Dengue disease surveillance: an updated systematic literature review

Objectives To review the evidence for the application of tools for dengue outbreak prediction/detection and trend monitoring in passive and active disease surveillance systems in order to develop recommendations for endemic countries and identify important research needs. Methods This systematic literature review followed the protocol of a review from 2008, extending the systematic search from January 2007 to February 2013 on PubMed, EMBASE, CDSR, WHOLIS and Lilacs. Data reporting followed the PRISMA statement. The eligibility criteria comprised (i) population at risk of dengue, (ii) dengue disease surveillance, (iii) outcome of surveillance described and (iv) empirical data evaluated. The analysis classified studies based on the purpose of the surveillance programme. The main limitation of the review was expected publication bias. Results A total of 1116 papers were identified of which 36 articles were included in the review. Four cohort-based prospective studies calculated expansion factors demonstrating remarkable levels of underreporting in the surveillance systems. Several studies demonstrated that enhancement methods such as laboratory support, sentinel-based reporting and staff motivation contributed to improvements in dengue reporting. Additional improvements for passive surveillance systems are possible by incorporating simple data forms/entry/electronic-based reporting; defining clear system objectives; performing data analysis at the lowest possible level (e.g. district); seeking regular data feedback. Six studies showed that serotype changes were positively correlated with the number of reported cases or with dengue incidence, with lag times of up to 6 months. Three studies found that data on internet searches and event-based surveillance correlated well with the epidemic curve derived from surveillance data. Conclusions Passive surveillance providing the baseline for outbreak alert should be strengthened and appropriate threshold levels for outbreak alerts investigated. Additional enhancement tools such as syndromic surveillance, laboratory support and motivation strategies can be added. Appropriate alert signals need to be identified and integrated into a risk assessment tool. Shifts in dengue serotypes/genotype or electronic event-based surveillance have also considerable potential as indicator in dengue surveillance. Further research on evidence-based response strategies and cost-effectiveness is needed. Objectifs Analyser les résultats de l'application d'outils pour la prédiction/détection des épidémies de dengue et la surveillance des tendances dans les systèmes de surveillance active et passive des maladies, afin d’élaborer des recommandations pour les pays endémiques et identifier les besoins importants de recherche.


Introduction
Dengue remains a major and growing public health threat worldwide. With the most recent study estimating that global infection rates of 390 million infections occur annually (Bhatt et al. 2013), the need for improved dengue surveillance is evident. Dengue surveillance is essential for the detection of outbreaks and, in the longer term, to monitor disease trends. In order to trigger timely interventions, outbreak alerts are particularly important to mobilise vector control and to prime or reorganise healthcare delivery services in preparation for a surge in suspected cases. Although vector control tools can be effective in principle, implementation remains an issue and effective dengue routine prevention is rarely achieved, particularly in high-density urban communities Pilger et al. 2010). Emergency vector control operations in response to dengue outbreaks 1116 are more typically applied, even though the efficacy of the most widespread method used, insecticide fogging or space-spraying, is dubious (Esu et al. 2010). Surveillance for dengue can include different indicators and systems (Harrington et al. 2013;Henning 2004;Stroup et al. 1989) to detect outbreaks and monitor trends. The authors' earlier systematic literature review (Runge Ranzinger et al. 2008) analysed 'the evidence on the structure, purpose and usefulness of dengue disease surveillance in dengue endemic countries' and described a general lack of evidence for the usefulness of dengue disease surveillance for early outbreak detection, especially the lack of indicators/alert signals available to trigger response. A stepwise adaptation of surveillance systems after evaluation in combination with active surveillance was recommended. Since then, new articles in the field of dengue surveillance have been published, and new initiatives towards early detection of dengue outbreaks have been launched (WHO expert meeting on dengue outbreak detection and response, June 2012). Common to all reports and recommendations is an increased recognition of the need for dengue control to focus on surveillance, vector control and adequate case management.
The aims of this review were to strengthen the evidence base and (where required) adjust the recommendations of Runge Ranzinger et al. (2008) by incorporating new trends and tools and to identify and summarise advances or improvements made. As in the earlier review, dengue vector surveillance was not covered here, but has been reviewed independently elsewhere (Bowman et al. 2014, in press).

