Similar Effectiveness of Paroxetine, Fluoxetine, and Sertraline in Primary Care: A Randomized Trial

Context Selective serotonin reuptake inhibitors (SSRIs) are the most commonly prescribed class of antidepressant, yet it is not known whether one SSRI is more effective than another.

Objective To compare the effectiveness of 3 SSRIs (paroxetine, fluoxetine, and sertraline) in depressed primary care patients.

Design Open-label, randomized, intention-to-treat trial, with patient enrollment occurring in April-November 1999.

Setting Thirty-seven clinics in 2 US primary care research networks.

Patients A total of 573 depressed adult patients for whom their primary care physician thought that antidepressant therapy was warranted and who completed a baseline interview.

Interventions Patients were randomly assigned to receive paroxetine (n = 189), fluoxetine (n = 193), or sertraline (n = 191) for 9 months. Primary care physicians were allowed to switch patients to a different SSRI or non-SSRI antidepressant if they did not adequately respond to or tolerate the initial SSRI.

Main Outcome Measures The primary outcome measure was change in the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) Mental Component Summary score (range, 0-100), compared across treatment groups at 1, 3, 6, and 9 months. Secondary outcomes included other depression and psychological measures, multiple measures of social and work functioning, and other domains of health-related quality of life, such as physical functioning, concentration and memory, vitality, bodily pain, sleep, and sexual functioning.

Results Follow-up interviews were successfully completed in 94% of patients at 1 month, 87% at 3 months, 84% at 6 months, and 79% at 9 months. Responses to the 3 SSRIs were comparable on all measures and at all time points. The mean change in the SF-36 Mental Component Summary score at 9 months was + 15.8 in the paroxetine group, + 15.1 in the fluoxetine group, and + 17.4 in the sertraline group. The drugs were also associated with similar incidences of adverse effects and discontinuation rates.

Conclusions The SSRI antidepressants paroxetine, fluoxetine, and sertraline were similar in effectiveness for depressive symptoms as well as multiple domains of health-related quality of life over the entire 9 months of this trial.

Selective serotonin reuptake inhibitors (SSRIs) have become the most commonly prescribed class of antidepressants, accounting for more than $3 billion of annual prescription costs in the United States and growing by approximately 25% each year.¹ Compared with tricyclic antidepressants, SSRIs have a more favorable adverse effect profile, simpler dosing, and less toxic effects in the event of an overdose.^2,3 Although each SSRI has demonstrated effectiveness, there are no clinical trial data supporting the superiority of one SSRI relative to another. Studies comparing SSRIs have had short follow-up periods, had a limited range of outcome measures, and have predominantly recruited participants from psychiatric inpatient settings.⁴ Thus, it would be important to compare the effectiveness of SSRIs using a broad range of clinically relevant outcomes (eg, social and work functioning, well-being, and other domains of health-related quality of life) that extends follow-up into the maintenance phase of treatment. Moreover, because most depression is treated in primary care, studies in this clinical venue are critical.⁵

A Randomized Trial Investigating SSRI Treatment (ARTIST) was designed to compare the effectiveness of 3 SSRI antidepressants in depressed primary care patients. The purpose of ARTIST was to examine patient response to different SSRIs in terms of depression and other health outcomes during the acute as well as maintenance phases of antidepressant therapy.

Patients were enrolled over an 8-month period (April-November 1999) from clinical practices in 2 primary care research networks. The Primary Care Network is a not-for-profit voluntary organization of more than 8600 family practitioners, internists, and pediatricians who care for more than 10 million patients throughout the United States. The group is not organized along reimbursement lines, but rather represents a network of primary care practitioners interested in optimizing the care they provide through continuing education and practice-based research initiatives. The Duke Primary Care Research Consortium is an academic site management organization within the Duke University Health System consisting of more than 150 family physicians, internists, and pediatricians who collaborate in adult and pediatric clinical outcome trials. The network has 22 community-based practices that provide care for more than 300 000 patients in 8 counties of North Carolina.

