Unsupervised Machine Learning for Assessment of Left Ventricular Diastolic Function and Risk Stratification

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Current 2016 American Society of Echocardiography guidelines for assessing left ventricular diastolic function have limitations.
• A need exists for improved classification and risk stratification methods.

Purpose of the Study:

• To develop a data-driven, unsupervised machine learning approach for classifying diastolic function.
• To stratify risk using parameters from the 2016 American Society of Echocardiography guidelines.

Main Methods:

• Utilized transthoracic echocardiography data from 24,414 adult patients.
• Applied unsupervised machine learning to nine left ventricular diastolic function variables.
• Excluded patients with prior valve intervention, congenital heart disease, or cardiac assist devices.

Main Results:

• Identified three distinct patient clusters: normal diastolic function, impaired relaxation, and increased filling pressure.
• Demonstrated significantly different 3-year cumulative mortality rates across the clusters (11.8%, 19.9%, 33.4%).
• The machine learning clusters outperformed guideline-based grading for prognostication (C-index 0.607 vs. 0.582).

Conclusions:

• Unsupervised machine learning effectively identified distinct physiological and prognostic clusters based on diastolic function Doppler variables.
• These clusters offer a simpler, replicable method for diastolic function-related risk stratification in clinical practice and trials.