Prediction of Pediatric No-Show Rates Before and During the COVID-19 Pandemic Utilizing Machine Learning.

BackgroundPatient no-shows significantly disrupt pediatric healthcare delivery, highlighting the necessity for precise predictive models, especially during the dynamic shifts caused by the SARS-CoV-2 pandemic. In outpatient settings, these no-shows result in medical resource underutilization, increased healthcare costs, reduced access to care, and lower clinic efficiency and productivity. MethodsThe objective is to develop a predictive model for patient no-shows using data-driven techniques. We analyzed five years of historical data retrieved from both a scheduling system and electronic health records from a general pediatrics clinic within the WVU Health systems. This dataset comprises a total of 209,408 visits from 2015 to 2018, 82,925 visits in 2019, and 58,820 visits in 2020, spanning both pre-pandemic and pandemic periods. The data includes variables such as patient demographics, appointment details, timing, hospital characteristics, appointment types, and environmental factors. ResultsOur XGBoost model demonstrated robust predictive capabilities, notably outperforming traditional "no-show rate" metrics. Precision and recall metrics for all features were 0.82 and 0.88, respectively. Receiver Operator Characteristic (ROC) analysis yielded AUCs of 0.90 for all features and 0.88 for the top 5 predictors when evaluated on the 2019 cohort. Furthermore, model generalization across racial/ethnic groups was also observed. Evaluation on 2020 telehealth data reaffirmed model efficacy (AUC: 0.90), with consistent top predictive features. ConclusionsOur study presents a sophisticated predictive model for pediatric no-show rates, offering insights into nuanced factors influencing attendance behavior. The models adaptability to evolving healthcare landscapes, including telehealth, underscores its potential for enhancing clinical practice and resource allocation.