Patients were eligible if they were at least aged 18 years, received their primary care from a participating physician, had access to a home telephone, and were determined by their primary care physician (PCP) to have a depressive disorder for which antidepressant therapy was warranted. Exclusion criteria included (1) cognitive impairment (eg, dementia, psychosis) severe enough to preclude an adequate interview; (2) unable to read, write, or speak English; (3) terminal illness; (4) nursing home residence; (5) actively suicidal; (6) taking an SSRI either currently or any time within the past 2 months; (7) taking a non-SSRI antidepressant either for depression (any dose level) or for a nondepressive disorder at more than low doses (eg, >50 mg of amitriptyline or its equivalent); (8) not eligible for the starting doses of paroxetine (20 mg), fluoxetine (20 mg), or sertraline (50 mg); (9) history of bipolar disorder; (10) active cocaine or opiate abuser; and (11) pregnancy, breast-feeding, or planning to get pregnant in the next 9 months. The study was approved by the institutional review boards of Indiana University, Research Triangle Institute, and Duke University, and by a central institutional review board for the Primary Care Network.

The study was designed to resemble real-world practice in several respects.^6,7 First, the decision to initiate an antidepressant was based strictly on the PCP's judgment that there was clinical depression warranting treatment rather than insisting that criteria for a specific diagnosis, such as major depressive disorder, be established. Second, neither patients nor PCPs were blinded to treatment assignment, because blinding would preclude typical clinical management. Third, all decisions regarding dose changes, medication discontinuation, or switch to a different antidepressant were made by patients and their PCP. Other than for reasons of patient safety (primarily suicidal ideation), information gathered by telephone interview was not disclosed to the PCP.

After seeing the PCP, written informed consent was obtained from patients by clinic personnel, who then used a touch-tone telephone procedure to randomly assign patients to treatment. Patients initiated treatment with the recommended starting dose of 20 mg of paroxetine, 20 mg of fluoxetine hydrochloride, or 50 mg of sertraline. The PCP could adjust the dose or change to a different antidepressant based on clinical response, and treatment was to be continued for the 9-month trial. The coordinating center had stored a prearranged list of randomized assignments to paroxetine, fluoxetine, and sertraline to ensure allocation concealment for this study. Patients were randomly assigned within PCP in blocks of 3. Each block was balanced with up to 7 assignments for each of the 3 drugs. This block randomization ensured overall balance between treatment assignments for each PCP and was also intended to limit the maximum number of patients per PCP to 21; one clinician with a particularly large practice enrolled 30 patients.

Immediately after enrollment, patients received a pharmacy benefits card that covered the costs for both SSRIs as well as any non-SSRI antidepressants that the PCP prescribed during the 9 months of the trial. Providing this card minimized differences in patient compliance that might be due to socioeconomic factors or variable copay of different insurance plans.

Computer-assisted telephone interviews were used to assess outcomes. The goal was to complete baseline interviews within 72 hours of enrollment although efforts to contact the patients were continued for up to a week after enrollment. For follow-up interviews, efforts to contact participants began a week prior to the target follow-up date and continued for up to 2 weeks after this date. Interviewers made up to 6 callbacks, if needed, to complete an interview. Supervisory staff silently monitored a 10% sample of interviews for quality control purposes. As reimbursement for their time, study patients received $20 for each completed telephone interview, with additional payments of $20 for 4 completed interviews and $30 for 5 completed interviews. Thus, a patient who completed all 5 interviews received a maximum of $150.

Depression was assessed with several measures. The primary depression outcome was the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) Mental Component Summary (MSC), which uses a regression algorithm that incorporates all 8 SF-36 subscales as a measure of mental health and has been established as a sensitive outcome measure in studies of clinical depression.⁸ Secondary analyses also examined 2 SF-36 subscales (mental health and role-emotional) that correlate highly with depression.^9,10 The Symptoms Checklist (SCL-20), a modified subscale of the Hopkins Symptom Checklist and Brief Symptom Inventory, has demonstrated sufficient sensitivity to detect differences in depression severity change between treatment groups in primary care trials.^11-13 The Primary Care Evaluation of Mental Disorders (PRIME-MD)^14,15 depression module was used to determine the number of Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV), depressive symptoms as well as the diagnostic subgroups of depressive disorders (major depression, dysthymia, and minor depression).

