Models, machine learning and Medicaid

How research using real-world data can benefit government health programs

By | 2 minute read | July 8, 2021

Graphic depicting using real-world data and technology to benefit government-supported healthcare.

At IBM Watson Health, we have a dedicated team of researchers who are using our rich real-world data resources and technology to uncover and investigate trends that could benefit government-supported healthcare and social programs. They collaborate with multiple agencies and different levels of government to see how real-world data, such as claims data and data from electronic health records (EHRs), can provide valuable insights to program leaders and other decisionmakers. Their most recent studies included testing a new machine learning method for the US Food and Drug Administration (FDA) as well as investigating key trends related to substance abuse disorders in Medicaid claims data. 

Here are their latest findings:

Machine learning could help federal investigators find health outcomes of interest

IBM Watson Health teamed up with collaborators at Harvard Medical School and multiple federal agencies to test whether machine learning could help uncover events called health outcomes of interest (HOI) that are useful for the FDA to monitor when evaluating the safety of a medical product. Using EHR data from an IBM MarketScan Explorys data set, the researchers applied machine learning techniques to predict the risk of a muscle degeneration condition called rhabdomyolysis. The model they created performed better than a previous expert-developed model and showed that machine learning methods could help investigators monitor claims data from EHRs for HOI.

Medicaid enrollees who abuse multiple substances use more intensive services

In this collaboration between IBM Watson Health and Cambridge Health Alliance, researchers looked at how Medicaid enrollees who were being treated for opioid use disorder (OUD) and at least one other substance use disorder (SUD) used services. The researchers used data on 58,745 Medicaid enrollees from the IBM MarketScan Multi-State Medicaid database to study the level of treatment these people needed over one year. They found that the 29,267 individuals who needed treatment for an additional SUD on top of OUD required more intensive services like emergency department visits, residential care or partial hospitalization. The results suggest that removing barriers to treatment for OUD and providing stronger screening for additional SUDs could help identify people who need more complex treatment earlier and help prevent them from needing more intensive services.

Abusing multiple substances leads to more adverse events among Medicaid enrollees

In a related follow-up study, IBM Watson Health and the Cambridge Health Alliance investigated the prevalence of adverse events, like poisoning or suicidal ideation, among Medicaid enrollees with OUD and at least one other SUD. Using the same data set from the IBM MarketScan Multi-State Medicaid database, they found that people who had both OUD and another SUD had higher risks for adverse events than people with OUD alone. According to the results, people who abused multiple substances were more likely to experience opioid-related poisoning (1.5 times as likely), poisoning from any cause (1.8 times), and to be diagnosed with suicidal ideation (1.8 times). Based on these outcomes, screening for multisubstance abuse could help identify appropriate treatments earlier and mitigate the risk of adverse events.