4.1.2 Distinguishing Dengue from other Febrile Illnesses in an early Stage of Disease…
4.1.2.3 Lower IL-2 Levels, higher IP-10 Levels as well as increased IFN-γ
Combination of basic clinical data and cytokine data resulted in two different decision trees that showed higher accuracies than trees based only on either clinical or cytokine data. The first constructed tree including IFN-α (DENPRE_INCYTA_291; Figure 3.5;
Page 101) did not change its first two splitting criteria compared to the tree which was based only on cytokine data. IFN-α was followed by IL-2 but, in turn, the white blood cell count was used as a third splitting criteria for patients having lower IFN-α along with higher IL-2 levels. This suggests that the white blood cell count is probably a better predictor than IL-10 which is underlined by the better overall performance compared to the tree based only on cytokine data and by the fact that the tree based only on clinical data was the better predictor by itself. Furthermore, the odds ratio of white blood cell count is four times higher than the one of IL-10. The decision node that further separates patients with a lower white blood cell count was represented by IL-1 which is also reported to be secreted during dengue infection (Hober et al., 1993).
Generally, IL-1 is mainly secreted by macrophages and dendritic cells and has an important function in the activation of a specific immune response during inflammation. Its effector functions are similar to TNF-α which might be the reason for its position in the tree. It is intriguing that higher levels of IL-1 are a significant characteristic of dengue positive patients whereas high TNF-α levels are more common in other febrile illnesses suggesting that IL-1 is more specific for classical dengue fever and TNF-α might be more important in DHF/DSS as reported in other studies (Kittigul et al., 2000; Nguyen et al., 2004).
Interestingly, our best calculated classification model excluded IFN-α and combined other cytokines as well as clinical data (DENPRE_INCYT_291; Figure 3.7; Page 105).
It achieved an overall performance of 94% with a sensitivity of 90%. In the calculated tree, a lower white blood cell count and lower absolute numbers of lymphocytes represented the clinical risks for dengue infection. On the other hand, the immunological factors were represented by IL-2, IP-10, TNF-α as well as IFN-γ. IFN-γ is a pro-inflammatory TH1 specific cytokine and is mainly produced by macrophages.
Our finding that elevated IFN-γ levels could be an indicator for dengue virus infection is consistent with earlier reports demonstrating the induction of IFN-γ secretion upon dengue infection (Kurane et al., 1991; Kurane et al., 1986). Due to the early time point of sample collection, we expected to observe elevated levels of TH1 cytokines which was shown by Pacsa and collaborators (Pacsa et al., 2000). Hence, it is striking that the observed cytokines included in our trees do not fully support this hypothesis but rather point towards an overlapping of TH1 and TH2 responses.
The decision tree explicitly shows the complex picture of cytokines interacting with lymphocytes and is highly predictive in distinguishing dengue patients from patients with other febrile illnesses. We can conclude that the cytokines and clinical features represented in the combined classification model are of significant importance during an early stage of dengue infection. Nevertheless, it is prudent that our classification tree has to be closely evaluated in several aspects. Firstly, more detailed statistical investigation is needed to identify correlations between the chosen parameters at the decision nodes because by only looking at the decision tree, we are unable to detect direct interactions. Secondly, cytokine production has to be investigated on a cellular level and in vivo to understand the sources, kinetics and functional impacts of
cytokines specifically produced upon dengue infection. Finally, the model has to be validated in other epidemic settings, especially in regions with a high prevalence of other flaviviruses to determine dengue specific cytokines. Moreover, the combined model is useful in providing a hint of which cytokines might be of greater interest to investigate but it would not be suitable as a diagnostic tool in epidemic settings. The major drawback lies in the procurement of cytokine data which entails more labour and has higher cost than the commonly used hematological methods.
In summary to the first aspect, we integrated immunological and clinical factors into a predictive model for dengue infection, we propose that the combined decision tree may be a biological tool for the better understanding of the interplay between humoral and cellular factors during an early stage of disease and highlight that the tree based only on clinical data may be a useful diagnostic tool in epidemic settings.