Assessment of the new World Health Organization's dengue classification for predicting severity of illness and level of healthcare required

The objective of this study was to assess the validity of the new dengue classification proposed by the World Health Organization (WHO) in 2009 and to develop pragmatic guidelines for case triage and management. This retrospective study involved 357 laboratory-confirmed cases of dengue infection diagnosed at King Abdulaziz University Hospital, Jeddah, Saudi Arabia over a 4-year period from 2014 to 2017. The sensitivity of the new classification for identifying severe cases was limited (65%) but higher than the old one (30%). It had a higher sensitivity for identifying patients who needed advanced healthcare compared to the old one (72% versus 32%, respectively). We propose adding decompensation of chronic diseases and thrombocytopenia-related bleeding to the category of severe dengue in the new classification. This modification improves sensitivity from 72% to 98% for identifying patients who need advanced healthcare without altering specificity (97%). It also improves sensitivity in predicting severe outcomes from 32% to 88%. In conclusion, the new classification had a low sensitivity for identifying patients needing advanced care and for predicting morbidity and mortality. We propose to include decompensation of chronic diseases and thrombocytopenia-related bleeding to the category of severe dengue in the new classification to improve the sensitivity of predicting cases requiring advanced care.


Introduction
Dengue fever (DF) is the most prevalent arthropod-borne viral disease in humans and one of the major re-emerging communicable diseases. The World Health Organization (WHO) estimates that around 50 million dengue infections occur annually and approximately 2.5 billion of the world's population live in dengue-endemic areas [1].
Dengue virus is endemic in Saudi Arabia primarily in the western and southern provinces [2,3]. It is a well-recognized cause of seasonal outbreaks in Jeddah and Makkah [4][5][6][7]. Two large epidemics occurred in Jeddah and Makkah; the first in 2011, when 2569 cases were reported, and the second, in 2013 when 4411 cases including 8 deaths were reported [6]. Dengue was also reported in other regions of Saudi Arabia, including Al-Madinah (2009), and Aseer and Jizan (2013) [2,7].
Since the 1970s, dengue has been conventionally classified into four main categories (Table 1): non-classical DF, classical DF, dengue hemorrhagic fever (DHF), and dengue shock syndrome (DSS). DHF definition requires the presence of four criteria: fever, thrombocytopenia (<100,000 platelets/mm 3 ), hemorrhagic manifestations, and plasma leakage manifesting as accumulation of fluids in the peritoneal, pleural, or pericardial spaces, lower limb edema, hypoalbuminemia, or hemoconcentration. Several studies reported lack of correlation between the categories of the conventional classification and the disease severity [8][9][10]. Despite high specificity of the DHF category, the sensitivity is unacceptably low in detecting severe cases of dengue that require specialized care and monitoring in a hospital setting [11][12][13]. As a consequence, a global expert consensus meeting at WHO in 2008 accorded on a new classification of DF. The revised classification divides dengue into two categories ( Table 2): non-severe and severe dengue (SDF); the non-severe dengue is further divided into two categories: dengue with warning signs (D+W) and dengue without warning signs (D-W). The new classification was developed based on the level of clinical severity to establish management guidelines and to facilitate dengue reporting and surveillance. Warning signs were proposed to facilitate triage and early detection of potentially severe cases that need hospitalization, particularly in primary care settings and during outbreaks.
Several studies were conducted to assess the utility of the new dengue classification scheme in clinical practice [14][15][16][17][18][19]. After the release of the new WHO classification in 2009, the level of care required by patients with dengue was used by researchers as a gold standard to grade the severity of the illness to identify patients with severe disease that is likely to require high level of healthcare as in-patients and those with milder disease that can be managed as out-patients [11,14]. Such classification would conceivably improve outcome and reduce cost of healthcare. Additionally, the level of care is the most reliable indicator of severity in retrospective studies that were primarily used to assess the utility of the new classification in clinical practice. However, current data remain insufficient to establish the validity of this classification in predicting or identifying severe dengue cases that may need close observation or hospitalization for proper management.
This study aimed to assess the validity of the new dengue classification scheme based on data from Jeddah city as part of global evaluation of the new dengue classification.

Population and setting
This retrospective study included patients with dengue virus infection reported to the infection control unit at King Abdulaziz University Hospital (KAUH), Western Saudi Arabia over a 4-year period from January 2014 through December 2017. Only laboratory-confirmed cases presenting within 7 days of disease onset were included in the study. The diagnosis was confirmed if at least one of the following criteria was met in acute phase serum: (1) positive reverse transcription polymerase chain reaction (RT-PCR), (2) positive serology for dengue IgM, or (3) positive dengue-specific non-structural antigen-1 (NS1). Exclusion criteria included patients who presented 7 days after the onset of symptoms, or those who were transferred to other hospitals, or whose data were unavailable.

