Collaborative and Stepped Care for Mental Disorders

Abstract

Background

Collaborative and stepped care (CSC) models are recommended in guidelines because of their effectiveness in treating depression and anxiety disorders. The evidence for other mental disorders is, however, limited. The aim of this study was to evaluate the effectiveness of a collaborative and stepped care model (COMET) for patients with depressive, anxiety, somatoform, and/or alcohol-related disorders and related comorbidities in the routine care setting in Germany.

Methods

A prospective, cluster-randomized, controlled, parallel-group superiority trial was conducted; the subjects were patients in primary care practices. The primary endpoint was the change in mental health-related quality of life, assessed with the SF-36 Mental Health Component Summary Score (MCS) at 6 months in the intention-to-treat population. The secondary endpoints were symptom severity, remission, and response.

Results

Forty-one primary care offices were randomized either to COMET (n = 20) or treatment as usual (TAU, n = 21), and 615 patients were recruited (CSC: n = 307; TAU: n = 308). Data were available for 62% (COMET) and 55% (TAU) of the participants at 6 months. No significant group difference was found with respect to the primary endpoint (-1.96, 95% confidence interval [-4.39; 0.48], p = 0.113) or any of the secondary endpoints.

Conclusion

We found no superiority of the COMET model for the mental disorders addressed. Methodological issues, including differences at baseline and high dropout rates, make these findings challenging to interpret. Future studies should ensure comparability of groups, allocate resources for quality management, and investigate more suitable outcome measures, paying attention to factors of implementation.

With a 12-month prevalence of 17.6% worldwide and 27.7% in Germany, mental disorders are a frequent phenomenon (1, 2). They impose a substantial burden on patients and healthcare, with high direct and indirect costs (3). Mental and substance use disorders are responsible for 22.8% of years lived with disability (4). The most prevalent are depression and anxiety, somatoform, and alcohol-related disorders (5), with significant symptom overlap. Indeed, 44% of patients report two and 22%, three or more comorbid mental disorders (2, 6).

Primary care is often the first point of contact for patients with mental disorders, yet referral to specialized care is difficult, even in highly structured healthcare systems (7, 8). Guidelines increasingly recommend collaborative and stepped care (CSC) as an appropriate model care provision, allowing efficient resource allocation and care for comorbid conditions (9, 10). CSC models are characterized by:

1) An interdisciplinary network of providers, including primary care physicians (PCP), mental health professionals (e.g., psychotherapists, psychiatrists), and nurses

2) Stepped care algorithms that allocate resources according to disorder severity and treatment response, based on systematic monitoring (11, 12).

Numerous studies have demonstrated the efficacy of CSC among patients with depressive and anxiety disorders (12–14). Promising evidence suggests that CSC improves the management of somatoform disorders (15, 16) and increases treatment uptake (e.g., psychosocial treatment and/or pharmacotherapy) in alcohol-related disorders (17). A systematic review of 39 studies found that only five of them addressed mental comorbidities, and just one investigated more than two mental disorders (18). Another CSC study on patients with anxiety, depressive and stress-related disorders found no significant differences compared to treatment as usual (TAU), but earlier treatment response in the CSC group (19).

The aim of the study presented here was evaluate the effectiveness of a guideline-based CSC model (Collaborative and Stepped Care in Mental Health by Overcoming Treatment Sector Barriers; COMET) in patients with depressive, anxiety, somatoform and/or alcohol-related disorders and their comorbidities under routine conditions of the German healthcare system (20). The primary hypothesis was that treatment with COMET leads to a greater improvement in mental health-related quality of life by 6 months after baseline than TAU. The secondary endpoints comprised changes in disorder-specific symptom severity, response rates, and remission rates, together with healthcare utilization.

Methods

Design

The study was a cluster-randomized, prospective, parallel-group superiority trial comparing the effectiveness of CSC (COMET) and TAU in primary care patients with depressive, anxiety, somatoform and/or alcohol-related disorders conducted in Hamburg, Germany (ClinicalTrials.gov ID: NCT03226743). The Hamburg Medical Chamber ethics committee approved the study protocol (PV5595). The results are reported in accordance with the CONSORT 2010 guidelines, and the protocol has been published (20). A detailed description of the methods is provided in the eMethods.

eMethods.

Design

The study was a cluster-randomized, prospective, parallel-group superiority trial comparing the effectiveness of the COMET model with treatment as usual (TAU), conducted within a real-world setting involving a consecutive sample of primary care patients with depressive, anxiety, somatoform and/or alcohol-related disorders in Hamburg, Germany (ClinicalTrials.gov ID: NCT03226743).

The study protocol was approved by the Hamburg Medical Chamber ethics committee (PV5595) prior to patient recruitment. Reporting adhered to CONSORT 2010 guidelines (e1), and the study protocol has been published (20). The authors assert that all procedures contributing to this work comply with the ethical standards of relevant national and institutional committees and the Helsinki Declaration of 1975, as revised in 2013. Written informed consent was obtained from all participants.

Inclusion and exclusion criteria

Clusters

Primary care practices that were licensed by the Association of Statutory Health Insurance Physicians in Hamburg and had entered into a cooperation contract with the study group were eligible to take part. All participating primary care physicians (PCPs) received detailed patient information, informed consent forms, a tablet computer and instructions for the screening and recruitment procedures of patients.