Besides depression, we evaluated 5 other psychological outcome measures: the positive well-being scale from the RAND Medical Outcomes Study (MOS) questionnaire¹⁶; the hopefulness scale from the Health Outcomes Study questionnaire¹⁷; the somatization severity scale and the 3-symptom anxiety screener from the Patient Health Questionnaire¹⁸; and the disposition (self-esteem) scale from the Health and Daily Living Form.¹⁹

Social function was assessed by 3 measures: the SF-36 social functioning scale, the Quality of Social Interaction scale,²⁰ and the Quality of Close Relationships scale.¹⁹ Work function was assessed by 3 scales (13 items) from the Work Limitations Questionnaire²¹ (output demands, time management, and interpersonal relations), plus questions about percentage of work effectiveness and number of days of impaired work functioning in the past 2 weeks. Other health-related quality of life measures included the 5 remaining SF-36 scales (physical functioning, role-physical, bodily pain, general health perceptions, vitality) and 3 additional MOS scales (concentration and memory, sleep, and sexual functioning).

Antidepressant use was assessed at 1, 3, 6, and 9 months using 38 structured questions addressing SSRI compliance, current antidepressant and its dose, and reasons for antidepressant changes, including occurrence of adverse effects.

Other variables measured at baseline included alcohol use, recent anxiety attacks, past history of treatment for depression, and all self-reported clinic and emergency department visits in the past 3 months and hospital days in the past year. Also, a brief 9-item measure of depression severity was administered in the clinic immediately after study enrollment.¹⁸ Two weeks prior to telephone interviews at 1, 3, 6, and 9 months, a questionnaire was mailed to each subject to assess clinic, emergency department, and hospital use, including visits to a mental health professional. At the end of the study, 3 questions about satisfaction with medication and treatment for depression were asked.

The internal reliability of outcome measures was good, with a Cronbach α for all scales being .70 or higher (except the MOS Sleep scale, which was .67).

The primary outcome measure was the MCS score. A 5-point MCS difference represents half of an SD, which is a medium effect size.^8,22 To detect this degree of difference with 90% power (α = .05; β = .10; SD = 12), 125 patients per SSRI drug group were required. Assuming a 10% attrition at each of the 4 follow-up assessments, we decided to enroll 600 patients, which exceeds the required number of (3)(125)/(0.9)⁴.

All patients who completed a baseline interview were included in the intent-to-treat analysis using the drug to which the patient was randomized as the treatment variable. This means that patients who switched treatments remained in the original SSRI group to which they were randomized for the purposes of analysis. Mixed model analysis of variance (using SAS PROC MIXED procedure; SAS Institute, Cary, NC) was used for most comparisons of continuous outcomes among the 3 drug groups. This method does not impute missing data or carry the last observation forward but rather uses a maximum likelihood approach that gives valid results under missing at random assumptions.²³ SAS glimmix macro for generalized linear mixed models was used for only a few secondary outcomes, namely counts (eg, number of days of impaired functioning) and categories (eg, proportion of patients with major depression). Site, sex, and baseline score were included in the models as covariates. Main effects were month, drug, and the drug-by-month interaction. Random effects were included for clinic, physician within clinic, and patient within physician. Patient within physician was treated as a repeated random effect to correlate the data from multiple interviews and allowed the variance estimates to be different at each interview. If the drug-by-month interaction was not significant, the drug main effect was used to determine significance. Because the MCS was the primary outcome measure, this P value was not adjusted for multiple hypothesis testing. For the remaining secondary outcome measures, the P values were adjusted for multiple hypothesis testing using the Sidak method²⁴: adjusted P value = 1 – (1 – unadjusted P value)^{# tests}. Thus the P values for the additional psychological outcome measures were adjusted for performing 9 tests and the P values for the various other outcome measures were adjusted for performing 16 tests.

Participating practitioners included 26 PCPs from 25 Primary Care Network practice sites, and 51 PCPs from 12 Primary Care Research Consortium practice sites. There were 353 patients enrolled from the Primary Care Network and 248 from the Primary Care Research Consortium sites. The median number of patients enrolled per practitioner was 6 (range, 1-30).

