Relative role of border restrictions, case finding and contact tracing in controlling SARS-CoV-2 in the presence of undetected transmission

  1. Rachael Pung
  2. Hannah E. Clapham
  3. Vernon J. Lee
  4. Adam J Kucharski

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This paper will be of interest to infectious disease epidemiologists and modellers working on situational assessment, and public health researchers focused on COVID19 response. Through an informative case study on the Singaporean COVID19 epidemic, the paper provides estimates of case ascertainment under different levels of border restrictions and public health measures, as well as estimates of the effectiveness of contact tracing in reducing transmission. The combination of data from multiple sources with mathematical modeling provides a powerful tool to assess effectiveness of interventions. Some of the key claims of the manuscript - while plausible - are not directly supported by the analyses, and the methods and modelling assumptions require more detailed exposition.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #3 agreed to share their name with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Background

Several countries have controlled the spread of COVID-19 through varying combinations of border restrictions, case finding, contact tracing and careful calibration on the resumption of domestic activities. However, evaluating the effectiveness of these measures based on observed cases alone is challenging as it does not reflect the transmission dynamics of missed infections.

Methods

Combining data on notified local COVID-19 cases with known and unknown sources of infections (i.e. linked and unlinked cases) in Singapore in 2020 with a transmission model, we reconstructed the incidence of missed infections and estimated the relative effectiveness of different types of outbreak control. We also examined implications for estimation of key real-time metrics — the reproduction number and ratio of unlinked to linked cases, using observed data only as compared to accounting for missed infections.

Findings

Prior to the partial lockdown in Singapore, initiated in April 2020, we estimated 89% (95%CI 75–99%) of the infections caused by notified cases were contact traced, but only 12.5% (95%CI 2–69%) of the infections caused by missed infectors were identified. We estimated that the reproduction number was 1.23 (95%CI 0.98–1.54) after accounting for missed infections but was 0.90 (95%CI 0.79-1.1) based on notified cases alone. At the height of the outbreak, the ratio of missed to notified infections was 34.1 (95%CI 26.0–46.6) but the ratio of unlinked to linked infections was 0.81 (95%CI 0.59–1.36). Our results suggest that when case finding and contact tracing identifies at least 50% and 20% of the infections caused by missed and notified cases respectively, the reproduction number could be reduced by more than 14%, rising to 20% when contact tracing is 80% effective.

Interpretation

Depending on the relative effectiveness of border restrictions, case finding and contact tracing, unobserved outbreak dynamics can vary greatly. Commonly used metrics to evaluate outbreak control — typically based on notified data — could therefore misrepresent the true underlying outbreak.

Funding

Ministry of Health, Singapore.

Research in context

Evidence before this study

We searched PubMed, BioRxiv and MedRxiv for articles published in English up to Mar 20, 2021 using the terms: (2019-nCoV OR “novel coronavirus” OR COVID-19 OR SARS-CoV-2) AND (border OR travel OR restrict* OR import*) AND (“case finding” OR surveillance OR test*) AND (contact trac*) AND (model*). The majority of modelling studies evaluated the effectiveness of various combinations of interventions in the absence of outbreak data. For studies that reconstructed the initial spread of COVID-19 with outbreak data, they further simulated counterfactual scenarios in the presence or absence of these interventions to quantify the impact to the outbreak trajectory. None of the studies disentangled the effects of case finding, contact tracing, introduction of imported cases and the reproduction number, in order to reproduce an observed SARS-CoV-2 outbreak trajectory.

Added value of this study

Notified COVID-19 cases with unknown and known sources of infection are identified through case finding and contact tracing respectively. Their respective daily incidence and the growth rate over time may differ. By capitalising on these differences in the outbreak data and the use of a mathematical model, we could identify the key drivers behind the growth and decline of both notified and missed COVID-19 infections in different time periods — e.g. domestic transmission vs external introductions, relative role of case finding and contact tracing in domestic transmission. Estimating the incidence of missed cases also allows us to evaluate the usefulness of common surveillance metrics that rely on observed cases.

Implications of all the available evidence

Comprehensive outbreak investigation data integrated with mathematical modelling helps to quantify the strengths and weaknesses of each outbreak control intervention during different stages of the pandemic. This would allow countries to better allocate limited resources to strengthen outbreak control. Furthermore, the data and modelling approach allows us to estimate the extent of missed infections in the absence of population wide seroprevalence surveys. This allows us to compare the growth dynamics of notified and missed infections as reliance on the observed data alone may create the illusion of a controlled outbreak.

