A Portrait of Sentinel Surveillance Networks for Vector-Borne Diseases: A Scoping Review Supporting Sentinel Network Design

Abstract

Vector-borne diseases (VBDs) are continuing to emerge globally, requiring new surveillance systems to follow increasing VBD risk for human populations. Sentinel surveillance is an approach that allows tracking of disease risk through time using limited resources. However, there is no consensus on how best to design a sentinel surveillance network in the context of VBDs. We conducted a scoping review to compare VBD sentinel surveillance systems worldwide with the aim of identifying key design features associated with effective networks. Overall, VBD surveillance networks were used most commonly for malaria, West Nile virus, and lymphatic filariasis. A total of 45 criteria for the selection of sentinel unit location were identified. Risk-based criteria were the most often used, and logistic regression showed that using risk-based criteria dependent on host animals is particularly correlated with surveillance system sensitivity (p < 0.018). We identify tools that could prove valuable for sentinel surveillance network design, including a standardized approach for evaluating surveillance systems and a tool to prioritize criteria for selecting optimal geographic locations for spatial sentinel units.

Introduction

In the last few decades, we have witnessed an expansion in the geographic distribution of many arthropod vectors of human disease, such as mosquitoes, biting flies, and ticks. Driving forces in vector range expansion include anthropogenic factors, such as urbanization and globalization, and climate change, which result in new environments becoming suitable for the survival of vectors (Maynard et al. 2017, Ogden 2017, Kraemer et al. 2019). With expanding vector ranges, the emergence of vector-borne diseases (VBDs), diseases transmitted by hematophagous arthropods, is becoming a growing public health problem. VBDs now account for an estimated 17% of the worldwide infectious disease burden and cause more than 700,000 deaths annually (Semenza and Suk 2018, Barker and Reisen 2019).

In many cases, VBD emergence requires public health authorities to develop new approaches to track VBD risk within their administrative boundaries. Public health surveillance is defined as the continuous and systematic collection of data to guide public health practice and interventions (WHOa 2020). In the surveillance of VBDs, additional complexities must be considered—their distribution in space is dependent on vector ecology and reservoir host ecology, and human exposure to vectors. This creates specific challenges for VBD surveillance: (1) as vector habitat range increases, public health authorities must survey an increasing geographic area with finite resources, (2) the presence of vectors and pathogens in the environment is often the target of surveillance since it is prerequisite to the emergence of human cases, and (3) the spread of vectors and pathogens in space can be difficult to predict due to heterogeneity of habitat suitability and patchy introduction of the vector associated with stochastic events or host movement patterns.

Sentinel surveillance offers the opportunity to overcome some of these ecological and logistical challenges. Sentinel surveillance involves the selection of a subgroup of the population to be measured repeatedly to provide a time series of surveillance data. As it uses a targeted sample, costs are limited. Despite a restricted sample size, sentinels have been shown to permit the surveillance of the presence or absence of the vector and/or the pathogen in previous research, for instance as seen in sentinel chicken programs for the prompt detection of West Nile virus circulation, to name a single example (Yukich et al. 2014, Bahuon et al. 2016, Petrić et al. 2016).

Although sentinel surveillance offers the possibility of establishing a cost-effective surveillance system for VBDs, its major limitation is the spatial interpolation of surveillance findings outside the sentinel units (Girond et al. 2017). Thus, to optimize the use of data obtained from the surveillance system, sentinel units must be carefully selected according to the aims of the surveillance program, taking into consideration the type of sentinel, its geographical location, sampling design, and laboratory methods used. Sentinel units, which form individual parts of the sentinel surveillance network, refer to the geographical unit where a sentinel (hospital, clinic, laboratory, sentinel surveillance site) or a group of sentinel individuals (animals, mosquito traps) are located. In this article, we use the term “sentinel unit location” to designate the selected geographic location of a sentinel unit. Animals (individuals or herds) are frequently used as sentinel units (Halliday et al. 2007), and a review of the use of sentinel herds for the surveillance of VBDs showed that sentinel unit location was a key factor determining the efficacy of the surveillance network (Racloz et al. 2006).

Evaluation of surveillance systems is crucial to assess the efficacy and functionality of the network (U.S. Department of Health and Human Services 2006). With this information, public health authorities can evaluate the benefits and challenges of using certain surveillance structure before establishing a new network. The Centers for Disease Control and Prevention (CDC) has built a framework for surveillance network evaluation based on nine key characteristics (hereafter referred to as performance parameters). These performance parameters represent characteristics that a network should have to favor efficient and effective public health surveillance: simplicity, flexibility, data quality, acceptability, sensitivity, predictive value positive, representativeness, timeliness, and stability (German et al. 2001). Sentinel unit location will influence many of these parameters. Selecting appropriate locations for the surveillance units has the potential to favor the detection of disease occurrence (sensitivity), at an early stage in disease emergence (timeliness), and in a manner to capture the risk to human populations (representativeness).

To our knowledge, sentinel surveillance networks for VBDs have not been globally inventoried, as reviews have focused on specific types of sentinel units that for example, sentinel herds (Racloz et al. 2006), or on VBDs in specific settings for example, urban environment (Eder et al. 2018). Thus, a comprehensive review assessing the structure and performance of VBD sentinel surveillance networks worldwide could provide valuable insight into which criteria to consider when designing new surveillance networks of this kind. To respond to this knowledge gap, we carried out a scoping review to characterize the main features of past and existing surveillance networks. By extracting selection criteria involved in determining the spatial distribution of sentinel units, we were able to further describe the rationale underlying the spatial design of these networks. Lastly, by correlating the use of given selection criteria with the performance of the sentinel surveillance network, we aimed to examine whether certain criteria were more likely to result in the implementation of successful surveillance networks.

