Heterogeneity in influenza seasonality and vaccine effectiveness in Australia, Chile, New Zealand and South Africa: early estimates of the 2019 influenza season

We compared 2019 influenza seasonality and vaccine effectiveness (VE) in four southern hemisphere countries: Australia, Chile, New Zealand and South Africa. Influenza seasons differed in timing, duration, intensity and predominant circulating viruses. VE estimates were also heterogeneous, with all-ages point estimates ranging from 7–70% (I2: 33%) for A(H1N1)pdm09, 4–57% (I2: 49%) for A(H3N2) and 29–66% (I2: 0%) for B. Caution should be applied when attempting to use southern hemisphere data to predict the northern hemisphere influenza season.

We compared 2019 influenza seasonality and vaccine effectiveness (VE) in four southern hemisphere countries: Australia, Chile, New Zealand and South Africa. Influenza seasons differed in timing, duration, intensity and predominant circulating viruses. VE estimates were also heterogeneous, with all-ages point estimates ranging from 7-70% (I 2 : 33%) for A(H1N1) pdm09, 4-57% (I 2 : 49%) for A(H3N2) and 29-66% (I 2 : 0%) for B. Caution should be applied when attempting to use southern hemisphere data to predict the northern hemisphere influenza season.
In Australia, Chile, New Zealand and South Africa, sentinel surveillance is conducted in primary care and/or hospitals to monitor the timing, intensity and impact of influenza seasons, and to estimate influenza vaccine effectiveness (VE). While the influenza epidemics of these four southern hemisphere countries often coincide, the type of epidemic experienced can vary. Nevertheless, the influenza season experienced in southern hemisphere countries has sometimes been interpreted as a forewarning to the northern hemisphere [1]. Here, we describe the heterogeneity experienced during the 2019 influenza season in these four countries and provide early VE estimates.

Influenza surveillance systems
The sentinel surveillance systems used in this analysis are described in detail in the Table. For Australia, influenza-like illness (ILI) surveillance data came from the Australian Sentinel Practices Research Network (ASPREN), supplemented by the Victorian Sentinel Practice Influenza Network (VicSPIN) [2]. Hospital surveillance data were obtained from the Influenza Complications Alert Network (FluCAN) [3]. In Chile, severe acute respiratory infection (SARI) sentinel surveillance included seven sentinel hospitals distributed across six of 16 administrative regions [4]. In New Zealand, ILI surveillance leverages general practiceregistered patients in all 20 district health boards, ca 540,000, while SARI surveillance includes four public hospitals in Auckland and Counties Manukau District Health Boards [5]. Syndromic surveillance data from South Africa came from outpatient presentations to a large private healthcare provider network, based on International Classification of Diseases (ICD-10) codes for pneumonia and influenza (J9-J11) [6,7]. Virological surveillance in South Africa was conducted through the Viral Watch network [8].

Seasonality
Weekly 2019 influenza activity rates, e.g. ILI consultations per week, were plotted against the mean weekly rate for influenza seasons from 2013 to 2018. All rates were smoothed using a 3-week moving average. The moving epidemic method (MEM) package [9] in R software version 3.6.1 (R Foundation, Vienna, Austria) was used for calculating means and seasonal thresholds using default values to show the onset and intensity of the season ( Figure 1A). The specifications used for the MEM may differ from published national surveillance reports. The onset and peak of the influenza season was at least 5 weeks early in Australia and 1 to 2 weeks early in Chile, New Zealand and South Africa. Activity was well above expected levels in South Africa and very high in Chile, but only reached moderate levels in Australia or New Zealand. The seasons experienced in Chile and South Africa were also much shorter in duration than in Australia and New Zealand.

Virological data
Virological data are shown in Figure 1B

Vaccine effectiveness estimation
The virological data depicted in Figure 1B formed the basis for VE estimation. All systems followed a testnegative design, where the odds ratio (OR) comparing the odds of vaccination among test-positive cases vs test-negative controls was used to derive VE, i.e. VE = (1−OR adj )×100% [10]. Estimates were made separately for each country, virus and age group, incorporating covariates considered important by each site (Figure 2). The heterogeneity among estimates within each virus/age group combination was measured by I 2 and τ 2 [11]. All networks were able to provide data for the A (H3N2

Discussion
We have shown that within countries of the southern hemisphere, the timing, duration and intensity of the influenza seasons, the predominant circulating viruses, and VE all varied in the 2019 influenza season, even between neighbouring countries such as Australia and New Zealand. Similar observations have been reported from Europe [9]. Thus, it appears that activity in one country is not indicative of activity in another country, even when influenza seasons are contemporaneous. The early VE estimates for the 2019 influenza season in the southern hemisphere presented here were highest for influenza A(H1N1)pdm09 and lowest for Influenza detections by type and subtype (B) for patients enrolled in hospital and primary care surveillance for VE estimation. The data used in vaccine effectiveness estimation are a subset restricted to those patients with complete information and recruited within the weeks used for estimation (Table).

