Skip to main content
Fig. 6 | BMC Infectious Diseases

Fig. 6

From: Creating symptom-based criteria for diagnostic testing: a case study based on a multivariate analysis of data collected during the first wave of the COVID-19 pandemic in New Zealand

Fig. 6

Receiver Operator Curves (ROCs) showing the relationship between sensitivity and specificity for different cut-off values used to determine a recommendation to test for SARS-CoV-2 by PCR based on predicted probabilities of being a case from machine learning / random forest models. The dataset comprised 1125 cases and 4750 non-cases presenting for testing during the first wave of COVID-19 in New Zealand. For example, B shows the relationship between sensitivity and specificity for a random forest model including 15 symptom variables and age group. The optimal cut-off, 0.53 (i.e. the point closest to the top left corner), maximises the sensitivity and specificity of the decision to test by PCR, comparing the model predictions produced by the training set with the actual status (confirmed or not a case) in the test set. In this example the sensitivity is 84% and the specificity of 81%. The area under the curve (AUC) is also provided and illustrates how the predictions improve as more covariates are added from (A–D)

Back to article page