Controlled Clinical Trials 25 (2004) 119 142

www.elsevier.com/locate/conclintrial

Design paper

Sequenced treatment alternatives to relieve depression (STAR*D): rationale and design

A. John Rush a,*, Maurizio Fava b, Stephen R. Wisniewski c, Philip W. Lavori d, Madhukar H. Trivedi a, Harold A. Sackeim e, Michael E. Thase f, Andrew A. Nierenberg b, Frederic M. Quitkin f, T. Michael Kashner g, David J. Kupfer f, Jerrold F. Rosenbaum b, Jonathan Alpert b, Jonathan W. Stewart e, Patrick J. McGrath e, Melanie M. Biggs a, Kathy Shores-Wilson a, Barry D. Lebowitz h, Louise Ritz h, George Niederehe h, for the STAR*D Investigators Group 1
Department of Psychiatry, University of Texas Southwestern Medical Center, Dallas, Texas, USA Clinical Psychopharmacology Unit, Massachusetts General Hospital, Boston, Massachusetts, USA c Department of Epidemiology, University of Pittsburgh, Pittsburgh, Pennsylvania, USA d Department of Veterans Affairs Cooperative Studies Program and Stanford University, Palo Alto, California, USA e New York State Psychiatric Institute and the Department of Psychiatry, College of Physicians and Surgeons of Columbia University, New York, New York, USA f Department of Psychiatry, University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania, USA g Health Services Research and Development Service, Department of Veterans Affairs, and Department of Psychiatry, University of Texas Southwestern Medical Center, Dallas, Texas, USA h National Institute of Mental Health, Bethesda, Maryland, USA
b a

Received 24 September 2001; accepted 8 August 2003

Abstract STAR*D is a multisite, prospective, randomized, multistep clinical trial of outpatients with nonpsychotic major depressive disorder. The study compares various treatment options for those who do not attain a satisfactory response with citalopram, a selective serotonin reuptake inhibitor antidepressant. The study enrolls 4000 adults (ages 18 75) from both primary and specialty care practices who have not had either a prior inadequate response or clear-cut intolerance to a robust trial of protocol treatments during the current major depressive episode. After receiving citalopram (level 1), participants without sufficient symptomatic benefit are

* Corresponding author. Department of Psychiatry, University of Texas Southwestern Medical Center, 5323 Harry Hines Boulevard, Dallas, TX 75390-9086. Tel.: +1-214-648-4600; fax: +1-214-648-4612. E-mail address: john.rush@utsouthwestern.edu (A.J. Rush). 1 See appendix for a complete listing and location of regional centers, regional center directors, and clinical sites. 0197-2456/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/S0197-2456(03)00112-0

120

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

eligible for randomization to level 2 treatments, which entail four switch options (sertraline, bupropion, venlafaxine, cognitive therapy) and three citalopram augment options (bupropion, buspirone, cognitive therapy). Those who receive cognitive therapy (switch or augment options) at level 2 without sufficient improvement are eligible for randomization to one of two level 2A switch options (venlafaxine or bupropion). Level 2 and 2A participants without sufficient improvement are eligible for random assignment to two switch options (mirtazapine or nortriptyline) and to two augment options (lithium or thyroid hormone) added to the primary antidepressant (citalopram, bupropion, sertraline, or venlafaxine) (level 3). Those without sufficient improvement at level 3 are eligible for level 4 random assignment to one of two switch options (tranylcypromine or the combination of mirtazapine and venlafaxine). The primary outcome is the clinician-rated, 17-item Hamilton Rating Scale for Depression, administered at entry and exit from each treatment level through telephone interviews by assessors masked to treatment assignments. Secondary outcomes include self-reported depressive symptoms, physical and mental function, side-effect burden, client satisfaction, and health care utilization and cost. Participants with an adequate symptomatic response may enter the 12-month naturalistic follow-up phase with brief monthly and more complete quarterly assessments. D 2004 Elsevier Inc. All rights reserved.
Keywords: Major depressive disorder; Antidepressants; Cognitive therapy; Randomized clinical trial; Multistep multicenter clinical trial; Utilization and costs

1. Introduction and background Major depressive disorder (MDD) is a common, typically recurrent, often chronic, and very disabling disorder, costing the United States over $44 billion/year in direct and indirect costs [1]. MDD has a lifetime prevalence of 4.917.9% [2,3]. Women are twice as likely to suffer MDD as men. MDD occurs more frequently among young adults and among those with general medical conditions (GMCs) [4]. Depressed adults have nearly twice the annual health care costs of those without depression [5]. MDD is usually an episodic disorder with an average of one episode every 5 years [4,6]. Of persons with MDD, 2035% experience a chronic, unremitting course [4,6,7]. A similarly sized (and partly overlapping) group has early-onset dysthymia, a condition involving milder but chronic depressive symptoms, which precede the onset of major depressive episodes. Relapses and recurrent episodes are most likely among patients with antecedent dysthymia or incomplete interepisode recovery (i.e., those with residual interepisode symptoms). Longer major depressive episodes appear to be more difficult to treat [8,9]. Depression is the fourth most disabling medical condition worldwide based on disability-adjusted lifeyears (years of life lost due to premature death and years lived with a disability of specified severity and duration) [10]. Depression is predicted worldwide to be second only to ischemic heart disease with regard to disability by the year 2020 [10]. Patients with MDD function more poorly than other outpatients with a variety of GMCs in terms of physical activity and occupational and social role responsibilities [11]. As established by randomized controlled trials (RCTs), many medications and several depressiontargeted psychotherapies have efficacy [1216]. Since most of these efficacy studies exclude many patients with general medical and psychiatric comorbidities, generalizability is limited. Thus, the effectiveness of established treatments in more representative clinical populations is not well established. About 50% of outpatients with nonpsychotic MDD initially treated with either a time-limited, depression-targeted psychotherapy or a single antidepressant medication respond (i.e., have z50% reduction in baseline symptom severity) to treatment in efficacy trials [4,12,13,17,18].

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

121

In acute phase RCTs with medications lasting 68 weeks, about 2030% attain symptomatic remission, while 2030% respond but have residual symptoms (i.e., have a z50% reduction in total symptom severity without remission). Thus, only about 5070% of those who respond attain remission [17,19].2 Similar outcomes are found with time-limited, depression-targeted psychotherapies [20,21]. Symptomatic response with residual symptoms is associated with continuing functional disability [8,22 24] and a worse prognosis [2527]. The management of patients with treatment-resistant depression (i.e., lacking remission to one or more adequate trials of treatment) is a major public health challenge [28]. RCTs that compare treatments for depressions that have not remitted after an initial antidepressant treatment are rare. Thus, clinical practice guidelines that suggest different second or third treatment steps rest largely on clinical consensus or on open uncontrolled trials [14,29,30]. In fact, most practice guidelines do not recommend specific next-step treatments or specific treatment sequences [4,13,15]. Scientific evidence to define the most specific, effective next-step treatment options for treatment-resistant MDD should improve clinical outcomes and may reduce the cost of care [3133].

