Spatial assessment of disease surveillance during the first wave of the COVID-19 pandemic

Abstract

<b>Background:</b> Early surveillance of COVID-19 in Scotland included routine monitoring of positive test rates and COVID-19-related NHS 24 calls. The COVID Symptom Study (CSS) provides another surveillance source, collating self-reported symptoms in the general population and predictions of likely infection. <b>Aim:</b> To capture spatial patterns of COVID-19 infection using Spatio-temporal (ST) analyses on three data streams: positive test rates, NHS24 calls, and CSS predicted cases. These were compared to assess which was best for early disease surveillance. <b>Methods:</b> Data streams recorded weekly counts of activity by postcode district (PCD) during the first wave of the pandemic. ST analyses assessed the relationship between COVID-19 testing, NHS 24 COVID-19 calls, and CSS predicted COVID-19 cases, applying a Leroux conditional auto-regression (CAR) spatial GLM, adjusting for spatial covariates. <b>Results:</b> Positive test rates were associated with the proportion of NHS 24 calls related to COVID-19 per PCD (OR=1.038, 95% credible interval, 1.024-1.052) and the proportion of CSS app users predicted as cases, (OR=1.014, 0.974-1.056). A temporal effect was seen between all streams, after adjusting for spatial covariates.&nbsp;Using both NHS24 and the CSS to model COVID-19 positive test rates accounted for more ST variability than with the separate models, implying that combining sources may improve surveillance accuracy. <b>Conclusions:</b> NHS 24 and the CSS can identify similar trends/clusters of COVID-19 and gold-standard testing data, particularly when used in parallel. In the early stages of a pandemic, when widespread testing might not be available, alternative sources of data may be used to inform outbreak management.