Methods

Search strategy

A systematic search was conducted in Ovid MEDLINE, Embase, CABAbstracts and Global Health and in the gray literature up to November 29, 2019. The search strategy, including searching terms related to sentinel surveillance and VBDs, terms for VBDs of major public health concern as determined by the WHO (WHOb), are detailed in the Supplementary Data. The search strategy was first validated by a public health librarian from the Public Health Association of Canada (PHAC). The second validation involved using a snowball strategy (manually screening references of relevant articles) and by crosschecking for the presence of 10 references provided by an expert in the field. Any references not found during these two validation steps and which met our inclusion criteria were added to the list.

Study selection

Titles and abstracts drawn from the literature search were screened for relevance independently by two reviewers, to determine whether the articles met each one the following inclusion criteria: (1) surveillance of a VBD, including vector, animal, human, or environmental surveillance; (2) involved a network of at least two sentinel units, whereby a unit is an entity in a predetermined, fixed location; and (3) written in English, French, or Spanish. Articles which did not contain primary data were excluded. Any disagreements were settled by reaching consensus after further discussion. During the screening step, the information was evaluated for relevance by using a standardized tool implemented in DistillerSR (Evidence Partners, ON). The screening tool was validated by pretesting 10 preselected scientific articles by three individual reviewers with agreements ≥85%. General study characteristics were also extracted, such as type of study and year of publication (Supplementary Data).

Data extraction: description of sentinel surveillance networks for VBDs

Articles which met the inclusion criteria in the relevance screening were brought forward to the data extraction step, consisting of a second standardized data extraction (DE) tool (Supplementary Data). The DE tool was validated by three individual reviewers with agreements ≥85% on closed-ended questions from three preselected articles using the tool.

The DE tool contained three parts. The first involved extracting criteria, which were used in the selection of the sentinel unit location. Articles which did not report this information were excluded. In the second part, general characteristics of the sentinel network where clarified, such as VBD investigated, vectors responsible for the disease, methods used for data collection, and type of data collected. Lastly, a global evaluation of the sentinel network was completed, following CDC framework, to gain an overall impression of the value of the surveillance network.

Evaluation of sentinel surveillance networks and their construction

Based on CDC guidelines, proxies for each performance parameters were developed, and associated questions were included in the data extraction form (Supplementary Data). This allowed reviewers to evaluate the performance parameters of the surveillance in a categorical fashion (i.e., whether or not the sentinel network met the performance parameter proxies, or if it was unknown from reading the article).

The next step was to examine whether certain sentinel unit selection criteria influenced the success of the surveillance system. As aforementioned, sentinel location was deemed to directly impact three of the surveillance network performance parameters: sensitivity, representativeness, and timeliness. However, timeliness was excluded from this analysis because in most articles it was not possible to evaluate whether the surveillance network had met this performance parameter.

Thus, surveillance network sensitivity and representativeness were independently inserted as response variables in logistic regressions using the criteria used for sentinel unit selection as fixed effects. These models were run using the glm function within the R environment (R Core Team 2020, Venables and Ripley 2002). Articles whose sensitivity or representativeness could not be evaluated by reviewers were excluded from these models. Fixed effects included only selection criteria that were used in ≥10 articles, which resulted in the exclusion of 28 criteria from the models. Collinearity between variables was excluded using Variation Inflation Factor as calculated with the vif function (Fox and Weisberg 2019) in R. An Akaike Information Criterion (AIC) backward stepwise approach was used to identify the best fitting model. Finally, the fit of the models was evaluated using standard regression diagnostics methods, including Pearson's and deviance residuals, Cook's distances and Pearson's chi-squared test. No use of human or animal subjects, no need for ethics approval within the methods.

Results

Search outcome

Our search strategy yielded a total of 7309 articles, from the 4 databases and the gray literature searched. A total of 4674 duplicates were removed, and 3 articles were added after crosschecking references during the search strategy validation steps. A total of 2638 articles progressed to the title and abstract screening step.

During title and abstract screening, 1006 articles were not based on surveillance of VBD, and 861 did not focus on sentinel networks and were thus excluded. Seventy other articles were excluded as they were in a language other than English, French, or Spanish (47 Chinese; 3 Croatian; 3 German; 3 Italian; 1 Korean; 10 Portuguese; 2 Serbian; and 1 Turkish) and 64 articles were excluded as they did not present primary data. One article was both written in another language and did not contain primary data. This resulted in a total of 638 articles which progressed to the data extraction step (Fig. 1).

FIG. 1.

Search outcome from a scoping review aimed at investigating sentinel surveillance networks for vector-borne diseases, performed between September and November 2019. The search outcome is reported based on PRISMA guidelines. GGL, governmental gray literature; NGGL, nongovernmental gray literature.

Description of included articles

A total of 246 articles were eliminated either because they did not mention selection criteria used for sentinel site location (n = 186), or because the full text could not be found (n = 60). Thus, of the articles (n = 638), which passed the title and article screening step, 206 (32.2%) passed the full data extraction step. From here on, only the articles that were retained through the data extraction step will be described.

The 206 articles retained were carried out in Africa (n = 88, 42.7%), Asia (n = 32, 15.5%), North America (n = 27, 13.1%), Western Europe (n = 18, 8.7%), Australia (n = 14, 6.8%), Central or South America (n = 13, 6.3%), Oceania (n = 8, 3.9%), and Eastern Europe (n = 7, 3.4%). One study had sites both in Australia and Oceania. They presented results from sentinel networks operating at local (n = 33, 16.0%), regional (n = 73, 35.4%), national (n = 94, 45.6%), or less commonly multinational (n = 6, 2.9%) scale.

The principal VBDs monitored by sentinel networks included malaria (n = 68, 33.0%), West Nile virus (n = 32, 15.5%), lymphatic filariasis (n = 22, 10.7%), and schistosomiasis (n = 19, 9.2%) (Table 1). A total of 24 different viruses, 9 parasites and 3 bacteria were surveying across the articles.

Table 1.

