Five-Year Impact of Quality Improvement for Depression:  Results of a Group-Level Randomized Controlled Trial FREE

Depressive disorders are a leading cause of disability worldwide.^1- 3 Yet, many depressedpatients, especially historically underserved minority groups, such as Latinosand African Americans, do not receive guideline-concordant care. Depressedprimary care patients who are Latino or African American tend to have poorerhealth outcomes, without practice interventions to improve quality, than dowhites.^2,4- 7 Qualityimprovement (QI) interventions for depression within primary care can improvequality of care and health outcomes over 6 to 28 months and yield cost-effectivenessratios relative to usual care (UC) that are comparable to commonly used medicaltherapies; however, not all interventions, particularly those focused onlyon providers, improve outcomes.^8- 15 Furthermore,interventions of longer duration may provide larger long-term benefits.¹⁶

Partners in Care (PIC) is a group-level randomized controlled trialof 2 QI interventions, compared with UC, for depressed primary care patients.¹⁷ Both interventions supported local teams for interventionmanagement and provider education and nurses for patient education and assessment.One intervention trained nurses to support medication management by primarycare providers (QI-meds), and the other trained local therapists in cognitivebehavior therapy (CBT) and lowered the patient copay for use of those therapists(QI-therapy), but patients could have any or no treatment in either intervention.During 6 to 24 months of follow-up, relative to patients receiving UC, thosein QI programs had a lower likelihood of probable depressive disorder, betterquality of life, fewer "depression burden days," and a higher rate of employment.^17- 19 In the first year,clinical outcome benefits were greatest for Latinos and African Americanscombined, but employment gains were significant only among whites.²⁰ Health benefits continued into the second year underQI-therapy,¹⁹ and increased antidepressantmedication use continued into the second year for QI-meds.²¹ TheQI programs increased direct costs $300 to $400 for 24 months, but the interventionshad 2-year cost-effectiveness ratios comparable to those of commonly usedmedical treatments.¹⁸

This study examines the effects of PIC interventions on health outcomesand quality of care 57 months after study enrollment. To our knowledge, nostudy has examined such long-term effects of short-term QI for any medicalcondition. Patients who participate in short-term QI could experience long-termbenefits if they acquire new skills, such as learning to seek appropriatecare when sick, that help them prevent or manage their depression. We hypothesizedlong-term health outcome benefits for the QI groups. Since improved healthmay reduce the need for services, we did not expect long-term QI effects onuse of treatments or services overall; instead, we expected QI to reduce unmetneed for appropriate care for depression. Nevertheless, practice and payerconcerns about increased utilization of services in the long run may serveas a barrier to implementing effective QI programs, so we report QI effectson long-term probabilities of having different health care services as a contextfor our main results.

Based on our previous findings, we hypothesized that any average effectson health outcomes at 5 years would favor QI over UC and that there wouldbe greater benefits of QI on outcomes in Latinos and African Americans combinedthan in whites.^17- 21 Wedid not expect any mean effects of QI on the use of services either for allparticipants or among Latinos and African Americans.²⁰ Lowlevels of utilization of, satisfaction with, and quality of care for mentaldisorders among Latino and African Americans have been the focus of nationalpolicy and research attention.^6,20,22 Disparitiesin mental health care for other ethnic groups have been studied less often,and ethnic minorities other than Latinos and African Americans were representedin insufficient numbers in PIC for separate study.^6,20

The data are from PIC, a group-level, randomized controlled trial ofpractice-implemented QI programs for depression.^17- 21,23 Sixmanaged care organizations, 46 of 48 eligible primary care practices, and181 of 183 eligible primary care clinicians participated. Within organizations,practices were matched into blocks of 3 clusters based on specialty mix, patientsocioeconomic and demographic factors, and having mental health specialistson-site. Practice clusters were randomized within blocks to UC (mailing ofwritten practice guidelines to medical directors), QI-meds, or QI-therapyby using a random-number generator.

Study staff screened 27 332 consecutive patients in a 5- to 7-monthperiod for each practice, between June 6, 1996, and February 28, 1997. Patientswere eligible if they intended to use the practice for 12 months and if theyhad positive screening results for current depressive symptoms and probablemajor depressive or dysthymic disorder in the past year, using lead-in itemsfrom the World Health Organization's 12-month CompositeInternational Diagnostic Interview (CIDI).²⁴ Patients were ineligible if they were younger than18 years, if they were not fluent in English or Spanish, or if practice providersor therapists participating in the interventions were not covered by theirinsurance. The study was approved by the institutional review boards of RANDand the practices.

