Development of Dengue Infection Severity Score

Objectives. To develop a simple scoring system to predict dengue infection severity based on patient characteristics and routine clinical profiles. Methods. Retrospective data of children with dengue infection from 3 general hospitals in Thailand were reviewed. Dengue infection was categorized into 3 severity levels: dengue infection (DF), dengue hemorrhagic fever (DHF), and dengue shock syndrome (DSS). Coefficients of significant predictors of disease severity under ordinal regression analysis were transformed into item scores. Total scores were used to classify patients into 3 severity levels. Results. Significant clinical predictors of dengue infection severity were age >6 years, hepatomegaly, hematocrit ≥40%, systolic pressure <90 mmHg, white cell count >5000 /μL, and platelet ≤50000 /μL. The derived total scores, which ranged from 0 to 18, classified patients into 3 severity levels: DF (scores <2.5, n = 451, 58.1%), DHF (scores 2.5–11.5, n = 276, 35.5%), and DSS (scores >11.5, n = 50, 6.4%). The derived score correctly classified patients into their original severity levels in 60.7%. An under-estimation of 25.7% and an over-estimation of 13.5% were clinically acceptable. Conclusions. The derived dengue infection severity score classified patients into DF, DHF, or DSS, correctly into their original severity levels. Validation of the score should be reconfirmed before application of routine practice.


Introduction
Dengue infection has become an international public health burden. Half of the world population are presently at risk of dengue infection. Approximately 50-100 million infected cases were reported annually. Among those infected, 500000 patients had severe infection and required hospital admission; most were children. Approximately 2.5% died from the infection [1]. The cost of care was as high as $US 2.1 billion per year in The United States of America [2]. No specific treatments are available except for symptomatic [3], which are effective in early detection [4]. In patients with severe infection, shock and hemorrhage usually follow [4,5]. If not treated, death may be a consequence. Early detection or correct prognostication may avoid such severe complications [4,6].
The scoring system, such as the Pediatric Logistic Organ Dysfunction (PELOD) Score and the Disseminated Intravascular Coagulation (DIC) Score, was used to forecast mortality in DSS [19]. A decision tree algorithm was used to differentiate dengue fever from other types of fever and also to forecast severity in dengue infection [20]. Some studies applied DIC scoring system to diagnose DIC more precisely, using clinical signs and symptoms and routine laboratory investigations to differentiate DHF from DF [21]. Most of the prediction systems in the past focused on clinical outcomes of the disease. There are few studies that focused directly on dengue infection severity.
The aim of the present study was to develop a simple clinical risk scoring system to predict dengue infection severity, based on patient clinical characteristics and routine laboratory investigations, obtained from the previous investigation [22].

Patients.
The medical records of children aged 1-15 years with dengue infection between 2007 to 2010 in the three university-affiliated general hospitals located in the northern region of Thailand, Sawanpracharak Hospital in Nakorn  Sawan, Uttaradit Hospital in Uttaradit, and Kamphaeng Phet Hospital in Kamphaeng Phet, were reviewed. The data were retrieved from the hospital database, under the following ICD-10: A90-dengue fever, A91-dengue hemorrhagic fever, and A910-dengue hemorrhagic fever with shock.

Data Analysis.
Potential predictors for dengue severity from a previous report [22] were tested with nonparametric test for trend. The predictive ability was analyzed by multivariable ordinal logistic regression and presented with coefficients and odds ratios. Missing data were replaced by the mean values of the parameters. Assigned item scores were derived by transformation of coefficient of parameters. The total (sum) scores were used to classify patients into 3 severity levels. The distribution of scores across the three severity groups were presented with box plots. The discriminative and predictive abilities of the scores were presented with probability curves.

Ethical
Approval. The present investigation was approved by the Research Ethical Committee of the Faculty of Medicine, Chiang Mai University, and the research ethical committees of the three hospitals.

The Scoring
System. Transformation of significant parameter coefficients into item scores was done by division of each coefficient with the smallest coefficient of the model (0.30) and rounded up or down to the nearest 0.5 integers. The item scores ranged from 0 to 8.5, and the total score ranged from 0 to 18 (Table 2).