List of Vector-Borne Diseases Investigated Within the Articles Included in the Scoping Review, Including Type of Arthropod Vector, Type of Pathogen, and Number of Articles Which Studied Each of These Diseases

Disease	Vector	Pathogen	No. of articles
Malaria	Mosquitoes	Parasite	68
West Nile virus infection	Mosquitoes	Virus	32
Lymphatic filariasis	Mosquitoes	Parasite	22
Schistosomiasis	Snails	Parasite	19
Western equine encephalitis	Mosquitoes	Virus	15
Bluetongue	Midges	Virus	14
Murray Valley encephalitis	Mosquitoes	Virus	11
Onchocerciasis	Black flies	Parasite	11
Japanese encephalitis	Ticks	Virus	10
Ross River virus	Mosquitoes	Virus	9
Arbovirus infection	Mosquitoes	Virus	7
Chikungunya	Mosquitoes	Virus	5
Zika	Mosquitoes	Virus	5
Barmah Forest virus infection	Mosquitoes	Virus	4
Yellow fever	Mosquitoes	Virus	4
Lyme disease	Ticks	Bacteria	4
Venezuelan equine encephalitis	Mosquitoes	Virus	3
Epizootic hemorrhagic disease	Midges	Virus	3
Arboviruses group A and B infection	Mosquitoes	Virus	2
Eastern equine encephalitis	Mosquitoes	Virus	2
Rift Valley fever	Mosquitoes	Virus	2
Usutu virus infection	Mosquitoes	Virus	2
Leishmaniasis	Phlebotomine sand flies	Parasite	2
Q Fever	Ticks	Bacteria	2
Saint Louis encephalitis	Ticks	Virus	2
Bovine trypanosomiasis	Tsetse flies	Parasite	2
Chaga's disease	Triatomine bugs	Parasite	1
Edge Hill virus infection	Mosquitoes	Virus	1
Jamestown Canyon virus infection	Mosquitoes	Virus	1
Bovine ephemeral fever	Midges	Virus	1
Schmallenberg virus infection	Midges	Virus	1
Bartonella infection	Ticks	Bacteria	1
Crimean–Congo fever	Ticks	Virus	1
Powassan virus infection	Ticks	Virus	1
Tick-borne diseases	Ticks	NA	1
African trypanosomiasis	Tsetse flies	Parasite	1

NA, not applicable.

Sentinel surveillance networks in the articles were largely active (n = 181, 87.9%) and less commonly passive (n = 14, 6.8%). Some articles (n = 11, 5.3%), used both active and passive surveillance within the sentinel network.

Description of the sentinel networks

Surveillance networks described in the articles operated for <1 to >20 years and comprised between 2 and >50 sentinel units (Table 2).

Table 2.

Number of Sentinel Surveillance Units in Networks Described in Articles Considered in the Scoping Review, According to the Duration of the Surveillance Network Operation

No. of sites	Duration of network operation (years)							Total
No. of sites	<1	1–2	3–5	6–10	11–20	>20	Unknown	Total
2	7	3	0	0	0	0	0	10
3–5	15	15	10	1	1	0	1	43
6–10	20	22	13	4	4	0	0	63
11–20	7	11	8	1	1	0	0	28
21–50	9	9	8	3	0	1	0	30
>50	3	3	2	3	1	8	0	20
Unknown	3	2	3	2	1	1	0	12
Total	64	65	43	14	8	10	1	206

The sentinel units in the networks were villages (n = 46, 22.3%), clinics (n = 42, 20.4%), sites in an urban setting (n = 1, 8.3%), sites in the countryside (n = 16, 7.8%), farms (n = 11, 5.3%), schools (n = 9, 4.4%), sites in the suburbs (n = 7, 3.4%), sites in the forest (n = 4, 1.9%), health zones (n = 4, 1.9%), houses (n = 2, 1.0%), and laboratories (n = 1, 0.5%), zoos (n = 1, 0.5%), or kennels (n = 1, 0.5%). However, 68 articles (33.0%) did not explicitly state or describe their sentinel site settings.

When animals were used within the network, and placed within the sentinel units, the most common animals used were chickens (n = 32, 15.5%), bovine (n = 22, 10.7%), wild birds (n = 10, 4.9%), sheep (n = 7, 3.4%), dogs (n = 6, 2.9%), goat (n = 4, 1.9%), horse (n = 4, 1.9%), hamsters (n = 3, 1.5%), mice (n = 3, 1.5%), rodents (n = 3), livestock (n = 2, 1.0%), pigs (n = 2, 1.0%), deer (n = 1, 0.5%), donkeys (n = 1, 0.5%), and zoo animals (n = 1, 0.5%). A total of 132 articles (64.1%) did not use any animals as sentinels.

Aim of articles on sentinel networks

The broad aims of the articles are included following disease trends (n = 128, 62.1%), testing intervention methods (n = 72, 35.0%), profiling risk factors (n = 32, 15.0%), acting as an Early Warning System (EWS) (n = 15, 7.3%), and evaluating the surveillance network (n = 10, 4.9%).

Disease detection methods used

A total of 49 different sampling and laboratory methods were identified in the articles (Table 3) allowing the collection of various types of data (Fig. 2).

FIG. 2.

Data reported to be collected through the sentinel networks described in the articles retained in our scoping review. Articles could have more than one type of data collected. This included prevalence of pathogen/disease in humans (human prevalence), vector densities, prevalence of pathogen/disease in animals (animal prevalence), prevalence of pathogen in vectors (vector prevalence), intervention or treatment efficiency (intervention efficacy), demographic data (demographics), mortality/mortality data for humans (mortality/morbidity), mortality data for vectors (vector mortality), vector biting rate, and geographical data.

Table 3.

