Treating depression after initial treatment failure: Directly comparing switch and augmenting strategies in STAR*D

Abstract

Objective:

Augmenting or switching antidepressant medications are common next-step strategies for depressed patients failing initial medication treatment. These approaches have not been directly compared, thus our objectives are to compare outcomes for medication augmentation versus switching for patients with Major Depressive Disorder in the Sequenced Treatment Alternatives to Relieve Depression clinical trial (STAR*D).

Methods:

We conducted a retrospective analysis of participants ages 18–75 with DSM-IV non-psychotic depression who failed to remit with initial treatment in STAR*D (N=1,292). We compared depressive symptom remission, response, and quality of life among participants in each study arm using propensity score matching to minimize selection bias.

Results:

The propensity score matched augment (n = 269) and switch (n = 269) groups were well balanced on measured characteristics. Neither the likelihood of remission (RR:1.14, 95% CI:0.82–1.58) or response (RR:1.14, 95%CI:0.82–1.58), nor the time to remission (Log Rank Test, p = 0.946) or response (Log Rank Test, p = 0.243) differed by treatment strategy. Similarly, quality of life did not differ. Post-hoc analyses suggested that augmentation improved outcomes for patients tolerating ≥12 weeks of initial treatment and those with partial initial treatment response.

Conclusions:

For patients receiving and tolerating aggressive initial antidepressant trials, there is no clear preference for next step augmentation versus switching. Findings tentatively suggest that those who complete an initial treatment of 12 weeks or more and have a partial response with residual mild depressive severity may benefit more from augmentation relative to switching.

INTRODUCTION

Major Depressive Disorder (MDD) is projected to be the second leading cause of disability worldwide by 2020.¹ While existing antidepressant treatments are often effective, as many as 40% of depressed patients fail to respond following first-line treatment.² Of greater concern, only one-third of patients fully recover, or remit, with the initial treatment attempt.³ Improving clinical remission rates early in the course of treatment could significantly improve the quality of care for patients with depression.

However, limited evidence is available to guide clinical decision-making after initial treatment failure. Adding an additional medication to augment initial treatment or switching to a different antidepressant are the most common secondary treatment strategies, but these approaches have not been directly compared. The Sequenced Treatment Alternatives to Relieve Depression trial (STAR*D), a large scale effectiveness clinical trial, was designed to provide evidence on preferred treatment of MDD after initial treatment failure.^{4, 5} Among patients who did not achieve remission with first-line citalopram treatment (level-1), approximately one-third of patients augmenting treatment and one-quarter of patients switching treatment achieved subsequent remission (level-2).

To mimic real world care, STAR*D used an equipoise randomization scheme for treatment assignment,⁶ where patient choice could influence treatment selection. In STAR*D, patients had clear preferences for treatment strategies, and patients who received augmentation differed from patients who switched antidepressants.^{7, 8} To date, the augmentation and switch arms of the trial have not been directly compared, leaving clinicians uncertain about next steps for patients who fail initial antidepressant treatment.⁷ A fundamental question remains: when a patient has failed a trial of medication at an adequate dose and for a sufficient period of time, should the physician augment the first medication or switch to a new drug?

To address this issue, we used a non-experimental design to re-analyze the existing STAR*D trial data, applying propensity scoring to minimize imbalances in patient, disease, and treatment characteristics that precluded an augmentation-versus-switch comparison in the original study. We used the public access STAR*D data set to address three questions:

1. Does the likelihood of or time to remission differ as a function of treatment strategy?

2. Does the likelihood of or time to response differ as a function of treatment strategy?

3. Does quality of life differ as a function of treatment strategy?

Among patients who tolerate an initial medication trial but only partially respond, clinical practice suggests that augmentation is favored over switching treatments.⁹ Therefore, we hypothesized that augmentation would remain the preferred strategy compared with switching after adjustment.

MATERIALS AND METHODS

Study Overview

The original STAR*D trial methods, design and rationale have been detailed previously.^{4, 5} Briefly, STAR*D was designed to assess which treatments would be most effective for outpatients with nonpsychotic MDD who had unsatisfactory responses to initial treatments. Its inclusive selection criteria enrolled participants with multiple comorbid medical and psychiatric illnesses. All participants were established clinic patients and many (40%) were from primary care settings. The distribution of depressive severity was similar to nationally-representative samples, and the racial and ethnic composition of enrolled patients approximated the US census.