Undifferentiated (non-classical) dengue
Non-specific febrile illness that does not meet the criteria for classical dengue fever, dengue hemorrhagic fever (DHF), or dengue shock syndrome (DSS).
• Supportive serology or epidemic context (occurrence at the same location and time as other confirmed cases of DF).

Dengue hemorrhagic fever (DHF)
All of the following must be present: • Fever or history of acute fever, lasting 2-7 days, occasionally biphasic • Hemorrhagic manifestations • Thrombocytopenia (<100,000 platelets/mm 3 ) • Evidence of plasma leakage due to increased vascular permeability manifest as either fluid accumulation (e.g. peripheral edema, pleural effusion, pericardial effusion, pulmonary edema, or ascites, diagnosed clinically or radiologically) or hemoconcentration (rise in hematocrit [HCT] �20% of the patient's baseline level or a drop in HCT �20% of the baseline level following rehydration.

Definition of cases and warning signs
Persistent vomiting was defined as vomiting at least 5 times per day or vomiting everything the patient ingested. Shock was defined as tachycardia (pulse rate > 100 beats/minute) with either hypotension-for-age or narrow pulse pressure (<20 mmHg). Severe bleeding was defined as major bleeding (hematemesis, melena, or menorrhagia) associated with systolic hypotension, haemoglobin <8 g/dL, or a drop of haemoglobin of >2 g/dL, or bleeding that required blood transfusion. Renal impairment was defined as serum creatinine rise of at least 50% over the baseline that failed to improve after two days of re-hydration. The baseline creatinine was defined as the minimum value following two days of re-hydration. Peak creatinine was defined as the highest creatinine value recorded following two days of re-hydration. The baseline haematocrit (HCT) value was defined as the minimum HCT value following a minimum of two days of re-hydration provided that the patient had passed the 6 th day of illness. If no baseline HCT could be defined for a given patient, the hospital's average value of normal HCT (41% for adults and 42% for pediatrics) was used as the estimated baseline value. Peak HCT was the maximum HCT value recorded during hospital stay. Patients' clinical outcomes were determined using data from the time of hospital presentation to the time of hospital discharge or demise, and from any subsequent follow up data when available.
Warning signs were only considered at patient's presentation. The day patients developed severe dengue was documented. Lethargy was not included as a warning sign due to inadequate documentation of this symptom in the patients' records.

Patients' stratification by level of care
Therapeutic interventions and healthcare required by patients were classified into three levels: level I, included patients who were treated on outpatient basis; level II, included hospitalized patients who received intravenous fluids for rehydration and/or those who received platelets due to thrombocytopenia that was not associated with major mucosal bleeding; level III, included hospitalized patients who required intravenous fluids for resuscitation, mechanical ventilation, blood transfusion, inotropic support, or specific treatment for organ failure.
The authors determined the level of care (the gold standard) for each patient based on the maximum healthcare required by the patient any time during his/her entire hospital stay. For instance, if a patient with dengue was stable at presentation but 2 days later the patient required ICU admission, level III of care (patients requiring intensive care) will be assigned to this patient. The aim of the classification was to determine the level of care the patient is most likely to require and to predict severe dengue before it happens. The gold standard was assessed during the phase of data collection (interventions needed were answered as yes or no) and phase of statistical analysis (patients were grouped as level I, II, III according to the pre-set definitions in the research proposal). Assignment of patients to the level of care they required was reviewed and confirmed by the senior author (TAM) who is a professor of medicine and infectious disease with a vast and long experience in managing dengue fever.

Statistical analysis
Data were analyzed using IBM SPSS, version 22. P values of <0.05 were considered statistically significant. Chi-square test was used to compare categorical variables. Diagnostic values including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the old and the new WHO classifications for identifying severe cases were analyzed in reference to the level of healthcare the patients required. Thus, the level of healthcare required by patients was the basis for determining severity of dengue in a retrospective manner. For the old classification, patients who were classified as DHF/DSS were considered to have severe dengue, while those classified as non-classical or classical DF were considered to have non-severe dengue. For the new classification, patients classified as SDF were considered to have severe dengue and those classified as dengue with or without warning signs were considered to have non-severe DF. With respect to the gold standard (management level), severe cases were defined as patients who were managed with level III care, whereas non-severe cases were patients who required level I or II care.