Patients

Eligible for inclusion were persons aged 18 years or over (no upper age limit) who provided informed consent and met at least one of the following ICD-10 diagnoses: depressive episode (F32), recurrent depressive disorder (F33), dysthymia (F34.1), agoraphobia (F40.0), social phobia (F40.1), panic disorder (F41.0), generalized anxiety disorder (F41.1), mixed anxiety and depressive disorder (F41.2), somatoform disorders (F45), mental and behavioral disorders due to use of alcohol (F10). Patients with insufficient proficiency in the German language and those who were already receiving treatment for their (mental) conditions were excluded from study participation.

Randomization and masking

The participating PCPs were cluster-randomized in a 1:1 ratio and a block length of four using a list of computer-generated random numbers without any stratification variables. After obtaining informed consent from the physicians’ practices, the study team forwarded the practice names to the coordinator, who entered them in the randomization list and communicated the randomization results. The study team was unaware of the randomization sequence and which allocation group would be next. Due to the nature of the intervention, it was not possible to blind patients and health professionals to group assignment. Research assistants who conducted the patients’ follow-up assessments were also not blinded for logistical reasons (study support and guidance of PCPs). The biometricians were blinded to group allocation during the prespecified analyses outlined in the statistical analysis plan and were not involved in study implementation or data collection.

Procedures

Patients were informed about the study by practice assistants and received a tablet for screening, which, in line with the recommendations of practice guidelines (9, 21–24), included modules of the German version of the Patient Health Questionnaire (MDD; PHQ-9) (26), generalized anxiety (GAD-7) (27), panic (PHQ Panic module) and somatoform modules (PHQ-15) (28), the Somatic Symptom Disorder-B Criteria Scale (SSD-12) (29) and the Alcohol Use Disorders Identification Test (AUDIT-C) (30). The screening results along with cut-off values (PHQ-9 score ≥ 5; GAD-7 score ≥ 5, PHQ-Panic module: all questions of 3a-d and ≥ 4 items of 4 (a-k) must be answered affirmatively; PHQ-15 score ≥ 5; SSD-12 score ≥ 23; AUDIT-C score ≥ 6) were presented to the PCPs via tablet. ICD-10 checklists supported the PCPs in determining the diagnosis and the severity of the disorder (mild, moderate, or severe). Patients with one or more diagnoses were included in the study. The PCPs received an expense allowance of up to €120 for each patient included. Patients received a €10 voucher for each completed interview (follow-up). Patient recruitment, intervention, and data collection took place between 12 July 2018 and 22 October 2021. The trial was conducted per protocol, with amendments due to recruitment delays and the COVID-19 pandemic. To meet target enrollment, we assisted some practices during the screening process (n=1 COMET; n=7 TAU).

Intervention

COMET

The intervention consisted of a CSC program to be carried out by 20 PCPs. The collaborative network included 20 psychotherapists (13 cognitive-behavioral and 7 psychodynamic psychotherapists), four psychiatrists, and four inpatient or day-care clinics. To expedite referrals, an online scheduling platform was implemented to facilitate direct booking of psychotherapeutic and psychiatric appointments by PCPs. All network partners received initial training on study procedures, evidence-based guidelines (e.g., the German S3-guideline or National Disease Management Guideline) for depressive, anxiety, somatoform and alcohol-related disorders (9, 21–24), as well as on the COMET interventions and treatment concepts. In addition, quarterly network meetings and an education structure throughout the study period promoted quality assessment and information exchange. Further elements of the intervention included computer-assisted, guideline-based diagnostic procedures (via the eDiagnostic tool) and care pathways, including self-management interventions (e.g., bibliotherapy, online self-help programs). Symptom monitoring was performed at disorder-specific, predefined intervals by the main care provider (PCPs or mental health specialist). To identify severe cases with inadequate treatment, the study team monitored the care pathways of the patients concerned. In the event of insufficient treatment response or complications in the referral process, the main care provider was notified (20; eSupplement 2, Table 1). In addition to reimbursement in accordance with the statutory health insurance (SHI) benefit catalog and the applicable fee schedules, the PCPs of the COMET group also had the option of billing for additional services that facilitated interprofessional collaboration within the COMET model.

Treatment as usual

PCPs in the comparison group provided TAU under the German statutory healthcare system, including potential referrals to psychotherapy or psychiatric or psychosomatic outpatient or inpatient facilities. The TAU group had unrestricted access to routine mental health care according to evidence-based guidelines, especially the clinical practice guidelines (S3 guidelines), which provide recommendations for diagnosis and therapy based on research and clinical experience. Costs were reimbursed according to the benefit catalog and the fee schedules established by the SHI and the National Association of Statutory Health Insurance Physicians („Kassenärztliche Vereinigung“). All patients in the TAU group underwent the same systematic screening and diagnostic process as in the COMET group to ensure a comparable inclusion process.

Endpoints

The primary endpoint was the change in mental health-related quality of life, assessed with the Short Form Health Survey (SF-36) mental component summary (MCS) score (25), from baseline to 6 months. Responses were documented on a Likert scale, weighted, summed, and transformed to the range 0–100, where higher values indicate better health-related quality of life. We opted for this general score because it was not possible to apply disorder-specific measures due to the heterogeneity of disorders.

The secondary endpoints included change in the primary endpoint from baseline to 12 months, as well as changes in the severity of disorder-specific symptoms from baseline to 6 and 12 months, as assessed by various patient-reported measures (PHQ-9 [26], GAD-7 [27], PHQ-15 and PHQ-Panic module [28], SSD-12 [29], and AUDIT-C [30]). Additionally, disorder-specific response to treatment, defined as a reduction of ≥ 50% in symptoms, and remission rates were assessed (with clinical cut-off values of PHQ-9: 5, GAD-7: 5, PHQ-15: 9, for SSD-12: 23, and AUDIT-C: 4 (women)/5 (men)). The remission criterion was considered as met if the respective endpoint value fell below the specified cut-off at each time point. Further secondary endpoints included physical health-related quality of life, as measured by the SF-36 physical component summary score (PCS), patient satisfaction in general as well as with the care received (31), and health care utilization. All patient-reported endpoints were measured by questionnaires assessing lived experiences and health status (32).

