Low interim influenza vaccine effectiveness, Australia, 1 May to 24 September 2017

In 2017, influenza seasonal activity was high in the southern hemisphere. We present interim influenza vaccine effectiveness (VE) estimates from Australia. Adjusted VE was low overall at 33% (95% confidence interval (CI): 17 to 46), 50% (95% CI: 8 to 74) for A(H1)pdm09, 10% (95% CI: -16 to 31) for A(H3) and 57% (95% CI: 41 to 69) for influenza B. For A(H3), VE was poorer for those vaccinated in the current and prior seasons.

The ongoing Australian 2017 influenza season was so far characterised by record-high laboratory-confirmed influenza notifications [1], high consultation rates, high hospitalisation and mortality rates, particularly in New South Wales [2], large numbers of institutional outbreaks [2] and media attention. The southern hemisphere influenza vaccine used in Australia for this season was a quadrivalent formulation comprised of an A/Michigan/45/2015 (H1N1)pdm09-like virus, an A/Hong Kong/4801/2014 (H3N2)-like virus, a B/ Brisbane/60/2008-like virus (of the B/Victoria/2/87 lineage) and a B/Phuket/3073/2013-like virus (of the B/Yamagata/16/88 lineage) [3,4]. This same vaccine composition is being used in the upcoming northern hemisphere for the 2017/18 influenza season [5]. Here we report interim influenza vaccine effectiveness estimates for 2017 in Australia, using sentinel surveillance data.

Data collection
The Australian Sentinel Practices Research Network (ASPREN) and the Victorian Sentinel Practice Influenza Network (VicSPIN) constitute Australia's two sentinel influenza general practice (GP) networks. VicSPIN operates in the state of Victoria, while ASPREN operates nationally. Both surveillance systems use similar data collection methods [6,7], with the key difference that the VicSPIN surveillance period is limited to weeks 18 to 43 (1 May 1-29 October), timed to start roughly 2 weeks after vaccination campaigns in mid-April. Briefly, sentinel GPs submit weekly reports of the number of patients seen with influenza-like illness (ILI), defined as fever/history of fever, cough and fatigue, and the total number of patients. Nose/ throat swabs are collected from a subset of patients with demographic data, date of ILI onset, vaccination status (self-reported or medical record) and indications for vaccination, such as belonging to an influenza risk group. Swabs are tested by RT-PCR and positive samples are referred to the World Health Organization (WHO) Collaborating Centre for Reference and Research on Influenza in Melbourne, for antigenic characterisation by haemagglutination inhibition assay (HAI) [8] or focus reduction assay (FRA) [9] and genetic sequencing, as described previously [6]. All data were managed and analysed using R version 3.4.1.

Virological characteristics
Virological analyses and vaccine effectiveness estimation were restricted to the period 1 May to 24 September (weeks . During this period, 2,456 patients were swabbed, but samples from 116 patients were excluded because of missing information on vaccination status, and one with missing influenza status. Among the remaining 2,339 patients, working-age adults comprised the majority (n = 1,604, 69%), 440 (19%) were aged < 15 years and 297 (13%) were aged ≥ 65 years. Around 37% (477/1,279) of test-negative patients were vaccinated in 2017.
Eighty-eight patients tested positive for A(H1)pdm09, 522 were A(H3), 75 were not yet subtyped, 11 were B/ Victoria, 259 were B/Yamagata and 105 were influenza In panel A the mean and maximum are calculated by first aligning the peaks for each season, then calculating the weekly mean and maximum consultation rate, as described in [22].
In panel B the shaded rectangular area shows the weeks for which data were included in the vaccine effectiveness analyses, 1 May 2017 to 24 September 2017.

Figure 2
Phylogenetic tree for the haemagglutinin gene of influenza A(H3) viruses
B, but the lineage was not yet determined ( Figure 1B). Virus isolation was attempted for samples with a cycle threshold value of 30 or less (Table 1). HAI testing indicated that isolates were generally antigenically similar to their respective vaccine strains. Thirty-seven percent (n=98) of A(H3) viruses yielded insufficient haemagglutinin titres for testing by HAI and were instead assessed by FRA. In HAI and FRA, 10% (7/67) and 0% (0/75) of A(H3) viruses, respectively, were low reacting to post-infection ferret antisera raised to cell-propagated A/Hong Kong/4801/2014-like viruses. However, these proportions increased to 33% (22/67) and 20% (15/75), respectively, when tested against egg-propagated reference virus.

Vaccines effectiveness estimates
Vaccine effectiveness (VE) was estimated following a case-control test-negative design, where VE is estimated from the odds ratio (OR) comparing the odds of vaccination among test-positive and test-negative patients. The limitations of this design have been discussed at length [10,11]. Estimates were adjusted for week of specimen collection (cubic spline with 4 knots), and age group (spline with knots at 5, 15, 35, 65, 75 years

Discussion
Our interim analysis suggests moderate VE against influenza A(H1)pdm09 and influenza B. However, VE was low against influenza A(H3). The antigenic data reflect ongoing issues with A(H3) candidate vaccine viruses which, when propagated in eggs, rapidly acquire adaptive changes in the haemagglutinin which alter antigenicity. Cell-based vaccines, which are less affected by this, are only licensed in the United States, were not available in Australia in 2017 and will also not be available for the upcoming European season.  The significant genetic diversity of circulating viruses, many of which exhibit amino acid substitutions in key antigenic and glycosylation sites, also makes it difficult to select candidate vaccine viruses with high coverage.
This was the second season for which the A/Hong Kong/4801/2015-containing vaccine was used in Australia [3,12], and campaigns currently underway in the northern hemisphere are also using it for a second time [5,13]. During the 2016/17 northern hemisphere season interim VE estimates ranged from 15% (95%CI: -11 to 35) to 43% (95%CI: 29 to 54) [14][15][16][17]. It is unclear whether sequential vaccination will result in lower estimates for 2017/18, but our VE estimates were particularly low for people who received vaccine in 2016 and for older adults, 76% of whom were sequentially vaccinated. This finding is consistent with a modelling study which predicts low VE for sequentially vaccinated persons when the vaccine composition is identical, but the antigenic distance between the vaccine and circulating strains is high [18]. However, confounding due to prior infection status and negative interference from pre-2016 vaccines could not be controlled for in our analysis, and may have introduced bias.
In contrast to A(H3), VE estimates for influenza B were moderate and the combined effects of vaccination in 2016 and 2017 did not blunt effectiveness for influenza B, even though the composition remained the same. Similarly, VE for the few A(H1)pdm09 cases recruited was moderate, although low for Australia at 50% (95%CI: 8 to 74), where VEs have ranged from 54% to 79% in the past [6,7]. This was the one component of the 2017 vaccine that was updated since 2016, from A/ California/7/2009 to A/Michigan/45/2015.
This study provides interim estimates of the 2017 southern hemisphere influenza vaccine in the outpatient setting and may not apply to inpatient settings or severe illness. Interim estimates can reliably predict final season estimates [19], particularly when made after the peak [20], as is the case here. Should the circulating A(H3) influenza viruses predominate in the 2017/18 northern hemisphere influenza season [21], our results suggest that the vaccine may confer limited protection. Health authorities should consider other influenza prevention measures, including antivirals and health promotion messaging, in the event of a severe season and low VE against A(H3).