Kaminski Jakob André, Bschor Tom
Charité-Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Department of Psychiatry and Psychotherapy CCM, Charitéplatz 1, 10117 Berlin, Germany; Berlin Institute of Health, Berlin, Germany.
Schlosspark Hospital Berlin and University Hospital/Technical University Dresden, Germany.
J Affect Disord. 2020 Apr 1;266:95-99. doi: 10.1016/j.jad.2020.01.107. Epub 2020 Jan 23.
A recent study by Hengartner and Plöderl describes a strong increase for suicides (odds ratio (OR) of 2.83, 95% CI=1.13-9.67) and suicide attempts (OR=2.38 95%, CI=1.63-3.61) in antidepressant treated patients as compared to placebo. The authors re-analyzed data presented by Khan et al. who found no drug-placebo differences in suicide and suicide attempt rates. Hengartner and Plöderl base their findings on calculating the OR from a 2×2 table of the sum of the events and the totals of the sample sizes across studies. We here argue that pooling data from all drugs may not be the adequate approach.
We applied a meta-analytical approach to account for between-drug variance and conducted several statistical analyses as a sensitivity analysis. We argue that a more suitable approach for finding an overall effect from several observations is a meta-analytical approach namely the Mantel-Haenszel method without continuity correction.
Our analysis leads to different conclusions as opposed to Hengartner and Plöderl. With the recommended method we estimate an OR of 1.98, 95% CI 0.71-5.50 for suicides and 1.63 (95%CI=1.09-2.43) for suicide attempts.
The conducted analysis was restricted to the data available from the previous studies. Possibly, a more extensive search of the literature would lead to different results. However, we showed that re-analysing the re-analysis with several different approaches underlines the necessity of sensitivity analysis.
We could show that, in the case of rare events, the data is very sensitive to different analytical approaches underlining the importance of further investigations.
亨加特纳和普洛德雷尔最近的一项研究表明，与安慰剂相比，接受抗抑郁药治疗的患者自杀率（优势比（OR）为2.83，95%置信区间=1.13 - 9.67）和自杀未遂率（OR = 2.38，95%置信区间=1.63 - 3.61）大幅上升。作者重新分析了汗等人提供的数据，后者发现自杀率和自杀未遂率在药物组与安慰剂组之间并无差异。亨加特纳和普洛德雷尔的研究结果是基于对各研究中事件总和与样本量总数的2×2表格计算优势比得出的。我们认为将所有药物的数据合并可能并非合适的方法。
我们采用荟萃分析方法来考虑药物间差异，并进行了多项统计分析作为敏感性分析。我们认为，从多个观察结果中找到总体效应的更合适方法是荟萃分析方法，即不进行连续性校正的曼特尔 - 亨塞尔方法。
与亨加特纳和普洛德雷尔的研究相反，我们的分析得出了不同的结论。采用推荐的方法，我们估计自杀的OR为1.98，95%置信区间为0.71 - 5.50，自杀未遂的OR为1.63（95%置信区间=1.09 - 2.43）。
所进行的分析仅限于先前研究中可用的数据。可能更广泛地搜索文献会得出不同的结果。然而，我们表明用几种不同方法对重新分析进行再分析突出了敏感性分析的必要性。
我们可以表明，在罕见事件的情况下，数据对不同分析方法非常敏感，这突出了进一步调查的重要性。
