Clinical Results for Patients With Major Depressive Disorder in theTexas Medication Algorithm Project

Context  The Texas Medication Algorithm Project is an evaluation of an algorithm-based disease management program for the treatment of the self-declared persistently and seriously mentally ill in the public mental health sector.

Objective  To present clinical outcomes for patients with major depressive disorder (MDD) during 12-month algorithm-guided treatment (ALGO) compared with treatment as usual (TAU).

Design  Effectiveness, intent-to-treat, prospective trial comparing patient outcomes in clinics offering ALGO with matched clinics offering TAU.

Setting  Four ALGO clinics, 6 TAU clinics, and 4 clinics that offer TAU to patients with MDD but provide ALGO for schizophrenia or bipolar disorder.

Patients  Male and female outpatients with a clinical diagnosis of MDD (psychotic or nonpsychotic) were divided into ALGO and TAU groups. The ALGO group included patients who required an antidepressant medication change or were starting antidepressant therapy. The TAU group initially met the same criteria, but because medication changes were made less frequently in the TAU group, patients were also recruited if their Brief Psychiatric Rating Scale total score was higher than the median for that clinic's routine quarterly evaluation of each patient.

Main Outcome Measures  Primary outcomes included (1) symptoms measured by the 30-item Inventory of Depressive Symptomatology–Clinician-Rated scale (IDS-C₃₀) and (2) function measured by the Mental Health Summary score of the Medical Outcomes Study 12-item Short-Form Health Survey (SF-12) obtained every 3 months. A secondary outcome was the 30-item Inventory of Depressive Symptomatology–Self-Report scale (IDS-SR₃₀).

Results  All patients improved during the study (P<.001), but ALGO patients had significantly greater symptom reduction on both the IDS-C₃₀ and IDS-SR₃₀ compared with TAU. ALGO was also associated with significantly greater improvement in the SF-12 mental health score (P = .046) than TAU.

Conclusion  The ALGO intervention package during 1 year was superior to TAU for patients with MDD based on clinician-rated and self-reported symptoms and overall mental functioning.

Major depressive disorder (MDD) is a prevalent, serious, debilitatingillness that affects 7% to 12% of men and 20% to 25% of women in their lifetime.^1,2 The course of MDD is typically chronicor recurrent.³ From 10% to 30% of patientshave major depressive episodes that last loner than 2 years, and 20% to 30%have MDD superimposed on dysthymic disorder (double depression).^4-6 Majordepressive disorder accounts for up to 60% of psychiatric hospitalizations,and 8% to 15% of these patients commit suicide.^7,8 Furthermore,depression worsens the morbidity and mortality of several general medicalconditions (eg, cardiac heart disease,^9-11 myocardialinfarction,^12-14 chronicpain,¹⁵ diabetes,¹⁶ andasthma).^17,18 The direct monetarycost of treatment, combined with the indirect costs from lost productivity,are substantial^19-21 andhave been estimated to be between $44 and $53 billion per year.^22-25

Despite the high prevalence of MDD and the wide availability of effectivetreatments, undertreatment is common.^8,26,27 Theaim of treatment is symptomatic remission and functional recovery²⁸ with continuation treatment to prevent relapse.^29-32 Symptomaticimprovement (response) is distinguished from remission (ie, minimal or nosymptoms), because remission, in contrast to a response with residual symptoms,is associated with better functioning^33,34 anda better prognosis.^35-39

Most randomized controlled efficacy trials typically have engaged symptomaticvolunteers with minimal concurrent psychiatric or general medical illnessesand minimal levels of treatment resistance. Consequently, findings from thesestudies may not generalize to self-declared patients seen in clinical psychiatricpractice. Moreover, few studies define how to treat those with an unsatisfactoryclinical response to the initial treatment or compare the benefits of differentmedication options given sequentially.⁴⁰ Theseefficacy trials indicate that approximately 35% of participants achieve remissionin 6 to 8 weeks,²⁹ although higher remissionrates are found in longer treatment trials.^41,42 Inthe longer term, 10% to 30% of patients who do not respond or enter remissionquickly subsequently develop depressive relapses during the ensuing 4 to 6months despite continued pharmacotherapy.⁴³ Becauseno one treatment is a panacea, clinicians often use a sequence of treatmentsteps (either monotherapies or combinations) to increase the likelihood ofresponse or remission. Recent efforts have aimed to define guidelines or algorithmsfor the application of pharmacotherapeutic options for MDD.^44-50 Decisiontree–based algorithms hold the promise of increased consistency of treatmentacross practitioners, which in turn should lead to better clinical outcomesand more efficient use of health care resources. Algorithm-guided treatmentprovides a basis for improving the quality of treatment in both the publicand private sectors.

