Figure 1.

Comparison of supervised machine learning algorithms and summary of prospective evaluation of an SVM with 21 routine variables. (a) Comparison of supervised machine learning algorithms for the prediction of positive microbiology during COVID-19 pandemic. (b) Receiver operator characteristics (ROC) for an SVM used to predict positive microbiology within 48 h of admission to hospital. From a prospective pilot of 54 patients admitted to hospital between July and August 2020; 28/54 (52%) were SARS-CoV-2 PCR positive. (c) Summary of sensitivity and specificity from the ROC for an SVM using 21 blood test results to predict positive microbiology when different cut-off values are selected. ROC, area under the curve receiver operator characteristic; G-mean, geometric mean.