JMIR Publications

ABSTRACT

Background:

Electronic medical records (EMRs) contain large amounts of detailed clinical information. Using chart review to identify conditions within large quantities of EMRs can be time-consuming and inefficient. EMR-based phenotyping utilizing machine learning (ML) and natural language processing (NLP) algorithms is a continually developing area of study that holds potential for numerous mental health disorders.

Objective:

This review outlines and evaluates the current state of EMR-based case phenotyping for depression.

Methods:

A systematic review of EMR-based algorithms for phenotyping depression was completed. This research encompassed studies published from January 2000 to May 2023. The search involved three databases: Embase, MEDLINE, and APA PsycInfo. This was done using selected keywords that fall into three categories: terms connected with EMR, terms connected to case identification, and terms pertaining to depression. This document adheres to the Preferred Reporting Items for Systematic Reviews and Meta-analyses Extension for Scoping Reviews (PRISMA) guidelines.

Results:

A total of 20 articles were assessed and summarized in the review. Most of these studies were undertaken in the United States, accounting for 75% (15/20). This was followed by the United Kingdom with 15% (3/20) and Spain with 10% (2/20). Both data-driven and clinical rule-based methodologies were identified. The development of EMR-based phenotypes and algorithms is indicative of the data accessibility permitted by each respective health system, leading to varying performance levels among different algorithms.

Conclusions:

It is crucial to understand the commonalities and disparities in health systems, data gathering techniques, data release procedures, and current clinical routes for successful algorithm development strategies. There have been several propositions for strategies that aid in defining cases based on phenotype.