This blog post is co-authored by Slawek Kierner, SVP for Enterprise Data & Analytics, Humana and Tie-Yan Liu, Deputy CEO, Microsoft Research China.
Using AI models to have real effects
Trips to the hospital happen. And while everyone in the industry strives to provide world-class care for patient experiences, everyone — both patients and care teams — would rather avoid staying in the hospital. The teams at Humana felt they had enough data to investigate the possibility of proactively identifying when patients were heading for a high-risk incident, and they put Microsoft Cloud for Healthcare and AI technology to the test.
Humana’s question was straightforward: How do we take the data we have today and use it proactively? How do we use AI to identify signals in our existing ecosystem that tell us that someone might be experiencing a scenario that puts them at risk? And most importantly, how do we proactively engage and meet our members in their own environment before ending up in an emergency room?
The first approach to monitoring chronic patients is often focused on remote monitoring of patients and IoT devices, but to approach this challenge we wanted to take a different and much larger approach with AI. Combining clinical data triggered key event triggers that could indicate a patient’s deteriorating health, and a combination of prediction models, Microsoft Research and Humana’s computer science team collaborated on research to investigate whether they could develop a system that would identify potential gaps in care among patients and engage high-risk patients with care teams that could reach out and offer support.
The power of AI model enhancement
The result of the research was an insight into the future of AI in health. Health organizations like Humana have spent the last many years developing powerful predictive models with a single focus. Humana had existing models that predicted the likelihood of acute hospitalizations in the near future across their 4.9 million Humana Medicare Advantage members, as well as additional models that predict the cost of care and the likelihood of readmissions. Microsoft Research and Humana data science teams brought these models together with structured data to create and test a combination of neural networks and tree-based models with Microsoft cloud technologies.
Cloud tools were essential to develop the multivariable model as well as technology in Microsoft Cloud for Healthcare to unify different patient data streams. In addition, Microsoft Research has designed an advanced deep learning-based sequential modeling method to capture the dynamics of health status, which is crucial for accurately predicting the likelihood of readmissions. To further increase the robustness of the learned research model, Microsoft Research developed stand-alone resampling techniques to address the challenge of sample imbalance in this readmission prediction scenario. The study showed that by integrating all of these technologies together, the accuracy of the model improved by over 20 percent. And most importantly, the advanced models were developed using unidentified data that protects patient information.
Give nursing teams the opportunity to help patients when they need it most
“Model precision is crucial here in identifying risky members,” shares Mike Hardbarger, director of computer science at Humana and a contributor to this project’s research. “Our members deserve personal, proactive care. By using this model with others, we can not only help them avoid hospitalization, but care teams can have the necessary data for follow-up with a customized plan. From effective prescription management to food insecurity management, a care manager can then work directly with the member to initiate the next best action.
Proactive problem solving like this depends on collaboration and innovation. Deep learning allowed research teams, including Sean Ma, Lead Data Scientist at Humana, to gain a comprehensive range of both science and industrial considerations. “Working directly with algorithm writers accelerated significantly progress. I’m excited about what’s to come,” says Ma.
Use Microsoft Cloud for Healthcare to do more with your data
This research project is only one step in the development of the Humana analytics engine. Improvements will continue over time as further research is conducted, the model continues to be validated.
Learn more about Microsoft Cloud for Healthcare.