To our knowledge, the present study is the first controlled trial toevaluate algorithm-based treatment of depression in a public sector populationtreated by psychiatrists. One open trial⁵¹ ofthe impact of algorithm-driven treatment on symptomatic outcomes in a psychiatric(inpatient) population showed effectiveness for an algorithm in the inpatientsetting but lacked a control condition.

A series of studies,^51-65 includingthose conducted by Katon et al,^52,64 haveevaluated clinical outcomes following the use of Agency for Health Care Policyand Research–based, guideline-driven treatments (see the "Comment" section).Katon et al⁵² conducted a randomized controlledtrial of a guideline-driven intervention vs usual care in the treatment ofpatients with major (n = 91) or minor depression (n = 126) in a primary caresetting. For major but not for minor depression, the intervention was associatedwith greater adherence to adequate medication doses and more favorable ratingsof antidepressant medications benefit, as well as higher ratings of the qualityof care and better symptomatic outcomes. Most other trials conducted amongprimary care settings evaluated broadly defined guideline-driven principles(eg, Did the patient complete the acute-phase trial or not? Was the recommendedvisit frequency achieved?).

The Texas Medication Algorithm Project (TMAP) aimed to compare the clinicaland economic outcomes achieved with the use of prespecified medication algorithmscombined with clinical support and a prespecified patient and family educationalpackage for algorithm-guided treatment (ALGO) with treatment as usual (TAU).To increase the probability of appropriate algorithm implementation, an extensiveprovider support system with additional personnel funded by research moneyswas used.

This study is an effectiveness, intent-to-treat, prospective trial thatcompares patient outcomes in clinics offering ALGO with matched clinics offeringTAU. Clinics were prematched based on mental health and mental retardationauthority and urban status. Evaluable patients were postmatched based on symptoms(30-item Inventory of Depressive Symptomatology–Clinician-Rated scale[IDS-C₃₀] and 30-item Inventory of Depressive Symptomatology–Self-Reportscale [IDS-SR₃₀] scores) and length of illness (see Rush et al⁶⁶ for a detailed review of the rationale and design).This multisite study evaluated the clinical benefits of ALGO provided in 4clinics compared with 6 clinics that offered TAU (TAUnonALGO) and an additional4 clinics that offered TAU to patients with MDD but also provided ALGO foreither schizophrenia or bipolar disorder (TAUinALGO). Physicians from all14 clinics had access to the same medications. The TAUinALGO clinics wereintended to assess the effect of an "algorithm culture" associated with theimplementation of any of these algorithms on treatment practices. If no differencesbetween TAU groups were found, they could be combined for comparison withALGO. Randomizing patients among physicians and clinics would require healthcare providers to ignore their algorithm training and consultation interventionswhen treating control patients. To randomize by health care providers withinthe same clinic risked the "water cooler" effects (ie, algorithm physicianswould talk to and affect the practice of TAU physicians). Furthermore, physiciansin a clinic typically cross-covered for each other, further limiting feasibility.

The primary aim was to assess whether ALGO produced better clinicaloutcomes in terms of either an earlier onset and/or a greater overall effectduring a 1-year treatment period. We hypothesized that ALGO would produce(1) a faster and more robust improvement in symptoms, (2) better functioning,and (3) a lower side effect burden than TAU.

