Using Large Scale Rapid Antigen Testing (RAT) to Inform Participatory Ad-Hoc Community Surveillance for Emerging Communicable Disease Epidemics

Abstract

Besides the diagnostic use for infectious diseases in a point-of-care clinical settings, the simplicity and ease of self-performed RAT can also be an alternative approach for informing disease surveillance at the community level, carrying the potential advantage of enhanced timeliness, acceptability, and flexibility. Commissioned by the Hong Kong Government during the catastrophic fifth wave of the COVID-19 pandemic, our team established and maintained an ad-hoc large-scale participatory daily antigen rapid testing surveillance (DARTS) system for real-time situational awareness of SARS-CoV-2 activity to inform policy consideration in a timely manner. This Chapter will describe the concept and design of the surveillance approach, examine the practical feasibility and challenges, related logistical consideration on implementation and maintenance, technical aspects of data analysis to cater for the unique surveillance need, and other potential additional contribution of the data on understanding the novel disease (estimating vaccine effectiveness, and symptomatology and viral shedding pattern).

1. Introduction

Since its emergence in late 2019, COVID-19 has caused unprecedented impact in terms of both mortality and morbidity in many countries. As of 8 February 2023, an estimated cumulative incidence of over 754 millions and mortality of more than 6.8 million were recorded globally [1].

As a novel infection causing the unprecedented global disease burden through successive rapidly evolving ways of epidemics in different countries, the painful experience from COVID-19 highlighted the need and importance of robust, timely, and representative community surveillance of novel and rapidly emerging communicable disease epidemics to guide a precise situational assessment and inform appropriate downstream public health strategy in an evidence-based manner. Although rapid antigen test (RAT) has been around for decades for different infectious diseases such as influenza and malaria, their use was limited mainly as a screening/diagnostic tool in a clinical point-of-care setting. With its rapid global popularization as a self-testing tool in the COVID-19 pandemic, the simplicity of the procedure and the ease of its usage has opened new opportunity for its use in a scaled manner in the community as an alternative approach for informing disease surveillance, which also offering the potential advantage of enhanced timeliness, acceptability, and flexibility.

This Chapter will give an overview of the concept of disease surveillance, discuss the inherent problems and difficulties of surveillance for novel and emerging epidemic disease, explore the potential for using RAT as a tool for enhancing disease surveillance, and basing on the experience of an ad-hoc RAT-based community surveillance system developed over two waves of the COVID-19 pandemic in Hong Kong to examine the feasibility and consideration of using RAT for surveillance during an evolving epidemic.

2. Concept of disease surveillance

2.1 Aim of surveillance

Public health surveillance is “the ongoing, systematic collection, analysis, and interpretation of health-related data essential to planning, implementation, and evaluation of public health practice” [2]. With appropriate analysis and interpretation, suitably collected surveillance can provide valuable data to inform the monitoring and action for disease control and prevention. Specifically, continuous surveillance over time allows for the monitoring of the temporal pattern of targeted infectious diseases among a population, so as to flag any abnormal change in disease activity in a timely manner (situational awareness) [3]. If allowable by the data granularity, stratified surveillance by geographic area or population subgroups would also allow for the pin-down of any aberration spatially or demographically. Moreover, when coupled with suitable additional data and analyses, well-collected surveillance data may also help to shed light on other important aspects of an emerging/novel infection, including risk factors and epidemiologic profile, disease burden, clinical disease symptomatology, and severity pattern, the effectiveness of vaccination or other public health interventions, and transmissibility and growth rate of the evolving epidemic.

2.2 Common approaches of infectious diseases surveillance

Surveillance can be conducted in two different ways, passive and active surveillance. Passive surveillance systems rely on the existing routine data reported from health facilities to the health agency. In active surveillance systems, public health workers take an active role to identify cases and monitor the population of interest, such as active calling and home visiting to follow-up individuals, thus significantly greater human and financial resources are required to depict a more comprehensive picture of disease burden and trends [3].

Different approaches have been commonly adopted for infectious diseases surveillance in a wide range of settings. In notifiable disease surveillance, practitioners are mandated to report cases of some pre-specified notifiable diseases of epidemic potential to the health agency, usually predefined by the World Health Organization, national or regional authorities. Routine administrative data, e.g. clinic/emergency room attendance and hospital admission episodes from public hospitals are another common statistics for passive surveillance. Syndromic surveillance can be conducted by examining syndromic groups of symptoms captured in an appropriate setting such as the emergency room, or using hospital billing data with ICD codes. Laboratory surveillance performed at reference laboratory can confirm the agent of infection, and monitor the disease incidence, and evolving trends of subtypes and variants of an infectious virus, and its infectiousness, seroprevalence [4].

