Analysis of longitudinal randomized clinical trials using item response models.

Area of Science:

• Clinical research methodology
• Psychometrics
• Biostatistics

Background:

• Patient-relevant outcomes (PROs) like quality of life are increasingly vital in clinical research.
• Questionnaires are common instruments for measuring these PROs.
• Item response theory (IRT) offers advanced analytical approaches for longitudinal studies using such instruments.

Purpose of the Study:

• To compare methods for estimating treatment effects in longitudinal randomized clinical trials (RCTs) using IRT.
• To address the challenge of incorporating estimation error from latent outcome variables into treatment effect estimation.
• To evaluate the statistical power and practical advantages of different IRT-based approaches.

Main Methods:

• Comparison of three IRT-based methods: plausible value imputation (PVI), concurrent marginal maximum likelihood (MML), and a two-step MML approach.
• Analysis of data from a longitudinal randomized clinical trial.
• Assessment of statistical power for detecting small to moderate effect sizes.

Main Results:

• PVI and concurrent MML demonstrated significantly higher statistical power than the two-step MML method.
• PVI allows for treatment effect estimation using standard statistical software.
• Comparable power was achieved even when analyzing data from different item sets across patient groups, highlighting analytical flexibility.

Conclusions:

• PVI and MML are powerful methods for analyzing treatment effects in longitudinal RCTs using IRT.
• PVI offers practical advantages in terms of software accessibility and flexibility in study design.
• These methods enhance the ability to detect treatment effects and provide flexibility in questionnaire design for clinical trials.