Portal de Periódicos da CAPES

Increased statistical power with combined independent randomization tests used with multiple-baseline design

2013; Elsevier BV; Volume: 66; Issue: 6 Linguagem: Inglês

10.1016/j.jclinepi.2012.11.006

1878-5921

Pascal N. Tyrrell, Paul Corey, Brian M. Feldman, Earl D. Silverman,

Advanced Causal Inference Techniques

Physicians often assess the effectiveness of treatments on a small number of patients. Multiple-baseline designs (MBDs), based on the Wampold-Worsham (WW) method of randomization and applied to four subjects, have relatively low power. Our objective was to propose another approach with greater power that does not suffer from the time requirements of the WW method applied to a greater number of subjects.The power of a design that involves the combination of two four-subject MBDs was estimated using computer simulation and compared with the four- and eight-subject designs. The effect of a delayed linear response to treatment on the power of the test was also investigated.Power was found to be adequate (>80%) for a standardized mean difference (SMD) greater than 0.8. The effect size associated with 80% power from combined tests was smaller than that of the single four-subject MBD (SMD=1.3) and comparable with the eight-subject MBD (SMD=0.6). A delayed linear response to the treatment resulted in important reductions in power (20-35%).By combining two four-subject MBD tests, an investigator can detect better effect sizes (SMD=0.8) and be able to complete a comparatively timelier and feasible study.