Quality of symptom-based diagnosis of rotavirus infection based on mathematical modeling

Rotavirus is the leading cause of severe childhood gastroenteritis worldwide. The laboratory diagnosis requires testing of fecal specimens with commercial assays that often are not available in low resource settings. Therefore, estimation of rotavirus presence based on clinical symptoms is expected to improve the disease management without laboratory verification. We aimed to develop and compare different mathematical approaches to model-based evaluation of expected rotavirus presence in patients with similar clinical symptoms. Two clinical datasets were used to develop clinical evaluation models of rotavirus presence or absence based on Bayesian network (BN), linear and nonlinear regression. The developed models produced different levels of reliability. BN compared with regression models showed better rotavirus detection results according to optimal cut-off points. Such approach is viable to help physicians refer patient to the group with suspected rotavirus infection to avoid unnecessary antibiotic treatment and to prevent rotavirus infection spread in a hospital ward.