Sentinel surveillance can also be conducted in designated sentinel sites in specific settings, such as general out-patient clinics, private clinics, associated health services centers to monitor the trends in general population; or in setting with high risk of outbreak such as child care centers, kindergartens, schools; or residential care home or elderly. Although conceptually distinct, active and passive surveillance for some important infectious diseases (e.g. seasonal influenza infection) are usually being performed in parallel in different community and health care settings to complement each other.

2.3 Important attributes of a surveillance system

Surveillance systems using different data in different settings are having important differences in a number of key characteristics, which ultimately affect their suitability in reflecting the changing disease activity and fulfilling surveillance needs for different stages and situations. Major important characteristics included representativeness, timeliness, sensitivity, and specificity.

Representativeness refers to the ability to give an accurate reflection of the epidemiological trend and disease burden in a defined population. For instance, data from sentinel out-patient clinics cover health care seeking activity in the general community while hospital admission data geared more toward capturing severe infections that need to be hospitalized. On the other hand, institutional data such as schools and care homes reflect more on the disease activity in those high risk setting, and may precede an increase in activity in the general community.

Timeliness refers to the ability of the system to issue signals efficiently to flag potential upsurge in community disease activity. Although it would be ideal for any changing disease activity to be alerted in a near real-time manner to inform situational assessment and the implementation of control measures, this relates to how effective and efficient the data collection/ analysis/and dissemination can be taken place, and carried non-trivial technical and logistical implications [4].

Specificity denotes clinically the reliable exclusion of people not having the infection or disease, while in a surveillance sense it refers to the ability to not flagging many false alarm of a change in disease activity when there is not any. This is an essential attribute to prevent overloading downstream manpower needed for follow-up and investigation of false negative signals, which may lead ultimately to desensitization and poor acceptance of the system.

There is generally a natural trade-off between data timeliness and data specificity. For instance, early data in the course of an infection (e.g. use of over-the-counter antipyretics/school absenteeism) tended to be very timely but being non-specific, while data in the later stage of a disease (e.g. clinical attendance/hospital admission/laboratory diagnosis) tended to be more specific but being much less timely. Earlier indicators also tended to better reflect disease activity in the community, including those mild infections without the need to attend medical care; while later indicators are generally geared more toward capturing patients in the more severe end of the clinical spectrum.

Sensitivity clinically represents the ability to accurately identify people having the disease, while in a surveillance sense it refers to the ability to flag a real change in disease activity when there is any. As all signals would need to be follow-up, choosing a suitable cut-off for flagging a signal would be an important consideration to strike a balance between the sensitivity and specificity when building any surveillance system.

2.4 Other issues to consider in building an effective surveillance system

Depending on the aims and objectives of surveillance, different systems would have their own sets of prioritized characteristics of high importance. Common issues crucial to build an effective surveillance system are discussed below [5].

2.4.1 Objective

A clear and specific objective is crucial for aligning the aim of surveillance and the planned uses of surveillance data with the parametrization of the systems. The disease, infection, or health event under surveillance should be specified with a predefined case definition. A consistent case definition throughout the surveillance period helps to monitor the temporal trend of the health event over time and avoid artifactually introduced change in observed disease activity.

2.4.2 Data source

A stable and reliable data source shall be identified to collect surveillance data in a timely and efficient manner. The legal, ethical, and privacy implication of data collation, collection, and management shall be considered, with proper informed consent if deemed necessary in an active system capturing personal data. For data to be reported from hospitals, clinics, or practitioners, or historic health records archived, a suitable streamlined and user-friendly automatic reporting system integrated with existing health information systems would help to enhance the availability, timeliness, and validity of surveillance data.

2.4.3 Sampling

After clearly defining the population under surveillance, the approach of data collection (sentinel or sampling) shall be decided. Theoretically, the data should be captured from a sample representative of the targeted population. The two commonest sampling strategies included random sampling, which helps to enhance representativeness; and convenient sampling which helps to improve participation for a quick establishment of the system. Other sampling consideration may include defining the catchment area for different districts to ensure a proper coverage of cases in different geographical locations. Depending on the purposes of the surveillance systems, the time period and intervals of the surveillance (continuous or intermittent, short term or long term) can be specified.

