PREDICTIVE MODELS FOR DENGUE FEVER AND SEVERE DENGUE

Abstract

Introduction: Dengue is a major public health problem in tropical and subtropical countries but its clinical presentation may be similar to many febrile illnesses. Since in endemic countries laboratory confirmation is frequently delayed, the majority of dengue cases are diagnosed based on patient’s symptomatology. This can often lead to misdiagnosis and potential serious health complications. The objective of this study was to identify clinical, hematological and demographical parameters that could be used as predictors of dengue fever among patients with febrile illness. Methods: We conducted a retrospective cohort study of 548 patients presenting with febrile syndrome to the largest public hospitals in Honduras. Patients’ clinical, laboratory, and demographical data as well as dengue laboratory confirmation by either serology or viral isolation were used to build a predictive statistical model to identify dengue cases. Results: Of 548 patients, 390 were confirmed with dengue infection while 158 had negative results. Univariable analysis revealed seven variables associated with dengue: male sex, petechiae, skin rash, myalgia, retro-ocular pain, positive tourniquet test, and bleeding gums. In multivariable logistic regression analysis, retro-ocular pain petechiae and bleeding gums were associated with increased risk, while epistaxis and paleness of skin were associated with reduced risk of dengue. Using a value of 0.6 (i.e., 60% probability for a case to be positive based on the equation values), our model had a sensitivity of 86.2%, a specificity of 27.2%, and an overall accuracy of 69.2%; allowing for the diagnosis of dengue to be ruled out and for other febrile conditions to be investigated. Conclusions: The application of predictive models can be valuable when laboratory confirmation is delayed. Among Honduran patients presenting with febrile illness, our data reveal key symptoms associated with dengue fever, however the overall accuracy of our model is still low and specificity remains a concern. Our model requires validation in other populations with similar pattern of dengue transmission.

Key Words: Dengue, fever, Predictive model, symptoms, Honduras