidelines for cohort studies. 150

The questions were developed to cover the five key domains for tools assessing study quality based on Sanderson and colleagues’148 systematic review, with studies being allocated high, low or unclear bias. Table 9 summarises these quality assessment criteria; further details of scoring can be found in Appendix 3.

Data synthesis

Given wide variability in outcomes, it was not appropriate to use meta-analytic procedures for this phase of the review. Instead, a narrative overview of the studies is given.

Study and sample characteristics and prevalence data

Table 10 shows the participant and study characteristics of included studies.

The eight studies were included as they met the following criteria:

• Four of the eight studies reported long-term data in relation to severe attachment problems at baseline. 45,151–153

  • Two studies measured the stability of attachment over time. 151,152

  • Two studies examined the relationship between severe attachment problems in infancy and later mental health problems. 45,155

• Four studies met the criteria in terms of reporting long-term outcomes and measured severe attachment problems at inception, but did not report outcomes of disorganised attachment separately; therefore, the data have not been presented. 154–157

The prevalence data for the study by Carlson and colleagues153 are not included in Table 10, as although this study reports some data on prevalence at baseline, these are reported as a mean rating score across the sample and were not therefore available in a usable format.

Population characteristics

Three of the papers reviewed45,152,153 were part of the Minnesota Longitudinal Study of Risk and Adaptation,158 a large ongoing study in the USA. The other studies were conducted in various different countries but the predominant ethnicity was Caucasian, with the exception of an Israeli study with a 100% Jewish population. 155 The length of follow-up ranged from 10 to 28 years.

Table 10 shows the sample size at enrolment and follow-up, which also varied among studies. Sample size at follow-up ranged from 21154 to 162153 and attrition ranged from 11%156 to 82%. 155

Three studies successfully followed up more than 60%153,154,156 and only one study managed to successfully follow up more than 70%. 156 The participant information is based on the sample of participants that was followed up longitudinally. This explains slight differences in participant details in the Minnesota papers, which all followed up infants recruited as part of the same original sample. All participants were recruited at 12 months of age or under. The Minnesota sample was defined as being at risk of poor developmental outcomes because the parents were living below the poverty line at the time of birth. Most other samples comprised participants from middle-class backgrounds. Two of these papers described their samples as low risk. 151,157 Although the authors were not explicit about what ‘low risk’ means, they appeared to be making reference to the SES of the parents. There was one population study that reflected all infants born during a 1-year period in a geographical region. Dan and colleagues155 approached all parents who had given birth, within the same year, in Greater Haifa, Israel. Thus, their sample covered the full range of SES from that region.

Rates of disorganised attachment in the various papers reviewed

Table 10 shows the prevalence of disorganised attachment patterns in the various samples of the papers found for this phase, as well as the organised patterns of attachment. The papers that defined their samples as at risk owing to poverty reported considerably higher prevalence of disorganised attachment than the other studies (35%45 and 40%152).

Similarly, the two papers in which the sample was described as low risk151,157 had some of the lowest prevalence of disorganised attachment. Steele and colleagues157 reported that 8% of participants showed a disorganised attachment pattern with their mother and just 2% with their father. Aikins and colleagues151 reported that 11% of their sample had a disorganised pattern. Interestingly, one of the lowest reported prevalences of disorganised attachment was found in Dan and collagues’155 sample. Dan and colleagues155 found that only 3% of their sample, which included families from the whole population of Greater Haifa in Israel, had disorganised attachment patterns.

Ammaniti and colleagues154 found that 20% of their middle-class sample had disorganised attachment patterns, and Jaffari-Bimmel and colleagues156 found that 16% of their sample, who had been adopted internationally into middle-class families, had a disorganised attachment pattern. The variability in rates of disorganised attachment reflects the very different populations studied.

Quality assessment

Table 11 summarises the quality of the studies. Some items were rated as low bias fairly consistently across the studies. For example, all studies used the same measure of attachment across all participants.

