An Augmented Likelihood Approach Incorporating Error-Prone Auxiliary Data Into a Survival Analysis

Area of Science:

• Biostatistics
• Epidemiology
• Health Data Science

Background:

• Big data and electronic health records (EHR) provide extensive clinical data, but accuracy varies.
• Automated algorithms and self-reported data can be error-prone, while gold standard data is often limited to subsets.
• Accurate analysis of time-to-event outcomes is crucial for understanding disease progression and risk factors.

Purpose of the Study:

• To propose a novel statistical method for regression analysis of gold standard time-to-event outcomes.
• To incorporate error-prone disease diagnoses, particularly when gold standard data is available only for a subset of individuals.
• To address challenges like left-truncation and interval-censoring in time-to-event data.

Main Methods:

• Developed a joint likelihood model for gold standard and error-prone outcomes.
• Integrated self-reported disease diagnosis information into the regression analysis.
• Applied the proposed model to the Hispanic Community Health Study/Study of Latinos dataset.

Main Results:

• The method successfully augments regression analysis with limited gold standard data.
• Effectively leverages information from error-prone self-reported diagnoses.
• Quantified risk factors associated with diabetes onset in the study population.

Conclusions:

• The proposed statistical model provides a robust approach for analyzing time-to-event outcomes with mixed data quality.
• This method enhances the utility of large observational studies and EHR data by accounting for data inaccuracies.
• Improved risk factor identification for diseases like diabetes can inform public health strategies.