ORF23-01 - Clinical Prediction Algorithms for Subtyping Primary Aldosteronism in a Multicentre Australian Cohort

Introduction: Primary aldosteronism (PA) is the leading endocrine cause of hypertension. Recent guidelines recommending universal screening for PA in all with hypertension will increase the cases requiring differentiation of unilateral (surgically correctable) from bilateral subtypes, traditionally achieved by adrenal vein sampling (AVS). AVS is invasive and resource-intensive, with limited availability. This study evaluated existing subtyping algorithms in an Australian cohort and developed a novel predictive model to streamline the diagnostic pathway.

Methods: This multicentre retrospective study included patients with PA who underwent AVS at expert centres in Victoria, Queensland, and Western Australia. The PA subtype, unilateral (UPA) or bilateral (BPA), was determined by AVS and post-surgical outcomes. Published algorithms with high accuracy in systematic reviews were tested. A novel predictive model was developed using a stepwise backward selection logistic regression model in a training set (80%) and tested in a validation set (20%).

Results: Of 330 participants, 151 (46%) had UPA and 179 (54%) BPA. Published algorithms showed lower specificity than originally reported. For BPA, a score ≥8 using the algorithm by Kobayashi et al (2018) had 74% specificity and 70% sensitivity (AUC 0.78), and the algorithm by Kamemura et al (2017) had 92% specificity and 32% sensitivity (AUC 0.72). For UPA, algorithms by Song et al (2022) and Umakoshi et al (2018) had high specificity (99–100%) and low sensitivity (1–15%) (AUC 0.69 and 0.68, respectively). Collectively, these algorithms would have reduced the need for AVS by 61%, though with variable accuracy.

Our model found BPA was rendered more likely with female sex (odds ratio [OR] 2.19), higher potassium (OR 5.61), lower sodium (OR 0.81), lower aldosterone-renin ratio (OR 0.998), and absence of an adrenal nodule on CT (OR 0.17) (AUC 0.84, 95% CI 0.79-0.90). The optimal threshold value, or probability score, to estimate the likelihood of BPA, ranged from 0.40 to 0.48 for achieving the maximum accuracy of 76% in the training data set, and ranged from 0.44 to 0.57 in the validation data set for a maximum accuracy of 85%. Applying a high-specificity threshold (0.84 probability score) to our model would allow 26% of patients to bypass AVS and proceed directly to medical management with a 7% false-discovery rate.

Conclusion: Prediction models utilising routine biochemical and radiological data can accurately identify PA subtypes. Existing algorithms derived from other populations show reduced specificity in an Australian population. Our novel probability-based model, which (unlike previous algorithms) incorporates serum sodium, offers a scalable tool for personalised decision-making. High-specificity thresholds can safely transition a significant proportion of patients to targeted therapy while reducing the burden on AVS resources.