Two items assessed the methods of selecting study participants. These items asked whether or not the cohorts were comparable by source and by age (items 1 and 2, respectively). Only one study154 drew its cohort from different samples/sources (item 1). The majority recruited from the same sources. Two were unclear. 151,155 All the papers showed that the samples were recruited at the same age (item 2); however, it was unclear whether or not the groups in two papers from the Minnesota study45,152 were comparable by age. Although the groups were recruited and assessed for severe attachment problems at the same age, data from some of the initial assessments were lost, and so the data analysed were, consequently, a mixture of two or three different assessment periods, conducted at different ages, from whatever data were available.

Items 3–7 assessed the methods of measuring exposure and outcome variables. Items relating to the measurement of exposure considered the ascertainment of exposure, blinding of coders measuring exposure and whether or not a description of exposure was provided for the entire sample at baseline (items 3, 4 and 6, respectively). The same validated measurement of attachment organisation or presence of RAD/DAD was used across the sample in all the papers (item 3). The majority of papers demonstrated that the coders of the initial attachment measure were blind to risk factors or predictive variables related to the exposure status (item 4). Two papers were unclear. 154,155 Two papers156,157 provided attachment classifications for the full original sample (item 6). The remainder did not report these data.

Items relating to the measurement of outcomes considered whether or not outcomes of interest were present at the start of the study, and whether or not coders of outcome measures were blind to exposure status (items 5 and 7, respectively). The findings from item 5 have not been displayed in Table 11 because this item was not considered to be applicable to any of the papers. It asked whether or not there was demonstration that outcome(s) of interest were present at the start of the study. Participants were recruited at an early age in all the papers and outcomes measured in late childhood, adolescence and adulthood were deemed not relevant or measurable in infancy, so item 5 could not be applied to any of the papers. Three papers151,152,157 reported that subsequent outcome measures were rated by blind coders who were not aware of attachment status (item 7). The remainder did not report this information.

Items 8–11 assessed design-specific sources of bias. Items considered baseline differences between participants retained and those lost to follow-up, and whether or not analysis was conducted to compensate. Most of the papers tested for and reported no significant differences at baseline between participants followed up and those lost to attrition (item 8). Only one paper154 provided insufficient information to judge whether or not baseline differences were present. Item 9 asked whether or not any analysis was conducted to compensate for significant differences at baseline. All papers, with the exception of Ammaniti and colleagues,154 were rated lower on this item because they reported no significant baseline differences. Ammaniti and colleagues154 reported insufficient information to permit judgement. Most of the studies had high attrition rates (item 10). Papers were rated as low bias on this item if they reported less than 20% attrition or provided a description of participants lost to follow-up; this applied to only one paper. 156 The study by Dan and colleagues155 was unclear. Although several of the studies did report that there were no significant differences in attachment status between those retained and those lost to attrition, it could not be determined whether or not there were differences in the reasons for attrition between exposed and unexposed participants in any of the papers (item 11).

Item 12 considered the declaration of conflicts of interest and item 13 checked for any other bias. It was unclear whether or not there were any conflicts of interest or funding sources that may cause bias in any of the papers. At least one source of important bias, not covered by the first 12, was found in five of the eight papers. Aikins and colleagues151 received a rating of high bias on this item because prevalence figures were inconsistent with percentages calculated. Jaffari-Bimmel and colleagues156 conducted an analysis to compensate for missing data but did not report the extent of missing data. In addition to this, part, but not all, of the sample received an attachment-based intervention at 5 months old. Ammaniti and colleagues154 did not report results for the full sample on all outcome measures. Two papers received a rating of high bias for inconsistent reporting about the sample. 45,152

Long-term outcomes

Two studies examined whether or not attachment problems continued from infancy to adolescence (Table 12). The two studies measure adolescent attachment at different ages, using different tools. Aikins and colleagues151 reported that there was no continuity from disorganised infant attachment on the SSP to unresolved adolescent attachment on the Adult Attachment Projective Picture System.

Weinfield and colleagues152 reported that infants with disorganised attachment were significantly more likely to be insecurely attached in late adolescence, as measured by the Berkeley AAI,160 and less likely to be autonomous than participants who were organised in their attachment during infancy, with 86% of those who were disorganised in infancy classified as insecure on the AAI. However, of the 42 participants classified as disorganised in infancy, only nine were unresolved in adolescence. The unresolved classification was categorised further into unresolved ‘loss’ and unresolved ‘abuse’. Weinfield and colleagues152 explored the relationship between a disorganised attachment pattern on the SSP in infancy and unresolved loss and unresolved abuse in adolescence, only in those participants who discussed abuse and loss. They suggested that disorganised attachment was related to unresolved abuse but not unresolved loss.