2.4.4 Frequency

The frequency of data collection is usually related to the targeted timeliness of the surveillance system to reflect the disease activity in the community. A short time interval with a frequent data collection, (e.g. daily or a near real-time system) may be necessitated for monitoring the trajectory of a rapidly evolving epidemic disease carrying huge public health implication. In practice, however, the optimal frequency of collecting data is also depended and limited by the sources and types of data, availability of the required reporting platform and manpower, and subject to different logistical and ethical constraints.

2.4.5 Data analysis

Data cleaning, collation, and management plan shall be prospectively planned and before the roll-out of surveillance. A proper process of data management, transfer, and storage should be chosen to ensure the validity of data and the compliance with applicable data security and confidentiality policy consideration. Important issue to decide on regarding the analysis methods included the frequency of analysis (hourly, daily, weekly, monthly), the detailed analysis algorithm, the need for standardization or stratified analysis, the cut-off threshold for issuing any alert signal and its performance in terms of sensitivity and specificity.

2.4.6 Dissemination

The results of the surveillance shall be disseminated to the relevant stakeholders and end users as feedback and to inform relevant downstream follow-up decisions and actions. A dissemination plan with detailed description of the frequency of result dissemination, mechanism of dissemination, targeted audience, and whether it matched and included all the recipients identified in the objectives set.

3. Unprecedented problems of surveillance for a novel emerging epidemic disease such as COVID

Comparing to regular surveillance systems used for routine ongoing situational assessment of infectious diseases such as seasonal influenza infection, the surveillance for novel/ emerging/evolving epidemic diseases such as COVID-19 pandemic, is much more complex and poses many unprecedented challenges and problems. A fundamental consideration would include how to make sure a robust surveillance data, either an existing routine one or some specific new data, can be maintained to inform a timely, specific, and stable surveillance [6]. Generally, the routinely reported daily case count of an infection may not be a good data stream for surveillance, especially during the midst of a rapidly changing epidemic.

3.1 The inadequacy of routine case count as a surveillance data

Although being routinely reported by most health authorities in different countries and smaller geographical sub-regions like provenances and cities, the daily PCR-confirmed case count, whether by self-collected deep throat saliva samples or professionally sampled specimens in a suitable clinical setting, carried major shortcomings for being used as a surveillance indicator. Being a lagging indicator [7], its timeliness can be affected by the turnaround time for laboratory confirmation, which can particularly be prolonged due to constrained testing capacity during periods of peak epidemic activity, when data timeliness is of utmost importance. Its general bias toward capturing infections in the more severe end of the clinical spectrum also affects the representativeness and ability to reflect prevailing disease activity in the general population [8].

3.2 Lack of consistency in case ascertainment

A consistent reporting pattern of disease episodes underlies the validity of any surveillance system, or else any change in observed disease activity may only reflect a varying testing/reporting/ascertainment practice [7]. As illustrated in the case of COVID-19, these can be a particularly serious problem during an emerging epidemic as case definition for infection and recovery can both vary widely between different health authorities and over time, such as whether probable or asymptomatic cases are officially counted [9, 10]. Inconsistency in reporting practice over jurisdictions has compromised the completeness and comparability of the surveillance data [11]. Spurious disease trend due to changing or update in case definition over time can also occur with evolving understanding about the new disease.

3.3 Issues of changing testing and reporting practice

Time-varying access to testing, medical consultations, and changing admissions to hospitals impeded the stability of a surveillance system using those data and thus its ability to monitor trends over time [7]. Changing testing criteria, policy, and arrangements can lead to a biased and inconsistent outcome capturing over the course of surveillance. Comparing to initial preference of laboratory-based PCR testing, the increasing availability and popularity of self-administered RATs testing can result in observed upsurge in observed disease incidence. Changing official testing policy over time, like the adoption or abandonment of mandatory/voluntary testing or the use of blanket testing as a community screening practice at different premises and subpopulations can also affect the observed temporal disease pattern. For a self-reported data such as symptoms or RAT testing result, changing reporting practice due either to concerns on downstream implication like isolation and quarantine, stigmatization, work or financial detriments can all affect the surveillance data.

