As COVID-19 cases are still rising around the world and new variants are emerging, nonpharmaceutical interventions (NPIs) are essential for controlling the spread of COVID-19 [1,2] in many countries, especially when vaccination programs are impeded by factors such as vaccine hesitancy, reduced efficacy against some new variants, and vaccine shortage. NPIs such as personal protective equipment, social distancing, contact tracing followed by quarantine, border controls, travel restrictions, and enforced or recommended “stay-at-home” policies and “lockdowns” reduce transmission arising from individual contacts [3].

Hong Kong is a densely populated cosmopolitan city with extensive connections with mainland China and the rest of the world. The city experienced and successfully controlled three waves of COVID-19 in mid-January, March, and July 2020, respectively. Early in the first wave, the government closed its borders with the mainland and enforced a 14-day mandatory quarantine for all arrivals from mainland China and the close contacts of confirmed cases. Quarantine was then extended to arrivals from high-risk countries, and eventually to all arrivals, and entry was denied for all nonresidents during the second wave [1,2]. Unlike the first two waves, during which cases were predominantly imported cases and their close contacts, the third wave was characterized by community transmission in local clusters [3]. In response, the government stepped up public health measures, by implementing mandates for physical distancing, expanded community testing, enhanced case detection, contact tracing, and quarantine [4].

Active identification and isolation of infected persons and quarantine of close contacts reduced subsequent transmissions and brought the third wave under control within two months [3]. The importance and effectiveness of contact tracing have been documented elsewhere [5-8]. According to Aleta et al [5], aggressive social distancing measures, robust testing, contact tracing, and quarantine kept the disease within the capacity of the health care system in the absence of herd immunity.

Ideally, all local cases can be epidemiologically linked, forming a meta-cluster of a closed transmission network in the absence of imported cases. However, in reality, resources for contact tracing are limited, and cases are not able to recollect (ie, recall bias) or choose not to disclose all their close contacts. Close contacts may also not be known to the index case and therefore cannot be traced by conventional methods. As a consequence, some of the cases in a cluster outbreak or in unrecognized chains of transmissions could eventually emerge as “unlinked cases,” either as the index case in a new cluster outbreak or as a sporadic case without known involvement in any identified cluster (Figure 1). As shown by a modelling study [7], unlinked secondary cases can markedly impede outbreak control due to delayed isolation. Another simulation study predicted that at least 71% of close contacts of infectors needed to be traced to control the epidemic (ie, reproduction number reduced to <1) [8].

Hong Kong has implemented stringent containment measures since January 2020 [9,10]. Individuals who contacted a confirmed case up to two days before the initial case’s symptom onset would be put under mandatory quarantine and be tested [11]. Contact tracing became even more important during the third wave, which was characterized by more local transmissions, the impact of which should be evaluated.

It is critical to obtain empirical evidence on the effect of contact tracing on the containment of the epidemic in jurisdictions with control strategies (eg, Singapore, South Korea, and Hong Kong), where intensive contact tracing systems are applied to suppress transmission, to try to avoid the need for extensive restrictions to socioeconomic life (eg, lockdowns). Such data will also be invaluable for informing policies regarding the development of digital tracing technologies that have the potential to minimize the proportion of epidemiologically unlinked cases as well as the perpetuation of outbreaks. In this study, we investigated the clinical and epidemiological characteristics of the unlinked COVID-19 cases in Hong Kong, and estimated the transmission potential of the unlinked cases in different socioeconomic settings. Our study period covers the first three waves of the COVID-19 pandemic in Hong Kong. The findings provide empirical evidence on the value of contact tracing by comparing the epidemiological characteristics of the linked and unlinked cases, and estimating the probability of unlinked cases seeding community outbreaks. We also identified “hot spots” of outbreaks that were triggered by unlinked cases, which can inform risk assessments for designing future public health interventions to control transmission and prevent the health care system from being overwhelmed.

In this study, we characterized the epidemiological profile of local unlinked cases in the first three epidemic waves of COVID-19 in Hong Kong, a Special Administrative Region with a well-established case detection and contact tracing system used as a key containment strategy. Nevertheless, conventional contact tracing is unable to identify all close contacts at risk of infection, resulting in a significant number of unlinked cases, detected either as sporadic cases or as index cases of clusters of infections. We showed that unlinked cases were positively correlated with time to hospital admission, serious/critical clinical condition, and epidemic size. The higher proportion of patients progressing to a serious/critical condition among the unlinked cases was particularly pronounced in the third wave when cases were mostly due to local transmission.

Compared with the linked cases that were quickly identified through contact tracing and could be diagnosed before symptom onset, unlinked cases—for which no information regarding epidemiological linkage had been obtained through conventional contact tracing methods—could only be ascertained when they sought medical consultation upon becoming symptomatic. This could result in a longer delay in receiving appropriate care and treatment, which could result in an increased likelihood of progressing to a serious/critical clinical condition, which is consistent with reports regarding other respiratory infectious diseases [17,18]. However, the overall mortality rate was not found to be significantly different between linked and unlinked cases. One possible reason is that the health care system provides effective clinical management of patients with COVID-19. In Hong Kong, the mortality rate has been relatively low [19] and the health care system still had the capacity to provide optimal care for patients with COVID-19 as the epidemic was brought under control rapidly with physical distancing measures during both epochs of the epidemic.

