Department of Anesthesiology and Pain Medicine, University of Ottawa, Ottawa, Canada.
ICES-Ottawa, Ottawa, ON, Canada; Clinical Epidemiology Program, Ottawa Hospital Research Institute, Ottawa, ON, Canada; Department of Medicine, University of Ottawa, Ottawa, ON, Canada.
Br J Anaesth. 2022 Jul;129(1):33-40. doi: 10.1016/j.bja.2022.04.007. Epub 2022 May 18.
Older people (≥65 yr) are at increased risk of morbidity and mortality after emergency general surgery. Risk prediction models are needed to guide decision making in this high-risk population. Existing models have substantial limitations and lack external validation, potentially limiting their applicability in clinical use. We aimed to derive and validate, both internally and externally, a multivariable model to predict 30-day mortality risk in older patients undergoing emergency general surgery.
After protocol publication, we used the National Surgical Quality Improvement Program (NSQIP) database (2012-6; estimated to contain 90% data from the USA and 10% from Canada) to derive and internally validate a model to predict 30-day mortality for older people having emergency general surgery using logistic regression with elastic net regularisation. Internal validation was done with 10-fold cross-validation. External validation was done using a temporally separate health administrative database exclusively from Ontario, Canada.
Overall, 6012 (12.0%) of the 50 221 patients died within 30 days. The model demonstrated strong discrimination (area under the curve [AUC]=0.871) and calibration across the spectrum of observed and predicted risks. Ten-fold internal cross-validation demonstrated minimal optimism (AUC=0.851, optimism 0.019 [standard deviation=0.06]) with excellent calibration. External validation demonstrated lower discrimination (AUC=0.700) and degraded calibration.
A multivariable mortality risk prediction model was strongly discriminative and well calibrated internally. However, poor external validation suggests the model may not be generalisable to non-NSQIP data and hospitals. The findings highlight the importance of external validation before clinical application of risk models.
老年人（≥65 岁）在接受紧急普通外科手术后，发病率和死亡率增加。需要风险预测模型来指导这个高危人群的决策。现有的模型有很大的局限性，缺乏外部验证，可能限制了它们在临床应用中的适用性。我们旨在推导和验证一个多变量模型，该模型可以预测接受紧急普通外科手术的老年患者的 30 天死亡率，并进行内部和外部验证。
在方案公布后，我们使用国家外科质量改进计划（NSQIP）数据库（2012-6；估计包含美国 90%的数据和加拿大 10%的数据），使用逻辑回归和弹性网络正则化，推导出一个预测接受紧急普通外科手术的老年患者 30 天死亡率的模型。内部验证采用 10 倍交叉验证。外部验证使用来自加拿大安大略省的一个单独的临时卫生行政数据库进行。
总体而言，50221 名患者中有 6012 名（12.0%）在 30 天内死亡。该模型在观察到的和预测的风险范围内具有较强的区分能力（曲线下面积[AUC]=0.871）和校准能力。10 倍内部交叉验证显示最小的乐观性（AUC=0.851，乐观性 0.019[标准差=0.06]）和良好的校准能力。外部验证显示出较低的区分能力（AUC=0.700）和校准能力下降。
一个多变量死亡率风险预测模型在内部具有较强的区分能力和良好的校准能力。然而，外部验证效果不佳表明该模型可能不适用于非 NSQIP 数据和医院。这些发现强调了在临床应用风险模型之前进行外部验证的重要性。
