Pilot study

In two schools, focus groups were conducted with management staff and class teachers following completion of questionnaires, and qualitative information was gathered from parents during home visits. Information and questionnaire packs were translated into Arabic, Urdu and Somali. Based on the estimated prevalence in our pilot study (which we estimated as 1%), we calculated that recruiting 1000 participants for phase 1 in the main study would give us a 95% CI of 0.98-1.98 around our prevalence estimate. We predicted a 30-50% attrition rate for phase 1.

Main study

All remaining schools in one large educational sector of the city were approached, via the Director of Education, with regard to participation and a presentation about the study was given to head teachers. All schools agreed to take part. As no changes in measures followed from the pilot study, we were able to include the pilot schools and our total mainstream school target sample therefore consisted of all 29 schools with a total of 1654 eligible children. Children whose birthdays fell within the 6-8 years age group attending specialist schools (for children with disabilities or emotional and behavioural disorders) (n = 8) or living in out-of-area foster care (n = 27) were also approached (Fig. 1).

Measures

The Strengths and Difficulties Questionnaire (SDQ) is a 25-item parent- and teacher-report screening instrument investigating common child mental health problems. ^{Reference Goodman and Scott8} The SDQ has been well validated against other screening instruments ^{Reference Goodman and Scott8} and against psychiatric diagnosis. ^{Reference Goodman, Ford, Simmons, Gatward and Meltzer9} It contains subscales for emotional problems, conduct problems, hyperactivity, problems with peer relationships and prosocial behaviour.

The Relationship Problems Questionnaire (RPQ) is a 10-item parent- and teacher-report screening instrument for RAD symptoms. ^{Reference Minnis, Reekie, Young, O'Connor, Ronald and Gray5} In a large general population twin sample, the RPQ had good internal consistency (Cronbach's apha (α) = 0.85) and factor analysis identified that six items describe inhibited RAD behaviours and four items describe disinhibited RAD behaviours. ^{Reference Minnis, Reekie, Young, O'Connor, Ronald and Gray5} Children with parent RPQ scores of seven or over were defined as ‘likely cases’ and, hence, were screened as positive. This cut-off point was chosen based on the distribution of RPQ scores in a previous study of over 13 000 7- to 8-year-old children from a general population twin study ^{Reference Minnis, Reekie, Young, O'Connor, Ronald and Gray5} and (unpublished) cluster analysis of these data.

The DAWBA (Development and Wellbeing Assessment) ^{Reference Goodman, Ford, Richards, Gatward and Meltzer10} is an online computerised parent-report interview, which investigates the common emotional, behavioural and hyperactivity disorders, and generates DSM-IV ¹¹ and ICD-10 ¹² diagnoses. It has been validated against clinician diagnosis and used in large-scale population research. ^{Reference Goodman, Ford, Richards, Gatward and Meltzer10}

Fig. 1 Study flow chart.

RPQ, Relationship Problems Questionnaire; RAD, reactive attachment disorder, ASL, additional support for learning.

The Child and Adolescent Psychiatric Assessment, Reactive Attachment Disorder module (CAPA-RAD) is a parental semi-structured interview investigating symptoms of RAD. ^{Reference Minnis, Green, O'Connor, Liew, Glaser and Taylor6} It contains 28-items for parents and its format is based on the CAPA - a well-validated, semi-structured parent-report interview for child psychopathology used in large epidemiological studies. ^{Reference Angold and Costello13}

The Observational Checklist for Reactive Attachment Disorder ^{Reference McLaughlin, Espie and Minnis14} was originally developed for use in a clinic waiting room and explores interactions between the child and stranger(s) on first meeting. It has good internal consistency (Cronbach α = 0.75). In a previous study, it had good specificity but modest sensitivity in detecting children with RAD, ^{Reference Minnis, Green, O'Connor, Liew, Glaser and Taylor6} therefore we have since used it as an adjunct to our parent-report diagnostic measure when making a diagnosis. It was modified, during the pilot study, for use at home, and was completed if the child was at home during the phase 2 assessment and/or at the school assessment.