Eligible participants who consented to STAR*D were initially given citalopram. Participants were required to continue citalopram for 12 weeks unless (1) intolerable side effects required medication change, (2) optimal dose increase was not possible due to side effects or participants’ choice, or (3) significant symptoms were present after 9 weeks at maximally tolerated dose. Participants meeting these criteria were eligible to enter a series of non-placebo controlled randomized clinical trials. This analysis will address the next-step treatments, known as level-2 treatments because they followed failure of the first step.

Study Population

STAR*D enrolled 4,041 participants 18–75 years of age with nonpsychotic depression. Of 4,041 entering initial treatment, 1,439 participants did not achieve remission and entered level-2. Excluding those who received cognitive behavioral therapy as a second-step treatment (n=147), we focused on participants who received treatment within the level-2 medication arms (n=1,292). The efficacy of alternative augment strategies tested (citalopram plus bupropion SR, n=279; citalopram plus buspirone, n=286) did not differ, nor did those of alternative switch strategies (Bupropion, n=239; sertraline, n=238; and venlafaxine XR, n=250).⁷ Accordingly, we combined augment and switch strategies, respectively, into a switch (n=727) and an augment (n=565) group.

Measures

Within each treatment level, visits were conducted at baseline and weeks 2, 4, 6, 9, and 12, with an additional visit at week 14 if needed. At baseline, research coordinators collected demographic information, self-reported psychiatric history, and current general medical conditions (using the Cumulative Illness Rating Scale^{10, 11} [CIRS]). Additionally, they administered the Hamilton Rating Scale for Depression (HRSD₁₇)^{12, 13} and patients completed a self-report version of the 16-item Quick Inventory of Depressive Symptomatology (QIDS-SR₁₆).^{14, 15} At each follow-up visit, research coordinators collected QIDS-SR₁₆ and side effects ratings, (using the Frequency, Intensity, and Burden of Side Effects Rating [FIBSER], with higher ratings indicating greater severity).¹⁶ At the exit from each treatment level, measures including the HRSD₁₇ were collected by outcome assessors, masked to treatment group. Interactive Voice Response (IVR) assessments of clinical symptoms, side effects, and self-reported quality of life were conducted quarterly.

Outcomes

Remission was defined as a score of ≤5 upon exiting level-2 on the QIDS-SR₁₆ (scores range from 0–27, higher scores represent greater symptom severity). The secondary remission outcome was defined as a score of ≤7 on the HRSD₁₇ (scores range from 0–52, with higher scores representing greater symptom severity). Per the original STAR*D protocol,¹⁶ participants for whom the level-2 exit HRSD₁₇ score was missing were designated as not achieving remission. The QIDS-SR₁₆ and HRSD₁₇ scores correlate closely,¹⁵ and can be used to produce equivalent scores. However, the QIDS-SR₁₆ was selected as the primary remission outcome as it was more likely than the HRSD₁₇ to be collected at a time close to a patient’s exit from a level.

Response was defined as ≥50% decrease in QIDS-SR₁₆ score from level-2 entry to level-2 exit.

Quality of Life was measured using the physical health and mental health subscales of the Short-Form Health Survey (SF-12) and the Quality of life Enjoyment and Satisfaction Questionnaire (Q-LES-Q). The two SF-12 subscales range from 0–100, with higher scores indicating better functioning. The population norm for each is 50±10.¹⁷ The 16-item Q-LES-Q measures the degree of enjoyment and satisfaction experienced by subjects in daily functioning.^{18, 19} The total score is based on responses to the first 14 items on the scale (each scored on a 5-point Likert scale with 1=very poor, 5=very good). Similar to the SF-12, higher scores represent greater life enjoyment and satisfaction.

Statistical Analysis

We examined missing data patterns for variables in the propensity score and outcomes models. The level of missing information in the analysis dataset was only 5%, but 82% of observations had at least one missing value. To replace the missing values, we generated 20 imputed datasets using the Markov chain Monte Carlo method in SAS PROC MI.^{20, 21}

Propensity scores were developed using multiple logistic regression to estimate each participant’s conditional probability of receiving medication augmentation. Forty-seven covariates were selected for inclusion in the models based on their relationship with the treatment and the outcome (or the outcome alone). Thirty interaction terms were selected using a forward stepwise selection process that minimized covariate imbalance. We individually matched patients from the switch group to patients in the augment group based on their estimated propensity scores, resulting in the selection of 269 augment participants, matched with an equal number of switch participants.