The study was conducted in accordance with international guidelinesfor good clinical practice and the Declaration of Helsinki and approved bythe institutional review boards at The University of Texas Southwestern MedicalCenter at Dallas and The University of Texas at Austin. On study entry, symptoms,function, quality of life, side effect severity and burden, and health careservice utilization and treatment costs were evaluated at baseline and quarterlyfor at least a 12-month period for all available participants.

ALGO included 2 consensus-driven, medication management algorithms (oneeach for psychotic and nonpsychotic forms of MDD)⁴⁸ andexpert consultation (offered on biweekly teleconference) and on-site clinicalsupport from clinical coordinators and a patient and family education programprovided by the clinical coordinators.⁶⁷ Thisintervention package was intended to optimize pharmacotherapy, thereby enhancingclinical outcomes. Each physician implemented ALGO in close collaborationwith a clinical coordinator. A 7-step medication algorithm for nonpsychoticMDD and a 5-step algorithm for psychotic MDD were provided (Figure 1 and Figure 2).Most steps in each algorithm included multiple treatment options, with earliersteps including those treatment options with the most evidence and the bestrisk-benefit ratios.

Multiple tools were used to enhance adherence to the algorithm. A detailedtreatment manual was used for initial didactic training and ongoing consultationswith clinicians (available at: http://www.mhmr.state.tx.us/centraloffice/medicaldirector/timamddman.pdf).⁶⁸ The manual identified critical decisionpoints (eg, weeks 4, 6, 8, 10, and 12) for each medication when revisionsin treatment strategies or tactics were to be undertaken based on degree ofsymptom change and side effect burden (Figure1 and Figure 2).

Symptom severity and side effect burden were routinely monitored ateach treatment visit to guide treatment implementation, with the aim of ensuringan adequate duration and dose of medication. Clinical assessments at eachvisit included a global assessment of symptoms and associated symptoms, IDS-C₃₀ and IDS-SR₃₀, and side effect burden by a 10-point globalscale. A standard clinical record form was completed at each clinic visitby those implementing the ALGO intervention. The symptom severity assessmentswere conducted by clinical coordinators before the physician visits.

Each ALGO patient also received a stepwise education package that providedinformation about the disease, prognosis, treatment options, and medicationside effects. This package encouraged patients to participate in treatmentdecisions and adhere to the treatment.⁶⁷

Male and female outpatients 18 years or older with a clinical diagnosisof MDD (psychotic or nonpsychotic) were eligible for the study. Patients enteredALGO if their treating physician judged that they required an antidepressantmedication change or were starting antidepressant therapy. Entrance into TAUinitially used the same criteria. However, because medication changes weremade less frequently in TAU, patients were also recruited if their quarterly,routinely administered 24-item Brief Psychiatric Rating Scale (BPRS-24)^69-71 total score was higherthan the median for that clinic's routine quarterly evaluation of each patient.Once approached, another BPRS-24 interview was conducted. Patients with BPRS-24total scores no more than 1 SD below enrolled ALGO patient average scoreswere asked to participate. This procedure ensured a minimal level of symptomseverity for participation in TAU in the absence of a medication change. Thus,in both ALGO and TAU clinics, a combination of procedure-cued and clinician-cuedmethods was used.

Exclusion criteria were minimal. Patients were excluded if they hadschizophrenic, bipolar, or schizoaffective disorder or a primary diagnosisof an obsessive-compulsive or eating disorder (anorexia nervosa or bulimianervosa). Also excluded were patients who required inpatient hospitalizationfor detoxification at the time of study entry, received mental retardationservices, or participated in an Assertive Community Treatment program.⁷²Table 1 givesthe ethnic composition and characteristics of the participating sites andclinics.

Study participants provided demographic and medical history at baselineand during outcome assessments every 3 months for at least 12 months. Enrollmentfor the study occurred throughout 13 months. Research coordinators not blindto treatment assignment but not involved in providing any treatment conductedthe research outcome assessments.