To address recruitment delays and maintain the study’s three-year time frame, we consulted with the funder and the study’s advisory board, subsequently modifying two study design parameters that deviated from the original trial registration. Given the complexity of the COMET trial and the difficulty in adhering to the timeline, the statistical power of the study was reduced from 90% to 80%. This led to a reduction in the required number of offices from 50 to 38 and patients from 750 to 570, increasing the likelihood of achieving the target sample size. Furthermore, in January 2019 the primary endpoint measurement was moved from 12 to 6 months after baseline, consistent with similar intervention studies on mental disorders, enhancing comparability. By that time, a total of 157 patients had been recruited by the participating PCPs (n = 128 in the COMET group and n = 29 in the TAU group). However, no statistical analyses had been performed regarding the study hypotheses. Since the recruited patients were already scheduled for interviews 6 months after baseline and had given their informed consent, their participation remained unchanged. The modifications were approved by the funder and the advisory board, after which the study registration was updated on ClinicalTrials.gov and the ethics committee was notified. The published study protocol and statistical analysis plan were finalized on the basis of the modified parameters.

Data collection

Data on screening, diagnostics, disorder severity, treatment decision, and baseline assessment of the primary endpoint, were collected on a tablet using specially developed web-based software. Additional baseline questions included reasons for PCP consultation, age, sex, and current psychotherapeutic or psychopharmacological treatment. Follow-up data assessments were conducted via telephone interviews at baseline and at 3, 6, and 12 months. Trained research assistants conducted the Composite International Diagnostic Interview (CIDI) to validate diagnoses and assess exclusion criteria following the standard procedure.

Statistical analysis

Sample size calculation was based on the detection of a small to moderate standardized mean difference (Cohen’s d of 0.35 [e2]) with regard to the primary endpoint (change in the SF-36, MCS after 6 months) between the COMET and the TAU group with a statistical power of 0.80 at a type I error rate of 0.05 (two-sided). Assuming a correlation of 0.50 between baseline and follow-up measurements, this required analyzable data from 95 patients per group (190 in total) for a linear model with the baseline measurement as covariate (e3), if randomization took place at the patient level. With an average cluster size (number of patients per practice) of 12 and an intra-cluster correlation (ICC) of 0.05, this sample size was multiplied by the design effect of 1.55 (e4), leading to 156 patients in 13 offices per group (312 patients in 26 practices in total). Assuming dropout rates of 30% (practices) and 20% (patients), we aimed to recruit 38 practices (19 per group) with 15 patients each, resulting in a total number of 570 study participants (285 per group).

We predetermined all data analyses in a statistical analysis plan that was finalized and signed before the primary analysis (eSupplement 1). The analysis of the primary endpoint was based on the intention-to-treat (ITT) population. The ITT population consisted of all practices and patients that were randomized and included in the study for whom at least one measurement was performed after baseline. A linear mixed model was applied to analyze the primary endpoint with group (COMET/TAU), time (3, 6, 12 months after baseline), and the group by time interaction as fixed effects, patients nested in practices as random effects, and the baseline value of the SF-36 MCS as covariate. Prior to any further analysis, the baseline-imputed data was based on the EM algorithm to address missing values due to technical errors (COMET: n=18 [5.9%]; TAU: n=25 [8.1%]), as the new software initially failed to save complete patient data for the primary outcome. The direct maximum likelihood method was used as the statistical estimation procedure to compare the adjusted SF-36 MCS values between the groups, resulting in unbiased estimators under the missing-at-random assumption. No additional imputation of missing values was performed in the primary analysis. We assessed the contrast between the groups at the 6-month follow-up in confirmatory manner. The analysis was repeated in the per-protocol (PP) population (data of patients who provided data both at baseline and at 6-month follow-up and met none of the exclusion criteria).

We carried out sensitivity analyses with different methods for imputation of missing values (e.g., multiple imputation, last observation carried forward). Due to imbalances between the randomized groups at baseline (age, PHQ-9 and GAD-7 scores, symptom severity, SSD-12 remission, main diagnosis determined by PCP), we conducted two post-hoc analyses to examine the contribution of the imbalances to the treatment effect. First, we repeated the primary endpoint analysis with these variables as additional covariates. Second, we conducted a propensity score analysis taking these baseline variables into account.

The secondary endpoints were analyzed using a linear mixed model with the same model parameters as in the primary analysis. For the secondary endpoints with a binary format (remission, response), a mixed logistical regression model was conducted with group and time as fixed effects, patients nested in practices as random effects, and the respective score values at baseline as covariate. The group by time interaction was included if the corresponding interaction p-value was below 0.05.

Subgroup analyses for the primary endpoint comprised main diagnosis (depressive, anxiety, somatoform and alcohol-related disorder), comorbidity (yes vs. no), sex (female vs. male), age, educational level (secondary general school, intermediate secondary school, high school, university or technical college degree and no school degree), employment status (no employment, minor employment, part-time employment and full-time employment), recruitment strategy (without vs. with assistance from the study team), symptom severity (mild, moderate, and severe), and intervention start during COVID-19 pandemic (yes vs. no). Adjusted means, effect sizes and odds ratios are reported with their 95% confidence intervals and p-values. For all analyses the two-sided type I error was set at 0.05. Adverse events were determined using frequency tables for the COMET group. To assess potential differences in treatment utilization between the COMET and TAU groups, we descriptively analyzed the treatments received at 6 months after baseline.