Data Collection and Laboratory Analysis Methods Used for Disease Detection in Sentinel Surveillance Systems Described in the Articles Retained in Our Scoping Review

Method	No. of articles
Mosquito trapping (any kind)	73
Blood sampling (human)	68
Symptom surveillance or questionnaire	60
Blood sampling (animal)	59
ELISA	51
PCR (vector)	42
Microscopy (human tissues)	40
Blood smear (thick and thin blood films)	36
Physical examination	27
PCR (human tissues)	24
Parasitology (excluding Plasmodium spp.)	21
PCR (animal tissue)	20
Sequencing	19
Microscopy (vectors)	16
Bioassay	15
SNT	15
RDTs	14
Stool sampling (human)	14
Human case reporting	12
Larval survey	12
PRNT90	12
Hemagglutination test	11
Human landing catch	11
ICT card	11
Immunoassay	11
Classic isotopic 48-h test	9
Kato Katz methods	8
Serology (unspecified)	8
Snail survey	8
Urine sampling	8
Biopsy	8
AGID	6
CSF sampling	5
Microscopy (animal tissues)	5
Stool sampling (animal)	5
Ophthalmic examination	4
CBC	3
Vector removal	3
Biochemistry	2
Rodent capture	2
Satellite imaging	2
FTS	1
Isotope	1
Drag sampling	1
HOLA	1
RFLP	1
Spot test	1
Ultrasonography	1

AGID, agar gel immunodiffusion; CBC, complete blood count; CSF, cerebrospinal fluid; ELISA, enzyme-linked immunosorbent assay; FTS, Alere Filariasis Test Strip; HOLA, hot oligonucleotide ligation assay; ICT, immunochromatographic card test; PRNT90, plaque reduction neutralization tests; RDTs, rapid diagnostic tests; RFLP, restriction fragment length polymorphism; SNT, serum neutralization tests.

Selection criteria for sentinel unit locations

A total of 45 criteria involved in the selection of sentinel locations were identified during the data extraction step (Table 4). These criteria were grouped into 6 broad categories: Risk, Environment, Population, Delimitation, Past information, and Logistics.

Table 4.

Criteria Used to Select Sentinel Unit Locations in Sentinel Surveillance Networks for Vector-Borne Disease, as Extracted from the Articles Included in Our Scoping Review

Type	ID	Criterion	Description	No. of articles
Risk	R1	Risk (human)	There is documented risk of disease due to the presence of human cases within the SUL	55
	R2	Variation in risk	There is a variation in degree of risk of the disease between the SUL	25
	R3	Risk (vector)	There is documented risk of disease due to the presence of appropriate vector disease within the SUL	22
	R4	Risk (unspecified)	There is documented risk of disease, however, the nature of the risk is not elucidated within the SUL	20
	R5	Risk (host animals)	There is documented risk of disease due to the presence of appropriate host species within the SUL	17
	R6	Proximity to risk	The SUL is in proximity to an area with documented risk of disease	8
	R7	Risk (geography)	There is documented risk of disease due to geography (abiotic) within the SUL	5
	R8	Suspected risk	There is suspected risk of disease within the SUL	3
	R9	No risk	There is no documented risk of disease within the SUL	2
	R10	Risk (interface)	There is documented risk of disease due to the presence of vector–human interface within the SUL	2
Environment	E1	Geographical features	The geography of the SUL has been taken into consideration during the selection	27
	E2	Ecology (vector)	The ecology of the SUL is appropriate for the presence of the vector	18
	E3	Variation in ecology	The SUL has been chosen due to variation in ecology between these units	23
	E4	Ecology (unspecified)	The ecology of the SUL has been taken into consideration during the selection, however, authors have not specified how	6
	E5	Variation in geography	The SUL has been chosen due to variation in geographical features between these units	7
	E6	Ecology (host animal)	The ecology of the SUL is appropriate for the presence of the host species	7
	E7	Proximity to area of interest	The SUL is near an area of interest, such as a school	5
	E8	Livestock population	Selection of the SUL to maximize the volume of livestock within the units	4
	E9	Climate	Climate has been taken into consideration in the selection of the SUL	4
	E10	Ecology (disease)	The ecology of the SUL is appropriate for the presence of the VBD	2
	E11	Variation in farming practices	The SUL has been chosen due to variation in farming practices between these units	1
	E12	Variation in housing type	The SUL has been chosen due to variation in housing type between these units	1
Population	P1	Population numbers	Selection of the SUL to maximize the population reached within the units of the study zone	17
	P2	Population demographics	Population demographics are considered during the selection of SUL	10
	P3	Population stability	The populations within the SUL are stable (no immigration/emigration)	6
	P4	Population instability	The population within the SUL are unstable (immigration/emigration)	2
	P5	Presence of human activity	There is presence of a specific type of human activity (e.g., fishing, hunting, and wild mushroom picking) within the SUL	3
	P6	Health clinic demographics	Demographics of the health clinics are considered during the selection of SUL	1
Distribution	D1	Administrative boundaries	Selection of SUL according to administrative boundaries	29
	D2	Random distribution	Random distribution of SUL in the study zone	16
	D3	Even distribution	Even distribution of SUL through the study zone	9
	D4	Minimal distance	Separation of SUL by a minimal distance	2
Past information	I1	Past surveillance	The SUL was chosen as they had been used previously in surveillance programs	34
	I2	Previous studies	The SUL was chosen as they had been used in previously in scientific studies	21
	I3	Previous PH interventions	There are previous public health interventions carried out within the SUL	12
	I4	No previous PH interventions	There are no previous public health interventions carried out within the SUL	8
	I5	Variation in PH interventions	There is a variation in public health interventions carried out with the SUL	5
	I6	Areas of scientific interest	The SUL has been chosen as they represent areas of increased scientific interest	3
	I7	Modeling	The SUL was chosen as there is modeling data to support their selection	1
Logistics	L1	Logistics	Logistical constraints (e.g., travelling distance, access) are considered for the SUL	31
	L2	Voluntary enrollment	SUL is based on voluntary enrollment	14
	L3	Stakeholders	The SUL is selected according to stakeholder preferences, suggestions or recommendations	3
	L4	Specialist centers	There are specialists or a specialist center within the SUL	2
	L5	Threshold of consultations	The SUL is selected to ensure that a minimal threshold of patient consultations is achieved	2
	L6	Communications	There are adequate communication facilities within the SUL	2

The criteria are classified into broad categories: Risk, Environment, Population, Distribution, Past information, and Logistics.

