Effectiveness of early warning systems in the detection of infectious diseases outbreaks: a systematic review

Abstract

Background

Global pandemics have occurred with increasing frequency over the past decade reflecting the sub-optimum operationalization of surveillance systems handling human health data. Despite the wide array of current surveillance methods, their effectiveness varies with multiple factors. Here, we perform a systematic review of the effectiveness of alternative infectious diseases Early Warning Systems (EWSs) with a focus on the surveillance data collection methods, and taking into consideration feasibility in different settings.

Methods

We searched PubMed and Scopus databases on 21 October 2022. Articles were included if they covered the implementation of an early warning system and evaluated infectious diseases outbreaks that had potential to become pandemics. Of 1669 studies screened, 68 were included in the final sample. We performed quality assessment using an adapted CASP Checklist.

Results

Of the 68 articles included, 42 articles found EWSs successfully functioned independently as surveillance systems for pandemic-wide infectious diseases outbreaks, and 16 studies reported EWSs to have contributing surveillance features through complementary roles. Chief complaints from emergency departments’ data is an effective EWS but it requires standardized formats across hospitals. Centralized Public Health records-based EWSs facilitate information sharing; however, they rely on clinicians’ reporting of cases. Facilitated reporting by remote health settings and rapid alarm transmission are key advantages of Web-based EWSs. Pharmaceutical sales and laboratory results did not prove solo effectiveness. The EWS design combining surveillance data from both health records and staff was very successful. Also, daily surveillance data notification was the most successful and accepted enhancement strategy especially during mass gathering events. Eventually, in Low Middle Income Countries, working to improve and enhance existing systems was more critical than implementing new Syndromic Surveillance approaches.

Conclusions

Our study was able to evaluate the effectiveness of Early Warning Systems in different contexts and resource settings based on the EWSs’ method of data collection. There is consistent evidence that EWSs compiling pre-diagnosis data are more proactive to detect outbreaks. However, the fact that Syndromic Surveillance Systems (SSS) are more proactive than diagnostic disease surveillance should not be taken as an effective clue for outbreaks detection.

Background

The global pandemic of COVID-19 had a profound impact on the health of the public and economy around the world. Global pandemics have occurred with increasing frequency over the past decade, yet the world has missed opportunities to invest in preparedness and surveillance.

The impact of global infectious disease can be reduced by improving the international exchange of information, and developing monitoring and early warning systems [1]. The COVID-19 pandemic and its emerging variants have urged questioning the effectiveness of the Early Warning Systems (EWSs) for detecting infectious disease outbreaks. Although Dr. Li Wenliang, a Chinese ophthalmologist, issued an emergency warning of abnormal pneumonia cases in December 2019, many countries including the US did not respond optimally [2]. The Economist recently built a machine-learning model to estimate the number of excess deaths due to the pandemic for 223 countries [3]. The model estimates that the total number of deaths is 2–4 times higher than the number of confirmed deaths [3]. Estimates of excess deaths due to COVID-19 are from 20 to 25 Million globally [3]. Such losses are attributed to direct virus effect and indirect consequences on health systems’ overburdened capacity in both developed and developing countries [4].

The recent Monkeypox virus spread is a reminder of the need for diligent surveillance data monitoring for adequate containment of outbreaks and timely initiation of response measures, given the uncertainties and global fear of a repeated pandemic calamity that had the impact of COVID-19 [5]. Nevertheless, even with widespread evidence, investment in public health pandemic data systems within the health sector continues to be overlooked by most governments globally [4]. For instance, the director of the Statistics Division of the UN Department of Economic and Social Affairs (UN DESA), Stefan Schweinfest, asserted that the lack of data impede proper estimation of the impact of abnormal health events [4]. A recent study in China identified data source collection, integration and analysis as core components of an effective infectious disease EWS [6]. For example, databases from Emergency Departments (EDs), hospital and public health records, pharmacies, or even laboratories, are serving as approaches to infectious diseases surveillance which would generate alerts initiating public health investigations and response. However, despite the wide array of current surveillance methods, the effectiveness of these indicators and systems varies with multiple factors, including resource availability, the context of diseases or symptoms under surveillance, and other population health and social factors.

There is widespread agreement that global human health is vulnerable to existing or emerging infectious diseases due to the sub-optimum operationalization of surveillance systems handling health informatics data [4]. These health informatics systems rely on various mechanisms ranging from paper- or digital-based systems to gather population data and are not limited to technology only, as it is commonly miscomprehended [7]. Thus, EWSs enable early detection of the peaking of symptoms levels above-threshold before cases surge, or prompt recognition of small clustering of cases before prevailing illness overwhelms health systems.

However, there is no ideal effective surveillance system yet available. Proactive recognition of abnormal health events consistent with early epidemics remains challenging. Here we report a systematic review of the effectiveness of different EWSs’ designs in terms of utilized methods for data collection, taking into consideration feasibility in different resource settings.

To our knowledge, this systematic review is the first to examine surveillance systems in terms of evaluating their effectiveness. We bridge this gap by synthesizing the peer-reviewed literature on EWSs’ strategies and efficacy based on the methods of data collection. We accomplished this by conducting a systematic review and narrative synthesis of published studies on Syndromic and Sentinel Surveillance for infectious diseases, observing patients’ symptoms and confirmed diseases, respectively.

We define surveillance as “the ongoing systematic collection, analysis, and interpretation of health data essential to the planning, implementation, and evaluation of public health practice, closely integrated with the timely dissemination of these data to those who need to know” [8]. “Early Warning Systems, (EWS)” include data-based detection systems using health informatics data and approaches for infectious diseases surveillance.

Methods

Search strategy and selection criteria

We conducted a systematic review to identify relevant peer-reviewed articles regarding Early Warning Systems (EWSs) for detecting infectious disease outbreaks. PRISMA guidelines were followed in the reporting of the review [9]. Published articles were searched on the following electronic databases; PubMed and Scopus on October 21, 2022. The search included relevant keywords and word variants for early warning systems and infectious disease outbreaks. [see Additional file 1]. There was no language restriction at the primary databases’ search; however, we applied language restrictions to English in further steps of the PRISMA flowchart. Upon authors’ consensus (RM, DS, AM), the adopted approach allowed us to investigate the available literature in terms of the presence of a considerable number of publications on the topic of interest.

PubMed syntax (“Early warning system” OR EWS OR Alert OR “syndromic surveillance” OR SSS OR “syndromic surveillance system”) AND ((“infectious disease” OR “communicable diseases”) AND (outbreak)).

Scopus syntax ((TITLE-ABS-KEY (“Early warning system”) OR TITLE-ABS-KEY (ews) OR TITLE-ABS-KEY (alert) OR TITLE-ABS-KEY (“syndromic surveillance”) OR TITLE-ABS-KEY (sss) OR TITLE-ABS-KEY (“syndromic surveillance system”))) AND (((TITLE-ABS-KEY (“infectious disease”) OR TITLE-ABS-KEY (“communicable diseases”))) AND (TITLE-ABS-KEY (outbreak))).

Inclusion/exclusion criteria

To be included in the review, articles had to describe and evaluate an early warning system that is either already implemented or currently being implemented. This includes data-based detection systems (e.g. using data from various sources, such as death registries, social media trends, lab results, data on relevant over the counter (OTC) medication from pharmacies). The infectious disease outcomes were reviewed independently by two authors (RM and DS) to estimate the scope of the article, and include only those focusing on infectious diseases with the potential to become pandemics. For example, articles on facility outbreaks (e.g. such as Intensive Care Units) and school food poisoning were excluded.