4. RAT usage as point-of-care test

With its logistical simplicity of usage, more affordable cost, and wider availability, rapid antigen test (RATs) has become increasingly popular for identifying SARS-CoV-2 infection during the COVID-19 pandemic in a wide variety of clinical and community settings [12, 13, 14]. Common sampling methods included nasal swab and pooled throat and nasal swabs, either collected by healthcare workers or the general public. In contract to laboratory PCR testing requiring sample collection, storage, and transfer to laboratory, which typically associated with a turnaround time of one to several days, RAT has the advantage of giving a real-time result, and a much lower implication on expertise and laboratory processing capacity.

Previous studies have reported RAT to have a high positive predictive value, which makes it useful in setting of RAT as a self-testing tool, is particularly crucial to facilitate testing in remote areas with limited access to laboratory testing [15].

4.1 Usual setting and usage for clinical purposes

Over the course of the COVID-19 pandemic, RAT was being used for three main purposes. For symptomatic individuals presenting to a healthcare setting, RAT was being used as a point-of-case diagnostic tool to inform triage decision and clinical management [16]. For patients under isolation and exposed person in quarantine, RAT was also being used for self-monitoring to inform the discharge decision [17]. For other people in the community, RAT was widely used as a screening tool to establish eligibility for entering different vicinities like schools, hospitals, restaurants, recreational premises, and other governmental or private facilities. When laboratory testing capability was being stretched beyond the capability to cater for the testing demand, many jurisdictions had also shifted altogether from PCR to RAT as the official case definition of an infection.

Being sensitive, cost-effective, and portable RATs have been regarded a better performance than PCR in identification of infectious individuals [17].

4.2 Potential advantage of RAT for informing disease surveillance

Although being mainly used thus far as a diagnostic and screening tool, RAT also carried some potential advantages, including timeliness, specificity, and coverage for its potential use as a surveillance tool for the continuous monitoring of the epidemic/pandemic disease activity in a community.

4.2.1 Timeliness

Although PCR-based case count was announced daily in many countries, it can be a very lagging indicator of the actual disease activity in the community, especially in areas having a limited laboratory testing capability and reserve. For RAT, as the test results of most brands are available within 20 minutes, this can minimize the turnover time required for specimen transfer and laboratory testing, and make it a suitable data to inform very timely or near real-time disease surveillance.

4.2.2 Specificity

From a systematic review examining the performance of RAT, high overall pooled specificity of 100.0% (95% CI = 98.8 to 100.0%) was reported among RATs for different COVID-19 variants [18]. Comparing to many traditional data being used for surveillance, including absenteeism, influenza-like illness, and other syndromic grouping, RAT result is having a much higher specificity for reflecting the ture activity pattern of the disease under surveillance by the system.

4.2.3 Coverage

As RAT can be easily performed by the general public, easy to manufacture and distribute, with wide availability in community settings, RAT can be readily accessible and conducted by individuals in remote areas, which helps to increase the coverage of testing on patient over the whole severity spectrum, and thus enhancing the representative of the surveillance system.

Among traditional surveillance data, there is generally an inherent trade-off between these three characteristics. More timely data are generally much less specific, and more specific data, like laboratory confirmed cases of influenza, are much less timely and reflecting covering those having more severe disease.

5. An example of RAT-based disease surveillance system for COVID-19

5.1 Background and context

Since its emergence in late 2019, COVID-19 has caused unprecedented impact in terms of both mortality and morbidity worldwide. The daily number of PCR-confirmed cases (case counts) represented the most fundamental data routinely reported in most countries to report the clinical and public health burden of COVID-19 during the evolving COVID-19 pandemic. However, its ability to accurately reflect the changing epidemic temporal trajectory in the community can be affected by many of the inherent bias of this data. These included its inherent bias toward preferentially capturing diseases over the more severe end of the clinical symptomatology spectrum or having an identifiable risk exposure, so that they would come forward for the PCR testing. In the contrary, self-performed RAT, because of its simplicity and high accessibility, may help to give a much more representative picture of the disease activity in the community when being used with an appropriate testing strategy. On the other hand, the better timeliness, flexibility, and scalability of RAT also facilitated its use in large-scale surveillance programme for informing situational awareness in an epidemic setting.

In this section, the feasibility and logistical considerations of using self-performed RAT to inform ad-hoc or regular disease surveillance in the context of a novel and evolving epidemic will be discussed, basing on the experience of a large-scale participatory Daily Antigen Rapid Testing Surveillance (DARTS) System, implemented to monitor the situation of COVID-19 in real time during the fifth wave of the COVID pandemic in Hong Kong.

