Syndromic Surveillance-Based Estimates of Vaccine Efficacy Against COVID-Like Illness from Emerging Omicron and COVID-19 Variants

1 Introduction

New infections from the omicron variant of SARS-CoV-2 have been increasing dramatically in South Africa since first identification in November 2021, and despite increasing uptake of COVID-19 vaccine. The majority of confirmed cases have occurred in the Gauteng province [1]. Omicron carries several mutations that potentially contribute to increased transmissibility [2], as well as reduce immune protection. Initial reports show an increased risk of reinfection when facing the omicron variant of the SARS-CoV-2 virus [3]. However, these early claims are based on relatively small sample sizes. Moreover, a recent large scale analysis of COVID-19 test results within South Africa suggests that the current COVID-19 vaccines may not protect as well against the omicron variant [4]. In the setting of novel variants such as omicron, particularly in the early stages of national and international spread, there is a need for preliminary estimates of vaccine efficacy as a contributor to variant emergence and dominance.

To further understand the early emergence of the omicron variant in Gauteng, South Africa, we combine real-time syndromic surveillance data from daily cross-sectional surveys with a COVID-like illness (CLI) prevalence-based vaccine efficacy estimator (V E_CLIP).

2 Methods

3 Results

3.1 Derived CLI Definitions

To characterize the constellation of symptoms that can be used to identify COVID-19 within Gauteng, we first identified self-reported symptoms that have appeared to be rising during the delta and omicron waves: cough, fever, and muscle aches and pains (here referred to in short as myalgias). These symptoms were combined with loss of smell or taste (here referred to in short as anosmia), the symptom that is most highly associated with COVID-19 infection throughout the pandemic, and was a consistently strong predictor across countries, time, and syndromic surveillance platforms [10]. These definitions were termed stringent CLI (anosmia with fever, cough, and/or myalgias), classic CLI (cough with anosmia, fever, cough and/or myalgias), and broad CLI (myalgias with anosmia, fever, and/or cough). We illustrate how the prevalence among respondents to UMD-CTIS cross-sectional surveys, using these CLI definitions, track with COVID-19 cases from official case count data, using 7-day moving averages. Figure 1 suggests that the classic CLI definition (cough and at least one of the following: fever, muscle pain, anosmia), provides similar trends to reported cases. This is true across the delta wave time period when the Delta variant was dominant (June to September), and the first two weeks of December when Omicron infections surged. Further, we find that the stringent CLI definition that is designed to be more specific to COVID-19 – but also relies on a less frequently reported symptom (anosmia) – has a weaker signal during the omicron wave period infection than both the classic and broad CLI definitions that include more frequently reported symptoms. Because Omicron cases are rising in Gauteng at present, the signal-to-noise is changing. We therefore include all three CLI definitions for the vaccine efficacy estimates for comparison going forward.

• Download figure
• Open in new tab

Figure 1:

(Top Left) Proportion of UMD-CTIS survey respondents with self-reported COVID-like illness (CLI) in the prior 24 hours using three symptom-based definitions, 1) Stringent CLI: anosmia plus fever, cough, and/or myalgias, 2) Classic CLI: cough plus fever, anosmia, and/or myalgias, and 3) Broad CLI: myalgias plus fever, cough, and/or anosmia, over a 7-day rolling window. (Bottom Left) 7-day rolling average of confirmed new case counts of COVID-19. Gray shaded regions depict the time windows used for delta and omicron comparisons. (Top Right) V E_CLIP estimates and confidence intervals using 3 CLI definitions, and over the two periods. (Top Right) Map showing Africa, South Africa and Gauteng.

4 Discussion

The omicron variant is surging in Gauteng, South Africa, renewing concerns about the ability of current COVID-19 vaccines to protect against novel variants. Utilizing regional data drawn from a global syndromic surveillance platform, we provide early evidence that a relative decline in vaccine efficacy, as estimated by V E_CLIP, may be contributing to the omicron variant taking hold in Gauteng, and likely elsewhere. Other investigations ranging from prediction models of genetic changes, analysis of medical databases, and in vitro models have similarly raised concerns about the ability of Omicron to evade vaccine-induced immune response, transmit, and cause disease [2, 4, 11]. Evidence from recent lab based studies suggest that the Omicron variant has the potential to produce more breakthrough cases in 2-dose vaccinated individuals than previous SARS-CoV-2 variants [11], an is consistent with changes that we have observed in our proxy of vaccine efficacy. The South Africa study that analyzed over 200 thousand medical records suggests that the relative vaccine efficacy for omicron is about half that for the delta variant [4], lower than the decline observed using V E_CLIP, though the outcome measures in that study are not identical to the symptom-base outcomes utilized here.