Participant disposition from initial contact to trial completion is shown in Figure 1. Enrollment occurred over 8 months, from April through November 1999. Of 601 patients who provided informed consent and who were randomized to treatment at their primary care clinic visit, 573 completed the baseline telephone assessment. The 28 prebaseline dropouts were demographically similar to the 573 baseline completers, had slightly less severe depression (mean Patient Health Questionnaire score of 12.5 vs 14.3), and included 13 patients randomly assigned to receive paroxetine, 7 to fluoxetine, and 8 to sertraline. Primary analysis for outcomes was conducted in 546 patients who completed 1 or more follow-up interviews; these patients were similar at baseline to the 27 patients with no outcome information. Allocation concealment was maintained by having assignment occur after the patient saw the physician and by using a touch-tone telephone randomization procedure performed through the central coordinating center. Because patients were randomized in small blocks of 3, we further tested for bias in treatment group assignment (using the method described by Berger and Exner²⁵) and detected no selection bias. Follow-up interviews were successfully completed in 94% of patients at 1 month, 87% at 3 months, 84% at 6 months, and 79% at 9 months.

Baseline patient characteristics are shown in Table 1 while baseline scores on all outcome measures are shown in Table 2 and Table 3. Treatment groups were comparable at baseline on all variables except sex (Table 1) and SF-36 bodily pain score (Table 3). Overall, the study sample had a mean age of 46 years (range, 18-96), was 79% women, and had a racial or ethnic distribution of 84% white, 13% black, and 3% other. Major depression was present in 74% of the sample, dysthymia in 18%, and minor depression in 8%. Depression was moderately severe as ascertained by the mean (SD) MCS score of 30.9 (12.0), mean (SD) SCL-20 of 1.66 (0.72), and mean (SD) number of DSM-IV depression symptoms 5.8 (2.2) out of a maximum of 9. One third of the patients reported a past history of treatment for depression. In the past month, 35% had experienced an anxiety attack and 45% reported some use of alcohol.

All 3 SSRI groups had substantial improvement in depression and other health-related quality of life domains. In the entire sample, the proportion of patients who met criteria for major depression dropped from 74% at baseline to 32% at 3 months, and 26% at 9 months. Similarly, depressive symptom severity improved considerably over the 9-month trial: the mean MCS score went from 30.9 to 48.3, the mean SCL-20 from 1.66 to 0.78, the mean SF-36 mental health score from 42.9 to 71.3, and the mean number of DSM-IV depressive symptoms from 5.8 to 2.9. The proportion of patients reporting a recent anxiety attack declined from 35% to 14%.

Table 2 summarizes depression and other psychological outcomes by SSRI treatment group, and Table 3 summarizes work, social, and other health-related quality of life outcomes. The magnitude of the mean change for most scales was quite similar for the 3 SSRI groups. After adjustment for multiple comparisons, there were no significant group differences for any of the 3- and 9-month outcomes. Although not shown in Table 2 and Table 3, there were also no group differences for any outcomes at 1 or 6 months. Examining common cut points for classifying recovery, the proportion of patients who achieved an MCS score of 40 or greater by 9 months was 81% in the paroxetine, 77% in the fluoxetine, and 84% in the sertraline groups while the proportion who reached an SCL-20 score of 1.0 or less was 69%, 67%, and 74%, respectively. Finally, there were no group differences in 3- or 9-month MCS scores when the 28 randomized but prebaseline dropout patients were included in an imputed worst-rank analysis.²⁶

Two subgroup analyses were conducted. Patients with major depression were examined as one subgroup because they have been the target population in the largest number of trials establishing the efficacy of antidepressants. Patients who continued receiving the SSRI to which they were initially assigned for the entire 9 months of the trial were also examined in a completers analysis. There were 418 patients who met diagnostic criteria for major depression at baseline: 135 (71%) taking paroxetine, 143 (74%) taking fluoxetine, and 140 (73%) taking sertraline. There were 256 patients who continued taking the SSRI to which they were initially randomized for the full duration of the trial: 77 (41%) taking paroxetine, 97(50%) taking fluoxetine, and 82 (43%) taking sertraline. As with the intent-to-treat analysis for the entire sample, there were no outcome differences among the 3 SSRIs in either subgroup—major depression or continuers.

