Calidad en Atención Primaria Acceso abierto

Abstracto

Machine Learning Prediction of Diabetic Patients Follow-Up Patterns at Family Medicine Clinic of Tertiary Health Care Provider, Riyadh, Saudi Arabia

Aleid M, Alfantoukh L, Alhuqail N, Abomelha F and Rayes Z

Background: Diabetes prevalence is predicted to rise dramatically over the next 20 years, and associated spending is expected to increase threefold. We hypothesized that adding appointment data and International Classification of Diseases (ICD) codes to patient demographic data would improve predictions of follow-up appointment attendance utilizing machine learning models. Our results showed that the random forest classifier was the most accurate and sensitive, reaching 73% and 77%, respectively.

Methods: This study was based on retrospectively extracted patient’s records of King Faisal Specialist Hospital & Research Centre patients diagnosed with diabetes mellitus type I or II who had a follow-up appointment at the Family Medicine Clinic between January 1, 2014, and December 31, 2018. We built several machine learning models, including logistic regression, decision tree, random forest, k-nearest neighbors (KNN), and support vector machine (SVM) models. We also implemented a deep learning algorithm, Deep Neural Network (DNN).

Results: A total of 2,403 patients participated in the study; 3 were excluded because they had only one appointment. Of the 2,400 remaining, around 50% were female, 32% were hospital employees, and 82% were married. Non-Saudis represented around 25% of participants. A total of 19,218 appointments were analyzed, 44.33% of which were classified as “no- show.” Prediction accuracy increased by an average of 7% and 10% when we added appointment data and ICD codes, respectively, to demographic data.

Conclusions: Our results indicate that knowing appointment-related data and ICD codes corresponding to a patient along with their demographic data is useful in predicting follow-up status.