The clinical rating of depressive symptoms by the research coordinatorswith the IDS-C₃₀⁷³ was the primaryoutcome. Confirmatory symptom measures included the IDS-SR₃₀⁷³ and the BPRS-24.^69-71 Health-relatedquality of life was assessed using the Medical Outcomes Study 12-item Short-FormHealth Survey (SF-12).⁷⁴ Participants wereasked about the burden of side effects from medication during the past monththat "bothered or interfered with daily functioning." Respondents were consideredto have no significant side effects if they reported "no side effects" or"only mild side effects, not really significant" and to have significant sideeffects whenever side effects " . . . bothered me, but could tolerate them"or " . . . really bothered me, I either need to change my medication or takesomething for the side effects" or " . . . was so severe I had to be hospitalized."

Demographic information was obtained from a patient questionnaire administeredduring the face-to-face baseline interview. Alcohol and other drug use wasassessed quarterly using the Drug Abuse Screening Test⁷⁵ (scores>5 indicate drug abuse) and the Michigan Alcoholism Screening Test⁷⁶ (scores ≥5 indicate alcohol abuse). The PatientPerception of Benefits (T.M.K., unpublished data, 2000) is a 10-item, self-reportinstrument developed for this study, with scores ranging from 0 (belief) to40 (disbelief) that indicate whether the patient will see improved functioningif he/she gets needed care.

Hierarchical linear models⁷⁷ were adaptedto assess the impact of ALGO on clinical outcomes based on declining effectsanalyses developed for this study by Kashner et al.^78,79 Decliningeffects models are growth curves with dependent outcome variables representedas change scores. Independent variables include dichotomous treatment, timesince follow-up began, time × treatment interaction terms, and a constantterm. This approach takes into account repeated observations nested withinpatients; missing observations; varying intervals between follow-up observations;effect sizes that vary with time; heteroscedastic, autocorrelated, and othercomplex level 1 covariance structures; and continuous, bivariate, or ordinalvalued outcome variables. Parameter estimates were computed using HLM version5 software.⁸⁰

Estimates were computed separately to assess the impact of treatmenton ALGO and TAU patients. These estimates included an initial change in outcomesbetween baseline and the first 3 months and a growth rate in outcomes duringthe subsequent 9-month follow-up period. Growth rates were measured in termsof change in outcomes per quarter. To assess differences in the impact ofALGO vs TAU on outcomes, we measured differences in both initial changes andgrowth rates between ALGO and TAU patients. All estimates were adjusted toreflect baseline differences in starting values (change scores) and baselinecharacteristics (covariates) with respect to baseline need (IDS-C₃₀,length of illness in years), enabling (family size, disposable income), predisposing(years of education, Patient Perception of Benefits total score), and otherfactors (African American and Hispanic status). Program effects were computedby taking differences between ALGO and TAU with respect to initial changes(initial effect) and growth rates (growth rate effect). The growth rate effectwas used to determine if any initial ALGO advantage (initial effect) realizedduring the first quarter increased, remained constant, or declined duringthe 9-month follow-up. Declining effects were expected if, following an initialALGO advantage, TAU patients began to catch up to their ALGO counterparts(see Rush et al⁶⁶). To adjust for regressionto the mean due to baseline differences in reported side effect burden, patientswere divided for analyses between those reporting and not reporting significantside effects at baseline.

A total of 634 patients met study entry criteria and signed informedconsent. Of these, 21 did not complete the baseline assessment, 62 failedto report for any postbaseline visit, and 4 had such a visit but did not completeat least 1 postbaseline primary outcome (IDS-C₃₀). The remaining547 evaluable patients completed the primary symptom measure for at least1 postbaseline visit, including ALGO (n = 181), TAUinALGO (n = 212), and TAUnonALGO(n = 154).

Preliminary analysis revealed that patients attending TAUinALGO clinicswere achieving numerically but not statistically larger initial reductionsin symptoms following intake than their TAUnonALGO counterparts (IDS-C₃₀ adjusted D = −1.89, SE = 1.62, t₁₀₈₀ = 1.17, P<.24). The larger reductionswere expected if clinics that participated in ALGO programs targeting otherdisorders also improved care for MDD patients. On the other hand, TAUinALGOpatients were found to have only slightly lower severity of baseline symptoms(mean IDS-C₃₀ score = 36.0 ± 13.8) than their TAUnonALGOcounterparts (mean IDS-C₃₀ score = 37.9 ± 13.2), althoughthe difference was not statistically significant (D = −1.9, t₃₃₈ = 1.4, P = .18, with equalvariances not assumed).

