Files
Abstract
Background: Infectious mononucleosis (IM) is a common disease most commonly caused by the Epstein-Barr virus (EBV), and is most often seen in adolescents and children. Ordering a serological test for infectious mononucleosis (IM) in all patients with a sore throat is costly and impractical. Clinical prediction rules (CPRs) for the diagnosis of IM that combines symptoms, signs, and hematologic parameters may improve the diagnosis of IM and help clinicians prioritize diagnostic testing. Methods: We performed a systematic review and meta- analysis of the accuracy of the clinical signs, symptoms, hematologic parameters, and serological tests in patients with suspected IM. The test threshold was estimated using a convenience sample of US primary care physicians. We then used the structured data extracted from the electronic health records of a university health center between 2015 and 2019 to develop and validate the CPRs. The CPRs for the diagnosis of IM were developed using four statistical methods: traditional logistic regressions, fast and frugal trees (FFTs), classification and regression trees (CARTs), and artificial neural networks (ANNs). The CPRs were developed based on the clinical symptoms and signs with (IM-Lab) and without hematologic parameters (IM-Nolab) and were internally validated. Results: Based on our systematic review, the most helpful hematologic parameters for ruling in IM include lymphocytes greater than4×109/L, lymphocytes greater than 40% to 50%, or atypical lymphocytes greater than 40%. A combination of lymphocytes greater than 50% and atypical lymphocytes greater than 10% was also found to be helpful to rule in disease. Most of the individual clinical findings have limited diagnostic value in ruling out the disease when absent. We used data from clinical vignettes to the estimate a test threshold for IM of 9.5% (95% CI: 8.2% to 10.9%), we identified the probability of IM in the low- and high-risk groups as 8.8% and 31.2% for IM-Nolab logistic regression model (AUC=0.76); 4% and 79.4% for IM-Lab logistic regression model (AUC-0.94); 7.3% and 32.2% for IM-Nolab CART model (AUC=0.69); 5.9% and 61.8% for IM-Lab CART model (AUC=0.93); 8.2% and 33.5% for IM-Nolab FFT model (AUC=0.71); 5% and 68.2% for IM-Lab FFT model (AUC=0.94); 8.8% and 50.4% for IM-Nolab ANN model (AUC=0.70); and 4.4% and 69.3% for IM-Lab ANN model (AUC=0.97). The discrimination plots showed good discriminations for the IM-Nolab models and excellent discriminations for the IM-Lab models. The Calibration plots in the validation groups showed fair agreement between our predicted outcome and the observed test results for both IM-Nolab and IM-Lab models using all statistical methods. Conclusion: The derived IM-Lab and IM-Nolab models provided useful tools to help clinicians make a rapid diagnosis of IM. The IM-Nolab score has potential utility in telehealth visits, but the IM-Lab score provides a more accurate result. When externally validated, such risk scores would be useful for improving the diagnosis of IM and helping clinicians prioritize diagnostic testing.