SUL, sentinel unit location; VBD, vector-borne disease.

The Risk category includes criteria which evaluate the presence or absence of an indicator of risk within the selected sentinel units (e.g., presence of vectors, host animals or human cases). The Environment category takes into consideration the natural features of the study zone, such as habitat suitability for vectors or host animals, based on ecological, meteorological, or geographical data. The Population category is directly related to the human population of the study zone, either for demographic data, presence of human activity, or population dynamics. The Distribution category refers to criteria guiding the spatial distribution of sentinel units across the study area. The Past Information category incorporates previous knowledge, from former articles or surveillance programs, which supported the selection of sentinel unit locations. Finally, the Logistics category groups criteria, which were used to maximize feasibility of the sentinel surveillance network, including access or diffusion of results. On average, articles used 2.4 criteria to determine sentinel unit locations.

The most common categories of criteria used to determine sentinel unit locations were Risk, Past information, and Environment, with a total of 122 (59.2%), 79 (38.3%), and 74 (35.9%) articles using criteria from those categories, respectively. The Distribution, Logistics, and Population categories were less commonly used, with a total of 51 (24.82%), 43 (20.9%), and 30 (14.6%) articles reporting using these criteria, respectively.

Evaluation of the sentinel surveillance networks

Of the 206 articles, a vast majority of sentinel networks were reported to be useful (n = 200, 97.1%) by the authors. Furthermore, acceptability of the sentinel network was high (n = 191, 92.7%), with good sensitivity (n = 187, 90.8%), and representativeness (n = 157, 76.2%). Meanwhile, a large proportion of the articles had an obvious degree of complexity (n = 105, 51.0%). Overall, data quality was high, with 99 articles (48.1%) reporting less than 10% of data missing.

Some of the evaluation parameters showed a higher degree of uncertainty. Overall, 182 (88.3%), 181 (87.9%), and 175 (85.0%) of the articles did not provide enough information to objectively assess stability, timeliness, and flexibility, respectively.

Best fitting models, as determined by backward stepwise model selection using AIC, are shown in Table 5. Minimal AIC were 132.6 and 232.3 for the sensitivity and representativeness regressions, respectively. Logistic regressions showed that choosing sites used in previous studies, based on population numbers or using a risk measure of host animals were strongly significantly associated with sensitive surveillance systems. Meanwhile, the use of logistical constraints, population numbers, use of sites from previous studies or past surveillance initiatives, sites with previous public health intervention, and variation in ecology were criteria significantly associated with representative surveillance systems.

Table 5.

Logistic Models Investigating the Effects of Using Specific Sentinel Site Selection Criteria on Surveillance Network Sensitivity and Representativeness

Criteria	Sensitivity				Representativeness
Criteria	Value	Std. Err.	\|z\|	p	Value	Std. Err.	\|z\|	p
Intercept	4.433	0.962	4.610	<0.001	0.501	0.260	1.923	0.054
D1	−1.632	0.864	−1.889	0.059	0.493	0.553	0.893	0.372
Administrative boundaries
D2	−0.350	1.195	−0.293	0.770
Random
E1	−0.700	0.982	−0.713	0.476
Geographical features
E2	−1.237	0.958	−1.291	0.197
Ecology (vector)
E3	−1.550	1.076	−1.441	0.150	0.869	0.330	2.630	0.009
Variation in ecology
I1	1.055	1.172	0.900	0.368	0.836	0.332	2.519	0.012
Past surveillance
I2	1.640	0.702	2.338	0.019	2.386	0.290	8.224	<0.001
Previous studies
I3					2.417	0.289	8.355	<0.001
Previous PH interventions
L1	−0.245	0.965	−0.254	0.799	1.338	0.648	2.065	0.039
Logistics
L2	−0.902	1.256	−0.719	0.472	0.675	0.819	0.824	0.410
Voluntary
P1	1.909	0.710	2.688	0.007	1.457	0.678	2.150	0.032
Population numbers
R1	−1.685	0.939	−1.795	0.073
Risk (human)
R2	−0.923	0.935	−0.988	0.323	1.115	0.652	1.710	0.087
Variation in risk
R3					−0.149	0.536	−0.278	0.781
Risk (vector)
R4	−2.06	1.4	−1.47	0.142
Risk (unspecified)
R5	2.952	1.244	2.374	0.018
Risk (host animals)
McFadden's R ²	0.190				0.069

Sensitivity, as defined by the CDC, refers to the ability of the network to monitor changes in prevalence in number of cases, vector density, or pathogen over time; the reference level for the analyses was “ nonsensitive.” Representativeness refers to the ability of the network to accurately describes the occurrence of a health-related event over time and its distribution in the population by place and person; the reference level for the analyses was “not representative.” Selection criteria, which were used in >5% of articles (or in over 10 articles) were included in the models. CDC, Centers for Disease Control and Prevention.

Discussion

VBDs are rapidly emerging worldwide, a phenomenon precipitated by climate change (Campbell-Lendrum et al. 2015, Semenza and Suk 2018, Rocklöv and Dubrow 2020). As the ecological ranges of vectors expand, public health authorities see the need to survey large geographical areas to detect changes in the distribution of vectors and associated changes in the spatial distribution of VBD risk. Sentinel surveillance provides an attractive surveillance method by limiting costs and offering targeted insight into the disease cycle through time. Our scoping review offers a portrait of how sentinel surveillance networks for VBDs are designed, including the type of sentinels used, the number of sentinel units, and the disease detection methods used. Furthermore, we have summarized key criteria used for the selection of sentinel unit location in the context of VBDs and provided a rudimentary evaluation of these networks.

