JMIR Publications

ABSTRACT

Background:

Efforts are underway to capitalise on the computational power of the data collected in Electronic Medical Records (EMRs) to achieve a Learning Health System (LHS). Artificial Intelligence (AI) in healthcare has promised to improve clinical outcomes and many researchers are creating and validating AI algorithms on fixed retrospective datasets. Transitioning these algorithms from demonstrated validation on retrospective datasets to effectively working with real-time data pipelines from EMR data is rare. There is a poor understanding of the current enablers and barriers to empower this shift from dataset-based utilisation to real-time implementation of AI in health systems. This study investigates the factors which influence the success of AI implementations and how these factors can be mapped to an existing framework towards digital transformations to enable action.

Objective:

The aim was to conduct a systematic review of published case studies and guidelines that explored the implementation of AI utilising EMR data to enhance the care of hospitalised patients. The second aim was to map the findings into a three-horizon framework for successful digital health transformation to enable action.

Methods:

The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines were adhered to. PubMed, Scopus, Web of Science and IEEE were searched for studies published between January 2010 to January 2022. Articles with case studies and guidelines on the implementation of AI analytics in hospital settings using EMR data were included. We excluded studies performed in primary and community care settings. All studies were double screened and data extraction conducted independently using the Covidence® systematic review software. We coded evidence from the included studies that related to (1) enablers of AI implementation and (2) barriers to AI implementation. Actionable recommendations for the implementation of AI analytics towards achieving LHS were provided by mapping the findings of this study to a three-horizon framework for LHS analytics implementation.

Results:

Of the 1,247 studies screened, 26 met the inclusion criteria. Eleven studies developed and implemented AI analytics for enhancing the care of hospitalised patients while the remaining eight provided implementation guidelines. A total of 28 enablers were identified, of which 10 were newly found in this study. A total of 18 barriers were identified; six were new in the present study. The majority of these newly identified enablers and barriers were related to information and technology factors. The findings were mapped to the three-horizon framework for digital health transformation.

Conclusions:

There are significant issues in implementing AI in healthcare. Shifting from validating datasets to working with live data is challenging. The present review incorporated the identified enablers and barriers into an evidence-based framework and offers actionable recommendations for implementing AI analytics to achieve a LHS.