The Scottish Index of Multiple Deprivation (SIMD) ranks small postcode areas in quintiles according to deprivation from one (most deprived) to five (least deprived). The SIMD is derived from 38 indicators across 7 domains: income, employment, health, education, skills and training, housing, geographic access and crime (www.scotland.gov.uk/Topics/Statistics/SIMD/BackgroundMethodology).

Phase 1

Schools distributed a questionnaire pack to every child in school years primary 2 and 3 (age range 6-8 years). We did not to offer translated questionnaires because so few families returned non-English questionnaires in the pilot (n = 1). The pack consisted of a cover letter from the school, a study information sheet with a form allowing parents to opt out of the study, and the questionnaire booklet containing the SDQ and RPQ. Parents were asked to complete either the questionnaire or the opt-out form and return it to the class teacher in a sealed envelope to be collected by the researcher. Parents who returned a questionnaire were entered into a prize draw to win shopping vouchers.

Researchers contacted non-responders by telephone and offered to post another questionnaire pack to participants, complete the questionnaire over the telephone, or accept a verbal opt out. Each non-responding family was telephoned at least three times, once in the morning, afternoon and evening.

Simultaneously, on a prespecified date, class teachers completed the questionnaires for each child in their class after brief training. Teacher questionnaires relating to children whose parents subsequently returned written opt-out forms were shredded and not included in the analysis (n = 118). Teacher questionnaires were retained for those children whose parents opted in, and those who remained non-responders (n = 1528; 92.8%, Fig. 1). If a school indicated that a family spoke no English, and they did not return a questionnaire, they were removed from the phase 1 sample pool (n = 23).

To establish the representativeness of our questionnaire sample, SIMD scores were obtained for both responders and non-responders. We also compared our study data with population norms for the SDQ (aggregate data in Ford et al: ^{Reference Ford, Goodman and Meltzer15} data disaggregated into relevant age bands kindly provided by Dr Tamsin Ford).

Phase 2

The sampling frame for phase 2 consisted of all children who screened positive (a score of seven or higher) on the RPQ according to either teachers or parents, and a random 1:1 sample of children who scored within normal ranges on the RPQ. We automatically included all children living in foster care (n = 27) in phase 2 because the prevalence of RAD is thought to be considerably higher in this population. ^{Reference Meltzer, Gatward, Corbin, Goodman and Ford3} Once the sampling frame was determined (n = 338), identification numbers were put on a second database and organised into school groups, masking researchers as to whether the child was in the high-scoring or randomly selected low-scoring group.

All potential participants for phase 2 were contacted initially by telephone, then the phase 2 pack, containing an information sheet, opt-in and opt-out form with a prepaid return envelope, was posted. Non-responders were called at least three times spanning morning, afternoon and early evening: 118 parents (35%) required follow-up telephone calls. The researchers then visited the homes of those remaining non-responders, again at least three times per family spanning morning, afternoon and evening. A total of 44 parents (13%) required recruitment visits. If the family was not in, a short letter was left, stating that a researcher had visited, would do so again to find out whether they were interested in taking part and encouraging them to contact us on a number provided to leave a confidential message with the secretary if they did not wish to be contacted again. Eleven families (3%) were uncontactable after six visits, despite confirming address details with the school. If the parent had agreed to take part, the research team were more persistent - in one case visiting the parent 12 times before desisting. Interviews were conducted at home by non-clinical graduate psychologists. Interpreters were used, if required, to complete phase 2 assessments (n = 4). Interviews were clinically rated by senior child psychiatry trainees under the supervision of an experienced consultant child and adolescent psychiatrist, H.M. Any diagnostic dilemmas were discussed in group conferences led by H.M. (n = 42, 25%).