Figure 2
Early vaccine effectiveness estimates against influenza A(H1N1)pdm09, A(H3N2) and B by age group and setting,    A(H3N2). Early estimates often approximate final estimates [12]. However, the utility of these estimates for the northern hemisphere may be limited because the 2019 southern hemisphere vaccine differed from the 2019/20 northern hemisphere formulation in three of four components, A(H1N1)pdm09, A(H3N2) and B/ Victoria. Nevertheless, these estimates or earlier versions of them were included with other data reviewed at the WHO Consultation and Information Meeting on the Composition of Influenza Virus Vaccines for Use in the 2020 Southern Hemisphere Influenza Season during 23-26 September 2019 in Geneva and provided a general impression of the performance of the 2019 vaccine.

Setting
While heterogeneity in our VE estimates did not exceed an I 2 of 60%, with so few studies, the sensitivity of statistical tests to detect heterogeneity is probably limited. This is exemplified by the I 2 of 0% for influenza B estimates among adults despite differences in VE point estimates of 75 percentage points ( Figure 2). Thus, low heterogeneity statistics do not alleviate concerns about how to interpret discrepant VE point estimates.
There are many potential sources for this heterogeneity that affect not only the VE estimates, but interpretation of weekly activity rates. First, with random sampling, we should not expect estimates to be the same [13]. Second, when samples are small they may be vulnerable to statistical biases, such as sparse data bias, and bias due to measurement errors may be more profound [14]. Third, there were many differences in study design (Table). Case ascertainment differed; for example, a SARI case definition was used in New Zealand and Chile, but not in Australian hospital surveillance. Exposure ascertainment also differed, with varying availability of registries to verify vaccination status and the use of different vaccines. In particular, the adjuvanted vaccines used among Australians ≥ 65 years of age might be expected to yield higher VE than standard vaccines [15]. Fourth, vaccine coverage varied (Table). Low vaccination coverage, as observed in South Africa, affects power and precision and can exacerbate the bias induced by measurement errors. Higher coverage, as seen in Chile and among elderly patients in New Zealand and Australia, may mean that many more people in the sample are repeat vaccinees. Repeat vaccination may negatively impact VE and could result in lower VE estimates in highly vaccinated populations [16]. Finally, although only limited virological data were available, we observed differences in circulating A(H3N2) virus clades and B lineages. This may impact both seasonality and VE, particularly as most A(H3N2) viruses sequenced appeared to be in different clades from the vaccine virus (3C.2a2). Notably, most A(H3N2) viruses were also in different genetic groups from the 2019/20 northern hemisphere vaccine (3C.3a).
In conclusion, we have attempted to briefly summarise and interpret the 2019 influenza season in four southern hemisphere countries and have presented early VE estimates. We observed substantial variation in available data on influenza seasonality and VE within the southern hemisphere in 2019, which is unsurprising given the many differences in surveillance among these countries. Caution should be applied when attempting to infer the impending northern hemisphere influenza season based on these observations.

Ethical statements
Australia: Data were collected, used and reported under the legislative authorisation of the Australian state and territory legislation, and thus did not require Human Research Ethics Committee approval.
Chile: The institutional review boards at the Pan American Health Organization and United States CDC reviewed the protocol and considered it a vaccination effectiveness evaluation (non-intervention study). Monitoring vaccine effectiveness in Chile is an objective of severe acute respiratory surveillance; thus, ethics committee approval was not needed for data collection and analysis. We did not collect personal identifiers.
New Zealand: Influenza surveillance in New Zealand is conducted in accordance with the Public Health Act and thus ethics committee approval was not needed for collection or use of these data.
South Africa: Influenza surveillance is conducted in accordance with the Public Health Act and thus ethics committee approval was not needed for collection or use of these data.