Comparing the Usefulness of the 1997 and 2009 WHO Dengue Case Classification: A Systematic Literature Review

The 1997 and 2009 WHO dengue case classifications were compared in a systematic review with 12 eligible studies (4 prospective). Ten expert opinion articles were used for discussion. For the 2009 WHO classification studies show: when determining severe dengue sensitivity ranges between 59–98% (88%/98%: prospective studies), specificity between 41–99% (99%: prospective study) - comparing the 1997 WHO classification: sensitivity 24.8–89.9% (24.8%/74%: prospective studies), specificity: 25%/100% (100%: prospective study). The application of the 2009 WHO classification is easy, however for (non-severe) dengue there may be a risk of monitoring increased case numbers. Warning signs validation studies are needed. For epidemiological/pathogenesis research use of the 2009 WHO classification, opinion papers show that ease of application, increased sensitivity (severe dengue) and international comparability are advantageous; 3 severe dengue criteria (severe plasma leakage, severe bleeding, severe organ manifestation) are useful research endpoints. The 2009 WHO classification has clear advantages for clinical use, use in epidemiology is promising and research use may at least not be a disadvantage.


INTRODUCTION
The World Health Organization (WHO), with its Special Program for Research and Training in Tropical Diseases (WHO/TDR), issued new dengue guidelines in 2009, 1 including the 2009 WHO dengue case classification: dengue and severe dengue (D/SD). Warning signs (WS) have been established for triage to help clinicians with symptomatic cases in need of closer surveillance and/or hospitalization (dengue with warning signs [D+WS]).
Historically, the 1997 WHO dengue case classification (dengue fever (DF), dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) was developed in 1975 by expert consensus based on studies on Thai children in the 1950s and 60s, with modifications in 1986 and 1997. 2 In the last modification in 1997, four grades of DHF were defined (DHF1, −2, −3, and −4), with DHF1 and −2 being DHF and DHF3 and −4 being DSS. 3 In this work, we will refer to the dengue case classification recommended by the WHO in 2009 as the D/SD classification and the WHO 1997 classification as the DF/DHF/ DSS classification.
The reasons for developing D/SD were the shortcomings of DF/DHF/DSS, which were established in many studies and furthermore, summarized in a systematic review. 2 DF/DHF/DSS (1) is poorly related to disease severity, (2) misdirects clinicians identifying severe disease, (3) is difficult to use (tests required are often not available/difficult to apply), (4) does not help for triage in outbreaks, and (5) leads to different reporting globally as a result of the difficulties in using the classification for reporting clinicians.
The main emphasis of D/SD is, therefore, to help clinicians to identify and manage cases of severe dengue timely. It helps to save resources and contributes to a reduction of dengue mortality.
Based on the largest prospective multicenter study, the Dengue and Control (DENCO) study, 4 D/SD describes dengue as it currently occurs globally, focusing on severe dengue, defined as plasma leakage (shock or fluid accumulation with respiratory distress, which includes the former DSS), severe bleeding, or severe organ manifestation. With the improved description of dengue cases, case reporting is facilitated. WS have been empirically validated to some extent in the DENCO study. A larger study is currently under way to evaluate and define the predictive value of WS in outpatients (for the need of hospitalization) and inpatients (for severe disease). 5 Furthermore, D/SD is based on best available evidence (evidence grade 1/2) 6 on each step of the development from basic to implementation research. 7 A discussion evolved internationally on the usefulness and applicability of D/SD compared with DF/DHF/DSS, and in a relatively short period of time, numerous studies have been published by many independent research groups. The objective of this study is to provide a systematic literature review of the studies published, comparing D/SD and DF/DHF/DSS to facilitate the discussion about the usefulness of the different classification systems.

METHODS
Reporting items for systematic reviews and meta-analyses (PRISMA) Statement for systematic reviews and metaanalyses 8 were followed. This study defines (1) case definition as the description of clinical and laboratory parameters to define a dengue case compared with other febrile illnesses and (2) case classification as the different stages of the spectrum of dengue severity, either D, including D+WS, or SD.
