OBJECTIVES : Machine learning (ML) techniques are revolutionizing the field of health outcomes research by generating novel insights using various datasets. The aim of this study was to summarize the available literature on the use of ML techniques in outcomes research and evaluation of ML use in various research methodologies.

METHODS : A literature search on PubMed was performed for identifying studies involving any of the following outcomes: safety, efficacy, economic, patient-reported outcomes. Findings were further abbreviated: (A) published within 5 years, (B) use of ML techniques for assessing outcomes using clinical or real-world datasets.

RESULTS : 32 studies met the inclusion criteria; 26 were observational studies and 6 were clinical studies. We identified three broad data solutions for ML techniques: (1) predicting outcomes (k=23); (2) doing patient classification (k=8); and, (3) developing patient clusters (k=2).

A detailed assessment of studies revealed various reasons that give ML algorithms an upper hand over traditional methods. Key reasons include:

• Handling complex/big datasets (k = 11): This capability allows users to look at big datasets (EMR/social data) and multiple variables in single go;
• Improved accuracy (k = 8): ML algorithms have performed better in terms of accuracy when compared with traditional methods like simple regression;
• Scalability/Automation (k = 4): ML algorithms provide more generalized approach to solving a complex problem and can easily be automated;
• Others (k = 2): Cost effective, ease of use

Apart from studies listed above, 7 studies were found that developed a novel approach using ML algorithms. All of the analyzed studies aimed at solving unique problem and ML came in as a handy tool in researcher’s toolkit.

CONCLUSIONS : This review supports the use of ML techniques in outcomes research and its use is primarily driven by unique and advanced features it has to offer over traditional ways.