Of individuals completing the screening, 3918 were potentially eligible,but many left the clinic (eg, for a laboratory test) before their insurancestatus could be checked; 2417 patients were available for confirming insurance,and 241 (10%) had ineligible insurance. Of individuals reading informed consent,1356 (70%) enrolled: 443 in UC, 424 in QI-meds, and 489 in QI-therapy (Figure 1). Enrollment did not differ by ethnicity(χ²₃ = 3.5; P = .33).

The interventions are published elsewhere²⁵;all intervention materials are posted at http://www.rand.org/organization/health/pic.products/order.html. We estimated each managed care organization's participation costsbased on expected enrollment and per case costs for health care and researchtime. We provided half that amount ($35 000-$70 000 per managedcare organization). The interventions provided practices with training andresources to initiate and monitor QI programs adapted to local goals and resources.Patients and clinicians retained choice of treatment and use of interventionmaterials; the randomization was to resources for improved care, not mandatedtreatment. The study provided training materials and staff training and offeredlimited implementation support.

For both interventions, local teams were trained in a 2-day workshopto educate primary care clinicians through lectures, detailing, or audit andfeedback and to supervise staff and conduct team oversight. Practice nurseswere trained to help with patient assessment, education, and activation fortreatment. Practice teams were given patient education pamphlets, videotapes,tracking forms, clinician manuals, lecture slides, and pocket reminder cards.The materials described guideline-concordant care for depression; for example,they presented psychotherapy and antidepressant medication therapy as beingequally effective for most patients, encouraged attention to patient preferences,and advised adjusting treatment plans to patient needs and preferences.^26- 27

In QI-meds, nurse specialists were trained to support medication adherencethrough monthly visits or telephone contacts for 6 or 12 months, randomizedat the patient level. In QI-therapy, practice therapists were trained to provideindividual and group CBT,^28- 29 availableat the primary care copay (approximately $5-$10) for 6 months after enrollment.All patients could have other therapy at their usual copay (approximately$20-$35). Supervision was provided by local experts, who were assisted bystudy experts. In all conditions, patients could have medications, therapy,both, or neither. Previously, for example, we showed that in the first andsecond 6 months of the study, 40% and 35% of QI-therapy patients, respectively,received some antidepressant medications and 38% and 34% received at least4 psychotherapy sessions; 52% and 43% of QI-meds patients received some antidepressantmedications and 30% and 29% received at least 4 psychotherapy sessions.^21,30

The study was advertised in each practice using posters featuring physiciansand staff from that practice. The patient videotape featured minority physiciansand patients. Study CBT was specifically developed for low-income minorities.^28- 29 The CBT training was provided inEnglish and Spanish, and practices were asked to provide bilingual depressionnurse specialists and therapists. Intervention staff received cultural sensitivitytraining, and physician training discussed issues in providing depressiontreatment to minorities.

Patients were asked to complete the screening instrument, a telephone CIDI for depression and economic status, and a mailed surveyat baseline. We mailed follow-up surveys every 6 months for 24 months, withtelephone follow-up, and we conducted a telephone survey at 24 months. At57 months, we fielded a telephone follow-up. We would expect any survey modeeffect (telephone vs self-administered via mail) to be equivalent across interventiongroups. Completion rates relative to all initial enrollees (N = 1356) are95% (n = 1286) for the baseline telephone survey and 73% (n = 991) for the57-month survey, representing 86% of 1152 participants still living and enrolled.

We used indicators for QI-meds and QI-therapy compared with UC.

