Association analysis of self-reported outcomes with a validated subset

Area of Science:

• Health Research Methodology
• Biostatistics
• Epidemiology

Background:

• Health outcome data from surveys can be biased compared to clinical assessments.
• Measurement error in outcome ascertainment can lead to inaccurate conclusions in exposure-outcome association studies.
• Cohort studies often face challenges with imperfect outcome data collection.

Purpose of the Study:

• To develop a statistical method to address bias in health outcomes ascertained through questionnaires.
• To enable accurate inference in cohort studies using both validated and unvalidated outcome data.
• To provide a flexible approach usable with standard statistical software.

Main Methods:

• Constructed a two-part likelihood function utilizing both validated and unvalidated outcome data subsets.
• Employed weighted generalized linear models for the proposed method.
• Compared performance against maximum likelihood estimates (MLE) using Monte Carlo simulations.

Main Results:

• The proposed method allows for statistical inference with standard software, expanding on previous work.
• Monte Carlo simulations were used to compare the finite sample performance of the new method against existing approaches.
• Demonstrated the application of the methodology in a cohort study of long-term childhood cancer survivors.

Conclusions:

• The developed statistical method effectively addresses bias in questionnaire-based health outcome data.
• This approach enhances the reliability of findings in cohort studies, particularly those with mixed data quality.
• The study provides a practical tool for analyzing health data from large cohort studies, exemplified by cancer survivor research.