An accurate assessment of the burden of dengue in many parts of the world is hampered by inability to make a laboratory diagnosis and difficulties in identifying less severe forms of dengue by clinical presentation alone. A clinically applicable diagnostic algorithm could improve reporting in areas where laboratory resources are limited, while simultaneously benefiting busy clinicians by identifying dengue cases so that appropriately targeted prevention messages and management can be provided. This analysis using data from our enhanced dengue surveillance system shows that a prediction model can be constructed from readily available data collected early in the disease course. A combination of four clinical features and one routine laboratory test predicted dengue positivity with moderate accuracy, and more accurately than either the current or proposed WHO dengue case definition while requiring the collection of fewer data. Additionally, our study shows that for our population the proposed WHO dengue case definition is, at best, no better than the existing WHO case definition in predicting laboratory-positive dengue.
One important feature of our study, which has received relatively little attention previously, is the variation in clinical and laboratory features of dengue by patient age. Although these findings have also been reported by other groups,12,15
the WHO case definition makes no distinction between pediatric and adult patients. Moreover, among 19 previous studies of clinical features of dengue,5,12–17,26–37
only three have examined children and adults separately.12,17,36
It is not surprising, given the differences between children and adults in prior immunologic exposure to dengue and other viral infections, that pathophysiological responses to dengue infection12,38
and clinical and laboratory features may vary significantly. For example, subgroup analysis of our data shows that leukopenia, one of the manifestations of dengue listed in the current WHO case definition, is predictive of dengue early in the course of infection only among adults aged ≥ 20 years. Leukopenia is a common clinical finding in many viral childhood infections,39
and as children have an average of 6 to 8 viral infections annually,40,41
the finding that this is not a good early predictor for dengue among children could be anticipated. Although a previous study from Thailand did identify leukopenia as a good early predictor of dengue infection in children,14
another study from Nicaragua found, similar to our results, that leukopenia was significantly associated with early dengue infection in adults but not in children.12
Another key finding of our study was the importance of retro-orbital pain as a discriminator of dengue among all age groups in our population. This is especially important given that the proposed new WHO dengue case definition does not specifically mention this symptom in its list of clinical features.8
Omitting retro-orbital pain from the dengue case definition or grouping retro-orbital pain in with other possibly less specific pain syndromes in the definition may reduce clinical diagnostic accuracy. In our study when we performed an alternative analysis including headache and retro-orbital pain along with myalgia and arthralgia in the new WHO criteria of “aches and pains,” the model performed non-significantly worse than when we included only myalgia and arthralgia (data not shown). This is not surprising given that univariate analysis showed that body aches, joint pain, and retro-orbital pain were associated with dengue positivity, but headache was not (). Previous studies that have combined retro-orbital pain with headache showed no association between retro-orbital pain/headache and dengue positivity,11
whereas studies that have examined retro-orbital pain separately have seen this association;29,33,35
although sometimes only in certain age groups.17
Retro-orbital pain has been traditionally considered a relatively nonspecific manifestation of DF, but to our knowledge, no other common acute febrile illnesses have retro-orbital pain or eye pain listed in their case definitions. Retro-orbital pain has also been criticized for lacking sensitivity in children, given the presumed limited ability of young children to express location of pain. However, of 1,211 children aged 1–14 years with laboratory-confirmed dengue in Nicaragua, 59% were documented to have retro-orbital pain.12
Recent articles have highlighted a variety of ocular manifestations associated with dengue infection, including macular and retinal hemorrhages,42,43
optic neuritis and neuropathy,46
retinal artery occlusion,49
and retinal and macular edema.43,47
In a recent prospective study from Singapore, Seet and colleagues51
observed “clinically significant” ocular symptoms, defined as symptoms severe enough to warrant referral to an ophthalmologist, in 18% of hospitalized patients with serologically confirmed dengue. A 2006 study from India examined 134 consecutive hospitalized dengue patients and discovered ocular findings on funduscopic examination in over 40%.42
Similarly, maculopathy was documented in 10% of consecutively admitted dengue patients in Singapore.44
Based on the timing of the development of ocular signs relative to the course of dengue illness, and the association between low complement C3 levels and “dengue-associated maculopathy,”44
an immune-mediated process has been proposed for the ocular manifestations of dengue.44,45
These studies suggest that eye involvement with dengue might be more common than is appreciated and it is plausible that retro-orbital pain could be a more specific indicator of dengue infection than previously realized.