In February 2022, Hong Kong was struck by its worse pandemic wave of COVID-19 with the Omicron variant (the fifth wave), which had caused an unprecedented impact and disruption to the societal functioning, healthcare delivery, and local economy. In particular, the sudden and rapid upsurge of case load had overwhelmed both the sampling manpower and laboratory capacity for PCR testing. The subsequent piling-up of specimens and the much prolonged turnaround time for diagnosis confirmation had jeopardized the timely implementation of downstream clinical management and public health measures, and had potentially contributed to the seedling of further transmission and disease propagation in the community. Commissioned by the Hong Kong Government during the catastrophic period, our team established and maintained an ad-hoc large-scale participatory daily antigen rapid testing surveillance (DARTS) system for real-time situational awareness of SARS-CoV-2 activity to inform policy consideration in a timely manner [19]. Established as an emergency ad-hoc public health initiative, the RAT-based disease surveillance system was aimed to provide a continuous, stable, and unbiased data for real-time situational assessment over the course of the SARS-CoV-2 epidemic in the local community in Hong Kong. Since its launch on March 3, 2022, the system has helped to track through the rapidly changing trajectory of the COVID-19 pandemic over its fifth and sixth wave of community activities in Hong Kong.

5.2 Practical considerations & challenges for implementation

5.2.1 Platform design

A suitable design of the surveillance platform should be chosen with a view to provide a robust results for reflecting the epidemic activity [20]. The collection of surveillance data for multiple time points generally adopts two common designs of observational study, serial cross-sectional and longitudinal cohort study [21].

Serial cross-sectional survey of samples drawn from the populations at multiple time points, though may be a practically and logistically simpler way to reflect the disease trajectory in the community, is subject to common problems of representativeness and comparability issues across samples involving different participants and with changing ascertainment practice over time [22, 23].

On the other hand, although cohort-based surveillance systems were logistically more challenging to build-up and maintain, the resultant representative cohort and the stable reporting behavior would be valuable in reflecting the temporal trajectory of an epidemic. The platform may also offer an opportunity for more detailed data to be collected for exploring various epidemiologic characteristics of the disease.

5.2.2 Recruitment

The lag-time required for the initial system setup of any ad-hoc surveillance system during the midst of an evolving and worsening epidemic is a crucial limiting factor on its ultimate utility. A simple and efficient recruitment approach should be adopted as far as possible to facilitate a speedy recruitment. A sufficient sample size would help to improve the precision and granularity of the data to allow for stratified monitoring of populations subgroups as defined by important demographic factors and geographical regions. On the other hand, although many surveillance systems are basing on convenience sample for the relative ease of recruitment, effort should always be paid to make sure the adopted recruitment approach, framework, and setting can help to enhance the representativeness of the sample to the general population, both in the beginning and over the course of the surveillance, or else a very biased intelligence may be resulted.

In our platform, a cohort of >10,000 individuals were recruited primarily by random digit dialing, which allowed the recruitment to be completed quickly and the system to be swiftly set up within a week. Representativeness of the cohort for the local population was ensured by adopting a pre-defined age-stratified quota, which was also statistically weighted in the calculation of the daily point prevalence. Dropouts and non-complying individuals were replenished by continuing recruitment to maintain the sample number and prevent surveillance fatigue.

5.2.3 The reporting procedure and details of information to be collected

As any concern regarding the workload and data privacy commitment would inevitably deter enthusiasm for participation, potential trade-off between the required data detail and hesitancy for participation should also be carefully considered. A realistic balance should be strived between any unnecessary obsession collecting a large amount of personal data against the potential impact on participation and compliance. In our system, besides the collection of basic demographic information and COVID-19 vaccination status in baseline, only a simple reporting of symptoms and the testing result was required on a weekly basis, which helped to minimize the excessive workload for result submission and privacy implication on participants.

5.2.4 Selection of test kit and quality consideration

RAT products of different assay brands may have very different technical complexity and variable sensitivity. To ensure a reasonable testing performance, the adopted RAT test kits should be meeting the World Health Organization’s (WHO) priority target product profiles for COVID-19 diagnostics (i.e. sensitivity ≥80% and specificity ≥97%) and the United States Food and Drug Administration (FDA) Emergency Use Authorization (EUA). In our system, RAT kits of a single brand meeting the standard and with stable supply were chosen and provided to all participants to ensure the testing performance by minimizing potential misclassification as a result of suboptimal sensitivity and specificity.

