Study: Precision care management could reduce severe respiratory infection admissions

A recent review of Medicare claims suggests a machine learning algorithm could flag individuals at higher risk of infection or help prevent outbreaks within skilled nursing facilities.
By Dave Muoio
04:15 pm

A new investigation of roughly two million Medicare beneficiary claims suggests that machine learning (ML) could be used to predict which patients are at greatest risk of severe respiratory infections, and to place them in more appropriate care facilities in their area.

The study was published online in the American Journal of Managed Care and funded by the HEALTH[at]SCALE Corporation, a precision care technology company that develops the algorithm in question.

Based on their results, the researchers (who also have financial ties to HEALTH[at]SCALE) suggested that with further investigation this approach could potentially be applicable to the ongoing COVID-19 pandemic.


Within a community setting, the ML algorithm identified the top 1% of the cohort estimated to have the greatest risk of an admission for severe respiratory infection within the next 90 days. Comparing the claims to actual outcomes, researchers saw a 13-fold increase in risk for an emergency department visit, an 18-fold increase in risk for hospitalization and a 15-fold increase in risk for either outcome.

For a second analysis conducted within a post-acute setting, the algorithm selected recommendations for nearby skilled nursing facilities. Comparing those who received care at one of the top three recommended facilities to those who did not, the researchers reported a 37% and 36% relative reduction in severe respiratory emergency department visits and inpatient hospitalizations, respectively.

While the researchers stressed that use of the algorithm for COVID-19 is hypothetical without more targeted research, they also noted a handful of shortcomings in this more general exploration of severe respiratory outcomes. These included the use of administrative claims data, the differences between Medicare beneficiaries versus the general population, no controls for community disease-prevalence or other socioeconomic factors, and the inherent limitations of retrospective study designs versus observational causal studies or randomized clinical trials.


The researchers collected administrative claims from community Medicare beneficiaries between 2017 and 2019. Claims from the first two years were used to train the algorithm, while data from 2019 were only used to evaluate its performance. For the second analysis, a similar training and evaluation process was repeated for beneficiaries receiving post-acute care at skilled nursing facilities.


Much of the impetus for this exploration of severe respiratory disease management was the country's ongoing COVID-19 outbreak, the researchers wrote. Total cases in the U.S. recently passed the 5 million mark, and poorer outcomes have been reported among patients already at risk for, or with existing, severe respiratory disease.

With this in mind, the researchers note that ML-based precision management could be a feasible way to reduce the strain on healthcare resources across the population once there is greater consensus regarding which patients are most likely to experience severe sequelae from acquiring COVID-19.

Across the community setting, those identified could receive targeted programs such as social distancing guideline reinforcement, symptom education, home medication delivery and more, they wrote. In post-acute settings, those identified could receive actionable facility recommendations from clinical decision support tools using this type of approach.

More regional news

Person in a telemedicine visit at home

(Photo credit: Alistair Berg/Getty Images) 

Piggy Bank

(Photo Credit: Kwanchai Lettanpunyaporn/Getty Images) 



The latest news in digital health delivered daily to your inbox.

Thank you for subscribing!
Error! Something went wrong!