CollAborative care for Screen-Positive EldeRs with major depression (CASPER plus): a multicentred randomised controlled trial of clinical effectiveness and cost-effectiveness

Approvals obtained

This study was approved by NHS Leeds East Research Ethics Committee (REC) on 28 September 2010 (REC reference number 10/H1306/61). Research management and governance approval was obtained for each trial centre thereafter (see Appendix 1). This trial was assigned the International Standard Randomised Controlled Trial Number of ISRCTN45842879.

Trial centres

Four centres in the north of England were selected as trial sites: (1) York centre (the core study centre) covering the city of York, Harrogate, Hull and the surrounding areas; (2) Leeds centre and the surrounding area; (3) Durham centre and the surrounding area; and (4) Newcastle upon Tyne centre, including Northumberland and North Tyneside. Each centre was responsible for co-ordinating the recruitment of participants into the study (trial and epidemiological cohort).

Duration of follow-up

All participants were followed up by questionnaire at 4, 12 and 18 months (see Chapter 4).

Participant eligibility

Sample size

To detect a minimum standard effect size of 0.35 (aligning with the US IMPACT study16 and our previous CASPER trial29,30) with 80% power and a two-sided 5% significance level, 260 patients (130 per arm) would be required. Although this is an individually randomised trial, there may be potential clustering at the level of each collaborative care case manager, and hence the sample size was inflated to account for this. Based upon an estimated intracluster correlation coefficient (ICC) of 0.02 and a case load size of 20, the design effect would be 1.38 {1 + [(20 – 1) × 0.02]} and 360 patients (180 in each arm) would be required. Allowing for 20% loss to follow-up, the final sample size needed was 450 patients (225 per arm).

Epidemiological cohort

During the first year of the CASPER plus trial, an epidemiological cohort was assembled. This consisted of people who had consented to participate in the trial but who were not depressed. Through our broad inclusion criteria we successfully recruited a total of 4668 patients aged ≥ 65 years into the CASPER cohort, from who we identified those with major depression who were eligible to participate in the CASPER plus trial. The reasons for this strategy were twofold: first, to recruit an adequate number of potential participants who would subsequently be identified as having depression, as we believed this would not always be recorded on GP records; and, second, to establish an epidemiological cohort of older adults who could be followed up and who would help inform the knowledge base around the health and well-being of older adults. This type of study design is termed a cohort multiple RCT. 31

Recruitment into the trial

Recruitment of all participants into the trial took place through primary care. GP practices agreed to participate after a member of the study team had introduced it to them with written information, followed by a face-to-face visit to explain the study and what participation would involve. Patients were identified by a computer search and then invited to participate in the CASPER study by their general practice, which posted an invitation pack to all eligible patients. The packs comprised an invitation letter (see Appendix 2) signed from the general practice, a consent form (see Appendix 3), a decline form (see Appendix 4), a participant information sheet (see Appendix 5), a background information sheet (see Appendix 6) and a prepaid return envelope addressed to the core study centre. No patient-identifiable data were available to the study teams until patients returned their consent form.

Consenting participants

During the consent stage, potential participants were asked to complete the Whooley questions,32 a two-item depression-screening/case-finding tool. These questions were asked at two different time points – on the background information sheet at invitation and in the baseline questionnaire – both times as self-reports. At the consent stage, participants were informed about the opportunity of participating in other related studies (e.g. qualitative studies) and were asked to indicate if they agreed to be approached in the future by ticking a box on the consent form. All participants who consented to take part in the CASPER study at this stage became part of the CASPER cohort. Participants did not become part of the CASPER plus trial until they had been subsequently assessed for suitability by completing a standardised diagnostic interview and randomisation.

Baseline assessment

On receipt of written consent from participants by the return of their consent form via post, baseline data were collected through a self-report questionnaire. All participants who returned completed consent forms to the core study centre were sent a baseline questionnaire (see Appendix 7). Participants were asked to respond to the Whooley questions32 for a second time and to provide self-report medication data. They were also asked to complete a range of health surveys, which consisted of the Patient Health Questionnaire-9 items (PHQ-9)33 – a measure of depression severity using a nine-item depression scale in reference to how a respondent has been feeling over the past 2 weeks; the Short Form questionnaire-12 items (SF-12)34 – a measure of health-related quality of life to obtain health-state utility by estimating the Short Form questionnaire-6 Dimensions (SF-6D); the EuroQol-5 Dimensions, 3 levels (EQ-5D-3L)35 – a standardised measure of health-state utility, designed primarily for self-completion by respondents; the Generalised Anxiety Disorder-7 item (GAD-7)36 scale – a severity measure of generalised anxiety used to gauge the past 2 weeks; the Patient Health Questionnaire-15 items (PHQ-15)37 – a measure of somatic complaints using a 15-item scale in reference to the last month; and the Connor–Davidson Resilience Scale-2 items (CD-RISC2)38 – used to measure an individual’s resilience and ability to bounce back.

Randomisation

Randomisation was carried out by the York Trials Unit Randomisation Service [www.yorkrand.com (accessed 23 June 2016)], accessed by a trained researcher from the study team. Participants were automatically randomised by a computer on a 1 : 1 basis by simple unstratified randomisation to either the intervention group or control group, following the completion of a diagnostic interview. All diagnostic interviews were conducted over the telephone by a trained researcher from the study team. The major depressive episode module of the Mini International Neuropsychiatric Interview (MINI) was used to ascertain the presence or absence of core depressive symptoms. 39 The MINI shows good agreement with other semistructured diagnostic interviews conducted to internationally recognised standards. 40–42 This allowed potential recruits to be identified as having major depressive disorder (five or more symptoms), subthreshold depression (two to four symptoms) or no depression (one symptom) (Table 1). 39,43,44 All participants diagnosed with major depressive disorder were randomised to either the intervention or the control arm.

Once participants had been randomised, they were sent a letter informing them of the outcome of their diagnostic interview. If their MINI outcome was major depression, they were informed of their group allocation, either collaborative care or usual care. The participant’s GP was also sent a letter informing them that the named patient was eligible to take part in the CASPER plus trial owing to the major depression outcome of their diagnostic interview. It also specified which arm of the trial they had been randomised to.

Ineligible participants

All participants whose outcome was not major depression (either non-depressed or subthreshold) were sent a letter informing them that they were ineligible for the CASPER plus trial but that they would remain in the CASPER epidemiological cohort and continue to be followed up via questionnaires. Their GPs were also informed of this. This process of following up non-trial participants was discontinued once the original CASPER trial completed (see Chapter 4).

Control group

Participants in the control group were allocated to receive usual GP care. They received no care additional to the usual primary care management of major depression offered by their GP in line with NICE depression guidance as implemented by their GP and local service provision. 7,8

Intervention group

Participants in the intervention group were allocated to receive a low-intensity programme of collaborative care using behavioural activation, designed specifically for those aged ≥ 65 years with major depression. Collaborative care was delivered by a case manager [a primary care mental health worker/Improving Access to Psychological Therapies (IAPT) worker] for an intended 8–10 weeks. This took place alongside participants’ usual GP care. The defining feature of collaborative care is a collaboration of expertise to help support the participant. A case manager works alongside the participant, sharing any relevant information with the GP and a mental health specialist (psychiatrist or psychologist). The case manager is a cohesive link between the participant and other professionals involved in their care. For example, a case manager who deemed a participant’s depressive symptoms to have deteriorated would pass this information on to the participant’s GP, who would optimise the management of the patient’s condition.

Collaborative care in the CASPER plus trial consisted of telephone support, symptom monitoring and active surveillance, facilitated by a computerised Patient Case-Management Information System (PC-MIS) [www.york.ac.uk/healthsciences/pc-mis (accessed 23 June 2016)] and low-intensity psychosocial management (behavioural activation). Participants randomised to the collaborative care intervention group were contacted by a case manager within 1 week of their randomisation to arrange their first collaborative care session. This was carried out face to face, usually at the participant’s home unless an alternative venue was preferred. After this initial meeting, subsequent sessions were carried out on a, more or less, weekly basis by telephone unless the participant had sensory impairments or preferred face-to-face visits. Case managers worked collaboratively with the participants, liaising with GPs and other health professionals involved in their care to discuss issues relating to participants’ mental and physical health, both during routine updates and when any concerns were identified. This included liaising with GPs as necessary to consider reviews of medication, which could relate to depression but also to comorbid physical health problems. It also included discussing with GPs referrals to other services, such as health services (e.g. pain clinics) or engagement with social services. Case managers worked with the participants to identify problems and agree goals for the intervention. They also worked with participants to identify, and subsequently provide, information about other services that may be useful, such as voluntary and statutory sector organisations and services.

Refinement of collaborative care/behavioural activation

The delivery of collaborative care and behavioural activation had been established in working-age adults for whom an appropriate training package and manual already existed. 28 However, these had not been tailored for use with older adults diagnosed with major depression. Before the study began, necessary changes were made to both the training package and manual (detailed in this section) to account for differences that may exist in the older adult population.

