The Pathways Study: Title and subTitle BreakA Randomized Trial of Collaborative Care in Patients With Diabetesand Depression FREE

Approximately 10% to 15% of patients with diabetes mellitus meet criteriafor comorbid major depression.^{1 - 2} Depressionis a risk factor for development of diabetes mellitus,^{3 - 4} itis associated with adverse diabetes outcomes,^{5 - 9} anddiabetes may worsen the course of depression.¹⁰ Researchsuggests that the presence of comorbid chronic physical disease, such as diabetesmellitus, is a negative prognostic factor for depression treatment outcomes.^{11 - 12} Patients with diabetes mellitus andmajor depression, compared with those with diabetes mellitus alone, have beenshown to have higher symptom burden⁵ ; increasedfunctional impairment^{5 - 6} ; pooreradherence to diet, exercise, and taking medications^{2 ,5 - 6} ;higher hemoglobin A_1c (HbA_1c) levels⁷ ;and more diabetes complications.^{8 - 9} Giventhe high comorbidity of depression and diabetes mellitus and the potentialreciprocal adverse impact of these conditions, there is a need for treatmenttrials to assess whether enhancing recognition and treatment of depressionimproves diabetes and depression outcomes.

Several small tertiary care–based trials have shown that antidepressantmedication was more effective than placebo in treating major depression inpatients with diabetes mellitus.^{13 - 14} Onetertiary care–based trial¹⁵ also foundthat cognitive behavior therapy was more effective than a diabetes educationalintervention in relieving depressive symptoms and significantly decreasedHbA_1c levels in patients with comorbid major depression and diabetesmellitus. A more recent, larger trial¹⁶ amongelderly, depressed, diabetic patients with relatively low baseline mean HbA_1c levels did not find improvement in glycemic control resulting fromimproved depression outcomes. These studies may not have been adequately poweredto detect effects on glycemic control due to the sample size or low baselineHbA_1c levels.

Most patients with diabetes mellitus and major depression are treatedwithin primary care rather than specialty care. Primary care diabetic patientsare more likely to have type 2 diabetes mellitus, less likely to be treatedwith insulin, and likely to have fewer diabetic complications and medicalcomorbidities.^{2 ,5 - 6 ,17} Healthservices research in primary care systems has improved the quality of diabetescare with telephone or in-person case management interventions.^{18 - 19} However,these studies have not addressed treatment of comorbid major depression.^{18 - 19} Only a minority of patients withdiabetes mellitus and major depression receive adequate treatment for depression.²⁰ Population-based strategies to improve quality ofcare and outcomes in patients with diabetes mellitus in primary care havefound depression to be an important barrier to enhancing diabetes self-management.²¹ The present randomized controlled trial tested theeffect of a health services intervention aimed at improving quality of depressioncare on both depression and glycemic control outcomes among primary care patientswith diabetes mellitus and depression.

The Pathways Study was developed by a multidisciplinary team in theDepartment of Psychiatry and Behavoiral Sciences at the University of Washingtonand the Center for Health Studies at Group Health Cooperative (GHC), Seattle.The GHC is a nonprofit health maintenance organization with 30 primary careclinics in western Washington State. The study protocol was reviewed and approvedby institutional review boards at the University of Washington and the GHC.All participants gave written informed consent.

Nine GHC primary care clinics in western Washington were selected forthe study. We estimated that 150 participants in both the intervention andusual care arms (assuming 15% patient attrition) were required to have 80%power to detect as significant a 0.23 (SD, 0.7) difference in the mean scoreof the 20 depression items from the Hopkins Symptom Checklist-90 (SCL-90).²² We estimated that 162 participants in both the interventionand control arms (assuming 15% patient attrition) were required to have 80%power to detect as significant a 0.5% (SD, 1.65%) mean difference in HbA_1c values.

Case identification was facilitated by prior development by the GHCof a population-based diabetes registry²³ thatsupports patient care. A survey mailed to patients on the diabetes registryassessed age, sex, years of education, employment status, race, and maritalstatus. Questions about clinical status included the following: age at onsetof diabetes, duration of diabetes, current diabetic treatments, and diabetestreatment at onset of disease. When surveys were not returned, second andthird mailings and telephone reminders were used to achieve a final responserate of 61.7%.