This study highlights the value of participatory, remote surveillance as an adjunctive tool for gleaning rapid insights about epidemiologic parameters in the face of changing disease dynamics. Specifically, the UMD-CTIS platform – with its daily study base sampling scheme and broad geographic coverage of thousands of sub-national regions in over a hundred countries worldwide – provides an existing and ongoing data stream for this type of analysis. Previous studies have shown the value of syndromic surveillance platforms for COVID-19 and other infections within specific countries [6, 12]. The use of short surveys delivered to a daily sample of Facebook users globally provides an opportunity to glean initial insights rapidly and on a small geographic scale from an existing data stream, and without the use of regional infrastructure, and with the ability to scale up analyses to other regions as omicron spreads globally.

While these early observational results suggest that UMD-CTIS cross-sectional self-reported symptom and vaccination status reports can be leveraged as a proxy for COVID-19 attack rate in a population with an ongoing epidemic. All three of the CLI definitions support a decline in V E_CLIP, and thus a potential decline in vaccine efficacy in the Gauteng province. However, this estimated decline in V E_CLIP can plausibly be attributed to biological mechanisms: 1) increased penetration of the Omicron variant to the COVID-19 vaccine, and/or 2) waning vaccine induced immunity (or confounding by time since vaccination). The present study does not contain enough information to disambiguate between these two alternatives, though UMD-CTIS is updated periodically and may incorporate questions regarding vaccine timing should there be a public health need for these data.

The early results of declining V E_CLIP, from the delta wave period to this initial omicron wave period presented here on key assumptions, and should therefore be taken with caution. For instance, our analyses do not account for the vaccine formula that an individual received. This is problematic, as studies have shown varying levels of vaccine efficacy across different vaccine formulations. Further, while our stringent, classic, and broad definitions of CLI can account for potential heterogeneities in COVID-19 symptomatology, the definition is general enough to include seasonal influenza and other febrile respiratory illnesses, as well as non-infectious causes of symptoms. Recent reports of increased influenza throughout the month of November, suggest that influenza and other respiratory pathogens could be contributing to the recent observation of reduced V E_CLIP within Gauteng. Although we have examined individual symptoms and additional definitions of CLI, we find that the use of the classic CLI symptom sets generally captures COVID-19 case trends in Gauteng during this early study period. As cases rise, and the proportion of respondents with anosmia similarly rise, the more specific CLI definition requiring anosmia may be more useful for ensuring COVID-19 specificity, and limiting bias from non-specific symptom trends. With the clinical presentation of Omicron still unknown, the syndromic definitions applied in this study were not used to distinguish between different variants of COVID-19 and instead we referred to outside confirmed reports. Lastly, we are assuming the UMD-CTIS respondents who were vaccinated and unvaccinated during the delta and omicron wave periods can be considered to be exchangeable. Vaccine uptake is not random and the force of infection is heterogeneous; those with higher risk and higher means may get vaccine earlier and have immunity that wanes earlier, while vaccinated individuals may be both healthier or more willing to engage in social or unmasked activities. UMD-CTIS is a statistical sample of Facebook users, and there may be important limitations to generalizing effect estimates from a selected sample [13]. Further studies could address this by incorporating covariates – such as gender, age, illness duration, and the potential risk behaviors of survey respondents – to fully understand the potential biases, and better characterize the potential risks of emerging virus variants.

While the syndromic surveillance informed proxy of vaccine efficacy may not be the gold standard for estimating vaccine efficacy, we have shown that analysis of regional data from UMD-CTIS can quickly point towards high-level changes leveraging remote, real-time CLI trends survey respondents across time. Still, this initial report has only described a very limited set of questions that can be asked using UMD-CTIS to better understand the epidemiological profile of Omicron in Gauteng, South Africa and globally. As we continue to face SARS-CoV-2 variants, and seasonal and emerging infectious disease in general, it is crucial that we combine insights from lab-based studies, with early insights from syndromic surveillance data streams.

Data Availability Statement

Access to the CTIS data can be can be requested from Facebook Data for Good website (https://dataforgood.facebook.com/dfg/docs/covid-19-trends-and-impact-survey-request-for-data-access). The UMD Global CTIS Open Data API, Microdata Repository, and contingency tables are available from The University of Maryland Social Data Science Center Global COVID-19 Trends and Impact Survey website (https://covidmap.umd.edu).

Competing Interests

Authors report research grant funding from the Massachusetts Consortium on Pathogen Readiness (JSB), the Rockefeller Foundation (JSB), and the National Institutes of Health (CMA, K23 DK120899), during the conduct of the study. The authors have no relevant financial conflicts of interest to disclose.