Statistical Deconvolution for Inference of Infection Time Series.

Area of Science:

• Epidemiology
• Biostatistics
• Computational Biology

Background:

• Accurate daily infection incidence measurement is vital for effective epidemic response.
• Stochastic delays in symptom onset, testing, and reporting obscure true transmission dynamics.
• Existing methods for estimating incidence are often sensitive to model misspecification and data censoring, leading to biased results.

Purpose of the Study:

• To develop a novel estimator that accurately measures daily infection incidence by accounting for stochastic delays.
• To address limitations of existing estimators, including sensitivity to model misspecification and censored observations.
• To provide a stable and reliable method for real-time epidemic monitoring.

Main Methods:

• Development of a robust incidence deconvolution estimator incorporating a regularization scheme.
• Comparison of the new estimator against existing methods through a comprehensive simulation study.
• Application of the estimator to COVID-19 case data in the United States.

Main Results:

• The robust incidence deconvolution estimator demonstrated superior accuracy and stability across various simulation conditions.
• The method proved robust to model misspecification and right-censored data.
• Analysis of US COVID-19 data using the new estimator provided reliable insights into transmission dynamics.

Conclusions:

• The robust incidence deconvolution estimator offers a significant advancement in accurately measuring daily infection incidence.
• The developed R package, incidental, provides a practical tool for real-time epidemic surveillance.
• This method enhances epidemic response capabilities by providing unbiased and stable infection data.