Eligible patients were ambulatory, were English speaking, had adequatehearing to complete a telephone interview, and planned to continue to be enrolledin GHC during the next year. Psychiatric exclusions were as follows: (1) currentlyin care with a psychiatrist; (2) a diagnosis based on GHC's automateddiagnostic data of bipolar disorder or schizophrenia; (3) use of antipsychoticor mood stabilizer medication based on GHC's automated pharmacy data;and (4) mental confusion on interview, suggesting significant dementia.

The Patient Health Questionnaire 9 (PHQ-9) was used to screen for depression.^{24 - 25} The PHQ-9 (at a cutting score of≥10 with ≥5 symptoms scored as being present more than half of the days)has been found to have high agreement with structured interview in establishinga diagnosis of major depression.^{24 - 25} Althoughwe did not require patients to meet criteria for major depression, they wererequired to have a score of 10 or greater on the PHQ-9 in the initial screeningand persistent symptoms, as evidenced by an SCL-90 depression²² meanitem score of higher than 1.1 at a second telephone screen 2 weeks later.Patients were not excluded if they were taking antidepressants in the prior3 months as long as they had persistent symptoms.

A total of 9063 questionnaires were mailed, and 7841 patients were foundto meet initial eligibility criteria (Figure 1); 4839 questionnaires (61.7% of those eligible) were returned,and 1038 were eligible for baseline screening based on a PHQ-9 score of 10or greater. A total of 851 (82.0%) of the 1038 respondents were successfullyreached by telephone for baseline screening, and 375 met criteria for therandomized trial (based on a second screening SCL-90 score of >1.1). Only46 (12%) of 375 eligible patients refused to participate.

For measuring change in depression, the SCL-90 depression scale²² was chosen as the primary dependent variable, basedon previous studies that showed the high reliability, validity, and sensitivityto change of this measure.^{16 ,22 ,26} Giventhe higher percentage of patients with dysthymia than with major depression,response to treatment at 6 and 12 months was defined as 40% or greater reductionin SCL-90 scores, based on a National Institutes of Health consensus panelrecommendation regarding measurement of outcomes in dysthymia.²⁷ Wealso report 50% or greater reduction in SCL-90 scores at 6 and 12 months.We used the Patient Global Impression²⁸ scoreat 6 and 12 months as a subject-rated global assessment of improvement indepression since baseline. Patients rated their satisfaction with depressioncare at baseline and 6 and 12 months on a 5-point ordinal scale that ratedtreatment from poor to excellent.²⁶

Hemoglobin A_1c measures exposure of red blood cells to glucoseduring a 90-day period.²⁹ Study participantsagreed to blood draws at their GHC primary care clinic to measure HbA_1c at baseline and 6 and 12 months.

The GHC's computerized pharmacy and utilization records wereused to measure the number of specialty mental health visits, for examinationof refills of antidepressant medications, and to determine whether the patientreceived an adequate dosage based on evidence-based guideline standards for90 days or more within each 6-month period.³⁰ Thelowest doses in the ranges recommended in the Agency for Health Care Policyand Research Guidelines and in guidelines developed for newer agents wereused to define minimum dosage standards.^{31 - 32}

Computerized pharmacy records were used to compute a chronic diseasecomorbidity score known as RxRisk.³³ The RxRiskhas been found to be comparable to using ambulatory care groups³⁴ inpredicting total future health costs.

A measure was developed based on previous literature³⁵ todefine 2 aspects of diabetes using automated diagnostic, pharmacy, and laboratorydata: (1) diabetes complications and (2) treatment intensity required. International Classification of Diseases, Ninth Revision, codeswere used to identify 7 potential types of diabetic complications (retinopathy,nephropathy, neuropathy, cerebrovascular, cardiovascular, peripheral vascular,and ketoacidosis).³⁵ Pharmacy data regardingthe use of oral hypoglycemic agents and insulin indicated treatment intensity.

