Re: Trauma care does not discriminate: The association of race and health insurance with mortality following traumatic injury.
Re: Trauma care does not discriminate: The association of race and health insurance with mortality following traumatic injury.
复制标题
回复：创伤护理不歧视：种族和健康保险与创伤后死亡率的关系。
DOI：
10.1097/ta.0000000000000779
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
Millham,Frederick
中科院分区：
文献类型：
--
作者：
Osler,Turner;Glance,LaurentG;Li,Wenjun;Buzas,JefferyS;Wetzel,MeganL;Hosmer,DavidW;Millham,Frederick
回复：我们感谢扎法尔博士对我们工作的兴趣，并感谢《创伤和急性护理外科杂志》给我们机会评论他的问题和担忧。我们很高兴扎法尔博士发现我们使用估算保险状况来解决幸存者治疗分配偏差是一种新颖的补充。我们希望其他人也会发现它对他们的研究有用。尽管我们未能在文章中强调这一点，但我们观察到保险覆盖范围与生存的明显关联是一种附带现象而不是真正的相关性，这是我们分析种族重要性的一个重要先决条件：因为种族与保险覆盖范围相混淆（拥有保险的黑人患者要少得多），因此从我们的最终模型中排除保险覆盖范围对于正确得出种族与生存无关的结论至关重要。我们同意 Zafar 博士的观点，即国家创伤数据库 (NTDB) 是一个宝贵的资源；事实上，我们在其他调查中广泛使用了 NTDB。然而，全国住院样本（NIS）也是一种宝贵的资源，它使我们能够提出一个根本不同的问题。由于 NIS 是美国所有医院 20% 的分层样本，因此它在统计上准确地反映了美国所有医院的入院情况。因此，它使我们能够评估种族和保险覆盖范围在整个美国的重要性。相比之下，NTDB 只允许推断向 NTDB 提供数据的医院。（顺便说一句，我们会注意到，由于 NIS 是分层样本，因此除了美国其他约 4,000 家医院中的 20% 之外，它实际上还包含 NTDB 中 20% 的医院）。 Zafar 担心 NIS 中的可用变量不允许进行充分的风险调整。虽然他同意我们在研究中使用的创伤死亡率预测模型 (TMPM1) 可以对解剖损伤进行充分的风险调整（我们同意这一点：根据 DRG 国际疾病分类第 V9 版 [ICD-9] 代码计算的 TMPM 实际上比根据简明损伤量表 [AIS] 代码计算的损伤严重程度评分 [ISS] 更好地区分幸存者和非幸存者）2，但他担心其他重要的因素NIS 中没有死亡率预测因素，并提到休克是一个特别值得关注的问题。我们最初同意他的怀疑，即临床变量休克可能无法在管理数据集中准确捕获。然而，单变量分析显示，NIS 中编码为患有创伤性休克的 1% 患者的死亡率 (26%) 远高于编码为未患有创伤性休克的 99% 患者（死亡率仅为 1.5%）。此外，在调整后的死亡率模型中，老年人（年龄9-64岁）创伤性休克的比值比（OR，1.90）远高于年轻人（OR，1.27），这一观察结果与临床经验相符，即老年人总体上对休克的耐受性较差。经过考虑，我们得出的结论是，与我们先入为主的观点相反，ICD-9 代码 958.4 可能是一个非常可靠的休克指标，因为它是编码者对医生在笔记中所写的考虑诊断的直接转录，而不是基于收缩压的任意二分值。实际上，我们将休克的诊断“众包”给了知识渊博的医生，而不是依赖于单一的血压测量。我们的最终模型基于对 NIS 中可用数据的仔细使用，并且通过客观的区分和校准措施，非常可靠。正如我们在文章的在线附录中指出的那样，年轻人的逻辑模型......
In Reply: We appreciate Dr. Zafar’s interest in our work and thank the Journal of Trauma and Acute Care Surgery for the opportunity to comment on his questions and concerns. We are pleased that Dr. Zafar found our use of imputed insurance status to address survivor treatment assignment bias to be a novel addition. We hope others will also find it useful in their research. Although we failed to emphasize it in our article, our observation that the apparent association of insurance coverage with survival is an epiphenomenon rather than a real correlation was an essential prerequisite to our analysis of the importance of race: because race is confounded with insurance coverage (far fewer black patients have insurance), excluding insurance coverage from our final model was crucial to correctly conclude that race is not associated with survival. We agree with Dr. Zafar that the National Trauma Data Bank (NTDB) is a valuable resource; indeed, we have made extensive use of the NTDB in other investigations. However, the National Inpatient Sample (NIS) is also a valuable resource, one that allows us to pose a fundamentally different question. Because the NIS is a stratified sample of 20% of all US hospitals, it is a statistically accurate reflection of all hospital admissions in the United States. It therefore allows us to assess the importance of race and insurance coverage in the entire United States. The NTDB, by contrast, allows only inferences to the hospitals that contribute data to the NTDB.(In passing, we would note that, because the NIS is a stratified sample, it actually contains 20% of the hospitals in the NTDB in addition to the 20% of the other roughly 4,000 hospitals in the United States).Dr. Zafar is concerned that the variables available in the NIS do not allow for adequate risk adjustment. While he agrees that the Trauma Mortality Prediction Model (TMPM1) that we used in our study allows for adequate risk adjustment of anatomic injuries (a point we concur on: TMPM computed on DRG International Classification of DiseasesV9th Rev.[ICD-9] codes actually discriminates survivors from nonsurvivors better than Injury Severity Score [ISS] computed on Abbreviated Injury Scale [AIS] codes), 2 he worries that other important predictors of mortality are not available in the NIS and mentions shock as a particular concern. We initially shared his suspicion that a clinical variable, shock, might not be accurately captured in an administrative data set. However, univariate analysis showed that in the 1% of patients coded in the NIS as having traumatic shock, mortality was much higher (26%) than in the 99% of patients coded as not having traumatic shock who had only a 1.5% mortality. Moreover, in adjusted mortality models, the odds ratio (OR) for traumatic shock was much higher in the elderly (age 9 64 years)(OR, 1.90) than in the young (OR, 1.27), an observation that comports with the clinical experience that the elderly are, in general, less tolerant of shock. After consideration, we concluded that, contrary to our preconceived opinion, ICD-9 code 958.4 may be a very reliable indicator of shock precisely because it is a direct transcription by coders of physicians’ considered diagnosis as written in their notes rather than being an arbitrarily dichotomized value based on a systolic blood pressure. In effect, we have ‘‘crowd sourced’’the diagnosis of shock to knowledgeable physicians rather than relying on a single blood pressure measurement. Our final models were based on careful use of the data available in the NIS and, by the objective measures of discrimination and calibration, were very reliable. As we note in the online appendix to our article, the logistic model for young …