Maximum likelihood estimation of age-specific incidence rate from prevalence

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Longitudinal studies are standard for estimating age-specific chronic disease incidence rates but are expensive and prone to participant attrition.
• Existing methods for incidence estimation from aggregated data lack robust accuracy assessment, often relying on less reliable bootstrap resampling techniques.

Purpose of the Study:

• To develop and validate novel methods for estimating chronic disease incidence rates and their confidence intervals using aggregated data.
• To introduce a maximum likelihood-based approach as a more accurate alternative to resampling methods.

Main Methods:

• Utilized the illness-death model and a related partial differential equation to link incidence, prevalence, and mortality.
• Developed maximum likelihood estimation techniques, incorporating a binomial likelihood function for incidence rate estimation.
• Applied the method to historical data on breathlessness in British coal miners and diabetes in Germany, considering scenarios with non-differential and differential mortality.

Main Results:

• Demonstrated the feasibility of estimating incidence rates and confidence intervals for chronic conditions using aggregated data.
• Showcased applicability in scenarios involving both non-differential and differential mortality, including specific data configurations (e.g., diseased vs. all-cause mortality).
• The maximum likelihood method proved effective and can potentially replace traditional resampling techniques for incidence estimation.

Conclusions:

• Maximum likelihood estimation provides a viable and accurate approach for determining chronic disease incidence rates from aggregated prevalence and mortality data.
• This method offers a cost-effective and reliable alternative to longitudinal studies and bootstrap resampling, enhancing epidemiological research capabilities.