Article activity feed

  1. eLife

    Author Response

    Reviewer #1 (Public Review):

    The authors estimated the effectiveness of border restriction, testing and contact tracing in managing transmission of Covid19, and in detecting "missed" Covid19 cases. They developed a standard branching process model to disentangle the effects of each control measure on the incidence of missed infections, by fitting their model to data on cases both at the border and in the community of Singapore, from the beginning of the Covid19 outbreak until December 2021.

    The Authors modelled detected and undetected community infections as two separate branches of the transmission tree. They then fitted their model to the observed incidence data to obtain an estimate of the number of missed infections. Through this method, they explained the importance and contribution of case ascertainment (through testing and contact tracing), as well as border and community restrictions, towards transmission reduction.

    This modelling and inference framework could be applied to data from anywhere in the world to estimate the number of undetected infections when in lack of infection prevalence data.

    Strengths:

    For each of the five phases of border and public health measures put in place in Singapore in 2021, the authors successfully provided estimates on the number of undetected community cases, effectiveness of contact tracing and testing in finding unlinked cases, and effective reproduction numbers of both detected and undetected cases. All these estimates can be valuable to Covid modellers worldwide to either benchmark or parametrise their own model parameters, and to Singapore's public health officials to decide on future strategies of transmission prevention.

    Estimating infection prevalence and case ascertainment rates is one of the main challenges of Covid modellers everywhere. The authors' method to reconstruct the transmission tree of both known infections and undetected ones, and the subsequent fitting to observed data, could be used to estimate case ascertainment rate in the absence of prevalence surveys.

    The authors also found that contact tracing is only useful for transmission reduction when coupled with a high rate of case ascertainment. This is a well-known but important result, highlighting the need for more timely and accessible community testing.

    Weaknesses:

    The authors' models and estimates are mostly well supported by data, but the Methods need to be clarified and extended, and the results could be presented in a clearer way.

    The transmission model in particular needs to be presented in a more detailed way to avoid confusion around the modelling assumptions and to allow easy reproduction of the model by the reader.

    Thank you for this comment. As there were several mathematical notations required, we have compiled all parameters and variables in an Appendix Table 1 (tracked edits page 28 and 29) and also indicated within the table the assumed parameters, distributions and priors, unknown parameters to be modelled and derived parameters. The details of the model parameters and assumptions were also introduced in the main text based on the sequence of model building and simulating disease transmission, so we hope that this additional table will facilitate better understanding of the model framework.

    It would also be very useful to readers to visualize the different restriction measures in place together with the result graphs, to strengthen the link between the two and to highlight the effect that different border and public health regimes have on transmission and on the proportion of undetected infections, which the authors mention in the main text.

    Thank you for this comment, we have amended Figure 1 to incorporate short labels on key outbreak events or control measures to help the reader understand the changes in epidemic trajectory. Further details are also documented in the Figure 1-figure supplement 2 (a tabulated version of the previous Appendix 1 Figure 5 - studied time periods for wild-type SARS-CoV-2 and Delta variant outbreaks)

    While these results can definitely help the Singapore decision-makers design an efficient transmission control strategy, this paper could also be useful to researchers abroad. It is therefore important that the model is explained more clearly and that results and assumptions be benchmarked against those from some other country.

    Thank you for this comment. We have compared our findings on the burden of disease for SARS-CoV-2 wild type with those from other countries in the initial submission and have also expanded our Discussion section for findings related to the Delta variant as more studies were published since our initial submission. The amendments are as follows:

