Investigation of the Impact of Childhood Immune Imprinting on Birth Year-Specific Risk of Clinical Infection During Influenza A Virus Epidemics in Hong Kong Free

Abstract

Influenza imprinting reduces risks of influenza A virus clinical infection by 40%–90%, estimated from surveillance data in western countries. We analyzed surveillance data from 2010 to 2019 in Hong Kong. Based on the best model, which included hemagglutinin group-level imprinting, we estimated that individuals imprinted to H1N1 or H2N2 had a 17% (95% confidence interval [CI], 3%–28%) lower risk of H1N1 clinical infection, and individuals imprinted to H3N2 would have 12% (95% CI, −3% to 26%) lower risk of H3N2 clinical infection. These estimated imprinting protections were weaker than estimates in western countries. Identifying factors affecting imprinting protections is important for control policies and disease modeling.

Imprinting is defined as a lifelong protection against the infection or disease caused by the strains encountered in childhood [1]. Recent evidence for imprinting shows imprinting shapes epidemiological risk [1–3]. Birth cohorts (groups of people born in the same year, and putatively exposed to similar strains in childhood) show stronger protection against influenza viruses that are antigenically similar to those from childhood. Birth-year–specific imprinting is associated with protection against hospitalization and death from avian influenza viruses of the same hemagglutinin (HA) group [3]. HA group 1 contains human seasonal subtypes H1, H2, and avian H5, whereas group 2 contains seasonal H3 and avian H7. Imprinting is also associated with protection against clinical infections with seasonal influenza A viruses of the same subtype [1, 2], and seasonal influenza B viruses of the same lineage [4]. One previous study in Hong Kong estimated that there was lower imprinting protection against subclinical H1N1 and H3N2 infection [5]. It is not clear whether this weaker estimated protection is due to the study's more stringent protective end point (subclinical, not clinical, infection), or due to other factors such as subtropical influenza seasonality, age-specific differences in case ascertainment, and geographic or season-specific effects affecting the estimated strength of imprinting. Here, we analyzed population-level data from Hong Kong to determine the degree of protection from imprinting. We analyzed surveillance data on medically attended influenza cases, by using the same approach as Gostic et al [1].

METHODS

Study Design

Population and Surveillance Data

Hong Kong population data by age were obtained from the website of Census and Statistics Department of Hong Kong. Influenza activity in the general community was monitored through a sentinel surveillance network in outpatient clinics, which report the number of patients with influenza-like illness (ILI) defined as a fever >37.8°C plus a cough or sore throat. Patients were classified into 5 age groups: < 5, 5–24, 25–44, 45–64, and 65+ years. The number of ILI patients by age groups were reported weekly. The public health laboratory also collects data on the weekly proportion of specimens from sentinel outpatient clinics and local hospitals that tested positive for influenza virus.

Circulating Subtypes in Previous Years

The information of circulating types from 1918 to 2019 were summarized following the approach in Ranjeva et al [5]. Prior to 1968, annual subtype frequencies were specified by well-known durations of subtype circulation between historical pandemics [6]. Between 1968 and 1997, the annual fraction of subtype-specific influenza A sequences in the Global Initiative on Sharing All Influenza Data (GISAID) database were used as annual subtype frequencies [7]. After 1998, influenza surveillance data in Hong Kong were used to compute the annual subtype frequencies [8]. This information was used to impute the subtype of primary influenza virus infection.

Statistical Methods

Estimating the imprinting effect required calculation of the fraction of each birth cohort with the first influenza A virus infection that was seasonal subtypes H1N1, H2N2, or H3N2, and the fraction that remained naive for each year. Therefore, we followed the approach in Gostic et al [3] to estimate these fractions, based on a previous age-seroprevalence study from Netherlands [9], and the proportion of infections that were H1N1, H2N2, or H3N2 at each year informed by global surveillance data (Supplementary Material 1).

For each week, we computed the number of cases of an influenza subtype for an age group by multiplying the number of ILI patients in that age group and the proportion of positive specimens for that influenza subtype [8, 10]. We used this approach to compute the case numbers of H1N1 (the 2009 pandemic strain) and H3N2 for age groups < 5, 5–24, 25–44, 45–64, and 65+ years from 2010 to 2019. In 2009, special Designated Flu Clinics, which were not part of the sentinel network, were established temporarily as a response to the H1N1 pandemic. Therefore, the surveillance data in 2009 may not be comparable to other years and was excluded from the analysis [11].

Then, we modified the approach [3] to estimate the imprinting effect by fitting models with assumed imprinting mechanism, including HA group-level imprinting, HA subtype-level imprinting, or neuraminidase (NA) subtype-level imprinting (Supplementary Table 1). We fitted multinomial distributions to the case distribution by age group, because we only observed the age groups instead of the exact birth years of cases (Supplementary Material 2). We extended the model to include influenza vaccination, with assumption that vaccine effectiveness was 50% [12], and the vaccine coverage informed by government press release (Supplementary Figure 1). Because the analysis is sensitive to age-specific differences in case ascertainment, as some individuals with ILI or acute respiratory illness (ARI) may not seek medical help, we conducted a sensitivity analysis in which observed case counts were scaled up based on the age-specific probability of health care seeking from a survey in Hong Kong [13]. Statistical analyses were conducted using R version 4.0.5 (R Foundation for Statistical Computing).

