Endocrine Abstracts

JOINT1304

Background: Subtyping primary aldosteronism (PA) into unilateral and bilateral disease is essential for effective treatment, as unilateral PA can be treated with adrenalectomy, while bilateral PA is managed medically. The gold standard, adrenal vein sampling (AVS), is creating a bottleneck because it is technically challenging and only available in specialized centers.

Aim: To create a predictive score for identifying bilateral PA with a high specificity to bypass AVS, while ensuring a high sensitivity for unilateral disease.

Methods: Retrospective observational study of patients, who underwent AVS at the University Hospital Basel, Switzerland between 2015 and 2023. A decision tree model, using rpart (R version 4.4.2), was used to predict the likelihood of bilateral disease.

Results: A total of 164 patients with PA underwent AVS. 20 patients were excluded due to insufficient selectivity of AVS, leading to a total of 144 patients that were included in the analysis (38% female, median age 51[45;59]). Final diagnosis, based on AVS results, was unilateral in 68% (n=98) and bilateral in 32% (n=46) of patients. The classification decision tree with the highest accuracy was built based on the following predictors: unilateral adrenal mass, potassium, diastolic blood pressure, age, glomerular filtration rate (GFR). The model achieved an overall accuracy of 81%, with a specificity of 95% and sensitivity of 52% for bilateral disease. Assuming a distribution of 50% of uni- and bilateral disease, this decision model would allow to bypass AVS in 50% of bilateral diseases, translating into 25% total reduction in AVS exams. This comes with a price of 5% of patients with unilateral disease not undergoing AVS and surgery (benefit ratio 8:1). Feature importance analysis indicated diastolic blood pressure, GFR, unilateral adrenal mass, and potassium as most influential variables, while age contributed minimally to the model.

Conclusion: Our findings indicate that a decision tree model, based on variables: unilateral adrenal mass, potassium, diastolic blood pressure, age, and GFR can serve as an effective tool for predicting bilateral PA. Importantly, our model works with robust clinical parameters and is not dependent on variables from validation tests (e.g. saline loading test) This classification model has the potential to first, minimize the need for complex AVS procedures in patients with bilateral disease, second, optimize AVS resources for patients with unilateral disease, and third reduce health care costs associated with AVS diagnostics. Prospective validation of this score is necessary.