Buchlak Quinlan D, Yanamadala Vijay, Leveque Jean-Christophe, Edwards Alicia, Nold Kellen, Sethi Rajiv
Neuroscience Institute, Virginia Mason Medical Center, Seattle, WA, USA.
Neuroscience Institute, Virginia Mason Medical Center, Seattle, WA, USA.
J Clin Neurosci. 2017 Sep;43:247-255. doi: 10.1016/j.jocn.2017.06.012. Epub 2017 Jul 1.
Complication rates in complex spine surgery range from 25% to 80% in published studies. Numerous studies have shown that surgeons are not able to accurately predict whether patients are likely to face post-operative complications, in part due to biases based on individual experience. The purpose of this study was to develop and evaluate a predictive risk model and decision support system that could accurately predict the likelihood of 30-day postoperative complications in complex spine surgery based on routinely measured preoperative variables.
Preoperative and postoperative data were collected for 136 patients by reviewing medical records. Logistic regression analysis (LRA) was applied to develop the predictive algorithm based on patient demographic parameters, including age, gender, and co-morbidities, including obesity, diabetes, hypertension and anemia. We additionally compared the performance of the predictive model to a spine surgeon's ability to predict patient complications using signal detection theory statistics representing sensitivity and response bias (A' and B″ respectively). We developed a decision support system tool, based on the LRA predictive algorithm, that was able to provide a numeric probabilistic likelihood statistic representing an individual patient's risk of developing a complication within the first 30days after surgery.
The predictive model was significant (χ=16.242, p<0.05), showed good fit, and was calibrated by using area under the receiver operating characteristics curve analysis (AUROC=0.712, p<0.01). The model yielded a predictive accuracy of 75.0%. It was validated by splitting the data set, comparing subset models, and testing them with unknown data. Validation also involved comparing the classification of cases by experts with the classification of cases by the model. The model significantly improved the classification accuracy of physicians involved in the delivery of complex spine surgical care.
The application of technology and data-driven tools to advanced surgical practice has the potential to improve decision making quality, service quality and patient safety.
在已发表的研究中，复杂脊柱手术的并发症发生率在25%至80%之间。大量研究表明，外科医生无法准确预测患者是否可能面临术后并发症，部分原因是基于个人经验的偏差。本研究的目的是开发并评估一种预测风险模型和决策支持系统，该系统能够根据常规测量的术前变量准确预测复杂脊柱手术30天术后并发症的可能性。
通过查阅病历收集了136例患者的术前和术后数据。应用逻辑回归分析（LRA），基于患者人口统计学参数（包括年龄、性别）以及合并症（包括肥胖、糖尿病、高血压和贫血）来开发预测算法。我们还使用代表敏感度和反应偏差的信号检测理论统计量（分别为A'和B″），将预测模型的性能与脊柱外科医生预测患者并发症的能力进行了比较。我们基于LRA预测算法开发了一个决策支持系统工具，该工具能够提供一个数字概率似然统计量，代表个体患者术后30天内发生并发症的风险。
预测模型具有显著性（χ=16.242，p<0.05），拟合良好，并通过受试者操作特征曲线分析下的面积（AUROC=0.712，p<0.01）进行了校准。该模型的预测准确率为75.0%。通过拆分数据集、比较子集模型并用未知数据进行测试对其进行了验证。验证还包括将专家对病例的分类与模型对病例的分类进行比较。该模型显著提高了参与复杂脊柱手术护理的医生的分类准确率。
将技术和数据驱动的工具应用于先进的外科实践有可能提高决策质量、服务质量和患者安全性。
