Staub Lukas P, Aghayev Emin, Skrivankova Veronika, Lord Sarah J, Haschtmann Daniel, Mannion Anne F
NHMRC Clinical Trials Centre, The University of Sydney, Locked Bag 77, Camperdown, Sydney, NSW, 1450, Australia.
Schulthess Klinik, Spine Center, Zurich, Switzerland.
Eur Spine J. 2020 Jul;29(7):1742-1751. doi: 10.1007/s00586-020-06351-5. Epub 2020 Feb 27.
Surgeons need tools to provide individualised estimates of surgical outcomes and the uncertainty surrounding these, to convey realistic expectations to the patient. This study developed and validated prognostic models for patients undergoing surgical treatment of lumbar disc herniation, to predict outcomes 1 year after surgery, and implemented these models in an online prediction tool.
Using the data of 1244 patients from a large spine unit, LASSO and linear regression models were fitted with 90% upper prediction limits, to predict scores on the Core Outcome Measures Index, and back and leg pain. Candidate predictors included sociodemographic factors, baseline symptoms, medical history, and surgeon characteristics. Temporal validation was conducted on 364 more recent patients at the same unit, by examining the proportion of observed outcomes exceeding the threshold of the 90% upper prediction limit (UPL), and by calculating mean bias and other calibration measures.
Poorer outcome was predicted by obesity, previous spine surgery, and having basic obligatory (rather than private) insurance. In the validation data, fewer than 12% of outcomes were above the 90% UPL. Calibration plots for the model validation showed values for mean bias < 0.5 score points and regression slopes close to 1.
While the model accuracy was good overall, the prediction intervals indicated considerable predictive uncertainty on the individual level. Implementation studies will assess the clinical usefulness of the online tool. Updating the models with additional predictors may improve the accuracy and precision of outcome predictions. These slides can be retrieved under Electronic Supplementary Material.
外科医生需要工具来提供手术结果的个性化评估以及围绕这些结果的不确定性，以便向患者传达现实的预期。本研究针对接受腰椎间盘突出症手术治疗的患者开发并验证了预后模型，以预测术后1年的结果，并将这些模型应用于在线预测工具中。
使用来自一个大型脊柱科室的1244例患者的数据，采用套索回归和线性回归模型，并设定90%的预测上限，以预测核心结局指标指数以及背部和腿部疼痛的评分。候选预测因素包括社会人口统计学因素、基线症状、病史和外科医生特征。通过检查观察到的结果超过90%预测上限（UPL）阈值的比例，并计算平均偏差和其他校准指标，对同一科室另外364例近期患者进行了时间验证。
肥胖、既往脊柱手术史以及拥有基本强制保险（而非私人保险）预示着预后较差。在验证数据中，不到12%的结果高于90%UPL。模型验证的校准图显示平均偏差值<0.5分，回归斜率接近1。
虽然模型总体准确性良好，但预测区间表明在个体水平上存在相当大的预测不确定性。实施研究将评估在线工具的临床实用性。使用额外的预测因素更新模型可能会提高结果预测的准确性和精确性。这些幻灯片可在电子补充材料中获取。
