Prvulovic Stefan T, Zoghi Sina, Gajjar Aryan, Sabet Cameron J, Covell Michael M, Pahwa Bhavya, Gupta Nithin, Schmidt Meic H, Moisi Marc D, Delashaw Johnny, Bowers Christian A
School of Medicine, Georgetown University, Washington, DC, USA.
Bowers Neurosurgical Frailty and Outcomes Data Science Lab, Flint, MI, USA.
Global Spine J. 2025 Aug 28:21925682251375072. doi: 10.1177/21925682251375072.
Study DesignRetrospective cohort study.ObjectivesFrailty and nutritional status are predictors of adverse spine surgery outcomes. This study evaluated the predictive utility of a combined Risk Analysis Index (RAI) and Geriatric Nutritional Risk Index (GNRI) model, and introduced a compound score integrating RAI, GNRI, American Society of Anesthesiologists (ASA) classification, and Preoperative Acute Severe Condition (PACS). To develop the CGARP score, we performed multivariable logistic regression with 30-day mortality as the dependent variable and GNRI, RAI, ASA, and PACS as independent variables.MethodsUsing the National Surgical Quality Improvement Program (NSQIP) database (2015-2020), we assessed predictive performance for mortality, morbidity, and length of stay in spine surgery patients. Demographics, comorbidities, and surgical risk factors were analyzed across RAI quartiles. Model performance was measured using area under receiver operating characteristic curve (AUROC).ResultsAmong 360 133 patients, increasing frailty and malnutrition were independently associated with worse postoperative outcomes. The RAI-GNRI model showed outcome-specific discrimination, C-statistics 0.619 (reoperation) to 0.882 (mortality). The CGARP compound model outperformed individual predictors across all outcomes, with AUROCs of 0.882 (mortality), 0.762 (non-home discharge), 0.686 (extended length of stay), 0.694 (any complication), and 0.641 (readmission). Internal bootstrapping confirmed model stability. Random Forest was the most predictive machine-learning algorithm (AUC = 0.9553). Threshold analysis using Youden's J statistic identified 4 risk categories, correlating with stepwise increases in mortality, complications, and non-home discharge.DiscussionFrailty and nutritional risk are independently predictive of adverse spine surgery outcomes. The CGARP model demonstrated superior predictive performance and provides clinically actionable risk stratification.
研究设计
回顾性队列研究。
目的
衰弱和营养状况是脊柱手术不良结局的预测因素。本研究评估了联合风险分析指数（RAI）和老年营养风险指数（GNRI）模型的预测效用，并引入了一个整合RAI、GNRI、美国麻醉医师协会（ASA）分级和术前急性重症情况（PACS）的复合评分。为了制定CGARP评分，我们以30天死亡率为因变量，GNRI、RAI、ASA和PACS为自变量进行多变量逻辑回归。
方法
使用国家外科质量改进计划（NSQIP）数据库（2015 - 2020年），我们评估了脊柱手术患者死亡率、发病率和住院时间的预测性能。在RAI四分位数中分析了人口统计学、合并症和手术风险因素。使用受试者操作特征曲线下面积（AUROC）来衡量模型性能。
结果
在360133例患者中，衰弱和营养不良程度增加与术后不良结局独立相关。RAI - GNRI模型显示出针对特定结局的区分能力，C统计量从0.619（再次手术）到0.882（死亡率）。CGARP复合模型在所有结局方面均优于个体预测指标，AUROC分别为0.882（死亡率）、0.762（未回家出院）、0.686（延长住院时间）、0.694（任何并发症）和0.641（再入院）。内部自助法证实了模型的稳定性。随机森林是预测性最强的机器学习算法（AUC = 0.9553）。使用约登指数进行的阈值分析确定了4个风险类别，与死亡率、并发症和未回家出院的逐步增加相关。
讨论
衰弱和营养风险是脊柱手术不良结局的独立预测因素。CGARP模型表现出卓越的预测性能，并提供了具有临床可操作性的风险分层。