2. Aims and treatment paradigm Sequenced Treatment Alternatives to Relieve Depression (STAR*D) aims to define prospectively which of several treatments are most effective for participants with MDD who experience an unsatisfactory clinical outcome following an initial and, if necessary, subsequent treatment(s). Eligible participants who consent to STAR*D are enrolled into the first level of STAR*D (level 1), which uses the selective serotonin reuptake inhibitor (SSRI) citalopram (CIT). Participants with an adequate clinical response with level 1 (as defined by the clinician, who is informed by a brief clinical ratingsee below) enter a 12-month, naturalistic, follow-up phase. Participants without a satisfactory clinical outcome to CIT are eligible to enter a series of randomized clinical trials. The first trial (level 2) compares the effectiveness of seven different treatments including four switch options (venlafaxine [VEN], sertraline [SER], bupropion [BUP], and cognitive therapy [CT]) and three augment options (CT, BUP, or buspirone [BUS] added to CIT) (Fig. 1). Thus, STAR*D evaluates the comparative effectiveness of different treatment strategies, such as switch of treatment (switch) versus augmentation of the initial treatment (augment) at level 2. It also compares the effectiveness of individual treatment options within and across these two strategies (switch versus augment). Those with a satisfactory therapeutic response in level 2 enter the 12-month naturalistic followup phase. Participants without a satisfactory therapeutic response to both CIT (level 1) and to CT (level 2), whether used as a switch or augment option, enter level 2A, an RCT comparing the effectiveness of two

According to Frank et al. [19], response is a clinically significant reduction in depressive symptoms (e.g., z50% reduction in baseline symptom severity). Remission is the complete absence of depressive symptoms, defined for this protocol as total 17item Hamilton Rating Scale for Depression (HRSD17) score of <8. A response without remission refers to a patient with a z50% decrease in baseline symptom severity at exit but whose HRSD17 total score is still z8. A partial response refers to persons whose depressions do not meet the response criterion but who do have clinically meaningful reductions in baseline symptom severity (e.g., z25% but <50% reduction in baseline symptom severity).

122

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

Fig. 1. Treatment strategies/substrategies and options.

pharmacological switch options (BUP or VEN),3 which ensures that all participants who enter level 3 have not responded to two different antidepressant medication treatments. Those with a satisfactory response to level 2A enter the 12-month follow-up, while those without such a response enter level 3. Level 3 is an RCT that compares the relative effectiveness of two switch options (mirtazapine [MRT] or nortriptyline [NTP]) and two augment options (lithium [Li] or thyroid hormone [THY]). Those with a satisfactory therapeutic response to level 3 enter the 12-month naturalistic follow-up phase. The remaining participants may enter level 4, an RCT comparing the effectiveness of two treatment switch options (tranylcypromine [TCP] or the combination of MRT+VEN). Those with an adequate response to level 4 enter the 12-month naturalistic follow-up, while the remaining participants exit the study and are treated as clinically indicated.

3. Major endpoints The primary outcome is depressive symptom severity, measured by the 17-item Hamilton Rating Scale for Depression (HRSD17) obtained at the end of each treatment level (levels 1 4) by
Level 2A did not include SER because we viewed this as an unlikely clinical choice, given the out of class switch entailed with CIT (i.e., a return to the original class would be clinically redundant). Second, with two as opposed to three cells at level 2A, the potential for adequate sample sizes was increased.
3

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

123

independent, telephone-based interviewers with no knowledge of treatment assignment [34,35]. Treatment remission, for research purposes, is defined as an HRSD17 total score <8 upon exit from a treatment level. The 30-item Inventory of Depressive SymptomatologyClinician-Rated (IDS-C30), obtained by the same telephone-based interviewers, is a secondary symptom outcome [36,37]. Other secondary outcomes, all obtained by an interactive voice response (IVR) telephone system, include brief self-reports of depressive symptoms, function, side-effect burden, quality of life, and participant satisfaction [38].

4. Study regimens 4.1. Selection of protocol treatments The specific treatments at each level were selected to balance safety while evaluating common practices and pharmacological reasoning. A detailed rationale for the treatments chosen at each level are reviewed elsewhere [39]. To illustrate, CIT was chosen to represent the SSRI class at level 1 since, as several SSRIs will shortly become generic, health care systems are likely to encourage use of an SSRI as a first agent. Thus, defining the next best step following an ineffective SSRI has high public health significance, though the available research on which next step is preferred is largely limited to open noncomparative trials of nonresponders following treatment with an SSRI [16]. Furthermore, CIT has minimal drug-drug interactions and a half-life (3235 hours) such that switching to another medication will not entail studying a combination of two medications in the early weeks of the next level, as would be the case, for example, with switching from fluoxetine (FLU). In addition, this half-life means that CIT does not need to be tapered to avoid discontinuation-emergent symptoms. Thus, switching to another treatment is straightforward, requiring neither a tapering nor a washout period. All participants without remission are eligible for both switch and augment strategies at level 2, while those with CIT intolerance are eligible only for the switch strategy at level 2. The level 2 switch options will compare an in class switch (i.e., SER) to an out of class switch (i.e., BUP), a dual-action agent (i.e., VEN), and psychotherapy (i.e., CT). 4.2. Treatment visits and movement to next level or follow-up Although the HRSD17, collected by telephone interview, is the primary outcome used to determine response for research purposes, the decision to proceed to the next level or to follow-up depends on clinical judgment regarding the benefit of the current treatment(s) as informed by the 16-item Quick Inventory of Depressive SymptomatologyClinician Rating (QIDS-C16), which is obtained by the clinical staff at each treatment visit without knowledge of the primary research outcome (i.e., HRSD17 score) [4042]. These clinician judgments (not the HRSD17 score) that may lead to a participant proceeding to the next level include intolerance, nonresponse, and response without remission. Intolerance is declared if a participant discontinues treatment within the first 4 weeks for any reason or due to intolerable side effects after that time, independent of the symptomatic status. Nonresponse is a <50% reduction in symptom severity measured by the QIDS-C16 at exit, as long as exit is not due to intolerance. Remission (the absence of depressive symptoms) is defined as V5 on the QIDS-C16, while

124

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

response without remission is a z50% reduction in baseline QIDS-C16 score but a QIDS-C16 score > 5 at exit from a treatment level (as long as exit is not due to intolerance). The decision to move to the next level is made by the clinician based on symptom severity (assessed by clinical judgment and informed by the QIDS-C16 total score) and by side-effect status. Participants with clear intolerance or minimal reduction in baseline symptom severity (e.g., <15% by week 6 or <25% by week 9), who have received adequate doses, are encouraged to move to the next treatment level. Those without response at completion of a treatment level may proceed to the next treatment level, while those with remission may enter follow-up. Those with response but without remission may enter follow-up, but they are strongly encouraged to proceed to the next treatment level after an adequate dose and duration have been achieved. Protocol medication clinic visits are required at weeks 0, 2, 4, 6, 9, and 12 at all treatment levels. The visit schedule is flexible (e.g., the week 2 visit may be held within F6 days of week 2). Extra visits may be held if clinically needed. If a participant exhibits a response or remission only at week 12, two additional visits may be used to determine if that status is sustained. Psychotherapy visits are to be scheduled twice a week for weeks 14, then once a week for 8 weeks. Participants with remission before week 12 (16 sessions) and with at least 3 consecutive weeks of remission may go to follow-up without completing all 16 visits. If response without remission is present at visit 16, the therapist may continue with CT for an additional four weekly sessions. If such patients then achieve remission, they may receive an additional four sessions (every other week) and six additional sessions (monthly) for up to a total of 30 sessions. The rationale for providing more extended CT for those without a stable remission by the end of acute phase CT rests on recent findings that such patients benefit from continuation phase CT [43]. If nonresponse is present at visit 16 (end of week 12) for CT or if response without remission remains after continuing medication up to week 14 or CT up to visit 20, participants may proceed to level 2A. 4.3. Protocol treatment delivery A natural tension exists between the desire for maximal participant retention with the most diligent implementation of protocol treatments and the desire for maximal generalizability of findings. In addressing this tension, STAR*D aims to ensure that proper treatment implementation and maximal patient retention both occur since the main objective is to compare the effectiveness of different treatments that are reasonably well implemented. Participant/family education is provided as part of general clinical management for participants [4,13,15]. Participants are compensated for each research assessment ($25/1-hour assessment). The clinical research coordinator (CRC) actively pursues participants who miss appointments to prevent premature protocol discontinuation. Newsletters and STAR*D website (http://www.star-d.org) updates are provided to participants to maintain interest and motivation. Pharmacotherapists and psychotherapists were trained in the protocol and treatment procedures at the Dallas national coordinating center (NCC). A clinical procedures manual details all relevant clinical procedures. The CRCs check protocol implementation and complete a clinical record form (CRF) at each clinic visit. Psychotherapy training was conducted by expert cognitive therapists from the University of Pittsburgh School of Medicine. Mastery of CT procedures and techniques was confirmed according to the methods described by Shaw prior to treatment of protocol cases [44]. After initial training,

