Reconstructing COVID-19 incidences from positive RT-PCR tests by deconvolution

Area of Science:

• Epidemiology
• Public Health
• Mathematical Modeling

Background:

• Emerging COVID-19 variants present significant public health challenges.
• Mass daily RT-PCR testing was implemented in China, but a delay exists between infection and positive test results.
• Understanding this delay is crucial for accurate infection incidence reconstruction and effective disease control.

Purpose of the Study:

• To develop and validate a deconvolution method for reconstructing daily COVID-19 infection incidences from RT-PCR test data.
• To estimate the delay pattern between infection and positive RT-PCR tests.
• To apply the method to real-world outbreak data for improved epidemic monitoring and control.

Main Methods:

• Formulated a convolution model treating incidence reconstruction as a linear inverse problem with positivity constraints.
• Employed the Richard-Lucy deconvolution algorithm and developed a real-time version for incidence reconstruction.
• Applied the method to Omicron variant outbreak data from Beijing and Wuxi, estimating delay functions using an E-M algorithm.

Main Results:

• The estimated delay function for a 2022 bar outbreak showed a shortened mode (4 days) compared to 2020 data.
• Deconvolved infection incidences accurately identified infection events, even after outbreak closure, and detected explosive increases in Wuxi.
• The method revealed outbreak progression, offering insights into prevention and control strategies, particularly the role of mass testing.

Conclusions:

• The deconvolution method is broadly applicable to other infectious diseases with appropriate delay models.
• Accurate reconstruction requires estimating the delay function in a context similar to the virus variant and testing protocol.
• Real-time deconvolution provides valuable, prompt feedback for modifying control measures during acute epidemic phases.