Eligibility criteria included (1) research on dengue case classification and/or dengue case definition, (2) comparison of D/SD and DF/DHF/DSS, (3) any comparative/analytical study design, and (4) published after the publication of the new WHO dengue guidelines. 1 All languages were included; the search was conducted in English only. Studies looking only at one classification were excluded for failing to compare the different models. Also excluded were studies not performed in dengue-endemic countries. Studies without a defined methodology, including expert opinion and conference proceedings, were collected and used for the discussion.
The literature search and analysis were developed and carried out through July 15, 2013 with two data extractors. The search terms derived from two major categories were (1) dengue disease (dengue) combined with (2) classification and/or definition.
The search strategy was applied to the following databases: US National Library of Medicine and the National Institutes of Health Medical Database (PubMed), the Latin American and Caribbean Health Sciences Database (Lilacs), Excerpta Medica Database (EMBASE), and the Cochrane Database of Systematic Reviews (CDSR). The WHO library database (WHOLIS) and Google Scholar were searched for grey literature. Relevant literature was screened for additional articles in the reference sections.
Because of the limited search options available to this field of research, a very broad search strategy was used, using a reduced number of combinations to increase the number of initial hits.
All results were screened for duplicates. In the next stage, results were screened based on the title and abstract only, and the full texts of potentially relevant studies were subsequently assessed. Relevant information was tabulated in evidence tables.
Study designs were divided into the following categories: (1) prospective studies of dengue cases, (2) studies analyzing post-hoc existing prospectively collected databases of dengue cases, (3) studies reviewing retrospectively existing medical charts of dengue cases, and (4) acceptability studies (mostly of qualitative design).
No studies were excluded in the analysis for quality reasons if the eligibility criteria were met, but limitations and possible biases are reported in the results section. The analysis followed the categories according to the use of the case classification: (1) clinical use (establishing severe dengue disease using WS for identifying severe dengue disease, detecting dengue, usefulness for triage, usefulness for outbreaks, and ease of application by clinicians), (2) use in surveillance detecting dengue cases and detecting dengue outbreaks as well as reporting of the different levels of severity of dengue disease (D/SD), and (3) use in research.

RESULTS
In total, 782 studies were identified during the electronic search that were potentially relevant to the research question, including duplicates. After the screening of titles and abstracts, 25 studies remained eligible.  Eligible studies are tabulated in Table 1. Ten studies [22][23][24][25][26][27][28][29][30][31] were relevant to the research question based purely on expert opinion, without stating a particular methodology as to how the results were derived. These studies were excluded from the analysis but reflected in the discussion for the purpose of highlighting the different opinions. Three additional studies were excluded after full assessment of the text, because they only assessed DF/DHF/DSS without comparison with D/SD 32,33 or dealt with a particular subgroup of cases not representative of dengue-endemic countries (dengue in travellers 34 ).
Of the remaining 12 studies, 10-21 11 studies were from published databases, and one study was from grey literature. No other studies were identified from the reference lists. Figure 1 summarizes the process of selection in a flowchart.
All studies were published after 2009-the date set for initiating the search. Most of the studies included (prospective, post-hoc analysis of existing datasets or review of existing medical charts, and any qualitative design) were performed in Asia, with the exception of three studies: one study that included 18 study sites worldwide, 10 one study from Nicaragua, 16 and one study from Peru. 18 Studies [10][11][12][13] are prospective studies published between 2010 and 2013. All other studies used datasets either prospectively developed and compared post-hoc with the use of D/SD and DF/DHF/DSS or retrospectively derived from case notes of dengue-positive cases developed over time and without a strict study protocol. One study 10 also included the acceptability of the different dengue case classifications with qualitative methods.
