Endocrine Abstracts

JOINT3924

Background: Non-functioning adrenal adenomas (NFAs) are considered benign incidental findings, yet their potential metabolic impact remains uncertain. Emerging evidence suggests that NFAs may be associated with metabolic disturbances and an increased prevalence of comorbidities, such as hypertension, type 2 diabetes, depression, osteoporosis and hepatic steatosis. It remains unclear whether these clinical and biological alterations are age-dependent or if obesity acts as a mediator in this association.

Methods: We analyzed a cohort of 150 patients with NFAs, stratified into 2 age groups(below50 and over 50 years). This cutoff was chosen based on physiological transitions associated with aging, including a higher prevalence of insulin resistance, altered immune function and increased cardiometabolic risk. Each of these subgroups is further divided based on the presence or absence of obesity. Clinical analyzed features are: arterial hypertension, osteopenia/osteoporosis, clinical depression, hepatic steatosis, type 2 diabetes. Biological parameters included total, LDL and HDL-cholesterol, serum triglycerides, fasting serum glucose, serum uric acid, serum phosphate, lymphocyte, eosinophil and basophil percentage and C-reactive protein levels. Machine learning (ML) techniques, including Permutation Feature Importance and attention mechanisms, were applied to identify the most relevant predictors of metabolic alterations and evaluate whether obesity mediates these associations. To enhance the accuracy and interpretability of our results, we applied the Permutation Feature Importance method which evaluates the relevance of each feature by measuring its impact on model performance when its values are randomly permuted.

Results: This study aimed to:assess differences in clinical, metabolic, and biochemical features between younger and older patients with NFAs, evaluate the prevalence of associated comorbidities and apply ML techniques to identify key metabolic risk patterns. Additionally, we investigated whether obesity mediates the relationship between NFAs and metabolic alterations, providing new insights into the potential interplay between NFAs, metabolic dysfunction and systemic disease. We selected the Permutation Feature Importance method due to its ability to provide a direct and intuitive measure of feature importance without requiring access to the model’s internal structure. Additionally, we utilized an attention mechanism to assess the importance of these features in predicting metabolic risk. This approach led to a significant increase in model accuracy- approximately 17% higher compared to traditional ML models-demonstrating the benefits of explainability and attention mechanisms in optimizing predictive algorithms.

Conclusion: Our findings suggest that while obesity is common in patients with NFAs, it does not fully explain the observed metabolic alterations. Instead, obesity may act as an aggravating factor, amplifying the metabolic impact of NFAs. Future studies should explore the potential role of insulin resistance and chronic inflammation in the NFA-obesity-metabolism axis.