Optimal Methods for Reducing Proxy-Introduced Bias on Patient-Reported Outcome Measurements for Group-Level Analyses

Area of Science:

• Health outcomes research
• Biostatistics
• Rehabilitation medicine

Background:

• Patient-reported outcomes (PROs) are crucial in healthcare research.
• Caregivers (proxies) often complete PROs for patients.
• Proxies tend to rate patient outcomes more negatively than patients themselves.

Purpose of the Study:

• To identify optimal methods for reducing bias introduced by proxy respondents in PRO analyses.
• To improve the interpretability of PRO data in group-level studies.

Main Methods:

• Simulated data from 200 stroke patients and their proxies completing 9 PROMIS domains.
• Varied sample sizes (50-500) and proxy respondent proportions (10%-50%).
• Evaluated six methods for handling proxy data, including complete case analysis, proxy substitution, and multiple imputation.

Main Results:

• Mean PROMIS T-score differences ranged from 0 to 1.75 across methods.
• Methods 1 (complete case) and 5 (multiple imputation) yielded estimates closest to observed patient-only means.
• Low bias was observed with 10%-20% proxy completions for most methods; proxy substitution with adjustment showed low bias even at 50% proxy completion.

Conclusions:

• Modest proxy-introduced bias was found in PRO scores and regression estimates.
• Bias remained low across domains and sample sizes, increasing with higher proxy proportions.
• Proxy substitution with adjustment is a viable method for handling high proportions of proxy respondents in stroke PRO research.