Home Magazines Editors-in-Chief FAQs Contact Us

Using machine learning for no show prediction in the scheduling of clinical exams


Abstract

For this work actual data of more than 1.000.000 appointments of large laboratories in São Paulo and Rio de Janeiro (Brazil) were used. The study has the first objective to study the profile of no show patients: patients who schedule a clinical examination but do not attend it. The second objective was to find out which variables are relevant to predict patient non-attendance on the day of examination. As a final objective, it is of interest to provide, at the time of the appointment, whether the patient will attend the scheduled procedure or not. In order to achieve these goals, machine learning based prediction models were developed and analyzed, resulting in a clustering of the scheduled patients according to their no show propensity. Clustering was used instead of a binary result in order to allow for different strategies to be defined as the resulting profiles were analyzed. The models were evaluated by their accuracy and the combination of Random Forest and Logistic Regression techniques presented the best results, identifying clusters which had up to 75% accuracy rate in a validation environment.The main discussions follow in two lines: the first one is related to the allocation of the available time, from a patient who will not attend the examination, to one who will. The second discussion is in the understanding of the needs and circumstances that lead to non-attendance in performing the exam.

Keywords

no show, machine learning, prediction models, clinical exams

Testimonials