Machine-learning prediction of 90-day readmission following primary TKA: Insights from 2,123 cases in the Michigan arthroplasty registry

利用机器学习预测初次全膝关节置换术后90天内再入院:来自密歇根州关节置换登记处2123例病例的启示

阅读:1

Abstract

BACKGROUND: Despite peri-operative advances, unplanned 90-day readmission after total knee arthroplasty (TKA) remains a costly quality metric. We leveraged statewide registry data and machine-learning analytics to quantify contemporary risk and uncover modifiable drivers. METHODS: We retrospectively reviewed all primary TKAs performed between 2016 and 2024 at a MARCQI-participating academic hospital. Demographics, comorbidities, peri-operative factors, and medication use were compared between patients with and without 90-day readmission. A multilayer perceptron neural network (MPNN) incorporating 10 routinely collected variables (age, sex, race, marital status, length of stay [LOS], smoking, alcohol, bleeding disorder, prior DVT/PE, ASA class) was trained (70 %) and validated (30 %) to predict readmission; model performance was assessed by accuracy, AUC, and variable importance. RESULTS: Among 2123 TKAs, 86 patients (4.1 %) were readmitted. Compared with non-readmitted patients, those readmitted had higher ASA scores (2.92 ± 0.49 vs 2.72 ± 0.48), longer LOS (2.97 ± 1.43 vs 2.32 ± 1.30 days), more prior DVT/PE (11.9 % vs 5.8 %), and greater pre-operative opioid use (5.0 % vs 3.2 %) (all p ≤ .036). LOS (ρ = 0.10) ranked highest on correlation analysis, while home/self-care discharge correlated negatively (ρ = -0.06). The MPNN achieved 96.0 % accuracy and AUC 0.704; LOS was the strongest predictor (importance 100 %), followed by ASA class, race, prior DVT/PE, and bleeding disorders. CONCLUSIONS: Although uncommon, 90-day readmission after primary TKA is chiefly associated with prolonged LOS, elevated ASA status, prior thromboembolism, opioid exposure, and non-home discharge. Targeted enhancements -- aggressive VTE prophylaxis, opioid-sparing analgesia, and streamlined discharge pathways promoting safe home return -- may further curb readmissions and related costs.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。