Using a computerized algorithm, patients were randomized to the interventionor usual care group. After completion of the baseline interview, interventionpatients were called by a nurse within 1 week to set up an appointment. Usualcare patients were given recommendations to work with their primary care physicianon issues related to depression. After randomization, telephone interviewswere provided at 3, 6, and 12 months by a telephone survey team who were blindedto intervention status.

The intervention was designed to improve quality of care and outcomesof depression but not to directly improve diabetes education or care. Theintervention was an individualized, stepped-care depression treatment programprovided by a depression clinical specialist nurse in collaboration with theprimary care physician.

Three half-time registered nurses were hired to implement collaborativecare treatment. Nurses received an initial 1-week training course on diagnosisand pharmacotherapy and an introduction to problem-solving treatment methodsbased on the intervention developed for the Improving Mood–PromotingAccess to Collaborative Treatment (IMPACT) study.^{36 - 37} Apsychiatrist (W.J.K.), primary care physician (E.H.B.L.), and psychologist(E.L.) participated in the training. An intervention manual from the IMPACTtrial was used to train nurses on collaborative care, stepped-care principles,pharmacology, and problem-solving treatment.³⁸

Nurses were also trained using the manual for problem-solving treatmentof depressive disorders in primary care (PST-PC) following the protocol describedby Hegel and colleagues,³⁹ which included didactics,role play, observation of a videotaped demonstration, and review of the PST-PCtreatment manual.⁴⁰ Each nurse was requiredto treat at least 4 depressed patients with 6 sessions of PST-PC during a2-month period. Each session was audiotaped, and sessions 1, 3, and 5 wererated using the PST-PC Adherence and Competency Rating Scale.^{39 - 40} Nurseswere required to have at least 3 audiotaped treatment sessions from differentpatients rated as satisfactory by the team psychologist (E.L.).

Patients were offered an initial choice of 2 evidence-based treatments:antidepressant medication or PST. Treatment included an initial 1-hour visitfollowed by twice-a-month, half-hour appointments (telephone and in person)in the acute phase of treatment (0 to 12 weeks).

A stepped-care approach was used in which patients received differenttypes and intensities of services tailored to their observed outcomes.⁴¹ If patients still had persistent depressive symptoms(<50% decrease in severity based on the PHQ-9) 10 to 12 weeks after initialtreatment with either PST or antidepressant medication, they could (1) switchto a second antidepressant with a different mechanism or side effect profile;(2) switch to the alternative treatment (from PST to medication or vice versa);(3) receive augmentation with PST or antidepressant medication with the firsttreatment they had received; or (4) receive a psychiatric consultation. Thischange in treatment at 10 to 12 weeks was labeled step 2 care. In situationswhere patients received 1 or more step 2 interventions, where symptoms persisted(<50% improvement), or where there was a lack of patient and cliniciansatisfaction with outcome after a second treatment (8 to 12 weeks), referralto specialty care by the GHC mental health system for longer-term follow-upwas made (step 3).

Once patients reached a significant decrease in clinical symptoms (≥50%decrement in symptoms), the nurse began continuation phase treatment, whichconsisted of monthly scheduled telephone contacts. For patients with persistentsymptoms or social isolation, nurses offered monthly continuation groups insteadof monthly telephone calls.

Each nurse had supervision twice a month with a team that included apsychiatrist (W.J.K., G.S., or E.W.), psychologist (E.L.) (pertaining to PST),and family physician (E.H.B.L.) to review new cases and patient progress.Nurses interacted regularly (via written notes and verbally) with the primarycare physician treating the patient. On alternate weeks, nurses reviewed casesby telephone with the psychiatrist supervisor. The psychiatrist supervisorregularly reviewed choices and dosages of medication and clinical responseand recommended changes, which the nurse discussed with the primary care physicianand patient. During the study, 45 audiotapes were reviewed by the team psychologistto provide ongoing feedback to nurses and to assess treatment fidelity.

Usual care patients were advised to consult with their primary carephysician regarding depression. Primary care physicians at the GHC frequentlyprescribe antidepressant medication and can refer patients to the GHC MentalHealth Services. Both intervention and usual care patients could also self-referto a GHC mental health care provider. Usual care for diabetes mellitus inthe GHC is provided by the primary care physician, with occasional supportfrom diabetes nurses for patients with persistently high HbA_1c levels.

