全髋关节置换术后机器学习算法的开发与验证:用于住院时间和支付模型
算法所需变量为年龄、种族、性别和共病评分,预测住院时间和费用的受试者工作特征曲线下面积分别为0.87和0.71;随着患者复杂程度增加,费用预测误差在中等、主要和极端共病组分别为3%、12%和32%。
要点
- 机器学习算法仅需年龄、种族、性别和共病评分即可预测住院时间和费用。
- 预测住院时间和费用的受试者工作特征曲线下面积分别为0.87和0.71。
- 费用预测误差随共病复杂程度增加而增大,中等、主要、极端共病组分别为3%、12%、32%。
核心结论
- 初步机器学习算法在初次全髋关节置换术前预测住院时间和费用方面表现出良好的结构效度、信度和反应度。
- 该算法有望在择期全髋关节置换术前实现基于风险的患者特异性支付模型,提供与病例复杂程度相称的分层报销。
研究设计与样本
- 采用受试者工作特征曲线下面积和百分比准确率评估模型性能。
- 住院费用分组为低于12000美元、12000至24000美元和高于24000美元;住院时间分组为1至2天、3至5天和6天及以上。
主要发现
- 机器学习算法需要年龄、种族、性别和共病评分(「疾病风险」和「发病风险」)即可展现良好效度、信度和反应度。
- 预测住院时间和费用的受试者工作特征曲线下面积分别为0.87和0.71。
- 随着患者复杂程度增加,费用预测误差在中等、主要和极端共病组分别增加3%、12%和32%。
关键数值与推荐
- 住院费用分组阈值:低于12000美元、12000至24000美元、高于24000美元。
- 住院时间分组阈值:1至2天、3至5天、6天及以上。
- 预测住院时间和费用的受试者工作特征曲线下面积分别为0.87和0.71。
- 费用预测误差在中等、主要和极端共病组分别为3%、12%和32%。
临床怎么用
- 原文未据此给出临床推荐。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: Value-based payment programs in orthopedics, specifically primary total hip arthroplasty (THA), present opportunities to apply forecasting machine learning techniques to adjust payment models to a specific patient or population. The objective of this study is to (1) develop and validate a machine learning algorithm using preoperative big data to predict length of stay (LOS) and patient-specific inpatient payments after primary THA and (2) propose a risk-adjusted patient-specific payment model (PSPM) that considers patient comorbidity. METHODS: Using an administrative database, we applied 122,334 patients undergoing primary THA for osteoarthritis between 2012 and 16 to a naïve Bayesian model trained to forecast LOS and payments. Performance was determined using area under the receiver operating characteristic curve and percent accuracy. Inpatient payments were grouped as <$12,000, $12,000-$24,000, and >$24,000. LOS was grouped as 1-2, 3-5, and 6+ days. Payment model uncertainty was applied to a proposed risk-based PSPM. RESULTS: The machine learning algorithm required age, race, gender, and comorbidity scores ("risk of illness" and "risk of morbidity") to demonstrate excellent validity, reliability, and responsiveness with an area under the receiver operating characteristic curve of 0.87 and 0.71 for LOS and payment. As patient complexity increased, error for predicting payment increased in tiers of 3%, 12%, and 32% for moderate, major, and extreme comorbidities, respectively. CONCLUSION: Our preliminary machine learning algorithm demonstrated excellent construct validity, reliability, and responsiveness predicting LOS and payment prior to primary THA. This has the potential to allow for a risk-based PSPM prior to elective THA that offers tiered reimbursement commensurate with case complexity. LEVEL OF EVIDENCE: III.
引用
- Ramkumar PN, Navarro SM, Haeberle HS, et al. Development and Validation of a Machine Learning Algorithm After Primary Total Hip Arthroplasty: Applications to Length of Stay and Payment Models. J Arthroplasty. 2018. PMID: 30665831. doi:10.1016/j.arth.2018.12.030
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。