A random-censoring Poisson model for underreported data

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Data underreporting in socially deprived, underdeveloped regions leads to biased statistical estimates.
• Existing methods like the censored Poisson model require precise identification of censored regions, which is often impractical.
• Accurate risk monitoring is crucial for targeted interventions in vulnerable populations.

Purpose of the Study:

• To introduce the random-censoring Poisson model (RCPM) to account for uncertainty in both data counts and reporting processes.
• To enable estimation of relative risk and censoring probability for each region.
• To address the challenge of underreported data in risk monitoring.

Main Methods:

• Development of the random-censoring Poisson model (RCPM).
• Proposal of a Markov chain Monte Carlo scheme using data augmentation for posterior sampling.
• Simulation studies comparing RCPM with existing models under various scenarios.

Main Results:

• The proposed RCPM effectively accounts for uncertainty in underreported data.
• The model allows for simultaneous estimation of relative risk and censoring probability.
• Simulation results demonstrate RCPM's advantage over competitive models in scenarios with data quality issues.

Conclusions:

• The RCPM offers a robust approach to risk monitoring in areas with poor data quality.
• This model provides more reliable estimates of disease risk by acknowledging reporting uncertainties.
• Application to early neonatal mortality in Brazil highlights the model's practical utility.