Monitoring respiratory infections in covid-19 epidemics

Unique challenges in understanding covid-19 epidemiology

Compartmental modelling has been crucial in understanding the covid-19 pandemic, estimating both the current scale (“nowcasting”) and its development over time (“forecasting”) to inform the epidemic response worldwide. But many of the assumptions made (such as probability of transmission per contact) are based on expert opinion because empirical evidence is often not available.5 Thus, a small error can be amplified into a large bias that can be consequential. The model that was used to inform the 2001 foot and mouth disease epidemic, which led to the culling of 1.2 million British livestock, was later found to have markedly overestimated the disease impact and underestimated the effectiveness of conventional public health intervention.6

Modelling for covid-19 faces several unique challenges. First, the proportion of undetected, mild, or asymptomatic cases, which might account for a substantial proportion of community transmission, remains largely uncertain.3 Second, epidemic curves from different populations are not comparable and do not necessarily approximate the underlying transmission dynamics. Third, the fatality rate of covid-19 varies dramatically from about 1% in Germany to 12% in Italy, warranting extra caution when inferring the actual number of infected people based on the number of deaths and fatality rates.1

A simple, rapid, and cost effective approach

Influenza (or influenza-like illness (ILI), defined by the World Health Organization as an acute respiratory illness with a measured temperature of ≥38°C and cough, with onset within the past 10 days) has been routinely monitored in 114 countries across five continents, coordinated by the Global Influenza Surveillance and Response System since 1952.7 Although high income countries tend to have stronger influenza surveillance, over the past few decades robust electronic surveillance systems have been established in Africa (13 countries), Latin America (23 countries), and the Middle East (14 countries).7

By monitoring the transmission patterns of ILI, we can approximate the overall performance of a population’s general infection control measures, including enhancement of personal and environmental hygiene and social distancing. Other measures that are specific to covid-19, such as travel restrictions, quarantine, and case isolation, are unlikely to alter local ILI patterns substantially. Unlike covid-19, which has a long incubation period of up to 16 days,8 ILIs (incubation period 1-2 days)9 are likely to be more sensitive to infection control measures. This allows a much more timely understanding of the changing transmission intensity of droplet and fomite borne disease, which includes covid-19.

Drastic reductions in various viral respiratory infections were also noted in Hong Kong during the SARS epidemic in 2003.10 Nonetheless, changes in healthcare seeking behaviours and disease surveillance practices, especially during stringent “lockdown” measures, must be carefully considered. The influence of behaviour and practice changes could be accounted for by examining ILI detection rates (which account for changes in the total number of tests or cases examined) and severe cases, deaths, or institutional outbreaks that are less likely to be confounded.

Epidemiology of influenza and other respiratory infections

Using published surveillance data, we compared the length and severity of the 2019-20 winter influenza season in Hong Kong with those in 2015-19.12 An influenza season in Hong Kong is the period in which the positive rate of influenza cases in respiratory tract specimens tested in the public health laboratory and the hospital admission rate for influenza are above the usual baseline (9.2% for specimens positive rate and 2.5 per 100 000 population for hospital admission rate; estimated as 1.96 standard deviation above the average weekly numbers during non-season periods from 2014 week 49 to 2019 week 48).12 We also compared the positive test rates of common respiratory infections in Hong Kong before and after week 5 of 2020, when most infection control measures began.

Compared to 2015-19, the 2019-20 winter influenza season was 63.2% shorter, the number of institutional (schools or homes for elderly people) ILI outbreaks was 68.4% lower, and the number of deaths from laboratory confirmed influenza in adults was 62.3% lower (table 1). Although the 2020 season began at a similar time (week 2) as the previous winter (week 1), there was a sharp decline in laboratory confirmed influenza cases from >18% in weeks 3-4 to <1% in week 8 and remained low (0.1%) into week 15, while the total number of respiratory specimens tested remained steady (~6000 a week) (fig 2, table 1). Similar trends have been observed for other respiratory infections in the same period (fig 2).

Fig 2

Time series of percentages of different pathogens in all respiratory specimens analysed in 2018-20 in Hong Kong.

• Download figure
• Open in new tab
• Download powerpoint

Consistently, the outpatient ILI rate (number of ILI cases per 1000 consultations in the sentinel general outpatient clinics) in Hong Kong started to fall after the first covid-19 case in week 4, and this trend continued with a >80% reduction between weeks 2 and 15 (fig 1).

We also looked at temporal ILI patterns (using population specific parameters) in Southern China (Hubei and 14 other provinces or municipalities, which recorded 92% of all reported covid-19 cases in China as at 18 April), South Korea, Singapore, and England.1213141516 We found similar rapid reductions of ILI transmission from week 4 in Southern China (fig 3) and South Korea (fig 4) and from week 5 in Singapore (fig 5). ILI transmission intensity in Singapore started to rise from week 8 to week 11, and then dropped again until week 15, which is similar to the trend observed in England over the same period (fig 6). Note that the ILI and covid-19 data across populations were only qualitatively compared because of substantial heterogeneity in the choice of indicator for ILI transmission and case identification strategies for covid-19 in different populations. For the same reasons, different scales were used for the y axes of the figures.