We excluded studies that describe projections or proposed EWS which have not yet been evaluated. Articles on computational optimization exercises for outbreaks detection or focusing on non-human outcomes are also excluded. Additionally, commentaries, editorials, correspondences, systematic literature reviews, and preprint articles were also excluded. Eventually, we excluded out-of-date articles – this is because only 2 were detected before 2000, and they used out-of-date technology which is no longer relevant to the current generation of models [10, 11]. Reasons for exclusion are in Fig. 1.

Fig. 1

PRISMA flow diagram

Data extraction, analysis, and quality assessment

Abstracts and potentially relevant full texts were reviewed independently by two authors (RM and DS) with any conflicts resolved by consensus. The two reviewers screened titles and abstracts, removed duplicates (automatic and manual), and extracted data with the EndNote reference management software.

We assessed the quality of both retrospective and prospective studies for clarity of the research aim and methodology, data reproducibility, comparability, and outcome ascertainment using a 6-question Yes/No questionnaire based on an adapted CASP Checklist and that utilized in evaluating a review on internet-based EWS. Two reviewers (RM and AM) independently assessed the risk of bias and any uncertainty was resolved by contacting the third independent reviewer (DS). An additional file shows the utilized quality assessment tool and the quality assessment results in more detail (see Additional file 2).

We identified the following items for extraction: the author’s name, article’s title, EWS name, country of study, research question, study design, surveilled diseases or manifestations, system’s strengths, weaknesses, recommendations, and effectiveness, as well as comparing systems and study limitations, then used Excel spreadsheets to document all key elements from the included manuscripts.

In some cases, not all of the findings of a study will be relevant to our scope; in such cases, we have included only the relevant findings. Lastly, we extracted a summary of the authors’ interpretations/conclusions. The review was registered with PROSPERO (CRD42021278123) on 18 November 2021 and is reported according to PRISMA guidelines.

Results

More than half (n = 44) of the 68 articles included in the final selection were assessing Early Warning Systems (ESWs) in high-income countries (HICs), including in the US (n = 11), Netherlands (n = 6), UK (n = 5), Australia (n = 3), Germany (n = 3), Canada (n = 3), France (n = 2), Italy (n = 2), Taiwan (n = 2), Japan (n = 2), Austria (n = 1), New Zealand (n = 1), Norway (n = 1), Singapore (n = 1), Spain (n = 1), Korea (n = 1). Finally, an additional study, not included in the above calculation, was evaluating EWSs in more than one HICs; including France, Germany, UK, and Spain. On the other hand, 12 studies covered EWSs in middle-income countries (MICs), including China (n = 7), India (n = 2), Brazil (n = 1), Federated States of Micronesia (n = 1) and the Republic of Macedonia (n = 1). While 10 studies were performed in low-income countries (LICs); Ghana (n = 2), Yemen (n = 2), Darfur (n-1), Madagascar (n = 1), Samoa (n = 1), Sierra Leone (n = 1), Iraq (n = 1) in addition to Pacific island countries and territories (PICTs) (n = 1).

All the studies were quantitative, either retrospective (n = 47) or prospective (n = 15). Three articles assessed the EWS under study both retrospectively and prospectively, while two used the study design of predictive simulation in addition to one cross-sectional study [12]. Table 1 provides a description of included studies with a summary of their key findings. n* is the number of articles evaluating EWSs in different categories; given that some systems were assessed in two articles; NHS Direct syndromic surveillance (n = 1 effective + 1 not effective) - ProMED-mail (n = 1 effective + 1 limited effective) - National electronic Disease Early Warning System (eDEWS) (n = 1 effective + 1 complement) (1 telephone triaging and 1 records)- Electronic Surveillance System for the Early Notification of Community-based Epidemics (ESSENCE) (n = 1 effective and 1 not effective but useful) - China Infectious Disease Automated-alert and Response System (CIDARS) (n = 2 effective).

Table 1 Data Extraction

Fifty studies reported the effectiveness of Syndromic Surveillance Systems (SSS) and the remaining 18 focused on EWSs monitoring diagnostic diseases. Forty-two articles found EWSs successfully functioned independently as surveillance systems for pandemic-wide infectious diseases outbreaks, 16 showed complementary roles thus having contributing surveillance features but cannot be relied upon solely, and 3 studies demonstrated EWS ineffectiveness. The EWS’s evaluation results of the remaining 7 articles are as follows; adjunct (n = 2), partial (n = 1), potential (n = 1), limited (n = 1), not effective but useful (n = 1), in addition to one effective EWS but with less capacity [66].

We summarize evidence on the effectiveness of EWS in the detection of infectious diseases outbreaks according to the source of data collection, comprised of 7 categories: emergency care and triage-based (n = 20), hospital/public health records (n = 13), web/internet-related (n = 11), healthcare workers-based (n = 13), pharmaceuticals sales (n = 2), and laboratory results (n = 1). Furthermore, EWSs involving combinations of former designs (n = 4) and those related to enhancing the existing traditional surveillance (n = 4) were classified as a separate category.

Emergency care and triage-based EWSs (n = 20)

Emergency care and triage-based early warning systems gather patients’ complaints at the first contact point with the healthcare system. This EWS category includes syndromic data from Emergency departments (n = 11), Telephone triaging (n = 7), and Ambulance dispatches (n = 2).

Emergency departments - ED-EWS (n = 11)

All ED-EWS studies were in high-income countries (HICs) and distributed as the following; among all the studies 3 were from Australia (New South Wales NSW), [14, 16, 18] and the rest was divided into 3,2,1,1,1 for the United States [17, 19, 23], Italy [13, 21], UK [15], Taiwan [20], and Korea [22], respectively. The majority demonstrated overall good capacity as an EWS where performance evaluation revealed the independent effectiveness of 5 systems (US [17], Italy [13], UK [15], Taiwan [20], Korea [22]) and the complementary usefulness of 3 (US [19], Italy [21], Australia [18]) in addition to the supplementary effectiveness of one US EWS [23]. Only 2 Australian ED-EWSs were deemed either of mere potential benefit or ineffective overall [14, 16].

The ED-EWS competency is attributed to the ability of public health institutions to rapidly respond to cases identified by emergency departments [13, 17]. For instance, the New York City Department of Health and Mental Hygiene syndromic surveillance system (DOHMH) managed to credibly prompt public health actions related to the rise in meningitis-related ED visits, [17] and the Italian ED-SSS successfully predicted the first measles case by 2 months despite the low virus circulation documented by other traditional systems [13]. Additionally, the UK ED-SSS highlighted the privileges of hospital emergency department data in terms of being representative and inclusive to severe cases and non-residents (open access EDs), respectively [15]. This facilitates monitoring the spectrum of common pathogens, augmenting community-based surveillance besides introducing novel clinical indicators to the chief complaint - such as data on discharge status, investigations, and treatment [15].

However, the 3 studies concluded that ED effectiveness as a complement to traditional surveillance, [18, 19, 21] was due to the effects of human factors and defects in categorizing syndromes [18]. For example, Muscatello et al. pointed out that data accuracy of the Australian system was influenced by the knowledge of ED staff for medical coding and their ability to concentrate during busy shifts, especially when manifestations are documented after the visit or event or not recorded at all [18]. Particularly, the Australian information system allowed entry of only one diagnosis code per patient which might reflect a limited part of the presenting syndrome [18].