5.2.5 Supply and distribution of test kits

During the midst of an evolving pandemic, accessibility of RATs can be a real problem. In our system, RATs for SAR-CoV-2 antigen were supplied to all participants on their consent to take part in the regular surveillance programme. Telephone hotlines were operated to address any enquiry and request for assistance.

Depending on the particular settings and scenario, a number of specific issues may need to be taken into consideration if testing kits were to be centrally supplied to participating individuals willing to take part in the surveillance to facilitate a rapid kick-starting of the programme. These issues may include the availability and stability of supply from the vendor, repacking needs subjecting to whether the supplied kits were originally in individual or bulk packing, and the logistical issues of test kit delivery to a large number of participants; each of those can become a non-trivial challenge given the community lockdown and/or infrastructure breakdown during the midst of an evolving epidemic.

5.2.6 Scheduling of testing to achieve regular data

A stable reporting behavior is the cornerstone of any useful surveillance data, as spurious temporal pattern can easily be resulted on changing reporting intensity as a result of surveillance fatigue or overzealous reporting. In our system, participants were being divided into seven cohorts of approximately equal size, with individuals in each cohort scheduled to perform the RAT on a designated day of the week irrespective of symptom or exposure history, so as to achieve a rolling testing schedule of around 1400 individuals on a daily basis.

5.2.7 Sampling and testing technique

In our system, RAT was performed by each participant, or sampled by parents for minors, with a self-sampled combined throat and nasal swab. Although the test is designed primarily for self-usage, our experience indicated that some people may still find it difficult to use and interpret, and the textual guidance provided in the pamphlet is difficult to understand and follow. In addition, there is also a potential problem of wrongly interpreting the testing results, especially among populations with advanced age and lower educational level. Simple instruction for the sampling and proper conduction of the self-tests, either for oneself or for minors, was therefore provided by us in both infographic format and step-by-step instructional video-clips specially recorded in-house by our team. Telephone hotlines were also operated to address enquiry and request for assistance from participants. Most individuals (99.22%) reported that they managed to successfully conduct the RAT testing independently under the video guidance.

5.2.8 Reporting platform

An in-house designed online platform was established to streamline direct enrolment and efficient registration of interested participants to the real-time surveillance system in a user-friendly manner. This online platform was also used for participants to report their demographic data initially and for the reporting of testing results and photos throughout the surveillance period through simple reporting steps. An individualized, password-protected account was assigned to each participating household for submitting their individual health records in a secured manner. To prevent any false positive due to reporting error, self-reported RAT positive results were validated with the RAT photos submitted on a daily basis.

5.2.9 Follow-up and cohort maintenance

Depending on the length of the targeted surveillance, proper maintenance of the surveillance cohort may be important to prevent surveillance fatigue when participants were losing interest or impetus on the regular testing and reporting activity, especially when their self-perceived risk was lowered with any apparent improvement in community epidemic activity. In our experience, following-up of participants regularly through phone calls, SMS, and WhatsApp was opined by the majority of our participants (98.96%) as useful for reminding them to complete the testing and reporting in a regular and timely manner. On the other hand, the operation of a responsive study hotlines would also be useful for addressing any problems that may be encountered by participants during the surveillance period.

5.2.10 Prospective data analysis

For many surveillance systems, including most seasonal influenza sentinel surveillance systems, analyzing on a weekly basis is generally good enough. Weekly or monthly data analysis with aggregated data across time would help to reduce data randomness and cyclical variability like weekend effects to give a more stable estimate over time. However, when more frequent data analysis (e.g. daily) is needed to inform a timely risk assessment and intelligence, the issue of data randomness may become a non-trivial. Data smoothing approaches, including random walk, simple moving averages, [24] exponential smoothing algorithm, or other algorithms [25], would help to remove noise due to random changes or seasonality, thus would help to avoid false alert and facilitate the identification real temporal trends of changing disease activity. The longer duration of data to be aggregated in the moving average, the more smoothed the data becomes, but may reduce the sensitivity and timeliness of noticing an increase in disease activity.