    “We estimated that the risk of ICU and fatality was 1.2% and 0.3% among wild-type SARS-CoV-2 infections across the study periods in 2020 and 0.2% and 0.2% among the Delta variant infections across the study periods in 2021 and our wild-type infection fatality ratio estimates corroborated with early studies in other countries and regions Centre for Disease Control and Prevention (2020), Brazeau et al. (2020); Myerowitz-Katz and Merone (2020).Early in the outbreak of Delta cases in Singapore (Apr 1 to May 12, 2021), more than 60% of the cases were unvaccinated. The risk of ICU and mortality among all notified cases during this period was 1.7% and 1.4%. When accounting for missed infections, the risk of ICU and mortality among all infections was 0.3% and 0.2%. This infection fatality ratio is comparable with that of the wild-type SARS-CoV-2. Over 90% of the fatalities occurred in the elder aged 60 and above and the 95%CI of the infection fatality ratio was 0.03–0.3% which overlaps considerably with the infection fatality ratio in the elderly during the H1N1 influenza pandemic in 2009 Riley et al. (2011); Wong et al. (2013). In most studies, the case fatality ratio are more commonly reported. This was estimated to be 3.4% in South Africa Sigal et al. (2022) and some studies reported approximately two times higher risk of death when compared to the wild type Li et al. (2021b). The healthcare system in these countries were under pressure arising from the surge in Delta variant cases Detsky and Bogoch (2021); Maslo et al. (2022) which potentially contribute to a higher case fatality ratio. Our estimates of comparable infection fatality ratio for the wild-type and Delta variant in the overall population could also be attributed to better clinical management of COVID-19 cases over time, and availability of new pharmaceuticals Beigel et al. (2020); Goldman et al. (2020); Ohl et al. (2021). Despite the lowered burden of infection estimates, it is prudent to vaccinate a large majority of the population before relaxing COVID-19 control measures to keep the absolute number of deaths arising from a highly transmissible Delta variant low Li et al. (2021).”

    Reviewer #2 (Public Review):

    This work combines multiple data sources and a branching-process mathematical model to assess the effectiveness of specific types of COVID-19 interventions including contact tracing, border screening, and case finding. The focus is on the original SARS-CoV-2 (2020) and the Delta (2021) variant outbreaks in Singapore. Utilizing data on both linked and unlinked cases, the model is also used to predict the total number of missed infections.

    Strengths:

    The study provides a way to utilize data to understand the importance of various simultaneously employed intervention strategies throughout the pandemic. Given the constantly evolving state of the epidemic, this retrospective study provides insight into what interventions worked under which conditions. This will be valuable for policymakers to understand what types of strategies to prioritize.

    The underlying model formulation has been used previously to understand components of transmission during the COVID-19 pandemic. Case data from Singapore is used to fit the model. Model conclusions on the number of true infections are consistent with a published seroprevalence study.

    Weaknesses:

    The paper currently provides an incomplete description of the model. Multiple terms (e.g. cases vs infections) are used throughout the manuscript to have a precise meaning, but that is not apparent until reaching the Methods section at the end. As these terms could be misinterpreted, a precise definition should come earlier. More information is needed regarding the parameters. It is unclear which parameters were fit, and no parameter values were given. For the lognormal, a mean is given, but no standard deviation.

    Thank you for the comments. We have added a short summary on the model structure and definition of key terms in the last paragraph of the Introduction section but referenced the Methods and Materials section for further elaboration. Furthermore, we have added a list of mathematical notations which contains both unknown parameters to be modelled, derived parameters and the observed distributions and their corresponding parameters.

    The identifiability of the model should be discussed. Given the wide range of some confidence intervals, it does seem that parameter identifiability could be a problem. The extent of this is hard to assess given the level of information on the parameters currently given.

    Thank you for this comment. Within the section “Effectiveness of case finding and contact tracing”, we have added a final line as follows:

    Across all time periods in 2020, ϵcf exhibits wide confidence interval as a result of some correlation with the factor, ρ, which scales the extent of missed imported infections (Figure 2-figure supplement 3).

    Furthermore, under the discussion section, we have, in the original submission, provided suggestions on how to overcome this challenge.

    Furthermore, unlinked cases were generated by either missed imported or local infections with the former modelled as a factor of the notified imported cases, ρ. As such, the interaction of model parameters results in ϵcf estimates characterised by wide 95% credible intervals. To improve these estimates, we could further stratify exposure histories of unlinked cases by their interactions with travellers from countries with ongoing outbreak for model fitting.

    Reviewer #3 (Public Review):

    This paper presents a new mathematical method to estimate case ascertainment (the fraction of infections identified as cases) and applies it to COVID-19 data from Singapore over the period early 2020 to late (pre-Omicron) 2021.

    The method relies upon access to line-listed case data that includes whether or not a case was an importation or was (epidemiologically) linked or unlinked.

    Through the application of this method, new results on the contribution of case identification and contact tracing to reducing transmission are derived. Reproduction numbers for different classes of infections (importations, linked, unlinked) are computed.

    A sensitivity analysis establishes that inclusion of the richer case line-list information results in tighter credible intervals for key parameters of interest, although there is no 'gold standard' on which to evaluate if those estimates are more accurate.

