Table 1 Synthesis of data from the included articles (n = 18)

Single data signal SyS system

Over-the-counter (OTC) sales of pharmacy sales

Edge et al., 2004 [30]*

NI

NI

In situations where infected individuals have symptoms prompting self-medication, OTC sales trend would provide a more sensitive, timely and geographically specific detection tool than monitoring emergency room visits and laboratory-based surveillance

Adaptations to the algorithm will have to be developed to adjust for a number of factors contributing to the general noisiness of these data such as seasonal effects, promotional sales and type of population served. The success of such system will rely on automatic collection, analysis and dissemination of results

Kirian et al., 2011 [25]*

NI

Sensitivity: 4–14%, specificity: 97–100%

It may capture symptoms in the population before a person with gastrointestinal illness seeks health care

It does not necessarily indicate the buyer’s location, their demographic status, or the reason for the purchase. Those who purchase OTC medications for their illness may not be representative of the sick population as a whole. Hoarding behaviour will also affect the outcome

Reimbursement of prescription drugs

Mouly et al., 2016 [20]*

NI

Sensitivity: 6% and 21% for two examined outbreaks

Prescription drug data can be considered for the development of a detection system of waterborne outbreaks given its ability to describe an epidemic signal. It could support authorities in slow developing outbreaks

The algorithm cannot be used directly in other countries because of their different health systems, types and sources of data, and medical practices. The accuracy depends on the medical consultation rate in the impacted population. The accuracy of using health insurance data to describe waterborne outbreaks depends on the medical consultation rate in the impacted population, however, as this is never the case, data analysis underestimates the total number of acute gastrointestinal cases

Calls to health advice line (‘telehealth’)

Bjelkmar et al., 2017 [21]*

 ~ 2.5 months

NI

Comparing call patterns between water distribution areas that were based on groups of postal codes gives timely indication of the underlying cause and therefore substantially increases the chances of effective countermeasures

Tradeoff between sensitivity and specificity in signal detection. Need for a protocol for signal evaluation and validation, especially for regions where the population size is small

Multiple data signal SyS systems

Emergency care data; medical dispatch, ambulance medical service, emergency department chief complaints

Balter et al., 2005 [27]*

NI

NI

Emergency department syndromic surveillance might prove useful for detecting a problem and quantifying its magnitude

This system cannot determine the true etiology. If insufficient information exists to initiate an investigation, the decision is often made to observe whether the signal continues the next day, thereby losing syndromic surveillance's theoretical advantage of timeliness

Ziemann et al., 2014 [24]*

NI

NI

This system could detect changes in local trends and clusters of statistical alarms

It is not likely to detect local gastrointestinal outbreaks with few, mild, or dispersed cases. The probability of detecting an outbreak increases with the outbreak size. The results cannot be generalized to region-level data or very sparse time series

Over-the-counter (OTC), web queries, calls to health advice line

Andersson et al., 2014 [19]*

NI

Calls to health advice line: sensitivity: 40–50%, specificity: 99%, web queries and OTC: no signal

SyS can serve as an early warning for waterborne outbreaks, especially with telephone triage data with sufficient temporal and spatial resolution. It may be suited to detecting widespread rises in syndromes and, rarely, small-scale outbreaks

The alarm does not contain information on the cases’ medical status to validate the cause of the alarm. Moderate and low outbreaks (< 1000 cases) are unlikely to be detected. Limitations to the reported results are linked to one of the four outbreaks were not waterborne

Telehealth, in-hours and out-of-hours GP, ED visits

Smith et al., 2010 [22]*

Peak of calls coincides with outbreak (95% CI) in one area

NI

Multiple syndromic data streams are an advantage

Telehealth may, in general, be driven by media bias

Chief complaints of patients reporting to emergency departments, over-the-counter and prescription pharmacy sales, and worker absenteeism

Heffernan et al., 2004 [28]*

NI

NI

Syndromic surveillance systems have proved useful for detecting substantial citywide increases in common viral illnesses (e.g. influenza, norovirus and rotavirus)

The studied systems have not detected more contained outbreaks earlier than traditional surveillance

Combined health, spatial and environmental data

Proctor et al., 1998 [29]*

Timeliness of learning about the peak was 15 days earlier in in monitoring treatment plant effluent turbidity compared to ER’s visits and clinical laboratory

NI

It is noted the value of alternate data sources as early warning systems which can complement laboratory diagnosis

There are weaknesses for all proposed surrogate waterborne surveillance systems. For example, turbidity did not give information on disease causing-organisms; and treated water meeting quality standard could still contain sufficient level of pathogens

Rambaud et al., 2016 [26]*

NI

NI

Combining two complementary methods protects against false positives, e.g. confusion of cases stemming from exposure from other types of food or swimming, for example

Pilot-study and not tested on a larger scale

Coly et al., 2017 [23]*

NI

Detected outbreaks < 100 cases

Increases sensitivity and timely detection of waterborne outbreaks

These systems are expensive in terms of resources and shared expertise in incorporating local knowledge regarding both environmental and health data

Simulations

Method evaluations via simulations of multiple signal SyS systems

Cooper et al., 2006 [36]**

Unlikely to detect local outbreak

NI

It may capture symptoms in the population before seeking health care

The alarm does not contain information regarding the cases’ medical status to validate the cause of the alarm. Moderate and low outbreaks (< 1000 cases) are unlikely to be detected. The detection ability varies seasonally. Telehealth may, in general, be driven by media bias

Burkom et al., 2011 [31]**

NI

Sensitivity: 80%, specificity: 99%

Use of multiple syndromic data streams is an advantage. The number of false alarms is greatly reduced

Simulation results must generally be improved with real epidemiological data

Xing et al., 2011 [35]**

NI

Of the simulated models, the regression method had higher sensitivity (range 6–14% improvement of sensitivity in the surveillance system)

Demonstrates possible improvement in the surveillance system to increase sensitivity

Simulations based on small number of data points

Zhou et al., 2015 [34]**

3.3 to 6.1 days

When reported, the sensitivity ranged from 24 to 77%, and the PPV was 90.5%

Sensitivity and timeliness increase with stratification

Study population perhaps not representative

Colón-Gonzales et al., 2018 [33]**

Unlikely to detect outbreaks < 1000 cases

NI

Framework applicable for other SyS systems

The detection ability varies seasonally

Mouly et al., 2018 [32]**

NI

Sensitivity: 73%, PPV: 90.5%

Space–time increases the likelihood of detecting outbreaks

The probability of detecting outbreaks increases with the outbreak size