Although change scores prevent factor loading baseline differences ontoestimates of effect sizes, issues of regression to the mean remain, leadingto upward biases of ALGO effect estimates. Because regression to the meanposed a more serious problem, we constructed a final analytic sample by matchingeach ALGO patient with the best match (without replacement) from either theTAUinALGO or TAUnonALGO groups with respect to baseline IDS-C₃₀ score(≤2), IDS-SR₃₀ score (≤10), and whenever possible, lengthof illness (≤20 years) independent of and blind to any outcomes. The approachis conservative, because including TAUinALGO patients would likely bias againstfinding an ALGO effect on reducing symptoms.

Comparing patients assigned to ALGO (n = 182) vs TAUnonALGO (n = 154),the unadjusted estimate for the ALGO initial effect is −5.58 ±1.45 for IDS-C₃₀ (t₁₂₄₅ = 3.85, P<.001) or −3.25 ± 1.49 for IDS-C₃₀ (t₁₂₃₆ = 2.19, P = .03) if adjusted to reflect baseline differences. These estimatescompare with −4.42 ± 1.36 for IDS-C₃₀ (t₁₃₂₀ = 3.24, P = .002) when calculationsare based on comparing ALGO (n = 175) with its matched TAU (n = 175) samples.The underlying hierarchical linear model is stable, because the adjusted estimatedinitial effect and robust standard error (IDS-C₃₀ D = −4.42± 1.36) are comparable to their least-squares equivalent (IDS-C₃₀ D = −4.80 ± 1.20).

Of the 181 evaluable ALGO patients, 175 (96.7%) from 4 clinics (38%,24%, 21%, and 17%, respectively) were matched successfully to 175 TAU patientsfrom 10 clinics (18%, 14%, 13%, 13%, 12%, 9%, 8%, 6%, 4%, and 3%, respectively),including 100 (47.2%) from TAUinALGO and 75 (48.7%) from TAUnonALGO. As expected,the final analytic ALGO (n = 175) and TAU (n = 175) samples had comparablebaseline IDS-C₃₀ scores (42.0 ± 13.1 vs 41.7 ± 12.7, Δ= 0.22, t₃₄₈ = 0.16, P<.87) and were comparable on most other demographic and healthvariables (Table 2).

Although resolving regression to the mean issues was important, theimpact of poststudy matching on external validity is unclear. Specifically,compared with unmatched evaluable patients (n = 197), the final analytic sample(n = 350) had more depressive symptoms (IDS-C₃₀ total score = 41.8vs 33.2; Δ = 8.7; t₅₄₅ = 7.4; P<.001), poorer mental functioning (SF-12 Mental HealthSummary [MHS] score = 27.8 vs 32.5; Δ = −4.7; t₅₂₇ = 5.3; P<.001), and shorterlength of illness (13.2 vs 19.8 years; Δ = −6.6; t₅₃₃ = 6.0; P<.001) and wereslightly younger (41.4 vs 43.7 years; Δ = −2.3; t₅₄₃ = 2.3; P<.02). No otherbaseline features distinguished these 2 groups.

Baseline covariates were not statistically significant predictors ofchange scores. Thus, potential biases introduced by these factors are likelyto be small. These analyses were limited to primary outcome measures overall analytic patients and not broken down by baseline symptom scores. Thisunderscores our finding of a difference between ALGO vs TAU on both initialand growth rate effects. Neither family size nor disposable income predictedoutcome. Medicaid and other public assistance variables, when included intothe existing model, did not improve the exploratory power of the includedcovariates.