Methodology
This review followed the protocol (York 2001) used in a previous systematic literature review on dengue disease surveillance, and the reporting guidelines set out in the PRISMA Statement for systematic reviews and metaanalyses (Liberati et al. 2009). The eligibility criteria of the reviewed literature were as follows: (i) population at risk of dengue, (ii) dengue disease surveillance, (iii) outcome of surveillance described and (iv) empirical data evaluated. After the recent dengue outbreaks in France and Croatia (29 and 30), the population at risk in the search was extended to include the European region. Literature reported in English, German and Spanish was included although the search was conducted in English only. Studies focusing on risk mapping, transmission dynamics, forecasting or prediction methods were excluded (e.g. Buczak et al. 2012; Chen & Chang 2013; Racloz et al. 2012), as they did not directly study surveillance systems.
The literature search and analyses were developed and continued until 15th February 2013, with two persons working as data extractors. Search fields included Medical Subject Heading (MeSH) terms/subjects and free text, considering population, intervention and outcome. The terms 'dengue' and 'severe dengue' (dengue fever (DF), dengue h(a)emorrhagic fever (DHF) and dengue shock syndrome (DSS)), 'surveillance' (disease, clinical, active, passive, sentinel, epidemiologic, population), 'communicable disease control', 'effectiveness', 'evaluation', 'disease notification', 'disease outbreaks', 'hospital and clinical laboratory information system' were used. The search strategy was adapted according to the databases, consistent with the process undertaken during the primary review published in 2008.
The search strategy was applied to the following databases to locate peer-reviewed studies: The United States National Library of Medicine and the National Institutes All results were screened for duplication by author, title, journal and publication date, and then screened for relevance, based on the title and abstract only. The full text of all studies considered to be relevant was then reviewed for final assessment by two independent data extractors. Where necessary, consensus was achieved by discussion. Relevant information, including study bibliographic information, study design and objectives, levels of endemicity and population, components of the surveillance system (surveillance subjects, scope and method), resources spent on the system, delivery of the surveillance system (information flow, outbreak and/or case definition, additional relevant information), purpose of the surveillance system and outcome attributes, was extracted and tabulated in evidence tables ( Table 1).
As many studies were descriptive or 'ecological studies' and therefore could not be ranked according to the 'hierarchy of study designs' (York 2001), the National Health and Medical Research Council (NHMRC 2000) evidence hierarchy (Merlin et al. 2009) was used to group the studies according to study design. Only studies at evidence 'level IV or level III-2 and III-3' were included. -Would be a low-cost option -Estimating an indicator near real-time -Dengue-related search queries are generally not as influenced by news coverage (panic-driven searches) -Despite strong historical correlations, it remains susceptible to false alerts.
-Sufficient search volume is needed.
-Rural areas and developing nations tend to lack or have limited Internet access currently.
-Intercountry comparisons may be difficult, each presented country and model must be considered independently. -In settings, with less developed surveillance systems, an internet search term-based system may yield significant gains in the rapidity of predictions.
-It is conceivable that an internet search term-based model may be a proxy for routine surveillance in specific settings.
-Individual models need to be developed for specific settings using local surveillance data and search terms.

Conclusions of study author:
The community-based programme captured twofold more fever and symptomatic dengue infections relative to study population size than the school-based system while monitoring five times as many people using the same number of personnel and the same amount of resources.
-Community-based surveillance allowed to identify symptomatic dengue cases in all age groups and was not solely limited to school-aged children.
-Several factors, including the research objective, sitespecific dengue epidemiology and cultural characteristics of the study population, will help determine the type of active surveillance system to implement. -On the civilian side, more time was necessary to detect the new dengue fever outbreak, in particular because it did not use any statistical tool to identify an increase in cases above a threshold.
-Several weeks were also necessary for the local vector borne disease committee to request a strengthening of existing vector control measures.

