Fig. 2: Generic model classification performance.

Median (grey dots) and interquartile ranges for receiver operating characteristics (ROC) for rapid antigen testing only approach (purple) and posterior median and interquartile range ROC for Syndromic-only (teal) and Syndromic-Rapid Antigen Test (RAT) Combined (yellow) models for n = 1172 biologically independent individuals predicted under temporally structured cross-validation. In the RAT-only model, the ROC is a single value (i.e., a dot rather than a line) as the binary test has a single sensitivity and specificity. In Syndromic-only and Syndromic-RAT Combined classes, the ROC values demonstrate the performance of the model for any hypothetical scenario as defined by the axes (as opposed to Fig. 5 which demonstrates model performance in specific epidemiological scenarios which are realisations of single points in this space). While ROC plots are often plotted as curves, we do not have continuous probability values due to the binary nature of predictor symptoms. This is important as discontinuity in the probabilities impacts the sensitivity of the model to classification thresholds, such as those used in the scenarios below.