The percentage of patients available for analyses at 3, 6, 9, and 12months were 100%, 99.5%, 83.2%, and 75.9%, respectively. As such, retentionfor the analyzable sample was considered excellent. The efficacy analyseswere conducted on the analytic sample of 350 patients (n = 175 each from ALGOand TAU). For the primary outcome measure, both TAU and ALGO groups had significantdecreases in IDS-C₃₀ scores during the first 3 months, with continuingreductions during the subsequent 9 months. The initial decline was significantlygreater for ALGO than TAU during the first 3 months. This advantage for ALGOover TAU persisted throughout the ensuing 3 quarters (ie, there was no catch-upby TAU) (Figure 3). Table 3 gives changes in IDS-C₃₀ scores in subgroupsdetermined by baseline level of symptom severity.

To explore whether ALGO effects varied depending on baseline symptomseverity, we further subdivided the sample into (1) very severe, (2) severe,and (3) mild/moderate baseline symptoms by IDS-C₃₀ score defineda priori. These analyses revealed that the effects obtained with ALGO werelargely accounted for by patients with severe and very severe baseline IDS-C₃₀ symptoms. The study was powered only to address the comparison betweenALGO and TAU for the overall groups. Therefore, these analyses comparing subgroups(based on different severity) are hypothesis generating rather than definitive.

The IDS-SR₃₀ revealed that ALGO was associated with significantlygreater symptom reduction during the first 3 months than was TAU. Both groupscontinued to improve during the subsequent 9 months, although TAU patientsshowed no evidence of catching up to their ALGO counterparts (Table 4 and Figure 4). Table 4 gives the overall changes in IDS-SR₃₀scores and changes based on subgroups defined by baseline symptomseverity using the IDS-SR₃₀. The differences between ALGO and TAUwere statistically significant for the very severely ill (IDS-SR₃₀ score,≥58) and for the severely ill (IDS-SR₃₀ score, 30-57) groups.

The BPRS₂₄ total scores revealed that both groups had significantsymptomatic improvements after the first quarter, with continued improvementduring the subsequent 9 months (data not shown). Mental functioning, as measuredby the SF-12 MHS score, improved initially and over time for both the ALGOand TAU groups, although the ALGO group experienced a significantly greaterinitial improvement, with no discernible catch up for their TAU counterparts(Table 5 and Figure 5). This effect was most profound for those with the lowestbaseline SF-12 MHS score. No significant between-group differences were observedfor the SF-12 MHS score for either group (data not shown).

ALGO patients with significant side effects at baseline (n = 77) tendedto report more side effects at the end of 3 months than their TAU counterparts(n = 76) (odds ratio [OR], 1.85; 95% confidence interval [CI], 0.99-3.47; t₅₆₇ = 1.94; P<.053).Differences in growth rates in side effects during follow-up were not significant(OR, 0.82 per quarter; 95% CI, 0.71-1.07; t₅₆₇ = 1.35; P<.18). Furthermore, ALGO patients(n = 87) who reported no significant side effects at baseline did not differfrom their TAU counterparts (n = 95) either after the first quarter (OR, 0.98;95% CI, 0.55-1.73; t₆₂₉ = 0.08; P<.94) or in growth rates during follow-up (OR, 1.02per quarter; 95% CI, 0.82-1.26; t₆₂₉ =0.16; P<.87).

This is the first study, to our knowledge, to assess the short- andlonger-term effects of treatment algorithms in the care of psychiatric patientswith MDD in the public mental health sector. At baseline, this patient populationwas characterized by substantial symptom severity, poor daily functioning,significant concurrent general medical conditions, and frequent alcohol andother substance abuse.

The ALGO intervention was associated with statistically and clinicallysignificantly better clinical outcomes than TAU in the primary (and most secondary)efficacy assessments, including IDS-C₃₀, IDS-SR₃₀, andSF-12 MHS scores. The magnitude of the difference between ALGO and TAU wasrobust (mean IDS-C₃₀ difference = 4.5 points; mean IDS-SR₃₀ difference = 7.5 points). The significant advantage for ALGO wasseen in the first quarter, with no evidence that TAU patients caught up withtheir ALGO counterparts during the ensuing 9-month period. Exploratory analysessuggested that ALGO was superior to TAU in those with greater symptom severityor worse function (SF-12) at baseline. The magnitude of the difference betweenTAU and ALGO is clinically substantial. By the clinician rating, twice asmuch (and by the self-report, 3 times as much) symptom reduction occurredin ALGO than in TAU. A 4.4-point difference in IDS-C₃₀ is roughlyequivalent to a 3-point difference on the Hamilton Rating Scale for Depression,which is the difference typically found in drug-placebo comparisons, yet herewe are comparing 2 active treatments (TAU and ALGO).