Sample sizes of the studies 10-21 were relatively small, ranging from 50 to 300 cases (mostly less than 100 cases), with the exception of three studies 10,17,20 with more than 1,000 cases each. Most of the studies included adults and children; only three studies were confined to pediatric centers. 12,16,21 All studies were performed in hospital settings of either secondary or tertiary level hospitals. The study in ref. 10with 18 study sites-had the biggest scale in case numbers and the most comprehensive scope with all levels of healthcare delivery systems included in Asian as well as Latin American countries.
Studies often addressed multiple questions; the following results are presented according to the categories described in the methods sections. In general, the prospectively collected studies are largely in favor of the use of D/SD in the clinical setting; the retrospective studies have similar results, apart from the study in ref. 15 Clinical use. Which classification describes the clinical picture of dengue cases better? The studies dealing with this question all agree that D/SD classifies dengue as it occurs clinically. The study in ref. 10 in 18 countries with the biggest sample overall concludes that "13.7% of cases could not be classified according to DF/DHF/DSS (missing data e.g. haematocrit, platelet counts, and tourniquet tests on 50% of charts), compared to 1.6% for D/SD. 32.1% of severe dengue cases could not be classified in the DF/DHF/DSS system." The study in ref. 17, also in a large sample, reconfirms that "DF/DHF/DSS and D/SD were discordant in defining severe disease (p: 0.001)." D/SD describes severe disease, whereas DF/DHF/DSS describes severe disease less well. Similarly, findings in the study in ref. 19 are "a cross tabulation [that] showed DF cases were distributed in all of the severity groups stratified by D/D+WS/SD (53.8% Group D/45.4% Group D+WS/0.8% Group SD). Of the DHF cases, 23 (79%) were categorized as Group D+WS, and six (20.7%) as Group SD." When looking at quality of studies, the studies in refs. 10, 17, and 19 of this systematic literature review provide probably the best available quantitative evidence for this question.

Conclusion
Prospective studies (follow-up of dengue cases) 10 Barniol and others 2011 Mixed methods, three arms: (1) quantitative (prospective and retrospective chart reviews), (2) triage and management usefulness; (3) user-friendliness and acceptance (1) Prospective/ retrospective medical chart reviews; (2) Self-applied questionnaires; (3) focus groups 13.7% of cases could not be classified according to DF/DHF/DSS (missing data; e.g., hematocrit, platelet counts, and tourniquet tests on 50% of charts) compared with 1.6% for D/SD. WS for SD (necessary for the D+WS category) were documented in a large proportion. The revised classification also proved to be more sensitive for timely recognition of SD. 32.1% of SD cases could not be classified in the DF/DHF/DSS system. D/SD was easily applicable in clinical practice, also seen to be useful for triage and case management by medical staff more frequently than the DF/DHF/DSS classification. Concerns in questionnaires and focus groups included (1)   (continued)  (continued) Step 2: An analysis of regional and national dengue guidelines: need to re-evaluate and standardize guidelines; actual ones showed a large variation of definitions, an inconsistent application by medical staff, and a lack of diagnostic facilities necessary for the DHF diagnosis in frontline services.
Step 3: A prospective cohort study in seven countries, a clear distinction between SD (defined by plasma leakage and/or severe hemorrhage and/or organ failure) and (non-severe) dengue can be made.
Step 4: Three regional expert consensus groups in the Americas and Asia concluded that dengue is one disease entity with different clinical presentations and often, unpredictable clinical evolution.
Step 5: Global expert consensus meeting at WHO in Geneva, Switzerland-the evidence collected in steps 1-4 was reviewed, and a revised scheme was developed and accepted, distinguishing D/SD.   Identifying dengue (case definition). Only one study dealt with the question of whether the case definitions for D are more useful to detect dengue compared with DF. 14 The results-retrospectively based on 164 dengue cases compared with 200 other febrile illnesses (OFIs)-show for DF (as defined by the DF/DHF/DSS classification) a sensitivity of 95.4% and a specificity of 36.0% and for D (as defined by the D/SD classification) a sensitivity of 79.9% and specificity of 57.0%. The study acknowledges that D/SD was not primarily designed for detecting dengue. When using an NS1 test, the sensitivity and specificity for both dengue case classifications improve considerably (Table 1).