The instruments used here for measuring attachment in later life (Adult Attachment Projective Picture System;151 Berkeley AAI152) measure constructs suggested by their authors as being related to attachment. Validity testing is difficult because attachment is likely to change over a child’s lifetime and be influenced by a range of factors. The tests used may also be assessing related or other constructs.

Table 13 shows the findings on long-term mental health outcomes of ‘severe attachment problems’. Only two papers reported long-term outcomes of 10 years or more in relation to disorganised attachment. Of all outcomes searched for, three findings were found; all were in relation to mental health and all from the Minnesota sample. The findings were that disorganised attachment in infancy was significantly correlated with overall history of psychopathology at age 17 years,45 dissociative experiences at age 19 years45 and borderline personality symptoms at age 28 years. 153 This last association was weak, and lost its significance when included with maternal hostility and other factors in a regression analysis.

These findings were surprisingly limited, and so after submission of the draft report, additional scoping of the literature exploring 5- to 10-year follow-up was conducted. Although this was not comprehensive or systematic (because of very limited time scales), some findings are included in Tables 29 and 30 in Appendix 4 for reference.

Inclusion criteria

The PICOS criteria were as follows:

• Population Parents or caregivers of young children (with a mean age under 13 years) who have severe attachment problems (as defined in Chapter 1) or are at high risk of developing such problems (e.g. children with a history of maltreatment).

• Intervention Interventions were aimed at parents or caregivers, including foster carers. Interventions were excluded if aimed at teachers or teaching assistants (without parents or caregivers) or not focused at an individual level (e.g. organisational structure change in a care setting).

• Comparators No intervention, an attention control or usual care.

• Primary outcome The child’s attachment to the caregiver. Although we felt that it was important to examine the studies identified with a disorganised (D) pattern of attachment or a diagnosis of attachment disorder (e.g. RAD/DAD), we felt that it was also essential to examine the studies that look at the promotion of attachment security. Therefore, we did not specify that it was a necessary requirement for studies to contain a D classification or RAD/DAD diagnosis if attachment pattern was measured.

Measures of attachment were relevant if the measure attempted to assess the nature of the child’s attachment to the caregiver. Only whole attachment instruments or instruments that included a subscale measuring attachment were included. For inclusion in the review, the primary or principal aim of the tool must focus on the measurement of attachment pattern or attachment disorder.

• Secondary outcomes Quality of life; psychological well-being; rates of mental ill health at any age; psychosocial development; educational attainment; entry into care or the criminal justice system; acceptability of the intervention to parents. We sought validated outcome measures in any of these domains. [All secondary outcomes, with the exception of parenting, are related to the outcome for the child and not the parent (i.e. we are interested in the child’s psychological morbidity and not that of the parents).]

• Study design RCTs.

Any paper that did not meet the above criteria but provided additional information to that given by a paper already included was automatically included in the review.

Conducting the quality assessment

The quality assessment was conducted with the risk-of-bias assessment for RCTs using the criteria recommended by the Cochrane Handbook. 174 The recommended approach for assessing risk of bias in studies included in the Cochrane Review is a two-part tool, addressing six specific domains:

• sequence generation and allocation concealment (selection bias)

• blinding of participants and providers (performance bias)

• blinding of outcome assessor (detection bias)

• incomplete outcome data (attrition bias)

• selective outcome reporting (reporting bias)

• other sources of bias.

The first part of the tool involves describing what was reported to have happened in the study. The second part of the tool involves assigning a judgement relating to the risk of bias for that entry, in terms of low, high or unclear risk. All quality assessments were conducted separately by two independent reviewers; any disagreements were resolved by arbitration and a third party when required.

Data extraction

Data were extracted by two independent reviewers and the information was standardised into an Excel^® spreadsheet (2010; Microsoft Corporation, Redmond, WA, USA); reviewers subsequently met to agree the data extraction. This information formed the basis of the tables for the report. For non-English language studies, one reviewer conducted the translation using the specific template provided.