We repeated the screening measure of probable major depressive disorderat follow-up, eliminating the dysthymic disorder item because it refers tothe previous 2 years.¹⁷ The screening measurehas a positive predictive value of 55% against the 12-month CIDI.⁹ Mental health–related qualityof life was measured using the Mental Health Composite Score (MCS12), a globalmental health scale scored from the Short-Form 12^31- 32;scores were normalized to a population mean (SD) of 50 (10), with higher scoresindicating better health.^17- 20

We used self-reports of 4 or more specialty visits as an indicator ofcounseling and of 2 months or more of antidepressant medication use duringthe 6 months before follow-up. At 6-month follow-up, having 1 or both typesof treatments greatly improved health and employment outcomes.³³ Atlong-term follow-up, patients may no longer require treatments if their healthhas improved. To account for this, we use an indicator of "unmet need forappropriate care," defined as having probable major depressive disorder inthe 6 months before follow-up but not receiving counseling or antidepressantmedication, contrasted with persons without probable disorder or who receivedtreatment.¹⁷

As a context for our main results, we used survey items to develop countsof general medical visits for any purpose, general medical visits for emotionalproblems, specialty mental health visits, and overnight hospital stays forany purpose. We focus on dichotomous indicators of each type of use, but conclusionswere similar using counts of visits and alternative transformations of counts.

For analyses of the main effect of intervention status on outcome, ethnicityis a covariate, and we contrast whites and all minorities grouped togetherby patient self-designation. For the analysis of QI effects on health disparities,we contrast Latinos and African Americans (combined) with non-Hispanic whites.Latinos and African Americans are 2 ethnic groups with strong national evidenceof mental health care disparities.^6,34- 35 Asa combined group, Latinos and African Americans had a favorable clinical outcomeof QI relative to whites at 1-year follow-up in PIC.²⁰ Combiningthese 2 groups affords greater statistical precision,^36- 37 butin sensitivity analyses, we compared QI effects for African Americans andLatinos relative to whites separately, with qualitatively similar conclusionsas the combined comparisons. We base the main conclusions on the combinedcomparisons but provide estimates for African Americans and Latinos separatelyto facilitate interpretation.

Other ethnic groups in PIC included Asian Americans (n = 17, 3 of whomself-designated themselves also as white), Native Americans (n = 10), andwhites/Native Americans (n = 40). Qualitative descriptive findings for NativeAmericans and white/Native Americans combined suggest levels of treatmentand outcomes similar to those for whites; for example, the weighted percentagewith appropriate care was 48.2% at baseline and 45.2% at follow-up, and changesin mental health status from baseline to follow-up were similar across interventiongroups (eg, for MCS12, increases of 7-9 scale points). Consistent with recentexpert recommendations,^6,38 wedid not consider it appropriate to aggregate all ethnic minority groups intoone category; instead, we focused on 2 ethnic minority groups with balancedrepresentation across intervention groups, national data on mental healthcare disparities, and previous outcome data from PIC.^20,34,38

Using the patient screening instrument, we measured age; sex; education(less than high school, completed high school, some college, completed college,or more); a count of having 0, 1, 2, or more than 2 of 19 chronic medicalconditions; and the global mental and physical health scores on the Short-Form12.^31- 32 We used data from thescreening measure and the baseline CIDI^17,24 tocategorize patients as having 30-day depressive symptoms plus 1-year disorder(ie, "disorder") vs symptoms only. Using items modeled after the Health andRetirement Survey,³⁵ we developed a baselinehousehold wealth variable. We used indicators for practice group.

We conducted patient-level, intent-to-treat analyses, maintaining participantsin the intervention group of the clinic in which they were screened for enrollment.For a group-level randomized trial with a moderate number of groups, controlfor baseline differences that may affect outcomes or that predict nonresponseis recommended.³⁶ To test the effects of QI,we estimated multivariate regression models with intervention status as theindependent variable, controlling for ethnicity (white vs minority) and allthe covariates listed in the previous subsection, which include factors predictiveof nonresponse in this study (see the "Results" section). For dichotomousmeasures, we estimated logistic regression models. For MCS12, we conductedordinary least squares regression on untransformed scores. To examine differencesin QI effects for African Americans and Latinos combined vs whites, we conductedan additional set of analyses with intervention status, ethnicity group, andintervention-ethnicity interactions as the main independent variables, excludingAsian Americans (n = 17), Native Americans (n = 10), and whites/Native Americans(n = 40) (see the "Race or Ethnicity" subsection). We also conducted sensitivityanalyses comparing whites with Latinos, and whites with African Americans,and stratifying the model interacting white and African Americans/Latinoscombined by baseline disorder status (1-year disorder vs symptoms only). Weadjusted for the clustering of patients within providers and clinics usinga bias reduction method for standard error estimation developed by Bell andMcCaffrey.³⁷

