Hey! When was your last visit to the emergency department (ED)? Hopefully, your answer is “never”. While most of us rarely visit the ED, a small portion of patients actually visit the ED more than three times a year. Since ED services usually cost more than other types of care, frequent ED users create a strain on the healthcare system. This means healthcare providers are challenged with finding ways to identify these heavy ED users and then change their behavior using case management (CM).
Researchers at CHIDS (Xia Hu, Sean Barnes, Margrét Bjarnadóttir, and Bruce Golden) recently developed machine learning algorithms to better classify heavy ED users. Their work, forthcoming at IISE Transactions on Healthcare Systems Engineering, focuses on predicting two types of users:
1) Among people who don’t currently use ED services frequently (4+ times a year), who is going to become a frequent user? These are the “Jumpers”.
2) Among people who already use ED services a lot, who is going to continue this pattern? These are the “Repeaters”.
Analyzing a set of insurance claims for Medicaid, their supervised machine learning framework was able to answer these questions using a wide range of information about patients (e.g., gender, use of mental health services). Compared to models that only use information about a patient’s current ED usage, their machine learning approach was much better at predicting frequent ED usage in the future.
So, should we focus on the “Jumpers” or the “Repeaters”? The answer is not so straightforward. Based on evidence from psychology, the researchers argue that “jumpers”, who don’t yet have the habit of relying on ED services, might be more responsive than “repeaters” to case management interventions. However, the researchers also note a trade-off in enrolling “jumpers” because “repeaters” are more accurately identified based on medical history. Reliance on enrolling too many predicted “jumpers” might lead to spending CM resources on patients who would never become frequent ED users in the first place.
In the end, the researchers found that enrolling the optimal mix of “jumpers” and “repeaters” is key to maximizing the cost-effectiveness of typical CM interventions. In sum, machine learning methods can help providers can leverage their patient data to improve their selection of patients who are most likely to reduce their reliance on ED services in response to CM programs.
Source: Hu, X., Barnes, S., Bjarnadóttir, M., & Golden, B. (2017). Intelligent Selection of Frequent Emergency Department Patients for Case Management: A Machine Learning Framework Based on Claims Data. IISE Transactions on Healthcare Systems Engineering, 0–0. http://doi.org/10.1080/24725579.2017.1351502