What are the appropriate methods for analyzing patient-reported outcomes in randomized trials when data are missing?

Area of Science:

• Health Services Research
• Biostatistics
• Psychometrics

Background:

• Patient-Reported Outcomes (PRO) are crucial in randomized trials for comparing patient groups.
• Classical Test Theory (CTT) and Item Response Theory (IRT) are primary analytical strategies for PRO data.
• The performance of CTT versus IRT with missing PRO data is not well-established.

Purpose of the Study:

• To investigate the appropriateness of IRT and CTT analytical strategies for PRO data with missing values.
• To compare the power and bias of different methods under various missing data mechanisms.

Main Methods:

• Simulated PRO data (quality of life) with missing responses under different missingness mechanisms (random, covariate-dependent, latent trait-dependent).
• Analyzed data using CTT methods (complete-case analysis, mean imputation, multiple imputation) and an IRT-based Wald test on a Rasch model.
• Evaluated statistical power and bias of each method.

Main Results:

• IRT and multiple imputation CTT methods demonstrated the highest statistical power.
• Both IRT and multiple imputation CTT were unbiased across all simulated missing data types.
• Traditional methods like listwise deletion and mean imputation showed bias and lacked power.

Conclusions:

• IRT models and multiple imputation are recommended for analyzing PRO data with missing values in randomized trials.
• Avoid traditional CTT methods (listwise deletion, mean imputation) due to bias and power issues.
• Software and modules are available to facilitate these advanced analytical approaches.