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

125

audiotapes from 20% of study sessions are reviewed; if therapists experience significant adherence problems, additional supervision is provided. Therapists unable to adhere to CT methods are not permitted to continue to see study patients. 4.4. Concurrent treatments Across all levels, adjunctive treatments for associated symptoms (e.g., insomnia) or medication side effects (e.g., sexual dysfunction) are allowed, thereby mirroring common practice, but antipsychotic (e.g., risperidone or olanzapine) and mood-stabilizing agents (e.g., valproic acid or lamotrigine) other than Li as one of the randomized augment options at level 3 are prohibited. Participants may receive treatment for concurrent GMCs, as long as these medications do not contraindicate the use of level 1 and all level 2 treatments. Participants may not be receiving any psychotherapy that is aimed at the treatment of depression other than CT as prescribed in the protocol. Any antidepressant medications taken at study entry must be discontinued (after consent) before beginning CIT. The discontinuation is guided by clinical judgment without the formal requirement of a washout period.

Table 1 Inclusion/exclusion criteria for study entry Inclusion criteria Age 18 75 Written informed consent obtained Score z14 on the HRSD17 Meets DSM-IV* criteria for single or recurrent nonpsychotic MDD a. At least one of the symptoms is either depressed mood or loss of interest and pleasure and b. Five or more symptoms have been present simultaneously during at least the past 2 weeks and represent a change from previous functioning Exclusion criteria 1. History of bipolar disorder (I, II, not otherwise specified) (lifetime) 2. History of schizophrenia, schizoaffective disorder, or psychosis not otherwise specified (lifetime) 3. Current anorexia or bulimia 4. Current primary obsessive compulsive disorder 5. History of clear-cut intolerability to, or lack of effect with, an adequate trial (e.g., 60 mg of fluoxetine for 6 weeks) of at least one protocol medication in the current episode of MDD 6. Lack of response to an adequate trial of an SSRI (CIT, FLU, paroxetine, and SER) in the current episode of MDD (e.g., adequate trial with CIT is at least 40 mg/d for 6 weeks or 60 mg/d for the last 2 weeks of a 6-week trial) 7. Currently taking CIT and has been taking it for more than 7 days 8. Did not respond to 16 or more sessions of CT in the current episode of MDD 9. Did not respond to seven or more sessions of electroconvulsive therapy in the current episode of MDD 10. Has general medical condition that contraindicates any level 1 or 2 treatment option 11. Is taking any concomitant medication that contraindicates any level 1 or 2 treatment option 12. Requires immediate hospitalization for substance/alcohol detoxification or treatment 13. Requires immediate hospitalization or day treatment for psychiatric disorder(s) 14. Requires antipsychotic medication or mood stabilizers 15. If female, is pregnant * DSM-IV: Diagnostic and Statistical Manual of Mental Disorders, 4th edition [45]. 1. 2. 3. 4.

126

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

4.5. Follow-up STAR*D also aims to describe the incidence, nature, and course of symptoms and function for those entering the 12-month naturalistic follow-up. This follow-up phase is available to all participants with at least a response to any treatment. Clinic visits are held every 23 months. All participants are encouraged to continue the same type and dose of medication found successful in acute treatment. 4.6. Study participants The broad inclusion and minimal exclusion criteria (Table 1) were chosen to include outpatients with nonpsychotic MDD who would typically receive medication or psychotherapy in most clinical settings. Those over age 75 were excluded because concomitant GMCs or medications may complicate treatment with a number of protocol medications, while adolescents/children were excluded because the efficacy and safety of most study medications have not been established and the delivery of CT would have required separate training/specific experience in this population. The study protocol was developed according to the principles of the Declaration of Helsinki [46]. All risks, benefits, and adverse events associated with each treatment within the randomized treatments are explained to study participants, who provide written informed consent prior to study entry and again prior to entry into each new level of treatment or into follow-up.

5. Study organization 5.1. Organizational structure The STAR*D infrastructure includes the NCC (Dallas), the data coordinating center (DCC) (Pittsburgh), and 14 regional centers (RCs), each of which oversees two to four primary and specialty care clinical sites (CSs). Each RC provides support, quality control, and coordination for the recruitment, retention, and safety of study participants and oversees the acquisition of clinical information from the CSs. STAR*D provides opportunities for ancillary studies relevant to improving the treatment of treatment-resistant depression that otherwise do not impede the conduct of the main protocol. 5.2. Site selection/training/recruitment Applications from potential RCs were reviewed, site visits were conducted at those that ranked highly, and RCs were selected based on the likelihood that CSs in either the public or private sector overseen by the RC could meet recruitment goals, ensure adequate overall minority representation, and provide adequate and safe protocol adherence. CSs largely include practices not typically engaged in efficacy RCTs. CRCs at each RC are trained and certified in protocol implementation and data collection methods. CRCs work very closely with the participants and clinicians, administer some of the clinician-rated instruments, ensure that all self-rated instruments are properly filled out, and function as study coordinators, providing a liaison among CSs, the RC, and the DCC. The CRCs minimize research burdens for clinicians.

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

127

5.3. Enrollment/randomization Eligible and consenting outpatients are screened and enrolled in protocol-defined treatment at the CSs. The CRC then calls the IVR system and enters a patient identification number. Consenting participants undergo randomization at levels 2, 2A, 3, and 4. Treatment assignment using randomization occurs via a blocked randomization scheme, stratified by CSs, and by the treatment options that are acceptable to the participant at that particular level (levels 2 and 3). To account for differences in participants acceptance of the various treatment strategies (e.g., augment or switch) or substrategies (e.g., augment exclusively with one of the two medications), we have created specific acceptability strata for levels 2 and 3. Participants are stratified according to their acceptance of the seven treatment options at level 2 and of the four treatment options at level 3. Participants are randomized among the treatment options included in the strategies (or substrategies) that they accept. Level 2 has two strategies (augment or switch) and four substrategies (medication switch, medication augment, CT switch, and CT augment), while level 3 has two strategies (switch and augment). A patient must accept all treatment options within a substrategy (levels 2 and 3). In analyses comparing strategies or particular sets of substrategies, outcome data will be restricted to all participants assigned to those treatment options under conditions of potential randomization to any of them. Uniform random number generation is used to randomize participants among treatment options within each level, with an equal chance of receiving any option. Randomization was blocked with the block size equal to the number of treatment options within the preferred strata. The randomization process for the next treatment level is initiated upon exiting the current treatment level, following participant written consent to proceed. The CRC notifies the IVR system that the participant is exiting the current level and entering the next treatment level and presents the participants treatment preferences. The IVR system identifies the next treatment option in the appropriate stratum and immediately responds to the CRC.