Identifying severe disease (case classification). Several studies looked at sensitivity and specificity of D/SD and DF/DHF/ DSS for identifying severe dengue. Sensitivity for detection of severe disease has been calculated by five studies, 11,13,16,18,21 with results in the range from 59% to 98% for SD (as defined by the D/SD classification). The prospective studies 11,13 estimate sensitivity at 88% and 98%, respectively. Specificity is estimated between 41% and 99% (99% for the prospective study presenting values 11 ). Sensitivity and specificity for DHF (as defined by the DF/DHF/DSS classification) range between 24.8% and 89.9% and between 25% and 100%, respectively (sensitivity of 24.8% and 74% for the prospective studies and 100% specificity for the prospective study presenting data) ( Table 2).
In summary, comparing DF/DHF/DSS with D/SD, it seems that, in most studies, D/SD turned out to be better in terms of sensitivity and specificity for distinguishing D and SD, which is summarized in the study in ref. 10 (here derived from qualitative data: "the revised classification also proved to be more sensitive for timely recognition of severe disease").
WS for SD disease. Several studies looked at the value of WS as predictors of severe disease. The study in ref. 11 suggests that having at least five WS may be a good predictor of severe disease; in particular, lymphocyte counts 1,500 cells/mm 3 , platelet counts 20,000/mm 3 , and raised AST levels were associated with SD. However, in the study in ref. 20, the three most common WS were lethargy, abdominal pain/tenderness, and mucosal bleeding. Also, no single WS alone or combined had a sensitivity of more than 64% in predicting severe disease. 20 Specificity was 90% for both DHF and SD with persistent vomiting, hepatomegaly, hematocrit rise, and rapid platelet drop. 20 In practice, however, it seems that WS are significant events that are well-documented in medical charts, 10 but using WS as compulsory hospitalization criteria may well increase workload.
Usefulness for triage and in outbreaks. The study in ref. 10 finds-with focus groups and questionnaires among clinicians in 18 countries-that D/SD "was easily applicable in clinical practice, also seen to be useful for triage and case management by medical staff, more frequently than the DF/DHF/DSS classification." The study in ref. 15 analyses medical charts of 274 confirmed dengue cases. When using DF/DHF/DSS, more intensive treatment (with careful monitoring and intravenous rehydration) was needed in 94 cases compared with 189 cases when using D+WS and SD as criteria. Similarly, the study in ref. 17 calculates the requirement for hospitalization increasing from 17.0% to 51.3% using D/SD. This finding is in agreement with the findings in the study in ref. 18: "DF/DHF/DSS may save health care resources with its more stringent definition of a DHF case requiring the most intensive hospital intervention." However, the same study points out that "the use of D/SD guidelines in this outbreak facilitated the early admission of those with life threatening dengue disease for appropriate clinical management." Ease of application by clinicians. Only two studies assessed the ease of application by health personnel, with positive results overall for the D/SD classification: the study in ref. 10 shows, with questionnaires and focus groups, that D/SD was easily applicable in clinical practice more frequently than the DF/DHF/DSS classification; the former was also seen to be useful for triage and case management by medical staff. The study in ref. 16 analyzed with interrater agreement methods how clinicians agreed with each other when classifying the same cases according to either D/SD or DF/DHF/DSS; the former had substantial agreement (κ = 0.62, P = 0.001), and the latter had moderate agreement (κ = 0.46, P = 0.001).
Use in surveillance and research. No study examined the use of D/SD compared with DF/DHF/DSS in the context of dengue surveillance and/or reporting; the study in ref. 10 argued that, with the improved description of cases and especially, the improved sensitivity for severe cases, data  collection for epidemiological returns should be improved. However, the study in ref. 16 concludes that the use in epidemiology needs to be evaluated.