Inclusion and exclusion criteria

Clusters

For inclusion, primary care practices had to be licensed by the Association of Statutory Health Insurance Physicians in Hamburg and had to have signed a cooperation contract with the study group.

Patients

The inclusion criteria were age > 18 years, informed consent to study participation, and one or more of the following ICD-10 diagnoses: depressive episode (F32), recurrent depressive disorder (F33), dysthymia (F34.1), agoraphobia (F40.0), social phobia (F40.1), panic disorder (F41.0), generalized anxiety disorder (F41.1), mixed anxiety and depressive disorder (F41.2), somatoform disorders (F45), mental and behavioral disorders due to use of alcohol (F10). Patients with insufficient proficiency in German and those already receiving treatment were excluded.

Intervention

COMET

The intervention was a CSC program to be carried out by 20 PCP. The collaborative network comprised 20 psychotherapists (13 behavioral and 7 psychodynamic), four psychiatrists, and four inpatient or day-care facilities. An online scheduling platform was implemented to facilitate direct booking of outpatient psychotherapeutic and psychiatric appointments by PCPs. Network partners received training on evidence-based guidelines for depressive, anxiety, somatoform and alcohol-related disorders (9, 21–24), and the COMET model. Furthermore, quarterly network meetings promoted quality assurance and information exchange. Further elements were computer-assisted and guideline-based treatment decision algorithms, including self-management interventions. The symptoms were monitored at disorder-specific intervals by the main care provider (PCP or mental health specialist). To identify severe cases with inadequate treatment, the study team tracked the care pathways of the patients concerned and notified the main care provider (see study protocol [20] and eSupplement 2, Table 1 for details).

Treatment as usual

The comparison group received treatment as usual under the German statutory healthcare system, with unrestricted access to evidence-based mental health care in accordance with the clinical practice guidelines (e.g., S3 guidelines), which provide recommendations for diagnosis and treatment.

Endpoints

The primary endpoint was the change in mental health-related quality of life, assessed with the Short Form Health Survey (SF-36) mental component summary score (MCS) (25) from baseline to 6 months on a scale of 0 to 100. A general score was chosen due to the heterogeneous conditions. The secondary endpoints were change in the primary endpoint from baseline to 12 months, change in disorder-specific symptom severity from baseline to 6 and 12 months as measured by validated instruments (PHQ-9 [26], GAD-7 [27], PHQ-15 and PHQ-Panic module [28], SSD-12 [29], and AUDIT-C [30]) as well as disorder-specific response to treatment and remission (eMethods). Further secondary endpoints were physical health-related quality of life (SF-36 physical component summary score [PCS]), patient satisfaction (31), and treatment utilization. All endpoints were assessed through patient-reported questionnaires (32). Due to recruitment delays, two parameters from the protocol, the power and primary outcome measure, were modified (eMethods).

Data collection

Data were collected on a tablet using web-based software. Follow-up assessments were conducted via telephone interviews, with trained research assistants administering the Composite International Diagnostic Interview (CIDI) to verify the diagnoses and the exclusion criteria.

Statistical analyses

The sample size calculation aimed to detect a small to moderate standardized mean difference (Cohen’s d of ≥ 0.35) in the primary endpoint between COMET and TAU with 80% statistical power and a significance level of 5%. Accounting for anticipated dropout rates of 30% for practices and 20% for patients, we targeted the recruitment of 570 patients (285 per group) across 38 practices.

All data analyses were predetermined in a statistical analysis plan (eSupplement 1). The primary endpoint analysis was based on the intention-to-treat (ITT) population. A linear mixed model was applied with group (COMET/TAU), time (3, 6, and 12 months), and group by time interaction as fixed effects, with patients nested in practice as random effects and baseline values of the SF-36 MCS included as covariates. The analysis was repeated in the per-protocol (PP) population. Sensitivity analyses explored different imputation methods. Due to baseline imbalances, post hoc analyses included additional covariates and propensity score analyses. The secondary endpoints were analyzed using linear mixed models and mixed logistic regression. The subgroup analyses for the primary endpoint covered main diagnosis, comorbidity, sex, age, educational level, employment status, recruitment strategy, symptom severity, and intervention start during COVID-19 pandemic. Adjusted means, effect sizes, odds ratios, and their 95% confidence intervals are reported.

Results

Participants

The PCP were recruited in the period from 17 January 2018 to 31 December 2019, with invitation letters sent in two waves to n = 2451 PCPs in Hamburg. A total of 41 practices were recruited, with 20 randomized to the COMET and 21 to the TAU group (Figure 1). Seventeen COMET offices (52.9% female PCPs) and 14 TAU offices (71.4% female PCPs) actively included patients.

Figure 1.

Flow chart showing the clusters and participants throughout the trial.

CIDI, Composite International Diagnostic Interview; COMET, intervention group; FU, follow-up; PCP, primary care physician; TAU, control group, treatment as usual; T0, baseline; T1, FU at 3 months; T2, FU at 6 months; T3, FU at 12 months

“Missing” refers to participants who did not provide data at the designated time but later re-engaged in the study and provided data. “Lost to FU” refers to participants who initially participated in the study but did not provide data at later time points, indicating that they had dropped out.