Interestingly, the only non-effective ED-SSS among the included articles was designed to only track a single symptom rather than a syndromic group of symptoms [14]. An example of this is that the passive surveillance through lab reporting of pertussis cases was 7 days more proactive than surveilling ED visits with cough [14]. On the other hand, the only included ED-EWS of adjunct effectiveness was in a recent study surveilling COVID-like symptoms rather than influenza-like illnesses [23].

The five effective ED-EWSs all noted that improving data quality is key for ED-based surveillance. Hope et al. emphasized the variation among hospitals’ data coding is an important challenge to ED-SSS [16]. Moreover, accurate diagnosis codes enabled the Korean ED-based EWS to predict influenza cases 2 weeks in advance via monitoring fever as a the chief complaint [22]. Accordingly, Joseph Wu et al. recommended the adoption of the error-detection function to enhance the efficiency and completeness of data entry, [20] given that the reliance on search-based text strings does not take into consideration misspellings, abbreviations, and synonyms [13]. Therefore, methodological approaches should target word analysis and ensure a standardized format for data extraction, and consequently, overcome the major challenge of the chief complaint syndrome surveillance model [15, 17].

Nevertheless, standardized data entry methods might have unforeseen drawbacks. Terry et al. mentioned that standardization of ED chief complaints will not be broadly applicable, given the idiosyncrasies of hospitals in entering chief complaints [19]. For instance, Westchester County’s SSS, which relies on less than three complaints to generate a signal, carries the risk of misclassification of text terms into syndrome categories which would trigger false alerts [19].

Telephone triaging (n = 7)

Only one of the seven included articles on telephone triaging-based EWS was conducted in a low-income country (Yemen) [27]. Evaluations demonstrated the effectiveness of this designated EWS category; however, the three UK National Health Service telephone helpline (NHS24) studies showed mixed results, [25, 26, 30] and the Canadian Telehealth Ontario toll-free helpline revealed a restricted role as a complement that could not be relied upon solely [24].

Assessments demonstrated telephone triaging effectiveness and strengths in terms of simplicity, health staff acceptability, and national representativeness [26, 30]. For instance, Caudle et al. declared that the lack of necessity to aggregate data from unlinked sources is an advantage for Telehealth over ED-based EWSs [24]. The Japanese telephone triaging successfully detected seasonal influenza and pediatric rotavirus outbreaks by surveillance of fever and diarrhea, respectively [28, 29].

Notably, the timeliness of the telephone triaging EWSs varies with resource availability. For example, in England and Wales, the NHS24 was capable of observing rises in syndromes within 12–36 hours from receiving calls and had successfully detected ILI outbreaks and abnormal rises in vomiting and diarrhea [26]. Additionally, Kavanagh et al. asserted the timeliness of the Scotland NHS24 in comparison to media outlets [30]. However, in low-resource settings, the timely response of phone-based surveillance systems is a major shortcoming [27]. For instance, the National Electronic Disease Early Warning System (eDEWS) in Yemen had a 2.85-day lag between the first reported case of Cholera and the initial public health response, and the duration to inform health authorities and responses’ timeliness vary from region to region [27].

Despite the above benefits, telephone triaging-based EWS has several system limitations. For example, not only is the NHS helpline likely to overlook small localized outbreaks, but also during routine surveillance the telephone calls could be a barrier to accurate data and tracking thus raising false positive alerts and triggering unnecessary public health responses [30]. Cooper et al. also revealed that the NHS helpline calls about diarrhea failed to detect a historical Cryptosporidiosis outbreak [25]. Moreover, during pandemic spread, the NHS24 alarm system becomes less sensitive, and its role becomes limited to tracking temporal changes [30]. Cooper et al. expanded that the NHS24’s full capacity will not be reached unless there is a huge surge in call rates [25]. Eventually, in the case of pediatric infectious diseases such as Rotavirus, objective reporting of symptoms by children’s parents represents a bias [28, 29].

Ambulance dispatches (n = 2)

Two studies in HICs reported that ambulance dispatches-based surveillance systems are sensitive with few false alerts [31, 32]. For instance, the independently effective US surveillance system identified the expected annual influenza epidemics in simulated serial daily analyses from 1994 to 1998 and prospectively from 1999 to 2002 [31]. While the Spanish System for Information on Detection and Analysis of Risks and Threats to Health (SIDARTHa) cannot be relied upon solely (complement) and successfully indicated the onset of high influenza activity 1 week before and at the same time as the sentinel system, in 2010–11 and 2011–12 influenza seasons, respectively [32].

Notably, surveillance systems based on ambulance service run sheets necessitate the availability of timely population-wide electronic data that are routinely collected and liable to categorization into syndromes [31, 32]. Without real-time data, the designated systems would reach an undesirable sensitivity where the missing number of influenza isolates for a particular day will not be associated with a rise in the alarm threshold of that day [31]. Moreover, the Spanish system for Information on Detection and Analysis of Risks and Threats to Health (SIDARTHa) was able to utilize the call logs and ED patient records at the same time from the medical dispatch center [32]. This shows the complementary action and the potential of use for more than one EWS [32].

Hospital/public health records (n = 13)

Hospitals and public health registries are indispensable for monitoring infectious diseases. The majority of the included EWSs in this category proved independent effectiveness (n = 10) apart from two complementary and one partially effective system in LICs.

Hospitals and health facilities records (n = 7)

Seven of the studies evaluated hospitals and health facilities-related EWSs covering electronic or paper-based results of the hospital admissions (n = 1) and the inpatient data (n = 6).

Admission-based EWSs are preferable to be implemented in states with “discrete population centers” and run by staff “aware of hospital admission patterns” to increase the likelihood of unusual event recognition [35]. Although admission-based systems do not capture outpatient illness-related outbreaks, they are advantageous in identifying and provoking investigations for unusual syndromes and those limited to one case [35]. For instance, the US Hospital Admissions Syndromic Surveillance statewide syndromic surveillance (HASS) proactively recognized the rare condition of West Nile virus encephalitis and one-case outbreaks such as smallpox or SARS [35]. Nevertheless, Dembek et al. stated that the inpatient SSS has an average lag of 1–2 days in comparison to the outpatient one given the time lapse from manifestations onset and admission [35]. Additionally, reliance on case counts overlooks potentially useful demographic data [35].

The articles on inpatient EWSs adopted either an ICD-diagnosis code-based system or Symptom-Clicking-Module (SCM) for automatic grouping of symptoms through “pre-defined syndrome definitions” [33, 34, 38].

Although inpatient EWSs investigate data on patients with different risk factors and conditions, a German study demonstrated the potential system’s in-season ineffectiveness since cases with comorbidities tend to have longer hospital stays with delayed inpatient data collection, in addition to the absence of laboratory confirmation results [34]. However, inpatient systems could permit manual exploration of time series figures of hospitals’ daily surveillance data by local epidemiologists [33]. Moreover, Ang. et al., evaluating the ICD-9-based Singaporean Patient Care Enhancement System (PACES), recommended that “repeat consults” would abolish the inherent background noise of the primary care consult-based system and help detect outbreaks of sizes undetectable by community-based systems [38].