Where there were multiple publications for the same study, data were extracted primarily from the original complete publication. In cases in which the duplicate publications reported additional relevant data, these data were also extracted.

Demographic information

Demographic information was collected on a range of different variables. These included age of parent, age of child, ethnicity and SES.

Data synthesis

The results section is split into two main tables. The first looked at studies that have focused on promoting the security of attachment, without RAD or DAD or those with a disorganised pattern of attachment. The second focussed specifically on studies using a population with a disorganised pattern or a population with a diagnosis of RAD/DAD. This is also replicated in the meta-analysis.

Meta-analysis

For the meta-analysis we explored each dichotomous outcome (secure or disorganised attachment pattern as measured by the Ainsworth scale6 or MacArthur Working Group scale49). We extracted data on the numbers of patients experiencing the outcome for each group. The odds ratio (OR) and 95% CI were calculated for each study outcome. The ORs were pooled using a fixed-effects model or random-effects model [the Mantel–Haenszel (M–H) method] and the corresponding 95% CIs were calculated. Statistical heterogeneity was assessed. This is the variability in the intervention effects being evaluated in the different studies. Statistical heterogeneity manifests itself in the observed intervention effects being more different from each other than one would expect from random error (chance) alone. Where the result indicated significant heterogeneity, a random-effects model was chosen; otherwise, a fixed-effects model was applied.

The data corresponding to the last reported follow-up were chosen. Subgroup analyses were undertaken for the following:

• duration of intervention

• length of follow-up

• number of sessions within the intervention (< 5, 5–16, > 16)

• age of child at the start of the intervention

• foster children involved

• middle class

• intervention location (home, mixed, other)

• male caregiver included

• video feedback provided

• intervention involving a component targeting maternal sensitivity

• primary focus to modify child attachment

• intervention involving caregiver and child together, caregiver and child separately, caregiver alone or mixed.

For continuous data, means, standard deviations (SDs) and sample sizes were extracted for each study. As a range of different outcome measures were used, Cohen’s d effect sizes were calculated to allow comparison between studies, but no formal pooling of data was undertaken.

Characteristics of included intervention studies

Table 14 shows the study, participant and intervention characteristics of included intervention studies without a disorganised classification or diagnosis of RAD. Table 15 shows the study, participant and intervention characteristics of included intervention studies that contained a disorganised classification or diagnosis of RAD.

In total, 30 studies were identified,175,178,180–184,188,189,191,194–196,198,200–203,205–208,211,212,218 29 of which were delivering an intervention in a hypothesised ‘at-risk’ group to improve attachment security129,175–178,180–186,189,191,194–196,198,200–203,205,206,208,211,212,218 and one of which provided treatment for children already with a diagnosis of RAD. 188 The studies were published between 1985 and 2012 and were undertaken in a variety of countries: five in the UK,176,188,189,207,208 14 in the USA,177,180–182,184–186,191,194,196,200,201,205,211 three in the Netherlands,178,198,203 three in Canada,175,195,218 one in Germany,202 one in Australia,130 one in Italy,206 one in Finland212 and one in Lithuania. 183

• Eight studies were identified in the main systematic review that evaluated the intervention using a measure that assessed a disorganised pattern. 189,190,194–196,198,200,218

• Only one study was identified that was focused on a population of children with an attachment disorder diagnosis. 187

• The remaining 21 studies evaluated the interventions with a measure that assessed only secure and insecure attachment patterns without a disorganised group. 129,175–178,180–186,201–209,211

The characteristics of the populations in each study were described and classified into nine categories based on the reporting of authors’ population descriptions and/or inclusion criteria. Figure 6 shows the distribution of the population characteristics across the 30 included studies.

FIGURE 6.

Study population characteristics for studies with and without a disorganised attachment pattern (D) or a diagnosis of RAD/DAD.

The nine population characteristic categories were:

• parent mental health

• low SES

• life events/homelessness

• child behavioural problems/disability/high irritability

• middle class

• poor parenting or parental sensitivity

• single/first-time/adolescent mothers

• low-birthweight/pre-term infants

• foster placement/child welfare/child’s maltreatment history.

