Faculty Receives $3 Million National Institutes of Health Grant to Develop Novel Bayesian Machine Learning Methods
Rutgers School of Public Health associate professor, Liangyuan Hu, has received a $3,301,474 grant (R01HL159077) from the National Institutes of Health National Heart, Lung, and Blood Institute, part of the National Institutes of Health.
Hu and her colleagues will develop a suite of novel Bayesian machine learning methods over the course of five years to address challenges posed by complex longitudinal data with censored survival outcomes, including missing exposure data and causal inference with time-varying confounding and sequential unmeasured confounding. The new methods will be used to advance the understanding of emerging cardiovascular health questions, using enhanced integrated data.
Causal inference methods play an important role in public health and health care service by allowing practitioners to make informed decisions about treatments and improving adoption of evidence-based practices. Population cohort studies funded by the National Institutes of Health are widely used in cardiovascular and population research and have provided fundamental knowledge for cardiovascular disease prevention strategies and public health policies. Pooling data across multiple cohorts provides a unique opportunity for in-depth investigations of emerging cardiovascular disease research questions that cannot be answered by a single cohort. For example, lowering blood pressure is an effective intervention to reduce the risk of cardiovascular disease, but the optimal blood pressure threshold values for initiating the treatment are unknown, particularly among the young and frail elderly. Although forming a fertile ground for stimulating cardiovascular disease research, the elevated complexity of the integrated data poses challenges for statistical analyses and demands specialized and refined techniques beyond traditional analysis approaches.
There are three major analytical challenges. First, many longitudinal cardiovascular disease risk factors have missing data with various missing patterns. Second, traditional causal inference techniques do not meet challenges posed by the issues of time-varying confounding, prolonged time course, and incomplete longitudinal covariates. Third, violations of the sequential ignorability assumption embedded in causal inference methodology can be a potential source of bias.
In response to the urgent need of innovative health data analysis approaches, Hu and her team propose a suite of generalizable statistical methods utilizing machine learning.
“We will develop a suite of novel Bayesian machine learning methods to address challenges posed by complex longitudinal data with censored survival outcomes, including missing exposure data, causal inference with time-varying confounding, and longitudinal unmeasured confounding,” says Hu. “Our work will provide a key apparatus for advancing the understanding of emerging cardiovascular health questions via the enhanced use of integrated data.”