Potentially preventable readmissions are a scourge on the US healthcare system.
Each year millions of patients are discharged from the hospital, only to return within 30, 60, or 90 days.
Not only do patients, their families, and their caregivers suffer as a result, but hospitals, insurers, and the government waste billions of dollars that could be spent on other public health priorities. Many if not most of these readmissions could have been avoided if clinicians had effective, scalable, and timely methods for identifying not only which patients were the highest risk, but what steps should have been taken to mitigate that risk.
In recent years there has been a proliferation of readmission risk assessment models, yet readmission rates have barely budged. Fundamental flaws exist in most approaches in the areas of Data, Model Adaptability and Clinical Workflow Integration.
Many tools rely solely on historical patient data mined from the EHR or are disease-specific models that cannot be scaled to address all readmissions challenges. Models that rely on data collected at discharge are not timely enough to enable clinicians to take meaningful action, and ones that are not well-integrated into clinical workflow are not easily adopted.
For a readmission risk assessment tool to achieve a meaningful and long-lasting impact, these common pitfalls must be avoided at all costs. Today, I’m going to address some of the many data challenges faced when trying to risk assess patients.
Historical Data does not Predict Future Readmissions
Anybody who has ever invested in the stock market, rooted for a local sports team, or stuck with a television show past its tenth season knows that past performance gives you no guarantee on future returns. Factors beyond our control and beyond our ability to predict may cause our fortunes to turn on a dime.
Consider the Dow Jones Industrial Average: Those who had any investments around July of 2007 remember the feelings of unabashed optimism and certainty inspired by the great bull run of the early 2000s. Unfortunately, those same investors also most assuredly remember what happened shortly thereafter, when the financial crisis of 2008 erased trillions of dollars’ worth of wealth.
A recent systematic review of readmission risk models concluded that many hospitals still model their approach to identifying high-risk patients based on historical admissions, claims data, and outdated information on patient populations [1].
Using these old data to model and predict readmissions is dangerous. And with increasing pressure on hospitals to reduce readmissions, this approach also runs the risk of becoming extremely costly. Just ask the guy who splurged on Brooklyn Dodgers tickets in 1958, or the guy who put all his money into 8-track cassettes in 1979, or the guy who started a Hummer dealership in 2005.
Any of these folks will tell you that past performance data can not only betray you, but it may also prevent you from recognizing the obsolescence of your sources. As a result, this data may cost you a fortune.