Machine learning approaches for electronic health records phenotyping: a methodical review

Abstract

Objective: Accurate and rapid phenotyping is a prerequisite to leveraging electronic health records for biomedical research. While early phenotyping relied on rule-based algorithms curated by experts, machine learning (ML) approaches have emerged as an alternative to improve scalability across phenotypes and healthcare settings. This study evaluates ML-based phenotyping with respect to (1) the data sources used, (2) the phenotypes considered, (3) the methods applied, and (4) the reporting and evaluation methods used.

Materials and methods: We searched PubMed and Web of Science for articles published between 2018 and 2022. After screening 850 articles, we recorded 37 variables on 100 studies.

Results: Most studies utilized data from a single institution and included information in clinical notes. Although chronic conditions were most commonly considered, ML also enabled the characterization of nuanced phenotypes such as social determinants of health. Supervised deep learning was the most popular ML paradigm, while semi-supervised and weakly supervised learning were applied to expedite algorithm development and unsupervised learning to facilitate phenotype discovery. ML approaches did not uniformly outperform rule-based algorithms, but deep learning offered a marginal improvement over traditional ML for many conditions.

Discussion: Despite the progress in ML-based phenotyping, most articles focused on binary phenotypes and few articles evaluated external validity or used multi-institution data. Study settings were infrequently reported and analytic code was rarely released.

Conclusion: Continued research in ML-based phenotyping is warranted, with emphasis on characterizing nuanced phenotypes, establishing reporting and evaluation standards, and developing methods to accommodate misclassified phenotypes due to algorithm errors in downstream applications.

Keywords: cohort identification; electronic health records; machine learning; phenotyping.

Figure 1.

Overview of the phenotyping process. Step 1 involves data preparation which includes (i) extraction and processing of relevant data from records of candidate patients from the data warehouse and (ii) manual review of a subset of charts to obtain gold-standard phenotype labels. Step 2 is the algorithm development phase in which researchers use the data from Step 1, often referred to as the data mart, to develop the phenotyping algorithm with a rule-based or machine learning (ML) method. Step 3 evaluates the accuracy of an algorithm by comparing the assigned phenotype from the algorithm to the gold-standard label, often with estimates of the positive predictive value (PPV), sensitivity, and other accuracy metrics. Step 4 applies the algorithm from Step 2 to obtain the cohort of patients with the phenotype for downstream analysis. The identified cohort can then be used in a variety of downstream applications.

Figure 2.

PRISMA diagram for article selection. Only 1 exclusion reason was chosen for each record during the screening process, although the reasons are not mutually exclusive.

Figure 3.

Number of articles that used the various machine learning paradigms.

Figure 4.

Types of structured data and clinical notes used to develop phenotyping algorithms in the selected articles (excluding articles using competition data). A data type is presented if it is used in more than 1 article. Encounters include encounter metadata, while medical history notes include both social history and cardiac surgical history.

Figure 5.

Top phenotypes considered within each machine learning category and the number of articles of each phenotype (excluding articles using competition data sources). Phenotypes are colored if they appear in more than 1 ML paradigm.