缺失数据可能导致髋部骨折数据库研究结论改变:美国外科医师学会国家手术质量改进项目研究
该研究基于美国外科医师学会国家手术质量改进项目(NSQIP)数据库,纳入2005年至2013年因髋部骨折接受手术的老年患者26066例,分析人口学、合并症和实验室检查变量的缺失比例,并比较三种缺失数据处理方法对术后不良事件危险因素识别的影响。结果显示部分变量缺失率高达77.9%,三种方法仅识别出35个危险因素中的7个(20%)为共同因素。作者认为缺失数据是国家数据库研究中的重要问题。
要点
- NSQIP髋部骨折手术患者中,许多变量缺失率高达77.9%。
- 三种缺失数据处理方法识别出的术后不良事件危险因素不同。
- 35个危险因素中仅7个(20%)在三种分析中一致。
核心结论
- 缺失数据是国家数据库研究中的重要问题,研究者在评价此类研究时必须加以考虑。
研究设计与样本
- 基于美国外科医师学会国家手术质量改进项目(NSQIP)数据库,识别2005年至2013年因髋部骨折接受手术的患者,共26066例。
- 记录人口学、合并症和实验室检查变量的缺失百分比,并采用基于常见缺失数据处理方法的多变量回归检验这些变量与「任何不良事件」的关联。
主要发现
- 比较三种缺失数据处理方法的多变量回归发现,术后不良事件的危险因素不同。
- 35个识别出的危险因素中,仅7个(20%)在三种分析中均被识别。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
AIMS: The aims of this study were to characterize the frequency of missing data in the National Surgical Quality Improvement Program (NSQIP) database and to determine how missing data can influence the results of studies dealing with elderly patients with a fracture of the hip. PATIENTS AND METHODS: Patients who underwent surgery for a fracture of the hip between 2005 and 2013 were identified from the NSQIP database and the percentage of missing data was noted for demographics, comorbidities and laboratory values. These variables were tested for association with 'any adverse event' using multivariate regressions based on common ways of handling missing data. RESULTS: A total of 26 066 patients were identified. The rate of missing data was up to 77.9% for many variables. Multivariate regressions comparing three methods of handling missing data found different risk factors for postoperative adverse events. Only seven of 35 identified risk factors (20%) were common to all three analyses. CONCLUSION: Missing data is an important issue in national database studies that researchers must consider when evaluating such investigations. Cite this article: Bone Joint J 2018;100-B:226-32.
引用
- Basques BA, McLynn RP, Lukasiewicz AM, et al. Missing data may lead to changes in hip fracture database studies: a study of the American College of Surgeons National Surgical Quality Improvement Program. Bone Joint J. 2018. PMID: 29437066. doi:10.1302/0301-620X.100B2.BJJ-2017-0791.R1
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。