Dengue is one of the most underreported tropical diseases
. On the other hand, during epidemics, overreporting can occur at some health sites. The lack of laboratory resources and the nonspecific clinical presentation of non-severe cases greatly contribute to this situation
. Diagnostic algorithms based on clinical data may enhance disease diagnosis and surveillance in endemic areas.
In this study, we identified some clinical and hematological features associated to confirmed dengue cases. We also constructed simple predictive models for dengue diagnosis based on clinical and hematological data for patients presenting early in the course of disease. These models showed moderate accuracy
, and performed better than the diagnostic features proposed by WHO for probable dengue
 in the study population.
Among clinical features, rash, taste disorder, conjunctival redness, and lymph node enlargement were all associated with dengue diagnosis. Rash is among the classic signs of dengue fever and its description includes different cutaneous manifestations, such as a diffuse erythema coincident with fever or a macular exanthema that is more frequent after the third day of disease
[1, 6]. Rash has also been reported to be more common in primary than in secondary infections
, an issue that we could not explore in our study.
The extraordinarily high prevalence of rash detected on the physical exam of dengue patients (72%) in our study can be related to major efforts to ascertain any kind of cutaneous manifestations in a research setting. These may have included a mild flushing that would otherwise be unnoticed. It should be noted that a history of rash was more accurate for dengue diagnosis than its detection on physical exam. The same finding was also reported by Chadwick et al.
 who found that history of rash was the only independent clinical predictor that remained in a regression model which included clinical and laboratory data. In this context, the possibility that self-examination could be more accurate than medical examination at ascertaining some mild cutaneous manifestation should be considered.
Comparing our results with other reports, we found that taste alteration, conjunctival redness, and lymphadenopathy, although described as common manifestations of dengue fever
, are not consistently investigated as predictive signs and symptoms in studies on dengue diagnosis
. Nevertheless, taste alteration was reported as a dengue predictor by the few researchers who investigated this symptom
[15, 39], and conjunctival redness was reported by Low et al.
 as one of the most accurate signs (p < 0.0005; OR = 4.49) for dengue diagnosis among adult outpatients with a fever lasting less than 72 hours. The frequency of lymph node enlargement in dengue patients has varied from 3%
 to about 20%
[17, 23] in studies that reported this sign. In these studies, it was not a useful sign to discriminate D from ND patients, despite being more frequent among the first ones in all of them.
The importance of ocular findings warrants better investigation as recent studies have described a variety of ocular manifestations in dengue
[40–42], suggesting that eye involvement may be more common than usually appreciated. Gregory et al. (2010) also highlighted the importance of ocular manifestations in dengue as they found retro-orbital pain as an important clinical feature to discriminate dengue from OFI in all age groups
Among hematological data, leukocyte count was the most discriminant feature. Although common in other viral illness, leukopenia has been consistently reported as an independent predictor of dengue diagnosis among febrile patients
[14, 16, 19], particularly in adults
[19, 20, 43–46], and seems to occur earlier than thrombocytopenia
[43, 44]. Indeed, it was the most important isolated predictor in our study.
Other authors have already proposed models and scales to predict dengue infection in febrile adults living in the Americas. Comparison between predictive rules and WHO criteria through ROC curve analysis was also done by some of them. In Colombia, Díaz et al. (2006) developed a scale which performed better as an early predictor in adults than the one produced with WHO criteria (AUC ROC of 81.0% versus 70.0%). It was composed by the presence of rash, positive tourniquet test, absence of nasal discharge, arthralgias, absence of diarrhea (1 point for each finding), leukocyte count <4,000/mm3 (3 points) and platelet count <180.000/mm3 (2 points)
. Similar accuracy was obtained in Puerto Rico with a predictive model for adults with suspected dengue (AUC ROC: 79.7%). Retro-orbital pain, rash, absence of sore throat, and leukopenia were the independent predictors in this age group
As strengths of this study, we may cite the collection of clinical data by a trained team, before the performance of laboratory tests, and the use of a comprehensive standardized protocol for history taking and examination. The recording of clinical features prior to any test result minimized the risk of observation bias, and the inclusion of clinical features other than the usual dengue signs and symptoms allowed us to identify some potentially useful new predictors.
