Collaborative care for patients with depression and diabetes mellitus: a systematic review and meta-analysis

Literature searches were conducted through March 27, 2013 using the electronic databases Medline (1946 to present), Embase (1980 to present), Cocharne library (present) and PsycINFO (1806 to present). The detailed search strategies were shown in Additional file 1. No restriction was placed on type of language. We also screened the references from retrieved articles and reviews to identify additional articles which met the eligibility criteria.

Studies that met the following criteria were included in this meta-analysis: (1) Participants: both male and female patients of any age, with a diagnosis of both depression and diabetes. Diagnosis of depression was according to one of the following: a, diagnosis made by primary care physicians; b, current prescription for an antidepressant; c, diagnosis according to International Classification of Diseases, Ninth Revision (ICD-9) code, Diagnostic and Statistical Manual (DSM) and/or Research Diagnostic Criteria (RDC); d, assessment through clinician-rated and/or self-rated validated instruments, for example, patient health questionnaire (PHQ); e, diagnosis made through structured psychiatric interview. Diagnosis of diabetes was according to one of the following: a, diagnosis made by primary care physicians; b current prescription for a glucose lowering medication; c, diagnosis according to ICD-9 code; d, diagnosis according to laboratory result. (2) Type of intervention: although there was considerable variability in the exact nature of the intervention [29], we regarded that if it fulfilled the following four criteria as the intervention of collaborative care: a, a multi-professional patient care; b, a structured management plan; c, scheduled patient follow up; d, enhanced inter-professional communication [30]. (3) Type of control: we regarded it as control according to one of the following: a. no additional intervention was provided; b. usual care was provided in the control group; c. enhanced usual care was provided in the control group. (4) Type of outcome measurements: the studies were included if one of the following outcomes were reported in the original article: a, depression treatment response; b, depression remission; c, hemoglobin A1c (HbA1c) control; d, adherence to medication (including adherence of oral hypoglycemic agents and/or antidepressants); (5) Type of studies: only clinical randomized controlled trials (RCTs) or cluster RCTs were eligible; (6) Setting: primary care settings.

Studies that did not meet the above criteria were excluded. In addition, those duplicated publications were excluded. Two authors (YH TW) independently evaluated the articles for inclusion. Any discrepancies were resolved by further discussion and consultation of a third author (XW). The selection process by means of a flow chart was presented in Figure 1.

A standardized data extraction form was used. The following information was extracted: first author name, publication year, country of corresponding author, study design, study period, age, gender, ethnicity, sample size (intervention/control), length of follow-up, conflicts of interest (Table 1), inclusion criteria of patients, description of “collaborative care” and “control”, main outcomes (see Additional file 2). All the corresponding authors of studies included in qualitative synthesis were contacted by e-mail a maximum of 3 times to obtain additional information. However, no information was received from the original authors.

The risk of bias for the included studies was assessed using the Cochrane Collaboration’s tool for assessing risk of bias[34]. Two authors (XW TW) independently assessed the methodological quality of RCTs using the Cochrane risk of bias tools. Any discrepancies were resolved by a third author (RC).

Primary clinical outcomes of interest, evaluated at the end of follow up, 6 months and 12 months of follow up (during which patients were on collaborative care intervention) respectively, included depression treatment response (defined as 50% or more decrease in the Hopkins Symptoms Checklist-20 (SCL-20) score from base line); depression remission (defined as SCL-20 score less than 0.5); diabetes clinical outcomes (HbA1c values) (defined as HbA1c measures exposure of red blood cells to glucose during a 90-day period). Secondary outcomes were adherence (defined as the percentage of prescribed doses taken, calculated as the number of doses taken divided by the number of doses prescribed over the observation period * 100%) to antidepressant medication and oral hypoglycemic agent (which was dichotomized at a threshold of 80%) [35].