Results are generalizable to public sector patients with psychotic ornonpsychotic MDD. This population is characterized by substantial socioeconomicdisadvantages, long-standing depressive illness, and likely varying degreesof treatment resistance. Whether similar results would be found with employed,better-educated depressed patients seen in private practice is unclear.

Despite robust benefits attributable to ALGO, even among the responders,substantial symptoms remained. The fact that significant symptoms and functionalimpairment persisted points to the severity, comorbidity, chronicity, or possibletreatment resistance in this population. Results also raise the question ofwhether the outcomes would be different (more robust) if ALGO was used inless severely and chronically ill populations or in different treatment environments.On the other hand, those patients with particularly high health service utilizationmight accrue even greater benefits than patients with less complicated illnesses.

Furthermore, the study intervention was directed only toward optimizingpharmacotherapy and patient adherence. These results suggest the need to studythe effects of a broader-based intervention that would integrate evidence-basedpsychotherapy with evidence-based pharmacotherapy, as well as changes in thehealth service provision systems, to enhance physician adherence to evidence-basedtreatments.

Also, ALGO physicians likely demonstrated varying levels of algorithmadherence. It is possible that the clinical impact would be greater if physicianalgorithm adherence was monitored more closely and facilitated with real-timefeedback provided to clinicians to enhance decision making and algorithm adherence.In future analyses, we will examine whether physician adherence to the algorithmsis related to patient outcomes. Physicians' average patient load was not alteredduring the study and thus could have negatively affected their adherence tothe algorithms.

Limitations to the present study include the fact that although thestudy clinics were matched, the clinics, patients, and physicians were notrandomly assigned to the study groups (ie, ALGO or TAU), which may have introduceda bias. The outcome assessors were not blinded to treatment assignment andcould have biased the results in favor of the ALGO group. However, self-reports(IDS-SR₃₀, SF-12) corroborated clinician ratings of the benefitsof ALGO. In addition, varying degrees of algorithm adherence were accepted.

Despite its limitations, the results of this study have significantimplications for the provision of mental health care. Our findings, togetherwith reports of enhanced outcomes reported in primary care^66,81-86 settingswith the use of enriched disease management programs, suggest ways to enhanceclinical practice and care systems that might improve clinical outcomes andpositively affect health care utilization. Evidence to date indicates thatcare systems and practice procedures that attempt to apply practice guidelines,improve the consistency of care provided, and improve patient adherence appearto provide improved patient outcomes (both depressive symptoms and function).

Future studies need to evaluate how we can ensure more consistent implementationof disease management programs. To accomplish behavioral change, these issuesneed to be examined at both the practitioner and organizational levels. Atthe practitioner level, we need to explore mechanisms to increase algorithmadherence, including academic detailing, continuous quality improvement, andcomputerized decision support systems.^87-89 Atthe organizational level, we need to explore modes that more efficiently implementchange and more effectively allow practitioners to provide care. As notedin the Institute of Medicine's report, "Crossing the Quality Chasm: A NewHealth System for the 21st Century,"⁹⁰ improvingthe quality of health care in the United States requires not only changinghealth care professionals and organizations but also better methods of disseminatinginformation, application of technology, communication systems, and the creationof payment systems that reward positive performance. This is obviously anevolutionary process. We hope that this study provides a step toward additionalresearch to improve practice procedures and to provide care to enhance theoutcomes for individuals with depressive disorders.

Correspondence: Madhukar H. Trivedi, MD, Department of Psychiatry,University of Texas Southwestern Medical Center, 6363 Forest Park Rd, Suite1300, Dallas, TX 75235 (madhukar.trivedi@utsouthwestern.edu).

