Zura Robert, Braid-Forbes Mary Jo, Jeray Kyle, Mehta Samir, Einhorn Thomas A, Watson J Tracy, Della Rocca Gregory J, Forbes Kevin, Steen R Grant
Dept. of Orthopaedic Surgery, Louisiana State University, New Orleans, LA, USA.
Braid-Forbes Health Research, Silver Spring, MD, USA.
Bone. 2017 Feb;95:26-32. doi: 10.1016/j.bone.2016.11.006. Epub 2016 Nov 9.
Fracture nonunion risk is related to severity of injury and type of treatment, yet fracture healing is not fully explained by these factors alone. We hypothesize that patient demographic factors assessable by the clinician at fracture presentation can predict nonunion.
A prospective cohort study design was used to examine ~2.5 million Medicare patients nationwide. Patients making fracture claims in the 5% Medicare Standard Analytic Files in 2011 were analyzed; continuous enrollment for 12months after fracture was required to capture the ICD-9-CM nonunion diagnosis code (733.82) or any procedure codes for nonunion repair. A stepwise regression analysis was used which dropped variables from analysis if they did not contribute sufficient explanatory power. In-sample predictive accuracy was assessed using a receiver operating characteristic (ROC) curve approach, and an out-of-sample comparison was drawn from the 2012 Medicare 5% SAF files.
Overall, 47,437 Medicare patients had 56,492 fractures and 2.5% of fractures were nonunion. Patients with healed fracture (age 75.0±12.7SD) were older (p<0.0001) than patients with nonunion (age 69.2±13.4SD). The death rate among all Medicare beneficiaries was 4.8% per year, but fracture patients had an age- and sex-adjusted death rate of 11.0% (p<0.0001). Patients with fracture in 14 of 18 bones were significantly more likely to die within one year of fracture (p<0.0001). Stepwise regression yielded a predictive nonunion model with 26 significant explanatory variables (all, p≤0.003). Strength of this model was assessed using an area under the curve (AUC) calculation, and out-of-sample AUC=0.710.
A logistic model predicted nonunion with reasonable accuracy (AUC=0.725). Within the Medicare population, nonunion patients were younger than patients who healed normally. Fracture was associated with increased risk of death within 1year of fracture (p<0.0001) in 14 different bones, confirming that geriatric fracture is a major public health issue. Comorbidities associated with increased risk of nonunion include past or current smoking, alcoholism, obesity or morbid obesity, osteoarthritis, rheumatoid arthritis, type II diabetes, and/or open fracture (all, multivariate p<0.001). Nonunion prediction requires knowledge of 26 patient variables but predictive accuracy is currently comparable to the Framingham cardiovascular risk prediction.
骨折不愈合风险与损伤严重程度及治疗方式相关，但仅这些因素并不能完全解释骨折愈合情况。我们推测，临床医生在骨折初诊时可评估的患者人口统计学因素能够预测骨折不愈合。
采用前瞻性队列研究设计，对全国约250万医疗保险患者进行研究。分析了2011年医疗保险标准分析文件中5%提出骨折索赔的患者；骨折后需连续参保12个月，以获取国际疾病分类第九版临床修订本（ICD - 9 - CM）骨折不愈合诊断代码（733.82）或任何骨折不愈合修复手术代码。采用逐步回归分析，若变量没有足够的解释力则将其从分析中剔除。使用受试者工作特征（ROC）曲线方法评估样本内预测准确性，并从2012年医疗保险5%标准分析文件中进行样本外比较。
总体而言，47437名医疗保险患者发生了56492处骨折，其中2.5%的骨折为不愈合。骨折愈合患者（年龄75.0±12.7标准差）比骨折不愈合患者（年龄69.2±13.4标准差）年龄更大（p<0.0001）。所有医疗保险受益人的年死亡率为4.8%，但骨折患者经年龄和性别调整后的死亡率为11.0%（p<0.0001）。18处骨骼中有14处发生骨折的患者在骨折后一年内死亡的可能性显著更高（p<0.0001）。逐步回归得出一个具有预测骨折不愈合的模型，该模型有26个显著的解释变量（所有变量，p≤0.003）。使用曲线下面积（AUC）计算评估该模型的强度，样本外AUC = 0.710。
逻辑模型预测骨折不愈合具有合理的准确性（AUC = 我们推测，临床医生在骨折初诊时可评估的患者人口统计学因素能够预测骨折不愈合。
采用前瞻性队列研究设计，对全国约250万医疗保险患者进行研究。分析了2011年医疗保险标准分析文件中5%提出骨折索赔的患者；骨折后需连续参保12个月，以获取国际疾病分类第九版临床修订本（ICD - 9 - CM）骨折不愈合诊断代码（733.82）或任何骨折不愈合修复手术代码。采用逐步回归分析，若变量没有足够的解释力则将其从分析中剔除。使用受试者工作特征（ROC）曲线方法评估样本内预测准确性，并从2012年医疗保险5%标准分析文件中进行样本外比较。
总体而言，47437名医疗保险患者发生了56492处骨折，其中2.5%的骨折为不愈合。骨折愈合患者（年龄75.0±12.7标准差）比骨折不愈合患者（年龄69.2±13.4标准差）年龄更大（p<0.0001）。所有医疗保险受益人的年死亡率为4.8%，但骨折患者经年龄和性别调整后的死亡率为11.0%（p<0.0001）。18处骨骼中有14处发生骨折的患者在骨折后一年内死亡的可能性显著更高（p<0.0001）。逐步回归得出一个具有预测骨折不愈合的模型，该模型有26个显著的解释变量（所有变量，p≤0.003）。使用曲线下面积（AUC）计算评估该模型的强度，样本外AUC = 0.710。
逻辑模型预测骨折不愈合具有合理的准确性（AUC = 0.725）。在医疗保险人群中，骨折不愈合患者比正常愈合患者更年轻。骨折与14种不同骨骼骨折后1年内死亡风险增加相关（p<0.0001），这证实老年骨折是一个重大的公共卫生问题。与骨折不愈合风险增加相关的合并症包括过去或现在吸烟、酗酒、肥胖或病态肥胖、骨关节炎、类风湿关节炎、II型糖尿病和/或开放性骨折（所有变量，多变量p<0.001）。预测骨折不愈合需要了解26个患者变量，但目前的预测准确性与弗雷明汉心血管风险预测相当。