Training occurred over 2 days and involved a combination of brief lectures and role-play. Topics covered were the collaborative care approach as applied to older adults, medication management in older adults, behaviour theory and behavioural activation as adapted for older adults.

Participant follow-up

All participants in the CASPER plus trial were followed up with questionnaires at 4 months (see Appendix 9), 12 months (see Appendix 10) and 18 months (see Appendix 11). All post-randomisation questionnaires were posted to participants from the York Trials Unit along with a pre-addressed prepaid envelope. Participants could complete the questionnaires manually and return them by post to York Trials Unit or they could complete the questionnaire online; an instruction sheet explaining how to log on to the CASPER study site and complete the process was included with each questionnaire. Reminder letters were sent by post at 2 weeks to any participants who had not returned their questionnaire. Telephone follow-up by one of the study team’s researchers was conducted for any participants who did not return the reminder questionnaire in order to complete the primary outcome measure (PHQ-9) at the very least.

Withdrawals

Withdrawal could occur at any point during the study at the request of the participant. If a participant indicated that he or she wished to withdraw from the study, a researcher would speak to the participant to clarify to what extent they wished to withdraw: from the intervention, from the follow-up or from all aspects of the study. When withdrawal was only from the intervention, then follow-up data continued to be collected. Data were retained for all participants up to the date of withdrawal, unless they specifically requested for their details to be removed.

Objective data

Once the CASPER plus trial participants of a general practice had completed their follow-up, objective data were collected for each trial participant. Objective data consisted of details on each participant’s prescribed medication and the number of contacts they had with their general practice during their time in the trial. The only exception was for those participants who had withdrawn in full, thereby withdrawing consent to access their medical records. Objective data were collected from general practices via request from the core study centre. A spreadsheet template was e-mailed to the key contact of each general practice that included the identification codes of each trial participant for the practice with prewritten frozen headings: there were no identifiable data. The search dates for each participant were also listed, from the date they were randomised until either the date they completed the study 18 months later or the date that they had died, if that was the case. Data were still collected on participants who had withdrawn from treatment or follow-up, as they had provided us with consent to access their health records for the 18 months that they would have been in the study. The transfer of all objective data via e-mail was approved on the basis that no identifiable data were shared either with the general practice at the request stage or with the core study centre at the stage that objective data were returned.

Suicide protocol

A small but elevated risk of suicide and self-harm was inherent in the study population, all members of which had been identified as having major depression. All participants (both usual care and collaborative care) were subject to usual GP care and GPs were responsible for the day-to-day management of major depression. GPs were accountable for all treatment and management decisions including prescribing of medication, referral and assessment of risk. This arrangement was made clear to all clinicians and general practices that agreed to participate in the study. The pragmatic nature of the CASPER plus trial meant that we did not seek to influence this arrangement. However, we did follow good clinical practice by monitoring for suicide risk during all our encounters with participants. When a patient expressed a risk through thoughts of suicide or self-harm, we followed the study-specific procedure for suicide risk (see Appendix 8).

Patient and public involvement in research

The CASPER plus trial was informed by the involvement of users of mental health services and carers throughout the research period. An advisory group was established in the early stages of study. This consisted of a number of older adults, some of whom had mental health conditions, along with a carer representative. This group provided valuable insights into the relevance and readability of the study documentation. In the future, we plan to engage patient and public involvement in our dissemination strategies to guide on how best to share the findings.

Further studies

Following completion of the CASPER trial, the Self Help At Risk Depression (SHARD) substudy (not described in this report) was introduced to randomise participants identified with subthreshold depression to receive a self-help workbook or usual GP care. Results from the SHARD study will follow.

Primary outcome

The primary end point for the trial was patient-reported depression severity, as measured by the PHQ-933 at 4 months’ follow-up. Each item is scored from 0 to 3; thus, PHQ-9 scores can range from 0 to 27, with higher scores indicating more severe depression. Total scores from 0 (non-depressed) to 27 (severely depressed) were calculated based on the nine PHQ-9 items. These data were collected via self-report on the follow-up questionnaires. Any participants who did not return a completed questionnaire were sent a reminder, and those participants who did not respond were telephoned by one of the study team’s researchers to ask them to complete the PHQ-9 over the telephone. Missing items were replaced with the mean of the remaining items if one or two items were missing.

The PHQ-9 data were collected at baseline and randomisation (at the diagnostic interview), as well as at 4, 12 and 18 months’ follow-up. Scores at baseline and randomisation are reported in Chapter 5, Baseline characteristics. When analyses were adjusted for initial PHQ-9 score, the score at randomisation was used. The primary end point for the CASPER plus trial was at 4 months’ follow-up. At that point, treatment differences in the magnitude of a standard effect size of 0.35 were sought, which is of moderate size for psychological interventions and in line with collaborative care effects observed in other studies. Cohen50 classifies a standard effect size of 0.3 as a small to medium effect size, and this is in line with NICE guidelines for depression, which adopts a similar grading of clinical significance. Four months was selected as the primary end point, because it would occur soon after the end of the planned treatment but allow some additional time in the event that it was not possible to see participants on a weekly basis for practical reasons (e.g. holidays).

Secondary outcomes

The secondary outcome measures used were:

• depression severity and symptomatology at 12 and 18 months (PHQ-9)

• binary depression severity at 4, 12 and 18 months (PHQ-9), using scores of ≥ 10 to designate moderate depression caseness

• quality of life at 4, 12 and 18 months (SF-12 and EQ-5D-3L)

• psychological anxiety at 4, 12 and 18 months (GAD-7)

• mental health medication at 4, 12 and 18 months

• physical health problems at baseline, 4, 12 and 18 months (PHQ-15)

• psychological resilience at baseline, 4, 12 and 18 months (CD-RISC2)

• mortality at 4, 12 and 18 months.

Mental health medication

Medication data were captured by self-report on the follow-up questionnaires. Participants indicated prescribed medication by selecting from a list of 10 antidepressants, as well as listing any other medications they were prescribed.

Mortality data

A data linkage service was established with the NHS Digital to provide regular updates from the Office for National Statistics (ONS) mortality data on any trial participants who had died while in the study. Members of the research team recorded any identified deaths, date and cause of death on the study management database.

Adverse events

The CASPER plus study was not a Clinical Trial of an Investigational Medicinal Product and was, therefore, not subject to any additional restrictions. Decisions regarding prescription of medications were made by the participant in conjunction with their GP: participation in the study had no bearing on this process. Any participants who asked a member of the CASPER plus study team for an opinion on medication issues were strongly encouraged to seek advice from their GP.

The study recorded details of all serious adverse events (SAEs). Any judged to have been related to the study were required to be reported to the REC under the terms of the standard operating procedures for RECs. 51 In the context of the older adult population of the CASPER plus study, many of the SAEs were expected: unscheduled hospitalisations, life-threatening conditions, incapacitating illnesses and deaths. These were not perceived as unexpected events; therefore, they would be reported as SAEs only if they appeared to be related to an aspect of taking part in the study (e.g. participation in treatment, completion of follow-up questionnaires, participation in qualitative substudies or telephone contact).

When a SAE was identified, the trial manager was informed by e-mail using a participant’s trial identification number, and not by any identifiable data. He or she then informed the chief investigator and two members of the Trial Management Group, who jointly decided if the event should be reported to the REC as a SAE. A SAE form was completed and a copy was filed securely at the core study centre. Any unexpected SAEs that were also judged to have been related should have been reported to the main REC within 15 days of the chief investigator becoming aware of the event. In the CASPER plus study, none of the SAEs were judged to have been related to the trial.

The occurrence of adverse events during the trial was monitored by an independent Data Monitoring Ethics Committee and the Trial Steering Committee. The Data Monitoring Ethics Committee/Trial Steering Committee would have seen immediately all SAEs thought to be treatment related.

Data collection schedule

An overview of the time points at which trial data were collected is presented in Table 2.

Statistical assumptions

Participants, care deliverers and the study team were not blinded to treatment allocation. However, allocations were concealed (group A and group B) for interim study reports, for example for the purpose of independent data monitoring reporting. The trial statistician who was responsible for the final statistical analysis was kept blind to group allocation until the primary analysis had been completed.

All analyses were conducted on intention-to-treat basis, using a two-sided statistical significance level of 0.05 unless otherwise stated. A full specification of the statistical analyses is documented in the CASPER plus statistical analysis plan (version 1.0). Any additional data assumptions for data, once received from the York Trials Unit data management team and the CASPER plus trial management team, for the purpose of this report, are documented separately.