In resource-limited areas, such case definition-based EWSs are associated with implementation challenges. For example, the challenging contextual situation in Yemen impedes timely reporting and alert verifications where “only 21% of all the National Electronic Disease Early Warning System (eDEWS)‘s alerts were verified within the first 24 h of detection in 2016” [36]. Accordingly, the local health system fragility along with poor “understanding of case definitions” contributed to having quality and timeliness as major eDEWS’ drawbacks [36]. Also, the Indian indicator-based surveillance (IBS) and event-based surveillance (EBS) systems encounter challenges in terms of gathering electronically-delivered data from reporting units, absence of previous baseline data, and incomplete indicator-based surveillance data capturing (admitted patients in the central hospital) [37]. Furthermore, the tendency for data-burden reduction in deprived settings is another hurdle in LMICs. For instance, the simplicity of the Pacific island countries and territories (PICTs) through reliance on a definite number of easily assessed syndromes entailed that cases will not be notified unless they meet the included case definitions and that a rise in symptoms will not generate alerts [39].

Public health records (n = 6)

Public health (PH) records from public health offices reflect epidemiological data ranging from the incidence of different notifiable infectious diseases to all-cause mortalities. PH records-based EWSs were assessed in six studies; among the total number of studies four were in LMICs [41,42,43,44] and two in HICs [40, 45]. All demonstrated independent effectiveness except the Ghana PH system of partial competency [42].

Public health records-based EWSs successfully rely on centralized systems facilitating information sharing across local public health departments [41, 43, 44]. For instance, the Chinese Infectious Disease Automated-alert and Response System (CIDARS) collects data from the existing electronic National Notifiable Infectious Diseases Reporting Information System (NIDRIS) and reports to CDCs via short message service (SMS) [43]. Zhang et al. asserted CIDARS’s high sensitivity and specificity with a 3-day median detection time has shown effectiveness in detecting Dengue Fever (DF) outbreaks [44]. Moreover, a prototype EWS in Brazil tracking epidemiological viral circulation data during the 2014 World Cup had effectively “predicted high risk of dengue for 57% of the microregions reporting high levels of dengue during the games” and provided warning 3 months in advance [41]. Also, through real-time data analysis from existing Provincial Electronic Health Record data, the Alberta Real Time Syndromic Surveillance Net (ARTSSN) in Canada overcame the traditional system fragmentation, taking the advantage of “timeliness, comprehensiveness, and automation [40]. Furthermore, Ghana PH-EWS utilizes the National Influenza Center (NIC) routine ILI data where it detected circulating influenza A and B lineages with an “average of 10 days between symptom onset and detection“ [42]. Eventually, the French death certificates-based EWS is an “all-cause mortality surveillance system” compiling data “from computerized city halls” [45].

However, in PH records-based EWSs, outbreaks would not be detected until clinicians reported cases [43]. Yang et al. demonstrated CIDARS’s potential less timeliness and sensitivity in comparison to EWS based “on data on pre-diagnosis of cases in hospitals, media reports or school absenteeism” [43]. In addition to the previous routinely accepted delays, a French study by Baghdadi et al. showed that there is an additional hurdling delay (90% of mortality within 7 days) in the web-based death certification by physicians in the French system “Reactive mortality surveillance system- syndromic surveillance system SurSaUD” [45].

Web and internet-based EWSs (n = 11)

Web-based early warning systems (n = 4)

Four studies evaluated Web-based EWS using computer software; three studies in LMICs (China [47], the Republic of Macedonia [48], India [49]) and a single one in HICs (Norway [46]). The designated systems monitored an array of syndromic groups except China’s infectious disease automated alert and response system which focused on Hand, foot, and mouth disease (HFM) [47]. Assessments demonstrated independently the effectiveness of Web-based EWSs except for the complementary role of ALERT in the Republic of Macedonia [46,47,48].

Web-based Early Warning Systems enhance communication across surveillance networks from health facilities at local levels to higher public health authorities [46, 48]. For example, Li et al., evaluating the Chinese web-based alert and response system, revealed its sensitivity especially with larger outbreaks (> 20 cases) than smaller ones (< 10 cases) with an average detection time of 1.7 and 2.1 days, respectively, and a 4.5 days-lag until reporting to public health authorities [47]. Additionally, a “digital disease surveillance system”, relying on cell phone applications, allowed the Indian health authorities to monitor mass gathering surveillance data through “near-real-time daily reports” from local health staff and epidemiologists [49].

Notably, facilitated reporting by remote health settings and rapid alarm transmission are key advantages of Web-based EWSs. For instance, the web-based ALERT system of Macedonia, relying on primary care facilities’ data, had successfully detected the onset of seasonal influenza and was more proactive than the routine diagnosis-based surveillance [48]. Moreover, Guzman-Herrador et al. revealed that in the Norway web-based system, Vesuv, efficient information exchange allowed the update of outbreak data, and had easy log-in by various stakeholders, besides sending automatic reminders to notifiers within 3 weeks [46].

Internet-based early warning systems (n = 7)

Seven studies evaluated internet-based systems; Email-based systems (n = 2) in Darfur [50] and Netherlands [51], and Social trends-related EWSs (n = 3) in the US [54], Europe [56], and Canada [55] which revealed independent effectiveness. In addition, 2 studies reported contradicting results regarding the effectiveness of the US Electronic Surveillance System for the Early Notification of Community-based Epidemics (ESSENCE) [52, 53]. A study additional to the previous seven studies- was categorized under pharmaceutical-based EWS category- was published by Dong et al. stated that the Chinese Baidu search queries–for the “fever” term–showed a strong correlation with influenza activity, and along with Sentinel hospital ILI and over the counter (OTC) drug sales could complement the routine surveillance based on lab-confirmed cases [70].

The covered email-based systems –which rely on emails as a means of communication between the EWS staff– compile surveillance data from various resources; media and official reports, online sources, and local observers, which are later transmitted to national and federal levels [50]. Studies revealed that email-based systems have unique considerations regarding data quality and the need for key informants. For instance, the email system in Darfur’s refugee camps demonstrated the inherent risk of inaccurate data documentation and imprecise information about the camps’ population [50]. Moreover, Zeldenrust et al., evaluating the Netherlands ProMED-mail (Program for Monitoring Emerging Diseases), stated that the system specificity is low as its information sources are liable to bias and some lack scientific terminologies [51].

Likewise, studies on social trends-related EWSs –tracing infectious diseases symptoms and signs shared by users of social platforms– raised concerns over unethical and biased data collection [54,55,56]. For instance, data storage and utilization of Google Flu Trends is done without consent, jeopardizing users’ privacy issues [54]. Samaras et al. emphasized that internet data cannot be 100% accurate which is subjected to underestimate or exaggeration depending on human reaction to situations [56]. Furthermore, educational level along with cultural and language backgrounds affect the accuracy of symptoms shared by internet users [54]. For instance, Dong et al. stated that the Chinese Baidu search queries for the “fever” term are affected by health-seeking behavior such as fear or curiosity that would affect the public search queries [70]. Nevertheless, according to Betancourt et al., the US Electronic Surveillance System for the Early Notification of Community-Based Epidemics (ESSENCE)–an internet-based system utilizing ICD-9–is effective given its data completeness, accuracy, and high specificity (95.5–96%) [53].

Moreover, representativeness is a huge hurdle to social trends-based EWSs. For example, data sampling and computational approximation methods influence the data accuracy of Google Flu Trends [54]. Dion et al., evaluating the Canadian Global Public Health Intelligence Network (GPHIN), also asserted the representativeness limitation of social trends-related EWSs since social media platforms are not accessible to everyone, as well as social tweets’ word limit might prevent the inclusion of potentially useful contextual information, respectively [55].

