Development of predictive models for differential diagnosis of hypertrophic cardiomyopathy

Abstract

Aim. To develop predictive models for differential diagnostics of hypertrophic phenotype in patients with concomitant diseases, as well as to validate through independent assessment.

Material and methods. The study included an analysis of 1169 medical records from the medical information system of patients with severe myocardial hypertrophy and a preliminary diagnosis of hypertrophic cardiomyopathy (HCM) (I42.1, I42.2). The patients were divided into 3 following groups: patients with a probable HCM, patients with moderate myocardial hypertrophy (>15 mm) due to a known disease, and patients with severe myocardial hypertrophy that cannot be explained by left ventricular pressure overload ("gray zone"). The original dataset contains 74 parameters. Machine learning models of the following classes were created and optimized: logistic regression, support vector machine, decision tree, and gradient boosting decision trees.

Results. All models have sufficient accuracy in detecting HCM, but the accuracy of ruling out the diagnosis is quite low. The use of a machine learning model using logistic regression reduced the HCM misdiagnosis risk in the group of questionable diagnosis to 31%.

Conclusion. Four predictive models for differential diagnosis of severe left ventricular myocardial hypertrophy were developed to improve the HCM diagnosis. Blinding validation showed that logistic regression is the most optimal model for clinical practice.