Based on the propensity-score-matched samples, we used binomial regression models to measure the association between medication treatment arm (augment vs. switch) and remission and response rates. We generated incidence proportions and risk ratios for remission and response using PROC GENMOD.²²

Time to first remission and response were analyzed using survival analysis. Kaplan-Meier curves were generated and a log-rank statistic was used to compare the curves. Also, Cox proportional hazards models were used to estimate hazard ratios. For the survival analysis, we used a singly-imputed dataset based on the expectation-maximization method.^{20, 21}

We used linear regression to estimate the differences between the switch and augment groups in mean SF-12 physical health score, SF-12 mental health score, and Q-LES-Q total score.

RESULTS

Prior to propensity scoring, the augmentation group differed from the switch group in several ways (Table 1). Augmenters had fewer lifetime depressive episodes (5.6 vs. 6.8), shorter current depressive episode duration (27.8 vs. 30.3 months), a greater proportion with drug (8.7% vs. 5.5%) or alcohol abuse (13.0% vs. 11.9%), and fewer with posttraumatic stress disorder (18.9% vs. 23.0%) compared to switchers. Upon exiting level-1 treatment, augmenters had lower depressive severity by QIDS-SR₁₆ (11.4 vs. 13.2) and slightly higher quality of life by SF-12 and Q-LES-Q measures. Furthermore, augmenters were less likely to exit level-1 due to side effects (10.3% vs. 62.8%), spent more time in level-1 treatment (11.9 weeks vs. 8.0 weeks), and had higher citalopram daily doses at level-1 exit (55.1 vs. 41.4 mg) than switchers.

Table 1:

Participant characteristics before and after propensity score matching

	Pre-Propensity Score		After Propensity Score Matching		Unmatched Patients
	Augment N = 565	Switch N = 727	Augment N = 269	Switch N = 269	Augment N = 296	Switch N = 458
Demographic Characteristics
Age - years	41.6 ± 12.7	42.4 ± 12.8	41.7 ± 12.8	41.8 ± 12.9	41.5 ± 12.6	42.7 ± 12.7
Female sex - %	58.8	58.7	59.0	58.4	58.5	58.9
Race - %
White	78.0	75.8	77.4	77.0	78.6	75.1
Black	18.0	18.7	18.4	18.9	17.7	18.6
Education - %
More than High School	22.3	23.1	20.4	21.2	24.1	24.2
Employment status - %
Employed	64.0	63.1	65.1	64.7	63.0	62.2
Marital status - %
Married or cohabitating	43.8	43.0	44.7	44.4	43.0	42.2
Divorced, separated or widowed	28.2	30.7	28.6	28.7	27.8	31.8
Psychiatric History (baseline)
No. of major depressive episodes	5.6 ± 8.8	6.8 ± 11.1	6.5 ± 10.4	6.3 ± 10.1	4.8 ± 7.0	7.1 ± 11.6
Duration of index depressive episode
Mean - mo	27.8 ± 56.1	30.3 ± 66.7	25.8 ± 52.0	25.9 ± 58.1	29.6 ± 59.6	33.0 ± 71.1
≥ 2 Years - %	27.6	27.6	26.2	25.7	28.9	28.7
Psychiatric Comorbidities - %
Drug Abuse	8.7	5.5	7.0	7.1	10.4	4.6
Alcohol Abuse	13.0	11.9	11.7	11.9	14.2	12
Posttraumatic Stress Disorder	18.9	23.0	21.5	20.6	16.5	24.4
Generalized Anxiety Disorder	23.4	22.2	23.0	22.8	23.8	21.9
Level-1 Exit Characteristics
Depressive Severity:
QIDS-SR16 score	11.4 ± 4.9	13.2 ± 4.9	12.5 ± 5.1	12.6 ± 4.9	10.3 ± 4.5	13.6 ± 4.9
HRSD17 score	16.0 ± 7.1	19.0 ± 7.3	17.6 ± 7.2	17.6 ± 7.3	14.6 ± 6.7	19.9 ± 7.2
SF-12
Mental	32.4 ± 9.2	29.7 ± 9.6	31.0 ± 8.9	30.8 ± 9.4	33.7 ± 9.2	29.1 ± 9.6
Physical	46.9 ± 11.8	45.6 ± 12.0	46.1 ±12.1	46.2 ± 11.8	47.7 ± 11.4	45.3 ± 12.2
QLESQ score	45.7 ± 16.3	41.1 ± 16.9	43.0 ± 16.5	43.1 ± 16.5	48.1 ± 15.7	39.9 ± 17.0
Side Effect Measures
FIBSER score	1.2 ± 1.3	2.8 ± 1.9	1.4 ± 1.4	1.4 ± 1.5	1.0 ± 1.2	3.7 ± 1.6
Exited level-1 due to side effects - %	10.3	62.8	19.9	20.2	1.6	87.8
Experienced a serious adverse event - %	4.6	2.6	3.5	3.5	5.6	2.1
Management Features
Duration of level-1 treatment - weeks	11.9 ± 2.9	8.0 ± 4.2	10.9 ± 3.2	10.9 ± 3.0	12.8 ± 2.4	6.3 ± 3.8
Citalopram dose - mg/day	55.1 ± 10.9	41.4 ± 17.7	52.8 ± 12.8	52.9 ± 12.4	57.1 ± 8.4	34.7 ± 16.8
Medication Adherence - %
Never / Rarely Missing	89.6	82.6	88.9	87.4	90.1	79.7