Statistical analysis

Secondary analyses

The primary analysis model was repeated (1) including case managers as a random effect to account for clustering within case managers, (2) including additional covariates predictive of PHQ-9 scores at 4 months as identified by univariate regressions, (3) including additional covariates predictive of non-response at 4 months as identified by univariate regressions and (4) using multiply imputed data. Results from the secondary analyses were compared with those from the primary analysis in order to ascertain the robustness of any observed treatment differences.

Direct referral

In order to maximise recruitment in an often difficult to reach group, an additional method of recruitment was introduced. In addition to the original strategy of sending an invitation pack by post to all patients (aged ≥ 65 years) who were identified by computer search as eligible for invitation by the general practice, this was supplemented with direct GP referral at patient consultation. GPs from participating practices were given a number of patient invitation packs (identical to those currently sent by post) that could be handed to patients aged ≥ 65 years who may be consulting about depression and who did not meet the exclusion criteria.

Targeted search

In order to optimise the search strategy, there was a move from an all-inclusive approach to a more targeted approach. The targeted search strategy was developed to include only patients who had a diagnosis of depression or those who were prescribed depression medication and had other conditions associated with an increased risk of depression (e.g. depression, low mood, antidepressant medication, ischaemic heart disease, diabetes, chronic obstructive pulmonary disease, arthritis or being a carer). This was done using the aforementioned Read codes (a coded thesaurus of clinical terms that have been used in the NHS since 1985), the choice of which was left to the discretion of the participating general practice.

Eighteen-month follow-up questionnaire

An 18-month follow-up questionnaire was introduced to obtain longer-term outcomes.

Cohort

In the original protocol, it was stated that all participants who returned screening questionnaires would be followed up at 4 and 12 months via post or online. This included participants both in the RCT and those in the epidemiological cohort. On completion of the CASPER trial of subthreshold depression, this policy was discontinued in order to maximise recruitment and retention to the CASPER plus trial. For the remainder of the trial, only CASPER plus trial participants were followed up. All potential cohort participants who had consented but did not meet the criteria for major depression at diagnostic interview were not followed up.

Recruitment and follow-up

Participants were randomised into the CASPER plus trial between September 2012 and August 2014 from four UK sites and their surrounding areas in the north of England: York, Leeds, Durham and Newcastle upon Tyne. A total of 74 general practices screened their practice lists and identified patients who met the initial inclusion criteria. Exclusion criteria consisted of any known alcohol dependency and/or psychotic symptoms as recorded on GP records, any known adverse comorbidities or any other factors that GPs deemed made it inadvisable to invite patients, such as recent bereavement.

A total of 64,214 patients were identified by GP practices between 5 May 2012 and 10 June 2014 and invited by letter to take part in the CASPER study. Of 10,686 patients who consented, 3224 patients were assessed for eligibility by diagnostic interview. Based on the diagnostic interview, 485 (15%) patients were identified to have a major depressive episode and were randomised into the CASPER plus trial. Of the 485 participants randomised, 249 were allocated to collaborative care and 236 to usual care. The remaining patients were classified as having either below threshold depression (n = 1525) or subthreshold depression (n = 1214). They became part of the epidemiological cohort or were entered into the CASPER or CASPER SHARD trials if within the recruitment window for these trials. The randomised number of 485 participants exceeded that of the planned sample size of 450. The flow of participants is illustrated in Figure 1.

FIGURE 1.

Consolidated Standards of Reporting Trials diagram. a, Reasons for not receiving collaborative care: Carer – no time (n=1), causing marital unrest (n = 2), cognitive impairment (n = 1), did not wish to engage (n = 10), died (n = 2), invasive (n = 6), lost interest (n = 3), not low in mood (n = 1), physical disability (poor hearing) (n = 1), physical ill health (n = 8), receiving other counselling (n = 1), too busy (n = 2), too severely depressed (n = 1) and unable to contact (n = 3). IQR, interquartile range.

Trial withdrawals

Participants were able to withdraw from the study at any point. They were offered the options of withdrawing from the intervention only, from questionnaire follow-up (allowing continued collection of objective data) or from all aspects of the study. Data up to the date of withdrawal were retained for all participants, unless they specifically requested for their details to be removed. This happened on one occasion. The total number of trial withdrawals by trial arm is given in Table 3. Participants could withdraw from only collaborative care treatment but remain in the trial for follow-up purposes. A total of 83 participants (33%) in the collaborative care arm withdrew from treatment at some point, and the numbers of full or partial withdrawals were greater in this arm (n = 55) than in the usual-care group (n = 24).

When reasons for withdrawal were provided by the participant, these were documented in the study management database. Following completion of the trial, reasons were grouped into common categories, and these are listed in Tables 4–6 for the different types of follow-up.

The trial sample size calculation allowed for losses to follow-up of 20% at the primary end point at 4 months. The primary outcome (PHQ-9 depression severity) was available for 390 patients at that point, equating to an actual loss to follow-up of 19.6% (25.3% in the collaborative care arm and 13.6% in the usual-care arm).

Score distribution

Figure 2 illustrates the distribution of PHQ-9 scores for each trial arm over time. At randomisation, scores were distributed approximately normal with a slight right skew, which became more pronounced over the follow-up period, as patients in both arms improved.

FIGURE 2.

Distribution of PHQ-9 scores by trial arm. (a) Randomisation, collaborative care; (b) randomisation, usual care; (c) 4 months’ follow-up, collaborative care; (d) 4 months’ follow-up, usual care; (e) 12 months’ follow-up, collaborative care; (f) 12 months’ follow-up, usual care; (g) 18 months’ follow-up, collaborative care; and (h) 18 months’ follow-up, usual care.

Unadjusted summary statistics

Summary statistics of the raw PHQ-9 scores are given in Table 13 and are illustrated in Figure 3. Average depression severity, as measured by the PHQ-9, was around 14 score points at randomisation. Scores in both treatment arms improved between randomisation and 4 months’ follow-up, but to a greater extent in the collaborative care group (to a score of around 9) than in the usual-care group (to a score of around 11). By 12 and 18 months’ follow-up, average depression scores continued to improve slightly in the usual-care group, whereas scores in the collaborative care group increased again to similar levels.

FIGURE 3.

Unadjusted mean PHQ-9 scores (with 95% CIs).

Adjusting for clustering by case manager

It was expected in the planning and sample size calculation for this trial that collaborative care case managers would work with an average case load of 20 patients, and the clustering of outcomes within case managers was expected to be described by an ICC of 0.02. In total, there were 20 case managers (four in York, four in Leeds, 10 in Durham and two in Newcastle upon Tyne) for a total of 246 participants in the collaborative care arm, that is, an average case load of 12.3 randomised patients per case manager. Case loads varied considerably between 1 and 46 patients. Three patients withdrew before they were assigned a case manager.

The average ICC for clustering within case managers was found to negligible: ICC ≤ 0.0001 (95% CI 0 to 0.0757) for PHQ-9 scores at 4 months.

In order to quantify the impact of the grouping by case managers with respect to the primary outcome, case manager identifiers were included as a random effect in the primary linear mixed-analysis model, nested within treatment arm. Participants in the usual-care arm were coded as their own case managers for the purpose of analysis, and the covariance structure was estimated separately for each treatment arm in order to account for the differences in variability for the random effect.

Adjusted PHQ-9 score means and group differences for this analysis are given in Table 15. Group differences remained significant in favour of collaborative care at 4 months’ follow-up (a difference of 1.92 PHQ-9 score points), and outcomes did not significantly differ between groups at 12 or 18 months. Thus, accounting for the clustering by case manager did not affect the size of the treatment effect compared with the primary analysis.

Adjusting for covariates predictive of Patient Health Questionnaire 9-items at 4 months

The primary analysis was adjusted for PHQ-9 depression at randomisation and baseline physical limitations (SF-12 PCS score). In order to identify any other relevant covariates of depression severity at follow-up, a number of selected demographics and baseline measures were used as predictors of PHQ-9 depression at 4 months in individual regressions followed by a combined regression to avoid issues of multicollinearity, using a non-conservative significance level of p < 0.10 at each stage. All analyses adjusted for PHQ-9 scores at randomisation.

Considered predictors were age, sex, an indicator of whether or not any selected antidepressants had been prescribed at baseline, a history of depression [as measured by two questions of the MINI at randomisation: (1) whether or not patients had ever been consistently depressed for a minimum of 2 weeks and (2) whether or not patients had ever experienced a lack of interest or enjoyment for a minimum of 2 weeks], baseline anxiety (as measured by the GAD-7) and baseline physical functioning (as measured by the PHQ-15).

Results of the individual regressions and summary regression are given in Table 16. Positive coefficients indicate increased depression at 4 months for higher values of the predictor variable (or for the condition specified in the table for categorical variables). Initial identified predictors following individual regressions were prescribed medication, a history of depression (both indicators) as well as baseline GAD-7 and PHQ-15 scores. Higher levels of anxiety, physical health problems, a greater likelihood of being described antidepressants and having a history of depression were associated with higher PHQ-9 depression severity at 4 months. Of these predictors, all but prescribed antidepressants remained significant in a summary regression and were included in the primary analysis model. Age and sex were not significant predictors of PHQ-9 scores.