Despite the above-mentioned concerns, monitoring social trends has successfully functioned as an early warning system in HICs with speedy data processing as a key advantage. For instance, the real-time operation on a 24/7 basis gave GPHIN a remarkable velocity where data retrieval and processing occurred every 15 and less than 1 minute, respectively [55]. The US Google Flu Trends was 1–2 weeks more proactive than CDC reports in estimating levels of influenza [54]. Also, the European Google Flu Trends in France, Germany, the UK, and Spain proactively detected measles outbreaks within 2 months [56]. However, as for Witkop et al., ESSENCE has a low positive predictive value (31.8%), triggering “time-consuming false alarms”, and failing to detect influenza outbreaks early enough with a time lag of 1–3 days [52].

Human resources-based (n = 13)

Thirteen articles evaluated a unique category of early warning system that relies on human resources to proactively detect outbreaks. The involved cadres range from general practitioners (n = 5) [58, 59, 62, 67], community health workers (n = 2) [60, 63], volunteers (n = 3) [57, 61] to school staff (n = 3) [64,65,66]. Unlike other EWS categories, human resources-based systems are implemented more in LMICs (Ghana [60], Sierra Leone [63], Madagascar [62], Iraq [68] and China [66]), and even in one of the HICs (Austria) [57] the system monitors outbreaks among former residents of developing countries. None of the ESWs in this category were ineffective; whereas, effectiveness varied from independent functioning (n = 6) [12, 57, 59,60,61, 63] to complementary effectiveness (n = 6) [58, 62, 64, 65, 67, 68], in addition to one showing less capacity than the artificial intelligence-based form [66].

General practitioner (n = 5)

GPs participation in EWSs ranged from voluntary participation from different provinces or recruited from defined clinics, to Out-of-Hours services and house calls network [58, 59, 62, 67]. The included studies were implemented in HICs (France [58], New Zealand [59], and the Netherlands [67]) except the one in Madagascar [62] and that in Iraq [68]. All had complementary functions (n = 4) except the independently effective one in New Zealand (n = 1) [59]. GP-based EWSs surveil a broader range of symptoms in comparison to other EWS categories [58, 59, 62, 67, 68]. The reason is that man-powered surveillance systems offer resilience and flexibility with data acquisition [58, 59, 62, 67].

Assessments of the general practitioner-based EWS underscored the real-time nature of this category of surveillance systems which has several pros both from the user and public health perspectives. From the physicians’ perspective, Flamand et al. stated acceptability of the French SOS Medecins by GPs as their role was limited to routine data gathering, without additional work requirements [58]. Furthermore, an evaluation of the Netherlands ICARES (Integrated Crisis Alert and Response System) asserted that the system does not impose a work burden [67]. Jones N. F. and Marshall R., evaluating the New Zealand EWS, reported the increasing compliance of the participating GPs [59]. Eventually, compliance of the participating Madagascar system’s GPs was evidenced by the 89% estimated rate of daily data transfer [62].

Regarding the public health perspective, the Netherlands ICARES generated alerts within 24 hours from data entry by the treating practitioner [67]. Also, Randriana et al. mentioned that the Madagascar EWS successfully detected rises in fever manifestations related to influenza, arbovirosis, and malaria [62]. Furthermore, deploying SSS assisted Iraqi health staff to recognize rises in influenza-like illness by combining surveillance data on “fever and cough symptoms” during mass gathering events [68]. Eventually, the French EWS showed high sensitivity and being human-operated, offering flexibility to changes in syndromes under surveillance, along with access to illness data that are outside of hospital admissions [58].

Additionally, human factors-related considerations might impede the GP-based EWSs’ full capacity. Firstly, the Madagascar Sentinel syndromic-based surveillance system highlighted that GP-based EWSs are likely to be associated with relative tardiness in eliciting public health responses, which might outweigh the advantage of rapid data gathering [62]. Accordingly, Jones N. F. and Marshall R. pointed out the importance of proper planning of resource allocation in terms of GP time to ensure system maintenance [59]. Nevertheless, the Madagascar experience was encouraging for developing countries; however, ensuring robust communication systems with competent internet connections is the main challenge [62]. Secondly, bias arising from GPs’ recording behavior is another human factors-related hurdle for GP-based EWSs.

Community health workers (n = 2)

Community-based surveillance in the included studies encompasses recruited community health monitors and surveillance supervisors where reporting followed a bottom-up approach from districts to public health authorities [60, 63]. Two studies demonstrated the independent effectiveness of community-based surveillance (CBS) in LMICs [60, 63]. For instance, the modified CBS in Ghana detected over 300 events that would otherwise go undetected [60]. Also, the new Ghana EWS identified 26% of all suspected vaccine-preventable disease cases reported by routine surveillance [60]. Additionally, the Community Event-Based Surveillance (CEBS) in Sierra Leone demonstrated rapid cases detection in comparison to the national system and it had successfully triggered alarms to one-third of Ebola cases identified [63].

The authors highlighted obstacles that are unique to Community-based Surveillance. The CBS broader-scale implementation is challenging given the required extensive training for community members and maintaining communication with the surveillance team [63]. Moreover, like the GP-based EWSs, Community-based Surveillance is liable to biases from staff miscategorization that would lead to inaccurate capturing of cases [63]. Furthermore, monitoring of death events (late indicator) by community health workers could be challenged by the practice of unsafe burials such as in Sierra Leone CEBS [63]. Accordingly, Ratnayake et al. recommended boosting community engagement, adjusting trigger definitions, and full involvement in the overall Integrated Disease Surveillance and Response system [63].

Others (n = 6)

Reliance on volunteers and school staff requires additional involvement of human resources for the early detection of infectious disease outbreaks. Three effective manpower-based EWSs recruited focal persons at asylum seekers’ reservation centers, and volunteers at Hurricane Katrina and Hurricane Harvey evacuees’ shelters in Austria and the US, respectively [12, 57, 61]. For instance, an acute gastroenteritis outbreak and tracking a rise in respiratory symptoms were successfully confirmed by the American volunteers-based EWS [61]. Lastly, two complementary effective School-Based Syndromic Surveillance System (SID-SSS) in Taiwan and UK gathered surveillance data “from school nurses or class teachers (for those preschools without nurses)” and school absenteeism reports, respectively [64, 65]. However, Yang et al. highlighted that utilization of information technology to record school absenteeism has better “simplicity, cost-effectiveness, data quality, sensitivity, and timeliness” metrics than manual approaches by school staff (completeness of 100% versus 86.7%, respectively) [66].

This EWS category surveilled an array of infectious diseases’ prodromal manifestations that are likely to develop in crowded settings [12, 57, 61, 64]. Gathering pre-defined data sets from confined populations allowed near real-time data collection with timely public health response [12, 57, 61]. For instance, the US Emerging Disease Syndromic Surveillance’s cot surveys took less than 60 seconds (to 5.2 minutes) per person, which enabled daily assessment of evacuees [12, 61]. Also, El-Khatib et al. pointed out the practicability of tally sheets of the Syndrome-Based Surveillance System (SbSS) for Infectious Diseases among Asylum Seekers which monitor easily recognizable manifestations of infectious diseases [57]. The Austrian system is a reliable one with high sensitivity and had the advantage of rapid implementation in emergencies [57]. Moreover, surveilling school children in Taiwan–for students’ absenteeism and family health status besides clinical and epidemiological data–addressed gaps within ED-SSS since the former tracked mild cases at pre-diagnosis levels while the latter collected data on severe illnesses from all age groups throughout the year [64].