Figure 6 shows the study populations for the studies that did and did not contain a disorganised pattern of attachment or a diagnosis of attachment disorder, that is, those which were for children at high risk of severe attachment difficulties. Interestingly, the majority of the disorganised or disorder studies assessed interventions using a population where the parents were at higher risk of raising children with poor developmental outcomes, that is, where the parents either had a mental health problem or were single, first-time or adolescent mothers.

Ten conducted the intervention across a population with low SES, which was the largest category. 175,176,185,191,196,198,200,205,208,211 The other studies were distributed evenly between the other eight population categories.

Specific interventions

The following are brief descriptions of some of the specific interventions included. Studies marked * concern participants with a disorganised pattern or a diagnosis of RAD/DAD.

Interventions which were not included

There were a number of studies that we were aware of that we might have expected to see in a review of this nature. Some of these parental interventions for attachment disorders, or interventions intended to improve attachment security, have not been uncovered by this review but were mentioned by the PPI or expert groups, and are described below. It should be noted that this is not an exhaustive list.

Quality assessment

Table 16 shows the results of the Cochrane quality assessment for included intervention studies without a disorganised category or a diagnosis of RAD. Table 17 shows the results of the Cochrane quality assessment for included intervention studies that contained a disorganised category or diagnosis of RAD. Three domains were consistently rated as high bias across the included studies. These were incomplete outcome data, selective reporting and ‘other’ bias. Incomplete outcome reporting was often rated as high bias because attrition was over 10% across the course of the trial. The reason for the selective reporting item predominantly receiving a rating of high bias across the studies was poor reporting of the secondary outcomes within the studies. Many of the studies received a rating of high in ‘other bias concerns’. There were various reasons for this including unexplained attrition, unexplained missing data, small sample size/low power and inconsistencies within the data.

Unclear reporting, where the author’s descriptions were not sufficient to rate the relevant information, was apparent, with a lack of detail about the random sequence and the method of allocation concealment. Blinding was conducted to some extent in approximately 60% of all trials. The mixed presentation of trial quality across the review suggests that any conclusions should be interpreted with caution.

Of the 34 interventions identified in this phase, 21 were established, named interventions. Many of these consisted of multifaceted treatment programmes with components such as home visits, video feedback, family therapy or sensitivity training for caregivers. In some cases, several studies assessed the same intervention programme, such as the UCLA FDP intervention, which was evaluated by both Heinicke and colleagues190–192 and Beckwith. 201

Of the included interventions, very few involved only the caregiver and not the child. The IPT intervention is one such example that used therapy sessions targeted at external problems regarding the caregiver. 182 The majority of studies included both caregiver and child in the intervention, particularly interventions that involved sensitivity training, video feedback or dyadic play sessions. 183,196,202–204

Alongside these named interventions, several unnamed interventions are identified in this phase. Some of these therapies involved the caregiver’s physical proximity to the child, such as massage therapy185 or utilising baby carriers. 180 Others used similar techniques to the named interventions; for example, many involved home visits from experienced mothers,129,133 volunteer coaches186 or other professionals.

The named and unnamed interventions have several overlapping themes in terms of their content, using similar techniques within the intervention strategies. Some of the more common themes or foci of the interventions included using positive feedback for the caregiver, exploring and changing the parent’s perception of the child, improving parental attachment to the infant and promoting sensitive caregiving.

Meta-analyses findings

Of the 39 papers (30 studies129,175–178,180–186,188,189,191,194–196,198,200–203,205–208,211,212,218) that were included in the main systematic review (see Figure 5), 21 papers had data that could be included in the meta-analysis. 129,133,180,189–200,202–204,208,209,218 The remaining 18 papers could not be included because they did not contain sufficient raw data to be analysed. We initially set out to do three meta-analyses.

1. Only one RCT identified had investigated the effectiveness of an intervention on RAD symptoms;190,191 therefore, we were unable to conduct a meta-analysis for the effectiveness of interventions on attachment disorders.

2. We investigated the effectiveness of the intervention in reducing disorganised patterns of attachment.

3. We investigated the effectiveness of the interventions in promoting a secure attachment as an outcome. This has been included in Appendix 6 to provide additional information.