Conclusions of study author:
-Specific areas of acceptability to enter data could be significantly improved -The high sensitivity and low specificity of syndromic surveillance is characteristic.
This lack of specificity may lead to costly false alarms.
-Sensitivity of the system means that if utilised correctly by s stakeholders, it is unlikely to miss any disease epidemic where the primary symptom is fever.
-The timeliness and sensitivity were major strong points.
-Logging of patients into the system took too much time

Conclusions of study author:
-This was not a classic capture-recapture study, but rather an ecological study comparing incidence rates in a cohort to national surveillance rates in the surrounding urban areas.
-The HCSFV district, which borders Lake Managua, may have higher dengue rates than other health centres, as there was more reported dengue among the non-study population of the HCSFV.
-However, another plausible explanation is that these higher numbers are due to the impact of the PDCS study protocol and increased awareness. -

Conclusions of study author
-This active surveillance found a higher disease incidence than reported to the national surveillance system, particularly in preschool children and that disease incidence was high in both rural and urban areas.
-It also confirmed the previously observed focal nature of dengue virus transmission.
C. Trend monitoring and outbreak detection -Peak season was Sept to Nov.
-Airport fever screening was successful in identifying 45% of imported dengue cases with fever.
-No statistical impact on community transmission comparing presence and absence of airport fever screening.
-Dengue cases appeared to be positively associated with population density ( Conclusions of study author: -55.0% (298/542) of imported cases were temporary, nonfebrile, that is, latent cases, undetected by airport fever screening.
-50-90% of dengue infection cases are asymptomatic, and therefore, transmission of DENV into Taiwan via incoming travellers may be inevitable.
-The efficacy of screening symptomatic passengers passing through Taiwan airports by NCIT was found to have a PPV = 30.5-62.6% when fever prevalence among passengers was <1%.
-The effect on mitigating community transmission in dengue epidemics was not significantly different between pre-2003 border control methods and post-2003 methods (fever screening).

Conclusions of study author:
-Dengue in Taiwan is an adult infectious disease; elderly have high morbidity and mortality rates.
-Secondary infection increases the disease severity, but not mortality, dengue-infected adults are more symptomatic.
-The transmission cycle in Taiwan is unique, beginning with importation in the summer and ending in the winter.
-Severe disease can be caused by secondary infection with DENV-2 or primary infection with DENV-1 or DENV-3.
1148 Studies were grouped according to study types: models, time-series, case studies, ecological studies, evaluations, expert consensus, descriptive studies, prospective and retrospective cohorts.
No studies were excluded in the analysis for quality reasons if the eligibility criteria were met, and the limitations and possible biases in such studies are reported in the results section. The analysis grouped studies into four categories based on the purpose of the surveillance approach under investigation: (A) outbreak prediction/ detection; (B) trend monitoring; (C) both outbreak prediction/detection and trend monitoring; and (D) low/nonendemic countries.

Results
A total of 1116 studies, including duplicates, were identified during the electronic search as potentially relevant to the research question. After screening of titles and abstracts, 90 studies remained eligible. Full assessment of the text eliminated 54 further studies, leaving 36 studies included (Figure 1). Data of the 36 studies were extracted to a table (Table 1), also assigning a unique identifier number for each study.
When grouped according to purpose of the surveillance system studied and compared with the 2008 review, an increase in research interest in early outbreak detection was apparent, particularly in endemic countries: (A) outbreak prediction or detection (14 studies/previously 5); (B) monitoring dengue trends (4 studies/previously 6); (C) outbreak prediction and trend monitoring (9 studies/previously 7); and (D) non-endemic countries (9 studies/ previously 6). Within each of these categories, key components essential for improving surveillance in endemic and non-endemic countries were identified. The detailed findings are summarised and presented in Table 1.