Nevertheless, low specificity and false-negative reports are potential drawbacks of the designated EWS category [61]. For example, the SbSS for Infectious Diseases among Asylum Seekers did not have consistent availability of staff at reservation centers nor accurate daily registration of refugees, and if asylum seekers were admitted to hospitals immediately, underreporting and under-detection of syndromes are potential challenges [57]. Additionally, Murray et al. disclosed that evacuees – to avert isolation placement – were likely to hide illnesses [61]. Moreover, Weng et al. revealed that school breaks and reporting variations are common limitations for SID-SSS [64]. Although the school-based influenza-like illness (ILI) surveillance in China was 1 week more proactive than lab confirmation, Dong et al. asserted that inability to track relevant data during holidays and weekends leads to inaccurate assessment of “community-level influenza activity during these periods” [70]. Eventually, specificity of school-based syndromic surveillance system could be enhanced by including virological surveillance of representative samples [64].

Pharmaceuticals sales (n = 2) and laboratory results (n = 1)

Monitoring over-the-counter (OTC) medications sales and diagnostic laboratory-based testing represent surveillance means where early warning systems trace patients’ data from paramedical sources. None of the 3 EWSs included in the designated category proved their independent functionality [69,70,71].

Laboratories compliance and clinical correlation are the main challenges to lab-based EWSs. For example, the Netherlands’ Infectious diseases Surveillance Information System (ISIS) had low coverage where only 18 out of the 85 included labs were connected to the central medical microbiology laboratories (MML) [71]. Thus, the collected positive and negative microbiological results were not routinely gathered into the laboratory information management systems in a representative manner for the Netherlands [71]. Moreover, laboratory influenza specimens do not reflect the total viral activity [70]. For instance, respiratory pathogens such as a respiratory syncytial virus (RSV) have similar manifestations to influenza, thus would weakly correlate with lab ILI positive results [70]. Nevertheless, discontinuation of the above ineffective systems is not recommended where lab EWSs provide up-to-date trends of micro-organisms incidence and their suspension would lead to loss of important epidemiological data [71].

On the other hand, the inability to trace back cases represents a major pitfall in relying on drug prescription sales for surveillance. For example, the unavailability of information of the OTC purchasers limited the effectiveness of the drug sales EWS in Tianjin as a complement to existing influenza lab surveillance [69]. Nevertheless, despite the New York City OTC-EWS data collection on prescription sales and information on medical visits besides OTC drugs, Das et al.’s also mentioned that the designated New York system acted as an adjunct for surveillance where gastrointestinal drug sales were less sensitive than ED diarrheal visits where its effectiveness is limited to monitoring patients’ level [69].

Additionally, consumer behavior and the evolving OTC market present unique challenges to pharmaceutical sales-EWSs. For instance, if the public is stockpiling OTC medicines, this would mask the real consumption from acute illness [69]. Moreover, new medications’ market entry and the variety of drug formulations represent challenges for categorizing syndromes of the OTC sales-based EWS [69].

Multi-data source (n = 4)

The multi-source EWSs in the included articles rely essentially on data records– for syndromic surveillance of infectious disease outbreaks–either by combining different kinds of health registries or those with various data-gathering methodologies. Three of these EWSs were conducted in the Netherlands; two retrospectively revealed successful combined effectiveness, and one showed a complementary role by Predictive simulation [72,73,74]. The fourth study demonstrated the complement effect of the Chinese multi-source EWS by retrospective assessment [75].

Leveraging Records and Staff was one of the successful EWS combined designs. For example, Wijngaard et al. reported that the Netherlands EWS used a combined design, hospital data from the National Medical Register, and ILI data from a sentinel network of GPs, facilitating a timely detection of localized outbreaks - especially those related to emerging pathogens, independently of lab surveillance [74]. Wijngaard et al. also pointed out that detection proactivity of different data sources–especially for respiratory infectious diseases surveillance–is as follows; “hospital data (+1 week), pharmacy purchases/GP consultations (+2 weeks), and deaths/lab diagnostic requests (+3 weeks)” [73]. However, the latter associations should be further validated, especially those related to “absenteeism and pharmacy data” [73]. Additionally, rural China (ISSC project); utilized manual labor from health facilities, pharmacies, and primary schools for data entry from paper forms into automated electronic data capture, enabling timely identification of epidemics with recognition of changing disease patterns [75].

Combining public health reports and email services is another successful multi-data source EWSs. In the Netherlands, monitoring the daily ECDC Round Table Report and ProMED-mail was associated with the timely reporting of 95% of threats [72]. Bijkerk et al. also highlighted that this combined EWS approach would “save at least 2.5 hours a week on human resources” [72].

Enhanced surveillance (n = 4)

The four included articles evaluating Enhanced Surveillance Systems revealed the effectiveness of their temporary adaptations for mass gathering events [76,77,78,79]. The systems’ enhancements targeted the frequency of epidemiological data transmission with 2 studies agreeing that the daily surveillance data notification was the most successful and accepted enhancement strategy [76, 79]. For instance, the Germany-based Enhanced Surveillance Systems during the 2006 FIFA World Cup had adaptations in terms of days instead of weekly transmission of notifiable diseases, marking World Cup-related cases in SurvNet – the routine surveillance system, additional daily reports from the hosting cities besides the daily analysis and reporting by the National Enhanced Surveillance Operations Centre (NESOC) [79]. Williams et al. confirmed that the latter modifications had strengthened the timeliness and detection capacity without the substantial need of additional resources and that it is unnecessary to implement syndromic surveillance in mass events unless the routine reporting system is not robust [79].

Another form of temporary enhancements, World Cup German Enhanced System, included accelerating data transmission to the existing electronic surveillance system through immediate telephone contacts, allowing free-text notifications even if outside the included case definitions, media monitoring for relevant events, and boosting communication between national and international stakeholders [76]. Schenkel et al. highlighted that the latter system adaptation sensitized the routine mandatory notification system where it has successfully detected a Norovirus outbreak related to the World Cup [76]. Additionally, 1 day was the average data transmission time to federal authorities, instead of three [76].

In LMICs, evaluating the benefits of enhancing existing systems over syndromic surveillance is controversial, given that sustaining daily data transmission is challenging in low-resource settings [76]. For instance, full utilization of a web-based tool of Suite for Automated Global Electronic bioSurveillance Open ESSENCE (SAGES OE) for data storage in the Federated States of Micronesia and Samoa, within the context of the 8th Micronesian Games and the third United Nations Conference on Small Island Developing States (SIDS), retrospectively, was challenged by connectivity issues and lack of enough computers and trained staff [77, 78]. Furthermore, a spreadsheet-based alternative was used in SIDS due to SAGES OE technical limitations [77]. Nevertheless, authors highlighted that the designated enhanced surveillance helped in decision-making and acted as an assurance system for health security with the potential of full integration into routine surveillance as it boosted communication channels between clinical, laboratory, and public health departments [77, 78].

Discussion

To our knowledge this review is the first to synthesize evidence on the evaluated effectiveness of Early Warning Systems (EWSs) for proactive detection of infectious diseases outbreaks. The majority of the studies were in HICs and those included from low and middle-income countries were almost equal. Most of the studies evaluated Syndromic Surveillance Systems (SSS) and a few focused on EWSs monitoring diagnostic diseases, assessing systems with an array of methods for surveillance data collection. Emergency departments (ED) and triaging data in addition to those from hospitals and public health records formed the main bulk of data sources for the included EWSs. Furthermore, data from the following surveillance methods were included: web/internet- and healthcare workers-based EWSs. Additionally, a few studies covered pharmaceutical sales, laboratory results, multi-design EWSs, and enhanced traditional surveillance systems.

