Areas of expertise

My research is dedicated to improving health outcomes through developing advanced statistical methods motivated by datasets such as the UK Biobank, AllofUs, and various cancer studies. I am particularly interested in areas like competing and semicompeting risks survival analysis, longitudinal analysis, Bayesian inference, model assessment tools, data integration, and statistical genomics. Here are a few recent highlights from my work.

Semicompeting risks model: In the context of complex diseases, individuals may encounter multiple nonterminal events alongside a single terminal event. The occurrence of a terminal event can censor the nonterminal events, but the reverse is not the case. Motivated by type 2 diabetes disease mechanism, we proposed a semicompeting risks model within the Bayesian framework accounting for two nonterminal events and one terminal event.

Longitudinal data analysis: Longitudinal collection of biomarkers play a significant role in disease diagnostics, screening, and also in classifying subjects into homogenous subgroups, which becomes challenging in the presence of multiple time-varying covariates. Our ongoing work focus on classifying subjects showing similar behavior accounting for complex nature of multiple time-varying covariates.

Joint model with competing risks: In medical studies, longitudinal biomarker and competing risks survival data are collected simultaneously along with other covariates. It is common to encounter competering risks survival data with partially masked causes. We developed a joint model for analyzing longitudinal and competing risks survival data accounting for the masked causes within the Bayesian framework.

Cure rate model: In clinical studies, some subjects may never experience the disease of interest (or become cured of it), regardless of the duration of follow-up. In the context of competing risks survival data with masked causes, our recent developments include a mixture cure rate model, a Bayesian joint model incorporating a mixed-effects model for longitudinal data and a promotion time cure rate model.

Model assessment tools: Quantifying the fit of survival data with respect to each cause becomes quite difficult in the presence of masked causes. Within the joint modeling framework, we propose cause-specific concordance (C)-index and cause-specific deviance information criterion (DIC). These model assessment tools can be useful to quantify predictive performance as well as fit of survival data for each observed cause. Our model assessment tools have also extended within the cure rate models.

Data integrataion: Leveraging historical or complementary data has been found to be useful in current data analysis. Our recent research works include integrating prevalent data under semicompeting risks and partial borrowing-by-parts power prior under joint modeling frameworks.

Interim analysis in clinical trials: Early decision making at phase Ib/II is crucial in the drug development process. Futility boundary is often desirable to make an appropriate decision at each interim look. During my internship, I have developed an optimized interim futility decision making algorithm for early phase clinical trials with binary endpoint using Bayesian predictive probability of failure.