RESULTS

From 2010 to 2019, we estimated that there were 3801 H1N1 cases and 7902 H3N2 cases (Supplementary Table 2). We reconstructed birth-year–specific probabilities of childhood imprinting to H1N1, H2N2, and H3N2 for cases ascertained from 2010 to 2019 (Supplementary Figure 2 as illustration).

We fitted 4 models with age groups, vaccination, and different assumptions on the mechanism of imprinting (Table 1), including HA group-level imprinting, HA subtype-level imprinting, NA subtype-level imprinting, and no imprinting. We found that the model included HA group-level imprinting was the best model. The model with HA subtype-level imprinting was comparable (ΔAIC = 2.4), and these 2 models fitted the surveillance data better than the model with NA subtype-level imprinting (ΔAIC = 8.4) and without any imprinting (ΔAIC = 19.8).

In the best model, we estimated the individuals imprinted to HA group 1 would have 17% (95% confidence interval [CI], 3%–28%) lower risk of H1N1 clinical infection, and the individuals imprinted to HA group 2 would have 12% (CI, −3% to 26%) lower risk of H3N2 clinical infection, compared to those without such imprinting. We estimated that the risk of clinical infection for individuals aged 0–4 years was significantly higher than other age groups. The model-predicted age group distribution from 2010 to 2019 was similar to the observed distribution (Supplementary Figure 3), suggesting the model fit was adequate.

In the sensitivity analyses in which the models were fitted to data scaled up by probability of seeking medical help in the presence of ILI or ARI (Supplementary Tables 3–5), we found that the best model was still the model with HA group-level imprinting in both scenarios. The estimated protection from imprinting was also similar (Supplementary Tables 6 and 7).

DISCUSSION

tIn this study, we explored the potential of imprinting effect against clinical infection based on surveillance data. By analyzing the age group distribution of H1N1 and H3N2 in the surveillance of clinically attended influenza cases, we found that a statistical model with HA group-level imprinting was the best model to explain the difference in age distribution among subtypes. A previous study [5] suggested that the imprinting effect against subclinical infection was limited, based on a household study in Hong Kong. Here, we also found that the imprinting effect against clinically attended influenza cases was weaker than other regions such as the United States [1, 2] for influenza A, and the imprinting effect against influenza B infection based on data for New Zealand [4]. Such difference may be attributed to the seasonality, in which there were usually 2 influenza seasons in Hong Kong, compared with a single season in other regions. Compared to individuals in areas with a single annual influenza season, individuals in Hong Kong may have more exposure on average, and therefore at a point in a person’s life, the probability that the person is protected from recent infections is higher on average. In this case, the short-term protection from recent infection may play a more important role than long-term imprinting protection. Other factors may also contribute to the weaker imprinting effects, including differences in health care seeking behavior and age-specific case ascertainment between Hong Kong and other countries, seasonal and geographic differences in the antigenic properties, diversity of circulating strains, or the accuracy of imprinting probabilities due to data availability.

It should be noted that we found the models with HA subtype-level imprinting and with HA group-level imprinting were comparable. This may be due to the low number of participants in the 1957–1967 cohort, which was the only difference between these models. One study also supported the imprinting effect for influenza B virus clinical infection for B/Victoria and B/Yamagata [4], but one previous analysis supported the NA subtype-level imprinting rather than HA group imprinting, although both our results and the study suggested that models with no imprinting performed substantially worse [1]. In reality, it may be possible that the protection of imprinting was from a combination of HA imprinting and NA imprinting. However, it may not be possible to explore this because there was limited diversity of subtypes circulating in the human population. Also, due to collinearity, by using models with surveillance data, it may not be possible to accurately identify the mechanism of imprinting [1].

Our study had several limitations. First, because the exact age of cases in the surveillance was unknown, we had to conduct an age-group–specific analysis, which may have lowered the statistical power compared with other studies with such information. Consequently, we were unable to account for the protection from maternal immunity in the first 3–6 months of life in the calculation of imprinting probabilities, which may blur estimates of imprinting protection. Second, the weekly numbers of influenza cases for a subtype were computed by multiplying the number of ILI cases in surveillance by the proportion of specimens positive for that subtype, which may be inaccurate. Finally, there may have been a change in surveillance over the 10 years that impacted our results, although we accounted for change in vaccine coverage.

In conclusion, we found evidence to support HA group and subtype imprinting protection against clinical influenza infection from surveillance data. The degree of protection in Hong Kong was lower than in western countries, suggesting there could be other factors affecting imprinting protections.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. The authors thank Dr Qiqi Zhang for providing health seeking behavior data from their previous study, and Xiaotong Huang for technical assistance.

Financial support. This work was supported by the Hong Kong Government Theme-based Research Scheme (grant number T11–712/19N) and the Health and Medical Research Fund, Food and Health Bureau (grant number 20190542); and the National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services (grant number 75N93021C00015).

References

Author notes