Patients were recruited between 12 July 2018 and 22 October 2021. A total of n = 1183 (COMET: n = 504; TAU: n = 679) patients were screened for eligibility, with n = 713 (COMET: n = 347; TAU: n = 366) meeting the study diagnoses. Of those, n = 615 patients gave informed consent to participate (COMET = 307; TAU: n = 308).

The patients in the COMET and TAU groups differed in the primary endpoint (SF-36, MCS) and disorder severity at baseline, suggesting greater impairment among COMET patients (Table 1; eSupplement 2, Table 2). The baseline characteristics of dropouts did not differ from those of the participants who continued the study (eSupplement 2, Table 3). Further details on post-hoc sensitivity analyses, secondary endpoints, and subgroup analyses are provided in the eResults.

Table 1. Baseline characteristics of the study groups.

	COMET (n = 307)	TAU (n = 308)
Age in years: M (SD)	37.7 (12.4)	43.3 (14.9)
Female sex: N (%)	188 (61.2)	205 (66.7)
Education level: N (%) – Secondary general schoola – Intermediate secondary schoolb – High schoolc – University or technical college – No school degree – Missing (N)	24 (9.1) 76 (28.8) 109 (41.3) 55 (20.8) 0 (0.0) 43	28 (11.5) 76 (31.1) 83 (34.0) 52 (21.3) 5 (2.0) 64
Employment status: no employment N (%) – Missing (N)	49 (18.8) 46	65 (26.6) 64
Relationship status: in partnership N (%) – Missing (N)	157 (59.5) 43	164 (67.2) 64
Mental health-related quality of life (SF-36, MCS): M (SD) – Missing (N)	28.04 (8.81) 18	35.60 (12.29) 25
Severity of depressive symptoms (PHQ-9): M (SD) – Missing (N)	14.09 (4.98) 1	11.07 (5.23) 2
Severity of anxiety symptoms (GAD-7): M (SD) – Missing (N)	11.78 (4.48) 1	8.99 (4.85) 4
Severity of somatic symptoms (PHQ-15): M (SD) – Missing (N)	14.74 (5.35) 1	13.13 (5.53) 1
Psychological burden related to somatic symptoms (SSD-12): M (SD) – Missing (N)	21.31 (9.99) 13	17.91 (9.85) 17
Symptoms of alcohol-related disorder (AUDIT-C): M (SD) – Missing (N)	3.19 (2.60) 2	2.87 (2.36) 1
Main diagnosis (ICD-10) diagnosed by PCP: N (%)  – Depressive disorder  – Anxiety disorder – Somatoform disorder – Alcohol-related disorder*1	226 (73.6) 63 (20.5) 18 (5.9) 0 (0)	241 (78.2) 30 (9.7) 29 (9.4) 8 (2.6)
Severity of main diagnosis: N (%)*2 – Mild – Moderate – Severe	42 (13.7) 235 (76.5) 30 (9.8)	172 (55.8) 118 (38.3) 18 (5.8)
One or more comorbidities: N (%)	90 (29.3)	97 (31.5)

Results are shown as N (%), M = mean and SD = standard deviation.AUDIT-C, Alcohol Use Disorders Identification Test; COMET, intervention group; GAD-7, Generalized Anxiety Disorder Scale-7; ICD-10, International Statistical Classification of Diseases and Related Health Problems; PCP, primary care physician; PHQ-9, Patient Health Questionnaire-9; PHQ-15, Patient Health Questionnaire-15; SF-36 MCS, Short Form Health Survey mental component summary score; higher values indicate greater health-related quality of life; SSD-12, Somatic Symptom Disorder-B Criteria Scale; TAU, control group, treatment as usual. For the disorder-specific instruments, higher scores indicate a higher symptom burden.

^aGerman: Hauptschule (9 years of education). ^bGerman: Realschule (10 years). ^cGerman: Gymnasium (13 years).

Numbers in bold show the variables with differences between COMET and TAU at baseline.

^*1 Patients with alcohol-related disorders were underrepresented in our sample.

^*2 Symptom severity was assessed by the PCP.

eResults.

Results

Participants

The recruitment of PCPs started on 17 January 2018 and ended on 31 December 2019, during which time invitation letters were sent by mail to n = 2451 PCPs in Hamburg in two waves. Following information events and personal visits by the study team, 41 offices were successfully recruited, with 20 randomized to the COMET group and 21 to the TAU group (Figure 1). Of those, 17 COMET practices (52.9% female physicians) actively enrolled patients, compared to 14 TAU practices (71.4% female physicians).

A total of n = 1183 patients (COMET: n = 504; TAU: n = 679) were screened for eligibility, of whom n = 713 (COMET: n = 347; TAU: n = 366) met at least one of the study diagnoses. Of those, n = 615 patients provided informed consent to participate in the study (COMET = 307 in COMET; TAU: n = 308). The patients in the COMET and TAU groups differed in primary endpoint (SF-36, MCS) and disorder severity at baseline, indicating a greater level of impairment among COMET patients compared to those in TAU (eSupplement 2, Table 2). We found no relevant differences regarding baseline characteristics between patients who continued their study participation and those who dropped out (eSupplement 2, Table 3).

Effectiveness

Table 2 presents the results of the primary endpoint analysis with baseline-imputed data using a baseline-adjusted linear mixed model. No significant group difference was found in the MCS scores (-1.96 [-4.39; 0.48], p = 0.113; see Figure 2). Sensitivity analyses with different methods for missing values led to comparable results (eSupplement 2, Tables 4 and 5). The results of the primary analysis also were confirmed to be robust in the post hoc covariate-adjusted analysis (eSupplement 2, Table 6) and the PP analysis (eSupplement 2, Table 7). According to the Bland-Altman plot, there was an indication of an interaction between MCS score and treatment effect. The propensity score plot revealed that more severely impaired patients were more likely to be allocated to the COMET group (eSupplement 2, Figures 1 and 2). However, analysis of the primary endpoint after propensity score matching also showed no relevant group differences (data not shown). There was a difference in the MCS score in favor of the TAU group at 3 months (-4.92 [-7.30; –2.54], p < 0.001), but not at 12 months (eSupplement 2, Tables 8 and 10).