    Of some interest, the results suggest that both the case- and infection-fatality-ratios were lower during the Delta wave in 2021 than the ancestral wave in 2020. The underlying reasons (particularly for the IFR) are unclear and not investigated, but are perhaps due to vaccination (although the paper reports approximately just 35% vaccination coverage in early 2021), age-specific effects or other socio-demographic effects.

    The method is based on a branching process model, for which the key elements are well described (noting some (likely) typographical errors in the typesetting of the equations), although the exact details of the computational implementation are not described, leading to some ambiguity in how the method is implemented.

    Thank you very much for the overall comments and suggestions.

  2. eLife

    Evaluation Summary:

    This paper will be of interest to infectious disease epidemiologists and modellers working on situational assessment, and public health researchers focused on COVID19 response. Through an informative case study on the Singaporean COVID19 epidemic, the paper provides estimates of case ascertainment under different levels of border restrictions and public health measures, as well as estimates of the effectiveness of contact tracing in reducing transmission. The combination of data from multiple sources with mathematical modeling provides a powerful tool to assess effectiveness of interventions. Some of the key claims of the manuscript - while plausible - are not directly supported by the analyses, and the methods and modelling assumptions require more detailed exposition.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #3 agreed to share their name with the authors.)

  3. eLife

    Reviewer #1 (Public Review):

    The authors estimated the effectiveness of border restriction, testing and contact tracing in managing transmission of Covid19, and in detecting "missed" Covid19 cases. They developed a standard branching process model to disentangle the effects of each control measure on the incidence of missed infections, by fitting their model to data on cases both at the border and in the community of Singapore, from the beginning of the Covid19 outbreak until December 2021.

    The Authors modelled detected and undetected community infections as two separate branches of the transmission tree. They then fitted their model to the observed incidence data to obtain an estimate of the number of missed infections. Through this method, they explained the importance and contribution of case ascertainment (through testing and contact tracing), as well as border and community restrictions, towards transmission reduction.

    This modelling and inference framework could be applied to data from anywhere in the world to estimate the number of undetected infections when in lack of infection prevalence data.

    Strengths:
    For each of the five phases of border and public health measures put in place in Singapore in 2021, the authors successfully provided estimates on the number of undetected community cases, effectiveness of contact tracing and testing in finding unlinked cases, and effective reproduction numbers of both detected and undetected cases. All these estimates can be valuable to Covid modellers worldwide to either benchmark or parametrise their own model parameters, and to Singapore's public health officials to decide on future strategies of transmission prevention.
    Estimating infection prevalence and case ascertainment rates is one of the main challenges of Covid modellers everywhere. The authors' method to reconstruct the transmission tree of both known infections and undetected ones, and the subsequent fitting to observed data, could be used to estimate case ascertainment rate in the absence of prevalence surveys.
    The authors also found that contact tracing is only useful for transmission reduction when coupled with a high rate of case ascertainment. This is a well-known but important result, highlighting the need for more timely and accessible community testing.

    Weaknesses:
    The authors' models and estimates are mostly well supported by data, but the Methods need to be clarified and extended, and the results could be presented in a clearer way.
    The transmission model in particular needs to be presented in a more detailed way to avoid confusion around the modelling assumptions and to allow easy reproduction of the model by the reader.
    It would also be very useful to readers to visualize the different restriction measures in place together with the result graphs, to strengthen the link between the two and to highlight the effect that different border and public health regimes have on transmission and on the proportion of undetected infections, which the authors mention in the main text.
    While these results can definitely help the Singapore decision-makers design an efficient transmission control strategy, this paper could also be useful to researchers abroad. It is therefore important that the model is explained more clearly and that results and assumptions be benchmarked against those from some other country.

  4. eLife

    Reviewer #2 (Public Review):

    This work combines multiple data sources and a branching-process mathematical model to assess the effectiveness of specific types of COVID-19 interventions including contact tracing, border screening, and case finding. The focus is on the original SARS-CoV-2 (2020) and the Delta (2021) variant outbreaks in Singapore. Utilizing data on both linked and unlinked cases, the model is also used to predict the total number of missed infections.

    Strengths:
    The study provides a way to utilize data to understand the importance of various simultaneously employed intervention strategies throughout the pandemic. Given the constantly evolving state of the epidemic, this retrospective study provides insight into what interventions worked under which conditions. This will be valuable for policymakers to understand what types of strategies to prioritize.

