4.1.3 Severity Prediction in an early Stage of Disease
4.1.3.4 IL-10 as an early Predictor of severe Dengue Infection shows higher
In a second round, we calculated a decision tree only based on cytokine data. The goal was to find early immunological predictors of thrombocytopenia in a later stage of disease. This approach resulted in two decision trees which were always using IP-10 as the first splitting criteria but exploited either IL-8 (SEVERE_CYTOA_IL8; Figure 3.28; Page 150) or IL-10 (SEVERE_CYTOA_IL10; Figure 3.29; Page 150) at the second decision node. We observed that the two models still showed a good overall performance but were slightly worse when compared to the clinical trees. Furthermore, the confidence intervals for the AUCs of the ROC curves were comparably large which suggests some instability of the classifiers possibly caused by the smaller dataset. However, the choice of taking IP-10, IL-8 and IL-10 with the defined
threshold into our decision tree modeling is backed by the calculated odds ratios and their confidence intervals. Univariate analysis also showed significant differences of these three variables between the severe and the mild group. It is important to note that analysis of second visit data still detected significant differences of IP-10, IL-8 and IL- 10 whereas the discrepancies were reduced between the two groups on the third visit.
These findings point towards an important general role of these three cytokines during dengue disease, especially in influencing the development of thrombocytopenia.
We identified IL-8 in combination with IP-10 as a better early predictor for severity which showed an overall accuracy of 75% with a sensitivity of 66%. In addition, it is worthwhile mentioning that IL-8 levels detected on the second visit correlated with viral load on the first visit (R=-0.368; 95%CI: -0.542, -0.162; p value = 0.001) which highlights the importance of an early high viral load in disease outcome. Interestingly, we did not detect any significant differences in IL-8 levels between dengue and other febrile illnesses on the 1st visit. This directly supports the finding of another study showing that IL-8 levels of classical dengue fever patients are not elevated within the first 4 days whereas patients suffering from more severe disease already showed elevated IL-8 levels between day one and day four (Raghupathy et al., 1998).
Furthermore, a study investigating IL-8 secretion in PBMCs detected delayed peaking of IL-8 levels between days three and four of infection which they explained that a certain threshold of viremia has to be obtained in order to induce chemokine transcription through NS5 (Medin et al., 2005).
IL-8 is secreted by macrophages as well as epithelial cells during an inflammation and is a chemokine that mainly attracts neutrophiles to the site of infection (Baggiolini et
al., 1997). Secretion of IL-8 is linked to pleural effusion (Baggiolini et al., 1995; Pace et al., 1999) and IL-8 levels are elevated in more severe dengue patients (Juffrie et al., 2000; Raghupathy et al., 1998). Induction of IL-8 secretion is controlled by the cytokine network which includes IL-1 and TNF-α. Because of elevated IL-1 levels in dengue patients with regard to our control group but no significant changes in levels TNF-α, we think that IL-1 is the main inducer of IL-8 during dengue infection and that there might be a direct induction by NS5 as proposed by Medin and collaborators (Medin et al., 2005). Expression of IL-8 may be used to counteract the anti-viral effects of innate immunity, allowing further spreading of dengue virus to neighboring cells (Medin et al., 2005). Such a mechanism was also observed to enhance viral replication in other viruses (Lane et al., 2001; Murayama et al., 1994).
Substitution of IL-8 with IL-10 in the final decision tree resulted in a higher sensitivity of 71% but lower overall performance of 73%. This reflects a slightly narrower predictive window but a higher predictive power in detecting severe cases in an early stage of disease. We have to consider the fact that our chosen model is skewed towards the major class which is in our case the mild group. In such class distributions, a better overall performance is achieved by having a higher specificity or lower misclassification rate of mild cases. On the other hand, Increasing the sensitivity and thus decreasing the specificity leads in a higher overall error rate caused by the higher misclassification rate of the major class which decreases the overall performance of the classifier. Thus, we can assume that IL-8 is a better general predictor and that IL-10 is more specific for more severe infections.
There are several reports describing elevation of IL-10 levels during dengue infection (Libraty et al., 2002; Nguyen et al., 2004) and association of IL-10 with thrombocytopenia as well as plasma leakage (Green et al., 1999b) which clearly maintained our calculated model. One explanation for its important role in an early prediction of severity could be its inhibitory effect on IL-2 and therefore on T-cell proliferation. We were able to observe this in the first part where we attempted to find a clinical as well as immunological model that was able to distinguish dengue from other febrile illnesses. This impaired T-cell proliferation may lead to a higher viral load and finally, as observed in our clinical severity model, combined with secondary infections to a more severe disease outcome. Briefly, more IL-10 in an early stage of infection and lower IL-2 may lead to a higher viremia which subsequently affects disease outcome.
In summary, we showed that increased levels of IP-10 in combination with higher serum concentrations of either IL-10 or IL-8 are a good predictive indicator for the development of more severe disease. Due to the fact that adults are at lower risk to develop DHF/DSS it is possible that the observed picture of these three cytokines may be important mostly in disease severity of adult patients which do not show DHF/DSS.
This is underlined by the fact that IL-8, IL-10 and IP-10 were still elevated on the second visit and that no differences in TNF-α were observed. Such a picture leads to the assumption that TNF-α is more characteristic for DHF/DSS which, in turn, is very rare in adults and therefore, other immunological factors such as IL-8, IL-10 and IP-10 may have a more important role in disease outcome of adult infections. Another explanation might be the fact that we predicted thrombocytopenia and therefore, only immunological factors specific for a low platelet count were identified.