OBJECTIVES: To better evaluate the effectiveness of antidepressant drugs in the treatment of major depression in clinical practice. METHODS: A simulation experiment was used to illustrate an application of marginal structural models (MSMs) via inverse probability of treatment weighting (IPTW) approach in the context of non-randomized data on N = 1,000 depressed subjects, initially subjected to "watchful waiting". In simulation we assumed that subjects with worse intermediate outcome have a higher probability of being subsequently assigned to antidepressant treatment while those who receive antidepressant treatment are more likely to reach remission and less likely to reach relapse state. The outcomes from multiple (500) simulated data sets are analyzed using simple unadjusted analysis based on logistic regression and using MSM. RESULTS: In contrast to unadjusted analysis, but consistent with the treatment assumptions, using MSM via IPTW results in strong evidence of the effectiveness of the antidepressant treatment. Furthermore MSM via IPTW substantially reduces the probability of wrongly rejecting the null hypothesis. However, the instability of weights due to the sparse data and incorrectly specified MSM may still result in inflation of Type I error rates. CONCLUSIONS: MSMs may allow evaluating the causal effects associated with antidepressant treatment from the data observed in clinical practice.
OBJECTIVES: To better evaluate the effectiveness of antidepressant drugs in the treatment of major depression in clinical practice. METHODS: A simulation experiment was used to illustrate an application of marginal structural models (MSMs) via inverse probability of treatment weighting (IPTW) approach in the context of non-randomized data on N = 1,000 depressed subjects, initially subjected to "watchful waiting". In simulation we assumed that subjects with worse intermediate outcome have a higher probability of being subsequently assigned to antidepressant treatment while those who receive antidepressant treatment are more likely to reach remission and less likely to reach relapse state. The outcomes from multiple (500) simulated data sets are analyzed using simple unadjusted analysis based on logistic regression and using MSM. RESULTS: In contrast to unadjusted analysis, but consistent with the treatment assumptions, using MSM via IPTW results in strong evidence of the effectiveness of the antidepressant treatment. Furthermore MSM via IPTW substantially reduces the probability of wrongly rejecting the null hypothesis. However, the instability of weights due to the sparse data and incorrectly specified MSM may still result in inflation of Type I error rates. CONCLUSIONS: MSMs may allow evaluating the causal effects associated with antidepressant treatment from the data observed in clinical practice.
Authors: Emanuel Severus; Florian Seemüller; Michael Berger; Sandra Dittmann; Michael Obermeier; Andrea Pfennig; Michael Riedel; Sophia Frangou; Hans-Jürgen Möller; Michael Bauer Journal: BMC Med Date: 2012-07-02 Impact factor: 8.775