Using machine learning to reduce waste in post-acute care

Share this content:
Michael Cantor, M.D
Michael Cantor, M.D

Until recently, efforts to more efficiently use post-acute care resources were managed by trial and error, rather than data-driven insights into identifying patients at high risk of readmission and collaborating with PAC facilities based on evidence of excellent performance.  

The increasing use of machine learning is changing the way health plans and provider systems manage hospital discharge and select PAC facility partners to provide skilled nursing facility, inpatient rehabilitation facility, long-term acute care hospitals and home health. This is good news for patients, and for our healthcare system, which needs to reduce waste. It is also an opportunity for PAC facilities to evaluate their care processes and use data to improve care, or risk losing patient volume to more efficient, higher-performing facilities.

What is machine learning, and how is it being used to reduce waste in PAC? Machine learning is defined as the ability of a machine to get better at a task over time by improving how it assesses data to achieve a goal. That's a long-winded way of saying that intelligent machines can change how they evaluate a problem and get better at making decisions without additional programming. Through an iterative process, the algorithm the machine uses evolves to provide better solutions to the problem being addressed.  

Machine learning algorithms are now being used to improve PAC by using data to determine the optimal path of care for patients after hospital discharge, and to define the PAC network that can deliver the best results. Machine learning-based discharge planning programs take in a wide variety of data, including claims data, electronic health record data, demographics, and consumer data (e.g., credit scores, zip-code specific financial information) and run the data through algorithms to identify the best path of care for the patient after discharge, including whether a PAC facility stay is needed, which facility has the best performance and is best to meet the needs of that patient, and/or whether home health services are needed.

Consider the example of an 81-year-old woman admitted to the hospital for a stroke with symptoms of speech problems and right-sided weakness. By her third day of hospitalization, her symptoms have almost resolved, and she is ready to leave – where should she go to get ongoing rehabilitation and recover? She also has atrial fibrillation (irregular heart beat), mild heart failure, diabetes, breast cancer in remission, and mild cognitive impairment, and lives alone in a condominium in an affluent suburb of Chicago. She takes nine medications each day, including the powerful blood thinner warfarin (Coumadin).  

In most hospitals, this patient and her family would be given a list of SNFs and told to pick one, with little or no information about the facilities other than location and STARs (quality) rating. Some hospitals are already using discharge planning software that incorporates machine learning to improve this process.  The software takes key data points known about this patient, including her medical history, nine medications, admitting diagnosis, past claims and costs for hospital, emergency room and outpatient care, as well as demographic and zip-code data to determines the best path of care for her based on its analysis of what has happened to similar patients in the past, and costs and performance (readmission rates, average cost/episode, STAR rating) for SNFs and home health agencies in her area. 

The software generates a short list of SNFs and home health agencies for the patient and her family to choose from, thus providing guidance based on data, which is more accurate and helpful than the tools available in most hospitals today. The machine learns by analyzing what happens to this patient (and many others) as she recovers, so that it can refine its recommendations for similar patients in the future.

Similar machine learning software is being used to shape the network of PAC facilities for health plans and provider organizations. The software can analyze utilization patterns and costs of PAC facilities and home health agencies in a specific geographic area, using the data unique to its population to understand performance in a more granular way, and incorporate multiple variables to determine how to narrow the network so that the wasteful expenditures due to readmissions, prolonged stays in SNF, and unnecessary home visits can be reduced by selecting higher-performing facilities and home health agencies. Instead of waiting for annual STAR ratings, the network analysis software can take in data more frequently and use machine learning to get better at identifying the best PAC partners. The software can also be used to generate reports that health plans and provider organizations can share with their PAC facility and home health agency partners so that they can identify opportunities for improvement and enhance their performance.

Machine learning-enabled software to help with creating better discharge paths and improve PAC networks can improve patient experience, lead to better outcomes, and reduce waste and costs. It will also require PAC facilities and home health agencies to rely on performance to gain referrals and market share. PAC facilities can benefit from this by creating collaborative relationships with health plans and provider organizations to get current, accurate data about their performance and use that data to enhance performance and grow market share.

Michael Cantor, M.D, JD, is Chief Medical Officer at CareCentrix.

Featured Listings

Anacapa Technologies, Inc.

Anacapa Technologies, Inc.

Anacapa Technologies, Inc. is a veteran-owned, federally certified small business based in San ...

Apollo Corporation

Apollo Corporation

The Shield™ Splash Guard is the most efficient and economical product available to ...

Bulk TV Internet

Bulk TV  Internet

Bulk TV & Internet is the leading provider of DIRECTV television solutions designed ...