No study addressed whether D/SD has advantages or disadvantages for research. The study in ref. 16 states that its use in pathophysiological and epidemiological studies needs additional evaluation.
Because there were no studies identified that formally evaluated D/SD compared with DF/DHF/DSS in these areas, we refer to the discussion section below, where expert opinions/ viewpoints were also included.
Additional results. Most studies conclude that additional research needs to address unsolved questions, especially in larger studies (locally adapted case definitions for diagnosing dengue in the absence of confirmatory laboratory testing and the predictive value of WS in outpatients and inpatients) and different settings. 11,12 It has also been emphasized that training needs to be addressed as well as the development of local clinical protocols for dengue.

DISCUSSION
Limitations. Because D/SD was only introduced in 2009, the number of studies available is surprising, probably underlining the importance of this question. Nevertheless, the number of studies available for answering the study question was limited, and therefore, all available data in this study, including lower-quality studies and retrospective studies, had to be considered in this review. Only very few prospective studies are available, and even fewer qualitative studies have been reported. Regarding the quality of retrospective studies, it is often unclear if all data are available retrospectively to conduct such studies. Also, almost all studies reviewed are on a secondary or tertiary level of care and not primary care. Most studies are not in resource-poor settings. Furthermore, sample sizes are mostly small, and data from Asian countries prevail.
Publication bias is another concern, especially for the expert opinion papers; therefore, they have been excluded from the analysis and are mentioned only in the discussion.
Relative value of the two dengue case classifications. The first question is which purpose does the classification serve: clinical care, surveillance, or research (see also ref. 29). In this section, we will consider these three main areas of application of the dengue case classification.
For the clinical use of D/SD, the three prospective studies dealing with this question confirm that D/SD has advantages in defining the severity of dengue cases. 10,11,13 The evidence is even stronger when considering sample size. The study in ref. 10 is, by far, the biggest study in this context, conducted globally in 18 different dengue-affected countries, and the study in ref. 11 is a medium-sized study. The sensitivity and specificity analyses 11,13,16,18,21 are equally in favor of D/SD. However, this finding is no surprise, because the classification has been designed 4 "to develop a revised evidence-based classification that would better reflect clinical severity." The ability to distinguish D and SD-as it currently occurs globally-is probably the biggest advantage of D/SD. Expert opinion underlines this finding. The studies in refs. 26 and 29 argue-mainly with reference to the study by Alexander and others 4 and the study in ref. 10-that D/SD offers a better categorization than the DF/DHF/ DSS classification.
In a letter responding to study, 13 it has been queried whether the classification is useful for predicting dengue severity and also, if the application of the classification can be considered to have an easy application. 31 Clinical use of the classification also includes the use of WS. It would be beneficial to have robust and validated WS, but the level of evidence for most WS as established in D/SD is not very high, because large quantitative studies are needed to establish the validity of single parameters and/or combinations of parameters in predicting the progress to SD. This finding has been acknowledged by the two quantitative studies dealing with this question 12,20 and the expert opinions: responding to the study in ref. 20, it has been mentioned that convincing studies about WS need to be prospective and should include primary care. 30 The WS, as defined in the D/SD framework, were considered to require more specific definitions. 22  The dengue case definition is important for diagnosing a dengue case, especially when diagnosing without laboratory confirmation in resource-poor settings. 14 The D/SD framework uses the same criteria for case definitions as defined by the DF/DHF/DSS framework (which is based on expert opinion), with some modifications also driven by expert opinion. However, it is well-known that using only clinical criteria for discriminating between different febrile illnesses is difficult or impossible. Similar to the WS, this question can only be addressed with large prospective global studies, which are under way (IDAMS study above).