Depending on the types of surveillance data collected and the surveillance purposes, either the incidence or prevalence may be used. Incidence refers to the occurrence of new cases of disease or a health event in a population over a specified period of time, while prevalence refers to the proportion of a population who have an infection or a health event, including newly occurring cases and existing cases, in each specific time point or period. A repeated community-wide sampling without elucidating the information to differentiate new or existing infection would give an estimate of prevalence. Our system reported a simple daily point prevalence, estimated by dividing the number of individuals reported positive results by the number of individuals submitted valid test results on the same day. When allowed by available and collected demographic data, stratified analysis by gender, age group, geographical region, and socioeconomic status would help to assess the changing disease activity across different population subgroups and helps to assess potential differential risk and health care burden over an evolving epidemic. Longitudinal pattern of the changing infection prevalence over time would then allow the situational assessment of the changing epidemic situation, or calculating the effective reproductive number (Rt) to assess how quickly the virus is spreading in the community.

5.2.11 Feedback of surveillance intelligence

Timely and efficient feeding back of the surveillance intelligence in the most accessible manner is of utmost importance for informing any relevant community stakeholders and the general public regarding the evolving trajectory of the pandemic to facilitate appropriate downstream public health planning, decision-making, and action. In our system, the daily point prevalence of COVID-19 infection was disseminated on a real-time basis through an open online digital dashboard (https://covid19.sph.hku.hk/dashboard) to the general public and any potentially relevant stakeholders.

5.2.12 System evaluation

Evaluation of the system after it was up and running after a suitable period or stage would be important to assess how effective and efficient the system is, and to inform areas needing improvement. The Donabedian framework originally developed by the University of Michigan, involving an assessment of three categories: the “structure,” “process,” and “outcomes”, represents a generic and useful framework for evaluating quality of health services and health care programmes [26]. In relation to a surveillance system, structure describes the context in which the system is built, process denotes the implementation and delivery of the surveillance steps, and outcomes refer to the effects of the surveillance intelligence on the understanding and handling of the evolving pandemic.

More specific, a more detailed evaluation of the nine key attributes of a surveillance system as listed below can be done using the CDC guidelines, which should be taken into consideration during both the planning and evaluation of any public health surveillance systems [27].

• Simplicity refers to the system’s structure and ease of operation.

• Flexibility is the ability of the system to adapt to changing information needs and operating conditions with minimal additional cost.

• Data quality is the completeness and validity of the data collected through the system.

• Acceptability is the willingness of persons and organizations to participate in the system, including those who operate the system, report cases of the disease, or use the data.

• Sensitivity is the proportion of cases of a disease detected by a surveillance system and the ability of the system to monitor changes in the number of cases over time, such as outbreaks.

• Predictive value positive is the proportion of cases reported through the system that are accurately diagnosed instances of the disease under surveillance.

• Representativeness is the extent to which the system accurately describes the occurrence of the disease over time and its distribution in the population by place and person.

• Timeliness reflects the delay between steps in a surveillance system and availability of information for control of the disease under surveillance when needed.

• Stability is the ability of a surveillance system to collect, manage, and provide data without failure and to be operational when needed.

Evaluation of the surveillance system after landmark stages is important and can inform the way forward in relation to the evolving epidemic. Our evaluation based on the Donabedian framework and CDC guideline had revealed the surveillance initiative to be a representative, stable, and timely surveillance system with high data quality and acceptability.

5.3 Conclusions

The successful establishment and maintenance of the DARTS system during the SARS-CoV-2 pandemic in Hong Kong demonstrated that results of regular self-performed RAT in the community can be used to inform situational awareness of the trajectory of an evolving epidemic as an ad-hoc participatory surveillance system. Our experience also demonstrated that it is logistically and technically feasible for establishing such a large-scale ad-hoc surveillance platform in a timely manner during an emergency, including the recruitment, follow-up, and maintenance of a sizably large number of representative surveillance participants with good compliance. The regular non-symptom and risk-based testing approach helped to give a more representative picture of disease activity of all severity spectrum, including subclinical cases who still carried an implication of downstream transmission. The use of RAT instead of PCR has helped to avoid the constraint of manpower and testing capacity, which the government has also quickly adopted for case definition.

Acknowledgments

This project was supported by the Henry Fok Foundation and a special commissioned fund (reference number COVID-19FHB, COVID-19HHB) from the Health Bureau of the Hong Kong Special Administrative Region Government. We acknowledge research support by Jerome Leung, Grace Ng, Oscar Kwok, Christine Ning, Kenny Mok, Clarice Guo, and Kaki Chen.

Conflict of interest

The authors declare no conflict of interest.