There were no relevant group differences at 6 months in terms of secondary endpoints (Table 2), disorder-specific response, or remission (Table 3), with the exception of marginal indications of potential group differences in symptoms of alcohol-related disorder in favor of the COMET group; however, this effect was clinically irrelevant, time-independent, and remained consistent over time. There were relevant group differences in mental health-related quality of life, symptom severity (depression, anxiety, and somatic symptoms), remission (anxiety, somatic symptoms), and response (depression and psychological burden related to somatic symptoms) at 3 months in favor of the TAU group. By 12 months, these effects had diminished and differences in psychological burden related to somatic symptoms and response in depression in favor of the COMET group were found (eSupplement 2, Tables 8–11).

None of the subgroup analyses revealed relevant differences between the groups (eSupplement 2, Tables 12a and b), despite indications that patients with moderate symptom severity in the TAU group may have benefited most from the intervention, regardless of the clinical relevance of the observed difference. Descriptive analyses have demonstrated a disparity in the utilization of mental health services between the two groups. Compared with the TAU group, COMET patients more often received psychotherapy (60% vs. 17%), psychopharmacotherapy (27% vs. 15%), and consultations with a psychiatrist (11% vs. 6%) eSupplement 2, Table 13). Preliminary results from the analysis of the implemented care pathways indicated that patients in the COMET group received a significantly greater proportion of needs-oriented and guideline-based treatment compared to those in the TAU group at both 3 and 6 months (results not presented). Due to insufficient response in the TAU group, data on adverse events (AE) are available only from the COMET group (reported by n = 5 primary care physicians and n = 11 psychotherapists), necessitating cautious interpretation. This limitation restricts a comprehensive understanding of AE across both treatment arms. A total of 26 COMET patients experienced AE: 11 reported personal burdens, 14 health-related burdens, 9 work-related burdens, and 16 impaired occupational functioning. One person required hospitalization, and four experienced (passive) suicidal ideation. We cannot determine whether AE were associated with study participation, further limiting the interpretability of these findings.

Conclusions

We found no evidence supporting the superiority of the COMET model over TAU, which does not align with previous findings on CSC for individual mental disorders. Despite defined inclusion criteria and sensitivity analyses, our results are likely to be constrained by baseline imbalances and potential uncontrolled selection bias from unobserved confounders. However, this complex randomized controlled trial is one of the few to address multiple mental disorders and their comorbidities, warranting cautious interpretation of its overall benefits. Future studies should explore general CSC models for patients with different mental disorders and comorbidities, along with more suitable or additional outcome measures to evaluate their effectiveness. Systematic monitoring should be conducted to ensure the quality and proper implementation of guideline-based care pathways. Additionally, contextual factors influencing the implementation of the CSC model for various mental disorders should be examined. Despite its lack of effectiveness, our complex CSC model offers valuable tools and intervention components for further research efforts in routine practice.

Effectiveness

Table 2 shows the results of the primary endpoint analysis with baseline-imputed data using a baseline-adjusted linear mixed model. No significant difference in MCS scores was found between the COMET and the TAU group (-1.96 [-4.39; 0.48], p = 0.113; Figure 2). Sensitivity analyses with different methods for missing values led to comparable results (eSupplement 2, Tables 4 and 5). The results of the primary analysis also proved to be robust in the post-hoc covariate-adjusted analysis (eSupplement 2, Table 6) and in the PP analysis (eSupplement 2, Table 7). Moreover, there were no group differences at 6 months in terms of secondary endpoints (Table 2), disorder-specific response, or remission (Table 3).

Table 2. Results of primary and secondary endpoint analysis as change from baseline to 6 months.

	COMET			TAU			COMET vs. TAU
	N	Observed mean (SD)	Adjusted mean [95% CI]	N	Observed mean (SD)	Adjusted mean [95% CI]	Adjusted difference [95% CI]	P-value	Cohen’s d	Group × time interaction*1	ICC
Primary endpoint
Mental health-related quality of life, SF-36 MCS	186	11.32 (11.55)	8.99 [7.34, 10.64]	164	8.30 (12.02)	10.95 [9.20, 12.70]	−1.96 [−4.39, 0.48]	0.113	−0.17	0.000	0.011
Secondary endpoint
Severity of depressive symptoms, PHQ-9	186	−5.82 (5.57)	−4.90 [−5.60, −4.20]	167	−4.69 (4.92)	−5.51 [−6.25, −4.77]	0.61 [−0.42, 1.64]	0.242	0.13	0.000	0.011
Severity of anxiety symptoms, GAD-7	186	−5.38 (5.30)	−4.42 [−5.04, −3.81]	166	−4.03 (4.75)	−4.87 [−5.52, −4.21]	0.44 [−0.47, 1.35]	0.334	0.10	0.000	0.003
Severity of somatic symptoms, PHQ-15	186	−5.67 (5.70)	−5.20 [−5.93, −4.46]	167	−5.13 (5.20)	−5.60 [−6.37, −4.83]	0.40 [−0.67, 1.47]	0.455	0.08	0.000	0.021
Psychological burden related to somatic symptoms, SSD-12	105	−8.30 (10.78)	−8.57 [−10.20, −6.94]	98	−4.96 (9.18)	−6.27 [−7.97, −4.57]	−2.30 [−4.66, 0.06]	0.056	−0.27	0.000	0.000
Symptoms of alcohol-related disorder, AUDIT-C	188	−1.08 (1.86)	−0.99 [−1.21, −0.77]	168	−0.55 (1.48)	−0.67 [−0.90, −0.44]	−0.32 [−0.61, −0.03]	0.030	−0.19	0.117	0.030
Physical health-related quality of life, SF-36 PCS	177	3.04 (10.20)	3.21 [1.84, 4.58]	152	1.97 (10.15)	2.14 [0.71, 3.57]	1.08 [−0.72, 2.87]	0.226	0.11	0.895	0.026
Patient satisfaction, ZAPA	167	3.19 (17.01)	3.13 [1.10, 5.15]	152	4.71 (11.93)	3.85 [1.77, 5.94]	−0.72 [−3.20, 1.75]	0.548	−0.06	0.081	0.004

