Abstract:
This article describes a new method for estimating weekly incidence (new onset) of symptoms consistent with Influenza and COVID-19, using data from the Flutracking survey. The method mitigates some of the known self-selection and symptom-reporting biases present in existing approaches to this type of participatory longitudinal survey data. The key novel steps in the analysis are: <b>1)</b> Identifying new onset of symptoms for three different Symptom Groupings: COVID-like illness (CLI1+, CLI2+), and Influenza-like illness (ILI), for responses reported in the Flutracking survey. <b>2)</b> Adjusting for <i>symptom reporting bias</i> by restricting the analysis to a sub-set of responses from those participants who have consistently responded for a number of weeks prior to the analysis week. <b>3)</b> Weighting responses by age to adjust for <i>self-selection bias</i> in order to account for the under- and over-representation of different age groups amongst the survey participants. This uses the survey package [22] in R [30]. <b>4)</b> Constructing 95% point-wise confidence bands for incidence estimates using weighted logistic regression from the survey package [21] in R [28]. In addition to describing these steps, the article demonstrates an application of this method to Flutracking data for the 12 months from 27th April 2020 until 25th April 2021.