Surveillance systems for outbreak detection and/or prediction (Groups A and C)
Most of these studies were from highly endemic settings and were intended to predict or detect outbreaks at an early stage.
Using electronic event-/search query-based surveillance for early detection of increased dengue activity. Two studies investigated the value of data quantifying the numbers of internet searches seeking dengue information in a number of countries (Bolivia, Brazil, India, Indonesia and Singapore; studies 1 and 2), by comparing with epidemiological data from the surveillance system using time-series analysis. The curve of the search queries over time was similar to the epidemic curve constructed from surveillance data, underlining the usefulness of this new and relatively simple approach. Study 13 used a real-time electronic approach based on Health Map in order to enhance timeliness and outbreak detection and to provide an added value for monitoring the ongoing spread of dengue.
Using the appearance of a new dengue serotype/genotype as an alert signal for dengue outbreaks. Six studies investigated serotype changes as a dengue outbreak signal (studies 3-8) using virus surveillance information, analysing laboratory data (genotyped or sequenced data) or hospital data (severity of cases) and examining how these correlated with the number of reported cases or dengue incidence. Five of these studies (3-7) also analysed whether a serotype shift or a clade replacement was positively associated with a subsequent epidemic. The sixth study investigated the correlation between population-wide serotype-specific data and an increase in cases (study 8) and showed that outbreaks occurred following the introduction of new serotypes in specific islands. Examining these studies in detail, retrospective studies in Singapore (studies 3,5) found that a serotype switch from DENV-2 to DENV-1 in 2004/2005 was associated with the 2005 epidemic. However, according to Schreiber (study 6), viral genome sequencing would not have been sufficient to predict this outbreak. A switch from DENV-1 back to DENV-2 in early 2007 was used as a warning sign and led to response actions that were believed to have reduced the impact of an outbreak 6 months later. A clade replacement within DENV-2 was also considered a contributing factor to the 2007 Singapore outbreak (study 3) and another outbreak at the end of 2010 (study 4). Similarly, three surveys in Surabaya (Indonesia) investigated prospectively the correlation of DENV type and disease incidence. Here, an increase in case numbers in 2010 was attributed to a genotype shift in DENV-1 from genotype IV to I between April and September 2009 (study 7). Retrospective analysis of serotype-specific surveillance data in the Pacific region (study 8) demonstrated that the rapid replacement of DENV-1 by DENV-4 in the region was associated with dengue outbreaks in 2008 and 2009 in Kiribati, New Caledonia, Samoa, Tonga and other islands.
Using syndromic surveillance to create alert signals for dengue outbreaks. Five studies investigated the value of syndromic surveillance for early outbreak detection. These included a comparison of community-based fever surveillance with surveillance of school absenteeism in Peru (study 9) and two studies in French Guiana (studies 10 and 12) that described the advantages of reporting dengue cases using a syndromic case definition compared with routine reporting. These two French Guiana studies and another in Madagascar (study 14) used sentinel sites and reported higher sensitivity and outbreak early warning capacity compared with the routine reporting systems (which were based on laboratory surveillance and passive case reporting). Studies 10 and 12 highlighted the need for maintaining the traditional surveillance and considering the increased potential for false alerts in syndromic surveillance systems.
The prospective study in Peru indicated that community door-to-door fever surveillance had higher sensitivity than school absenteeism records as an indicator for dengue (study 9); the community-based fever cohort captured twice as many cases as the school-based approach.
In French Guiana (study 10), the syndromic clinical surveillance in a military population and the routine laboratory reporting systems were found to be complementary: the syndromic approach detected an outbreak 3-4 weeks earlier and was six times more sensitive than laboratory-based surveillance, but the specificity was lower in the former. Further analysis (study 11) using CDC criteria (CDC 2001) showed that the ideal reporting time was often not achieved due to barriers at data entry and that an increased risk of false alerts needed to be considered. However, all respondents perceived that this system detected outbreaks adequately and subsequent countrywide introduction of sentinel-based syndromic reporting in French Guiana identified 80 signals for confirmed cases and 64 for clinical cases and predicted three major epidemics (study 12). In Madagascar, a sentinel-based syndromic surveillance system for six diseases was evaluated: it detected ten outbreaks, five were confirmed and two of which were dengue (study 14).
Use of other sentinel site-based approaches to increase capacity for outbreak detection. Three studies analysed sentinels sites for early outbreak warning, either in the form of sentinel-based reporting and virus surveillance (Cambodia, study 20) or for non-endemic countries (studies 22, 23, see group D below). One study described an enhanced routine surveillance system in Puerto Rico by motivating public health staff, which resulted in an increase in reported dengue incidence three times above the incidence during the two most recent epidemics in 1994 and 1998 (study 21). In Cambodia, passive surveillance plus sentinel site surveillance including virus surveillance increased the sensitivity of detecting outbreaks (defined as numbers of cases exceeding two standard deviations [SD] above the mean) although the response was delayed, mainly due to inadequate financial management (study 24).
In Europe, ten new strains of dengue viruses were detected in travellers returning from Africa, and increased observation of dengue in travellers by surveillance networks (TropNetEurop) was correlated with outbreaks documented in national data (study 23).