The significance of comparisons by intervention status and tests ofinteractions were based on regression coefficients. Consistent with recommendationsfor group-level randomized trials,³⁶df are based on the number of practices (43 had patientsin the follow-up). The presented results are standardized predictions generatedfrom each regression model. We used the regression parameters and each individual'sactual values for covariates other than intervention status to calculate predictedoutcomes assuming the patient had been assigned to each intervention condition,in turn; we then calculated the mean prediction under each intervention condition.The analytic sample size for the main effect analysis is 991. For analysisof whites vs African Americans and Latinos, the total is 924. Specific modelsdiffer by a few individuals owing to missing data in the dependent variables.We used multiple imputations for missing items in independent variables; weaveraged predictions from 5 randomly imputed data sets and adjusted standarderrors for uncertainty due to imputation.^39- 40

Nonresponse weighting is applied to the observed data, weighting eachobservation by the reciprocal of the probability of study enrollment and follow-upresponse, so as to facilitate generalization to the sample eligible for thestudy. For sensitivity analyses, we conducted unweighted, unadjusted analyseswith no change in qualitative results or substantive conclusions.

We applied a 2-tailed significance level of P≤.05,and we report actual P values. As a first study ofsuch long-term outcomes, our analysis is somewhat exploratory; for most hypotheses,findings are consistent across measures, suggesting that formal correctionfor multiple statistical comparisons may be too conservative.

Table 1 indicates that theintervention groups had similar ethnic distributions at baseline among participantscompleting 57-month follow-up. However, baseline sex and global mental health(MCS12) differed by intervention status, with QI-therapy patients more likelyto be female and to have worse global mental health; a similar health differencewas reported for the baseline sample.^17- 18

Additional factors related to nonresponse at the 57-month follow-upwere ethnicity (African Americans were less likely to respond than whites, t = 2.89; P = .006), poorer physicaland mental functioning, and education (high school graduates were less likelyto respond than individuals with some college). Intervention status did notsignificantly predict nonresponse (F_2,41 = 1.76; P = .18), and there was no statistically significant interaction betweenethnicity (African American, Latino, other, white) and intervention status(combined QI vs UC) in follow-up response (F_3,40 = 1.52; P = .22, unweighted analysis).

Relative to UC, QI-therapy lowered the adjusted rate of probable depressivedisorder at follow-up by 7.4 percentage points (95% confidence interval [CI],0.1 to 14.6 percentage points) and QI-meds lowered the rate by 5.7 percentagepoints (95% CI, −0.5 to 11.9 percentage points), with QI-therapy significantat P = .05 (Table2). Pooling QI-meds and QI-therapy vs UC yields a 6.6–percentagepoint difference (95% CI, 0.4 to 12.8 percentage points, t₄₂ = 2.08; P = .04). As seen in Table 2, the effect of each QI interventionon the MCS12 score is small and nonsignificant. The reductions in unmet needwere also not statistically significant.

Table 3 indicates that underUC, African Americans and Latinos combined have worse health outcomes thanwhites. Participation in QI-therapy lowered the adjusted rate of probabledisorder among African Americans and Latinos but had little effect among whites;the interaction is significant at P = .04. Resultsare qualitatively similar for QI-meds but are not statistically significant;pooling the 2 QI groups vs UC yields a 16.1–percentage point reduction(95% CI, 5.2 to 27.0 percentage points, t₄₂ = 2.74; P = .009) in probable disorder amongAfrican Americans and Latinos. Participation in QI-therapy improved globalmental health (MCS12) among African Americans and Latinos by 4.5 scale points(95% CI, 0.5 to 8.5 points), a moderate to large effect.³² Thereis no effect of QI-therapy on MCS12 among whites; the interaction term issignificant at P = .02.

The QI-therapy intervention reduced unmet need for appropriate careamong African Americans and Latinos, an effect comparable to the pooled QIeffect during the active intervention period.¹⁷ Useof QI-meds also reduced unmet need for appropriate care among African Americansand Latinos, but with weaker statistical significance (P = .07); pooling intervention groups yields a reduction of 11.4 percentagepoints (95% CI, 3.1 to 19.7 percentage points, t₄₂ = 2.65; P = .01), relative to UC. No noteworthyQI effect on unmet need for appropriate care occurred among whites, but interactionterms are not statistically significant.