The four VBDs most frequently targeted by sentinel surveillance systems in the reviewed literature were malaria, West Nile virus, lymphatic filariasis, and schistosomiasis. Overall, mosquito-borne diseases accounted for two thirds (67%) of retained articles, possibly reflecting global concern that by 2050 half of the world's population could be exposed to disease-spreading mosquitoes, such as Aedes spp., due to their expanding habitat range (Kraemer et al. 2019). Malaria represents a major global health burden, accounting for ∼435,000 deaths annually (WHO 2020) and it was therefore not surprising that one third (33%) of retained articles focused on this parasite. However, despite dengue showing the greatest increase in global incidence in the last 50 years, with a 30-fold rise, none of the articles included in our review targeted this disease (WHO 2014). Meanwhile, sentinel surveillance has been shown to be effective in monitoring transmission for other diseases spread by Aedes aegypti, such as Zika and chikungunya (Barrera et al. 2019). As collection methods and laboratory techniques are often similar for analysis of pathogens transmitted by a specific vector species, we suggest that existing sentinel networks could increase their impact by diversifying surveillance targets to include a larger breadth of neglected tropical diseases in regions where they are emerging or endemic.

Mosquito trapping, blood sampling, or symptom surveillance/questionnaires in humans, followed by blood sampling in animals were the most commonly used data collection methods. As for laboratory methods, Enzyme-Linked Immunosorbent Assay (ELISA) and PCR tests were most often employed. These relatively simple methods require limited training and resources (in comparison to more extensive methods such as rodent capture, physical examinations, and specialized laboratory tests) and may therefore have logistical advantages over more complex or expensive techniques in terms of network feasibility and sustainability.

A total of 45 selection criteria were identified from the different articles retained in our scoping review, reflecting the large variety of criteria considered by public health authorities when designing surveillance networks. This list of criteria provides a starting point for researchers and public health authorities seeking to identify selection criteria that will help design a sentinel network capable of fulfilling their surveillance objectives. Overall, an average of 2.4 criteria were used in the articles examined in this review, suggests that, in practical terms, working with a limited number of criteria may be sufficient to achieve satisfactory sentinel unit selection.

Risk-based criteria were the most frequently used for sentinel unit selection. Risk measures included incidence of human cases of VBD, presence of host animals, variation of risk between sentinel site locations, presence or abundance of the primary vector, or lastly, some unspecified measure of risk of disease. As sentinel surveillance targets a limited subset of the population, using risk criteria to orientate the location of sentinel units should allow for more sensitive data collection. Logistic regression analysis showed that using a selection criterion based on risk as measured by host animals is significantly correlated with sensitive surveillance systems (p = 0.018). This supports the use of sentinel animals as a sensitive initial indicator of risk (Herrer 1982, Julian et al. 2002, Kwan et al. 2010, Achazi et al. 2011, Petrović et al. 2018), with the caveat that care must be taken to select an epidemiologically appropriate sentinel animal for the disease being studied (Halliday et al. 2007).

Environment was the second most frequent category of criteria used in selection of sentinel unit locations—particularly geographical features. Variation in ecology (without explicit detail into which features were taken into consideration) were correlated with a more representative surveillance system (p = 0.009). None of the environmental criteria was correlated with sensitive surveillance systems. Despite this finding, we argue the importance of considering environment and landscape features in the selection of a sentinel unit, as VBDs are known to be sensitive to environmental changes, including ecological, landscape, and geographical characteristics (Campbell-Lendrum et al. 2015, Lowe 2018). However, choosing sentinel unit locations based solely on environment may hinder representativeness if anthropogenically driven factors, such as human movement, urbanization, and poverty, are not taken into consideration (Ali et al. 2017). Environmental criteria nonetheless play an increasingly important role in the era of climate change, and so we suggest that careful selection of sentinel unit locations based on both ecological and geographical criteria will help insure sensitivity and long-term relevance of the surveillance system, especially in the context of disease emergence.

Many surveillance networks used past information to select sentinel unit locations, taking advantage of sites used during previous surveillance initiatives (34 networks) or research projects (21 networks). These selection criteria are likely to lead to a longer time series and reduce resources required to establish new sentinel locations. Furthermore, surveillance systems that used past study sites (including past surveillance sites, or sites which have been previously targeted for public health interventions) as a selection criterion were significantly more representative (p < 0.001). Considering this information, we suggest that using existing sites established in previous initiatives and converting these to sentinel unit locations may be advantageous, provided that the previous study's objectives and data collection methods are compatible with those of the new surveillance network. Meanwhile, surveillance networks which evaluate intervention methods represent a special subgroup; in this case it is also important to determine if sentinel unit locations are to be places in areas with or without application of the targeted public health interventions.

Researchers should evaluate the added value of considering the geographic distribution of their sites—administrative boundaries may have a slightly negative impact on sensitivity of the sentinel surveillance network (p = 0.059), but could ensure equity in resource allocation across a territory. This could be overcome by selecting an even distribution of sites within the study zone or random distribution of sites which do not appear to have a negative impact.

Taking into consideration population-based criteria, such as population numbers, for selection of sentinel unit location, was associated with increased representativeness and sensitivity of the surveillance network (p = 0.032; p = 0.007). This is unsurprising since representativeness describes the accurate representation of the disease distribution in the population by place and person. Thus, targeting the right population, or maximizing the population reached by the sentinel units can provide a better portrait of the disease situation.

Finally, logistical criteria (for instance, taking into consideration travel distance and access to sentinel sites) was associated with representativeness of the surveillance network (p = 0.039). This suggests that sound strategic planning of sentinel unit locations, taking into account logistical constraints and aiming for voluntary participation, is more likely to result in effective and feasible data collection, resulting in a better understanding of the disease and its repercussions on human populations.

