Simultaneous variable selection in regression analysis of multivariate interval‐censored data

• Published in: Biometrics Vol. 78; no. 4; pp. 1402 - 1413
• Main Authors: Sun, Liuquan, Li, Shuwei, Wang, Lianming, Song, Xinyuan, Sui, Xuemei
• Format: Journal Article
• Language: English
• Published: United States Blackwell Publishing Ltd 01.12.2022
• Subjects: Algorithms; Computer applications; Computer Simulation; Data analysis; Data structures; EM algorithm; Feature selection; Humans; interval censoring; Likelihood Functions; Longitudinal Studies; Maximization; minimum information criterion; Models, Statistical; Multivariate analysis; Optimization; Parameter estimation; patients; Regression Analysis; risk; Risk analysis; Risk factors; Time Factors; transformation models; interval censoring; EM algorithm; multivariate analysis; transformation models; minimum information criterion
• ISSN: 0006-341X; 1541-0420; 1541-0420
• DOI: 10.1111/biom.13548

Multivariate interval‐censored data arise when each subject under study can potentially experience multiple events and the onset time of each event is not observed exactly but is known to lie in a certain time interval formed by adjacent examination times with changed statuses of the event. This type of incomplete and complex data structure poses a substantial challenge in practical data analysis. In addition, many potential risk factors exist in numerous studies. Thus, conducting variable selection for event‐specific covariates simultaneously becomes useful in identifying important variables and assessing their effects on the events of interest. In this paper, we develop a variable selection technique for multivariate interval‐censored data under a general class of semiparametric transformation frailty models. The minimum information criterion (MIC) method is embedded in the optimization step of the proposed expectation‐maximization (EM) algorithm to obtain the parameter estimator. The proposed EM algorithm greatly reduces the computational burden in maximizing the observed likelihood function, and the MIC naturally avoids selecting the optimal tuning parameter as needed in many other popular penalties, making the proposed algorithm promising and reliable. The proposed method is evaluated through extensive simulation studies and illustrated by an analysis of patient data from the Aerobics Center Longitudinal Study.