We used Revman software (version 5.2), which was available through the Cochrane Collaboration (http://​www.​cochrane.​org). Heterogeneity was quantified with a Chi-square heterogeneity statistic and by means of I square, with a predefined significance threshold of 0.1 [36]. If a significant trend for heterogeneity was observed, a random effect model via generic inverse variance weighting was used to combine the effect [37]. Otherwise, we used a fixed-effects model to calculate the pooled effects. Results were expressed as the relative risks (RRs) with 95% confidence intervals (CIs) for dichotomous variables and mean differences (MDs) with 95% CI for continuous variables. Results were considered statistically significant when P < 0.05. The possibility of publication bias was initially planned to be evaluated by funnel plots, but not adopted eventually either because of the limited number of trials included or the significant heterogeneity among trials [38]. No additional analysis was performed. No protocol of the present review has been published or registered.

We identified 1,467 citations (Figure 1). After excluding 103 duplicate records, two authors (YH XW) screened 1,364 titles and abstracts to identify the potentially relevant studies. Totally 217 full-text articles were assessed for eligibility. Of these, 10 studies were included in qualitative synthesis [2, 19, 23, 24, 26, 27, 31‐33]. A total of 8 articles met the final eligibility criteria for meta-analysis [19, 23, 26, 27, 31‐33]. The detailed selection process was described in Figure 1.

Table 1 summarized the basic characteristics of the included randomized controlled trials. There were 8 studies with a total of 2,238 patients with both depression and diabetes, which compared collaborative care with usual care [19, 23, 26, 27, 31‐33]. All trials were from the United States. One trial included only African Americans [27]. Two trials included only old patients (aged or above 50 and 60 respectively) [27, 28]. Three trials used enhanced usual care for the control group [19, 31, 32] and five trials used normal usual care [23, 26‐28, 33]. Duration of trials varied from 13 months to 30 months.

Random sequence generation [23, 28, 31, 32] and allocation concealment [19, 23, 28, 32] were described adequately in four studies respectively. Blinding of patients was not possible. Therefore, all studies had high risk of bias in blinding. Five studies [19,23,28,36,38,] described adequate blinding of outcome assessment. Two studies [19, 23] did not provide sufficient information for assessment of incomplete outcome data. Six studies [19, 23, 28, 31‐33] reported all expected outcomes. The detailed risk of bias of all included trials was shown in Table 2.

Four trials provided information on treatment response rate to calculate the overall effect size, with 1,096 patients [19, 23, 32, 33]. The length treatment ranged from 12 to 24 month. All of these trials reported an increased treatment response rate in the collaborative group, and one was significant (RR = 1.96, 95% CI = 1.38-2.78) [19]. Pooling of the mean proportion showed that 44.8% of patients in intervention group and 34.3% of patients in control group had treatment responses. The meta-analysis showed that collaborative care was associated with a significant increase in treatment response rate at the end of follow up (RR = 1.33, 95% CI = 1.05-1.68; P = 0.06 for heterogeneity; I² = 59%) (Figure 2A). The RR of 1.33 indicated a 33% relative increase in treatment response rate was added to collaborative care. There was significant heterogeneity in the studies.

Two trials reported information on depression remission rate [32, 33]. Of these, one showed a significant improvement on remission rate in collaborative care group at the end of the 24 month’s treatment (adjusted RR = 1.53, 95% CI =1.11-2.12) [32]. Another reported a non-significant improvement [33]. There were 552 patients to calculate the overall effect size. The meta-analysis indicated that there was a non-significant effect of collaborative care (RR = 1.15, 95% CI = 0.87-1.52; P = 0.33 for heterogeneity; I² = 0%) (Figure 2B).

Four trials evaluated treatment response rate at 6 months follow up, with totally 1,118 patients [19, 23, 32, 33]. All reported an increased treatment response rate in the collaborative group, and three were significant. Combining the four trials also demonstrated a statistically significantly beneficial effect of collaborative care at 6 months follow up (RR = 1.64, 95% CI = 1.28-2.10; P = 0.09 for heterogeneity; I² = 54%) (Figure 3A).

Four trials provided information to calculate the overall effect size of treatment response rate at 12 months follow up, with 1,344 patients [19, 23, 32, 33]. All of these trials reported an increased treatment response rate in the collaborative group, and two were significant. The meta-analysis showed that collaborative care was associated with a significant increase in treatment response at 12 months follow up (RR = 1.42, 95% CI = 1.14-1.76; P = 0.10 for heterogeneity; I² = 52%) (Figure 3B).