Ethics statement
The study was approved by the Institutional Review Board of King Abdulaziz University Hospital, Jeddah, Saudi Arabia. All data analyzed were anonymized.
The clinical presentation and laboratory characteristics of the 357 eligible patients are summarized in Tables 3 and 4. The mean interval from the onset of illness to the hospital presentation was 4 ± 1�8 days (range 1-7, median 4 days). One of the 357 eligible patients met the inclusion criteria despite having no fever. The inclusion criteria were any patient with clinically suspected dengue that was laboratory-confirmed with positive IgM, PCR, or NS1 and presenting within 7 days after the onset of symptoms. This patient was a 25 y old Somalian woman with one day history of generalized body ache, vomiting, menorrhagia, and dyspnea Examination revealed jaundice and splenomegaly. Platelet count was 9.0 x10 3 /mm 3 , WBC, 1.0 x10 3 /mm 3 , and hemoglobin, 4.6 g/dl. Dengue was clinically suspected because of her thrombocytopenia and leukopenia and the fact that she presented during an outbreak of dengue. Dengue IgM was positive. She was managed with blood transfusion and supportive care and the patient fully recovered.
Since the old and the new classifications developed by the WHO were applicable to both adults and children, we decided to analyze and present the data for the whole group. Despite the small number of children (53 child or 15% of our study population), subgroup analysis was performed to see if there were any major differences between adults and children and there was none.

Evaluation of the classifications for screening for level III of care
Cross-tabulations of the level of care (level I/II versus level III) with the old and the new classifications showed 86% and 93% proportional agreement (PA) of the old and the new classifications with the level of care, respectively (Tables 5 and 6).
Diagnostic values including sensitivity, specificity, PPV and NPV of the old and the new classifications in identifying patients who required level III of care is presented in Table 7.

Proposed revision of the new WHO classification
Of the 40 patients who needed level III of care, 14 (35%) patients were not classifiable as SDF using the WHO 2009 classification. Of these 14 cases, 8 had co-existing hematological conditions (sickle cell anemia and hereditary spherocytosis) that caused severe drop in hemoglobin level necessitating urgent blood transfusion, 5 patients had severe thrombocytopenia (<20,000 platelets/mm 3 ) with evidence of bleeding, and one patient had a neurological deficit that was exacerbated by dengue. Thus, the 14 patients with SDF as per the gold standard (ie requiring level III of care) who were not identified by the new WHO classification were either patients with decompensated chronic medical diseases, hematological diseases presenting with aplastic crisis, or thrombocytopenia <20,000 platelets/mm 3 and minor or major bleeding.
To improve the sensitivity of the clinical assessment in identifying patients who would require level III of care, we propose two revised versions of the new classification by integrating new criteria to the definition of SDF. The first proposed revision integrates patients with acutely decompensated chronic disease; e.g. patients with known cardiomyopathy who present with acute heart failure or patients with known hematological disease, such as sickle cell anemia, presenting with aplastic crisis. This first revision improves the sensitivity from 65% to 85%, the PPV from 74% to 79%, and the NPV from 96% to 98%, with no change in specificity (97%). The second proposed revision also integrates (besides decompensated chronic disease) patients with thrombocytopenia of <20,000 platelets/mm 3 with any bleeding even if minor. This second revision results in further improvement of the sensitivity from 65% to 98%, the PPV from 74% to 80%, and the NPV from 96% to 99�7%, with without affecting specificity (97%). The diagnostic value of the first and second proposed revisions of SDF in the new classification is presented in Table 8 and the revised criteria are presented in Table 9.

Screening for severe outcomes
Severe outcomes were defined as patients who recovered after having complications directly or indirectly related to dengue or those who died. The sensitivity of the new classification, the 1 st proposed revision, and the 2 nd proposed revision, for identifying patients with severe outcomes was 72%, 84%, and 88%, respectively, and the NPV was 97�7%, 99%, and 99%, respectively (Table 10).  (Table 11). Correlation between the number of warning signs (<2 versus �2), the old and the new classifications, and the level of care is presented in Table 12.

