76 – Contributed Poster Presentations: Section on Statistics in Epidemiology
Relaxing the Interaction Test in the Cox Proportional Hazard Model for Epidemiological Studies
Eiji Nakatani
Shizuoka General Hospital
The interaction test in epidemiological studies is used to determine the subclass with a strong therapeutic effect, but one of its practical problems is the less power and the larger sample size required to confirm treatment efficacy, compared with that of a randomized trial. From this perspective, we proposed a test that had enough power to find the subclasses that had significant effects in the Cox proportional hazards model. The estimator was the difference in hazard ratios for a particular factor between the subclass and the whole. The estimator and confidence interval were estimated using the bootstrap method. For the discovery of subclass, the proposed test can be used by epidemiological researchers as an alternative to the traditional interaction test and may enable easy interpretation and powerful subclass determination. We performed simulations that applied our proposed method for the interaction test.