Measurement error, time lag, unmeasured confounding: Considerations for longitudinal estimation of the effect of a mediator in randomised clinical trials. — CFSMEATLAS
Measurement error, time lag, unmeasured confounding: Considerations for longitudinal estimation of the effect of a mediator in randomised clinical trials.
Goldsmith, K A, Chalder, T, White, P D et al. · Statistical methods in medical research · 2018 · DOI
Quick Summary
This study examined how treatments work in the PACE trial by using advanced statistical methods to track changes in patients' symptoms and functioning over time. The researchers tested different ways of analyzing data to understand whether improvements in one area (like activity levels) might explain improvements in another area (like fatigue). They found that accounting for measurement errors and considering the timing of measurements was important for getting accurate results.
Why It Matters
This study improves how researchers analyze treatment trials for ME/CFS by identifying best practices for understanding treatment mechanisms. Better statistical methods help determine which specific changes in patients' condition are actually responsible for improvements, which can guide future treatment development. These methodological advances ensure that conclusions about how treatments work are more reliable and trustworthy.
Observed Findings
Contemporaneous (immediate) mediator-outcome effects provided a better fit to the data than lagged effects
Accounting for measurement error in the mediator was an important consideration in mediation analysis
Unmeasured confounding between the mediator and outcome significantly affected estimates
Assuming a constant mediator-outcome relationship over time improved statistical precision
Wide spacing between measurements limited the ability to detect lagged temporal relationships
Inferred Conclusions
Longitudinal mediation analyses should account for measurement error and unmeasured confounding to avoid biased estimates
Measurement timing and spacing are critical design considerations for future trials aiming to study treatment mechanisms
Cross-sectional mediation approaches may miss important temporal dynamics and should be replaced with proper longitudinal methods
Statistical assumptions about how mediators relate to outcomes over time should be explicitly tested rather than assumed
Remaining Questions
How do different measurement intervals affect the ability to detect lagged mediator-outcome relationships in ME/CFS trials?
What This Study Does Not Prove
This study does not prove whether any specific treatment mechanism actually works or whether the PACE interventions are effective. It is a methodological study about how to properly analyze data, not a clinical outcomes study. The findings about statistical approaches may or may not apply to other ME/CFS trials with different measurement schedules or treatment types.
About the PEM badge: “PEM required” means post-exertional malaise was an explicit required diagnostic criterion for participant inclusion in this study — not that PEM was studied, observed, or discussed. Studies using criteria that do not require PEM (e.g. Fukuda, Oxford) are tagged “PEM not required”. How the atlas works →