JMIR Publications

ABSTRACT

Background:

Machine learning (ML) and big data analytics are revolutionizing healthcare, particularly in disease prediction, management, and personalized care. With vast amounts of real-world data (RWD) from sources like electronic health records (EHRs), patient registries, and wearable devices, ML offers significant potential to improve clinical outcomes. However, data quality, transparency, and clinical integration challenges remain.

Objective:

This study aims to systematically review the use of ML in real-world data for disease prediction and management, identifying the most common ML methods, disease types, study designs, and sources of real-world evidence (RWE).

Methods:

A systematic review followed the PRISMA guidelines to identify studies that utilized machine learning methods for analyzing real-world data in disease prediction and management. The review focused on extracting data related to the machine learning algorithms used, disease categories, types of studies, and sources of RWE, such as electronic health records (EHRs), patient registries, and wearable devices.

Results:

The systematic review revealed that the most frequently employed machine learning methods were Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM). These methods were applied across various disease categories, with cardiovascular diseases, cancers, and neurological disorders being the most common. Real-world evidence primarily originated from EHRs, patient registries, and wearable devices, with a predominant focus on predictive modeling to improve clinical outcomes.

Conclusions:

ML and big data hold significant promise for enhancing healthcare through better disease prediction and management. However, data quality, model interpretability, and generalizability must be addressed to integrate ML models fully into clinical practice.