Surveillance for describing endemic⁄epidemic trends (Group B and C)
These surveillance systems under investigation were mostly population-based and passive. Some included additional sentinel sites or virus surveillance but they were used only to monitor viral trends and were not applied to early warning.
Four cohort-based studies calculated the level of underreporting, either using capture-recapture approaches comparing two independent surveillance systems or by comparing cohort-based data with the national routine reporting. The expansion factor indicating the level of underreporting was calculated to be: The results demonstrated remarkably high levels of underreporting in the surveillance systems, particularly for non-hospitalised cases. It was a common experience that a large proportion of the affected population was not captured by passive routine reporting (e.g. non-users of health services, users of private/traditional sectors or certain age groups (e.g. adults in Cambodia).
Four evaluations of routine dengue surveillance systems (studies 24-27) in 6 countries (Brazil, Bolivia, Cambodia, Indonesia, Maldives and Thailand) were conducted using a similar protocol for evaluations based on CDC Guidelines (CDC 2001). Both trend monitoring and outbreak detection were evaluated. All evaluations found that a clear understanding of the objectives of the surveillance system by all stakeholders was crucial. The routine reporting systemssome of them with laboratory supportwere perceived to be useful for trend monitoring and national planning but, as they did not apply appropriate thresholds/alert signals or include additional surveillance components, they had little capacity for early outbreak detection. In particular, reporting timeliness was perceived to be low, ranging from a few days for notification in the Maldives (study 27) to six to seven weeks until data analysis in Cambodia (study 24). Moreover, the responses were delayed, as shown in the Maldives, where no threshold for taking action was implemented, and in Cambodia, where lack of sufficient financial management and other constraints undermined any response to the alert signal of 'increased transmission (above two SD) in low transmission season'. In Thailand, where the system relied exclusively on clinically confirmed cases, respondents felt that outbreak responses were delayed because decision-makers did not trust the data and feared false alerts (study 24). All evaluations reported that timeliness could have been increased by electronic reporting or simplified reporting forms and that data analysis should have been performed at the lowest possible level (e.g. every district, once per week), given that sufficient capacity was available.

Dengue surveillance in low⁄non-endemic countries (Group D)
The value or effectiveness of primarily laboratory-supported active dengue surveillance systems in non-endemic settings was described in several studies from Asia and Europe. Timeliness of the system and laboratory support were reported to be crucial elements.
Three European studies described the recent detection of dengue in France (study 30) and in Croatia (study 29) and imported dengue cases in Italy (study 28). In Croatia, the notification of returning travellers led to the detection of autochthonous cases, while survey-based investigations revealed additional cases (29).
An evaluation of routine reporting in Korea (study 31) reported a 2-to 15-day delay from disease onset to reporting, which was shortened when electronic reporting components were introduced.
Four studies from Taiwan (studies 32-36) demonstrated the effectiveness of linking routine reporting with strong laboratory support and active and syndromic reporting elements in monitoring epidemiological, virological and clinical trends. Airport fever screening (studies 32 and 36) detected around 45% of imported dengue cases, but any impact this might have had on subsequent autochthonous transmission could not be determined.