Phase 3

Children were invited to phase 3 if their parents had reported key DSM-IV RAD symptoms on the CAPA-RAD interview performed in phase 2. We did not conduct phase 3 assessments on children who were not thought likely to have RAD because in previous research ^{Reference Minnis, Green, O'Connor, Liew, Glaser and Taylor6} our assessment package identified no false positives in typically developing children.

For disinhibited RAD the key CAPA-RAD questions were:

1. (a) seeking comfort from strangers;

2. (b) indiscriminate friendliness;

3. (c) demanding and attention-seeking behaviour;

4. (d) minimal checking in unfamiliar settings;

5. (e) cuddliness with strangers, asking personal questions of strangers;

6. (f) invading social boundaries.

For inhibited RAD these were:

1. (a) avoids eye contact;

2. (b) frozen watchfulness;

3. (c) hypervigilance;

4. (d) unpredictable reunion responses.

Of the 28 children identified for phase 3, 22 (79%) parents gave consent for their child to be assessed. These 22 children were seen by a researcher in school or, if in foster care out-of-area, in the foster home. Observations of child behaviour were made using the Observational Checklist for Reactive Attachment Disorder. Assessments were conducted in a private room and were videoed for diagnostic purposes.

RAD diagnoses

Diagnoses of RAD were made, based on DSM-IV criteria, by H.M. and the research team, following review of the CAPA-RAD, the teacher RPQ, the Observational Checklist for Reactive Attachment Disorder, comorbid diagnoses (from the DAWBA) and videotaped interaction between the child and researcher (who was a stranger to the child at the assessment visit). We have used this method of diagnosing RAD in previous research and have shown that it is highly sensitive and specific in discriminating children with RAD from typically developing children. ^{Reference Minnis, Green, O'Connor, Liew, Glaser and Taylor6} Where the diagnosis was not absolutely clear, or where we were unable to see the child in school and were relying simply on interview and questionnaire data, the child was given a ‘borderline/suspected’ diagnosis. Both DSM-IV and ICD-10 suggest that RAD should only be diagnosed in the presence of a history of ‘pathogenic care’. We decided that it would be upsetting for participants and reduce response rates if we asked parents direct questions about abuse and neglect of their child, although this was explored in the post-traumatic stress disorder (PTSD) section of the DAWBA, which included abuse as an example of a trauma. Examples of maltreatment reported in this section included physical abuse and witnessing domestic violence.

Statistical methods

Although we had no bias in the selection of schools (because all agreed to take part) and almost full screening data from teachers on our population (teacher SDQ and RPQs), only 50% of our invited sample took part in the second (case-control) phase. It is likely that there was some bias in our sampling at this stage, therefore we wished to estimate the number of cases that would have been ascertained if we had also managed to assess the other 50% of our population. The first step was to establish the sensitivity and specificity of our algorithm for deciding that a child had a likely diagnosis of RAD. We performed binary logistic regression, using the total teacher RPQ score, the total score for core CAPA-RAD disinhibited items and the total score for core CAPA-RAD inhibited items. A receiver operating characteristic (ROC) analysis was performed to determine the optimum sensitivity and specificity when determining RAD caseness. In order to apply the logistic regression model to the entire data-set and identify likely cases of RAD, we imputed CAPA-RAD scores for the missing 50% of our phase 2 sample, based on the variables for which we had virtually full data.

For the multiple imputation, the following variables were identified as having virtually full data: whether or not the child was selected as a case or control for phase 2, gender, SIMD and teacher RPQ. These were used to impute missing values for the core CAPA-RAD inhibited and disinhibited items listed earlier. In total, 100 data-sets were imputed with constraints added to ensure the values predicted in the multiple regression on the predictive variables were within the correct limits. Once the total inhibited and disinhibited CAPA-RAD parameters had been estimated from the individual components, the predictive logistic regression model was used to classify each of the participants as probable cases/controls.