Late-arriving patients have become a prominent concern in several ambulatory care clinics across the globe. Accommodating them could lead to detrimental ramifications such as schedule disruption and increased waiting time for forthcoming patients, which, in turn, could lead to patient dissatisfaction, reduced care quality, and physician burnout. However, rescheduling late arrivals could delay access to care. This paper aims to predict the patient-specific risk of late arrival using machine learning (ML) models. Data from two different ambulatory care facilities are extracted, and a comprehensive list of predictor variables is identified or derived from the electronic medical records. A comparative analysis of four ML algorithms (logistic regression, random forests, gradient boosting machine, and artificial neural networks) that differ in their training mechanism is conducted. The results indicate that ML algorithms can accurately predict patient lateness, but a single model cannot perform best with respect to predictive performance, training time, and interpretability. Prior history of late arrivals, age, and afternoon appointments are identified as critical predictors by all the models. The ML-based approach presented in this research can serve as a decision support tool and could be integrated into the appointment system for effectively managing and mitigating tardy arrivals.

中文翻译：

---

一种基于机器学习的方法来预测门诊护理中心的患者守时。

迟到患者已经成为全球数家门诊诊所中的主要关注点。容纳它们可能导致有害的后果，例如日程安排中断和增加对即将到来的患者的等待时间，进而可能导致患者不满意，护理质量下降和医生精疲力尽。但是，将晚到的时间重新安排可能会延迟获得护理的时间。本文旨在使用机器学习（ML）模型来预测特定于患者的迟到风险。提取来自两个不同的门诊医疗机构的数据，并从电子病历中识别出预测变量的完整列表。四种ML算法（逻辑回归，随机森林，梯度提升机，和人工神经网络），它们的训练机制有所不同。结果表明，机器学习算法可以准确地预测患者的迟到情况，但是就预测性能，训练时间和可解释性而言，单个模型无法达到最佳效果。所有模型都将迟到，年龄和下午约会的先前历史确定为关键预测因素。本研究中提出的基于ML的方法可以用作决策支持工具，并且可以集成到约会系统中，以有效地管理和缓解迟到的到达。所有模型都将下午和约会指定为关键预测因素。本研究中提出的基于ML的方法可以用作决策支持工具，并且可以集成到约会系统中，以有效地管理和缓解迟到的到达。所有模型都将下午和约会指定为关键预测因素。本研究中提出的基于ML的方法可以用作决策支持工具，并且可以集成到约会系统中，以有效地管理和缓解迟到的到达。