Statistical Methodology Development in Blood Transfusion Protocol Research

描述（由申请人提供）：在美国，损伤是丧失生产寿命的主要原因，并且消耗所有捐献的红细胞（RBC）输注的10 - 15%。虽然25%的1级创伤中心患者接受至少一个单位的RBC，但那些接受大量输血（MT）（定义为24小时内10个或更多单位）的患者消耗了71%的RBC输血，院内死亡风险为40%。血浆和血小板（单独成分）分别给予90%和71%的MT患者。尽管最近的临床和转化研究水平很高，但在知识和障碍方面仍然存在重大差距。最紧迫的问题包括：1）是否更准确地预测需要MT的患者，2）是否采用最佳MT方案进行早期干预（即，足够的血浆、血小板和RBC单位的体积和比率）可以改善患者的结果。创伤输血实践中这些未解决的问题持续存在，主要是因为研究设计（回顾性）和统计分析方法（标准回归模型）的限制，这些方法不适合数据的高度动态性。根据MT的标准定义对患者进行亚组，排除了真正需要MT方案但在接受第10个RBC单位之前因手术或其他干预而死亡或实现止血的患者，从而引入了生存偏倚。生存偏倚也威胁到以前的研究，因为累积24小时输血率的标准使用和死亡率的回归建模不能解决治疗是否延长了生存期或患者必须存活足够长的时间才能接受治疗（例如，以获得高的血浆：血小板：RBC比率）。使用替代的统计策略，如时间依赖性比例风险回归可能无法克服这些问题，因为潜在的信息删失和时间依赖性混杂。我们的目标是解决这些问题，通过开发相关的方法，潜在的类分析和经常性的事件数据分析。将进行两个具体目标：1）开发和评估潜在类别模型，以准确识别真正需要MT的老年患者，并取代现有MT定义作为评估预测算法性能的金标准。此外，新的黄金标准将帮助我们通过增加新的候选预测因子来增强现有预测算法的性能; 2）开发多类型复发事件模型，用于估计时间依赖性RBC，血浆和血小板输注率，并评估其对患者生存的影响。所开发的方法将通过模拟研究进行广泛测试，然后用前瞻性观察性多中心重大创伤输血（PROMMTT）研究的数据进行验证。这项研究的结果将指导未来比较有效性研究的设计和实施，并促进MT协议从实验室到床边的创新改进的更快翻译。我们的新统计方法有望在许多不同的临床背景和动态数据集上得到广泛应用。

DESCRIPTION (provided by applicant): In the U.S., injury is the leading cause of productive years of life lost and consumes 10-15% of all donated red blood cell (RBC) transfusions. While 25% of patients admitted to level 1 trauma centers receive at least one unit of RBCs, those receiving massive transfusion (MT), defined as 10 or more units within 24 hours, consume 71% of all RBC transfusions with a 40% risk of in-hospital mortality. Plasma and platelets (separate components) are given to 90% and 71% of MT patients, respectively. Despite high levels of recent clinical and translational research, significant gaps in knowledge and barriers remain. The most urgent include whether 1) more accurate prediction of the patients in need of MT, and 2) earlier intervention with the optimum MT protocol (i.e., sufficient volumes and ratios of plasma, platelet and RBC units) can improve patient outcomes. These unresolved issues in trauma transfusion practice persist largely because of constraints in study design (retrospective) and statistical analysis methods (standard regression modeling) that are poorly suited to the highly dynamic nature of the data. Subgrouping patients according to the standard definition of MT introduces survival bias by excluding the hemorrhaging patients who truly needed an MT protocol, but died or achieved hemostasis due to surgical or other intervention before receiving the 10th RBC unit. Survival bias also threatens previous studies because the standard use of cumulative 24 hour transfusion ratios and regression modeling of mortality cannot resolve whether the treatment prolonged survival or patients had to survive long enough to receive treatment (e.g., to achieve high plasma:platelet:RBC ratios). The use of alternate statistical strategies like time-dependent proporational hazards regression may not overcome these problems because of the potential for informative censoring and time-dependent confounding. Our objective is to address these issues by developing relevant methodology for latent class analysis and recurrent event data analysis. Two specific aims will be undertaken: 1) to develop and evaluate a latent class model to accurately identify the hemorrhaging patients who truly needed an MT and replace the existing MT definition as the gold standard in assessing the performance of prediction algorithms. Furthermore, the new gold standard will help us enhance the performance of existing predictive algorithms with the addition of new candidate predictors; and 2) to develop a multi-type recurrent event model for estimating time-dependent RBC, plasma, and platelet transfusion rates, and evaluating their impact on patient survival. The developed methods will be extensively tested by simulation studies and thereafter validated with data from the PRospective Observational Multicenter Major Trauma Transfusion (PROMMTT) study. Results from this research will guide the design and conduct of future comparative effectiveness research and facilitate more rapid translation of innovative improvements in MT protocols from bench to bedside. Our new statistical methods are expected to have broad application across many different clinical contexts and dynamic data sets.