Adjusted PHQ-9 score means and group differences for the primary analysis model [additionally adjusting for history of depression (two questions), GAD-7 and PHQ-15 at baseline] are given in Table 17. Group differences remained significant in favour of collaborative care at 4 months’ follow-up (a difference of 1.95 PHQ-9 score points), whereas differences at 12 and 18 months remained not statistically significant. Thus, accounting for additional predictors of the primary outcome did not affect treatment differences. An overall comparison of treatment effects for PHQ-9 depression severity from different analyses is presented in Table 18.

Adjusting for missingness

No valid PHQ-9 response at the primary end point of 4 months’ follow-up was available for 25.3% (n = 63) of patients in the collaborative care arm and 13.6% (n = 32) of patients in the usual-care arm. In order to investigate the impact of missing data on the treatment effect, any baseline predictors of non-response at 4 months’ follow-up (no valid PHQ-9 score) were identified by individual and a summary logistic regression using p < 0.10 and included as covariates in the primary analysis model.

Considered predictors were age, sex, an indicator of whether or not any selected antidepressants had been prescribed at baseline, a history of depression (as measured by two questions of the MINI at randomisation), depression at randomisation (PHQ-9 score), baseline mental well-being (SF-12 MCS score), baseline anxiety (GAD-7 score) and baseline physical functioning (PHQ-15 score, SF-12 PCS score).

The results of the individual and summary regressions are presented in Table 19. Odds ratios > 1 indicate a greater likelihood of non-response at 4 months for higher values of the predictor variable (or for the condition specified in the table for categorical variables). The initial identified predictors were age, GAD-7 score, SF-12 MCS score, PHQ-15 score, prescribed antidepressants and ever having lost interest or enjoyment for ≥ 2 weeks. PHQ-9 response at 4 months was more likely to be missing for older participants, participants with greater anxiety, reduced mental functioning or more physical problems, those not on antidepressants and those who reported ever having lost interest or enjoyment for more than 2 weeks. Of these predictors, age, the SF-12 MCS score, PHQ-15 score, antidepressant use and loss of interest remained significant in a summary regression and were included in the primary analysis model.

Adjusted PHQ-9 score means and group differences for the primary analysis model are given in Table 20. Group differences remained significant in favour of collaborative care at 4 months’ follow-up (a difference of 1.97 PHQ-9 score points), and remained not statistically significant at 12 and 18 months’ follow-up. Thus, accounting for predictors of non-response did not affect the treatment effect. An overall comparison of treatment effects for PHQ-9 depression severity from different analyses is presented in Table 18.

In addition, the primary analysis was repeated using complete data derived from multiple imputation by chained equations. Data were imputed from all additional predictors identified in the previous two analyses (age, baseline SF-12, GAD-7 and PHQ-15 scores, antidepressant use and depression history) as well as treatment allocation and available PHQ-9 scores at any time points. Adjusted PHQ-9 score means and group differences for the primary analysis model are given in Table 21 (results based on 20 imputations). Group differences remained significant in favour of collaborative care at 4 months’ follow-up (a difference of 1.93 PHQ-9 score points), and remained not statistically significant at 12 and 18 months’ follow-up. Thus, using complete data did not affect the treatment effect. An overall comparison of treatment effects for PHQ-9 depression severity from different analyses is presented in Table 18.

Summary of Patient Health Questionnaire-9 items analysis models

Table 18 provides an overview of group means and treatment effect estimates from the primary analysis and secondary analyses of depression severity at 4, 12 and 18 months as measured by PHQ-9 scores. Unadjusted means are presented for reference. Adjusted average estimates of group differences at the primary end point at 4 months ranged from 1.92 to 1.97 PHQ-9 score points in favour of collaborative care.

Binary Patient Health Questionnaire 9-items outcome

Using the cut-off point of ≥ 10 PHQ-9 score points, Table 22 presents the number and percentage of moderately to severely depressed participants at randomisation and follow-up by treatment arm. The figures are illustrated in Figure 4. Approximately 77% of randomised CASPER plus participants were depressed at randomisation. At 4 months’ follow-up, this percentage improved to 40% in the collaborative care arm and 55% in the usual-care arm. This difference was not maintained at 12 or 18 months’ follow-up.

FIGURE 4.

Unadjusted per cent of patients (with 95% CIs) with moderate to severe depression.

Data were analysed by logistic mixed-effects modelling, including moderate to severe PHQ-9 depression (yes or no) at 4, 12 and 18 months as the outcome, predicted by trial arm, time (4, 12 or 18 months), group by time interaction, depression severity at randomisation (PHQ-9 score) and baseline physical functioning (SF-12 PCS score). Resulting treatment effect estimates are presented in Table 23.

The greater reduction in moderately to severely depressed cases seen in the collaborative care arm compared with the usual-care arm was statistically significant at 4 months’ follow-up (odds ratio 2.18, 95% CI 1.36 to 3.51; p < 0.001), but was not statistically significant at 12 or 18 months.

Antidepressants

Patients indicated on questionnaires whether or not they were currently prescribed any of a list of 10 antidepressants (see Table 24 for details and the frequencies of prescriptions by trial arm). Citalopram was the most commonly prescribed antidepressant.

A binary variable was created to indicate whether or not patients had been prescribed any of the listed antidepressants. Table 25 presents the number and percentage of patients on antidepressants at baseline and follow-up by treatment arm. The figures are illustrated in Figure 5. Approximately 33% of patients were prescribed antidepressants at baseline. Of the participants remaining in the trial at follow-up, antidepressants were prescribed to a greater percentage at 4, 12 and 18 months than at baseline, around 36% on average. Differences between treatment arms were small.

FIGURE 5.

Unadjusted per cent of patients (with 95% CIs) who were prescribed antidepressants.

Data were analysed by a logistic mixed-effects model, including prescribed medication (yes or no) at 4, 12 and 18 months as the outcome, and were predicted by trial arm, time (4, 12 and 18 months), group by time interaction and prescribed antidepressants at baseline. Treatment effect estimates are presented in Table 26.

The adjusted relative odds of being prescribed antidepressants were higher in the collaborative care arm than the usual-care arm at 4 months (odds ratio 0.39, 95% CI 0.17 to 0.89; p = 0.025); however, differences between treatment arms were not statistically significant at 12 or 18 months’ follow-up.

Generalised Anxiety Disorder-7 item scale psychological anxiety

The GAD-7 is a brief measure of symptoms of anxiety with a score range of 0–21, with higher scores indicating more severe anxiety. Unadjusted means for psychological anxiety based on the GAD-7 are presented in Table 27 and Figure 6, and the results of the formal statistical analysis by mixed modelling are given in Table 28.

FIGURE 6.

Unadjusted mean GAD-7 scores (with 95% CIs).

The figures indicate that anxiety was on average at around nine score points for all participants at baseline. Both trial arms improved anxiety levels at 4 months’ follow-up, significantly more so in the collaborative care arm (mean score difference 1.68, 95% CI 0.77 to 2.59; p < 0.001). Group differences decreased in magnitude but remained statistically significant at 12 months’ follow-up (mean score difference 1.09, 95% CI 0.14 to 2.03; p = 0.024), but not at 18 months.

Short Form questionnaire-12 items physical component summary score

The SF-12 PCS score ranges from 0 to 100, with higher scores indicating better functioning. Unadjusted means for physical functioning are presented in Table 29 and Figure 7, and the results of the formal statistical analysis by mixed modelling are given in Table 30.

FIGURE 7.

Unadjusted mean SF-12 PCS scores (with 95% CIs).

The figures indicate that physical functioning was below average adult physical health status (scores of < 50) for participants throughout the trial period, as would be expected in an elderly population. Patients maintained similar functioning levels between baseline and 18 months, and group differences in physical functioning were not statistically significant based on the mixed-effects analysis at 4 months (mean score difference –0.44, 95% CI –2.00 to 1.23; p = 0.583) or any other follow-up.

Short Form questionnaire-12 items mental component summary score

The SF-12 MCS scores range from 0 to 100, with higher scores indicating better functioning. Unadjusted means for psychological functioning are presented in Table 31 and Figure 8 and the results of the formal statistical analysis by mixed modelling are given in Table 32.

FIGURE 8.

Unadjusted mean SF-12 MCS scores (with 95% CIs).

The figures indicate that participants’ average mental functioning was below the general average for adults (scores of < 50) throughout the trial period, which may be expected in a population with major depressive episodes. At 4 months’ follow-up, mental functioning had improved in patients in both arms, but to a greater extent in the collaborative care arm, and this difference between arms was statistically significant (mean score difference –3.02, 95% CI –5.04 to –0.99; p = 0.004). Participants in both trial arms maintained a score of around 40 at 12 and 18 months’ follow-up, and group differences were not statistically significant at these time points.