Table 3 illustrates thatresults were qualitatively similar for Latinos vs whites and for African Americansvs whites, but intervention effects were especially strong among African Americans.For example, the effect of QI-therapy on probable disorder is significantfor this group relative to whites (t = 3.07; P = .004).

Sensitivity analyses stratified by baseline disorder status suggestqualitatively similar effects of pooled QI on reducing probable disorder forthe combined minority groups, regardless of baseline disorder status. Thetrend for more favorable QI effects among minorities, relative to whites,holds among those with baseline disorder (n = 565, t =1.66, P = .10) but not those with symptoms only (n= 360, t = 0.15, but P =.88). Similarly, the greater effect of the interventions on reducing unmetneed among minorities, relative to whites, applies primarily among those withbaseline disorder (t = 3.25, P = .002), not symptoms only (t = 0.76, P = .45).

As expected, there were no noteworthy or statistically significant QIeffects on the probability of having any type of visit or hospitalizationor on counts of visits or hospital stays for the whole sample or for whitesor African Americans and Latinos combined.

We found a modest mean reduction across the whole sample in the likelihoodof having probable depressive disorder at follow-up for pooled QI interventionscompared with UC, equivalent to a 3–percentage point reduction in theprevalence of depressive disorder.¹⁷ Even sucha modest reduction in continued depression could have important public healthimplications given the high prevalence and effect on morbidity of depression.Our findings build on promising results from other studies^15- 16 showingsubstantial health outcome improvements from QI for depression in primarycare at 24 to 28 months.

This result, however, was mainly due to a very large effect of QI, especiallyQI-therapy, on health outcomes among Latinos and African Americans at 5-yearfollow-up. Without intervention, 56% of minorities and 36% of whites had probabledepressive disorder at follow-up, indicating a large outcome disparity. WithQI-therapy, however, outcomes were similar for whites and for the combinedAfrican American and Latino group. The effect of QI-therapy on the globalmental health of African Americans and Latinos, almost 0.5 SD, is within arange considered clinically meaningful on the Short-Form 12.³² Wealso found that QI-therapy reduced unmet need for appropriate care among AfricanAmericans and Latinos at 57 months, which might lead to subsequent healthoutcome improvements, a subject for future study. Because the minority individualsparticipating in PIC preferred psychotherapy for treatment of depression atbaseline,⁴¹ QI-therapy may have been a particularlyculturally appropriate intervention. These findings are relevant to nationaldebates over how to reduce disparities in the context of few evidence-basedapproaches.^6,22

Miranda et al²⁰ observed a similar differentialQI effect during the first year of follow-up and proposed that this findingmay have been due to the lower previous exposure of minorities to appropriatedepression treatment. The observation of this effect at 5-year follow-up,however, suggests that the interventions may have also addressed fundamentalcauses of health care disparities. Factors cited in the literature as contributingto health care disparities²² include (1) culturaldiscordance of provider and patient,^42- 43 whichthe cultural sensitivity training in PIC may have attenuated somewhat; (2)provider uncertainty about diagnosis or treatment benefit, enhancing relianceon previous conceptions,⁴² which PIC educationmay have reduced; (3) language or other incompatibilities between patientsand providers,^44- 45 which theuse of bilingual intervention staff [ie, depression nurse specialists andtherapists] reduced; (4) health care system problems in settings frequentedby minorities, such as poor services availability^46- 47 anddifficulty accessing specialty referrals,⁴⁸ whichPIC resources may have reduced; (5) interpersonal communication and participationstyle factors,^42,49 which PICtraining in patient preference may have improved; (6) discrimination and racism,^50- 51 which PIC clinician training acknowledgedbut did not directly address; (7) overuse of services by whites,^52- 53 whichPIC indirectly addressed through advocating tailoring treatment to need; (8)insurance coverage differences,^54- 56 whichPIC partially addressed through reducing the CBT copay; (9) differences inhaving a regular provider,⁵⁶ which PIC didnot address; and (10) differences in clinician diagnosis and treatment,^57- 59 which PIC standardizedthrough clinician training. Causes of disparities could be mitigated by featuresof the underlying collaborative care approach, such as enhancing leadershipcommitment to support appropriate care.^60- 61 Supportfor this view is provided by recent studies showing the effectiveness of astepped-care program for major depression among impoverished women in Chile⁶² and of group interpersonal psychotherapy for depressionin rural Uganda.⁶³