One limit of this scoping review is that the literature search targeted VBDs, which can be transmitted to humans. However, articles that reported on VBDs solely impacting animal health were not excluded, as they are still considered of public health importance, and can impact human populations indirectly. These diseases, such as bluetongue, are nevertheless under-represented in our results. Although many of the criteria identified will be broadly applicable to any type of sentinel network targeting VBDs, specific additional criterial may be important to consider when building sentinel surveillance networks for VBDs affecting principally animal populations, or even plants (Huang et al. 2020).

Our logistic regressions provided insight into ways of prioritizing criteria selection to optimize sentinel surveillance network performance. However, these regressions should mainly be used as a guide for performance of the selection criteria, as several limitations of this approach are important to note. First, lack of sensitivity or representativeness may not necessarily be due to sentinel unit location, but due to the surveillance methods used (e.g., failing to use an appropriate species as sentinel animals) (Crans 1986). Next, due to the complex nature of evaluating surveillance networks and the limited information available in the articles, we used simplified proxies to determine whether a particular surveillance network successfully met the CDC system attributes (German et al. 2001). However, there was still a high volume of missing information, where the reviewers were unable to determine whether the parameter was fulfilled.

Our results from the descriptive analyses of evaluation parameters and the performance score highlight an important gap in thorough reporting of surveillance functionality in publications. In addition, in articles where the aim was specifically to evaluate the surveillance system, the evaluation usually focused on a single key aspect of the overall performance of the network, such as sensitivity, representativeness, timeliness, or stability (Crans 1986, Koukounari et al. 2011, Pultorak et al. 2011, Yukich et al. 2014, Bleyenheuft et al. 2015, Healy et al. 2015, Petrović et al. 2018, Sanou et al. 2018, Wahnich et al. 2018). The need for a comprehensive approach to evaluate surveillance systems, which should be complete, flexible, and operational has been identified in the past (Calba et al. 2015). We add to this conclusion that clear and concise reporting of surveillance network evaluations should be incorporated into this approach. This would allow researchers and public health authorities implementing new surveillance networks or adding new sentinel unit locations to their network to assess potential benefits and challenges associated with different surveillance network designs.

Conclusion

Our scoping review characterized different elements required for the construction of a sentinel surveillance network for VBDs. Findings from the literature can act as a reflection exercise for those wishing to establish a new sentinel surveillance system. We have identified tools that could prove valuable for such aims, including a standardized and comprehensive approach to evaluating surveillance systems and a tool to prioritize criteria that will aid in spatial design of an effective surveillance network. In particular, given the high number of criteria and the particularities of individual VBDs, the development of an algorithm which could be applied by researchers or public health authorities to prioritize criteria to meet surveillance objectives would be a useful future development in this area.

Footnotes

Acknowledgments

The authors would like to extend their thanks to the Public Health Library of Public Health Agency of Canada and Katherine Merucci for their support in conducting their literature search.

Authors' Contributions

C.G., P.L., and C.B. conceptualized the project. C.G., C.B., P.B., and M.M. designed the scoping review. P.L. and C.S. contributed to this design. P.L. and C.B. supervised the project. C.G., C.S., and C.-A.V. collected the data by acting as reviewers for the scoping review. C.G. wrote the article, and results were analyzed with the support of C.S., P.B., M.M., C.B., C.S., and P.L. and commented on the article, and all coauthors reviewed the final draft.

Author Disclosure Statement

No conflicting financial interests exist.

Funding Information

This research was funded by the Canadian Lyme Disease Research Network (CLyDRN) and the Faculty of Veterinary Medicine, Université de Montréal.

Supplementary Material

Supplementary Data

References

Achazi

, Růžek

, Donoso-Mantke

, Schlegel

, et al. Rodents as sentinels for the prevalence of tick-borne encephalitis virus. Vector Borne Zoonotic Dis, 2011; 11:641–647.

Ali

, Gugliemini

, Harber

, Harrison

, et al. Environmental and social change drive the explosive emergence of Zika virus in the Americas. PLoS Negl Trop Dis, 2017; 11:e0005135.

Bahuon

, Marcillaud-Pitel

, Bournez

, Leblond

, et al. West Nile virus epizootics in the Camargue (France) in 2015 and reinforcement of surveillance and control networks. Rev Sci Tech, 2016; 35:811–824.

Barker

, Reisen

. Chapter 4 - Epidemiology of vector-borne diseases. In: Mullen

, Durden

, eds. Medical and Veterinary Entomology (Third Edition). Cambridge, MA: Academic Press, an Imprint of Elsevier, 2019:33–49.

Barrera

, Amador

, Acevedo

, Beltran

, et al. A comparison of mosquito densities, weather and infection rates of Aedes aegypti during the first epidemics of Chikungunya (2014) and Zika (2016) in areas with and without vector control in Puerto Rico. Med Vet Entomol, 2019; 33:68–77.

Bleyenheuft

, Lernout

, Berger

, Rebolledo

, et al. Epidemiological situation of Lyme borreliosis in Belgium, 2003 to 2012. Arch Public Health, 2015; 73:33–33.

Calba

, Goutard

, Hoinville

, Hendrikx

, et al. Surveillance systems evaluation: A systematic review of the existing approaches. BMC Public Health, 2015; 15:448.

Campbell-Lendrum

, Manga

, Bagayoko

, Sommerfeld

. Climate change and vector-borne diseases: What are the implications for public health research and policy?. Philos Trans R Soc Lond B Biol Sci, 2015; 370:20130552.

Crans

WJ.

Failure of chickens to act as sentinels during an epizootic of eastern equine encephalitis in southern New Jersey, USA. J Med Entomol, 1986; 23:626–629.

10.

Eder

, Cortes

, Teixeira de Siqueira Filha

, Araújo de França

, et al. Scoping review on vector-borne diseases in urban areas: Transmission dynamics, vectorial capacity and co-infection. Infect Dis Poverty, 2018; 7:90.

11.

Fox

, Weisberg

An {R} Companion to Applied Regression, Third Edition. Thousand Oaks CA: Sage, 2019. Available at https://socialsciences.mcmaster.ca/jfox/Books/Companion/

12.

