Study: Machine learning tool predicts avoidable emergency room and hospital utilization with 90% accuracy

New technology from public benefit corporation Waymark predicted avoidable emergency room and hospital utilization with 90% accuracy, according to a new study released this month.

The machine learning tool, Waymark Signal, looks at “social risk factors and patient risk trajectories,” along with healthcare utilization, to determine which Medicaid patients are at risk for costly visits.

Researchers used Centers for Medicare and Medicaid Services data from 2017–2019 across 26 states and Washington, DC, to determine that “Waymark Signal was 3x better at identifying at-risk patients and 10x better at predicting costs compared to conventional [Medicaid risk] models.”

They also found that Waymark Signal identified a racial bias in existing Medicaid risk models, noting that “because Black individuals typically have less access to higher-cost tertiary care centers, traditional cost-based models often under-predict their future costs and assume the lower costs reflect lower health needs.”

“This study makes an important contribution to the underserved area of Medicaid risk prediction,” Will Shrank, venture partner at venture capital firm Andreessen Horowitz, said in a statement. “While Medicare and commercial insurers have benefitted from numerous risk modeling advances, Medicaid programs have seen far fewer peer-reviewed studies achieving such a significant performance gain.”

Medicaid programs can apply Waymark’s findings “to their own data and populations,” according to the company, which works directly with health plans and primary care doctors to increase access to care for lower-income and underserved patients. Waymark supports about 50,000 Medicaid enrollees in Washington state and Virginia, Fierce Healthcare reported.

Waymark raised $42 million last fall, with backers including Lux Capital, CVS Health Ventures, New Enterprise Associates, and Andreessen Horowitz, according to Fierce.