Two trials reported depression remission rate at 6 months follow up, with an evaluation of 595 patients to calculate the overall effect size [32, 33]. Both of the trials reported an increased treatment response rate in the collaborative group, neither of them was significant. However, the combined results indicated a significant increase with collaborative care (RR = 1.33, 95% CI = 1.01-1.75; P = 0.70 for heterogeneity; I² = 0%) (Figure 3C).

Two trials reported information on depression remission at 12 months follow up, with an evaluation of 569 patients to calculate the overall effect size [32, 33]. The meta-analysis resulted in a non-significant effect on depression remission at 12 months follow up (RR = 1.20, 95% CI = 0.93-1.55; P = 0.48 for heterogeneity; I² = 0%) (Figure 3D).

Seven trials reported HbA1c values. One trial [26] showed a significant reduction on HbA1c values in collaborative care group at the end of follow up. Another trial [24] showed that the percentage of patients with HbA1c values less than 7 was 60.9% and 35.7% in collaborative care group and usual care group respectively (no statistical analysis provided). The other five trials [19, 23, 27, 28, 32] provided information to calculate the overall effect size at the end of follow up, with totally 1,094 patients. Of these, only one was significant (MD = -0.48, 95% CI = -0.91--0.05) [19]. The analysis of the pooled data from the five trials demonstrated a reduction in HbA1c values at the end of follow up in favor of collaborative care (MD = -0.13, 95% CI = -0.46-0.19; P = 0.08 for heterogeneity; I² = 51%) although this was not statistically significant (Figure 4).

Four trials evaluated HbA1c values at 6 months follow up, with 1,101 patients to calculate the overall effect size [19, 23, 28, 32]. There was a non-significant reduction in HbA1c values in favor of collaborative care (MD = -0.06, 95% CI = -0.24-0.12; P = 0.31 for heterogeneity; I² = 16%) (Figure 5A).

Four trials evaluated HbA1c values at 12 months follow up, with 1,053 patients to calculate the overall effect size [19, 23, 28, 32]. The analysis of the pooled data from the four trials demonstrated a reduction in HbA1c values at 12 months follow up in favor of collaborative care (MD = -0.07, 95% CI = -0.28-0.13; P = 0.20 for heterogeneity; I² = 36%) although this was not statistically significant (Figure 5B).

Four trials reported rates of adherence to antidepressant medication, all provided information to calculate the overall effect size [23, 26, 27, 31]. There were a total of 891 patients. All the four trials reported a significant improvement on adherence to antidepressant medication. The meta-analysis also indicated a statistical significant positive effect on rates of adherence to antidepressant medication (RR = 1.79, 95% CI = 1.19-2.69; P < 0.001 for heterogeneity; I² = 84%) (Figure 6A).

Two trials provided information to calculate the pooled effect size on adherence to oral hypoglycemic agent, with a total of 238 patients [26, 27]. Both of them showed a significant improvement on adherence to oral hypoglycemic agent. The pooled data indicated that collaborative care was associated with a significant improvement of adherence to oral hypoglycemic agent (RR = 2.18, 95% CI = 1.61-2.96; P = 0.60 for heterogeneity; I2 = 0%) (Figure 6B).

Combining the results from different trials generated significant variations in outcomes. In meta-analyses of primary outcomes, there existed a significant heterogeneity in the overall results of depression treatment response. The values of I² are 54%, 52% and 59% for treatment response at 6 months, 12 months and the end of follow up respectively, which were all represented moderate levels of heterogeneity [42]. We therefore used random effects models to combine the data of these outcomes. Collaborative care is a kind of complex intervention with a considerable variability involving separate mechanisms, which is difficult to specify and define [29]. Nevertheless, compared with previous reviews [22, 43], which using broader inclusion criteria for collaborative care, this review was based on more precise definition of collaborative care [12, 30].

Last but not least, this meta-analysis was based on 8 RCTs with 2,238 patients from different primary care practices in the United States, which might not be representative of all patients with both diabetes and depression around the world. Further research is needed to help clarify whether collaborative care can be implemented outside the United States.