DEVELOPMENT OF CLINICAL SCORING SYSTEM AND MACHINE LEARNING-BASED MODEL TO PREDICT DIAGNOSIS OF PRIMARY ALDOSTERONISM IN PATIENTS WITH HYPERTENSION. (June 2022)

• Record Type: Journal Article
• Authors: Buffolo, Fabrizio; Burrello, Jacopo; Burrello, Alessio; Heinrich, Daniel; Adolf, Christian; Muller, Lisa Marie; Chen, Rusi; Forestiero, Vittorio; Sconfienza, Elisa; Tetti, Martina; Veglio, Franco; Williams, Tracy Ann; Mulatero, Paolo; Monticone, Silvia
• Is Part Of: Journal of hypertension. Volume 40(2022)Supplement 1
• Page Start: e32
• Publication Date: 2022-06
• Subjects: Hypertension -- Periodicals; Hypertension -- Periodicals; 616.132005
• DOI: 10.1097/01.hjh.0000835556.56797.ae
• Languages: English
• ISSNs: 1473-5598

Abstract

Abstract : Objective: Current guidelines recommend screening for PA in patients with hypertension on the basis of individual factors, that considered together are present in more than 50% of patients. Recently, some experts proposed to further expand the screening for PA to all patients with hypertension, potentially increasing the burden and the costs on the health care system. Design and method: We designed a study to build and validate prediction models based on supervised learning and a conventional scoring system to define the risk of PA in arterial hypertension and tailor the diagnostic workup to the individual risk of each patient. We developed a clinical score and supervised machine learning algorithms in a retrospective internal cohort of 4059 patients with hypertension, and an external cohort of 584 patients with hypertension. Primary aldosteronism was confirmed by confirmatory tests, in agreement with major international recommendations, and subtype diagnosis was achieved by computed tomography and adrenal vein sampling. Results: On the basis of 6 widely available parameters (male sex, systolic blood pressure, antihypertensive treatment, body mass index, lowest potassium, and organ damage) we developed a numerical scoring system (SToP-PA score) and 308 machine learning based models, selecting the one with the highest predicting performances. At internal validation we obtained high predictive performance with SToP-PA score (ptimised sensitivity of 90.7% for PA and 92.3% for unilateral PA) and even higher with machine learning based model (ptimised sensitivity of 96.6% for PA, and 100.0% for unilateral PA). The application of these models allowed the identification of a subgroup of patients with very low probability of having PA (0.6% with both models) and null probability of having unilateral PA. Finally, we validated our models within an independent external cohort, with performance not significantly different from internal validation. Conclusions: The SToP-PA score and the machine learning model accurately predict the individual pretest probability of PA in patients with hypertension, avoiding the screening for PA in up to 32.7% of patients using a machine learning algorithm, without omitting patients with unilateral UPA.