6. Data collection 6.1. Screening and clinical data Table 2 summarizes the data collected during the screening process. Data collection forms can be ordered at www.star-d.org. At all clinic visits, information is obtained to guide clinicians implementing study treatments (Table 3). Symptomatic status is measured by the QIDS-C16 [4042]. The QIDS-C16 is used to gauge response because previous experiences suggest that global ratings may be insensitive to detecting residual symptoms [49,50]. The QIDS-C16 rates all nine criterion symptoms of MDD, is easily administered, and is distinct from the primary symptom measure (the HRSD17). The self-rated form of the QIDS (QIDS-SR16) is also administered at all visits [41]. The clinical record form records the Clinician Global ImpressionImprovement scale [51], the Patient-Rated Inventory of Side Effects, and global ratings according to the Frequency and Intensity of Side Effects Rating (FISER), and Global Rating of Side-Effect Burden (GRSEB), which are seven-point Likert scales developed for this study.

128 Table 2 Data collection at screeninga Domain Consent Characteristics Eligibility Concurrent psychiatric diagnosis(es) Symptoms

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

Measure

Time Method (minutes) Interview Interview Interview Self-report Interview

Occasion At At At At intake intake intake intake

Administrator Location CRC CRC CRC Participant Clinic Clinic Clinic Clinic Clinic

General medical conditions

a

Consent 20 Clinical/demographic features 3 Inclusion/exclusion 5 Psychiatric Diagnostic 20 30 Screening Questionnaire [47] 15 Hamilton Rating Scale for Depression (17-item) [34,36]/ Quick Inventory of Depressive SymptomatologyClinician Rating (16-item) [40 42] Quick Inventory of Depressive 4 SymptomatologySelf-Report (16-item) [40 42] Cumulative Illness Rating Scale [48] 5

At intake CRC

Self-report At intake Patient

Clinic

Interview

At intake CRC

Clinic

Screening and baseline research outcomes assessments are obtained on all participants (anticipated n = 4000).

6.2. Research outcomes Research outcomes (Table 4) are collected by IVR and by telephone interviews conducted by Research Outcomes Assessors (ROAs), independent of and masked to treatment. ROAs collect the HRSD17, the IDS-C30, and the five-item Income and Public Assistance Questionnaire measuring the participants monthly income by source (employment, assistance and welfare, unemployment
Table 3 Data collection at clinical visitsa Domain Symptoms Measure Clinical Global ImpressionImprovement (1-item) Quick Inventory of Depressive SymptomatologyClinician Rating (16-item) Quick Inventory of Depressive SymptomatologySelf-Report (16-item) Frequency and Intensity of Side Effect Rating/Global Rating of Side-Effect Burden (3-item) Patient-Rated Inventory of Side Effects (8-item) Clinical record form Medication adherence questionnaire Time (minutes) 0.5 3 4 1.5 Method Interview Interview Self-report Self-report Administrator Clinician CRC/clinician Patient Patient

Side effects

Medication Medication compliance

a

3 3 3

Self-report Self-report Self-report

Patient Clinician Participant

These measures are used to provide consistent information to the clinicians who use this information in the protocol and are recorded on the CRF. These measures are collected at clinic visits for participants in protocol treatment. They are not collected at clinic visits for participants seen in follow-up.

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142 Table 4 Research outcomesa Domain Symptoms Measure Hamilton Rating Scale for Depression (17-item)/Inventory of Depressive SymptomatologyClinician-rated (30-item) Quick Inventory of Depressive SymptomatologySelf-Report (16-item) Short-Form Health Survey (12-item) [52] Work and Social Adjustment Scale (5-item) [53] Work and Productive Activity Impairment Questionnaire (6-item) [54] Quality of Life Enjoyment and Satisfaction Questionnaire [55] Frequency and Intensity of Side Effect Rating/Global Rating of Side-Effect Burden (3-item) Patient Satisfaction Inventory (2-item) Utilization and Cost Patient Questionnaire (15-item) [56 61] Income and Public Assistance Questionnaire (5-item) Current Treatment Questionnaire (5-item) Time (minutes) 20 25 Method

129

Telephone (ROA)

6 4 2 2.5 6 1.5 1 5 3 2

Telephone (IVR) Telephone (IVR) Telephone (IVR) Telephone (IVR) Telephone (IVR) Telephone (IVR) Telephone (IVR) Telephone (IVR) Telephone (ROA) Telephone (ROA)

Function

Quality of life Side effects Patient satisfaction Utilization and cost Income Current treatment
a

Research outcomes are obtained at entry and exit from each treatment level and in follow-up at months 3, 6, 9, and 12.

benefits, and unearned income including retirement benefits, interest, royalties, and leases). The IVR collects the QIDS-SR16 and ratings of function, quality of life, side effects, and health care utilization. The IVR method was chosen because self-reporting of symptomatic and functional outcomes in nonpsychotic, depressed outpatients reliably corresponds to observer ratings obtained through standard clinical interviews [38,62,63]. IVR has the advantages of: (1) high participant acceptance and convenience, (2) high interrater reliability with well-standardized measures, (3) high correspondence with clinical interview data, (4) high sensitivity to change over time, and (5) equivalent capacity to distinguish symptomatic outcomes between treatment groups when compared to standard clinician ratings in depressive and other psychiatric disorders [63]. Research outcomes are collected at pretreatment, at exit from each treatment level, and at months 3, 6, 9, and 12 in follow-up. Interim research outcomes (QIDS-SR16, five-item Work and Social Adjustment Scale [WSAS], six-item Work and Productive Activity Impairment Questionnaire [WPAI], and FISER/ GRSEB) are also collected by IVR, at week 6 in each treatment level and at months 1, 2, 4, 5, 7, 8, 10, and 11 in follow-up. 6.3. Utilization/cost data STAR*D gathers information on participant use of health services, measured in both physical units and costs by type (depression-related, other psychiatric or mental health care, general medical care) and setting (outpatient clinic, emergency room, inpatient care). Indirect costs include changes in market productivity measured in terms of employment status, hours worked, and earned income (wages, salaries, tips, commissions, and earnings from self-employment) [6467]. Changes in both market work and household

130

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

management functioning due to depression symptoms (WSAS) and health status (WPAI) are also measured. Indirect costs also include the administrative costs associated with welfare transfer payments. 6.4. Data management Data are collected on paper forms, through an IVR system, and through direct electronic transfer from CS administrative databases. The paper data collection forms are faxed to the DCC. The fax transmissions are sent directly to a microcomputer where the forms are scanned using optical character recognition software, and data are immediately transferred into an ACCESS database on a secure server. The data are verified and pass through an editing process to check for logical inconsistencies within the data. Any discrepancies are e-mailed to the appropriate CRC for resolution. The resolutions are e-mailed back to the DCC for inclusion in the database. IVR data are transferred routinely from Health Technology Systems, Inc. (Madison, WI) to the DCC, where they are appended to the other study data. Reports are available to study personnel via the study website for monitoring data quality, recruitment, and treatment adherence. These reports are updated three times a week to assure the most recent data are being assessed. Data abstracted from the CS administrative databases are used solely for the economic analyses and as a check to determine patient protocol adherence. All data collection forms and data reports include only the subjects study identification number (ten alpha-numeric characters). The first two characters indicate the RC, the third and fourth indicate the CS, the fifth through seventh indicate the sequential recruitment of the subject at that CS, and the last three characters indicate the subjects name code (first three letters of the last name). No names or personal identification information are sent to the DCC or stored in the STAR*D database. 6.5. Statistical considerations Each level of treatment past level 1 is considered to be a unique RCT. The primary analyses make no attempt to control for prior treatments, assuming that the random treatment assignment will balance the prior treatments across the treatment groups. Secondary analyses may be conducted to determine if there is a prior treatment effect, especially if there is an imbalance of the prior treatments across the treatment groups.

