Accounting for underreporting in mathematical modelling of transmission and control of COVID-19 in Iran

Meead Saberi, Homayoun Hamedmoghadam, Kaveh Madani, Helen Dolk, Andreï S. Morgan, Joan K. Morris, Kaveh Khoshnood, Babak KhoshnoodmedRxiv 2020 · DOI

Abstract: Iran has been the hardest hit country by the outbreak of SARS-CoV-2 in the Middle East with 74,877 confirmed cases and 4,683 deaths as of 15 April 2020. With a relatively high case fatality ratio and limited testing capacity, the number of confirmed cases reported is suspected to suffer from significant under-reporting. We developed a compartmental transmission model to estimate the effective reproduction number and its fluctuations since the beginning of the outbreak in Iran. The estimation method accounts for underreporting due to low case ascertainment by estimating the percentage of symptomatic cases using delay-adjusted case fatality ratio based on the distribution of the delay from hospitalization-to-death. Our estimates of the effective reproduction number ranged from 0.66 to 1.73 between February and April 2020, with a median of 1.16. We found that non-pharmaceutical interventions were effective in slowing the spread, but argue that if the impact of underreporting is overlooked, the estimated transmission dynamics could mislead public health decisions.

Introduction

In the early months of 2020, Iran emerged as a major epicenter for SARS-CoV-2. However, researchers identified a glaring discrepancy: the official Case Fatality Ratio (CFR) was unusually high—reaching nearly 17% in the first week—suggesting that for every confirmed case, many more were likely going undetected due to limited testing capacity.

Standard epidemiological models relying solely on confirmed case counts can produce "garbage in, garbage out" results. If a model doesn't account for this "hidden" part of the epidemic, it may miscalculate the effective reproduction number ($R_t$) and lead policymakers to believe that interventions are working better than they actually are. This paper seeks to bridge that gap by integrating underreporting estimates into a transmission model to provide a more accurate timeline of the outbreak in Iran.

Methodology

The authors employed a variation of the SEIR (Susceptible-Exposed-Infected-Recovered) compartmental model to capture the transmission dynamics and the impact of interventions.

Compartmental model structure and delay distributions used to estimate transmission dynamics.

Key technical features of the approach include:

  • Delay-Adjusted CFR: Instead of using raw death counts, the researchers estimated the percentage of symptomatic cases by accounting for the time lag between hospitalization and death. They assumed a baseline "true" CFR of 1.4% and used a lognormal distribution (mean of 13 days) to estimate when deceased individuals would have first been reported as infected.
  • Parameter Optimization: To fit the model to the data, they used Ordinary Least Squares (OLS) minimization combined with a pattern search algorithm—a derivative-free global optimization method—to minimize the Root Squared Error across infected, recovered, and removed cases.
  • Rolling $R_t$ Estimation: The effective reproduction number ($R_t$) was calculated over a rolling 7-day window to capture how specific events, such as school closures or holiday travel, influenced the spread.

Transmission Dynamics and Intervention Impact

The study utilized official time-series data from February 19 to April 15, 2020. The results highlight a critical "underreporting trap": when underreporting is ignored, the initial $R_t$ appears much higher (estimated at 5.67) than when it is accounted for (estimated at 1.73).

Timeline of interventions (a) compared with the estimated effective reproduction number $R(t)$ (b) and the delay-adjusted Case Fatality Ratio (c).

Key Insights

  • Intervention Paradox: Ignoring underreporting makes non-pharmaceutical interventions (NPIs) look more effective than they truly are. A perceived drop from 5.0 to 0.7 might lead to over-reliance on measures, whereas the reality (a drop from 1.7 to 0.7) suggests a more modest, though still significant, impact.
  • Behavioral Backlash: The model captured a secondary spike in transmission around late March. This coincided with Nowruz (Persian New Year), where despite school closures, millions of citizens traveled across provinces, spreading the virus from urban centers like Tehran to the rest of the country.
  • Reproduction Number Trends: The estimated $R_t$ in Iran fluctuated between 0.66 and 1.73, with a median of 1.16. It dropped from 1.73 on March 1 to 0.69 by April 15 due to the cumulative effect of school closures, bans on public gatherings, and partial business shutdowns.
  • ICU Fragility: Even under scenarios where $R_t$ remains below 1.0, the authors found that the demand for ICU beds would likely exceed Iran's national capacity, highlighting the "long tail" of the epidemic.

Limitations and Conclusion

The authors acknowledge several limitations, including the assumption of constant underreporting over time and the reliance on a baseline CFR of 1.4% borrowed from other regions. Furthermore, the model treats the population as relatively uniform, whereas age-stratified data might reveal different transmission risks among demographics.

In conclusion, while NPIs and public compliance were effective in slowing the spread of COVID-19 in Iran, the study emphasizes that accounting for underreporting is essential for accurate modeling. Overlooking these factors can mislead public health decisions and policy makers, especially during the critical early stages of an outbreak.

Summary updated

Abstract

SUMMARY Background Iran has been the hardest hit country by the outbreak of SARS-CoV-2 in the Middle East with 74,877 confirmed cases and 4,683 deaths as of 15 April 2020. With a relatively high case fatality ratio and limited testing capacity, the number of confirmed cases reported is suspected to suffer from significant under-reporting. Therefore, understanding the transmission dynamics of COVID-19 and assessing the effectiveness of the interventions that have taken place in Iran while accounting for the uncertain level of underreporting is of critical importance. We use a mathematical epidemic model utilizing official confirmed data and estimates of underreporting to understand how transmission in Iran has been changing between February and April 2020. Methods We developed a compartmental transmission model to estimate the effective reproduction number and its fluctuations since the beginning of the outbreak in Iran. We associate the variations in the effective reproduction number with a timeline of interventions and national events. The estimation method also accounts for the underreporting due to low case ascertainment by estimating the percentage of symptomatic cases using delay-adjusted case fatality ratio based on the distribution of the delay from hospitalization-to-death. Findings Our estimates of the effective reproduction number ranged from 0.66 to 1.73 between February and April 2020, with a median of 1.16. We estimate a reduction in the effective reproduction number during this period, from 1.73 (95% CI 1.60 – 1.87) on 1 March 2020 to 0.69 (95% CI 0.68-0.70) on 15 April 2020, due to various non-pharmaceutical interventions including school closures, a ban on public gatherings including sports and religious events, and full or partial closure of non-essential businesses. Based on these estimates and given that a near complete containment is no longer feasible, it is likely that the outbreak may continue until the end of the 2020 if the current level of physical distancing and interventions continue and no effective vaccination or therapeutic are developed and made widely available. Interpretation The series of non-pharmaceutical interventions and the public compliance that took place in Iran are found to be effective in slowing down the speed of the spread of COVID-19 within the studied time period. However, we argue that if the impact of underreporting is overlooked, the estimated transmission and control dynamics could mislead the public health decisions, policy makers, and general public especially in the earlier stages of the outbreak. Funding Nil.
BibTeX
@misc{saberi2020accounting,
  author = {Meead Saberi and Homayoun Hamedmoghadam and Kaveh Madani and Helen Dolk and Andreï S. Morgan and Joan K. Morris and Kaveh Khoshnood and Babak Khoshnood},
  title = {{Accounting for underreporting in mathematical modelling of transmission and control of COVID-19 in Iran}},
  year = {2020},
  doi = {10.1101/2020.05.02.20087270},
  howpublished = {medRxiv},
}
Figure