Tailoring capture-recapture methods to estimate registry-based case counts based on error-prone diagnostic signals

Area of Science:

• Epidemiology
• Biostatistics
• Cancer Surveillance

Background:

• Effective epidemiological monitoring relies on accurate case counts and disease prevalence data.
• Traditional capture-recapture (CRC) methods have limitations in efficiency and defensibility.
• Existing surveillance data streams can suffer from non-representative sampling and diagnostic signal errors.

Purpose of the Study:

• To extend the "anchor stream" sampling design for more efficient and accurate estimation of disease recurrence.
• To develop a methodology that accounts for false positive and negative diagnostic signals in surveillance data.
• To provide a robust alternative to traditional CRC methods for cancer registry data.

Main Methods:

• Utilized a small random sample with medical records abstraction combined with existing signaling data streams.
• Developed a method to estimate true case counts using an estimable positive predictive value (PPV) parameter.
• Employed multiple imputation for standard errors and an adapted Bayesian credible interval approach for statistical inference.

Main Results:

• The proposed "anchor stream" design effectively handles false positive/negative signals in surveillance data.
• Demonstrated valid estimation of true case counts by incorporating an estimable PPV.
• Simulation studies and a real-world cancer registry data example confirmed the method's benefits.

Conclusions:

• The extended "anchor stream" methodology provides a more efficient and defensible approach to epidemiological surveillance.
• This method accurately estimates disease recurrence, such as breast cancer, even with imperfect signaling data.
• The approach offers improved statistical properties for case count estimation in cancer registries.