Discriminations.
The mean total severity scores in patients with DF, DHF, and DSS were 3.6 ± 2.1, 5.1 ± 3.2, and 11.0 ± 4.1 ( Table 3). The distribution of the derived severity scores was different among the three severity groups (Figure 1). The derived scores also discriminated DHF from DF and also discriminated DSS from DHF ( Figure 2).

Clinical Predictions.
Our scoring system discriminated DSS and DHF from DF with an area under the receiver operation curve (AuROC) of 0.7416 (95% CI = 0.7294-0.7537, figure not shown) and discriminated DSS from DHF and DF with a higher AuROC of 0.8877 (95% CI = 0.8788-0.8964, figure not shown).
The scores between 2.5 and 11.5 predicted DHF correctly in 136 out of 296 patients, with an underestimation in 46 patients (5.9%) and an overestimation in 94 patients (12.1%).

Discussions
Dengue infection is an urgent condition requiring prompt diagnosis and treatment before patients enter into bleeding or shock states. Previous scoring systems trying to evaluate or forecast disease severity included the Dengue Fever Scoring System based on epidemiological information and clinical signs or symptoms, which might be useful in detecting DF very early prior to laboratory results [25]. Other scoring systems were the Pediatric Logistic Organ Dysfunction (PELOD) Score and the Pediatric Risk of Mortality III (PRISM III), used to evaluate the mortality rates [26], and the Disseminate Intravascular Coagulation (DIC) Score to diagnose DIC and to discriminate DF and DHF from other febrile illnesses [27]. Decision tree algorithms were also applied to classify dengue infections into DF, DHF grade I, DHF grade II, and DHF grade III [28]. They may also predict patients with low risk who may be discharged safely and select patients with high risk who should be admitted for close monitoring [29].
These studies used clinical symptoms and/or laboratory tests to evaluate or forecast the risks. Some studies included only adult patients [25,29] which may or may not be relevant for children. In studies which included children, some studies used information on the first few days of admission in pediatric intensive care unit (PICU) [19]. In some studies, included DHF patients did not experience shock and may not be inferred to patients with DSS [21]. Risk prediction in many studies was based on nonroutine laboratory investigations [19,20,26,28,29]. Despite their high predictive powers, those nonroutine predictors were not available in many primary care centers. An investment on such facilities may not be feasible and may not be cost effective.
Our study developed a scoring system based on clinical risk and routine laboratory parameters and categorized patients with dengue infection into DF, DHF, or DSS.
(1) Patients scoring <2.5 were classified as DF which is the mildest form. These patients normally do not require hospital admission even in the feverish stage. Patients may safely be handled as outpatients with symptomatic treatments and may be advised to observe any abnormal signs or symptoms with a follow-up appointment. (2) Patients scoring between 2.5 and 11.5 were classified as a risk group for DHF. These patients should be admitted to be closely monitored for early signs of plasma leakage, hemoconcentration, coagulopathy, thrombocytopenia. (3) Patients scoring >11.5 were classified as a risk group for DSS. These patients should also be admitted to monitor any early signs of shock.
Our scoring system predicted DSS correctly with a positive predictive value (PPV) of 88%, similar to other studies which reported a PPV of 82-95% [20,26]. The score may be used to discriminate DSS from DF and DHF.
However, the fact that the derived scores in DF and DHF were more or less overlapping made it less likely to be used to differentiate DHF for DF. Other nonroutine predictors specifically for DHF may be required.
In case of continuous routine laboratory investigations, our score may also be used to monitor the patients risk for DSS as the disease progresses from day to day. In case of outpatients, this will help clinicians make decision when to admit the patients to hospital. Application of this scoring system into routine patient care may help reduce unnecessary admission and also reduce case facility in severe cases that are admitted based on high risk scores.

Conclusions
The derived dengue infection severity score classified patients into DF, DHF, or DSS, correctly into their original severity levels. Validation of the score should be reconfirmed before application into routine practice.

Conflict of Interests
This investigation was partially supported by the Research Grant for Basic and Applied Sciences, the Faculty of Medicine, Chiang Mai University.