Plus–minus values are means ± SD.

QIDS-SR16, the 16-item Quick Inventory of Depressive Symptomatology, Self-Rated (scores can range from 0 to 27; higher scores indicate increased severity of depressive symptoms); HRSD17 the 17-item Hamilton Rating Scale for Depression (scores can range from 0 to 52; higher scores indicate increased severity of depressive symptoms); SF-12 Short Form Health Survey, two SF-12 subscales (Mental and Physical) range from 0 to 100; higher scores indicate increased perceived functioning. The population norm for each is 50 ± 10; QLESQ Quality of Life Enjoyment and Satisfaction Questionnaire (scores can range from 0 to 100; higher scores indicate greater satisfaction); FIBSER (Frequency Intensity and Burden of Side Effect Rating), is a global measure of the intensity and the burden of side effects. Scores were averaged to calculate a composite score. PRISE, Patient-Rated Inventory of Side Effects assesses the presence of specific side effects in nine biologic systems.

Level-1 refers to initial treatment with citalopram. Level-2 refers to second-step treatment with either medication augmentation or switching.

The propensity score matching corrected for these important differences between augmentation and switch groups in level-2, resulting in a smaller, but well-balanced sample. The matched sample included an average of 269 patients in each group (augment and switch) across the 20 imputations (Table 1). This sample had a mean age of approximately 42 years and was primarily female (60%), Caucasian (approximately 77%), and employed (approximately 65%). Overall, sociodemographic characteristics were similar before and after propensity score matching.

Patients excluded from the matched sample had distinct characteristics that differed as function of whether they received augmentation or switched treatments (Table 1). Those excluded who received augmentation tended to be less severely depressed, remained in level-1 treatment longer, received a higher dose of citalopram during level-1 treatment, and were less likely to exit level-1 because of side effects than were excluded patients who received switch treatment.

Despite an unadjusted risk ratio in the full dataset favoring augmentation (RR:1.47, 95%CI:1.23–1.75), after matching the likelihood of remission (QIDS-SR₁₆ ≤5) did not differ between groups (RR:1.14, 95%CI:0.82–1.58) (Table 2). Similarly, the likelihood of remission using the HRSD₁₇ did not differ. Furthermore, the time to remission was similar between augmentation and switch strategies (Hazard Ratio:1.01, 95%CI:0.76–1.34, p = 0.946).

Table 2:

Comparison of Remission, Response and Quality of Life by Treatment Strategy: Full Study Sample and Propensity Score Matched Sample

