464Basics of survival analysis: age is not appropriate as time scale in Cox regression model. (2nd September 2021)

• Record Type: Journal Article
• Authors: Liu, Yangyang; Zhang, Jingjing; Mitsuhashi, Toshiharu; Matsuo, Toshihiko; Yorifuji, Takashi; Hamada, Jun
• Is Part Of: International journal of epidemiology. Volume 50(2021)Supplement 1
• Publication Date: 2021-09-02
• Subjects: Epidemiology -- Periodicals; 614.4
• DOI: 10.1093/ije/dyab168.397
• Languages: English
• ISSNs: 0300-5771

Abstract

Abstract: Background: Many previous methodological studies suggested to use age as time scale in Cox regression model, and some statistical analyses directly applied this conclusion. In the present study, we explain why age is not a more appropriate time scale compared to the time-on-study time scale. Methods: We address this argument based on five aspects: Cox regression model, conditional likelihood estimation, dataset of left-truncation or right-censoring, algorithms and software for Cox model, and inferring survival function. Furthermore, logical and algorithmic errors arise in the procedure of parameter inference with age time scale, and that certain evaluation indicators proposed by previous studies are inappropriate. Results: The function of time scale is mainly a sampling method for maximum likelihood estimation to infer coefficient of Cox regression model, and the method defined by the age time scale is incorrect in logics and algorithms. Furthermore, age as time scale creates new problems, such as the omission of covariates, loss of information as a continuous variable, increase in dropout, and inability to obtain the survival function. Conclusions: For the Cox regression model, the classic time-on-study time scale is more appropriate compared to age as time scale. Key messages: It is an important discussion because using age as time scale was first proposed decades ago, meaning that lots of turnovers in researchers, newbies tend to accept the assumptions of their predecessors, but the suitability has never been rigorously verified.