Application of a Classification Method for Studies of Allostatic Load

Sun Y. Jeon, Utah State University
Eric N. Reither, Utah State University

Since initially studied in the MacArthur Study of Successful Aging, health researchers have used the concept of allostatic load (AL) to help understand the consequences of repeated psychological stress on physical health. In the majority of empirical studies, AL is operationalized through a 4-step procedure: (1) attain 8-15 biomarkers for the AL construct from a given dataset; (2) dichotomize each biomarker at the upper quartile (75th percentile), assuming that 25% of the population is exposed to the risk factor; (3) assign one point for each biomarker that lies beyond the risk threshold; (4) sum the points across all biomarkers to acquire the total AL score. In this paper, using the Random Forest classification model, we show that this standard operationalization of AL can produce misleading AL scores through the arbitrary selection of particular biomarkers and risk threshold settings.