Local prevalence of transmissible SARS-CoV-2 infection: an integrative causal model for debiasing fine-scale targeted testing data

George Nicholson, Brieuc Lehmann, Tullia Padellini, Koen B. Pouwels, Radka Jersakova, James W. Lomax, Ruairidh King, Ann‐Marie Mallon, Peter J. Diggle, Sylvia Richardson, Marta Blangiardo, Chris HolmesmedRxiv 2021 · DOI

Abstract

Targeted surveillance testing schemes for SARS-CoV-2 focus on certain subsets of the population, such as individuals experiencing one or more of a prescribed list of symptoms. These schemes have routinely been used to monitor the spread of SARS-CoV-2 in countries across the world. The number of positive tests in a given region can provide local insights into important epidemiological parameters, such as prevalence and effective reproduction number. Moreover, targeted testing data has been used inform the deployment of localised non-pharmaceutical interventions. However, surveillance schemes typically suffer from ascertainment bias; the individuals who are tested are not necessarily representative of the wider population of interest. Here, we show that data from randomised testing schemes, such as the REACT study in the UK, can be used to debias fine-scale targeted testing data in order to provide accurate localised estimates of the number of infectious individuals. We develop a novel, integrative causal framework that explicitly models the process underlying the selection of individuals for targeted testing. The output from our model can readily be incorporated into longitudinal analyses to provide local estimates of the reproduction number. We apply our model to characterise the size of the infectious population in England between June 2020 and January 2021. Our local estimates of the effective reproduction number are predictive of future changes in positive case numbers. We also capture local increases in both prevalence and effective reproductive number in the South East from November 2020 to December 2020, reflecting the spread of the Kent variant. Our results illustrate the complementary roles of randomised and targeted testing schemes. Preparations for future epidemics should ensure the rapid deployment of both types of schemes to accurately monitor the spread of emerging and ongoing infectious diseases.

Introduction: The Challenge of Ascertainment Bias

Throughout the COVID-19 pandemic, public health decisions have heavily relied on "targeted" testing data (known in the UK as Pillar 1 and 2). In these schemes, individuals are tested because they exhibit symptoms or belong to high-risk groups. While these datasets provide high-volume information, they suffer from a fundamental ascertainment bias: because testing is non-random, the positivity rate is significantly higher than the actual prevalence in the general population.

Conversely, randomized surveillance studies like the UK’s REACT-1 provide an unbiased "gold standard." However, these studies are expensive and often lack the spatial resolution needed for local policy—they may provide regional insights but fail to capture neighborhood-level dynamics. This paper introduces an integrative causal framework designed to "debias" high-resolution targeted data by calibrating it against randomized survey results.

Methodology: An Integrative Causal Framework

The researchers developed a model based on a Directed Acyclic Graph (DAG) to explicitly represent the selection process for testing. This framework models how infection status leads to symptoms, and how those symptoms (or other factors) influence the decision to be tested.

Causal Directed Acyclic Graphs (DAGs) for (a) targeted testing and (b) randomized surveillance, illustrating the different selection mechanisms.

Key technical components of the model include:

  • The Bias Parameter ($\delta$): Defined as the log odds ratio of an infected person being tested versus a non-infected person.
  • Empirical Bayes (EB) Priors: The bias parameter is estimated at a coarse geographical level where randomized data is available and then applied as a "temporally smooth" prior to debias data at the fine-scale level of Lower Tier Local Authorities (LTLAs).
  • PCR-to-Infectious Mapping: To account for the fact that PCR tests can remain positive long after a person is no longer contagious, the model incorporates viral shedding and generation time data to estimate the number of actually infectious individuals.

Data Integration and Calibration

The study utilized several key UK datasets from June 2020 to January 2021, including Pillar 1 and 2 PCR counts. These were calibrated against randomized data from the REACT-1 study and the ONS COVID-19 Infection Survey. To validate the model's epidemiological relevance, the resulting $R_t$ estimates were compared against S-gene target failure (SGTF) data, which tracked the emergence of the Alpha (Kent) variant.

Key Results: Precision and Predictive Power

The model successfully reduced average estimated bias to near zero (-0.07%) when validated against independent REACT survey rounds. By combining the two data sources, the model produced prevalence estimates that were consistent with randomized surveys but featured much tighter confidence intervals, offering higher precision for local decision-makers.

Validation of the model against REACT round 8 data, illustrating the relationship between model estimates and survey results.

Furthermore, the debiased $R_t$ estimates proved highly predictive of future case changes. The Spearman’s correlation ($\rho$) between current $R_t$ and case growth one week later reached 0.68 in high-prevalence areas. Notably, the model captured local increases in prevalence and $R_t$ in the South East of England weeks before they became apparent in raw case counts, demonstrating its utility as an early-warning system.

Discussion and Limitations

While the model provides a powerful tool for debiasing, it relies on the assumption of spatial homogeneity—that the bias parameter ($\delta$) is relatively uniform across small areas within a larger region. If testing behavior varies drastically between neighboring towns, accuracy may be affected. Additionally, the current model does not explicitly account for demographic factors like age or socio-economic status, which are known to influence both infection risk and testing uptake. Future iterations of the model could benefit from a stratified analysis to address these granularities.

Summary updated

Abstract

Abstract Targeted surveillance testing schemes for SARS-CoV-2 focus on certain subsets of the population, such as individuals experiencing one or more of a prescribed list of symptoms. These schemes have routinely been used to monitor the spread of SARS-CoV-2 in countries across the world. The number of positive tests in a given region can provide local insights into important epidemiological parameters, such as prevalence and effective reproduction number. Moreover, targeted testing data has been used inform the deployment of localised non-pharmaceutical interventions. However, surveillance schemes typically suffer from ascertainment bias; the individuals who are tested are not necessarily representative of the wider population of interest. Here, we show that data from randomised testing schemes, such as the REACT study in the UK, can be used to debias fine-scale targeted testing data in order to provide accurate localised estimates of the number of infectious individuals. We develop a novel, integrative causal framework that explicitly models the process underlying the selection of individuals for targeted testing. The output from our model can readily be incorporated into longitudinal analyses to provide local estimates of the reproduction number. We apply our model to characterise the size of the infectious population in England between June 2020 and January 2021. Our local estimates of the effective reproduction number are predictive of future changes in positive case numbers. We also capture local increases in both prevalence and effective reproductive number in the South East from November 2020 to December 2020, reflecting the spread of the Kent variant. Our results illustrate the complementary roles of randomised and targeted testing schemes. Preparations for future epidemics should ensure the rapid deployment of both types of schemes to accurately monitor the spread of emerging and ongoing infectious diseases.
BibTeX
@misc{nicholson2021local,
  author = {George Nicholson and Brieuc Lehmann and Tullia Padellini and Koen B. Pouwels and Radka Jersakova and James W. Lomax and Ruairidh King and Ann‐Marie Mallon and Peter J. Diggle and Sylvia Richardson and Marta Blangiardo and Chris Holmes},
  title = {{Local prevalence of transmissible SARS-CoV-2 infection: an integrative causal model for debiasing fine-scale targeted testing data}},
  year = {2021},
  doi = {10.1101/2021.05.17.21256818},
  howpublished = {medRxiv},
}
Figure