Discussion
This study compared the old and the new WHO dengue classifications as predictors of both levels of healthcare and patients' outcomes among 357 patients with confirmed dengue. It demonstrated that both classifications were inadequate in identifying patients who required advanced level of healthcare. Both classifications had a low-to-moderate sensitivity (30% to 65%), although the new classification had an acceptable sensitivity in predicting severe outcomes (72%). Consequently, we propose to revise the new classification by integrating 2 new criteria in the definition of SDF, namely, acute decompensation of chronic diseases and evidence of minor bleeding in association with thrombocytopenia <20,000 platelets/mm 3 . The proposed revision improved the sensitivity to 98% in detecting patients who need level III healthcare without significantly altering specificity, which remained high (97%). It also improved sensitivity of predicting severe outcomes to 88%. Furthermore, the proposed revisions had 99�7% and 99% NPVs for the level of care and outcome severity, respectively. This means that a patient who would not be classified as SDF according to our proposed criteria would have 0�3% probability to require level III of healthcare and 1% probability of having severe outcome including complications and mortality; versus 5% each for the original classification, respectively. Pre-existing comorbidities have been considered as risk factors for progressing to severe dengue in the literature. In this study, patients with pre-existing morbidities that were clinically stable at presentation and throughout the duration of dengue illness did not need to be managed as severe cases (level III of care). On the other hand, patients known to have chronic diseases who presented with acute decompensation (unstable disease) at presentation or any time during their dengue illness needed to be managed as severe dengue and received level III of care; e.g. patients with known cardiomyopathy who presented with acute heart failure or patients with known hematological disease, such as sickle cell anemia, who presented with aplastic crisis.
Data from literature generally suggest that the new classification had a higher sensitivity as well as a higher or comparable specificity in identifying severe cases requiring higher level of healthcare in comparison with the old classification [14][15][16][17][18][19]. Several studies that compared the WHO classifications showed that a significant proportion of patients with SDF were misclassified as classical DF using the old classification, while they were readily identified by using SDF in the new classification [14][15][16]. Other studies compared the accuracy of the two classifications in certain clinical contexts. For example, a Brazilian study of 267 pediatric cases, showed that the old classification had a lower sensitivity (62% versus 87%) but a higher specificity (93% versus 73%) in discriminating severe cases compared to the new classification [20]. Another Brazilian retrospective study evaluating 121 autopsied individuals who died during Assessment of the new dengue classification 2011-2012 dengue epidemics, showed that the new classification had a higher sensitivity to discriminate dengue deaths and that the absence of plasma leakage and thrombocytopenia were the main reasons for failure of the old classification to discriminate DHF cases [21]. A review by Bandyopadhyay et al., analyzed 37 studies using the old WHO dengue classification, and demonstrated that this classification had a low sensitivity with frequent overlap between the different classes, especially in endemic areas [22]. Additionally, the new classification has been shown to be a better measure for case reporting and surveillance [21][22][23]. Conversely, other authors found no difference between the two classifications in term of sensitivity [24,25].
The new classification intended to develop a system that helps in directing patients' management and improving clinical outcome by reducing morbidity and mortality. Beyond the controversy over the usefulness of the old and the new WHO definition of SDF in identifying severe cases, the practicability of the new definition in epidemic contexts was contested, as it was perceived to be entailing heavier workload to healthcare personnel [24]. According to the new classification, all patients presenting with dengue warning signs should be admitted for observation. This would raise the hospitalization rate from 10% (DHF/DSS) to 75% (D+W/ SDF) according to our data, resulting in unnecessary observation/admission, which might overwhelm healthcare personnel and resources particularly during outbreaks. As shown in our study, almost half of D+W patients were managed as outpatients, which makes the relevance of warning signs in terms of disease severity questionable [14][15][16][17]. Only hemoconcentration with concurrent drop in the platelet count was a significant predictor of severe outcome and   Assessment of the new dengue classification the need for advanced healthcare. Most of the previous studies found that none of the suggested warning signs was a significant predictor of dengue severity [24,26,27], although one of them showed that the presence of five or more warning signs was a significant predictor of SDF [27]. A multi-center study across 18 countries, assessing user-friendliness and acceptance of the new classification from health professionals' viewpoint, showed that 24% of health professionals had concerns with the new classification including: a possible increase of hospitalization rates, non-specificity of warning signs, a possible increase of cost if more patients were admitted, and the need for more training and dissemination of more concise clinical protocols [17]. These disadvantages called for a revision to improve its practicability and specificity in identifying severe cases.