There is consistent evidence that Inpatient SSS has an average lag of 1–2 days in comparison to the outpatient one, given the time lapse from manifestations onset and admission [35]. Moreover, in Singapore Ang. et al. assessment demonstrated that the retrospective study was not able to evaluate the staff sick leave’s impact on surveillance systems relying on PACES (in-patient) [38].

Most studies that included Public Health records-based EWSs highlighted that outbreaks would not be detected until clinicians reported cases [43]. For instance, Yang et al. demonstrated (PH-EWS) CIDARS’s potential less timeliness and sensitivity in comparison to EWS based “on data on pre-diagnosis of cases in hospitals, media reports or school absenteeism” [43]. Moreover, Kavanagh et al. asserted the timeliness of the Scotland NHS24 (telephone triaging) in comparison to media outlets [30].

Studies on General Practitioner-EWS underscored the real-time nature of this category. The few studies that looked at school surveillance pointed out that students’ absenteeism and family health status, besides clinical and epidemiological data, addressed gaps within ED-SSS since the former tracked mild cases at pre-diagnosis levels while the latter collected data on severe illnesses from all age groups throughout the year [64]. There was also evidence that school surveillance was 1 week more proactive than lab confirmation [70]. However, for example, passive surveillance through lab reporting of pertussis cases was 7 days more proactive than surveilling ED visits with cough [14]. Additionally, Das et al. mentioned that the US OTC-EWS merely acts as an adjunct for surveillance; for instance, gastrointestinal drug sales were less sensitive than ED diarrheal visits [69].

With multi-source EWSs, Wijngaard et al. stated that detection proactivity of different data sources–especially for respiratory infectious diseases surveillance–is as follows; “hospital data (+1 week), pharmacy purchases/GP consultations (+2 weeks), and deaths/lab diagnostic requests (+3 weeks)” [73]. However, the latter associations should be further validated, especially those related to “absenteeism and pharmacy data” [73]. The few studies on death-based EWSs agreed that delayed reporting is a major drawback, such as in the web-based death certification by physicians in the French (90% of mortality within 7 days) [45]. Also, web-based EWS primary care facilities are not the optimum notification source [48]. It is noteworthy that in LMICs, evaluating the benefits of enhancing existing systems over syndromic surveillance is controversial, given that sustaining daily data transmission is challenging in low-resource settings [76]. For example, Sentinel hospital ILI, OTC drug sales, Baidu search query, and school-based ILI surveillance are a complement to traditional surveillance systems in China [70].

There is also consistent evidence that the competency of ED-EWSs is attributed to the ability of public health to rapidly respond to cases, besides ED’s data representative and inclusion of severe cases and non-residents [15]. Any recognized pitfalls were related to human factors and defects in categorizing syndromes. To sum up, using emergency department data on chief complaints and presenting symptoms is an effective EWS [19]. Nevertheless, it requires staff training on medical coding with wise utilization of standardized formats among hospitals’ EDs [18].

On the other hand, the effectiveness and strengths of telephone triaging systems were in terms of simplicity, health staff acceptability, and national representativeness [26, 30]. Notably, the timeliness of the telephone triaging EWSs varies with resource availability. However, there was evidence that telephone triaging-EWSs has several system limitations such as overlooking small localized outbreaks and less sensitivity during pandemics, besides objective reporting [28, 29].

Ultimately, evidence has demonstrated the sensitivity of ambulance dispatches EWSs with few false alerts [31, 32]. Recommendations necessitated the availability of timely population-wide electronic data that are routinely collected and liable to categorization into syndromes as crucial for ambulance dispatches EWSs [31, 32].

Evidence has revealed that admission-based EWSs are preferable to be implemented in states with “discrete population centers” and run by staff “aware of hospital admission patterns.” [35] Although they do not capture outpatient illness-related outbreaks, they are advantageous in identifying and provoking investigations for unusual syndromes and those limited to one case [35]. On the other hand, the evidence demonstrated that inpatient EWSs–relying on an ICD-diagnosis code-based system or Symptom-Clicking-Module (SCM) for automatic grouping of symptoms through “pre-defined syndrome definitions”–had in-season ineffectiveness [33, 34, 38]. However, inpatient systems could permit manual exploration of time series figures of hospitals’ daily surveillance data by local epidemiologists, with recommendations that “repeat consults” would abolish the inherent background noise of the primary care consult-based system [38]. Nevertheless, in resource-limited areas, such case definition-based EWSs are associated with implementation challenges.

The key advantage of Public Health records-based EWSs is successfully relying on centralized systems, facilitating information sharing across local public health departments [41, 43, 44]. However, in this case, outbreaks would not be detected until clinicians reported on their cases [43].

Web and internet-based Early Warning Systems carry huge potential for outbreak detection. Web-based EWSs enhance communication across surveillance networks from health facilities at local levels to higher public health authorities [46, 48]. Evidence revealed that facilitated reporting by remote health settings and rapid alarm transmission are key advantages of Web-based EWSs. On the other hand, internet-based EWSs include Emails and Social trends-related systems. Email-based systems compile surveillance data from various resources: media and official reports, online sources, and local observers that are later transmitted to national and federal levels. However, they have unique considerations regarding data quality and the need for key informants. Social trends-related EWSs raised concerns over representativeness and unethical biased data collection [54,55,56]. Despite the above-mentioned concerns, monitoring social trends has successfully functioned as an early warning system in HICs with speedy data processing as a key advantage.

Our analysis demonstrated that human resources-based systems are implemented more in LMICs where none of the included EWSs in this category were ineffective. The included cadres ranging from general practitioners, community health workers, and volunteers, to school nurses. First, general practitioners ranged from voluntary participation from different provinces or recruited from defined clinics to Out-of-Hours services and house calls network. The designated system monitored a broader range of symptoms in comparison to other EWS categories given that man-powered surveillance systems offer resilience and flexibility with data acquisition [58, 59, 62, 67]. Assessments of the general practitioner-based EWS underscored the real-time nature of this category of surveillance systems, which has several pros both from user and public health perspectives. However, human factors-related considerations might impede the GP-based EWSs’ full capacity, relative tardiness in eliciting public health responses, and biases arising from GPs’ recording behavior.

The community health workers’ (CHWs) include recruited community health monitors and surveillance supervisors. Obstacles of relying on CHWs encompass extensive training for community members, maintaining communication with the surveillance team, and staff syndromes’ miscategorization [63]. On the other hand, reliance on volunteers and school staff enables gathering pre-defined data sets from the confined population, and surveilling an array of infectious diseases’ prodromal manifestations that are likely to develop in crowded settings [57, 61, 64]. However, low specificity and false-negative reports are potential drawbacks [61].

The included systems for pharmaceutical sales and laboratory results did not prove solo effectiveness. Laboratories’ compliance and clinical correlation are the main challenges to lab-based EWSs; laboratories’ low coverage and results may not reflect the total viral activity [70]. Nevertheless, discontinuation of the above ineffective systems is not recommended [71]. Furthermore, the inability to trace back cases represents a major pitfall of relying on drug prescription sales for surveillance. Additionally, consumer behavior and evolving OTC market embody unique challenges to pharmaceutical sales-EWSs [69].

