Hypothalamic-pituitary-adrenal axis reactivity to acute stimulation is frequently assessed by repeated sampling of salivary cortisol. Researchers often strive to distinguish between individuals who show (responders) and those do not show (nonresponders) cortisol responses. For this, fixed threshold classification criteria, such as a 2.5-nmol/l baseline-to-peak increase, are frequently used. However, the performance of such criteria has not been systematically evaluated.
Repeated salivary cortisol data from 504 participants exposed to either the Trier Social Stress Test (TSST; n = 309) or a placebo protocol (n = 195) were used for analyses. To obtain appropriate classifications of cortisol responders versus nonresponders, a physiologically plausible, autoregressive latent trajectory (ALT) mixture model was fitted to these data. Response classifications according to the ALT model and information on the experimental protocol (TSST versus placebo TSST) were then used to evaluate the performance of different proposed classifier proxies by receiver operating characteristics.
Moment structure of cortisol time series was adequately accounted for by the proposed ALT model. The commonly used 2.5-nmol/l criterion was found to be overly conservative, resulting in a high rate of 16.5% false-negative classifications. Lowering this criterion to 1.5 nmol/l or using a percentage baseline-to-peak increase of 15.5% as a threshold yielded improved performance (39.3% and 26.7% less misclassifications, respectively).
Alternative classification proxies (1.5 nmol/l or 15.5% increase) are able to effectively distinguish between cortisol responders and nonresponders and should be used in future research, whenever statistical response class allocation is not feasible.