Data-driven approach in a compartmental epidemic model to assess undocumented infections

TL;DR

This paper presents a simple data-driven compartmental model that estimates undocumented COVID-19 infections and the effective reproductive number from reported case data, accounting for heterogeneity across locations.

Contribution

A novel compartmental modeling approach that estimates undocumented infections and R_t using only reported data, adaptable to various epidemiological contexts.

Findings

Estimated heterogeneity of under-reporting across Brazilian municipalities.

Reproductive number R_t is robust to asymptomatic diagnosis and infection rates.

Method can be extended to other diseases and data types.

Abstract

Nowcasting and forecasting of epidemic spreading rely on incidence series of reported cases to derive the fundamental epidemiological parameters for a given pathogen. Two relevant drawbacks for predictions are the unknown fractions of undocumented cases and levels of nonpharmacological interventions, which span highly heterogeneously across different places and times. We describe a simple data-driven approach using a compartmental model including asymptomatic and presymptomatic contagions that allows to estimate both the level of undocumented infections and the value of effective reproductive number R t from time series of reported cases, deaths, and epidemiological parameters. The method was applied to epidemic series for COVID-19 across different municipalities in Brazil allowing to estimate the heterogeneity level of under-reporting across different places. The reproductive number derived within the current framework is little sensitive to both diagnosis and infection rates during the asymptomatic states. The methods described here can be extended to more general cases if data is available and adapted to other epidemiological approaches and surveillance data.