SEMIPARAMETRIC ANALYSIS OF INTERVAL-CENSORED DATA SUBJECT TO INACCURATE DIAGNOSES WITH A TERMINAL EVENT

Area of Science:

• Biostatistics
• Epidemiology
• Chronic Disease Research

Background:

• Interval-censored data is common in chronic disease studies, often relying on imperfect biomarker diagnostics.
• Existing analysis methods typically assume accurate disease diagnosis, which is unrealistic for biomarkers like cerebrospinal fluid or cognitive function tests.

Purpose of the Study:

• To develop a semiparametric modeling framework to analyze interval-censored data with inaccurate disease diagnosis.
• To incorporate diagnostic accuracy (sensitivity and specificity) into statistical models for improved inference.
• To extend the framework to handle terminal events and accurate diagnoses.

Main Methods:

• Utilized the Cox proportional hazards model for interval-censored data with imperfect diagnosis.
• Developed a nonparametric maximum likelihood estimation (NPMLE) method.
• Implemented an efficient expectation-maximization (EM) algorithm for computational feasibility.

Main Results:

• The proposed model accounts for uncertainty in disease onset detection due to diagnostic inaccuracies.
• Regression coefficient estimators demonstrated asymptotic normality and semiparametric efficiency.
• Amyloid-beta was found to be significantly associated with Alzheimer's disease (AD).
• Tau was identified as predictive of both AD and mortality.

Conclusions:

• The novel framework effectively analyzes interval-censored data with inaccurate diagnoses in chronic disease studies.
• The approach provides reliable inference and is computationally feasible.
• Biomarker analysis, specifically for AD, can be significantly improved using this method, revealing distinct predictive roles for amyloid-beta and Tau.