Our study reported that older age, the absence of diarrhea, leukopenia, thrombocytopenia, and normal ESR are significant variables with excellence in discriminating dengue and other fever illnesses. Dengue patients are older, which is similar to other findings [7, 13,14,15]. This can be explained as older children often work in open fields during the day when Aedes mosquitoes are active, making them prone to Aedes bites [16].
The absence of diarrhea is also found to be significant in differentiating dengue from other febrile illnesses. Previous studies have reported that dengue patients are less likely to experience diarrhea [7, 13]. A study reported that little is known about the relationship between diarrhea and dengue [15], meanwhile diarrhea is known to be the symptoms of acute uncomplicated diarrhea and dysentery [9].
This study found leukopenia is more severe in the dengue group compared to other febrile illnesses, supported by previous studies [7, 13, 17,18,19,20]. Further analysis on this study also demonstrated leukopenia confers a 13 times risk towards dengue compared to other febrile illnesses. Leukopenia in dengue infection can be explained by the destruction or inhibition of myeloid progenitor cells, which is indicated by mild hypocellularity in the first 7 days of fever during bone marrow examination [21, 22]. We also recognized that instead of leukopenia, the control group appears to have elevated white blood cell (WBC). Our study included children with hepatitis as part of our control group. It has been reported that children with hepatitis A have a significantly lower WBC count than children without hepatitis A. However, when compared to our study, we observed that dengue patients have a lower median leukocyte (5.44 × 103/mm3 vs. 3.91 × 103/mm3) [23]. In addition, it is reported that leucocytosis is present in most pneumonia cases [24], and normal-elevated leukocyte is found in both non-bacterial and bacterial diarrhea [25].
Platelets play a role in the coagulation system and the inflammatory system for the host defence. Platelets can release multiple pro-inflammatory cytokines; thus, the increase in platelet has been reported as a normal response to infection [26]. However, we found the decrease in platelet level is more severe in the dengue group compared to the control group, supported by other studies [14, 17, 18, 20]. Furthermore, we found that thrombocytopenia is a significant variable after multivariate analysis, similar to Gregory et al. [18] Thrombocytopenia in dengue fever can be explained by various mechanisms infected bone marrow megakaryocyte, apoptosis platelet, and increased platelet destruction in reticuloendothelial, spleen, and liver. The reason for increased platelet destruction is not known precisely, but it is believed that the dengue viral itself, complement system, and endothelial cell damage are the cause. Moreover, cross-reaction of antibodies against platelet also resulted in thrombocytopenia [21, 22, 27]. This is in contrast to platelet count in the control group, such as pneumonia. Ghoneim et al. [26] found that 13.2% of patients with community-acquired pneumonia experience thrombocytosis, 80.8% of patients had normal platelet, and only 6% experience thrombocytopenia.
Our study reports that dengue patients are more likely to have normal ESR, supported by other studies [28, 29]. A study by Kalayanarooj et al. [30] compare the mean ESR value between dengue hemorrhagic patients (DHF) (10.71 mm/h), other viral infection (20.46 mm/h), bacterial infection (34.81 mm/h), and various illnesses (20.46 mm/h). It showed DHF patients had the lowest ESR mean, with more than half of patients having normal ESR. This can be explained by plasma leakage in dengue. Since blood is composed of plasma and blood cells, plasma leakage will increase blood cell percentage and decrease plasma proportion. Thus, during the Westergren method, the reading of column plasma will be reduced, and the ESR value will be within normal limits [29].
Previous multivariate studies had reported several predictive models. A study by Sawant et al. [7] has built a predictive model of dengue with three variables (myalgia, WBC count less than 5000/mm3, and aspartate aminotransferase (AST) > 40 IU) with a sensitivity of 86.7% and specificity of 83.3%. Another study by Kumar et al. [31] reported three significant variables after logistic regression: age, rash, and elevated serum alanine aminotransferase (sALT) level > 40 units. All three variables yielded a specificity of 99.2%, while the presence of one or more variables yielded a sensitivity of 89.3%. Gregory et al. [18] also proposed a predictive model to differentiate dengue for children aged 1–9. The study reported three variables: retro-orbital pain, no cough, and platelet < 240,000 cells/mm3. This model yielded an AUC of 0.7435. Our study reported five variables in differentiating dengue with an AUC value of 0.96. The presence of two or more significant variables yielded a sensitivity of 91.4% and specificity of 87.5%.
However, our study could not assess some of the symptoms that other studies reported as significant, such as retro-orbital pain [18], AST level [7, 31], and rash [31], because those variables were not recorded in the control group’s medical record; thus, we could not compare those variables and only limited clinical information can be analyzed in this study. This is our limitation as we only collect data from medical records. The control group also focuses more on gastrointestinal, urinary, and pneumonia conditions. Although the clinical diagnosis of dengue group was confirmed based on WHO 2009 guidelines, the control group was selected based on ICD10 codes registered in medical record, thus there may be classification bias in the process. Another limitation is that we could not test our model in a perspective model; therefore, the validity and reliability need to be studied again upon a different population. However, this study serves as a starting point for researchers to validate our findings. This study also included a large number of samples from the last 5 years in an endemic region, which is expected to provide a good representation of the dengue population. This model can be used in some rural areas in Indonesia with limitations in diagnosing dengue to increase suspicion of dengue, which is abundant in the monsoon season in Indonesia.
