The Use of Machine Learning to Predict Late Arrivals at the Adult Outpatient Department

Introduction: Patient unpunctuality leads to delays in the delivery of care and increased waiting times, resulting in crowdedness. Late arrivals for adult outpatient appointments are a challenge for healthcare, contributing to negative effects on the efficiency of health services as well as wasted time, budget, and resources. This study aims to identify factors and characteristics associated with tardy arrivals at adult outpatient appointments using machine learning and artificial intelligence. The goal is to create a predictive model using machine learning models capable of predicting adult patients arriving late to their appointments. This would support effective and accurate decision-making in scheduling systems, leading to better utilization and optimization of healthcare resources. Methods: A retrospective cohort review of adult outpatient appointments between January 1, 2019, and December 31, 2019, was undertaken at a tertiary hospital in Riyadh. Four machine learning models were used to identify the best prediction model that could predict late-arriving patients based on multiple factors. Results: A total of 1,089,943 appointments for 342,974 patients were conducted. There were 128,121 visits (11.7%) categorized as late arrivals. The best prediction model was Random Forest, with a high accuracy of 94.88%, a recall of 99.72%, and a precision of 90.92%. The other models showed different results, such as XGBoost with an accuracy of 68.13%, Logistic Regression with an accuracy of 56.23%, and GBoosting with an accuracy of 68.24%. Conclusion: This paper aims to identify the factors associated with late-arriving patients and improve resource utilization and care delivery. Despite the overall good performance of the machine learning models developed in this study, not all variables and factors included contribute significantly to the algorithms' performance. Considering additional variables could improve machine learning performance outcomes, further enhancing the practical application of the predictive model in healthcare settings.