7. Anticipated analytic approaches Data analyses from levels 2 and 3 are complicated by the fact that participants may accept or decline specific treatment strategies or substrategies (e.g., medication switch), in a so-called equipoise-stratified randomized design [68]. For level 2 analyses, data will be stratified by acceptability strata, and MantelHaenszel v2 tests will be used to make pairwise comparisons among the treatment options within the medication switch substrategy and between-treatment options within the medication augment substrategy. A satisfactory outcome for research purposes is defined as an HRSD17 <8. (Note: Participants who begin a level with HRSD17<8 will be excluded from analyses.) If no significant differences are detected within a substrategy, the data within the substrategy will be pooled. Otherwise the most effective treatment will be

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

131

used at the next level of testing. For example, if the pairwise comparisons of the medication switch show that SER is the most effective of the switch options, only the SER data will be used to represent the medication switch substrategy in comparisons with the other substrategies (i.e., medication augment, psychotherapy switch, or psychotherapy augment). If the pairwise comparisons of the medication switch option show that no one medication switch is more effective than the others, then all the data on medication switch options will be pooled when comparing this substrategy with other substrategies (i.e., medication augment, psychotherapy switch, or psychotherapy augment). Next, all pairwise comparisons will be made between the medication switch, medication augment, psychotherapy switch, and psychotherapy augment strategies. All comparisons will be made from contrasts among outcomes for participants randomly assigned to all elements of the contrast. For example, the comparison of medication switch to medication augment would include only those who had included both medication switch and medication augment as acceptable substrategies [68]. The analytic approach for level 3 is the same as the analytic approach for the medication switch and medication augment substrategies in level 2. The analyses of binary secondary endpoints will be conducted using the same analytic approach as used for the primary endpoints. Continuous secondary endpoints will be analyzed by a standard F-test for treatment effect applied to the two-way layout of strata and group. Exploratory analyses are planned of the naturalistic follow-up data outcomes and utilization of health care services during the 12-month follow-up period, including descriptive analyses of symptom severity, daily function, and the likelihood of time to relapse. Logistic regression and time-to-event (e.g., Cox proportional hazards) models will be used to assess the effect of factors associated with relapse and time to relapse, respectively.

8. Sample size, power, and effect size Levels 2 through 4 of STAR*D are considered separate RCTs, addressing the question, What should be done next considering that the prior treatment(s) failed? Therefore, we have made efforts to maintain an overall type I error rate of 0.05 within each level. 8.1. Level 2 The original National Institute of Mental Health Request for Application capped the total at 4000 participants. Given this studys design and previous reports in the literature and an expected response rate to CIT (level 1) of 50%, it was estimated that 2000 participants would enter level 2. To estimate the sample size of each pairwise comparison at level 2, one needs to first make assumptions about acceptability strata. Fig. 2 shows the mutually exclusive acceptability strata for the level 2 treatments that we anticipate. The number of participants in each stratum (in parentheses) is also included. For example, those entering stratum 1.1 have declined the medication switching substrategy altogether but are willing to accept randomization to medication augment or to psychotherapy either as a switch or augment substrategy. The numbers shown in each stratum are only estimates, since such an experiment has not been previously conducted. We arrived at these estimates using independent estimates by three experienced practitioners/clinical investigators (Drs. Fava, Rush, and Thase) with surprisingly close interrater

132

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

Fig. 2. Mutually exclusive level 2 acceptability stratum-specific randomizations.

agreement. These estimates were derived largely from inferences based on the results of published efficacy trials [13,16]. To determine the detectable effect size (v/N), consider the seven treatment options in level 2 [69]. We propose a step up procedure, taking advantage of the fact that there are two substrategies (medication switch and medication augment) where we can organize outcomes (response proportions): pVEN, pSER, pBUP for the three medication switch strategies and pCIT+BUP and pCIT+BUS for the two medication augment strategies. In addition, there are the response proportions from psychotherapy switch (pCT) and psychotherapy augment (pCIT+CT). First, we will test the null hypotheses of the equality of treatment options within the medication switch substrategy pVEN pSER ; pSER pBUP and medication augment substrategy pCITBUP pCITBUS , at level aH3 (two-sided). This is test 1. The next test, test 2, will compare the four substrategies (medication switch, medication augment, psychotherapy switch, psychotherapy augment) (4 3H2 6 pairwise comparisons). If any of the three null hypotheses from test 1 are rejected, then the treatment option with the superior outcome will be used to represent the treatment substrategy. If the null hypotheses from test 1 are not rejected, then the data within the medication switch and medication augment substrategies will be pooled. It is possible that within test 1, a null hypothesis would be rejected for one of the substrategies (e.g., medication switch) but not within the other substrategy (e.g., medication augment). In this situation, the

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

133

data within the substrategy that did not reject a null hypothesis would be pooled. Within the substrategy with a null hypothesis rejected, the treatment option with the superior outcome will be used to represent the treatment substrategy. If a = 0.025 for test 1 and test 2, the overall rate is <0.05. The Bonferroni corrected type I error used in the three test 1 comparisons will be 0:025H3 0:0083 and for the six test 2 comparisons will be 0:025H6 0:0043. These type 1 rates can then be used to identify the effect size that can be detected given the sample sizes described above [68]. Using the Mantel-Haenszel v2 test for association, and assuming 80% power, a type I error rate of 0:025H3 0:0083, a two-sided alternative hypothesis, a sample size of 287 per cell in each medication switch option, and a sample size of 354 in each medication augment option, an effect size of 0.205 can be detected for the medication switch comparisons and an effect size of 0.185 can be detected for the medication augment comparison. The estimate of 287 patients per medication switch option was obtained by summing the expected numbers of patients per option across the pertinent acceptability strata from Fig. 2. For each stratum, the number of patients per option was estimated by dividing the stratums expected sample size by the number of treatment options to which patients within the stratum accepted randomization (e.g., within stratum 0 the estimated sample size of 168 was divided by the number of treatment options, 7). The estimated numbers per option were then summed across those strata that included medication switch substrategy (i.e., strata 0, 1.2, 1.3, 1.4, 2.2, 2.3, 3.1). A similar approach was used for estimating the sample sizes for the treatment options in the medication augment substrategy. The detectable effect size for test 2 will depend on the results of test 1. If differences are found in test 1, either within the medication switch or medication augment substrategies, then only those participants who received the more advantageous treatment within that substrategy will be used in the following analyses. If not, the results will be pooled across treatment options within the substrategy. Therefore, the sample sizes for calculating the power of the next set of tests will be based on the results of test 1 and the data from Fig. 2. Table 5 presents the detectable effect sizes for each comparison, made according to the results of test 1 (if significant differences are not detected in test 1 within a substrategy, the results are pooled across treatment options; otherwise, the data for the most effective treatment are used). In the instance when the most effective treatment is chosen and an imbalance exists in the sample sizes of the two groups, the detectable effect size estimate is based on a balanced design and the larger of the two sample sizes.
Table 5 Estimated effect sizes by planned comparison for level 2 substrategies Substrategies Group 1 Medication switch Medication switch Medication switch Medication augment Medication augment Psychotherapy switch
a b

Not pooleda Group 2 Medication augment Psychotherapy switch Psychotherapy augment Psychotherapy switch Psychotherapy augment Psychotherapy augment n1 132 151 31 31 112 141 n2 132 151 31 31 112 141 Detectable effect size 0.228 0.213 0.470 0.247 0.247 0.220

Pooledb n1 396 453 93 93 224 141 n2 264 151 31 31 112 141 Minimum detectable effect size 0.131 0.123 0.271 0.271 0.175 0.220

Pooled: a single most effective treatment option is not identified within substrategies. Not pooled: a single most effective treatment option is identified within each substrategy.

