On Identifying and Mitigating Bias in the Estimation of the COVID-19 Case Fatality Rate

Abstract

The relative case fatality rates (CFRs) between groups and countries are key measures of relative risk that guide policy decisions regarding scarce medical resource allocation during the ongoing COVID-19 pandemic. In the middle of an active outbreak when surveillance data is the primary source of information, estimating these quantities involves compensating for competing biases in time series of deaths, cases, and recoveries. These include time- and severity- dependent reporting of cases as well as time lags in observed patient outcomes. In the context of COVID-19 CFR estimation, we survey such biases and their potential significance. Further, we analyze theoretically the effect of certain biases, like preferential reporting of fatal cases, on naive estimators of CFR. We provide a partially corrected estimator of these naive estimates that accounts for time lag and imperfect reporting of deaths and recoveries. We show that collection of randomized data by testing the contacts of infectious individuals regardless of the presence of symptoms would mitigate bias by limiting the covariance between diagnosis and death. Our analysis is supplemented by theoretical and numerical results and a simple and fast open-source codebase at https://github.com/aangelopoulos/cfr-covid-19 .

Associations between meteorology and COVID-19 in early studies: Inconsistencies, uncertainties, and recommendations

“…At the pandemic's current stage, efforts to characterize the environmental sensitivity of COVID-19 rely on surveillance data, but these data are heterogeneous and impacted by biases in the time series of deaths, cases, and recoveries [ 12 , 44 , 45 ]. Even for a fixed location with quasi-homogeneous interventions, human behavior, and demographics, there are regular changes in testing policies, which could result in a different proportion of cases detected with time [ 46 ].…”

Preparedness of countries to face COVID-19 pandemic crisis: Strategic positioning and factors supporting effective strategies of prevention of pandemic threats

“…In addition, countries can use different approaches for counting deaths. In fact, variations in data may be also due to dissimilar quality of healthcare systems and/or to interventions applied at different stages of the illness between countries, making comparative analysis in some cases problematic ( Angelopoulos et al, 2020 ; Antony et al, 2020 ; Lau et al, 2021 ).…”

Pandemic catch-22: The role of mobility restrictions and institutional inequalities in halting the spread of COVID-19

“…imperfect reporting, the release of incorrect data, and delays in reporting, to name a few. Angelopoulos et al (2020) [31] discusses how the problems could bias estimation in either direction depending on their relative magnitude, and Millimet and Parmeter (2019) [32] provides a discussion of cases when data is skewed in one direction due to one-sided measurement errors. While many of these data issues cannot be resolved, we interpret our estimates with caution and perform various robustness checks to minimize the bias in the comparisons we make.…”

“…There are three possible ways to limit the sample to countries that provide reliable test data. First, in the absence of testing a randomly selected sample from the population that most countries lack the resources to implement, Angelopoulos et al (2020) [31] expands on how contact tracing can be a powerful tool that allows otherwise intractable biases to be controlled. Contact tracing expands to include a much larger section of the target population, specifically a larger portion of mild and asymptomatic cases, that are otherwise left out from the testing pool.…”