The findings of our review suggest that several clinical and laboratory measures could potentially distinguish patients with dengue from those with OFI. Low platelet count and decreases in WBC and neutrophils were independently associated with the presence of dengue, when compared to patients with OFI, in both adults and children. These variables, as well as signs of rash and liver damage, were also used in multivariable models to distinguish patients with dengue from those with OFI. However, it is unlikely that any single indicator will be useful in clinical practice because these signs and symptoms are present in other diseases, such as viral hepatitis and leptospirosis, which are also endemic in areas with a high prevalence of dengue.
Alterations in the microvascular endothelium in patients with dengue are thought to lead to a higher likelihood of hemorrhage (Bandyopadhyay et al., 2006
; Cardier et al., 2005
). In this review, an increased frequency of hemorrhage was observed in adults with dengue but was not associated with dengue in studies that only included children; however, Hammond et al demonstrated that some types of hemorrhage (e.g., hematemesis and melena) were associated with dengue in children, suggesting that the types of hemorrhagic manifestations seen in dengue may depend on the age of the patient.
Signs of rash and indicators of liver damage, in combination with other variables such as age, myalgia, WBC count, and platelet counts, may help to establish a diagnostic algorithm to distinguish dengue from OFI patients. Several studies used multivariable regression models to discriminate dengue from OFI; however, most published models had lingering statistical questions. Wilder-Smith et al (2004)
presented a model with very large odds ratios; however, the confidence intervals for their model were also large and questions of over-fitting and co-linearity were not discussed. Deparis et al (1998)
presented a model with an unusually small odds ratio for a categorical variable (low platelet count), which may not be applicable in a clinical setting. None of the regression models was validated using a training and testing dataset approach. Furthermore, the generalizability of these models is questionable since most were derived from single outbreak studies. For example, Karande et al (2005)
was a single outbreak study and presented a model with 100% PPV, but they only had 13 patients with dengue in the model. Since the authors selected the variables for analysis, our review is unable to determine whether a specific value versus an increase (or decrease) in a particular variable is most useful.
Any algorithm to identify patients with dengue would need to be applied early in the illness in order to be useful in reducing unneeded hospitalizations. This review highlights a weakness in the literature as few studies indicated which day of illness clinical and laboratory measures were assessed. Only Kalayanarooj et al (1997)
and Deparis et al (1998)
separately analyzed clinical and laboratory measures according to day of illness. Kalayanarooj et al (1997)
showed that positive and negative predictive values for individual variables differed depending on the stage of illness. Deparis et al (1998)
showed that the frequency of clinical and laboratory symptoms varied according to day of illness.
Five of the included studies were case-control studies that relied on review of medical records or patient recall, which could bias the findings of these studies. Furthermore, two of the case-control studies did not use a standardized data collection form. Six studies relied on serologic testing of a single blood sample, which could increase the risk of misclassification of patients with dengue. Only two studies serologically confirmed all diagnoses in the OFI group, and differences found between patients with dengue and patients with OFI depended on the specific comparison febrile illness. Bruce et al (2005)
used a leptospirosis comparison group and was the only study that found no differences in platelet count or AST/ALT. Illnesses with similar characteristics, such as dengue and leptospirosis, will clearly be more difficult to discriminate on the basis of any clinical algorithm.
Duration of illness prior to study enrollment did not distinguish patients with dengue from those with OFI in four out of five studies. Duration of illness prior to presentation may be more applicable in distinguishing patients with DHF from patients with DF. On average, patients with DHF have a more severe illness and may require hospitalization for a more extended period of time after defervescence in comparison to DF. No study in this review prospectively compared clinical signs and symptoms in patients with DHF to patients with DF or OFI. We are, therefore, unable to make any conclusions from this review of which signs and symptoms, if any, can prospectively distinguish patients with DHF from patients with DF or OFI. It is perhaps surprising that 4/9 (44%) of studies that measured hematocrit found no significant differences between dengue and OFI. However, hemoconcentration is a feature of DHF and not of DF, and is defined based on comparison to a patient’s baseline hematocrit rather than a single measurement.
This review has several limitations. There were no inter-rater or intra-rater reliability of quality assessment ratings and the STROBE is mainly a score of reporting and may impact the ability to extract information rather than quality of the study itself. There is a lack of established quality assessment rating scales for evaluating observational studies. The STROBE gives merit to a study for addressing its limitations, which may explain why the retrospective studies had the highest quality assessment ratings. Not all studies had robust statistical methods. Both the dengue and OFI groups were heterogeneous, the former including both milder and more severe disease and the latter including a wide variety of possible etiologies. Some studies failed to include duration of fever or illness in their analysis, which could affect the interpretation of time-dependent variables. Also, most dengue outbreaks occur in countries where English is not the primary language. Thus, the restriction to English-language studies may have affected the findings of this review. Finally, many studies did not note what day of illness the clinical and laboratory data were measured, which makes it impossible to determine whether these data can distinguish dengue from OFI early in the course of illness.
Until low-cost, rapid, and sensitive laboratory assays are widely available for diagnosis of dengue, diagnostic algorithms will continue to have an important place in clinical management. Additional prospective studies are needed to establish an algorithm that can be validated and generalized to distinguish dengue from OFI and DF from DHF in the early stages of illness. Furthermore, longitudinal studies that routinely document clinical and laboratory signs and symptoms throughout each patients’ course of illness would provide much needed data to develop a predictive model that can distinguish patients with dengue who will require hospitalization from patients with OFI. An easily applicable clinical algorithm could have a favorable impact on the economies of dengue-endemic developing countries.