In the present study, the new definition of SDF missed 14 (35%) patients who needed advanced (level III) medical care. Of these 14 cases, 8 had co-existing hematological conditions (sickle cell anemia and hereditary spherocytosis) that caused severe drop in hemoglobin level necessitating urgent blood transfusion, 5 patients had severe thrombocytopenia (<20,000 platelets/mm 3 ) with evidence of bleeding, and one patient had a neurological deficit that was exacerbated by dengue. Adding two new criteria, namely, thrombocytopenia <20,0000 platelets/mm 3 with evidence of bleeding, even though minor, and decompensated chronic illness, to the definition of SDF improved its sensitivity to identify patients who needed advanced level of care from 65% to 98% and its sensitivity to predict cases likely to have morbidity or mortality from 72% to 88%. The first criterion suggested to be added to our proposed revision of the new classification as an indication of SDF is "acute decompensation of a preexisting comorbidity". Hematological conditions constituted majority of the cases where 8/10 patients who had a co-existing sickle cell anemia or hereditary spherocytosis developed aplastic crisis. Dengue is a hemorrhagic virus and it's known to cause thrombocytopenia and leukopenia but not anemia. However, a pre-existing defect along the line of red blood cells potentiated acute anemia.
The second criterion suggested to be included in our proposed revision of the new SDF definition is "thrombocytopenia<20,000 platelets/mm 3 with any bleeding, even though minor". Thrombocytopenia �100,000 platelets/mm 3 constitutes only one of the warning signs in the original version of the new WHO classification (2009) but it is not included in the definition of SDF. In a French-Polynesian study, thrombocytopenia <20,000 platelets/mm 3 was associated with a longer hospital stay, more frequent admission to ICUs, and higher mortality [28]. However, only two thirds of cases were classified as DHF/DSS and the remaining one third was classified as DF resulting in underestimation of the severity of illness [28]. This is consistent with our results demonstrating that severe thrombocytopenia (<20,000 platelets/mm 3 ) is a strong indicator for severe dengue and the need for advanced level of care. Therefore, the new classification should be revised to include severe thrombocytopenia in the definition of SDF to improve identification of severe cases.
In our study, 9/35 patients classified as SDF received level II of healthcare and the remaining 26 patients were hospitalized for close observation and monitoring. Of these 26 patients, 3 patients had suspicion of liver damage (aminotransferases >1000 U/L), which proved to be a transient elevation of liver enzymes resolving spontaneously without complications. Among the patients who had neurological manifestations, impaired consciousness was the most frequent sign (5%), followed by meningeal signs (5%), and convulsion (3%) with one patient presenting with encephalitis. All cases with neurological manifestations benefited from conservative management. Similar to our findings, a Vietnamese study reported that impaired consciousness was the most frequent neurological manifestation [29]. Impaired consciousness is part of SDF criteria according to the new classification. A study by Gupta et al, demonstrated a greater risk for neurological complications among patients classified as classical DF in the old classification which did not include neurological symptoms or low level of consciousness at presentation as signs of SDF [30]. Recent reviews also highlighted a rising trend of neurological complications of dengue and that it is associated with more frequent hemorrhagic manifestations, higher prevalence of DSS, and increased hospital stay and mortality [31]. A study from Brazil reported 21% of neurological manifestations with confusion being the most frequent sign [32]. In Europe, an even higher prevalence (24%) of neurological manifestations was reported among imported cases of dengue infection, mainly including cases acquired in Asia and the Americas [33]. This variability in the prevalence of neurological manifestations may be explained by varying time of clinical assessment with respect to patient's first presentation and symptoms onset. Furthermore, lack of clear definitions of organ damage that reflects the actual disease severity may lead to misplacement of patients under SDF category. In our cohort, only 61% of patients having organ damage needed level III care. Thus, standardized definition of organ damage is needed to improve the classification specificity.
Having a more sensitive classification will help capture severe cases and place patients in the right interventional care level they need to improve their quality of care and subsequently decrease morbidity and mortality associated with dengue. Having a firm classification will direct patient management as per guidelines instead of relying on individualized decisions and experiences in directing patient care. Additionally, having a more sensitive classification without altering specificity will help direct healthcare resources and avoid unnecessary hospitalization/observation which usually overwhelms health care systems especially during outbreaks.
Limitations of our study include the retrospective design yielding incomplete observations, especially for cases treated on outpatient basis, which may have compromised the robustness of the analysis. Moreover, most of the severe cases presented at a late stage, which limited the accuracy of warning signs assessment and analysis as predictors for severe dengue. Furthermore, this is a limited study in a specific area and transferability to other countries is questionable.
In conclusion, the new WHO dengue classification had low sensitivity for identifying patients in need of advanced level of care and for predicting morbidity and mortality. This classification needs to be revised to improve its sensitivity in predicting the required level of care. It is proposed to include 2 additional criteria to the definition of SDF, namely, decompensation of chronic disease and thrombocytopenia <20,000 platelets/mm 3 with evidence of any bleeding, even though minor. Adding these two criteria substantially improves the sensitivity to predict cases requiring advanced level of care and to predict severe outcomes. Further prospective controlled two-arm study designs are needed from more than one region and different countries to confirm our results. Further, criteria such as organ damage and impaired consciousness should be distinctively defined to avoid overlap of the definitions and misclassification.