EuroQol-5 Dimensions, 3 levels

Quality of life using the EQ-5D-3L is measured on five dimensions – mobility, self-care, usual activities, pain/discomfort and anxiety/depression – and participants are given three response options to indicate their level of problems for each dimension. The weighted summary index derived from these dimensions is summarised and formally analysed as part of the CASPER plus health economic evaluation. For the purpose of exploring differences in quality of life between treatment arms, the frequencies of responses for each category in each dimension are presented descriptively in Table 33 and illustrated in Figures 9–13.

FIGURE 9.

The EQ-5D-3L mobility dimension: per cent of patients in each severity category.

FIGURE 10.

The EQ-5D-3L self-care dimension: per cent of patients in each severity category.

FIGURE 11.

The EQ-5D-3L usual activities dimension: per cent of patients in each severity category.

FIGURE 12.

The EQ-5D-3L pain/discomfort dimension: per cent of patients in each severity category.

FIGURE 13.

The EQ-5D-3L anxiety/depression dimension: per cent of patients in each severity category.

The majority of CASPER plus participants indicated no problems or some problems in each of the EQ-5D-3L areas, with few patients having severe difficulties. The most frequent use of the severe category was in the pain/discomfort dimension (around one-quarter of participants). The greatest trial arm differences were seen for usual activities, with the number of patients who had no problems performing usual activities increasing from 40% to 49% at 4 months’ follow-up in the collaborative care arm, whereas rates remained stable in the usual-care arm. This difference was maintained at 12 months but not at 18 months. Relatively greater improvements in favour of the intervention arm were also seen for anxiety and depression at 4 and 12 months, the number of people not anxious or depressed being higher in the collaborative care arm, although group differences were of moderate magnitude (around 6%). There were no substantial group differences in the mobility, self-care or pain/discomfort dimensions.

Patient Health Questionnaire-15 items physical health problems

The PHQ-15 is a measure of physical health problems. In this study it had a score range of 0–28 (usual maximum is 30), as a question regarding menstrual problems was removed for the elderly CASPER plus patient population.

Unadjusted means for physical health problems are presented in Table 34 and Figure 14, and the results of the formal statistical analysis by mixed modelling are given in Table 35. Physical health problems significantly decreased in the collaborative care arm at 4 months’ follow-up; in contrast, in the usual-care group, symptoms remained constant throughout follow-up (mean score difference 1.67, 95% CI 0.98 to 2.36; p < 0.001). This difference became smaller but remained statistically significant at 12 months (mean score difference 1.19, 95% CI 0.47 to 1.90; p = 0.001), whereas follow-up scores returned to near baseline levels for both groups at 18 months.

FIGURE 14.

Unadjusted mean PHQ-15 scores (with 95% CIs).

Connor–Davidson Resilience Scale-2 items resilience

The two-item CD-RISC2 resilience measure has a score range of 0 to 8, with a higher score indicating greater psychological resilience. Unadjusted means for psychological resilience are presented in Table 36 and Figure 15, and the results of the formal statistical analysis by mixed modelling are given in Table 37. Average resilience at baseline was around 5 score points and remained consistent over the 18 months’ follow-up for patients in the usual-care group. Among patients in the collaborative care group, average resilience marginally improved but dropped back to baseline levels at 18 months. The group difference was statistically significant at 12 months’ follow-up (mean score difference –0.35, 95% CI –0.68 to –0.03; p = 0.034).

FIGURE 15.

Unadjusted mean CD-RISC2 scores (with 95% CIs).

Mortality

A total of 13 participants died during the 18-month follow-up period, six patients in the collaborative care arm (2.4% of randomised patients) and seven patients in the usual-care arm (3.0% of randomised patients). Causes of death are summarised in Table 40. A chi-squared test revealed that the difference in mortality rates between treatment arms was statistically significant (χ²₁ = 0.14; p = 0.705).

All deaths were further recorded as SAEs, and potential relatedness to the trial treatment was assessed as part of the adverse event processing. In total, 92% (12 events) of deaths were categorised as being unrelated to treatment, and 8% (one event) as unlikely to be related to treatment.

Collaborative care: required resources and associated costs

Case managers were psychological well-being practitioners (PWPs) employed at NHS band 5. Case managers each received training to provide collaborative care as part of the CASPER plus trial. In total, three training events were held covering four regions of the study (York, Leeds, Durham and Newcastle upon Tyne), each consisting of 2 consecutive days of training. The number of attendees per training event varied and efforts were made to provide training in a manner that ensured that the overall costs of travel and accommodation were minimal.

During the training, PWPs were orientated to the case managers’ manual, which outlined the overall principles of collaborative care and a ‘session-by-session overview’ of what case managers aimed to achieve with patients. The training courses for case managers were predominantly provided by two trainers; subsequently, these trainers also supervised case managers during implantation of the collaborative care programme implementation.

The manual stipulated that the programme of treatment should consist of ‘8–10 mainly telephone contacts with occasional face-to-face contacts over a period of 12 weeks’. In terms of the expectation for each session, it further stated that ‘contacts last 45 minutes for session one and 20–30 minutes for each subsequent contact’. The first session was generally held face to face and took place at participants’ homes, GP surgeries or other community venues.

Case managers received weekly supervision from a designated supervisor. The schedule of supervision followed a standardised agenda whereby for each patient there was a weekly discussion and case managers would prepare feedback to discuss each case with their supervisor. Supervisors were responsible for providing support to case managers on the process of collaborative care and medication management and on specific psychological interventions. On average, each patient contact was discussed between the case manager and supervisor for approximately 5 minutes.

Case managers provided participant-specific feedback to GPs. In the first instance, case managers worked with and advised participants’ GPs on their care. During treatment, case managers would provide a letter to update the GP on participants’ progress and, when appropriate, whether or not GPs might consider further treatment. At the end of the programme, case managers also sent a participant-specific summary report to the GP. Supervisors were available to advise case managers on next steps and consultation with GPs. Three letters were prepared and sent over the 12 weeks, requiring approximately 30 minutes of administration per letter. Case managers would also speak to GPs directly if they had any concerns about a participant’s medication or overall well-being.

Case managers were also charged with a duty of care to engage outside agencies (such as social services or in response to safeguarding issues) in situations in which they became aware of safety or risk (including abuse). However, the client group had a generally low level of clinical need in this respect, and this additional responsibility was not generally required.

To estimate the personnel costs required to provide collaborative care (as intended within the manual), estimates of NHS unit costs were derived from national reference costs56 (Table 41).

Table 42 summarises the resources required over the 12-week programme of collaborative care and indicates our estimate of the direct cost for base-case analysis. The direct cost of collaborative care (based on the prior estimation within the manual) was calculated to be £494.73. This cost is adopted for the base-case cost as, ex ante, there is insufficient information to anticipate actual levels of required care; however, deviation that did occur will be explored within our sensitivity analyses.

Consequences for health care by trial arm

Patient contacts over the duration of the trial are presented in Table 43, which compares the summary statistics for those who accessed collaborative care with the summary statistics for those who accessed usual care. Initial observation suggests that collaborative care in depression results in a small marginal increase in contacts with most services (except GP home visits). However, the mean contact rate with any service is dependent on access to the service and the subsequent level of utilisation.

To test whether or not differences in service use may be attributed to collaborative care, statistical tests must accommodate highly skewed distributions with significant numbers of zero service users and, therefore, specific analytical procedures are required. 60 Applying zero-inflated negative binomial regression61 allows inference on the effect of collaborative care on two factors: access (using the logistic model) and overall change in the contact rate (using the full model). For full regression outputs see Appendix 12.

Having any access to services is indicated by outputs of the logistic models. Across all five resource use categories we may conclude that participants are generally unlikely to access any services. Examining logistic regression outputs related to nurse appointment (see Appendix 12, Table 55) suggest that collaborative care may increase the likelihood of access (log odd = 14.1944; p = 0.01). However, small sample numbers available from this trial mean that inferences regarding the effect of collaborative care should be made with caution.

The full model specification accounts for access and subsequent use to test any overall change in the contact rate. Over resource use categories, there is generally no significant difference between groups. However, inference of the effect of collaborative care on nurse telephone consultations suggests an overall increase in the contact rate of 2.25 (95% CI 0.9285 to 5.4403; p = 0.073). Again, given the sample size, inferences on the effect of collaborative care should be made with caution.

Cost–consequences and total costs

Unit costs (as presented in Table 41) were multiplied by resource utilisation to derive patient-level costs of health care (Table 44). Health-care costs of treatment therefore extend beyond the cost of the collaborative care programme (£494.73), increasing wider costs by a mean of £682.27. Overall, the mean total cost in the collaborative care group was £1171.45 (95% CI £1166.99 to £1175.92, n = 226), compared with £654.14 (95% CI £650.78 to £657.52, n = 221) in the usual-care group.