This finding leads to the question of whether a clinically based dengue diagnosis and the application of WS can be useful for triage. Some studies, based on qualitative arguments 10 and expressing the viewpoints of clinicians, come to a positive answer; others refer to the fact that the ease of application is also documented in studies. 16 One retrospective study argues that the caseload for hospitalization increases dramatically. 15 However, the same study acknowledges that only 68% of cases with SD disease can be classified with the DF/DHF/DSS framework. In other words, 32% of cases are not classifiable and maybe missed, reflecting the low sensitivity of DHF. This result has been summarized in the study in ref. 18, which acknowledges that the workload may potentially increase with the broad use of WS but that cases with severe disease are correctly identified. It seems that the application of the WS in the algorithm needs adjustment to the locally available resources, which was mentioned in the WHO guidelines 1 : "depending on the clinical manifestations and other circumstances, patients may . . . be referred for in-hospital management (Group B)." Whether increased admissions with early treatment may reduce severe/ fatal illnesses needs to be analyzed as well.
The expert opinion papers vary on this issue. One study 22 supports the view that D/SD is beneficial for triage; another study 31 reports the excellent experience of using D/SD for grouping of patients for additional management.
It would be useful to see more studies using the algorithms for treatment as suggested in the WHO guidelines 1 and the Clinical Handbook on Dengue Management, 9 with local adaptations in prospective studies-maybe in outbreak situations-and use of qualitative methods to assess the usefulness of D/SD in real-life circumstances.
No quantitative studies assessed the use of the D/SD versus the DF/DHF/DSS classification in dengue surveillance and case reporting. However, the difficulties in applying DF/ DHF/DSS led to the development of multiple local adaptations and put global comparability at risk. Studies assessing the use of D/SD in clinical settings argued that the improved description of dengue as it occurs globally, leading to more specific categorizations according disease severity, should lead to improved reporting. Likewise, the opinion-based papers in this systematic review argue that D/SD 23 "maybe useful for surveillance and reporting" and provide the opportunity of improved international comparison of data. 26 However, to confirm this answer, additional research is warranted.
For the use of the D/SD classification as an outcome in basic research or pathogenesis research, no quantitative data were available.
A viewpoint/opinion paper 24 argues that "pathogenesis research should be conducted as much as possible on carefully defined categories of human disease response. This requires splitting, not lumping." This work was responded to by a group of authors 29 arguing that "DF/DHF/DSS was too complicated to use in clinical or public health settings, yet was not sufficiently precise for detailed pathogenesis studies" and that "D/SD brings clarity, clinical and epidemiological utility, and the potential for development of more precise definitions of clinical phenotype for pathogenesis studies." Therefore, D/SD, with its three severity markerssevere plasma leakage, severe bleeding, and severe organ manifestation-can be further refined for basic research.
Two other views were expressed: (1) quantitative studies confirmed that the DF/DHF/DSS framework does not reflect levels of severity 10,17,19 ; and (2) critical evaluation of D/SD is limited to the use in basic research; however, this critical evaluation is not based on specific studies, but is based on expert opinion only. 23,24,27 Some studies suggest solutions-based on their analysisto reclassify the dengue disease classifications (for example, introducing subgroups for dengue with organ failure). However, it has to be underlined that these suggested solutions are based on each individual study presenting individual solutions 15,17 and not the result of the highest available evidence for the particular study question, the latter being one of the strengths of D/SD. 25 In conclusion, this systematic review confirms-almost 5 years after its introduction-that the D/SD classification is able to detect disease severity with high sensitivity, thus assisting clinical management and potentially, contributing to reducing dengue mortality.
The limitation of not having more appropriate studies available is also a result of this study, with the urgent call for studies to be designed prospectively and include primary care.
Diagnosis of dengue by clinical parameters only continues to be a challenge. It is recommended to await the findings of the ongoing large clinical studies for a possible adap-tation of case definitions and WS. Furthermore, it is recommended to study the performance of D/SD for triage, especially in outbreak situations. For surveillance and global reporting, a unified classification system with accurate information on disease severity would be advantageous.
For pathogenesis research, the D/SD classification may open new opportunities, with a fresh look at underlying pathology now that the spectrum of disease is better described.