Predictive Models May Aid MACS Diagnosis 

A retrospective study published in the Journal of the American College of Surgeons suggests that clinical and radiologic features may help distinguish mild autonomous cortisol secretion (MACS) from nonfunctional adrenal incidentalomas (NFAI) in patients evaluated at an artificial intelligence-driven adrenal incidentaloma clinic. 

MACS occupies a diagnostic gray zone: Patients do not have overt Cushing’s syndrome, but may have cortisol-related comorbidities that affect cardiovascular and metabolic risk. In clinical practice, diagnosis is typically based on dexamethasone suppression testing, with evaluation for related conditions such as hypertension and type 2 diabetes.  

For the study, which was led by Mehmet Kostek, MD, FACS, of the Department of Surgery at University of Alabama at Birmingham, and colleagues, researchers analyzed data from 397 patients evaluated between February 2022 and August 2024 who were diagnosed with either MACS or NFAI. The mean age was 62.3 years, 55% were women, and MACS was present in 34% of the study population. 

Researchers developed models to distinguish MACS from NFAI using routinely collected clinical and imaging variables. Patients were randomly divided into training and test cohorts in a 3:1 ratio, and the models were developed using Least Absolute Shrinkage and Selection Operator (LASSO) regression and random forest algorithms. 

Body mass index, Posterior Adiposity Index, and the number of antihypertensive medications emerged as the strongest predictors of MACS, suggesting that metabolic and body composition-related features may help identify patients more likely to have the condition. 

The random forest model showed higher discrimination than the LASSO model. In the test cohort, the random forest approach achieved an area under the curve of 0.736, compared with 0.686 for the LASSO model. At Youden Index thresholds balancing sensitivity and specificity, the random forest model achieved 65% sensitivity, 76% specificity, and 72% accuracy, compared with 59% sensitivity, 76% specificity, and 70% accuracy for the LASSO model. 

The authors suggest that predictive models incorporating clinical and radiologic characteristics may help identify patients at higher likelihood of MACS in adrenal incidentaloma clinics.

However, model performance was moderate, and further validation will be needed before the models could be used more broadly. 

The authors reported no relevant conflicts of interest.