χ² Analyses with correction for continuity and 2-tailedindependent group t tests were used to examine differencesbetween the intervention and usual care groups on demographic and clinicalvariables. Group differences in the adequacy of dosage of antidepressants,patient global improvement, and percentage of patients with at least a 40%and 50% decrease in depression were examined using logistic analyses to accountfor baseline group differences and to calculate odds ratios and 95% confidenceintervals (CIs). To examine treatment group trends over time (baseline and3, 6, and 12 months) in antidepressant medication use, we used mixed-effectlongitudinal logistic regression models with 2 random effects (intercept andtime) and 2 fixed effects (treatment group and its interaction with time).We initially tested models with time as both a random and a fixed effect.In both models, the intercept was always assumed to be random. The 2 modelshad similar levels of significance for their main effects and interaction.Comparing the log likelihoods for the 2 models revealed a better fit for themodel with time as a random effect (eg, slopes of the line over time beingrandom effects). For the clinical outcomes of the continuous SCL-90 depressionscale scores and HbA_1c values, mixed-effect continuous longitudinalmodels were used, following the same strategy described herein. In the eventof significant effects, planned post hoc analyses were performed using F teststo elucidate the findings (adjusted for baseline values).

Effect modification of the pattern of change in depression over timebetween the treatment groups was examined for patient subgroups with majordepression, patients with dysthymia, and those taking antidepressants at baseline.This was tested by individually examining the 3-way interaction of group bytime by effect modifier (major depression, dysthymia, or antidepressant use).

Figure 1 illustrates the flowof patient selection for the study, including reasons for exclusion at variouspoints. Of the 329 patients enrolled (164 intervention patients and 165 usualcare patients), the following percentages completed 3-, 6-, and 12-month assessments:3-month assessment, 151 (91.5%) intervention patients and 154 (93.3%) usualcare patients; 6-month assessment, 143 (87.8%) intervention patients and 149(90.9%) usual care patients; and 12-month assessment, 146 (88.5%) interventionpatients and 142 (86.1%) usual care patients. A total of 132 interventionpatients (80.5%) and 131 usual care patients (79.4%) completed all 3 assessments.

There were no significant differences between groups in any variable(Table 1). The population was middle-agedto elderly, with approximately 1 patient in 5 from a racial/ethnic minoritypopulation. Most patients had type 2 diabetes mellitus, and approximatelytwo fifths were treated with insulin. The mean glycosylated hemoglobin levelwas 8.0%, and the mean number of complications was 1.5. This population hada high rate of lifetime dysthymia (approximately 70%), and approximately twothirds met criteria for major depression. Only 13% of patients did not meetmajor depression or dysthymia criteria at baseline. Approximately half weretreated with an antidepressant medication within the last 3 months.

Most (97.6%) of the 164 intervention patients completed an initial visitwith a nurse. Intervention patients had a mean ± SD of 5.06 ± 3.43in-person visits and 5.87 ± 4.32 telephone contacts witha nurse, and 4.9% were seen for a consultation by a team psychiatrist. A totalof 84 intervention patients (51.3%) took medication and underwent PST, 13(7.9%) underwent PST only, 53 (32.3%) took medications only, and only 14 (8.5%)did not take medication or undergo PST. Of the 97 patients receiving PST,62 (64%) had 4 or more in-person treatment sessions.

Patients receiving the intervention were more likely to receive 4 ormore specialty mental health treatment visits (including nurse interventionvisits and GHC specialty mental health visits) compared with patients receivingusual care (67.7% vs 6.7%). Controlling for prebaseline 6-month rate of adequacy,the intervention group had significantly higher rates of adequate dosage inthe first 6-month period (57.3% for intervention vs 40% in the usual caregroup) and the second 6-month period (53.0% for intervention vs 38.2% in theusual care group) (Table 2).

