Evaluation of undetected cases during the COVID-19 epidemic in Austria

Abstract

Background Knowing the number of undetected cases of COVID-19 is important for a better understanding of the spread of the disease. This study analyses the temporal dynamic of detected vs. undetected cases to provide guidance for the interpretation of prevalence studies performed with PCR or antibody tests to estimate the detection rate. Methods We used an agent-based model to evaluate assumptions on the detection probability ranging from 0.1 to 0.9. For each general detection probability, we derived age-dependent detection probabilities and calibrated the model to reproduce the epidemic wave of COVID-19 in Austria from March 2020 to June 2020. We categorized infected individuals into presymptomatic, symptomatic unconfirmed, confirmed and never detected to observe the simulated dynamic of the detected and undetected cases. Results The calculation of the age-dependent detection probability ruled values lower than 0.4 as most likely. Furthermore, the proportion of undetected cases depends strongly on the dynamic of the epidemic wave: during the initial upswing, the undetected cases account for a major part of all infected individuals, whereas their share decreases around the peak of the confirmed cases. Conclusions The results of prevalence studies performed to determine the detection rate of COVID-19 patients should always be interpreted with regard to the current dynamic of the epidemic wave. Applying the method proposed in our analysis, the prevalence study performed in Austria in April 2020 could indicate a detection rate of 0.13, instead of the prevalent ratio of 0.29 between detected and estimated undetected cases at that time.

Model Based Estimation of the SARS-CoV-2 Immunization Level in Austria and Consequences for Herd Immunity Effects

“…For the estimation of the immunization level, we ran our simulation model for the time span between February 2020 and February 2021. We identified the parameter for the detection rate based on study data and the method published in [3].…”

“…In specific, applying the method introduced in [3], the results of a prevalence study (SORA [6]) conducted in April 2020 result in a detection probability of 13% (CI 95%: [2%,86%]) [3]. The second study (Statistics Austria [7]), performed during the second wave of COVID-19, indicates an increase of the detection probability to 35% (CI 95%: [24%, 58%]).…”

“…To guarantee that the detected cases in the model match the officially reported case numbers, we proceeded analogous to [2] and [3] and calibrated the infection probability parameter as well as all parameters corresponding to governmental policies for this time period using an iterative bisection method. Hereby, the match of the cumulative detected cases and the officially reported numbers for the positive tests is the calibration target.…”

“…Furthermore, for all scenarios, the timeline of the new detected cases is delayed by about 9 days compared to the timeline of new infections. This delay occurs because cases are detected about 9 days after their infection as the mean incubation time is 5.1 days and the mean reaction time between symptom onset and positive test results is 3.8 days [3]. The detected cases are also reduced to 35% of all cases which corresponds to the detection probability applied for these scenarios.…”

“…For both tasks, we apply an agent-based model (ABM) in combination with data from performed cross-sectional SARS-CoV-2 prevalence and serological tests in Austria. This ABM has already been used for several scientific studies such as the investigation of tracing measures [2] or the general dynamics of undetected cases [3], and is one of three models that participate in the official Austrian COVID forecasting panel by the ministry of health (COVID Prognose Konsortium, [4]), which publishes short time forecasts on a weekly basis. For the herd effect, we will also perform cross model validation of the ABM with analytical results from a classic differential equation based SIR-model.…”