    The underlying model formulation has been used previously to understand components of transmission during the COVID-19 pandemic. Case data from Singapore is used to fit the model. Model conclusions on the number of true infections are consistent with a published seroprevalence study.

    Weaknesses:
    The paper currently provides an incomplete description of the model. Multiple terms (e.g. cases vs infections) are used throughout the manuscript to have a precise meaning, but that is not apparent until reaching the Methods section at the end. As these terms could be misinterpreted, a precise definition should come earlier. More information is needed regarding the parameters. It is unclear which parameters were fit, and no parameter values were given. For the lognormal, a mean is given, but no standard deviation.

    The identifiability of the model should be discussed. Given the wide range of some confidence intervals, it does seem that parameter identifiability could be a problem. The extent of this is hard to assess given the level of information on the parameters currently given.

  5. eLife

    Reviewer #3 (Public Review):

    This paper presents a new mathematical method to estimate case ascertainment (the fraction of infections identified as cases) and applies it to COVID-19 data from Singapore over the period early 2020 to late (pre-Omicron) 2021.

    The method relies upon access to line-listed case data that includes whether or not a case was an importation or was (epidemiologically) linked or unlinked.

    Through the application of this method, new results on the contribution of case identification and contact tracing to reducing transmission are derived. Reproduction numbers for different classes of infections (importations, linked, unlinked) are computed.

    A sensitivity analysis establishes that inclusion of the richer case line-list information results in tighter credible intervals for key parameters of interest, although there is no 'gold standard' on which to evaluate if those estimates are more accurate.

    Of some interest, the results suggest that both the case- and infection-fatality-ratios were lower during the Delta wave in 2021 than the ancestral wave in 2020. The underlying reasons (particularly for the IFR) are unclear and not investigated, but are perhaps due to vaccination (although the paper reports approximately just 35% vaccination coverage in early 2021), age-specific effects or other socio-demographic effects.

    The method is based on a branching process model, for which the key elements are well described (noting some (likely) typographical errors in the typesetting of the equations), although the exact details of the computational implementation are not described, leading to some ambiguity in how the method is implemented.

  6. ScreenIT

    SciScore for 10.1101/2021.05.05.21256675: (What is this?)

    Please note, not all rigor criteria are appropriate for all manuscripts.

    Table 1: Rigor

    NIH rigor criteria are not applicable to paper type.

    Table 2: Resources

    No key resources detected.

    Results from OddPub: Thank you for sharing your code and data.

    Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:
    There are some limitations to our study. Firstly, the model assumes that each of the four parameters remains constant in a specified time period. As such, we are unable to provide a time-varying measure to characterise the impact of different outbreak detection and control measures that were progressively rolled out in the population at a granular level. Instead, time periods were chosen based on prior knowledge of major policies that would affect at least one of the four model parameters. Secondly, those imported cases subjected to home quarantine and no quarantine were assumed to have the same potential of making contact with members of the community as household transmission might occur, while in reality the amount of community contact would be different as those under home quarantine could potentially only contact household members. In the absence of data, we made a conservative assumption that the imported cases under home quarantine were capable of generating local infections. Further model calibration would require data on the outcome of the various quarantine measures. Finally, missed infections could arise from asymptomatic or mildly symptomatic infections, or underreporting of symptomatic cases. We assumed that R is the same among these cases. More information is needed to determine the temporal variation in the types of cases to account for lowered transmission potential in asymptomatic or mildly symptomatic cases. At the same time, a key strength to our analysis i...

    Results from TrialIdentifier: No clinical trial numbers were referenced.

    Results from Barzooka: We did not find any issues relating to the usage of bar graphs.

    Results from JetFighter: We did not find any issues relating to colormaps.

    Results from rtransparent:
    • Thank you for including a conflict of interest statement. Authors are encouraged to include this statement when submitting to a journal.
    • Thank you for including a funding statement. Authors are encouraged to include this statement when submitting to a journal.
    • No protocol registration statement was detected.

    Results from scite Reference Check: We found no unreliable references.

    About SciScore

    SciScore is an automated tool that is designed to assist expert reviewers by finding and presenting formulaic information scattered throughout a paper in a standard, easy to digest format. SciScore checks for the presence and correctness of RRIDs (research resource identifiers), and for rigor criteria such as sex and investigator blinding. For details on the theoretical underpinning of rigor criteria and the tools shown here, including references cited, please follow this link.

  7. Version published to 10.1101/2021.05.05.21256675v1 on medRxiv