134

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

For example, if we look at the comparison of the medication switch strategy to the medication augment strategy, patients from acceptability strata where both medication switch and medication augment are acceptable are included in the analyses. This would include the following acceptability strata: 0, 1.3, 1.4, and 2.3. The estimate of the number of patients assigned to each treatment option in the medication switch and medication augment strategies (n = 132) is obtained by summing the expected number of patients per option across the pertinent acceptability strata from Fig. 2 (stratum 0: 24 (168/ 7)+stratum 1.3: 4 (22/6)+stratum 1.4: 4 (22/6)+stratum 2.3: 100 (500/5) = 132). If the results of test 1 indicate that differences exist in the treatment options within the medication switch strategy and within the medication augment strategy, then only one treatment option for each strategy will be used in test 2. Thus, only 132 subjects in each treatment option would be used for test 2 (those assigned to the best treatment option within each treatment strategy). In this case, the detectable effect size would be 0.228. However, if no differences are detected within either strategy, then the patients within each strategy will be pooled. The medication switch strategy would then have 396 patients (three medication augment treatment strategies132 patients per treatment strategy) and the medication augment strategy would have 264 patients (two medication augment treatment strategies132 patients per treatment strategy). In this case, the minimum detectable effect size would be 0.131. 8.2. Level 3 The analytic approach for level 3 will be similar to that for level 2. For level 3, the number of acceptability strata and the number of comparisons are smaller. A total of 698 participants are expected to enter level 3. We assume that 25% will choose a switch strategy, 25% will choose an augment strategy, and 50% will be willing to accept all possible treatments. There will be only one test (e.g., test 1) in each of the strategies to determine if there is a more effective treatment in the switch strategy and in the augment strategy. A type I error rate of 0.025H2 will be used for each of these tests. Given this information, the sample size per treatment option for the test within each strategy would be 174 participants and the detectable effect size would be 0.253. The next step will be to compare the two treatment strategies, using the test 2 approach as described above, with a type I error rate of 0.025. The detectable effect sizes for this comparison are 0.234 for the nonpooled results and 0.165 for the pooled results. This error rate is larger than the error rate used for test 2 in level 2 due to the fact that there are fewer acceptability strata and fewer treatment comparisons involved in level 3. 8.3. Level 4 Assuming a sample size of 292 (146 per group), a type I error rate of 0.05, 80% power, and a twosided alternative hypothesis, an effect size of 0.164 can be detected among the response rates of the two treatment groups. Generally, an effect size of 0.2 is considered small, while 0.5 is considered moderate [68]. Therefore, at levels 2, 2A, 3, and 4, the study has the necessary power to detect small to moderate effects. Sample size and thus power will depend on the acceptability of strategies (e.g., switch versus augment) or substrategies (e.g., medication switch, psychotherapy switch, medication augment, psychotherapy augment). If we have underestimated the acceptability of certain substrategies, then power to compare them to other strategies or substrategies is obviously increased (i.e., more participants than anticipated will be more widely accepting of more treatment substrategies). If we overestimated the

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

135

acceptability of certain substrategies, then the power of comparisons involving them, and of treatment option comparisons within them, is reduced. Even if our estimates are overstated by 50%, however, nearly every substrategy still contains a sufficient sample to detect clinically meaningful differences in effectiveness. Furthermore, if some substrategies (e.g., medication augment) are rarely accepted, study resources will then be appropriately focused on comparing those treatments that have a good chance of acceptance in representative participant populations.

9. Utilization and cost data analyses Differences in utilization of services will also be calculated for each health outcome comparison among patients randomly assigned to treatment options within the same acceptability strata. Classifying care by type (depression-related, other mental, general medical) and setting (outpatient, emergency room, hospital), between-group differences from the censored and bimodal distributions often encountered with utilization data are tested separately by first comparing use versus no-use (e.g., logistic regression) and then volume-of-use among users [7072]. Differences are computed from twolevel hierarchical models that account for timed observations nested within patient. The skewed volume distributions are normalized with a log, square root, or other appropriate transformation. Cost subtotals are computed for each care type and setting by multiplying the expected probability of use by the predicted volume among users, and then by a schedule of unit-costs representing market-rates computed from national transaction charges. Total costs are computed by summing the type and setting subtotals. Sensitivity analyses are applied to contrast findings from different unit-cost schedules (e.g., Medicare fees, Department of Veterans Affairs Reasonable Charges). Confidence intervals are computed by bootstrapping samples.

10. Data monitoring and safety reporting The data and safety monitoring board (DSMB) (three psychiatrists, one medical ethicist, one patient advocate, one statistician, and one physician/patient advocate) meets every 3 months to monitor various aspects of the study, including participant recruitment, protocol compliance, and adverse events. Adverse events are recorded by study clinicians or CRCs at each clinic visit, based on spontaneous reports and on a participant-completed side effects checklist. Serious adverse events (those that result in hospitalization, prolongation of hospitalization, persistent disability, congenital anomaly, death, or that are lifethreatening) are reported verbally and in writing to the RC director, the STAR*D safety officer, the DSMB, the RC institutional review board, the Food and Drug Administration, and the manufacturer of the medication. Interim analyses were initially proposed but not requested by the DSMB so they were not included in the final design.

11. Current status As of June 1, 2003, 2555 subjects had been enrolled into level 1 at 41 CSs from 14 RCs. Of these, 771 had entered level 2, 164 had entered level 3, and 42 had entered level 4.

136

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

12. Conclusions STAR*D uses a randomized, controlled design to evaluate both the theoretical principles and clinical beliefs that currently guide the management of treatment-resistant depression in terms of symptoms, function, satisfaction, side-effect burden, and health care utilization and cost estimates. Given the dearth of controlled data, results should have substantial public health and scientific significance, since they are obtained in representative participant groups/settings, using clinical management tools that can easily be applied in daily practice. To ensure that treatment resistance is present before allowing participants to proceed to the next treatment level, we provide training and oversight for both the pharmacotherapy and psychotherapy to ensure reasonably high-quality care. This design (sometimes called a hybrid design) engages representative patients and providers as participants, while ensuring adequate care [73]. STAR*D entails several innovations in study design and methods as compared to classical efficacy trials in MDD (i.e., broad inclusion criteria to increase generalizability of results, involvement of both primary and specialty care settings, outcome measures to assess multiple domains, the routine clinical use of symptom ratings to inform clinical decisions, and the concurrent use of two [ROAs and IVR] independent methods of evaluation). One of the most innovative aspects of the study is the flexible approach to randomizations at levels 2 and 3, where participants are randomized only to those treatment substrategies that they find acceptablethe equipoisestratified randomized design [68]. Because of these unique design features, the results of the study will be generalizable to a large set of patients. Acknowledgements This project has been funded in part with federal funds from the National Institute of Health, National Institute of Mental Health under Contract N01-MH-90003. Additional funds were provided by the Betty Jo Hay Distinguished Chair in Mental Health, Rosewood Corporation Chair in Biomedical Science and the Sara M. and Charles E. Seay Center for Basic and Applied Research in Psychiatry (A.J.R.), and by MH-30915 (Mental Health Intervention Research Center) (M.E.T., D.J.K.). The following pharmaceutical companies are providing medication for STAR*D participants: Bristol-Myers Squibb, Forest Laboratories, GlaxoSmithKline, King Pharma, Novartis, Organon, Pfizer, and Wyeth-Ayerst. The authors appreciate the contributions of the NIMH STAR*D Scientific Advisory Group, the NIMH Protocol Oversight Committee, the thoughtful consultation by Health Technology Systems, Inc., and the guidance of the NIMH Data and Safety Monitoring Board. We appreciate the provision of educational materials by the Texas Department of Mental Health and Mental Retardation. The authors appreciate the secretarial support of Fast Word Information Processing Inc. (Dallas, Texas) and David Savage, and the administrative support of Kenneth Z. Altshuler, M.D., Stanton Sharp Distinguished Chair (past Chairman) and Eric Nestler, M.D., Ph.D., Lou and Ellen McGinley Distinguished Professor and Chairman, Department of Psychiatry, University of Texas Southwestern Medical Center. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. government.

