Skip to main content

We’ve updated our Terms & Conditions and Privacy Policy. By using this site, you agree to these terms.

The problem:

Patient no-shows and late cancellations are common problems that adversely affect healthcare organizations’ financial performance and quality of service and can significantly limit access for other patients seeking care. 

The goal:

The goal of this research project was to create an innovative, high-performing machine learning model to predict no-shows and late cancellations for outpatient neurology appointments.

How we got there:

To mitigate the adverse effects of no-shows and late cancellations, many clinics implement an overbooking strategy. Overbooking is usually understood as booking two patients for the same appointment time, where one of the patients is preliminarily identified as high risk for no-show. In order to assess such risk, an accurate prediction of the no-shows is essential. An inaccurate overbooking approach can create other issues, such as scheduling collision, which leads to increased patient wait times and overtime costs. 

These issues motivated us to propose a model that can predict the probability of patient no-shows and late cancellations for neurology appointments. This department was chosen because there are a limited number of neurologists available to meet an ever-growing demand, and the field suffers from high rates of no-shows and late cancellations.


The outcome:

The Steele Institute’s Healthcare Re-Engineering team, along with their PhD student intern, developed an approach for feature selection methodologies by utilizing genetic algorithm (GA) and non-dominated sorting genetic algorithm II (NSGA-II) as the search strategies. Results indicated that the developed models are highly comparable to that of well-known methods such as recursive feature elimination.

We contributed to the literature by designing a stacking model, in which the base models are constructed based on different subsets of features for a given predictive technique. The results showed that the stacking model has a better performance than each individual base classifier.

We then designed a hierarchical process to weigh and rank the importance of each feature. We searched among an extensive number of features and introduced new ones that can significantly contribute to the prediction of no-shows and late cancellations for neurology. Our results showed that lead time, appointment type and department history of no-shows over the six months before the appointment date are among the leading factors for predicting patient no-shows and late

The proposed model could potentially be used to assist schedulers with decisions about overbooking to decrease the risk of scheduling collision. Other interventions can also be incorporated such as sending multiple reminders and making follow-up phone calls only to high-risk patients instead of reminding all patients, which has been shown in the literature to be an ineffective strategy.

Read the full article:

Machine learning case study predicts no-shows and late cancellations