Primary Care Visit Cadence Based on Risk: A Comparison of Machine Learning and Provider Judgment

Abstract

Summary More frequent primary care visits are associated with improvements in health outcomes and overall well-being and a reduction in health care disparities. Visit cadence is a core part of Oak Street Health’s value-based care model. Although its providers are generally good at predicting a patient’s risk of mortality, they are less accurate at predicting nonmortality outcomes, and they may not update each patient’s risk level in a timely manner. Oak Street launched a pilot initiative to assess whether machine-learning models could outperform provider visit cadence assignment in predicting three key outcomes: acute inpatient admissions, medical cost, and mortality. Pilot results demonstrated that the models were more accurate than provider judgment alone at identifying patients’ risk of admission and future medical cost. On the basis of these positive findings, Oak Street deployed the machine-learning tiering across the organization, with more than 90% of all eligible patients assigned a tier that is routinely refreshed using the predictive models.