We also performed an analysis in which all patients who dropped out, stopped the drug to which they were initially assigned, or switched to a different antidepressant were considered treatment failures for the initial drug. Using this most conservative intent-to-treat analysis for the entire sample, the proportion of patients who continued on the SSRI to which they were initially randomized for 9 months and achieved an MCS score of 40 or greater was 34% in the paroxetine, 37% in the fluoxetine, and 37% in the sertraline groups.

We evaluated whether treatment effects in older patients or in those with comorbid anxiety differed by drug group. Analyzing age as a binary variable (≥60 years vs younger), there were no interactions between SSRI group and age for the outcomes of MCS, SCL-20, or number of DSM-IV depressive symptoms. As shown in Table 2, there were no differences between drug groups in anxiety symptom count at baseline or at follow-up. Also, there were no interactions between SSRI group and anxiety for the 3 depression outcomes. Finally, there was no relationship between the number of patients enrolled by a PCP and depression outcomes.

At the end of the study, patients were asked to rate their satisfaction with depression treatment on a 5-point scale (excellent, very good, good, fair, or poor). Among the 451 respondents, 81% rated their satisfaction as good or better with the SSRI medication prescribed, 83% with the physician's interest in their depression, and 84% with their overall depression treatment. Satisfaction did not differ among the 3 SSRI treatment groups. Visits to a mental health professional were infrequent, averaging 0.96 mental health professional visits per patient over 9 months, and did not differ among treatment groups.

The proportion of patients who stopped or switched to another antidepressant was 13% at 1 month, 23% at 3 months, 32% at 6 months, and 40% at 9 months and did not differ by drug group. The final dose on initially randomized drug was 23.5 mg for the paroxetine group (with 91% taking ≥20 mg), 23.4 mg for the fluoxetine group (with 96% taking ≥20 mg), and 72.8 mg for the sertraline group (with 94% taking ≥50 mg).

Table 4 highlights the medication outcomes, reasons for stopping or switching antidepressants, and the most common adverse effects prompting discontinuation or switching. Since the latter represents only those adverse effects severe enough to prompt a medication change, it may underestimate the proportion of patients who develop clinically significant problems. Therefore, we also examined how many participants reported, at their 1-month interview, being "bothered a lot" by any of the 14 PRIME-MD physical symptoms not present at baseline. Using this threshold, the emergence of new bothersome symptoms did not differ by drug group and was uncommon in the overall sample, including bowel complaints (2.0%), stomach pain (1.5%), nausea or dyspepsia (1.3%), insomnia (1.3%), dizziness (1.1%), and headache (0.4%).

Since sexual dysfunction is a particular concern with SSRI antidepressants, we examined the 4 individual items constituting the sexual functioning scale: sexual satisfaction, erectile dysfunction or inadequate lubrication, difficulty having orgasm, and ability to satisfy sexual partner. Mean changes from baseline to 12 weeks were small and typically suggested slight improvement rather than worsening (Table 5). There were no significant differences between drugs for any of the 4 items.

ARTIST results unequivocally demonstrate the lack of differences among 3 SSRIs across a broad range of outcomes over 9 months. Paroxetine, fluoxetine, and sertraline were similar in the magnitude and time course of their effectiveness in ameliorating depression as well as improving other psychological outcomes, social and work functioning, and multiple other domains of health-related quality of life. The 3 SSRIs were also associated with a similar incidence of clinically significant adverse effects and rates of discontinuing or switching medication. Strengths of ARTIST include its large sample size, random assignment to an SSRI agent, rich battery of outcome measures, outcome assessment during both acute and maintenance periods of depression therapy, and a study design reflecting real-world practice. The involvement of multiple practices across the United States increases the generalizability of ARTIST findings.

Other than patients being randomly assigned to their initial SSRI treatment, all subsequent treatment decisions were under the control of the patients and their PCPs who could adjust medication dosage level or change antidepressants as they would in clinical practice. However, outcomes were assessed by telephone interviewers using validated measures rather than relying on evaluation by the treating PCP.

Three theoretical explanations for ARTIST findings would be inadequate sample size, inappropriate patient selection, and a restricted range of outcome measures. None of these factors seem particularly likely. Regarding sample size, attrition was even less than initially estimated. Thus, the number of patients still participating in follow-up assessments at 9 months meant our study had 94% power to detect a 5-point MCS difference between SSRI drug groups. The power to detect 4- and 3-point MCS differences (effect sizes of .33 and .25) was 81% and 57%, respectively. Thus, failure to detect SSRI differences is not due to inadequate power.

