Biometrics Unit - CTD INCa, Institut du Cancer Montpellier, Univ. Montpellier, Montpellier, France.
Institut de Recherche en Cancérologie de Montpellier Inserm U1194, University Montpellier, 208 rue des Apothicaire, Montpellier Cedex 5, 34298, Montpellier, France.
Qual Life Res. 2021 Jan;30(1):91-103. doi: 10.1007/s11136-020-02605-3. Epub 2020 Aug 18.
Health-related quality of life (HRQoL) is assessed by self-administered questionnaires throughout the care process. Classically, two longitudinal statistical approaches were mainly used to study HRQoL: linear mixed models (LMM) or time-to-event models for time to deterioration/time until definitive deterioration (TTD/TUDD). Recently, an alternative strategy based on generalized linear mixed models for categorical data has also been proposed: the longitudinal partial credit model (LPCM). The objective of this article is to evaluate these methods and to propose recommendations to standardize longitudinal analysis of HRQoL data in cancer clinical trials.
The three methods are first described and compared through statistical, methodological, and practical arguments, then applied on real HRQoL data from clinical cancer trials or published prospective databases. In total, seven French studies from a collaborating group were selected with longitudinal collection of QLQ-C30. Longitudinal analyses were performed with the three approaches using SAS, Stata and R software.
We observed concordant results between LMM and LPCM. However, discordant results were observed when we considered the TTD/TUDD approach compared to the two previous methods. According to methodological and practical arguments discussed, the approaches seem to provide additional information and complementary interpretations. LMM and LPCM are the most powerful methods on simulated data, while the TTD/TUDD approach gives more clinically understandable results. Finally, for single-item scales, LPCM is more appropriate.
These results pledge for the recommendation to use of both the LMM and TTD/TUDD longitudinal methods, except for single-item scales, establishing them as the consensual methods for publications reporting HRQoL.
健康相关生活质量（HRQoL）是通过自我管理问卷在整个治疗过程中进行评估的。传统上，主要使用两种纵向统计方法来研究 HRQoL：线性混合模型（LMM）或用于恶化时间/直到明确恶化时间（TTD/TUDD）的时间事件模型。最近，还提出了一种基于分类数据广义线性混合模型的替代策略：纵向部分信用模型（LPCM）。本文的目的是评估这些方法，并提出建议，以标准化癌症临床试验中 HRQoL 数据的纵向分析。
首先通过统计、方法和实际论据描述和比较这三种方法，然后应用于来自癌症临床试验或已发表前瞻性数据库的真实 HRQoL 数据。共选择了来自协作组的七个法国研究，对 QLQ-C30 进行纵向收集。使用 SAS、Stata 和 R 软件，通过三种方法进行纵向分析。
我们观察到 LMM 和 LPCM 之间的结果一致。然而，当我们将 TTD/TUDD 方法与前两种方法进行比较时，观察到了不一致的结果。根据讨论的方法和实际论据，这些方法似乎提供了额外的信息和补充解释。在模拟数据中，LMM 和 LPCM 是最强大的方法，而 TTD/TUDD 方法则提供了更具临床意义的结果。最后，对于单项量表，LPCM 更适用。
这些结果支持使用 LMM 和 TTD/TUDD 纵向方法的建议，除了单项量表，将它们确立为报告 HRQoL 的共识方法。