A mixed-effect logistic regression model using baseline and 3-, 6-,9-, and 12-month refills of antidepressant medication based on automated datawas performed. The 2 random-effects models (slope and intercept) showed asignificant time × treatment group interaction (z = 3.30, P < .001)and nonsignificant main effects of time (z = 0.29)and intervention group (z = 1.08). At baseline,the usual care and intervention groups did not differ in adherence (43.6%vs 35.4%, χ²₁ = 2.02, P = .16). At each of the follow-up assessments, the interventiongroup had significantly greater medication adherence than the usual care group,controlling for baseline adherence, with the odds ratios ranging from 2.18to 3.20 (Table 2).

At baseline, rates of satisfaction in the intervention and usual caregroups were very similar, but at 6 and 12 months, the intervention group reportedsignificantly greater satisfaction (Table 2)than the usual care group.

Mixed-effect regression models using baseline and 3-, 6-, and 12-monthfollow-up SCL-90 continuous data were performed. The 2 random-effects models(slope and intercept) showed a significant group × time interaction(z = 2.84, P = .004);both the main effects of time (z = 8.92, P < .001) and intervention group (z = 2.14, P = .03) werestatistically significant. Figure 2 showsthe depression means over time for the treatment groups (all follow-up meansadjusted for the baseline SCL-90 score). The baseline SCL-90 depression meanscores (F_1,327 = 2.21, P = .14)and the 3-month assessment (F_1,302 = 1.45, P = .23) did not differ significantly between the groups.However, by 6 months, the intervention group had a significantly lower adjustedmean than the usual care group (F_1,290 = 4.11, P = .04), and this difference continued to bestatistically significant at the 12-month assessment (F_1,285 = 4.96, P = .03). The average change from baseline to6 months was 0.39 (95% CI, 0.28-0.49) for the usual care group and 0.56 (95%CI, 0.46-0.67) for the intervention group. The average change between baselineand 12 months was 0.44 (95% CI, 0.33-0.56) for the usual care group and 0.65(95% CI, 0.54-0.76) for the intervention group.

Results of our effect modification analyses showed trend-level effectmodification for patients treated with an antidepressant in the 3 months beforerandomization (z = 1.59, P = .11) and nonsignificant modification for major depression(z = 0.65, P = .52)or dysthymia (z = 0.45, P = .65). The trend-level modification by antidepressantsshowed a greater intervention vs usual care treatment effect over time forthose who had not had previous exposure to antidepressants.

Table 3 shows significant differencesin the percentage of patients with a 40% decrease in SCL-90 depression scoresbetween the intervention and usual care groups at 12 months and similar butnonsignificant trends at 6 months. Approximately 10% more intervention vsusual care patients also improved 50% or more from baseline on SCL-90 scoresat 6 and 12 months, but these trends were not statistically significant.

At 6 months, a significantly higher percentage of the intervention patientsreported improvement on the Patient Global Impression measure from baseline,compared with usual care patients. These differences were greatest at the12-month assessment, when 71.9% of the intervention group reported improvementin their depression from baseline, compared with 42.3% of the usual care patients(Table 3).

A mixed-effect regression model compared HbA_1c values atbaseline and 6 and 12 months. There was no statistically significant group× time interaction (z = 0.60, P = .55) and no main effect of intervention;however, the time effect was statistically significant. We refit the modelwithout the interaction term, and there was no statistically significant treatmentgroup effect (z = 0.89, P = .37), but there was a statistically significant timeeffect (z = 3.65, P < .001). Figure 3 shows that HbA_1c levels decreasedover time for both groups: baseline (overall mean ± SD),7.99% ± 1.55%; 6-month assessment, 7.58% ± 1.47%;and 12-month assessment, 7.64% ± 1.57%. The follow-up meanswere adjusted for baseline HbA_1c levels. There were no statisticallysignificant group differences at any of the time points: baseline, F_1,315 = 0.24, P = .62;6 months, F_1,282 = 0.67, P = .41;and 12 months, F _1,274 = 0.61, P = .44.