This study also has some limitations. The first one is related to the small sample size used in multiple regression analysis. This may be attributed in part to the characteristics of our clinic, which attends mainly referred patients, to a long period without epidemics when we had slow subject enrollment and to our option to analyze only data from patients in the first three days of disease. The small sample size precluded any subgroup analyses.
We also had considerable losses because of indeterminate diagnosis (40/182), as only 66 out of 182 (36.3%) eligible patients collected two blood samples. Although the use of NS1 test helped to diagnose dengue in patients with samples collected from day 0 to day 2, those with only one negative sample collected between day 3 and day 7 were classified as indeterminate, as a negative result in both IgM and NS1 could be a false-negative by this time. As a consequence, we had relatively few confirmed ND patients. This kind of problem, however, affects most studies with outpatients, in such a way that losses greater than 30% for indeterminate diagnoses are common in this setting
We cannot exclude the possibility of some misclassification by our reference standard in dengue diagnosis. Although sensitivities above 90% for NS1 on days 0 to 2
[26, 48] and sensitivities above 93% for IgM in samples collected after the seventh day of disease have been reported in patients with a primary infection
[1, 28, 49], both tests have lower sensitivities in secondary infections
[28, 50]. As dengue is endemic in Rio de Janeiro, it is possible that some dengue patients have been classified as non-dengue. The consequence of this eventual misclassification would be a decrease in the measured odds ratios with underestimation of the accuracy of analyzed predictors.
The interpretation and generalization of the results of studies such as this one must consider the fact that diagnostic accuracy of clinical manifestations also depends on their frequency in the non-dengue group. This means that it varies according to the incidence of other febrile illnesses (OFI) in the same period and place. For instance, low platelet counts may be useless to discriminate dengue from OFI in a place where malaria is endemic, as low platelet counts are also frequent in the later
. In our study, laboratory tests for OFI were required as indicated by clinical suspicion and we are unable to describe the prevalence of other diagnoses, except for malaria, which was rare (2%) in the study population.
Although our model outperformed the diagnostic accuracy of the WHO criteria, its predictive value is still poor, with some 20% false negative among dengue patients and 30% false positive among those with OFI. The low accuracy of WHO case definition, mainly because of its low specificity has been described by Martinez et al. (2005), who also explored the accuracy of different number of WHO criteria
. The difficulty to identify early clinical predictors of dengue infection in adults has been described by Ramos et al. (2009) in a large study in Puerto Rico. Although they identified eye pain, diarrhea and absence of upper respiratory symptoms as independently associated to confirmed dengue cases, they also highlighted the low predictive value of these features, alone or in combination, for early infection in adults
. In spite of this relatively low accuracy, the use of management protocols based on clinical diagnostic scales proved useful to reduce hospitalizations due to dengue in Colombia
Nonetheless, the results of this study are potentially helpful for surveillance in adults, as they suggest that the proposed criteria, derived from simple clinical and hematological data, can be more accurate than the criteria for reporting suspected dengue cases currently in use. The use of this alternative algorithm could enhance the ascertainment of dengue cases by clinical and epidemiological criteria, and enable a more accurate estimation of the disease burden. Meanwhile, it should be stressed that in endemic areas, while early accurate laboratory tests are not widely available, dengue fever should be considered in every patient presenting with an acute undifferentiated febrile illness. Monitoring all these patients for the development of signs of severity, however, may impose a great burden on the health services. Once validated, algorithms that enable early identification of dengue cases could influence clinical outcomes as they would allow closely monitoring of selected patients. This procedure may warrant timely identification of alarm signs and the adoption of simple and widely available therapeutic support measures that are effective in preventing fatalities