The fundamental question posed by our findings is why there were long-termhealth benefits of QI at all. Figure 2 illustrates7 alternative pathways for long-term (time 2) gains after initial (time 1)benefits. The first pathway is a main QI effect on the use of treatments atlong-term follow-up, that is, quality of care; we found no evidence of suchmain effects. The second pathway is a targeted long-term QI effect on reducingunmet need for appropriate care. Our findings support this explanation amongAfrican Americans and Latinos. The third pathway is an indirect effect throughQI reducing risk factors for depression, such as the number of episodes. ThePIC interventions reduced the likelihood of probable depressive disorder inthe first year, particularly among African Americans and Latinos.²⁰ The fourth pathway is an early (first 2 years) QIeffect on increasing maintenance treatment, which would reduce future recurrences.Unützer and colleagues²¹ found increaseduse of antidepressant medications through the second year under QI-meds, providingpartial support. Pathways 5a and 5b (Figure2) reflect indirect QI effects on outcomes through patient learning(eg, symptom self-management) as a result of education or experiences withcare. We cannot test these pathways with our data. Pathway 6 is a long-termeffect on clinician learning. Although QI increased clinician knowledge at18 months,⁶⁴ most patients switched providersand most practices switched management by 2 years, reducing the plausibilityof clinician-based explanations. The seventh pathway is a persistent effectof QI on barriers/incentives, such as reduced long-term financial barriersto care. We do not have the data to test this pathway, and we have no reasonto think that QI would have had such persistent structural effects. Thus,PIC findings support several explanations for long-term QI effects, especiallyamong African Americans and Latinos. Further exploration of such pathwaysin future studies may help design interventions that enhance long-term benefits.

Our findings are subject to limitations. Measures are based on self-report.We recruited patients from specific community-based managed care practices.Underserved minorities, especially Latinos, are more likely than whites tobe uninsured³⁴; our findings apply mostly toinsured patients, but one public sector site had uninsured patients. We hadrelatively few African Americans, who had lower response at long-term follow-up,but intervention groups were balanced in ethnic distribution, and there wasno statistically significant interaction between ethnicity and interventionstatus in the probability of response. Our conclusions are largely based onpooled analyses of Latinos and African Americans; however, separate analysesshow large effects of these QI programs in African Americans. This exploratoryfinding should be reexamined in larger representative groups of African Americans.The effectiveness of QI should also be examined in ethnic groups we did notspecifically study, such as Asian Americans, Pacific Islanders, and NativeAmericans.

We conclude that implementing QI programs for depressed primary carepatients can lead to long-term improvements in health and reduced unmet needfor appropriate care among Latinos and African Americans, reducing substantialhealth outcome disparities in UC for these groups relative to whites. ThePIC therapy-resource intervention seemed particularly effective in this regard.In this study, gains for minority groups led to an overall modest health improvementacross participants, without noticeable long-term increases in the use ofservices. In the long run, QI improved equity, a major goal of medicine,⁶⁵ after being effective and cost-effective in the shortrun.¹⁸

Corresponding author: Kenneth Wells, MD, MPH, RAND, 1700 Main St,Santa Monica, CA 90407-2138 (e-mail: kwells@ucla.edu).

Submitted for publication May 19, 2003; final revision received September4, 2003; accepted September 23, 2003.

This work was funded by grants R10MH57992 and P50MH54623 from the NationalInstitute of Mental Health, Bethesda, Md, and by grant R01-HS08349 from theAgency for Healthcare Research and Quality, Rockville, Md.

We thank Maureen Carney, MS, for coordination of follow-up; BarbaraLevitan, BA, for oversight of the survey; Bernadette Benjamin, MS, for excellentprogramming support; the practice organizations participating in this studyfor providing access to their expertise and patients, implementing interventions,and providing in-kind resources: Allina Medical Group (Twin Cities, Minn),Patuxent Medical Group (Maryland), Humana Health Care Plans (San Antonio,Tex), MedPartners (Los Angeles), PacifiCare of Texas (San Antonio), and Valley-WideHealth Services (Colorado); their associated behavioral health organizations:Alamo Mental Health Group (San Antonio), San Luis Valley Mental Health/ColoradoHealth Networks, and Magellan/GreenSpring Behavioral Health (Maryland); andthe clinicians and patients who contributed their time and efforts to thisstudy.