Modeling longitudinal core temperature in a crossover trial of farmworkers in California

Analyzing core body temperature in field settings presents unique challenges, including high-frequency longitudinal measurements, individual physiological variability, and the environmental noise of active work shifts. This presentation discusses a comprehensive workflow in Stata for processing and modeling data from a crossover trial designed to evaluate cooling interventions (bandanas and mitts) among California farmworkers. I detail the steps necessary to move from raw, minute-by-minute sensor data to statistical inference. Key methodological hurdles addressed include (1) high-frequency data cleaning and the use of mipolate for data interpolation; (2) data smoothing using lowess to manage data artifacts; (3) the calculation of area under the curve (AUC) using integ; and (4) the application of mixed-effects REML regression to account for the crossover design, including trial week, carryover effects, and time-invariant physiological covariates (BMI, age, and sex). While the primary focus is on the analytical steps rather than the efficacy of the interventions, I demonstrate how Stata’s margins, contrast, and coefplot packages can be used to visualize complex longitudinal results and their sensitivity to model specifications. This toolkit offers a reproducible framework for researchers handling complex thermal or physiological time-series data in occupational health.