Estimating biomarker-based HIV incidence using prevalence data in high risk groups with missing outcomes.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• The sensitive/less sensitive testing algorithm for detecting recent HIV seroconversion (STARHS) is widely used for estimating HIV-1 incidence from cross-sectional data.
• Uncertainty in STARHS-based incidence estimates is poorly understood, particularly with missing data or in high-risk populations.
• Current ad hoc methods for handling missing data in STARHS produce inaccurate confidence intervals, potentially compromising statistical inference.

Purpose of the Study:

• To develop and validate robust statistical methods for accurately estimating uncertainty in HIV-1 incidence derived from STARHS.
• To address challenges posed by missing data within the STARHS algorithm.
• To extend these methods for use in regression analyses of incidence data.

Main Methods:

• Proposed maximum likelihood and Bayesian approaches for estimating uncertainty in incidence estimates with missing data.
• Applied these methods to a New York City injection drug user cohort study.
• Extended the methodology to regression settings for analyzing incidence trends.

Main Results:

• Demonstrated that ad hoc methods underestimate uncertainty in incidence estimates, using the injection drug user study as a case example.
• The proposed statistical methods provide more accurate confidence limits for incidence estimates.
• The developed methods are applicable to scenarios with missing data in serologic testing algorithms.

Conclusions:

• Accurate estimation of uncertainty is crucial for reliable HIV-1 incidence surveillance using STARHS.
• Maximum likelihood and Bayesian methods offer statistically sound solutions for handling missing data in STARHS.
• These improved methods enhance the validity of epidemiological studies relying on incidence data from similar testing algorithms.