Sensitivity analysis for nonignorable missingness and outcome misclassification from proxy reports.

Area of Science:

• Epidemiology
• Biostatistics
• Gerontology

Background:

• Epidemiologic studies of older adults often rely on proxy respondents (relatives, caregivers) when participants cannot self-report due to illness or cognitive impairment.
• Typically, only one report (either participant self-report or proxy report) is available per participant, leading to potential bias if proxy reports replace missing self-reports.
• Excluding participants with missing self-reports can also introduce bias, necessitating robust analytical methods.

Purpose of the Study:

• To propose a novel pattern-mixture model to effectively utilize error-prone proxy reports in statistical analyses.
• To reduce selection bias stemming from missing outcomes in epidemiologic studies of older adults.
• To conduct a sensitivity analysis to address potential bias from differential outcome misclassification.

Main Methods:

• Development of a pattern-mixture model incorporating proxy reports to mitigate bias from missing outcomes.
• Application of propensity-score stratification and multiple imputation for model estimation with high-dimensional covariates.
• Validation through simulation studies and application to the Baltimore Hip Studies cohort.

Main Results:

• The proposed pattern-mixture model effectively reduces selection bias associated with missing outcomes.
• Sensitivity analysis helps address bias arising from differential outcome misclassification.
• Simulation studies demonstrated the favorable performance of the developed methods.

Conclusions:

• The proposed statistical methods offer a valuable approach for analyzing data involving proxy reports in epidemiologic studies of older adults.
• These methods can improve the accuracy of parameter estimates and reduce bias when dealing with missing self-reported outcomes.
• Accessible SAS programs are provided to facilitate the implementation of these advanced statistical techniques.