Unadjusted Analysis - Full Study Sample
	Augment (N = 565)		Switch (N = 727)		Augment vs. Switch (N = 1,292)
	IP	(95% CI)	IP	(95% CI)	RR	(95% CI)
Measures of Remission
QIDS-SR16	0.33	(0.29, 0.37)	0.23	(0.20, 0.26)	1.47	(1.23, 1.75)
HRSD17	0.28	(0.25, 0.32)	0.20	(0.17, 0.23)	1.40	(1.15, 1.70)
Measures of Response
QIDS-SR16	0.27	(0.23, 0.30)	0.24	(0.20, 0.27)	1.14	(0.94, 1.37)
Measures of Quality of Life	Mean	(95% CI)	Mean	(95% CI)	Mean Difference	95% CI
Q-LES-Q	53.3	(50.9, 55.6)	50.9	(48.9, 53.0)	2.3	(−0.8, 5.5)
SF-12
Mental Health Status	40.2	(38.7, 41.7)	38.0	(36.8, 39.3)	2.2	(0.3, 4.1)
Physical Health Status	45.9	(44.5, 47.3)	45.1	(43.9, 46.3)	0.8	(−1.1, 2.7)
Adjusted Analysis - Propensity Score Matched Sample
	Augment (N = 269)		Switch (N = 269)		Augment vs Switch (N = 538)
	IP	(95% CI)	IP	(95% CI)	RR	(95% CI)
Measures of Remission
QIDS-SR16	0.28	(0.21, 0.34)	0.24	(0.19, 0.30)	1.14	(0.82, 1.58)
HRSD17	0.24	(0.18, 0.29)	0.22	(0.16, 0.28)	1.07	(0.76, 1.50)
Measures of Response
QIDS-SR16	0.27	(0.21, 0.33)	0.23	(0.18, 0.29)	1.14	(0.82, 1.58)
Measures of Quality of Life	Mean	(95% CI)	Mean	(95% CI)	Mean Difference	95% CI
Q-LES-Q	49.7	(45.8, 53.6)	50.8	(47.0, 54.6)	−1.1	(−6.6, 4.4)
SF-12
Mental Health Status	38.0	(35.3, 40.6)	37.6	(35.3, 39.9)	0.4	(−3.0, 3.7)
Physical Health Status	45.1	(42.6, 47.6)	45.5	(43.3, 47.8)	−0.4	(−3.8, 2.9)

Remission on the QIDS-SR16 is defined as a level-2 exit score of ≤ 5. Remission on the HRSD17 is defined as a level-2 exit score ≤ 7. Response was defined as ≥ 50% decrease in QIDS-SR16 score from level-2 entry to level-2 exit. QIDS-SR16, the 16-item Quick Inventory of Depressive Symptomatology, Self-Rated (scores can range from 0 to 27; higher scores indicate increased severity of depressive symptoms); HRSD17 the 17-item Hamilton Rating Scale for Depression (scores can range from 0 to 52; higher scores indicate increased severity of depressive symptoms); QLESQ Quality of Life Enjoyment and Satisfaction Questionnaire (scores can range from 0 to 100; higher scores indicate greater satisfaction); SF-12 Short Form Health Survey, two SF-12 subscales (Mental and Physical) range from 0 to 100; higher scores indicate increased perceived functioning. The population norm for each is 50 ± 10. IP= Incidence Proportion; RR=Risk Ratio

Similar to the unadjusted estimates of response by QIDS-SR₁₆, in the matched analysis, treatment strategy did not significantly affect the likelihood of response (RR:1.14, 95%CI:0.82–1.58). Additionally, treatment strategy did not affect the time to response (Hazard Ratio:0.85, 95%CI:0.64–1.12, p = 0.243).

Finally, treatment strategy did not affect quality of life, as measured by either the Q-LES-Q or the SF-12 Mental Health and Physical Health subscales.

In STAR*D, the mean doses of switch medications were lower than the adequate doses of the medications defined in the study protocol.²³ Therefore, we assessed whether dosing affected our results. We created variables to indicate each patient’s dose(s) based on standard dosing ranges²⁴ and assessed dose effects. In a sensitivity analysis, we adjusted for dosing between groups; none of the dose variables had a meaningful effect on any of the outcomes.

Post-Hoc Analyses

Although outcomes did not differ in the matched sample overall, the treatment effect may have been heterogeneous. Propensity score stratification facilitates the assessment of treatment effect heterogeneity through comparisons across strata. Stratification controls for pre-treatment differences by matching several groups on the propensity score rather than creating 1:1 matches between individuals. Using the full sample, we assessed whether the likelihood of remission differed by the propensity to receive augmentation by stratifying into quintiles, from low (most of this group switched) to high (most of this group augmented) propensity to receive augmentation.

Those with the highest propensity for augmentation were 1.65 times as likely to remit with augmentation as with switching (95%CI:0.86–3.16), suggesting a possible benefit of augmentation over switching for this subgroup of patients. Those with the lowest propensity for augmentation were also more likely to remit with augmentation (RR=2.42), but that group contained few augmenting participants (n=4) and the estimate was imprecise (95%CI:0.73–8.05).

To explore the reasons for this heterogeneity and inform clinicians’ treatment decisions, we identified two clinically relevant variables that strongly influenced the propensity to receive augmentation: depressive severity and number of weeks in first step treatment. We stratified on each variable to assess its effect on the likelihood of remission with augmentation vs. switch. Comparisons made within strata were based on the full study sample and were unadjusted for other covariates.