German

, Lee

, Horan

, Milstein

, et al. Updated guidelines for evaluating public health surveillance systems recommendations from the Guidelines Working Group. MMWR Recomm Rep, 2001; 50:1–35; quiz CE31.

13.

Girond

, Randrianasolo

, Randriamampionona

, Rakotomanana

, et al. Analysing trends and forecasting malaria epidemics in Madagascar using a sentinel surveillance network: A web-based application. Malar J, 2017; 16:72.

14.

Halliday

JEB

, Meredith

, Knobel

, Shaw

, et al. A framework for evaluating animals as sentinels for infectious disease surveillance. J R Soc Interface, 2007; 4:973–984.

15.

Healy

, Reisen

, Kramer

, Fischer

, et al. Comparison of the efficiency and cost of West Nile virus surveillance methods in California. Vector Borne Zoonotic Dis, 2015; 15:147–155.

16.

Herrer

Use of golden hamster as sentry in endemic areas of uta (cutaneous leishmaniasis). Rev Inst Med Trop Sao Paulo, 1982; 24:162–167.

17.

Huang

, Reyes-Caldas

, Mann

, Seifbarghi

, et al. Bacterial vector-borne plant diseases: Unanswered questions and future directions. Mol Plant, 2020; 13:1379–1393.

18.

Julian

, Eidson

, Kipp

, Weiss

, et al. Early season crow mortality as a sentinel for West Nile virus disease in humans, northeastern United States. Vector Borne Zoonotic Dis,, 2002; 2:145–155.

19.

Koukounari

, Touré S, Donnelly CA, Ouedraogo A, et al. Integrated monitoring and evaluation and environmental risk factors for urogenital schistosomiasis and active trachoma in Burkina Faso before preventative chemotherapy using sentinel sites. BMC Infect Dis, 2011; 11:191.

20.

Kraemer

MUG

, Reiner

, Brady

, Messina

, et al. Past and future spread of the arbovirus vectors Aedes aegypti and Aedes albopictus . Nat Microbiol, 2019; 4:854–863.

21.

Kwan

, Kluh

, Madon

, Nguyen

, et al. Sentinel chicken seroconversions track tangential transmission of West Nile virus to humans in the greater Los Angeles area of California. Am J Trop Med Hyg, 2010; 83:1137–1145.

22.

Lowe

The impact of global environmental change on vector-borne disease risk: A modelling study. Lancet Planet Health, 2018; 2:S1.

23.

Maynard

, Ambrose

, Cooper

, Chow

, et al. Tiger on the prowl: Invasion history and spatio-temporal genetic structure of the Asian tiger mosquito Aedes albopictus (Skuse 1894) in the Indo-Pacific. PLoS Negl Trop Dis,, 2017; 11:e0005546.

24.

Ogden

NH.

Climate change and vector-borne diseases of public health significance. FEMS Microbiol Lett, 2017; 364:1–8.

25.

Petrić. D, Petrovic T, Hrnjakovic I, Zgomba M, et al. West Nile virus ‘circulation’ in Vojvodina, Serbia: Mosquito, bird, horse and human surveillance. Mol Cell Probes, 2017; 31:28–36.

26.

Petrović. T, Šekler M, Petrić D, Lazić S, et al. Methodology and results of integrated WNV surveillance programmes in Serbia. PLoS One, 2018; 13:e0195439.

27.

Pultorak

, Nadler

, Travis

, Glaser

, et al. Zoological institution participation in a West Nile Virus surveillance system: Implications for public health. Public Health, 2011; 125:592–599.

28.

R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, 2020. RL: Available at https://www.R-project.org/.

29.

Racloz

, Griot

, Stärk

KDC

. Sentinel surveillance systems with special focus on vector-borne diseases. Anim Health Res Rev, 2006; 7:71–79.

30.

Rocklöv

, Dubrow

. Climate change: An enduring challenge for vector-borne disease prevention and control. Nat Immunol, 2020; 21:479–483.

31.

Sanou

, Dirlikov

, Sondo

, Kagoné TS, et al. Building laboratory-based arbovirus sentinel surveillance capacity during an ongoing dengue outbreak, Burkina Faso, 2017. Health Secur, 2018; 16(S1):S103–S110.

32.

Semenza

, Suk

. Vector-borne diseases and climate change: A European perspective. FEMS Microbiol Lett, 2018; 365:fnx244.

33.

U.S. Department of Health and Human Services. Principles of Epidemiology in Public Health Practice: An Introduction to Applied Epidemiology and Biostatistics. Atlanta, GA: Centers for Disease Control and Prevention (CDC), Office of Workforce and Career Development, 2006.

34.

Venables

, Ripley

. Modern Applied Statistics with S. Fourth Edition. New York, NY: Springer, 2002.

35.

Wahnich

, Clark

, Bloch

, Kubinson

, et al. Surveillance for mosquitoborne transmission of Zika virus, New York City, NY, USA, 2016. Emerg Infect Dis J, 2018; 24:827.

36.

WHOa. Vector-Borne Diseases. 2020. Available at https://www.who.int/news-room/fact-sheets/detail/vector-borne-diseases.

37.

WHOb. Public health surveillance. 2020. Available at https://www.who.int/immunization/monitoring_surveillance/burden/vpd/en/.

38.

WHO. WHO global health days. About vector-borne diseases. 2014. Available at https://www.who.int/campaigns/world-health-day/2014/vector-borne-diseases/en/#:∼:text=Vector%2Dborne%20diseases%20account%20for,627%20000%20deaths%20in%202012.

39.

WHO. World Health Data Platform: Malaria. 2020. Available at https://www.who.int/data/gho/data/themes/malaria.

40.

Yukich

, Butts

, Miles

, Berhane

, et al. A description of malaria sentinel surveillance: A case study in Oromia Regional State, Ethiopia. Malar J, 2014; 13:88.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.56 MB

0.00 MB