Key findings
A greater number of the studies included in the present study (19/36) were performed in Asia than in the Americas (8/36; previously 17/24), illustrating a shift in research attention to Asia from the Americas since the 2008 review, when 17 and 6 studies, respectively, of 24 were recorded. In the present study, four studies (4/36) had a global focus, one study was from Africa and three studies were from Europe, most likely reflecting the global spread of, and consequent interest in, dengue disease in these regions in recent years.
Tools for trend monitoring (Group B and C), and as baseline for 'excess reporting' for outbreak detection The surveillance systems deployed for this purpose were mainly population-based and passive. Some included additional sentinel sites or virus surveillance, but in those cases, the data were used only to monitor viral trends and were not applied to early warning. Four cohort-based prospective studies calculated an expansion factor with a range between 1.1 and 2.6 for inpatients in Cambodia and Thailand, respectively, and between 3.9 and 29 in Cambodian, Nicaraguan and Thai cohorts for non-hospitalised cases.
The results demonstrate remarkable levels of underreporting in the surveillance systems, particularly for nonhospitalised cases. It was a common experience that a large proportion of the affected population was not captured by passive routine reporting (e.g. non-users of health services, users of private/traditional sectors or certain age groups, e.g. adults in Cambodia). However, while less than satisfactory, this does not mean that such a system is entirely inadequate, because as long as it is accurately reflecting the disease trend, it may still be used effectively as a baseline for detecting excess reporting (e.g. more than 2xSD above the mean of the previous lance systemhas potential for dengue outbreak prediction. Studies that evaluate sensitivity, specificity and positive predictive values of such a threshold are likely to be particularly valuable. Throughout the studies, reporting time was slow, and without any threshold, responses were delayed while poor financial management and lack of trust in the data by decision-makers hindered further the delivery of adequate and timely response measures. Despite that, all evaluations reported that timeliness could have been increased by electronic reporting or the use of simplified reporting forms and that data analysis should have been performed at the lowest possible level (e.g. once per week in every district) if sufficient capacity was available.
In summary, the country evaluations consistently highlighted that immediate improvement is possible using a number of options, many of which are already available and easily implementable: (i) simplified data forms/data entry protocols/electronic-based reporting, (ii) clearly defined and easily understood system objectives, (iii) appropriate and regular/frequent data analysis at the lowest possible level (iv) and regular data feedback from top to bottom levels. As evidence becomes available, two additional components will be required to complete the model: (i) clearly defined and locally appropriate triggers for an outbreak response (no studies were found exploring the optimal sensitivity and specificity of such thresholds) and (ii) implementation of evidence-based response strategies.

Alert signals 1 (triggers/indicators/thresholds) for epidemic response (Group A, C and D)
Predicting outbreaks through the introduction or shift of a dengue sero-/genotype: six studies (studies 3-8) investigated serotype changes as a dengue outbreak signal demonstrated a positive correlation with the number of reported cases or dengue incidence, although the lag times could extend up to 6 months. However, viral genome sequencing alone would, according to Schreiber (study 6), not have been sufficient to predict an outbreak.
But these events are highly site-specific and are influenced by herd immunity, population size, co-circulation of additional dengue viruses and potentially numerous other factors. Moreover, only those countries with reliable serotype-/genotype-specific surveillance would be able to monitor changes in any patterns. Genotypic shifts were used as an early warning signal in Singapore prior to the 2007 epidemic and initiated an early response (study 3). Taking into consideration the possibility that publication bias (i.e. that only positive results are likely to be published) would have excluded additional studies where serotype shifts were not associated with subsequent outbreaks and that numerous potential confounding factors would have been possible in all studies, it is not yet possible to draw any firm conclusions on the value of this as a measure in surveillance. Nonetheless, the sensitivity, specificity and positive predictive value of this parameter merit evaluation in prospective and comparative studies.
Predicting or detecting dengue outbreaks by syndromic surveillance data: Five studies investigated the value of syndromic surveillance for early outbreak detection. These included a comparison of community-based fever surveillance with school absenteeism in Peru (study 9), and two studies in French Guiana (studies 10 and 12) describing the advantages of reporting dengue cases using a syndromic case definition as compared to routine reporting. The prospective study in Peru indicated that community door-to-door fever surveillance had higher sensitivity than school absenteeism records. In French Guiana (study 10), the syndromic approach detected an outbreak 3-4 weeks earlier and was six times more sensitive than laboratory-based surveillance, but specificity was lower. However, in another study in French Guiana (11), the ideal reporting time of 60 min for a real-time syndromic surveillance approach was often not achieved due to barriers at data entry, while a risk of false alerts was expected, given the high sensitivity of the system. In Madagascar (14) and French Guiana (12), syndromic sentinel-based surveillance built on clinical syndromic case definitions showed promising results, increasing the sensitivity of dengue case detection in comparison with routine reporting and allowing the early detection of epidemic events.
Two studies investigated the value of data quantifying internet searches for dengue information carried out in a number of countries (Bolivia, Brazil, India, Indonesia and Singapore; studies 1 and 2). The curve of the search queries over time was similar to the epidemic curve constructed from surveillance data underlining the usefulness of this new and remarkably simple approach. Study 13 used a real-time electronic event-based approach based on Health Map to enhance timeliness, outbreak discovery and provide an added value for monitoring the ongoing spread of dengue.
A number of studies that were included in the earlier 2008 review also dealt with this topic; in summary, the following syndromic surveillance-based indicators were identified: • Proportion of virologically confirmed cases (study 3,

