Precision of incidence predictions based on Poisson distributed observations

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Disease incidence prediction is crucial for public health planning and resource allocation.
• Simple trend extrapolation methods are commonly used but lack robust uncertainty quantification.
• Accurate prediction intervals are needed for reliable administrative and scientific decision-making.

Purpose of the Study:

• To develop and present methods for calculating approximate confidence prediction intervals for disease incidence.
• To apply these methods to both total case numbers and age-adjusted incidence rates.
• To demonstrate the utility of these methods using cancer incidence data.

Main Methods:

• Utilizing trend extrapolation on arithmetic or logarithmic scales for disease incidence predictions.
• Assuming a Poisson distribution for age and period-specific incident cases to derive prediction intervals.
• Generalizing prediction models to include power families and extra-Poisson variation for enhanced accuracy.

Main Results:

• The study provides a framework for calculating approximate confidence prediction intervals for disease incidence.
• Demonstrated application to cancer incidence data from the Stockholm-Gotland Oncological Region.
• The methods account for uncertainty in both total cases and age-adjusted rates.

Conclusions:

• The proposed methods offer a statistically sound approach to quantifying uncertainty in disease incidence predictions.
• These validated methods can improve the reliability of administrative and scientific disease forecasting.
• The approach is applicable to various disease prediction scenarios, enhancing public health preparedness.