Choice of time-scale in Cox's model analysis of epidemiologic cohort data: a simulation study

摘要: SUMMARY Cox's regression model is widely used for assessing associations between potential risk factors and disease occurrence in epidemiologic cohort studies. Although age often a strong determinant of risk, authors have frequently time-on-study instead as the time-scale, clinical trials. Unless baseline hazard an exponential function age, this approach can yield dierent estimates relative hazards than using even when adjusted for. We performed simulation study order to investigate existence magnitude bias degrees association covariate interest. Age onset was generated from exponential, Weibull or piecewise distributions, bothxed time-dependent dichotomous covariates were considered. observed no upon time-scale. Upon time- on-study, we veried absence exponentially distributed onset. For non- found that could occur interest independent age. It be severe case substantial with especially dependent covariates. Thesendings illustrated on data 84 329 French women followed prospectively breast cancer occurrence. In view our results, strongly recommend not time-scale analysing data. Copyright ? 2004 John Wiley & Sons, Ltd.