Patient selection also seemed appropriate. Patients initiated SSRI therapy based strictly on the PCP's judgment that there was a clinical depression warranting active treatment rather than requiring that a specific psychiatric diagnosis or depression severity threshold be established by a structured interview. Despite this pragmatic approach, however, most patients proved to have either major depression or dysthymia, both of which are established indications for antidepressant therapy.²⁷ We confirmed that all patients had at least moderate symptom severity according to their mean baseline scores on all depression measures. Finally, the degree of improvement in the mean SF-36 MCS score—15 points at 3 months and 17 points at 9 months—is at least as great as that seen in longitudinal studies of patients recovering from clinical depression for which the average increase in MCS score is 10.9 points.⁸

Outcome assessment in ARTIST was comprehensive and longitudinal. The assessment battery consisted of a broad array of depression and other psychological scales, multiple measures of social and work functioning, and other depression-relevant domains of health-related quality of life. Moreover, outcomes were assessed at 4 times during the acute and maintenance phases of antidepressant treatment. Despite this rather exhaustive approach to evaluation, differences in SSRI effectiveness were not demonstrated.

The pattern of rapid improvement during the first 4 weeks of therapy followed by more gradual improvement over the ensuing months was consistent across multiple domains. Although Mintz and colleagues²⁸ reported that improvement in work functioning may lag several months behind improvement in depressive symptoms, this conclusion was based on secondary analysis of heterogenous trials involving small numbers of patients, many of whom received psychotherapy rather than antidepressant medication. A large placebo-controlled trial showed improvement across multiple domains, including role functioning, within the first 6 weeks of starting an SSRI.²⁹

Patients were no more likely to stop or switch from one SSRI to another, and the reasons for discontinuation and adverse effect profiles for the 3 SSRIs were similar. Interestingly, sexual function tended to remain unchanged or slightly improved whether measured by the MOS composite sexual functioning scale or its 4 individual items. The reliability and validity of this MOS measure is well established,¹⁶ and there is preliminary evidence supporting its sensitivity to change.³⁰ Future studies including additional sexual function measures can further verify whether, as ARTIST data suggest, sexual functioning in depressed patients treated with an SSRI is, on average, more likely to improve than to worsen.

Several study limitations should be mentioned. First, the naturalistic design meant the number of patients remaining on their initially assigned drug declined as the 9-month trial progressed. However, discontinuation or switch rates did not differ among SSRI groups, and our findings were unchanged when analysis was restricted to patients who continued for the entire 9 months on the SSRI to which they were initially randomized. Second, a detailed inquiry about medication dosing, compliance, adverse effects, and reasons for switching or discontinuation was performed at the beginning of the telephone interview; therefore, telephone interviewers were technically not blinded to treatment assignment. However, primary and secondary outcome measures in this study were based on fully structured questions using standard response options that minimize interviewer interpretation or bias. Third, all antidepressants were provided to subjects at no cost to minimize the influence of socioeconomic status on outcomes. However, this is a difference from usual clinical practice for which the ability to pay may be one factor that could affect adherence. Fourth, one commonly prescribed SSRI—citalopram—was not investigated, as well as newer SSRIs that may be approved in the near future. However, the several comparator studies of citalopram with another SSRI have not demonstrated superiority.⁴

Previous studies, typically comparing 2 antidepressants in psychiatric inpatients, have shown that SSRIs are equally efficacious with one another as well as with other newer antidepressants in alleviating depressive symptoms.⁴ A retrospective chart review study suggested comparable effectiveness as well as discontinuation rates of paroxetine, fluoxetine, and sertraline.³¹ Our primary care–based randomized clinical trial demonstrates that these 3 SSRIs do not differ across a wide array of psychological, social, work, or other health-related quality of life domains in either the magnitude or the time course of response. With generic SSRI antidepressants now available, ARTIST results have important implications for health care costs. Although there may be some characteristics of any medication that distinguish its use in a particular patient, our findings suggest that in general none of the 3 SSRIs in this study can be recommended over another in terms of effectiveness.