JMIR Publications

ABSTRACT

Background:

Patient navigation (PN) programs have demonstrated efficacy in improving health outcomes for marginalized populations across a range of clinical contexts by addressing patient barriers to healthcare, including the Social Determinants of Health (SDoH). However, identifying SDoH can be challenging due to a myriad of factors. Machine learning offers new opportunities to predict social determinants of health to enhance the effectiveness of patient navigation in improving health outcomes for diverse patient populations.

Objective:

In this case study, we describe novel machine learning approaches to predict social determinants of health based on patient navigators’ interactions with patients in two Chicago area patient navigation studies. The paper provides recommendations for data collection and the application of machine learning techniques.

Methods:

We conducted two experiments exploring the feasibility of using machine learning to predict patients’ SDoH using data collected from patient navigation research. The machine learning algorithms were trained on data collected from two Chicago area patient navigation studies. In the first experiment, we compared several Machine Learning algorithms (Logistic Regression, Random Forest, Support Vector Machine, Artificial Neural Network, and Gaussian Naive Bayes) to predict SDoH from both patient demographic and navigator’s encounter data over time and determined whether one or more algorithms were suitable for this task. In the second experiment, we used multi-class classification with augmented information, such as transportation time to a hospital, to predict multiple SDoH for each patient.

Results:

In the first experiment, the Random Forest classifier obtained the highest accuracy among the classifiers tested. In the second experiment, multi-class classification was able to effectively predict a handful of patients’ SDoH based purely on demographic and augmented data. The case studies yielded some valuable lessons, including: (1) being aware of model limitations and bias; (2) planning for standardization of data sources and measurement; and (3) identifying and anticipating the intersectionality and clustering of social determinants of health.

Conclusions:

This study is the first approach to applying patient navigator encounter data, as well as multi-class learning algorithms, to predict SDoH. Although our focus was on predicting patients’ SDoH, machine learning can have a broad range of applications to the field of patient navigation, from tailoring intervention delivery (e.g., supporting patient navigation decision-making), to informing resource allocation for measurement, and augmenting PN supervision. However, any application of machine learning to patient navigation work will be constrained by the availability of data from which to train algorithms. Thus, attention is needed on bolstering data that can be made useful for machine learning algorithms. Clinical Trial: N/A