Fig. 3

Applied BN v2.2 performance as compared with observations, with Log Loss and AUROC across four scenarios. Each panel presented the distribution of the Applied BN predicted probabilities of isolating E.coli from urine sample given available patient’s information under the specified scenario. The predicted probabilities were compared with the reported culture result of each patient, where brown, blue and grey indicated E.coli was isolated, not isolated and no data, respectively. Scenario (a): age, sex, history of UTI, urinary tract comorbidities. Scenario (b): scenario (a) + reported diarrhoea, urine tract pain or discomfort, abdominal pain, haematuria, foul smelling urine, respiratory symptoms, parent reported fever, temperature, irritability, lethargy, nausea/vomiting, poor oral intake. Scenario (c): scenario (b) + urine collection methods, urine dipstick results (leucocyte esterase & nitrite). Scenario (d): scenario (c) + urine microscopy (leucocytes, bacteria, epithelial cells), leucocyte and neutrophil count (from full blood count), C-reactive protein level and ultrasound result