[Machine learning methods in differential diagnosis of ACTH-dependent hypercortisolism].

Area of Science:

• Endocrinology
• Medical Informatics
• Machine Learning

Context:

• ACTH-dependent hypercortisolism presents diagnostic challenges, with ectopic ACTH syndrome (EAS) and Cushing's disease (CD) requiring distinct management.
• Accurate differential diagnosis is crucial for appropriate patient treatment and outcomes.
• Noninvasive diagnostic methods are highly desirable to reduce patient burden and healthcare costs.

Purpose:

• To develop a noninvasive method for differentiating ACTH-dependent hypercortisolism.
• To evaluate a machine learning-based algorithm for predicting the probability of EAS.
• To optimize diagnostic accuracy using clinical data analysis.

Summary:

• A cohort study analyzed 223 patients with ACTH-dependent hypercortisolism (175 CD, 48 EAS).
• Key clinical variables including ACTH, potassium, cortisol levels, and pituitary adenoma size were identified.
• A Gradient Boosting Model (GBM) demonstrated superior predictive performance, achieving an AUC of 0.920 in the test sample.

Impact:

• The developed prognostic model facilitates the differentiation of EAS and CD.
• It serves as an effective primary screening tool for patients with ACTH-dependent hypercortisolism.
• This approach enhances diagnostic efficiency and supports timely, targeted therapeutic interventions.