Constancy of relative indices of treatment effect across different definitions of response derived from continuous scales: secondary analyses of individual-patient trial database of second-generation antipsychotics

Article type
Authors
Furukawa T1, Wagenpfeil S2, Leucht S3
1Department of Psychiatry and Cognitive-Behavioral Medicine, Nagoya City University Graduate School of Medical Sciences, Nagoya, Aichi, Japan
2Institute of Medical Statistics and Epidemiology, Technische Universität München, Munich, Germany
3Department of Psychiatry and Psychotherapy, Technische Universität München, Munich, Germany
Abstract
Objective: In randomized controlled trials and their meta-analyses, continuous outcomes are often dichotomized to define response, using various cutoffs. We set out to find a summary index of treatment effectiveness which remains constant across different definitions of such scale-derived response. Design: Secondary analyses of individual patient data from 10 randomized controlled trials examining second generation antipsychotics for schizophrenia (n = 4278). Methods: We ran meta-analyses to produce odds ratios (OR), risk ratios (RR) and risk differences (RD) and their 95% confidence intervals (CI) for different definitions of response, using cutoffs of 10% through 90% reduction on the symptom severity rating scales. Constancy of these indices was first examined through visual inspection, by way of I-squared statistics to quantify heterogeneity, and by way of coefficients of variation. If any of these indices were found to remain reasonably constant, we next examined the concordance between the number needed to treat (NNT) predicted from them and the observed NNT. Results: The visual inspection, I-squared statistics and coefficients of variation all suggested that OR and RR remained reasonably constant across various definitions of response, especially for those using thresholds of 10% through 70% reduction in the symptom severity. The NNTs predicted from OR and RR agreed well with the observed NNTs, with ANOVA intraclass correlation coefficients of 0.96 (95% CI: 0.92 to 0.98) and 0.86 (0.72 to 0.93), respectively. Conclusion: The relative measures of treatment effectiveness may remain reasonably constant across different scale-derived response definitions and, in conjunction with varying control event rates, can give accurate estimates of NNTs. This will allow individually matched estimates of NNT, depending on the level of response that he/she wishes to achieve.