Rigau-P erez & Clark 2005)
• Malaria negative rate in fever patients in a malaria endemic areas (Carme et al. 2003, Talarmin et al. 2000 • Fever alerts (Pirard et al. 1997;Kourı et al. 1998) • Clinical syndromic case definitions (study 10, 11, 12 and 14) • School absenteeism (study 9) • Google search queries or event-based surveillance (Study 1, 2 and 13) Fever alert for the purpose of outbreak detection was not found to be useful in Cuba and Bolivia (Pirard et al. 1997;Kourı et al. 1998). None of the studies included in this update analysed syndromic surveillance based on laboratory parameters or the proportion of virologically confirmed cases. One study from Singapore (study 3) mentioned that during the 2007/2008 epidemic, the proportion of DENV-positive samples detected by PCR rose from 57.9% in January 2007 to 91.0% in July 2007 at the peak of transmission. A similar trend has been shown in Puerto Rico previously (Rigau-P erez & Clark 2005).
In summary, detection of increases in proportions of positive tested samples and quantification of electronic search queries are both promising approaches to dengue outbreak detection. They are inexpensive and offer near real-time data and their value for operational use should be considered and investigated. Syndromic surveillance based on a clinical case definition remains a complementary tool to national routine reporting.

Limitations
The main limitation of this review was its restriction to English, German and Spanish. However, as the bulk of literature accessible on electronic databases today is indexed in English by title and abstract, and no additional articles in other languages were found during the extensive search, the impact of a language bias is likely to be limited. While publication bias is a potential concern, by screening carefully the reference lists of assessed articles and grey literature, the bias has been reduced.
A 'research hot spot' in Singapore and Taiwan was identified: these two countries accounted for 10 of the total of 36 studies, potentially introducing some level of bias in the overall assessment of the published literature. Potential for bias also may have occurred with respect to the evidence demonstrating an association between newly introduced dengue serotypes and subsequent outbreaks (see below), because no studies reporting the absence of any association (i.e. new serotypes not followed by an increase in dengue; a phenomenon that is arguably, less likely to be published) were found.
Two key knowledge gaps were identified: none of the studies investigated whether the thresholds currently in use for triggering an outbreak response were at an appropriate level of sensitivity or geographical scale, and none indicated how outbreaks were distinguished from standard or 'expected' seasonal changes in transmission. Further research in this area remains of the highest priority and is strongly recommended.

Conclusions
Following the systematic review of the evidence of the value or potential of various tools or approaches for dengue outbreak prediction or trend monitoring, the following conclusions can be drawn: • Passive surveillance remains the backbone of disease monitoring, also providing the baseline for outbreak alert. All opportunities for improvement should be exploited to ensure that disease trends are accurately reflected. While underreporting could be tolerated to a certain extent, further research will be required to determine how much.
• The usefulness of the new dengue case classification for epidemiological use should be evaluated, as it is currently underway for its clinical use.
• Country evaluations of dengue surveillance systems should be conducted and published following CDC criteria.
• More research is necessary to identify appropriate thresholds of excess reporting that can be used to trigger an outbreak response; such studies must take into account both the geographical scale as well as the level of sensitivity.
• Appropriate additional alert signals need to be identified and tested and integrated risk assessment tools need to be developed.
• Additional well-designed and well-implemented enhancement tools (such as active surveillance components, laboratory support or motivation strategies) would strengthen surveillance.
• Shifts in dengue serotypes or genotype have considerable potential in dengue surveillance, and the value of these data merits evaluation in prospective and comparative studies. It is crucial that both negative and positive results be published to overcome publication bias in favour of positive associations.
• Syndromic surveillance approaches have potential as useful complementary tools offering increased timeliness and sensitivity but with an increased risk of false alerts. Further studies investigating laboratory parameters (e.g. the proportion of confirmed-torequested laboratory tests) are also merited. Internet searches or electronic event-based surveillance strategies also show promise, although their operational usefulness remains to be demonstrated.
• Further research on evidence-based response strategies and cost-effectiveness is still needed.