Testing Multiple Predictive Models to Reduce Patient No-Show Rates in a Safety Net Healthcare Facility

Research Category: Engineering and Technology
Presenter: Demetri Lemonias
Additional Authors: Dr. James Benneyan
PI: Demetri Lemonias, James Benneyan

Patients who are of high risk of health complications occasionally miss appointments important to improving their health. This project sought to reduce the no-show rate among patients that are likely to no-show, and as a result worsen their health. To accomplish this, predictive models were created to determine which patients are most likely to no-show. Four predictive modeling techniques were used and compared, with the Adaptive Boosting model resulting in the most accurate with an Area under ROC curve of 0.83 and an accuracy rating of 84.01%. Using this model, the patient cohort over the five-week period was developed resulting in a population size of 423 patients (23.8% of total population). In conjunction with creating the no-show model, the hospital staff created a morbidity rating system to determine which patients are at risk of health complications or lower quality of care metrics due to missed appointments. Specific interventions focusing on fostering a positive relationship between patients and care givers were used to reduce the patient cohort no-show rate within a five-week long test pilot. The cohort had a historical no-show rate of 29.7% which was reduced to 27.6. A Chi Squared goodness of fit test proved this to be statistically insignificant due to the small sample size.

Conclusions: The 2.1% no-show reduction is promising and reinforces the idea that creating a positive relationship between the hospital and patient is crucial to reducing the no show rate. It also illustrates the need for more time and patients to statistically prove effectiveness.