Evidence has also revealed that multi-data source EWSs and enhancing traditional systems are promising for early outbreak recognition. Leveraging Records and Staff was one of the successful EWS combined designs. Also, enhanced surveillance temporary adaptations showed effectiveness during mass gathering events. The systems’ enhancements targeted the frequency of epidemiological data transmission with two studies agreeing that the daily surveillance data notification was the most successful and accepted enhancement strategy [76, 79]. However, in LMICs, evaluating the benefits of enhancing existing systems over syndromic surveillance is controversial, given that sustaining daily data transmission is challenging in low-resource settings [76]. Authors highlighted that the designated enhanced surveillance helped in decision-making and acted as an assurance system for health security with the potential of full integration into routine surveillance [77, 78].

Based on our extensive analysis of both syndromic and diagnosis-based EWSs, we suggest the following combinations as appropriate enhancement strategies for infectious diseases surveillance. In HICs, we recommend that investment be directed towards the spectrum of syndromic surveillance systems –tracking pre-diagnosis data- ranging from social trends, emergency care, and triage-based EWSs, to hospitals and health facilities records. This is attributable to the abundance of facilitations and financial resources in HICs that would enable the adoption of standardized speedy monitoring of human health data. On the contrary, LMICs -given their manpower availability- are advised to allocate investment towards the existing diagnosis-based EWSs such as centralized Public Health records, web, and email-related surveillance systems in addition to human resources-based EWSs.

There is scope to explore the One-health approach, including environmental and veterinary surveillance systems besides human-based ones. Our recommendation is aligned with a recent systematic review on EWSs for “vector-borne diseases” where Baharom et al. highlighted that incorporating meteorological and environmental surveillance systems with other “epidemiological tools” is a very promising strategy for outbreaks detection [80].

Also, the fact that Syndromic Surveillance Systems (SSS) are more proactive than diagnostic disease surveillance should not be taken as an effective indicator/clue for outbreaks detection. Given that in literature, LMICs and their vulnerable highly populated areas were under-represented in comparison to high-income ones, there is a need to investigate more the implementation contextual feasibility of different EWSs’ categories in low resource settings. For instance, studies revealed that in deprived areas staff training is integral for the implementation of electronic automated surveillance.

A key strength of our study lies in the fact that we captured all human-based EWSs of pandemic scope regardless of the surveilled symptoms or ailments. Additionally, by including both SSS and diagnostic disease surveillance systems, we were able to synthesize evidence on the level of effectiveness of EWSs’ source of data collection considering the broad spectrum of mild and severe clinical presentations. However, despite it being broad in scope, the review still had some weaknesses and limitations.

First, regarding the quality of the included articles, a few studies did not have specific aims; their focus was on making general observations of the efficacy of an intervention. For instance, some did not detect outbreaks but authors reported it to be effective. Also, some of the studies included did not make comparisons with the existing mode of surveillance.

Second, we applied language restrictions to English. However, we have quite well EWSs assessed from all over the world apart from areas that usually do not implement EWSs nor publish relevant credible sources.

Third, we did not include gray literature or pre-print publications. Nevertheless, the 61 included articles covered the broad spectrum of EWSs methods we aimed for which fitted the scope of our study in terms of effectiveness for detecting potential pandemic-wide outbreaks.

Fourth, we limited our search to PubMed and Scopus databases. However, the databases of choice are comprehensive where most of the well-recognized articles are indexed. Furthermore, the yielded results from PubMed and Scopus were reassuring to be a representative sample for the available literature, as each EWS category was analyzed in a considerable number of studies, and nearly a consensus was found per each category of studies in terms of pros, cons and the recommended course of action for improvement. Nevertheless, we recommend the inclusion of additional databases for further research.

Fifth, we did not include studies focusing only on facility settings outbreaks or outbreaks on a small scale. This might have added insights into potential unforeseen warning systems for proactive public health interventions. However, we were reassured that our search strategy did not restrict the inclusion of infectious diseases under surveillance, reflecting the different data collection methodologies that were captured.

Finally, the operationalization of evaluating EWS effectiveness varied across the studies. For example, the surveillance coverage of some studies was on pre-defined groups such as refugees and schools. Although the majority were of wide scale, population-based, comments on EWSs’ specificity, sensitivity, completeness as well as subjectivity, and transparency of reporting were not included in all studies. Moreover, monitored symptoms, case definitions, and EWS alert thresholds varied. This heterogeneity in assessment parameters makes meaningful comparison difficult.

The EWSs’ assessment results are plausible with rational justifications of effectiveness for each data collection method. However, as most of the current outbreaks with potential to become pandemics are due to zoonotic diseases, there is a need for more studies on the One-health approach, including environmental and veterinary systems with human data. Such studies could explore more proactive interdisciplinary public health strategies, particularly for respiratory transmitted diseases.

Only a few studies reported on enhancing surveillance of traditional systems that were all in mass gathering events. Therefore, making conclusions from our findings should be done with caution, particularly within the context of all-year surveillance systems. There is therefore the need for exploring the potential of enhancing surveillance systems throughout the year and specifically in LMICs.

Conclusion

Our study was able to evaluate the effectiveness of Early Warning Systems (EWSs) in different contexts and resource settings based on the EWSs’ method of data collection. There is consistent evidence that EWSs compiling pre-diagnosis data are more proactive to detect outbreaks. Emergency departments (EDs) data on chief complaints is an effective EWS but it requires utilization of standardized formats among hospitals’ EDs. Simplicity, health staff acceptability, and national representativeness are strengths of telephone triaging systems. Eventually, inpatient systems could permit manual exploration of hospitals’ daily surveillance data by local epidemiologists.

Centralized Public Health records-based EWSs facilitate information sharing; however, they rely on clinicians’ reporting of cases. Interestingly, human resources-based systems are implemented more in LMICs where none of the included EWSs in this category were reported ineffective.

Facilitated reporting by remote health settings and rapid alarm transmission are key advantages of Web-based EWSs. Email-based systems have unique considerations regarding data quality and the need for key informants. And, social trends-related EWSs successfully functioned but raised concerns over representativeness and unethical biased data collection.

Notably, pharmaceutical sales and laboratory results did not prove solo effectiveness. The EWS design combining surveillance data from both health records and staff was very successful. And, daily surveillance data notification was the most successful and accepted enhancement strategy especially during mass gathering events.

The fact that Syndromic Surveillance Systems (SSS) are more proactive than diagnostic disease surveillance should not be taken as an effective clue for outbreaks detection. Although HICs are recommended to focus investment on EWSs tracking pre-diagnosis data, in Low Middle Income Countries (LMICs), working to improve and enhance existing systems was more critical than implementing new Syndromic Surveillance approaches.

Leveraging the full capacity of Early Warning Systems to proactively detect infectious disease outbreaks is imperative more than ever before. Sources for collecting surveillance data are ample in most countries and each encompasses under-utilized pros and unforeseen cons. Nevertheless, with an exploration of EWSs functionality in different contexts and resource settings, policymakers and public health authorities should take tailored actionable steps to monitor human surveillance data and intervene at optimum times.

Many surveillance systems, whether syndromic or diagnostic disease-based, are in place in different countries globally that are run throughout the year or for mass gatherings. However, both high and LIMCs should have active surveillance to detect abnormal infectious events and utilize their primary methodology for real-time surveying pathogens. Such preparedness appears significantly urgent with the unprecedented pandemic era of current and emerging public health threats.