Results are presented as n, adjusted mean differences, SD with 95% CI using the ITT sample for patients with at least one follow-up measurement after baseline. The COMET group showed an average improvement in mental health-related quality of life of 11.32 (SD 11.55) points, while the TAU group improved by 8.30 (SD 12.02). The adjusted difference between groups was non-significant (p = 0.113). Higher values indicate a greater improvement in mental health-related quality of life.

^*1 An interaction p value above 0.05 means that the intervention effect was the same at all time points; otherwise, a time-dependent intervention effect was present. No relevant group differences in symptom severity were found, except for marginal indications of potential group differences in symptoms of alcohol-related disorder in favor of the COMET group; this effect was independent of time. Lower values indicate a greater improvement in symptom burden.

AUDIT-C, Alcohol Use Disorders Identification Test; CI, confidence interval; COMET, intervention group; GAD-7, Generalized Anxiety Disorder Scale-7; ICC, Intraclass Correlation Coefficient (variance of practices divided by total variance); PHQ-9, Patient Health Questionnaire-9; PHQ-15, Patient Health Questionnaire-15; SD, standard deviation; SF-36, Short Form Health Survey (MCS, mental component summary score; PCS, physical component summary score); SSD-12, Somatic Symptom Disorder-B Criteria Scale; TAU, control group, treatment as usual; ZAPA, questionnaire on satisfaction with outpatient care, focus on patient participation; higher values indicate higher satisfaction with care received.

Figure 2.

Results of the primary endpoint analysis presented as adjusted change from baseline with 95% CI, using the ITT sample for patients with at least one follow-up measurement after baseline.

Higher values indicate a greater improvement in mental health-related quality of life. COMET, intervention group; TAU, control group, treatment as usual.

t0 = baseline

t1 = follow-up at 3 months

t2 = follow-up at 6 months

t3 = follow-up at 12 months

Table 3. Results of analysis of secondary endpoints (binary) after 6 months.

	COMET		TAU		COMET vs. TAU
	n/N	%	n/N	%	Adjusted odds ratios (95% CI)	P-value	Group × time interaction*3	ICC
PHQ-panic module	23/36	63.9	12/17	70.6	0.47 [0.21, 1.07]	0.071	0.823	0.020
Disorder-specific remission*1
Severity of depression, PHQ-9	49/187	26.2	61/168	36.3	0.86 [0.47, 1.54]	0.604	–	0.043
Severity of anxiety, GAD-7	65/187	34.8	86/168	51.2	0.61 [0.32, 1.17]	0.139	0.000	0.000
Severity of somatic symptoms, PHQ-15	91/187	48.7	100/168	59.5	0.54 [0.24, 1.22]	0.137	0.000	0.045
Mental burden related to somatic symptoms, SSD-12	87/107	81.3	83/103	80.6	1.63 [0.75, 3.51]	0.216	–	0.000
Symptoms of alcohol-related disorder, AUDIT-C	160/190	84.2	129/169	76.3	2.63 [1.48, 4.67]	0.001	0.265	0.000
Disorder-specific response*2
Severity of depression, PHQ-9	85/186	45.7	77/167	46.1	0.79 [0.43, 1.46]	0.450	0.000	0.003
Severity of anxiety, GAD-7	96/186	51.6	87/164	53.1	0.79 [0.50, 1.23]	0.291	–	0.000
Severity of somatic symptoms, PHQ-15	75/186	40.3	65/167	38.9	1.09 [0.66, 1.78]	0.739	0.055	0.022
Psychological burden related to somatic symptoms, SSD-12	37/104	35.6	31/98	31.6	1.48 [0.69, 3.21]	0.315	0.004	0.000
Symptoms of alcohol-related disorder, AUDIT-C	60/157	38.2	46/144	31.9	1.83 [1.16, 2.90]	0.010	0.439	0.000

Results are presented as n/N (events/sample size), %, and adjusted odds ratios (OR) with 95% CI using the ITT sample for patients with at least one follow-up measurement after baseline.

^*1 Remission = value below clinical cut-off values on disorder-specific instruments: PHQ-9 score < 5, GAD-7 score < 5, PHQ-15 score < 9, SSD-12 score < 23, and AUDIT-C < 4 (women)/< 5 (men)

^*2 Response = at least 50% reduction on disorder specific instruments

^*3 An interaction p-value above 0.05 means that the intervention effect is the same for all time points; otherwise, a time-dependent intervention effect is present. No relevant group differences in remission and response were observed, except for marginal indications of potential group differences in symptoms of alcohol-related disorder in favor of the COMET group; this effect was independent of time.