Another important finding was the association between the unlinked infections and the local epidemic situation. As indicated by Kretzschmar et al [6], the delay in detection and isolation would reduce the effectiveness of contact tracing measures for COVID-19 control. Across the different settings, the probability of an outbreak was highest in households, a phenomenon reported by studies in mainland China [20] and Korea [21]. However, it should be noted that household members are usually less subject to recall bias than others, leading to an overrepresentation of household secondary cases in our data set and an upward bias in the transmission potential for household settings. As shown by Adam et al [13], frequencies of secondary transmission were highest in households, while clusters were largest in social settings like restaurants, where superspreading events occurred. In line with our finding, a previous study showed that eateries and workplaces were the settings more likely to generate transmission cascades across more than a single setting in the second wave of the epidemic [12]. Such enclosed settings where large numbers of people were in close contact were a risk factor for transmission, while the congregation of people not known to each other was a risk factor for not being able to identify close contacts, allowing the dispersion of infections in a wide range of the settings.

Globally, it was estimated that 10% of cases outside China had caused 80% of secondary cases at the start of the pandemic [22]. Superspreading events have driven the pandemic significantly and could quickly overwhelm the contact tracing system; for example, for a single large cluster, the epidemic investigation team traced up to 1000 people in Hong Kong [23]. Although venues like gymnasiums and sport venues were closed in Hong Kong and the government encouraged people to work from home during the pandemic, most businesses were operational. Dining in in restaurants was permitted with restrictions in numbers, operational hours, and capacity. Unsurprisingly, we found a higher probability of cluster transmissions in eateries and workplaces compared to other settings. However, considering the scenario in which sporadic unlinked cases are not likely to have been involved in clusters in health care, shopping, and party settings, these settings would also have high potentials for cluster transmission. Our analysis thus provides a retrospective risk assessment for optimizing the relaxation of social/physical distancing restrictions in different settings; in these settings, risk can be managed with a robust and strictly enforced digital tracing system. Resurgences of infections could be prevented, obviating the need for a complete lockdown and the associated social and economic costs.

Many studies have focused on identifying hot spots for superspreading events. Although indoor environments with poor ventilation are generally riskier, human activities and behaviors (whether people have taken precautions) are even more crucial [12,24]. Therefore, health care, shopping, and party settings should have strictly implemented precautions for preventing cluster transmission, because of the risk and large number of contacts associated with such settings. Our findings offer empirical evidence for the differential influence of unlinked cases on epidemic control, which could help prioritize mitigation strategies.

Many jurisdictions in the Western Pacific Region heavily relied on stringent case detection, contact tracing, isolation, quarantine, and intensive testing programs together with border restrictions for COVID-19 control [19]. However, the conventional contact tracing approach is labor-intensive and feasible for low-incidence settings only [7], capturing only known contacts and being subject to cases’ recall bias. Mobile phone data are a promising complementary tool. First, mobility data can inform incidence forecasts—for example, modeling combined with local mobility data showed that restaurants, bars, and gyms were hot spots for transmission [25]. Second, apps can be used to alert users instantaneously if their contacts are confirmed positive and recorded in the system [26,27]. This would prompt the alerted users to seek testing, especially for presymptomatic or asymptomatic transmissions [28-31]. In this study, we found that the percentage of asymptomatic infections was significantly lower in the unlinked cases than in the linked cases, primarily due to the fact that the unlinked cases would only be identified when they became symptomatic and sought medical consultation. However, even though contact tracing can identify some asymptomatic cases, others are missed by the system. App-based tools could prove useful for tracing these infections by providing improved coverage and timeliness and have the potential to address the challenge of anonymous close contacts that the index patients do not know. Digital contact tracing has been deployed in many countries [32], including India [33], Switzerland [26], and the United Kingdom [27]. Nevertheless, we acknowledge that the effectiveness of app-based monitoring tools on outbreak control remains controversial when variations in exposure relative to the infectiveness period, testing accuracy, isolation adherence, and coverage of the app-based technology are taken into account [34,35]. Most importantly, the use of smartphone data for digital contact tracing in such contexts raises a number of ethical and privacy concerns [36]. In Hong Kong, for example, residents were skeptical of the LeaveHomeSafe app implemented after the third wave of the epidemic, despite the government’s reassurance that user registration was not required and the check-in data would not be uploaded to the government’s system or any other systems [37,38]. The critical role of digital technology in complementing conventional contact tracing needs to be evaluated, and ethical and privacy concerns must be addressed.

One major limitation of this study is that the location history for the sporadic unlinked cases was unavailable. We estimated the transmission probability of the unlinked cases by considering a range of scenarios that varied by the proportion of sporadic unlinked cases. In addition, we did not quantify the frequency of outbreaks (eg, the number of clusters per 1000 unlinked cases), which would be of interest for decision-making due to the unknown number of sporadic unlinked cases in each setting. Instead, we provided the probability that an unlinked case could result in secondary transmission to convey the outbreak potential in that setting. This indicates a need for an exposure tracking system that automatically computes the risk level for different settings. Second, cases were linked epidemiologically without validation through other approaches such as the routine use of phylogenetic analysis due to a lack of viral sequencing data [39]. Fingerprinting of SARS-CoV-2 via phylogenetic information can assist in the identification of chains of transmission; therefore, such analyses should be performed in future investigations. Third, the retrospective data cannot infer transmission potential for some settings that were already closed during the study period (eg, kindergartens that were closed since the early phase of the first wave). Fourth, some of the unlinked cases may have been infected by the imported cases exempted from quarantine and testing [40], limiting the generalizability of our results to other places with different border control measures.

In conclusion, our findings suggest a need to promote the use of digital tracing methods on top of current conventional testing and tracing. Contact tracing measures can potentially shorten the time to admission, reduce serious cases, and prevent further spread. With the considerable probability of secondary transmission from the unlinked cases, digital tracing measures should be strictly enforced in high-risk social settings and the local epidemic should be closely monitored such that the measures can be adjusted in a timely manner.