Compared with usual care patients, the Pathways intervention patientsreceived more adequate depression treatment, were more satisfied with theircare for depression, and showed significantly greater improvements in depressivesymptoms during a 12-month period. These results add to the expanding literaturethat shows that depression can be effectively treated in the context of majorchronic medical illness.⁴²

Unlike many trials in younger adults, where differences between interventionand control patients in depressive outcomes at 4 to 6 months tended to decreaseby 9 to 12 months,^{26 ,43} the datafrom this trial suggest sustained intervention effects at longitudinal follow-upfor 1 year. A similar pattern of sustained benefits during a 12-month periodwas also seen in the recently published IMPACT trial, which tested a similarnurse intervention vs usual care in elderly patients.³⁶ Theseresults may reflect the fact that both the current study and the IMPACT trial³⁶ built in a continuation phase of the interventionin which nurses continued to monitor adherence and outcomes by telephone duringa 1-year period.

An extremely high rate of dysthymia (approximately 70%) and prior depressiontreatment was found in these patients. In contrast, we found rates of dysthymiaof 20% to 30%^{26 ,43} in primarycare studies that tested collaborative care depression interventions withmixed-age populations in the same health maintenance organization. The datathat show high rates of chronic depression are consistent with other datathat have shown that chronic medical illness is a negative prognostic factorin recovering from depression.^{11 - 12} Althougha higher percentage of intervention patients compared with controls improvedduring the 12 months, approximately 45% of intervention patients still hadsignificant depressive symptoms, suggesting continued need for improved interventionmodels. The high rates of coexisting chronic depression in this populationsuggest that many patients may benefit from longer-term interventions thatcombine medication and evidenced-based psychotherapy.⁴⁴ Arecently reported trial that enrolled patients with 2 or more years of depressionfound that a combined antidepressant medication and cognitive behavior analysistherapy intervention was more effective than either medication or cognitivebehavior analysis therapy alone in improving outcomes.⁴⁴

In this trial, the enhancement in quality of care and outcomes of depressionwas not accompanied by significant differences in HbA_1c levelsbetween intervention and control patients during a 12-month period. Only 1of 4 prior trials with patients with depression and diabetes mellitus hasfound that an effective depression intervention was associated with improvedHbA_1c levels.^{13 - 16} However,the mean baseline HbA_1c level in that efficacy study was approximately10.0% vs approximately 8.0% in the current trial.¹⁵ Also,the controls in that trial received diabetes education only,¹⁵ whereasapproximately half of the usual care controls in this study received antidepressanttreatment, which probably decreased intervention vs usual care differencesin depression outcomes. Similar to research showing that focusing only onbiomedical aspects of diabetes is not an optimal treatment for patients withcomorbid diabetes mellitus and depression,²¹ ourresults suggest that the alternative approach of focusing on depression careonly is not likely to achieve optimal diabetes outcomes. Given that depressedpatients with diabetes mellitus have more severe disease² andhigher numbers of behavioral risk factors (obesity, smoking, and sedentarylifestyle) than diabetic patients without depression,^{2 ,45} anintegrated biopsychosocial intervention program that focuses on improvingboth depression and diabetes mellitus management may be needed to improveclinical outcomes in both of these chronic illnesses.

This randomized trial was completed in 1 large health care system inthe Pacific Northwest, limiting generalizability. Participants had enhancedusual care, since routine care patients were encouraged to discuss depressionwith their primary care physician. Primary care physicians treated both interventionand control patients, leaving room for a spillover effect due to potentialphysician improvements in knowledge and skills in treating depression. Therelatively low baseline HbA_1c levels in this primary care populationmay have limited the effectiveness of the intervention on glycemic control.These potential biases would tend to result in underestimation of the effectivenessof the intervention, not overestimation.

In conclusion, the collaborative care model used in this study seemsto be a feasible and effective approach for improving the quality of careand outcomes of depression in primary care patients with diabetes mellitus.Enhanced depression care did not result in improved glycemic control. Furtherresearch is needed to determine how to improve diabetes outcomes in patientswith depression and diabetes mellitus.

Correspondence: Wayne J. Katon, MD, HealthServices Research and Psychiatric Epidemiology, Department of Psychiatry andBehavioral Sciences, Campus Box 356560, University of Washington School ofMedicine, 1959 NE Pacific, Seattle, WA 98195-6560 (wkaton@u.washington.edu).

Accepted for Publication: April 29, 2004.

Funding/Support: This study was supported bygrants MH4-1739 and MH01643 from the National Institute of Mental Health ServicesDivision, Bethesda, Md (Dr Katon).