AUDIT-C, Alcohol Use Disorders Identification Test; CI, confidence interval; COMET, intervention group; GAD-7, Generalized Anxiety Disorder Scale-7; ICC, Intraclass Correlation Coefficient (variance of practices divided by total variance); PHQ-9, Patient Health Questionnaire-9; PHQ-15, Patient Health Questionnaire-15; SSD-12, Somatic Symptom Disorder-B Criteria Scale; TAU, control group, treatment as usual

None of the subgroup analyses revealed relevant differences between the groups (eSupplement 2, Tables 12a and b). Descriptive analyses showed that COMET patients received psychotherapy (60% vs. 17%), psychopharmacotherapy (27% vs. 15%), and consultations with a psychiatrist (11% vs. 6%) more often than those in the TAU group (eSupplement 2, Table 13). Due to insufficient responses from the TAU group, data on adverse events are limited to 26 COMET patients, limiting interpretability (eResults).

Discussion

This cluster-randomized controlled trial evaluated the effectiveness of a guideline-based CSC model for patients with depressive, anxiety, somatoform or alcohol-related disorders and their comorbidities treated in a multiprofessional care provider network, compared with TAU. Under routine conditions of the German healthcare system and the design of the COMET study, no statistically significant group difference in mental health-related quality of life was observed from baseline to 6 months. Secondary analyses also showed no relevant differences. However, both groups showed clinical improvements in mental health-related quality of life over time, possibly reflecting high quality of routine care. Patients in the TAU group also underwent digital screening and diagnostic procedures, potentially enhancing both the detection of mental disorders and the care provided. Our results diverge from evidence in previous research suggesting the superiority of CSC models over routine care for patients with depression (33, 34) or anxiety disorders (12, 13), taking into account the heterogeneity in results, design and implementation of these models (13). Similar to our results, the CSC trial by Oosterbaan and colleagues (19) examined multiple conditions but did not find general superiority of the collaborative approach, yet reported significant differences in response and remission rates favoring the CSC group at 4 months.

Patient characteristics and selection

A major challenge in interpreting our results arose from baseline differences between the groups. Our qualitative survey indicated that PCPs in the COMET group may have selected more severely ill patients, anticipating easier access to specialized care. Differences in enrollment might also have been influenced by the support offered to PCPs in the TAU group in overcoming recruitment delays. The challenge of enrolling primary care patients in the TAU group and the risk of selection bias have been described in other studies (35). Consequently, despite cluster randomization, significant baseline differences emerged. Although the statistical analyses were adjusted for these imbalances, unobserved characteristics may still differ between groups and influence treatment response. Indeed, the propensity score plot and the baseline scores show that COMET patients appeared more burdened and possibly more severely ill, as shown by their lower MCS scores and higher symptom scores. Secondary analyses at 12 months showed differences in psychological burden related to somatic symptoms and depression-specific response favoring the COMET group. This delayed treatment response suggests that the benefits of the CSC model may not emerge until later in severely ill patients, a hypothesis to be examined in the pending 24-month follow-up analyses. In contrast to previous studies, which focused on single conditions, our trial included patients with multiple disorders and comborbidities, potentially complicating the response to treatment.

Implementation challenges

While the model was designed to be feasible for PCPs and mental health specialists, its implementation for patients with multiple disorders, alongside guideline training for network partners, may have been overly complex for routine care. The lack of rigorous monitoring of the response to treatment in the COMET model improved the systematic case-based communication among care providers only marginally.

Treatment delivery and quality

Descriptive analyses showed that COMET patients received more mental health treatments. However, this does not necessarily reflect the quality or appropriateness of treatment. Merely providing access to specialized care without enhancing collaboration may not be sufficient to improve outcomes (36), as highlighted by our qualitative survey of the care providers, which identified potential for improvement in communication, feedback loops and remuneration of collaborative activities (37). Further analyses are needed to examine how accurately the received treatments adhered to current guidelines and to determine which care pathways were implemented, particularly regarding the severity of disease (38).

Strengths and limitations

The COMET study is among the few to address multiple mental disorders and their comorbidities in a single model, conducted under German routine care conditions. Despite initial recruitment challenges, the targeted number of PCPs and patients was achieved. With sufficient statistical power and active engagement of PCPs, the trial’s implementation addressing comorbid mental health conditions highlighted its applicability in clinical practice. Furthermore, patient-reported outcome measures, cluster-randomization, and the blinded statistical analyses bolster the trial’s rigor, objectivity, and credibility.

Nonetheless, our trial is limited by methodical constraints. Alongside unequal baseline samples, we found high dropout rates of 38% (COMET) and 45% (TAU) at 6 months, comparable with previous studies (33). Notably, we found no differences between patients who continued treatment and those who dropped out. A significant proportion of patients were not captured between screening and the baseline telephone survey. While patient-reported endpoints, including health-related quality of life, are increasingly being used in efficacy trials (39), the SF-36 may lack the sensitivity required to detect subtle changes in mental health symptoms, particularly among populations with more severe conditions (40). Furthermore, the low response rates among the PCPs may limit external validity and generalizability, as those PCPs who participated may have been more open to mental health conditions, potentially affecting both groups equally. Lastly, the simultaneous implementation of study procedures and the CSC model complicated the evaluation by blurring start-up difficulties and assimilation processes with intervention assessment. Future studies should ensure group comparability, allocate resources for quality assurance, investigate additional outcome measures, and systematically analyze the determinants crucial for implementation. A comprehensive overview of the conclusions can be found in the eResults.