∑ Survival analysis
∑ Nonparametric functional estimation
∑ Efficient estimation
∑ Empirical processes
∑ Semiparametric inference
Sundar Subramanianís principal research is non-parametric and semi-parametric statistical inference from censored time-to-event-data.
In cancer clinical trials the time to a terminal event, such as mortality due to a specific cancer, cannot be always observed because
a) either a patient is lost to follow-up (or)
b) a competing terminal event preempts the occurrence of the event of interest
The competing causes induce what is called dependent censoring, posing new statistical challenges for the analysis of such data.
Frequently, however, such competing risks data are subject to both independent and dependent censoring and, in some cases, the cause of failure may even be missing.
Subramanianís current focus is on semi-parametric models for analyzing competing risks data, with the main research thrust being investigations into model selection, model diagnostics, robustness studies giving insights into model misspecification, and extensions addressing missing cause of failure (censoring) information.
His other ongoing research interests are
∑ estimation and inference for median regression parameters, including checking for model adequacy;
∑ multiple imputations for improved semi-parametric survival function-estimation from homogeneous right censored data when the censoring indicators are partially missing;
∑ bootstrap methods for model selection and bandwidth computation;
∑ model misspecification studies using simulation
His investigations involve study of the large sample behavior of estimators using techniques from counting processes and martingales, empirical processes, kernel estimation, and information bound theory.
The procedures have strong theoretical basis and find applications in Biostatistics.