Health-state utility by time point

Utility scores for each participant were estimated from the responses to the SF-6D at baseline and at 4, 12 and 18 months. Table 45 presents a summary of the unadjusted utility scores by time point and trial arm across all available respondents at each time point.

However, as we can observe, the available sample number by group and across time points declines as the study progresses. For the purpose of illustration, health-state utilities were estimated using a linear-mixed model and estimate group marginal effect for the mean for each time point; Figure 16 plots the outputs and illustrates trends in estimated utilities by trial arm over the trial period.

FIGURE 16.

Plot of mean (95% CI) of SF-6D indexes over the trial period, by trial arm.

Observing differences in baseline utility scores by trial arms suggests that control for baseline utility to estimate cost-effectiveness is important.

Sensitivity analysis: fidelity to intervention sessions and ex post adjustment of the expected direct cost of collaborative care

This analysis seeks to examine documented fidelity of participants to treatment (as observed from data collected using PC-MIS) and to consider how this may adjust our expectation of the cost of implementing collaborative care. Figure 20 summarises distribution in the number of contacts. The total number of registered sessions shows significant variation, and a bimodal distribution of participant sessions appears evident. This raises the question of whether or not there exists a selection process in early sessions (by consumer, provider or both) that divides patients into two groups. For example, 128 of participants (51%) received five or fewer sessions in the early stage of care and the remaining 49% were most likely to receive 10 sessions.

FIGURE 20.

Number of sessions of collaborative care. CC, collaborative care.

To examine if health status explains the number of received sessions, Table 48 presents baseline scores to PHQ-9, GAD-7 and SF-6D contingent on whether or not participants received more or fewer than five sessions. With respect to all three measures, this suggests that the group that received six or more sessions, on average, had poorer health status at baseline. However, reference to 95% CIs would suggest that the between-group difference is not significant.

The next question is, therefore, whether or not the number of sessions is influential on the treatment effect. Figure 21 illustrates the mean (95% CI) of SF-6D indexes over the trial period comparing usual care with treatment. The trial arms are subdivided into participants who received five or fewer sessions of collaborative care and those who received more than six sessions of care. This provides a clear illustration that a dose–response relation is likely to exist between the number of sessions received and generic health status. Fidelity to, and engagement with, the treatment programme appears to be an important feature in threshold depression.

FIGURE 21.

Mean (95% CI) of SF-6D indexes over the trial period comparing the usual-care group with trial arms that received either five or fewer sessions of collaborative care or more than six sessions of collaborative care. CC, collaborative care.

Table 49 calculates the adjusted direct costs of collaborative care using data from PC-MIS based on session from 174 trial participants. This suggests that, on average, collaborative care received by participants cost £198.25 (95% CI £196.16 to £200.35). Given the available information on health gains, it is difficult to determine how this should be interpreted compared with the expected ex ante cost of £494.73. One interpretation is that, in practice, collaborative care cost £296.48 less than expected.

To examine the value underlying the observed rate of fidelity to treatment, Figure 22 presents an adjusted CEAC (with ICER) to examine whether, with an intention-to-treat perspective, collaborative care for threshold depression represents value for money or not.

FIGURE 22.

Cost-effectiveness acceptability curve (controlling for baseline utility) using ex post estimate of the direct costs of collaborative care.

Firstly, given the ratio of average treatment effect to the adjusted cost of collaborative care, an ICER of £10,216 per incremental QALY can be estimated. Costs and QALYs can also be examined by subgroup (i.e. contingent on whether or not participants received more than five sessions of collaborative care). Table 50 performs a seemingly unrelated regression to inform the cost-effectiveness analysis related to these subgroups.

The results indicate that receiving five or fewer sessions of collaborative care is associated with an average cost of £12 (95% CI –£120 to £143) and results in an average QALY gain of –0.0004 (95% CI –0.0517 to 0.0509); therefore, this strategy is dominated by usual care.

We can also determine that the overall cost of receiving six or more sessions of collaborative care is associated with an average cost of £307 (95% CI £193 to £421.93; p < 0.001) and an average QALY gain of 0.0311 (95% CI –0.01375 to 0.0760; p = 0.174). Although the statistical significance of the difference in QALY gain is low, despite the reduction in sample size, it remains higher than the average treatment effect presented in Table 47. Overall, this suggests that, for individuals who receive six or more sessions of collaborative care, the ICER will be £9876 per QALY.

Figure 23 presents confidence ellipses on the cost-effectiveness plane for each subgroup and clearly illustrates that collaborative care requires a strict minimum number of sessions (i.e. six). Examining the CEAC, we can observe that (for session numbers greater than six) the probability that collaborative care is cost-effective is significantly higher over the explicit reimbursement range (£20,000–30,000 per QALY). These findings suggest that collaborative care may be cost-effective with improved fidelity and that further research to better understand reasons why certain participants do not adhere to the treatment programme (e.g. patient preferences or supply-side competing priorities) is required.

FIGURE 23.

(a) Confidence ellipses (comparing collaborative care with more or fewer than six sessions: session number ≤ 5 vs. ≥ 6) and (b) CEAC (for collaborative care with six or more sessions: session number ≥ 6).

Ethics approvals

Ethics approval for the RCT and this qualitative study was gained by Leeds East Research Ethics Committee, Yorkshire & Humber (reference number 10/H1306/61).

Design

We conducted semistructured interviews with trial participants, case managers and GPs to gather in-depth information on their views and experiences of receiving and delivering the intervention and how they perceived the acceptability, engagement and implementation of patient and collaborative care, respectively. Interviews were conducted with trial participants at the end of the intervention period and with case managers delivering the intervention and patient GPs during the intervention.

Sampling

Our aim was to interview a purposive sample of GPs and trial participants, including some participants who declined to take part or who withdrew from the intervention, alongside all the case managers who delivered the intervention. Our approach was to sample participants and GPs from recruiting practices in both urban and rural areas in the north of England and to gather data from areas of differing deprivation indices to achieve a spread in sex, age and socioeconomic status (SES). We aimed to interview all 12 case managers and supervisors and up to 15–20 GPs (or until category saturation was achieved) along with 7–10 participants who did not engage in the intervention and 15 participants who completed the intervention (or until category saturation achieved).

Initially, as numbers were small and recruitment to the trial was slow, we invited all participants who had completed the intervention to take part in a semistructured interview of up to 1 hour. All case managers were invited to be interviewed once they had delivered a course of treatment to at least three participants, and GPs from practices with at least five participants from the collaborative care arm of the trial were invited to be interviewed. Once we had recruited approximately half of our participants this way, we then used a purposive sampling strategy with the aim of gaining a more varied sample of patient and GP participants.

At the start of the CASPER plus qualitative study, following the order of GP practice recruitment, all participants invited to take part were from the central site of York, which included urban and rural practices in the surrounding areas from Harrogate to Hull. Given that most of these areas are of relatively low to moderate deprivation, we used an active selection process to ensure some participants from areas of higher deprivation were invited to be interviewed, such as inner-city Hull.

Participants were sent an invitation pack by post, which comprised a letter, an information leaflet and a consent form with a pre-paid envelope to return to the research team. GPs and case managers were sent an invitation letter, information leaflet and a consent form by e-mail. Before interviews commenced, written informed consent was obtained from all participants (see Appendices 13–15).

Data collection

Interviews were carried out by Karen Overend, Katherine Bosanquet and Sarah Nutbrown in a place convenient to the participant and lasted between 20 and 60 minutes. The majority of GPs chose to be interviewed at their practice, although 5 out of 12 asked to be interviewed by telephone. Ten out of the 12 participants requested to be interviewed in their home, with the remaining two choosing to be interviewed by telephone. Nearly all case managers were interviewed in the researcher’s office, with three opting for a telephone interview. Interviews were conducted between May 2013 and November 2014. All interviews were digitally recorded (with participants’ signed consent), transcribed verbatim and anonymised before data analysis.

A topic guide was developed for each of the three groups (see Appendices 16–18). The topic guides were designed with reference to the literature, approved by the research team and developed iteratively as data collection commenced.

Consent

In accordance with ethics guidelines, informed consent was gained by the researcher from each study participant before the interview commenced. An information sheet was sent to the participant in advance, as part of the invitation pack. Before starting the interview, the researcher (interviewer) ensured that this was signed by the participant (interviewee), repeated the main points of the information sheet and aim of the study and gave the participant an opportunity to ask any questions. The researcher assured the participant of the anonymity and confidentiality of their personal information. GPs and case managers were also given the opportunity to ask questions about the study and were assured anonymity and confidentiality. Consent was obtained from GPs and case managers using the same process as for trial participants.

