Geographic Data Analytics Identifies Health System Leakage Patterns

Health systems with incomplete patient records can use geographic data to identify patients who may be receiving care at external organizations, which could improve care management and reduce system leakage, a study published in JMIR Medical Informatics revealed.

Healthcare organizations frequently have incomplete data on their patients, as patients often seek care from providers in other health systems for specific needs. More than half of hospital encounters came from patients who also had encounters at other hospitals, existing research has shown.

Incomplete patient data can significantly hinder predictive modeling and population health management. Ideally, health systems should be able to use predictive modeling approaches to identify future high-cost patients for care management, which can prevent excess spending and improve care quality. Organizations that don’t have complete patient records may not be able to detect these patients, resulting in increased costs and poor patient outcomes.

In order to ensure that providers have complete visibility into patient activities across disparate systems, researchers examined data from the University of Washington Medicine (UWM), an organization that treats many patients referred from other health systems. Less than one-third of hospital encounters for all UWM patients occur within UWM.

The group used a geographic constraint to identify a large subset of patients who tend to receive care from UWM. The results showed that the best predictor of patients who will seek the majority of their care from UWM are whether patients have a primary care physician at UWM and if they live within five miles of at least one UWM hospital.

Researchers found that about 16.01 percent of UWM patients satisfied these conditions. Of these patients, about 69.38 percent had inpatient stays or emergency department visits that occurred within UWM over the following six months, more than double the corresponding percentage for all UWM patients.

The results demonstrate that using geographic data is a viable method for identifying likely patients in organizations that may have incomplete patient records.

For a data analysis task requiring relatively complete data, such as predictive modeling using historical data, we can conduct the task on this subset of patients and obtain useful results, even if conducting the task on all UWM patients is impractical,€ the team wrote.

For example, we can build a predictive model to identify future high-cost patients among this subset. Enrolling such patients in care management can help prevent high costs and improve outcomes.€

The study also showed that patients who lived farther from UWM tended to receive less of their care from UWM. This suggests that organizations should use different care management strategies for patients living at different distances from healthcare facilities. If a patient will only receive a small portion of care from a specific organization, then that organization can adjust its care management interventions to be more cost-effective.

Additionally, the group said that this method could apply to multiple health systems that exchange their data with each other. Researchers could use the method to identify patients that are likely to receive most of their care from these organizations, enabling data analytics tasks across numerous health systems.

The study has several limitations, including the fact that the team tested their method in only one urban health system. To understand how the approach works in different settings, the researchers said they will need to conduct analyses at organizations serving rural populations, where patients are often more scattered.

Still, the researchers are confident that this method could help other organizations generate actionable insights from their patient information.

Our method opens the door for conducting several major analysis tasks on incomplete medical data, which were previously deemed impractical to undertake.

Published by: