Characteristics, cost/effect consideration of clinical examinations, and construction of machine learning models of restrictive cardiomyopathy: insights from Peking Union Medical College Hospital

Abstract

Background: Restrictive cardiomyopathy (RCM) is an uncommon condition with heterogeneous causes. Amyloidosis, a major subtype, presents with diagnostic complexity, economic burden, and prognostic implications. This study aimed to apply machine learning (ML) techniques to improve the diagnosis of amyloidosis among RCM patients and assess the cost-effectiveness of laboratory tests.

Methods: This study included patients with RCM who underwent transthoracic echocardiography (TTE) and cardiac magnetic resonance imaging (MRI). Feature selection was performed using the least absolute shrinkage and selection operator regression based on variables that showed statistical differences between groups. These selected features were then used to construct eight ML models, which were trained and validated using leave-one-out cross-validation. The best-performing model was evaluated for sample size and interpreted using Shapley additive explanations to enhance model transparency. Laboratory testing costs related to autoimmune, infection, tumor, and amyloidosis evaluations were compared across subgroups.

Results: The Random Forest (RF) model achieved the best performance, with an area under the curve of 0.977, an accuracy of 0.908, a sensitivity of 0.869, and a specificity of 0.927. The model also showed a favorable Brier score and a satisfying effect size, indicating good performance in distinguishing amyloidosis from other RCM subtypes. Cost analysis revealed that patients without underlying autoimmune, infectious, or tumor-related etiologies incurred unnecessary expenditures. Multivariate regression identified key imaging features associated with amyloidosis, including left ventricular posterior wall thickness and left ventricular ejection fraction (LVEF) from TTE, and left ventricular short axis diameter, LVEF, and interventricular septum thickness from cardiac MRI.

Conclusion: This study established an interpretable ML model based on the RF algorithm and accurately distinguished amyloidosis among RCM patients. By guiding more targeted use of amyloidosis-specific testing, the model offers a potential cost-saving strategy while improving diagnostic efficiency. These findings support the clinical integration of ML-based tools to streamline decision-making and optimize the allocation of healthcare resources.

Keywords: amyloidosis; cardiac magnetic resonance imaging; health economics; machine learning; restrictive cardiomyopathy; transthoracic echocardiography.