Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland School of Medicine, Baltimore, MD, USA.
Department of Radiology, University of Alabama at Birmingham, Birmingham, AL, USA.
J Digit Imaging. 2022 Dec;35(6):1690-1693. doi: 10.1007/s10278-022-00670-3. Epub 2022 Jun 29.
The term "no-show" refers to scheduled appointments that a patient misses, or for which she arrives too late to utilize medical resources. Accurately predicting no-shows creates opportunities to intervene, ensuring that patients receive needed medical resources. A machine-learning (ML) model can accurately identify individuals at high no-show risk, to facilitate strategic and targeted interventions. We used 4,546,104 non-same-day scheduled appointments in our medical system from 1/1/2017 through 1/1/2020 for training data, including 631,386 no-shows. We applied eight ML techniques, which yielded cross-validation AUCs of 0.77-0.93. We then prospectively tested the best performing model, Gradient Boosted Regression Trees, over a 6-week period at a single outpatient location. We observed 123 no-shows. The model accurately identified likely no-show patients retrospectively (AUC 0.93) and prospectively (AUC 0.73, p < 0.0005). Individuals in the highest-risk category were three times more likely to no-show than the average of all other patients. No-show prediction modeling based on machine learning has the potential to identify patients for targeted interventions to improve their access to medical resources, reduce waste in the medical system and improve overall operational efficiency. Caution is advised, due to the potential for bias to decrease the quality of service for patients based on race, zip code, and gender.
“失约”是指患者错过或到达太晚而无法使用医疗资源的预约。准确预测失约情况为干预提供了机会，确保患者获得所需的医疗资源。机器学习 (ML) 模型可以准确识别高失约风险的个体，从而促进有针对性的干预措施。我们使用了我们医疗系统中 2017 年 1 月 1 日至 2020 年 1 月 1 日的 4546104 次非同日预约数据进行训练，其中包括 631386 次失约。我们应用了八种 ML 技术，其交叉验证 AUC 为 0.77-0.93。然后，我们在单个门诊地点对表现最佳的模型（梯度提升回归树）进行了为期 6 周的前瞻性测试。我们观察到 123 次失约。该模型在回顾性（AUC 0.93）和前瞻性（AUC 0.73，p<0.0005）方面准确地识别出可能失约的患者。高风险类别中的个体失约的可能性是所有其他患者平均水平的三倍。基于机器学习的失约预测模型有可能识别出需要针对性干预的患者，以改善他们获得医疗资源的机会，减少医疗系统中的浪费并提高整体运营效率。需要注意的是，由于种族、邮政编码和性别等因素导致服务质量下降的偏见风险。
