Development and validation of machine learning algorithms to predict posthypertensive origin in left ventricular...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Left ventricular hypertrophy (LVH) is frequently associated with hypertension, but non-hypertensive causes significantly impact patient management.
• A thorough workup is crucial for identifying the etiology of LVH, especially with increasing prevalence of mild cases.
• There is a need for tools to accurately assess the pretest probability of hypertensive LVH.

Purpose of the Study:

• To develop and validate machine learning models for predicting the hypertensive origin of LVH.
• Utilize first-line clinical, laboratory, and echocardiographic variables for prediction.
• Improve diagnostic accuracy and streamline patient management.

Main Methods:

• Retrospective analysis of 591 patients with LVH (maximal wall thickness ≥12mm).
• Data split into training and testing sets.
• Trained and validated decision tree, random forest, and support vector machine algorithms.

Main Results:

• All models demonstrated strong predictive performance (AUCs ranging from 0.82 to 0.90).
• Support vector machine achieved high specificity (0.96) and sensitivity (0.31) after threshold selection.
• Key predictor variables were consistent across algorithms; online calculators were developed.

Conclusions:

• Machine learning models accurately predict the hypertensive origin of LVH.
• Clinical implementation can decrease the number of etiological workups required for LVH patients.
• These tools support efficient and targeted patient evaluation.