Predicting antidepressant response by monitoring early improvement of individual symptoms of depression: individual patient data meta-analysis

Abstract

Background: Improvement in depression within the first 2 weeks of antidepressant treatment predicts good outcomes, but non-improvers can still respond or remit, whereas improvers often do not.AimsWe aimed to investigate whether early improvement of individual depressive symptoms better predicts response or remission.

Method: We obtained individual patient data of 30 trials comprising 2184 placebo-treated and 6058 antidepressant-treated participants. Primary outcome was week 6 response; secondary outcomes were week 6 remission and week 12 response and remission. We compared models that only included improvement in total score by week 2 (total improvement model) with models that also included improvement in individual symptoms.

Results: For week 6 response, the area under the receiver operating characteristic curve and negative and positive predictive values of the total improvement model were 0.73, 0.67 and 0.74 compared with 0.77, 0.70 and 0.71 for the item improvement model. Model performance decreased for week 12 outcomes. Of predicted non-responders, 29% actually did respond by week 6 and 43% by week 12, which was decreased from the baseline (overall) probabilities of 51% by week 6 and 69% by week 12. In post hoc analyses with continuous rather than dichotomous early improvement, including individual items did not enhance model performance.

Conclusions: Examining individual symptoms adds little to the predictive ability of early improvement. Additionally, early non-improvement does not rule out response or remission, particularly after 12 rather than 6 weeks. Therefore, our findings suggest that routinely adapting pharmacological treatment because of limited early improvement would often be premature.Declaration of interestNone.

Keywords: Antidepressants; depression; early improvement; individual patient data meta-analysis; treatment response.

Fig. 1

Actual probability of response at week 6 according to participants' predicted outcome (non-response versus response) for each model. Results are based on the test data-set. The dashed line indicates the baseline probability of response. The models predicted non-response for 42% (baseline), 38% (total improvement), 46% (item improvement) and 47% (item interactions) of participants.

Fig. 2

Proportion of participants who responded or remitted according to the percentage improvement from baseline. Error bars (dashed lines) indicate the 95% CIs.