A recent systematic review identified 15 studies that have examined the differences in clinical and laboratory features between dengue and other febrile illnesses,11
and since that publication at least three additional studies have been published.16,27,34
Our study differs from most previous studies in that data were recorded at the time of the initial clinic or emergency department visit rather than at the time of hospitalization. Hence, we were able to identify features that were predictive of dengue positivity early in the disease course, rather than mid-to-late in the course. When we reanalyzed our data excluding patients who presented more than 3 days after symptom onset (7.4% of our total patients), the only significant change in the findings was that rash was no longer associated with dengue-positivity in patients ≥ 20 years of age (data not shown). This is not surprising as the association between rash and dengue in this age group may have been driven by patients who presented for medical care at the time of the appearance of the relatively classical convalescent rash of dengue. Several of the age-specific predictive features for dengue identified in our study have been previously noted, including the presence of rash and leukopenia in adults,12,26,33,52
reduced platelet counts in children,12,14
and absence of upper respiratory symptoms in children.5,17
Apart from the collection of data early in the disease course, our study has several other strengths. Unlike some previous studies, it includes only cases that could be laboratory confirmed as dengue infection or not dengue infection, thereby reducing the potential for misclassification bias. Data were collected over more than one dengue season and in a period when all four dengue virus serotypes were circulating; thereby giving a more accurate overall clinical picture than studies that took place in an outbreak setting. Finally, our study had a relatively large sample size compared with other similar studies11
and we stratified by age group to reduce potential confounding by age.
Nonetheless, the study has several limitations. We used enhanced surveillance data which, while it is extremely accurate and complete overall, was missing information for certain variables such as petechiae and the results of the tourniquet test. The presence of petechiae was seen almost exclusively among patients with dengue, but because of the small numbers of patients for whom this variable was recorded, it could not be incorporated into our multivariate model. The tourniquet test is not routinely performed in Puerto Rico. Because both current and proposed WHO case definitions for probable DF include this as one possible example of the hemorrhagic manifestations criteria, absence of information on this feature may have led to an underestimation of the sensitivity of both case definitions. A prospective study is underway in Puerto Rico to evaluate the effect of inclusion of the tourniquet test on the diagnostic accuracy of the current and proposed WHO case definitions and whether including tourniquet test results would improve the performance of our predictive model. Because of the retrospective nature of our study, we were unable to assess the variation in clinical and laboratory features predictive of dengue by day of illness. Patients were seen only once or twice in the course of their illnesses so a final determination of case severity was not possible in many cases. Therefore, this study is unable to identify early predictors that are specific to severe dengue, a pressing research need. Finally, the data are from only one region of Puerto Rico and may not be representative of other areas with different dengue transmission patterns, population demographics, underlying causes for acute febrile illness (e.g., typhoid, malaria), or where patients present for medical care later in the course of illness.
This study suggests that simple clinical and laboratory data can be used to identify early dengue infections in both adults and children. Further efforts should be made to validate these findings in other geographic settings and time periods. If the models continue to perform well over time in these settings, they could be used to develop clinical diagnostic algorithms. Future studies should also seek to detect early clinical and laboratory markers that can predict the development of severe dengue. Findings from our study and others11,12,15,17
suggest that separate clinical case definitions or diagnostic algorithms may be needed for children and adults. Additionally, changes to the DF case definition that deemphasize retro-orbital pain should be carefully evaluated, as this symptom was one of the strongest predictors of dengue infection in our study.