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

137

Appendix A. Participating regional centers and clinical sites

138

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

139

140

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

References
[1] Greenberg PE, Stiglin LE, Finkelstein SN, Berndt ER. The economic burden of depression in 1990. J Clin Psychiatry 1993;54:405 18. [2] Weissman MM, Bruce ML, Leaf PJ, Florio LP, Holzer III C. Affective disorders. In: Robins LN, Regier DA, editors. Psychiatric disorders in America: the Epidemiologic Catchment Area Study: The Free Press. New York; 1991. p. 53 80. [3] Kessler RC, McGonagle KA, Zhao S, et al. Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the United States. Results from the National Comorbidity Survey. Arch Gen Psychiatry 1994;51:8 19. [4] Depression Guideline Panel. Clinical practice guideline. Number 5. Depression in primary care: volume 1. Detection and diagnosis: U.S. Department of Health and Human Services, Public Health Service, Agency for Health Care Policy and Research. Rockville, MD; 1993. AHCPR Publication No. 93-0550. [5] Simon GE, Von Korff M, Barlow W. Health care costs of primary care patients with recognized depression. Arch Gen Psychiatry 1998;52:850 6. [6] Mueller TI, Leon AC. Recovery, chronicity, and levels of psychopathology in major depression. Psychiatr Clin North Am 1996;19:85 102. [7] Keller MB, Harrison W, Fawcett JA, et al. Treatment of chronic depression with sertraline or imipramine: preliminary blinded response rates and high rates of undertreatment in the community. Psychopharmacol Bull 1995;31:205 12. [8] Judd LL, Akiskal HS, Maser JD, et al. A prospective 12-year study of subsyndromal and syndromal depressive symptoms in unipolar major depressive disorders. Arch Gen Psychiatry 1998;55:694 700. [9] Thase ME, Howland RH. Refractory depression: relevance of psychosocial factors and therapies. Psychiatr Ann 1994;24: 232 40. [10] Murray CJ, Lopez AD. Evidence-based health policylessons from the Global Burden of Disease Study. Science 1996;274:740 3. [11] Wells KB, Stewart A, Hayes RD, et al. The functioning and well-being of depressed patients: results from the Medical Outcomes Study. JAMA 1989;262:914 919. [12] Frank E, Karp JF, Rush AJ. Efficacy of treatments for major depression. Psychopharmacol Bull 1993;29:457 475. [13] Depression Guideline Panel. Clinical practice guideline. Number 5. Depression in primary care: volume 2. Treatment of major depression: U.S. Department of Health and Human Services, Public Health Service, Agency for Health Care Policy and Research. Rockville, MD; 1993. AHCPR Publication No. 93-0551. [14] Thase ME, Rush AJ. Treatment-resistant depression. In: Bloom FE, Kupfer DJ, editors. Psychopharmacology: fourth generation of progress: Raven Press. New York; 1995. p. 1081 97. [15] American Psychiatric Association. Practice guideline for major depressive disorder in adults. Am J Psychiatry 1993;150(Suppl 4):1 26. [16] Rosenbaum JF, Fava M, Nierenberg AA, Sachs GS. Treatment-resistant mood disorders. In: Gabbard G.O, editor. Treatments of psychiatric disorders. 3rd ed. American Psychiatric Press, Inc. Washington, DC; 2001. p. 1307 84. [17] Fava M, Davidson KG. Definition and epidemiology of treatment-resistant depression. Psychiatr Clin North Am 1996;19:179 200. [18] Jarrett RB, Rush AJ. Short-term psychotherapy of depressive disorders: current status and future directions. Psychiatry: Interpers Biol Process 1994;57:115 32. [19] Frank E, Prien R, Jarrett RB, et al. Conceptualization and rationale for consensus definitions of response, remission, recovery, relapse and recurrence in major depressive disorder. Arch Gen Psychiatry 1991;48:851 5. [20] Riso LP, Thase ME, Howland RH, et al. A prospective test of criteria for response, remission, relapse, recovery, and recurrence in patients treated with cognitive behavior therapy. J Affect Disord 1997;43:131 42. [21] Rush AJ, Thase ME. Psychotherapies for depressive disorders: a review. In: Maj M, Sartorius N, editors. WPA series in evidence and practice in psychiatry. Depressive disorders, vol. 1. 2nd ed. John Wiley & Sons, Ltd. Chichester, UK; 2002. p. 161 206. [22] Miller IW, Keitner GI, Schatzberg AJ, et al. The treatment of chronic depression, Part 3: psychosocial functioning before and after treatment with sertraline or imipramine. J Clin Psychiatry 1998;59:608 19. [23] Judd LL, Paulus MP, Wells KB. Socioeconomic burden of subsyndromal depressive symptoms and major depression in a sample of the general population. Am J Psychiatry 1996;153:1411 6.

