Performance of summary measures of core body temperature in heat-related health outcome regression models

Climate-related hazards are a leading health and safety concern for working populations. Technological advancements have made real-time collection of core body temperature, a measure of heat strain, more feasible in non-clinical settings. Researchers and practitioners are now faced with how to best summarize a plethora of core body temperature data. Core body temperature data were collected over the course of the workday for 128 agricultural workers in Guatemala from December 2021 to April 2023. Concurrently, measures of heat-related health effects (reduced kidney function, muscle breakdown, and dehydration) were collected. Using linear mixed-effect regression models we assessed the performance of 14 summary measures of core body temperature on the heat-related health effects of interest. Model performance was compared using AIC and BIC. There was no consistent summary statistic that provided the best fit based on AIC or BIC for all four heat-related health effects. The difference between the observed CBT and the average of the first 60 min of the cleaned workday data (delta) provided the best performance for percent change in creatine kinase (muscle breakdown), serum creatinine (reduced kidney function), and serum osmolality (dehydration). The area under the curve (AUC) performed best for dichotomous indicator of acute kidney injury. Choice of how to operationalize longitudinal measurements of core body temperature generated using ingestible core body temperature pills should be driven by the research question and health effect of interest.