Use of Machine Learning and Statistical Inference Methods for Identification of Risk Factors Associated with Atrial Fibrillation in Indian Patients: A Real-World Retrospective Study (Preprint)

BACKGROUND Atrial Fibrillation (AF) is the most common type of cardiac arrhythmia in the general population. Machine Learning (ML) models can be utilized to identify the most pertinent risk factors and build reliable risk factor-based screening algorithms. OBJECTIVE The aim of current study was to evaluate ML in identifying risk factors associated with AF in Indian patients in tertiary care settings. METHODS This real-world, retrospective, observational, multicenter study was conducted by collecting data from anonymized Electronic Medical Records (EMRs) of patients between 1 Jan 2016 and 31 Dec 2019. The ML logistic regression model was used to calculate the Odds Ratio (OR) for identifying the risk factors. RESULTS A total of 5044 patients were included in the study; cohort of AF (cases; n=2516) and non-AF patients (controls; n=2528). The OR showed age above 65 years and female gender as risk factors for incident AF. Smoking, alcohol consumption, dyslipidemia, and hemiplegia, were strongly associated risk factors for AF with high ORs. Chronic diseases such as cardiac arrhythmia, paroxysmal tachycardia, rhematic heart disease, valvular heart disease, chronic obstructive pulmonary disease, stroke, surgery, neurological history, and hypothyroidism also emerged from the model as risk factors for developing incident AF. The model had an F1-score of 0.6930, predicting incident AF with an accuracy of 70.27% (accuracy=0.7027), sensitivity of 0.7042, and specificity of 0.7013. The model also had a positive predictive value of 0.6820 and a negative predictive value of 0.7230 with a threshold of 0.51 for the classification of true AF patients and an area under curve (AUC) of 0.7756. CONCLUSIONS The ML logistic regression model could be a useful predictive tool for incident AF that aids in AF risk management and individualized clinical decision-making.