Data analysis

The interview transcriptions formed the data, through the use of thematic analysis and principles of constant comparison. 64 This was developed iteratively and the topic guides were modified as analysis progressed. The main qualitative researcher on the project (KO) worked closely with the data to identify descriptive coding; this was informed by regular discussion with qualitative supervisor (CC-G) and co-researcher (KB). Analysis was undertaken by individual researchers Karen Overend, Carolyn Chew-Graham and Katherine Bosanquet. Data analysis involved a process of organising the data, descriptive coding, interpretive coding, writing and theorising. Deviant cases were actively sought throughout the analysis and emerging ideas and themes modified in response. Following analysis by individual researchers, themes were agreed during discussion with the full research team.

Revealing hidden depression

For most of the older people we interviewed, being invited to participate in the CASPER plus study seemed to raise their awareness of low mood:

It crept up on me really, how I felt. I think it had been coming on for a long time and I didn’t realise how bad I’d got until I filled that form in and I just ticked the boxes and posted it.

Participant (PT)6

Several GPs described how taking part in the CASPER plus trial helped to raise awareness of depression in their older population. One GP said:

I think it’s probably alerted us to one or two of the . . . more needy participants who perhaps were not coming to us for help . . . people have been brought into the system that . . . had sort of dropped out from seeing the GP.

GP3

Some case managers described how some participants admitted they had not spoken to others, including their GPs, about how low they felt:

One gentleman that I saw, he said the most useful thing had been the diagnostics, as risk was identified, and so we wrote to the GP about that. And it was . . . the risk was still there when I saw him for the first time so I put that in a letter as well and he said that had kind of opened the door. He would have never gone and spoken to his GP about it.

Case manager (CM)2

. . . they [the patient] wouldn’t do anything and they wouldn’t commit suicide but they feel ashamed I guess of having some thoughts [that they’d be better off dead] . . . and those are the sorts of things they don’t always like us to share with the GP because it’s back to that stigma, isn’t it?

CM1

Although some patient participants did not use labels such as ‘depression’ or ‘low mood’, those who did suggested that other older people may fail to recognise or admit their feelings because of the perceived stigma of doing so:

. . . people don’t talk about it do they, they think it’s a weakness don’t they? But it is something that you can’t help when you are in it, you know as I say you don’t realise you are going in it and as much as you try you know sometimes you can’t get out it, it gets deeper you know.

PT6

A few patient participants commented on the ‘invisibility of depression’:

. . . you know if I broke an arm I’d get a sling wouldn’t I, you know it’s fairly obvious, but I suppose with any mental illness you can’t see it, you don’t know.

Withdrawn participant (PTW)1

Several GPs reported an awareness of the stigma associated with depression, especially in this age group, that may impact on whether or not the patient would raise it within a consultation:

It’s sort of an age group where they’re not as open about depression as maybe younger people are, there’s a bit of a stigma attached to it still.

GP8

A few GPs described how they normalised depression in older people; one admitted possibly colluding with the patient in ignoring cues within the primary care:

You’re sort of aware there are people who have depressive episodes that aren’t possibly addressed, they may themselves not really recognise it, and they just think it’s part of, you know, getting older.

GP3

You’d like to think that primary care is fairly aware of it [depression] anyway. But maybe the temptation is to let sleeping dogs lie, I don’t know. So you know, if you diagnose someone with depression you’ve got to do something about it haven’t you?

GP6

Some GPs described a tension between a desire to consider the ‘whole’ patient and, owing to limited time and treatment options, a tendency to prescribe antidepressants to older people:

We often go down a medication route because, well it does help them, and it’s very difficult to get other services. And the psychiatry for the elderly tends to be more focused on dementia.

GP8

Several GPs recognised that depression in older people often materialises alongside complex physical conditions or social problems, including loneliness. Some of these GPs disclosed a reluctance to identify the condition, partly because of the absence of a psychological treatment pathway for depression in the over-sixty-fives and a tendency to prioritise physical symptoms over emotional health:

I suppose in a busy clinic we probably don’t have time to sort of delve into depression along with the sort of 12 and a half minutes of consulting on chronic diseases that’s squeezed into 10 minutes, so depression would take another 5 or 6, so . . . we’ll probably skip over that unless they bring it to us.

GP12

Being invited to participate in the CASPER plus trial provided an opportunity for some people to talk about depression, enabling them to recognise and seek help for low mood.

Reducing the ‘blind spots’

Several case managers and three GPs described how two practitioners working with a patient helped to reduce the ‘blind spots’, as each professional offered a different perspective:

So you’ve got the benefit of somebody who’s looking at a person, never having met them before who can see certain things, versus somebody who has known somebody for some time and can see certain things but, those two people, will have, probably have, blind spots . . . because one person doesn’t know that person very well and the other has maybe, over the years, has just sort of formed a fixed idea about somebody. Collaborative working, not only will it progress the patient forward but it will also . . . reduce blind spots, I think, in their care.

GP1

One GP saw the case manager as helping to ‘patch up’ the gaps in the patient’s support network:

I think a lot of the difficulty . . . is their support networks have become a bit more fragmented . . . especially those that are bereaved, or have families spread around the country or spread around the world . . . so I can see that maybe we can patch that fragmentation up a little bit . . . it’s not the same as having your relatives but having some kind of support, I can see that as a benefit.

GP3

The case managers viewed their role as a facilitator, or ‘go-between’, who is able to convey information to the GP that the patient may be reluctant to disclose directly:

Sometimes, if people can’t talk to their GP or don’t understand that maybe they had a problem like depression, and don’t know how to approach a GP because of stigma and things like that then I’ve been that facilitator, I’ve helped them with that process.

CM1

For example, one case manager reported advocating on behalf of a patient who was having problems with pain:

. . . she was using cannabis to manage the pain and she felt there was nothing else the doctors could do, so I spoke to her GP and they said she could get a referral to the pain clinic . . . She [the patient] had given up all hope, but she was happy for me to pester them a little bit.

CM3

The GPs and case managers offered different perspectives on participants’ health needs, which was seen to reduce ‘blind spots’ in depression care.

An opportunity to talk

Being offered an opportunity to talk outside the GP consulting room was valued by the majority of participants:

The most startling thing about the experience was all my life I’ve never had anybody to talk to, there’re things I wouldn’t even discuss with my wife and to have an outsider person that didn’t really know me who was impartial . . . that helped me a great deal, just by having someone to discuss things with.

PT5

. . . having someone to talk to . . . about things in my life that I would talk to say the family about or friends unless they were extremely close friends, it gave me someone objective to talk to you know, that was removed from my situation.

PT2

Some participants suggested that GPs were not always receptive to discussing problems with mood:

You know and the GPs, well they don’t, they don’t seem to be interested I don’t think. Oh, it’s depression, take a pill, go away.

PT12

I just have a bit of a problem with doctors because I just don’t think they do the job that they maybe should be doing, it’s a 2-minute interview or whatever, they don’t really know your records, they don’t know the history, they don’t tie things up.

PTW1

In contrast, most participants described the case manager as providing empathic support, being able to offer more time than the GP and knowing how to direct participants to voluntary organisations:

. . . she did everything she possibly could . . . I mean she went the extra mile. She spoke to the people at Parkinson’s – Parkinson’s UK – to see if there was a network somewhere, an advice centre, and things I didn’t know she found out for me.

PT7

Patient participants spoke about the benefit of having someone to talk to in confidence, outside the primary care consultation, who was said to listen without judging, allowing them to talk openly about feelings and personal issues:

I thought it was very good. And I think the fact that people were bothered, to see how the older people felt . . . I think that was good. You didn’t feel like you just got a script thrown at you and you were waiting for God sort of thing . . . it was the fact that someone was interested in how you felt.

PT1

Giving participants an opportunity to talk outside the clinical setting of the primary care consultation room appears to be valued by most of the older people we interviewed and by their GPs.

‘Moving on’ from depression

Some participants reported how the case manager encouraged them to increase activity and social contact, which the participants felt had improved both their physical health and mood. For example:

The telephone conversations for me were helpful. She got me to think about doing things. I’m doing a computer course now and there’s a chance I might be able to help them at [voluntary organisation].

PT9

It has helped me thinking about things I can do . . . I go in the pool, only in the baby pool but it’s good for my legs and my shoulder . . . and you know it makes you feel better once you’ve done it, not just my legs, but in yourself, you know . . .

PT6

A few participants valued the practical aspects and the techniques learned from the case manager:

I’ve kept a diary all my working life and by going – a daily diary that is – and by going through it we could highlight various things that tip the balance if you like of the scales of happiness and depression and it was highlighted [depression] and between us we figured out a way of coming through it basically.

PT5

When we moved onto the technical part of it where they are asking specific questions and giving specific ideas, I find these very useful and in fact I’ve continued to do those. The ones I am talking about are where you identify things to do . . . and make a list.

PT4

Case management with behavioural activation provides older people with tools to help manage their depressive symptoms and to understand that behaviour and mood are closely linked. Behavioural activation promotes participation in social and physical activity, which may enable older people to ‘move on’ from depression and to experience improved well-being.

