Handling missing data in fibromyalgia clinical trials-considerations for the use of linear mixed models in longitudinal analyses: perspectives in rheumatology.

Abstract

To critically examine current statistical practices for handling missing data in fibromyalgia randomized controlled trials (RCTs) and to provide practical guidance for the implementation of linear mixed models (LMMs). This methodological and narrative perspective reviews recent RCTs in fibromyalgia, highlighting common limitations in the handling of missing data and longitudinal analyses. We contrast traditional approaches, such as repeated-measures ANOVA, with LMM-based strategies within the intention-to-treat (ITT) framework. Additionally, we provide a step-by-step guide for implementing LMMs in SPSS. Evidence indicates that many fibromyalgia RCTs continue to rely on suboptimal statistical methods, including the exclusion of participants with incomplete data and the use of ANOVA-based approaches. These practices may reduce consistency with the ITT principle, decrease statistical efficiency, and contribute to potentially biased treatment effect estimates. In contrast, LMMs can incorporate partially observed longitudinal data, explicitly model within-subject correlations, and accommodate unbalanced longitudinal designs under the Missing At Random (MAR) assumption. However, their validity depends on correct model specification and the plausibility of underlying assumptions. The persistent gap between methodological recommendations and analytical practices in fibromyalgia RCTs may compromise the interpretability and reliability of treatment effect estimates. LMMs may represent a more flexible and methodologically appropriate approach for longitudinal analyses involving incomplete follow-up data, particularly when aligned with study design characteristics and available statistical expertise. Improving statistical rigor in this field remains essential to support more reliable clinical interpretation and evidence-informed decision-making.