SAT-376 Dynamic Modeling of the 1 mg Overnight Dexamethasone Suppression Test: Sources of Variability and Strategies to Improve Diagnostic Accuracy for Detection of Autonomous Cortisol Secretion in Adrenal Adenoma Patients

Abstract

Among patients with adrenal adenoma, the 1 mg overnight dexamethasone suppression test (DST) is commonly used to help identify patients with biochemical evidence of autonomous cortisol secretion (ACS). The diagnostic accuracy of the DST in this setting is only fair, as illustrated by the application of different cut-scores for DEX-suppressed cortisol concentrations (e.g. 50 vs. 138 nM) favoring sensitivity or specificity, respectively (1). In the present study we sought to model the predictors of the DST to better characterize sources of variation and error in the diagnostic test. Forward solutions for DEX-suppressed total and free cortisol concentrations from midnight to 8 AM were obtained using simultaneous solution of differential equations in a dynamic 3-compartment model of free cortisol appearance, elimination, and reversible binding to serum proteins (2). Monte Carlo simulation was performed based on published distributional data for controls (n=1000) and ACS patients (n=1000). Predictor variables were assessed using standardized beta (STB) coefficients, which represent change in the standard deviation (SD) of the outcome (numerator, i.e. DEX-suppressed cortisol concentration) for each SD change in a predictor (denominator) in a multivariable context (2). The analysis confirmed diagnostic accuracy of the DST consistent with published guidelines (1). Significant predictors of DEX-suppressed cortisol included ACTH-independent cortisol secretion rate (CSRmin), free cortisol half-life, and the concentration of midnight serum total cortisol, corticosteroid binding globulin (CBG), and albumin (all P<0.01). The predictor variables having STB coefficients >0.4 included CSRmin (STB=0.67), free cortisol half-life (STB=0.53), and CBG (STB=0.44), with smaller influence of midnight total serum cortisol (STB=0.14), and albumin (STB=0.04) concentrations. STBs for DEX-suppressed free cortisol concentration were like those observed for total cortisol except that the influence of CBG was eliminated. Using a DEX-suppressed free cortisol cut-score of 2.5 nM improved sensitivity and specificity compared to DEX-suppressed total cortisol (P<0.001). We conclude that our model for DST agrees with experimental literature demonstrating limited diagnostic accuracy of the test and we have identified multiple factors that affect the DEX-suppressed cortisol concentration in both control and ACS populations. In addition, strategies to adjust for confounding variables, such as use of DEX-suppressed free cortisol, might improve diagnostic accuracy of the DST.