VA Eastern Colorado Geriatric Research, Education, and Clinical Center (GRECC), VA Eastern Colorado Health Care System, Aurora, Colorado, USA.
Physical Therapy Program, Department of Physical Medicine and Rehabilitation, University of Colorado, Aurora, Colorado, USA.
J Am Med Inform Assoc. 2022 Oct 7;29(11):1899-1907. doi: 10.1093/jamia/ocac123.
Prediction models can be useful tools for monitoring patient status and personalizing treatment in health care. The goal of this study was to compare the relative strengths and weaknesses of 2 different approaches for predicting functional recovery after knee arthroplasty: a neighbors-based "people-like-me" (PLM) approach and a linear mixed model (LMM) approach.
We used 2 distinct datasets to train and then test PLM and LMM prediction approaches for functional recovery following knee arthroplasty. We used the Timed Up and Go (TUG)-a common test of mobility-to operationalize physical function. Both approaches used patient characteristics and baseline postoperative TUG values to predict TUG recovery from days 1-425 following surgery. We then compared the accuracy and precision of PLM and LMM predictions.
A total of 317 patient records with 1379 TUG observations were used to train PLM and LMM approaches, and 456 patient records with 1244 TUG observations were used to test the predictions. The approaches performed similarly in terms of mean squared error and bias, but the PLM approach provided more accurate and precise estimates of prediction uncertainty.
Overall, the PLM approach more accurately and precisely predicted TUG recovery following knee arthroplasty. These results suggest PLM predictions may be more clinically useful for monitoring recovery and personalizing care following knee arthroplasty. However, clinicians and organizations seeking to use predictions in practice should consider additional factors (eg, resource requirements) when selecting a prediction approach.
预测模型可以成为医疗保健中监测患者状况和个性化治疗的有用工具。本研究的目的是比较两种不同方法预测膝关节置换术后功能恢复的相对优势和劣势：基于邻居的“像我这样的人”（PLM）方法和线性混合模型（LMM）方法。
我们使用两个不同的数据集来训练和测试膝关节置换术后功能恢复的 PLM 和 LMM 预测方法。我们使用常见的移动性测试 Timed Up and Go（TUG）来操作身体功能。这两种方法都使用患者特征和术后基线 TUG 值来预测手术第 1 至 425 天 TUG 的恢复情况。然后，我们比较了 PLM 和 LMM 预测的准确性和精度。
共使用 317 份患者记录（1379 次 TUG 观察）来训练 PLM 和 LMM 方法，使用 456 份患者记录（1244 次 TUG 观察）来测试预测。这两种方法在均方误差和偏差方面表现相似，但 PLM 方法提供了更准确和精确的预测不确定性估计。
总体而言，PLM 方法更准确、更精确地预测了膝关节置换术后 TUG 的恢复情况。这些结果表明，PLM 预测可能更有助于在膝关节置换术后监测恢复情况和个性化护理。但是，临床医生和组织在选择预测方法时，应考虑其他因素（例如资源要求）。