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

141

[24] Judd LL. The clinical course of unipolar major depressive disorders. Arch Gen Psychiatry 1997;54:989 91. [25] Rush AJ, Trivedi MH. Treating depression to remission. Psychiatr Ann 1995;25:704 5. [26] Van Londen L, Molenaar RP, Goekoop JG, Zwinderman AH, Rooijmans HG. Three- to 5-year prospective follow-up of outcome in major depression. Psychol Med 1998;28:731 5. [27] Sackeim HA, Prudic J, Devanand DP, et al. A prospective, randomized, double-blind comparison of bilateral and right unilateral electroconvulsive therapy at different stimulus intensities. Arch Gen Psychiatry 2000;57:425 34. [28] McCombs JS, Nichol MB. Pharmacy-enforced outpatient drug treatment protocols: a case study of Medi-Cal restrictions for cefaclor. Ann Pharmacother 1993;27:155 61. [29] Crismon ML, Trivedi M.H, Pigott TA, et al. Texas Consensus Conference Panel: The Texas Medication Algorithm Project. Report of the Texas Consensus Conference Panel on medication treatment of major depressive disorder. J Clin Psychiatry 1999;60:142 56. [30] Fava M, Uebelacker LA, Alpert JE, et al. Major depressive subtypes and treatment response. Biol Psychiatry 1997;42:568 76. [31] Gilbert DA, Altshuler K.Z, Rago WV, et al. Texas Medication Algorithm Project: definitions, rationale and methods to develop medication algorithms. J Clin Psychiatry 1998;59:345 51. [32] Rush AJ, Crismon ML, Toprac MG, et al. Implementing guidelines and systems of care: experiences with the Texas Medication Algorithm Project (TMAP). J Pract Psychiatry Behav Health 1999;5:75 86. [33] Rush AJ, Rago WV, Crismon ML, et al. Medication treatment of the severely and persistently mentally ill: The Texas Medication Algorithm Project. J Clin Psychiatry 1999;60:142 56. [34] Hamilton M. A rating scale for depression. J Neurol Neurosurg Psychiatry 1960;23:56 62. [35] Hamilton M. Development of a rating scale for primary depressive illness. Br J Soc Clin Psychol 1967;6:278 96. [36] Rush AJ, Giles DE, Schlesser MA, et al. The Inventory for Depressive Symptomatology (IDS): preliminary findings. Psychiatry Res 1986;18:65 87. [37] Rush AJ, Gullion CM, Basco MR, Jarrett RB, Trivedi MH. The Inventory of Depressive Symptomatology (IDS): psychometric properties. Psychol Med 1996;26:477 86. [38] Mundt J. Interactive voice response systems in clinical research and treatment. Psychiatr Serv 1997;48:611 23. [39] Fava M, Rush AJ, Trivedi MH, et al. Background and rationale for the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) study. In: Dunner DL, Rosenbaum JF, editors. Psychiatric clinics of North America, vol. 26. W.B. Saunders. Philadelphia; 2003. p. 457 94. [40] Rush AJ, Carmody T, Reimitz PE. The Inventory of Depressive Symptomatology (IDS): Clinician (IDS-C) and self-report (IDS-SR) ratings of depressive symptoms. Int J Methods Psychiatric Res 2000;9:45 59. [41] Rush AJ, Trivedi MH, Ibrahim HM, et al. The 16-item Quick Inventory of Depressive Symptomatology (QIDS) Clinician Rating (QIDS-C) and Self-Report (QIDS-SR): a psychometric evaluation in patients with chronic major depression. Biol Psychiatry 2003;54:573 83. [42] Trivedi MH, Ibrahim HM, Carmody TC, et al. A psychometric evaluation of the Quick Inventory of Depressive Symptomatology (QIDS) in public sector patients with mood disorders. Psychol Med, [in press]. [43] Jarrett RB, Kraft D, Doyle J, et al. Preventing recurrent depression using cognitive therapy with and without a continuation phase: a randomized clinical trial. Arch Gen Psychiatry 2001;58:381 8. [44] Shaw BF. Specification of the training and evaluation of cognitive therapists for outcome studies. In: Williams JBW, Spitzer RL, editors. Psychotherapy research: where we are and where we should go: Guilford Publications. New York; 1984. p. 173 89. [45] American Psychiatric Association. Diagnostic and statistical manual of mental disorders. 4th ed. American Psychiatric Press. Washington, DC; 1994. [46] World Medical Association Declaration of Helsinki. Ethical principles for medical research involving human subjects. Adopted by the 18th WMA General Assembly, Helsinki, Finland, June 1964; and amended by the 29th WMA General Assembly, Tokyo, Japan, October 1975; 35th WMA General Assembly, Venice, Italy, October 1983; 41st WMA General Assembly, Hong Kong, September 1989; 48th WMA General Assembly, Somerset West, Republic of South Africa, October 1996; and the 52nd WMA General Assembly, Edinburgh, Scotland, October 2000. Note of Clarification on Paragraph 29 added by the WMA General Assembly, Washington; 2002. [47] Zimmerman M, Mattia JI. The reliability and validity of a screening questionnaire for 13 DSM-IV Axis I disorders (the Psychiatric Diagnostic Screening Questionnaire) in psychiatric outpatients. J Clin Psychiatry 1999;60:677 83. [48] Linn BS, Linn MW, Gurel L. The cumulative illness rating scale. J Am Geriatr Soc 1968;16:622 6.

142

A.J. Rush et al. / Controlled Clinical Trials 25 (2004) 119142

[49] Fava GA, Rafanelli C, Grandi S, Conti S, Belluardo P. Prevention of recurrent depression with cognitive therapy: preliminary findings. Arch Gen Psychiatry 1998;55:816 20. [50] Paykel ES, Scott J, Teasdale JD, et al. Prevention of relapse in residual depression by cognitive therapy. A controlled trial. Arch Gen Psychiatry 1999;56:829 35. [51] Guy W. ECDEU Assessment manual for psychopharmacology: U.S. Government Printing Office, U.S. Dept. of Health. Washington, DC; 1976. Education and Welfare Publication No. 76-338. [52] Ware Jr J, Kosinski M, Keller SD. A 12-item short-form health survey: construction of scales and preliminary tests of reliability and validity. Med Care 1996;34:220 33. [53] Mundt JC, Marks IM, Shear MK, Greist JH. The work and social adjustment scale: a simple measure of impairment in functioning. Br J Psychiatry 2002;180:461 4. [54] Reilly MC, Zbrozek AS, Dukes EM. The validity and reproducibility of a work productivity and activity impairment instrument. Pharmacoeconomics 1993;4:353 65. [55] Endicott J, Nee J, Harrison W, Blumenthal R. Quality of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q): a new measure. Psychopharmacol Bull 1993;29:321 6. [56] Kashner TM. Agreement between administrative files and written medical records: a case of the Department of Veterans Affairs. Med Care 1998;36:1324 36. [57] Clark RE, Ricketts SK, McHugo G.J. Measuring hospital use without claims: a comparison of patient and provider reports. Health Serv Res 1996;31:153 69. [58] Dresser MVB, Feingold L, Rosenkranz SL, Coltin KL. Clinical quality measurement: comparing chart review and automated methodologies. Med Care 1997;35:539 52. [59] Kashner TM, Rost K, Smith GR, Lewis S. An analysis of panel data: the impact of a psychiatric consultation letter on the expenditures and outcomes of care for patients with somatization disorder. Med Care 1992;30:811 21. [60] Magaziner J, Simonsick EM, Kashner TM, Hebel JR. Patient-proxy response comparability on measures of patient health and functional status. J Clin Epidemiol 1988;41:1065 74. [61] Quam L, Ellis LBM, Venus P, et al. Using claims data for epidemiologic research: the concordance of claim-based criteria with the medical record and patient survey for identifying a hypertensive population. Med Care 1993;31:498 507. [62] Kobak KA, Greist JH, Jefferson JW, Mundt JC, Katzelnick DJ. Computerized assessment of depression and anxiety over the telephone using Interactive Voice Response. MD Comput 1999;16(3):64 8. [63] Kobak KA, Greist JH, Jefferson JW, Mundt JC, Katzelnick DJ. Computer-administered clinical rating scales: a review. Psychopharmacology 1996;127:291 301. [64] Rice DP, Hodgson TA, Kapstein AN. The economic costs of illness: a replication and update. Health Care Financ Rev 1985;7:61 80. [65] Rice DP, Miller LS. Economic burden of affective disorders. In: Hu T-W, Rupp A, editors. Advances in health economics and health services research, vol. 14. JAI Press. Greenwich, Connecticut; 1993. p. 37 54. [66] Rice DP. Economic burden of mental disorders in the United States. TEN: The Economics of Neuroscience 1999;1:40 4. [67] Stoudemire A, Frank R, Hedemark N, Kamlet M, Blazer D. The economic burden of depression. Gen Hosp Psychiatry 1986;8:387 394. [68] Lavori PW, Rush AJ, Wisniewski S, et al. Strengthening clinical effectiveness trials: equipoise-stratified randomization. Biol Psychiatry 2001;50:792 801. [69] Cohen J. Statistical power analysis for the behavioral sciences: Lawrence Erlbaum Associates. Hillsdale, New Jersey; 1988. [70] Duan N, Manning WG, Morris C.N, Newhouse JP. A comparison of alternative models for the demand for medical care. J Bus Econ Stat 1983;1:115 26. [71] Pohlmeier W, Ulrich V. An econometric model of the two-part process in the demand for health. J Hum Resour 1995;30:339 61. [72] Maddala GS. Limited-dependent and qualitative variables in econometrics: Cambridge University Press. New York; 1988. [73] Rush AJ. Linking efficacy and effectiveness research in the evaluation of psychotherapies. In: Miller NE, McGruder KM, editors. Cost-effectiveness of psychotherapy: a guide for practitioners, researchers, and policy makers: Oxford University Press. New York; 1999. p. 26 32.