Submitted for publication June 11, 2002; final revision received December31, 2003; accepted January 28, 2004.

This research was supported by National Institute of Mental Health (NIMH)grant MH-53799; NIMH R01 No. R01MH064062-01A2 (Dr Trivedi); the Robert WoodJohnson Foundation (Princeton, NJ); the Meadows Foundation (Dallas); the Lightner-SamsFoundation (Dallas); the Nanny Hogan Boyd Charitable Trust (Dallas); the TexasDepartment of Mental Health and Mental Retardation (Austin); the Center forMental Health Services (Washington, DC); the Department of Veterans Affairs(Washington, DC); Health Services Research and Development Research CareerScientist Award (RCS92-403); the Betty Jo Hay Distinguished Chair in MentalHealth and the Rosewood Corporation Chair in Biomedical Science (Dr Rush);the United States Pharmacopoeia Convention, Inc (Rockville, Md); Mental HealthConnections, a partnership between Dallas County Mental Health and MentalRetardation and the Department of Psychiatry of the University of Texas SouthwesternMedical Center, which receives funding from the Texas State Legislature andthe Dallas County Hospital District and the University of Texas at AustinCollege of Pharmacy; and the Southwestern Drug Corporation Centennial Fellowshipin Pharmacy (Dr Crismon). The following pharmaceutical companies providedunrestricted educational grants: Abbott Laboratories (Abbott Park, Ill), AstraZenecaPLC (Wilmington, Del), Bristol-Myers Squibb (New York, NY), Eli Lilly &Company (Indianapolis, Ind), Forest Laboratories Inc (New York), GlaxoSmithKline(Research Triangle Park, NC), Janssen Pharmaceutica (Titusville, NJ), NovartisInternational AG (Basel, Switzerland), Organon International Inc (Roseland,NJ), Pfizer Inc (New York), and WyethAyerst Laboratories Inc (Madison, NJ).

Results of this study were presented in part at the annual meeting ofthe American College of Neuropsychopharmacology; December 12, 2000; SanJuan, Puerto Rico; the annual meeting of the American PsychiatricAssociation; May 7, 2001; New Orleans, La; the annual meeting of theAmerican College of Clinical Pharmacy; October 21-24, 2001; Tampa, Fla;and the annual meeting of the American Psychiatric Association; May 21,2002; Philadelphia, Pa.

We deeply appreciate the consultations provided by Barbara Burns, PhD,Robert Drake, MD, Susan Essock, PhD, William Hargreaves, PhD, Teh-wei Hu,PhD, Anthony Lehman, MD, and Greg Teague, PhD, without whose expertise thisstudy could not have been accomplished. We also thank the National and TexasAlliance for the Mentally Ill and the National and Texas Depressive and ManicDepressive Association for providing patient and family educational materials.We appreciate greatly the assistance of Gus Sicard, PhD, for his translationof the educational, clinical, and research evaluation materials into Spanish.We thank Karla Starkweather, BJ (TMAP communications director, Texas Departmentof Mental Health and Mental Retardation, Austin). Most important, we wishto express our gratitude to the personnel at each community mental healthcenter for allowing us to conduct this research with their staff and patients,including Andrews Center, Tyler; Center for Health Care Services, San Antonio;Life Management Center, El Paso; Lubbock Regional Mental Health and MentalRetardation Center; Mental Health and Mental Retardation Authority of HarrisCounty, Houston; Tri-County Mental Health and Mental Retardation Services,Conroe; and Tropical Texas Mental Health and Mental Retardation Services,Edinburgh, Harlingen, and Brownsville. We appreciate the secretarial supportof Fast Word Inc (Dallas) and David Savage and thank Andrew Sedillo, MD, forestablishing the TMAP Web site (http://www.mhmr.state.tx.us/centraloffice/medicaldirector/tmap.html). We also thank the assistant module directors, Sherwood Brown, MD,PhD, John Chiles, MD, and Teresa Pigott, MD, and the algorithm managementteam coordinators, Judith Chiles, BSN, Ellen Dennehy, PhD, Ellen Habamacher,BS, and Tracie Key, BSN, RN.