Fig 3

Incidence of covid-19 and indicators of transmission intensity of influenza-like illness from week 2 to 16 of 2020 in Southern China (Hubei, Shanghai, Guangdong, Jiangsu, Zhejiang, Anhui, Fujian, Jiangxi, Hunan, Guangxi, Hainan, Chongqing, Sichuan, Guizhou, and Yunnan). Lockdown of Wuhan City and Hubei Province involved shutting down all transportations in and out of the region and either restricting or banning local travel. Data source: Chinese National Influenza Centre.13 The latest data available at the time of writing were dated 17 April.

• Download figure
• Open in new tab
• Download powerpoint

Fig 4

Incidence of covid-19 and indicators of transmission intensity of influenza-like illness from week 2 to 16 of 2020 in South Korea. Data source: Korea Center for Disease Control and Prevention.14 The drastic case surge in weeks 8-10 was primarily driven by “superspreading” clusters, including the Shincheonji Church cluster, in which 5016 members were infected (60% of all cases in South Korea) as at 18 March.17 The latest data available at the time of writing were dated 13 April.

• Download figure
• Open in new tab
• Download powerpoint

Fig 5

Incidence of covid-19 and indicators of transmission intensity of influenza-like illness from week 2 to 16 of 2020 in Singapore. Data source: Singapore Ministry of Health.15 The latest data available at the time of writing were dated 18 April.

• Download figure
• Open in new tab
• Download powerpoint

Fig 6

Incidence of covid-19 and indicators of transmission intensity of influenza-like illness from week 2 to 16 of 2020 in England. The 2019-20 winter influenza season started in week 49 of 2019 and ended in week 3 of 2020. UK nationwide lockdown involved shutting all non-essential shops (except supermarkets, pharmacies, newsagents, vets, petrol stations, hardware shops, banks, laundrettes, and undertakers), banning gatherings of more than two people except for households, mandating all residents to stay at home unless for essential travel (shopping for food or medicine, one type of exercise a day, medical appointments, as a carer visiting a vulnerable person, for essential work). Data source: Public Health England.16 The latest data available at the time of writing were dated 18 April.

• Download figure
• Open in new tab
• Download powerpoint

Limitations

Two major concerns for the proposed approach are that during a pandemic, as for most non-urgent illnesses, people with ILI might stop (or be told to stop) attending clinics, and the number of tests ordered for influenza surveillance might also be reduced. In such circumstances, however, the total consultation counts (the denominator of outpatient ILI rate) also decrease. The covid-19 tests in many countries might, at least partly, compensate for the changes, as people with covid-19 (non-specific respiratory) symptoms who test negative are documented as having ILI. A simple syndromic surveillance system for self-reported covid-19 or ILI symptoms (or both) could augment the proposed approach, with the advantages of being low cost and adaptive as it requires no additional clinical staff or laboratory capacity and could capture a more complete picture of covid-19 and ILI transmission in the community even with changing behaviours in patient and healthcare providers during the epidemic.

Importantly, the ILI data across populations cannot be quantitatively compared due to the substantial heterogeneity in the choice of indicator for ILI transmission, the virology of the specific strains, and vaccination efficiency and uptake rate. The same is true for covid-19, given the widely different case identification systems across populations. The drop in ILI GP consultation rate in England occurred before the local covid-19 outbreak and might not be relevant to this analysis. Overall, the ecological nature of our analysis prevents us from inferring causality or disentangling the independent effect of specific measures, and individual level analysis is needed for further investigation.

Lesson learnt

From the populations examined, we have shown that the transmission intensity of ILI was qualitatively consistent with that of covid-19, showing the potential value of the well established ILI surveillance data. Variations in the second wave in Hong Kong, primarily driven by imported cases, serve as a negative control showing how the approach might tease out the potential effects of general infection control measures. The rapid changes in ILI transmission support our argument that ILI is likely to be more sensitive to infection control measures than covid-19. The analysis of ILI might enable rapid appraisal of the impact of control measures on the transmission of covid-19 without mass serology testing.

The indirect evaluation method capitalising on the more widely available and reliable influenza surveillance data could be a complementary and cost effective way to understand different countries’ performance in controlling covid-19, especially for low and middle income countries with limited resources. Syndromic surveillance would be a valuable addition to the existing Global Influenza Surveillance and Response System and for countries not yet included in this system. We have shown the feasibility of using a simple analytical framework with minimal demand on statistical expertise in populations with little testing capacity for covid-19. We suggest the use of nowcasting and time series analyses for enhancing the value of ILI monitoring during this pandemic.