In a randomized clinical trial, the treatment population is generally heterogeneous having a differential response to the received investigational treatment. There may be several genomic and baseline characterestics for such response. One of the key roles for statisticians is to determine the relationship of several characteristics such as age, sex, BMI, comorbidities, genetics, and environmental factors for a clinical outcome of interest. For example, researchers have found that the human epidermal growth factor receptor 2 (HER2) gene controls the rapid growth and spread of breast cancer and have developed HER2 protein blocking medical products such as trastuzumab1. With the availability of electronic health record data, longitudinal data (repeated data collection on same respondents over time) from clinical trials, together with efficient existing and newly-developed statistical modeling, enable better therapeutic outcomes and contribute to personalized medicine. The health care industry has highly adopted these subgroup analysis methods in the recent past, to understand the efficacy and survival of patients with a new drug in consideration.  Some of the subgroup analysis methods that are of essential value in the literature includes FiveStar method, VirtualTwins, PRISM among others.

The FDA has recently published a regulation (21CFR813, subpart H) that allows the use of PFS or other surrogate clinical endpoints other than survival or irreversible morbidity in the accelerated approval of new drugs for serious or life-threatening illnesses.

Along with finding an association of essential features with the outcome, most phase 3 clinical trials use progression free survival as the surrogate endpoint in disease areas such as oncology,  readmission, predicting the course of Alzheimer’s disease, getting ahead of the surgery outcomes, and deterioration of patients in ICU. Hospitals under Medicare’s Hospital Readmissions Reduction program2 must ensure unplanned readmission to avoid penalties, reduce cost and resources. Statistics form the core for such machine learning and predictive algorithms guiding clinicians in disease diagnosis and therapeutic development. According to a survey, these algorithms have been adapted in about 60% of the healthcare industry. Once a drug passes through phase 1, and 2 of the clinical trial process, the newly developed compound is tested against comparative treatment/ placebo and progression free survival is an essential endpoint where the survival of patients is compared before and after initiation of the treatment.  Weibull regression has been widely used in modelling progression free survival. Bayesian shrinkage priors has been a successful variable selection technique. Here the presentations are focused on to capture a unified approach of subgroup analysis with penalized weibull regression through shrinkage priors. Further the utility based methods and non-parametric approaches their application in clinical trials are also discussed.

Through this session, we aim to bring together the best of both worlds by combining statistical knowledge and concepts of variable selection (i) identifying new challenges in research (ii) providing a platform for interdisciplinary discussions and (iii) help shape future directions for understanding the pathway of its utility in clinical research. The session will have four confirmed presentations actively involved in developing statistical methodologies and applying such models to understand the fundamentals of subgroup analysis It will facilitate floor discussion and exchange of new ideas and cross-domain knowledge transfer. 

Organiser: Arinjita Bhattacharyya 

Chair: Arinjita Bhattacharyya 

Speaker: Arinjita Bhattacharyya 

Speaker: Riten Mitra