High-dimensional survival analysis using penalized hazard models to integrate genomic, environmental, and socioeconomic variables for precision public health decisions

Abstract

High-dimensional survival data have become increasingly common in modern public health research, especially with the rapid growth of genomic sequencing technologies, satellite-based environmental monitoring, and detailed socioeconomic profiling. These multidomain datasets offer enormous potential for understanding population-level health risks, but they also introduce significant analytical challenges, including overfitting, multicollinearity, and the difficulty of selecting meaningful predictors from thousands of correlated variables. To address these challenges, this study applies penalized hazard model specifically the LASSO-Cox and elastic-net Cox approaches which are well-suited for variable selection and robust risk prediction in high-dimensional settings. Unlike traditional Cox models, penalized methods can efficiently shrink irrelevant coefficients toward zero while identifying a small, interpretable subset of influential predictors across genomic, environmental, and socioeconomic domains. Because real-world multidomain datasets are often inaccessible or restricted, this research uses a carefully constructed simulated dataset that mimics realistic public health conditions. The simulated cohort incorporates hundreds of genomic markers, multiple environmental exposures such as PM2.5 and temperature variability, and socioeconomic indicators reflecting income and neighborhood disadvantage. By applying penalized survival models to these simulated data, the study demonstrates how key predictors can be identified and how model performance metrics such as the concordance index, time-dependent AUC, and calibration quality can be evaluated. Overall, the abstract presents a rigorous, results-based framework that illustrates how penalized hazard models can support precision public health by integrating complex, high-dimensional data into actionable survival predictions.