Cumulative remission rate after sequential treatments in depression: reappraisal of the STAR*D trial data

The Sequenced Treatment Alternatives to Relieve Depression (STAR*D) was a seminal clinical trial of 4,041 outpatients with major depressive disorder (MDD), examining the effectiveness of sequential treatment steps ¹ . One of the key findings of the study was that the theoretical cumulative remission rate up to four treatment steps was 67% among those who initiated antidepressant treatment ¹ .

Although this finding has had a significant impact on clinical research and policy‐making ² , this estimated rate is subject to two significant limitations. First, the estimate assumes that there were no dropouts during the study (and those who exited the study would have had the same remission rates as those who stayed in the protocol) ¹ . In fact, only 995 subjects (24.6%) had complete data; 2,487 (61.5%) had dropout missing data patterns; and 559 (13.8%) had non‐dropout intermittent missingness even within 12 weeks in Step 1 treatment ³ . Second, our recent re‐analysis of the individual patient‐level data from the trial revealed that 1,108 subjects (27.4%) had taken at least one antidepressant medication during the index episode prior to study entry. Given that those who required more treatment steps were less likely to achieve subsequent remission ¹ , the cumulative remission rates likely differ between the drug‐naïve and previously treated subjects.

We therefore estimated the cumulative remission rate of the STAR*D trial by utilizing the inverse probability of censoring weighted (IPCW) Kaplan‐Meier method. Furthermore, we investigated the cumulative remission rates among individuals with and without prior antidepressant treatment history during the ongoing episode.

We extracted sociodemographic factors at baseline, and the scores on the 16‐item Quick Inventory of Depressive Symptomatology, self‐report (QIDS‐SR₁₆) and the Global Rating of Side Effect Burden (GRSEB) at weeks 0, 2, 4, 6, 9, 12 and 14 in Step 1‐4 treatments from the STAR*D dataset. Remission was defined as a score ≤5 on the QIDS‐SR₁₆ at any time during the treatments. The IPCW method can incorporate possible influential factors for dropouts in the estimation of survival curves ⁴ , ⁵ . We utilized the stabilized weights estimated through a Cox regression ⁴ , ⁶ that included both time‐dependent covariates (the QIDS‐SR₁₆ and the GRSEB scores at successive measurements) and time‐independent covariates (age, sex, education history, the Hamilton Depression Rating Scale total score at baseline, family history, and history of taking any antidepressant medications). Missing data were addressed using the multiple imputations by chained equations with 100 imputed datasets ⁷ .

We calculated the cumulative remission rates at 90, 180 and 360 days along with corresponding 95% confidence intervals (CIs). We applied the same methods separately to those who had received at least one antidepressant medication during the index episode prior to study entry and those who had not, and compared the cumulative remission rates between the two groups using the weighted log‐rank test.

The cumulative remission rates among all samples were estimated to be 53.8% (95% CI: 51.6‐55.9) at 90 days, 74.5% (95% CI: 72.1‐76.9) at 180 days, and 87.5% (95% CI: 82.4‐92.6) at 360 days. The median time to remission was 84 days (see also supplementary information). The estimated rates of cumulative remission among those who had received no antidepressant prior to study entry were 55.4% (95% CI: 53.0‐57.9) at 90 days, 76.3% (95% CI: 73.7‐78.9) at 180 days, and 89.1% (95% CI: 85.0‐93.2) at 360 days. These rates were higher (hazard ratio: 1.28; 95% CI: 1.16‐1.41, p<0.001) than among those who had had antidepressant exposure: 49.3% (95% CI: 45.5‐53.2) by 90 days, 70.1% (95% CI: 65.6‐74.6) by 180 days, and 82.1% (95% CI: 71.8‐92.3) by 360 days. The median time to remission in these two groups was 80 days and 91 days, respectively (see also supplementary information).

So, our re‐analysis of the STAR*D data shows a cumulative remission rate approximately 20% points higher than that reported in the original paper ¹ . That paper did not account for those who dropped out from or discontinued the study, while we applied a survival analysis taking into account time‐independent and time‐dependent patient characteristics such as longitudinal symptoms and side effects.

Previous research has typically assumed that dropouts had the same outcome as non‐dropouts (completer analysis) or would not have achieved remission in the intent‐to‐treat analysis (worst case scenario analysis). However, in a 9‐week single‐blind clinical trial of sertraline and mirtazapine in MDD, 147 participants who had dropped out from the intervention but were subsequently assessed had lower depressive scores and better treatment outcomes compared to 1,499 participants who completed the intervention and the assessment in the study ⁸ . Among those who dropped out, 32 participants who were difficult to contact had even lower depressive scores than 82 participants who were easily contactable ⁸ . These findings suggest that a greater tendency to drop out may be associated with better treatment outcomes.

The present findings align with prior reports of high rates of cumulative remission in individuals with depression who initiated antidepressant treatment for the first time. Specifically, a pro‐spective follow‐up study consisting of 90 drug‐naïve patients diagnosed with MDD reported that 85% of the subjects achieved asymptomatic or minimally symptomatic status by 12 months ⁹ .

There are several limitations to this study. First, the STAR*D trial included only outpatients with non‐psychotic MDD who received citalopram in primary or secondary care settings in the US, limiting the generalizability of the findings to other populations in different treatment settings. Second, remission was defined only on the basis of the QIDS‐SR₁₆ scores, without considering functional outcomes. Third, remission in the acute phase treatment does not necessarily mean stable remission, as 40‐71% of individuals who achieved remission were reported to experience relapse within one year ¹ .

In conclusion, our re‐analysis using the IPCW Kaplan‐Meier method shows a much higher cumulative remission rate (i.e., 87.5% within a year after the initiation of treatment) than the 67% reported in the original STAR*D paper. This promising finding provides an opportunity to revisit the therapeutic potential of currently available treatment options for MDD, and underscores the relevance of employing sequential treatments until remission is achieved.