ABSTRACT

Many clinical trials of treatments for patients hospitalised for COVID-19 use an ordinal scale recommended by the World Heath Organisation. The scale represents intensity of medical intervention, with higher scores for interventions more burdensome for the patient, and highest score for death. There is uncertainty about use of this ordinal scale in testing hypotheses. With the objective of assessing the power and Type I error of potential endpoints and analyses based on the ordinal scale, trajectories of the score over 28 days were simulated for scenarios based closely on results of two trials recently published. The simulation used transition probabilities for the ordinal scale over time. No one endpoint was optimal across scenarios, but a ranked measure of trajectory fared moderately well in all scenarios. Type I error was controlled at close to the nominal level for all endpoints. Because not tied to a particular population with regard to baseline severity, the use of transition probabilities allows plausible assessment of endpoints in populations with configurations of baseline score for which data is not yet published, provided some data on the relevant transition probabilities are available. The results could support experts in the choice of endpoint based on the ordinal scale.

摘要

针对COVID-19住院患者的治疗方法的许多临床试验均使用世界卫生组织推荐的序数表。该量表代表医疗干预的强度，干预得分越高，患者的负担就越重，死亡得分越高。在检验假设时使用此序数表存在不确定性。为了评估潜在终点的功效和I型误差并基于序数尺度进行分析，基于最近发表的两项试验的结果，针对情景模拟了28天的得分轨迹。该模拟使用随时间变化的序数标度的转换概率。在所有情况下，没有一个端点是最佳的，但是在所有情况下，对轨迹的排名衡量都表现得很好。所有端点的I型错误均控制在接近标称水平的水平。由于在基线严重性方面与特定人群没有联系，因此，使用过渡概率可以对尚未发布数据的基线得分配置的人群中的端点进行合理评估，前提是可以提供有关相关过渡概率的一些数据。结果可以支持专家根据顺序量表选择终点。