Chief investigator

Simon Gilbody.

Trial managers

Katharine Bosanquet and Helen Lewis.

Principal investigators

David Ekers, Simon Gilbody, John Holmes and Esther Cohen-Tovee.

Trial co-ordinators

Katharine Bosanquet, Emily Clare, Lesley Haley, Jahnese Hamilton, Shaista Meer and Gemma Traviss-Turner.

Statisticians

Rhian Gabe, Catherine Hewitt and Ada Keding.

Health economists

Steve Parrott and Dominic Trépel.

Qualitative researchers

Carolyn Chew-Graham and Karen Overend.

Trial administrators

Pauline Holloway, Sarah Mercer, Alice North and Denise Womersley.

Data management team

Matthew Bailey, Ben Cross, Tanya Pawson, Kevin Sherratt, Claire Stewart and Val Wadsworth.

Research team

Katie Atherton, Della Bailey, Jules Beresford-Dent, Jacqueline Birtwistle, Daniel Brett, Simon Carver, Catherine Donegan, Deborah Foster, Samantha Gascoyne, Rebecca Hargate, Gillian Johnson, Rachel Mann, Sarah Nutbrown, Karen Overend, Jodi Pervin, Norah Phipps, Katherine Richardson, Helen Riding and Rebecca Woodhouse.

Trial Management Group

Joy Adamson, Katie Atherton, Della Bailey, Jacqueline Birtwistle, Katharine Bosanquet, Emily Clare, Esther Cohen-Tovee, Jaime Delgadillo, David Ekers, Deborah Foster, Samantha Gascoyne, Simon Gilbody, Lesley Haley, Jahnese Hamilton, Catherine Hewitt, John Holmes, Ada Keding, Peter Knapp, Helen Lewis, Rachel Mann, Dean McMillan, Shaista Meer, Sarah Nutbrown, Karen Overend, Steve Parrott, Jodi Pervin, Gemma Traviss-Turner, Dominic Trépel and Rebecca Woodhouse.

Papers

Lewis H, Hems D, Bosanquet K, Overend K. Is enough being done to treat depression in the elderly? Aging Health 2013;9:243–5.

Pasterfield M, Bailey D, Hems DJ, McMillan D, Richards D, Gilbody SM. Adapting manualized Behavioural Activation treatment for older adults with depression. The Cognitive Behaviour Therapist 2014;7:e5.

Overend K, Lewis H, Bailey D, Bosanquet K, Chew-Graham C, Ekers D, et al. CASPER plus (CollAborative care in Screen-Positive EldeRs with major depressive disorder): study protocol for a randomised controlled trial. Trials 2014;15:451.

Bosanquet K, Mitchell N, Gabe R, Lewis H, McMillan D, Ekers D, et al. Diagnostic accuracy of the Whooley depression tool in older adults in UK primary care. J Affect Disord 2015;182:39–43.

Overend K, Bosanquet K, Bailey D, Foster D, Gascoyne S, Lewis H, et al. Revealing hidden depression in older people: a qualitative study within a randomised controlled trial. BMC Fam Pract 2015;16:142.

Bosanquet K, Bailey D, Gilbody SM, Harden M, Manea LE, Nutbrown SE, McMillan D. Diagnostic accuracy of the Whooley questions for the identification of depression: a diagnostic meta-analysis. BMJ Open 2015;5:e008913.

Donoghue HM, Traviss-Turner GD, House AO, Lewis H, Gilbody S. Life adversity in depressed and non-depressed older adults: a cross-sectional comparison of the brief LTE-Q questionnaire and life events and difficulties interview as part of the CASPER study. J Affect Disord 2016;193:31–8.

Lewis H, Keding A, Bosanquet K, Bailey D, Gilbody SM, Torgerson D. An randomized controlled trial of Post-it^® notes did not increase postal response rates in older depressed participants. J Eval Clin Pract 2016;23:102–7.

Presentations

Overend K. Collaborative Care for Depression: What is the Magic Ingredient? Systematic Review and Qualitative Meta-Synthesis of Provider and Patient Perspectives. Presented at the Primary Care Mental Health Research Conference, Amsterdam, the Netherlands, 12 May 2016.

Bosanquet K. Geographic Variation in Consent Rates during CASPER plus Randomized Controlled Trial. Presented at the Primary Care Mental Health Research Conference, Durham, UK, 26 March 2015.

Overend K. The CASPER Plus Qualitative Study: Collaborative Care for Older People with Depression. Presented at the Primary Care Mental Health Research Conference, Durham, UK, 26 March 2015.

Chew-Graham C. Case Management for Older People with Depression – A Qualitative Study: ‘Reducing the Blind Spots’. Presented at the Society for Academic Primary Care Annual Scientific Meeting Conference, Edinburgh, UK, 20 June 2014.

Overend K, Bosanquet, K. Reducing ‘Blind Spots’ to Achieve Patient-Centred Care: Early Insights from a Qualitative Study. Presented at the Society for Academic Primary Care North Conference, Kendal, UK, 21–22 November 2013.

Posters

Traviss G, Holmes J, Lewis H, Mitchell N, McMillan D, Hems D, et al. CASPER – Collaborative cAre for Screen Positive EldeRs. Presented at British Psychological Society Annual Conference, Harrogate, UK, 2013.

Bosanquet K, Mitchell N, Lewis H, Bailey D, Gabe R, McMillan D, Gilbody S. Diagnostic Accuracy of Whooley Depression Tool in Older Adults Based in Primary Care in the UK. Presented at the UK Primary Care Mental Health Research Conference, Manchester, UK, 2013. Won best academic poster award.

Overend K, Lewis H, Bosanquet K, Mann R, Hems D, Bailey D, Chew-Graham C. The CASPER Plus Nested Qualitative Research Study. Presented at the UK Primary Care Mental Health Research Conference, Manchester, UK, 2013.

Overend K, Lewis H, Atherton K, Bosanquet K, Hall R, Hems D, et al. The CASPER Plus Trial. Presented at the UK Primary Care Mental Health Research Conference, Manchester, UK, 2013.

Overend K, Bosanquet K, Lewis H, Nutbrown S, Bailey D, Hems D, et al. Reducing ‘Blind Spots’ to Achieve Patient-Centred Care: Preliminary Results from a Qualitative Study. Presented at Primary Care Mental Health Research Conference, Exeter, UK, 2014.

Lewis H, Hems D, Bailey D, McMillan D, Overend K, Woodhouse R, et al. Self-Help for Those at Risk of Depression: The SHARD Trial, A Study Protocol. Presented at Primary Care Mental Health Research Conference, Exeter, UK, 2014.

Chew-Graham C. Case Management for Older People with Depression – A Qualitative Study: Offering a New Perspective to Patients. Presented at World Association of Social Psychiatry Conference, London, UK, 2014.

Bosanquet K. Objective Data: Measuring Cost-Effectiveness of Healthcare Interventions. Presented at NIHR CRN Mental Health National Scientific Meeting Conference, York, UK, 2015.

Overend K. Case Management for Older People with Depression: A Qualitative Study. Presented at NIHR CRN Mental Health National Scientific Meeting Conference, York, UK, 2015.

Overend K. The CASPER Plus Qualitative Study: Collaborative Care for Older People with Depression. Presented at Primary Care Mental Health Research Conference, Durham, UK, 2015.

Bosanquet K. Objective Data: Measuring Cost-Effectiveness of Healthcare Interventions. Presented at Primary Care Mental Health Research Conference, Durham, UK, 2015.

Bailey D. Collaborative Care is Valued by Older Adults with Depression: The CASPER Trial. Presented at Primary Care Mental Health Research Conference, Durham, UK, 2015.

Keding A, Lewis H, Bosanquet K, Gilbody S, Torgerson D. Post-It Notes to Improve Questionnaire Response Rates in RCTs Findings from a Randomised Sub-Study. Presented at the International Clinical Trials Methodology Conference, Glasgow, UK, 2015.

Traviss-Turner GD. Measuring Life Events in Health Research. Presented at Primary Care Mental Health Research Conference, Amsterdam, the Netherlands, 2016.

Overend K, Groom M, Kirby N, Teahan A, Delgadillo J. Using Outcome Feedback in Psychological Therapy: Qualitative Findings from a Feasibility Study. Presented at Primary Care Mental Health Research Conference, Amsterdam, the Netherlands, 2016.

Radio broadcast interviews

• British Broadcasting Corporation (BBC1). Inside Out North East & Cumbria [interview with Gilbody S, Foster DJ and Bailey D]. London, 15 October 2012.

• British Broadcasting Corporation (BBC Radio York). BBC Radio York [interview with Gilbody S]. York, 15 October 2012.

• British Broadcasting Corporation (BBC1). Look North [interview with a GP whose practice participated in the CASPER study]. London, 15 October 2012.