Predictive modeling is good, but care coordination would be better


In the past week, two organizations in different parts of the country announced efforts to harness information technology to improve predictive modeling. The Heritage Provider Network in southern California is offering $3 million to whoever comes up with the best algorithm for predicting which patients will be hospitalized. And a healthcare system in the Dallas-Fort Worth area is testing a new IT tool that forecasts which patients with congestive heart failure (CHF) are most likely to be readmitted within 30 days of hospital discharge.

In each case, there's an economic motive behind the search for a reliable method of forecasting illness. Heritage is a managed care company that takes financial risk for care, so it wants to be able to manage that risk better. And hospitals are facing new Medicare regulations that will penalize them financially for excessive readmissions, starting in October.

Preliminary trials of the CHF modeling algorithm showed reductions in readmissions of up to 40 percent. But the identification of patients who were most at risk was only one factor in this result. The main reason for the drop in readmissions was that dedicated clinical teams coordinated the selected patients' post-discharge care.

What if, instead just focusing on the 30 percent of CHF patients who have the highest risk of readmission, the healthcare system was designed to provide excellent transitions of care and post-acute care for everyone who has been hospitalized? We'd still need software to identify risk factors for the stratification of care, but we'd also need a lot of other applications and a paradigm shift in healthcare to facilitate population health management.

The same holds true for the idea of predicting which patients are most at risk of being hospitalized. Instead of developing a single technology tool in a vacuum, we could use many specialized, interconnected apps, including health risk assessment tools, registries, personal health records, and remote patient monitoring to better coordinate and manage health so that fewer people with chronic diseases were hospitalized.

Of course, that's the infrastructure that a whole lot of people are engaged in building right now. The important thing to remember, as this edifice rises, is that technology by itself is not a solution (whatever vendors say). The solution lies in how the technology is used.

A recently published evaluation of the TransforMed medical home pilot makes this point painfully clear. In discussing the role of health IT, the authors note: "Many practices implemented disease registries without reconfiguring work processes to use them effectively for population management, such as identifying all patients with asthma and proactively suggesting that they obtain influenza vaccination."

The study's authors also point out that the applications available for this purpose resemble a "jigsaw puzzle" and are not "plug and play." As a result, they suggest, we're still a long way from being able to use them to provide coordinated care.

What all of this indicates is that the best minds in the health IT community should focus on figuring out how to transcend the current limitations of the technology for supporting population health management. When we can do that, we won't need predictive modeling programs. - Ken

Suggested Articles

Payers and providers have made significant investments in digitizing the healthcare system but have yet to see a return on that investment.

Fewer than 4 in 10 health systems can successfully share data with other health systems, which presents a number of challenges.

As telehealth programs continue to expand, it’s crucial to understand how facility management will shift with these advancements in healthcare.