Time on initial therapy consistently increased the likelihood of remission with augmentation; however, the effect was strongest for patients who received 12 weeks of initial therapy. For these patients, augmentation was nearly twice as likely to produce remission as was switch (12–13 weeks, augment n=164, switch n=89, unadjusted RR:1.9, 95%CI:1.16–3.11). Depressive severity following initial therapy also demonstrated a consistent trend with increased likelihood of remission with augmentation (as compared with switching) for those with lower depressive severity. This effect was strongest for patients with residual depressive severity scores less than 9 (augment n=175, switch n=119, Unadjusted RR:1.32, 95%CI:1.03–1.70).

DISCUSSION

For patients who received an aggressive initial antidepressant trial and tolerated it well, there is no clear preference for either switching or augmenting as a next-step treatment. Our analyses provide the first direct comparison of the benefit of augmentation vs. switch strategies for the large proportion of patients whose depression does not remit with the initial treatment attempt. Using propensity score matching, we were able to minimize imbalances in patient, disease, and treatment characteristics to test directly whether augmentation produced better clinical outcomes than switching following failure of an initial antidepressant treatment for MDD. While the unadjusted results suggested greater benefit from augmentation, the propensity score matched comparison, which accounted for measured differences between the groups, showed no greater clinical benefit from choosing augmentation over switching. Contrary to our predictions of better outcomes for augmentation, we found that the likelihood of remission or response, time to remission or response, and quality of life measures did not differ by treatment strategy.

Importantly, the matched sample does not reflect the full STAR*D trial population, but reflects patients who were moderately depressed, able to tolerate an aggressive (high dose) level-1 treatment for approximately 11 weeks, and tended not to discontinue initial treatment due to side effects. Patients who were excluded from the matched sample had distinct characteristics that differed as a function of whether they received augmentation or switch treatments. Specifically, within the unmatched sample, the switch group (compared to the augmentation group) had larger proportions of patients intolerant of citalopram and patients with greater depressive severity. Accordingly, our matched sample results apply to patients who received and were tolerant of a well-dosed acute phase depression trial and who were assigned to either augmenting or switching in sufficient numbers. For such a group, the selection of a switching vs. augmenting strategy did not change the likelihood of a better clinical outcome. However, based on post-hoc analyses of the full STAR*D trial population, augmentation produced better outcomes for patients tolerating 12 or more weeks of level-1 treatment and who had partially responded to initial treatment (mild depressive severity, QIDS-SR₁₆ < 9). For patients fitting this profile, augmentation may be preferred to switch.

There were limitations to our analyses. While propensity score matching minimized the selection bias introduced by the original STAR*D trial design, it did so at the cost of statistical power and generalizability. First, the group overlap was not ideal, leading to a smaller sample size with our matched analysis. As a result, our matched sample represents only a subset of the full STAR*D trial population. Second, it is difficult to make conclusions about differential treatment effects from the propensity-score-stratified analysis because of limited sample size. We also stratified on two clinically relevant variables that strongly influenced the propensity score; conclusions from those analyses are not as limited by sample size but are subject to residual confounding within strata. Third, remission and response were uncommon in our propensity-score-matched sample, resulting in limited power to detect a treatment effect. Fourth, our propensity score was restricted to those variables that were measured during the original trial. While the STAR*D study collected many variables indicative of the likelihood of depression remission, unmeasured confounding still may exist. Finally, the analysis included only a limited number of augmentation and switch options which, at the time the study was designed, represented treatment options with the strongest scientific support. The use of other augmentation strategies, such as atypical antipsychotic medications,²⁵ different second step antidepressants,^{26, 27} or different medication combinations²⁸ might produce different results.

For many patients not tolerating an initial antidepressant treatment, the choice is clear—stop the offending medication, and switch to a different antidepressant. However, the choice is less clear for those who tolerate but do not fully recover following the initial treatment. For these patients, clinicians need evidence-based guidance about next-step selection. Our propensity matched sample represents that group. Our analyses indicate that for these patients, augmentation vs. switch strategies yield comparable outcomes, whether measured by remission, response, or quality of life. However, our results also provide some tentative evidence that for patients who tolerate initial medication trials for 12 weeks and have depressive severity scores indicating a partial response, clinicians may consider recommending augmentation over switch.