Personalized Noninvasive Diagnostic Algorithms Based on Urinary Free Cortisol in ACTH-dependant Cushing's Syndrome

Abstract

Abstract Context Current guidelines for distinguishing Cushing's disease (CD) from ectopic ACTH secretion (EAS) are questionable, as they use pituitary magnetic resonance imaging (MRI) as first-line investigation for all patients. CRH testing is no longer available, and they suggest performing inferior petrosal sinus sampling (BIPPS), an invasive and rarely available investigation, in many patients. Objective To establish noninvasive personalized diagnostic strategies based on the probability of EAS estimated from simple baseline parameters. Design Retrospective study. Setting University hospitals. Patients Two hundred forty-seven CD and 36 EAS patients evaluated between 2001 and 2023 in 2 French hospitals. A single-center cohort of 105 Belgian patients served as external validation. Results Twenty-four-hour urinary free cortisol (UFC) had the highest area under the receiver operating characteristic curve for discrimination of CD from EAS (.96 [95% confidence interval (CI), .92-.99] in the primary study and .99 [95% CI, .98-1.00] in the validation cohort). The addition of clinical, imaging, and biochemical parameters did not improve EAS prediction over UFC alone, with only BIPPS showing a modest improvement (C-statistic index .99 [95% CI, .97-1.00]). Three groups were defined based on baseline UFC: < 3 (group 1), 3-10 (group 2), and > 10 × the upper limit of normal (group 3), and they were associated with 0%, 6.1%, and 66.7% prevalence of EAS, respectively. Diagnostic approaches performed in our cohort support the use of pituitary MRI alone in group 1, MRI first followed by neck-to-pelvis computed tomography scan (npCT) when negative in group 2, and npCT first followed by pituitary MRI when negative in group 3. When not combined with the CRH test, the desmopressin test has limited diagnostic value. Conclusion UFC accurately predicts EAS and can serve to define personalized and noninvasive diagnostic algorithms.