Datasets were received from 11 groups of investigators, consisting of data from 10 published and 3 unpublished studies. Data were received from NZ, Singapore, and New South Wales (NSW), the Northern Territory (NT), Queensland, Tasmania, Victoria, and Western Australia in Australia. Datasets are listed by study design and population, with pandemic phases referring to investigators’ definitions, which resulted in 19 datasets for analysis. Study designs consisted of 4 prospective cohorts (E–H), 3 randomized controlled trials (L, O, R), 2 prepandemic cross-sectional studies (A, D), 1 retrospective cohort study (M), and 6 unpaired pre-and postpandemic cross-sectional studies (I, J, K, N, P, S). Eleven datasets were community based, whereas 8 were from groups with potential risk factors.

Laboratory techniques common to all studies included HI assay, per inclusion criteria, and use of egg-grown, β-propiolactone–inactivated A/California/07/2009 reference virus as the antigen source. All studies provided titers and patient’s age in years for each assay, and all Australian studies provided geographic data to at least state/territory level. Table 1 illustrates study characteristics that differed between included studies, with studies differing by design, enrolment criteria, specimen type, and specific HI method. All studies but 2 (I, K) provided data on patient’s sex for all assays. All studies attempted to avoid contamination of specimens from vaccination effect, but approaches used to achieve this differed.

Received data consisted of 18,279 individual specimens, of which 18,131 assays (from 14,036 persons) were eligible for analysis, whereas 148 did not meet inclusion criteria. Samples were reclassified as prepandemic (4,414), intrapandemic (6,002), or postpandemic (7,715), according to the criteria described, with intrapandemic assays excluded from further analysis (Figure). The timing of the pandemic phases and the sample taking is summarized in the Technical Appendix. Of the assays eligible for analysis in the pre- or postpandemic groups, 125 prepandemic specimens and 2,065 postpandemic specimens were from risk groups, while the remainder were from community-based datasets.

Tables 2 and 3 show seropositive proportions in the pre- and postpandemic periods. Overall, the age-standardized prepandemic seropositive proportion was 9.4%, with regional estimates of 10.6% in Australia, 11.9% in NZ, and 3.5% in Singapore. Higher levels of immunity were seen with increasing age, with only 1 of 5 studies (A) of children finding evidence of preexisting immunity in age group 0–4 years, whereas markedly higher seropositive proportions were seen in those >75 years of age.

In the postpandemic period, the age-standardized seropositive proportion was 24.3%, giving an attack rate of 14.9% (Table 3). Attack rates by country were 13.1% for Australia, 19.0% for NZ, and 14.0% for Singapore. For all regions in which children 5–14 years of age were assessed, this age group had the highest levels of postpandemic seropositivity, except for NSW, in which those 15–34 years of age showed the greatest seropositivity. Among risk groups, unweighted seropositive proportions were 28.4% in HIV-positive persons in NSW, 21.5% in hemodialysis patients in NSW, 26.7% in NZ health care workers, 9.5% in Singaporean health care workers, 33.9% in Singaporean military personnel, 6.8% in staff and residents of Singaporean residential care facilities, 14.7% in pregnant women in WA, 29.5% in indigenous residents of the NT, 34.3% in Maori in NZ, and 43.7% in Pacific Peoples in NZ.

Logistic regression performed in assays from postpandemic, community-based collections showed that the age groups 5–14 years and 15–34 years, as well as residence in NZ, were associated with increased seropositivity. Negative effects were seen for older age groups and those with residence in Singapore (Table 4).

In 2 instances, the same demographic group was assessed by 2 studies using different methods, allowing for assessment of the effect of study design. The 2 cross-sectional studies performed in adults in NSW in the postpandemic phase (performed in different laboratories) were included in a regression model that examined the effect of using serum specimens collected to identify pathologic agents from outpatients by comparing them to results with plasma units of donated blood. The 2 collections obtained from WA adults in the postpandemic phase were analyzed similarly to examine the effect of using volunteers in a randomized-controlled trial by comparing those results to results using plasma units of donated blood. Neither the use of opportunistically collected blood specimens nor the use of adults enrolled in randomized controlled trials showed a significant difference on the outcome of seropositivity from results using plasma units of donated blood (Table 5).

We compared several other pairs of datasets by logistic regression to examine the effect of specific risk factors on the outcome of seropositivity. The odds ratio for the binary variable of study of origin (comparison study vs. reference study) is displayed as an estimate of the effect of the risk factor on the outcome. In NSW during the prepandemic phase, living in a residential care facility was associated with lower levels of preexisting seropositivity than was living in the general community. However, in the postpandemic phase, we found that persons with HIV infection or those who were undergoing hemodialysis were not significantly more likely to be seropositive than were community control subjects. In WA in the postpandemic phase, pregnancy was associated with lower levels of seropositivity. Health care workers in NZ had levels of postpandemic seropositivity similar to community controls, but those of health care workers in Singapore were lower. In Singapore, military personnel had similar levels of postpandemic seropositivity, while staff and residents of residential care facilities had lower levels compared to community controls. Aboriginal and Torres Strait Islander residents of the NT had higher levels of postpandemic immunity than other ethnic groups, as did Pacific Peoples of NZ (Table 5).

We obtained estimates of the full epidemiologic effects of A(H1N1)pdm09 in the 2009 Southern Hemisphere winter by pooling data from several serologic studies performed across our region. We believe that population-based serologic studies give a more direct measure of community exposure to the virus than notification-based data, which are inherently limited by the proportion of cases of infection that are captured by the notification system. The individual-level data enabled us to apply consistent statistical methods across studies. This enabled estimates of seropositivity to be made across more directly comparable groups, as well as assessments of the effects of specific risk factors on seropositivity.

Our community-based, age-standardized estimates of prepandemic seropositive proportions ranged from 3.5% to 11.9%, with Singapore demonstrating a lower level of prepandemic immunity than Australia and NZ. The increased levels of prepandemic immunity in those >75 years of age are likely to be partially due to cross-reacting antibody responses to influenza A/South Carolina/1/1918 and related viruses that were circulating in the early 20th century (20). However, the steady increase in seropositivity with age across age groups suggests more recent circulation of influenza viruses with the potential to elicit cross-reacting antibody responses (21).

The finding of peak postpandemic seropositivity in the 5- to 14-year age group is consistent with greater social mixing of school-aged children, lower prepandemic immunity, and results from other population-wide studies (22,23). Despite the low level of prepandemic cross-reactive antibodies to the virus, Singapore remained the region with the lowest proportion of population-standardized seropositivity in the postpandemic phase (17.5%), whereas estimates from Australia and NZ ranged from 22.1% to 32.8%. The implication of lower postpandemic seropositivity in more tropical regions is consistent with estimates from India and Hong Kong (23,24), as well as with the nonsignificant trend toward greater seropositivity in the more northerly Australian regions (NT and Queensland) on logistic regression. Although overall attack rates were similar for Australia and Singapore, this finding highlights the geographic heterogeneity of influenza spread and suggests that latitude may be a critical predictor of susceptibility to influenza, which might be explained by increased efficiency of transmission of influenza in cold temperatures (25) or population levels of vitamin D stores (26). The negative overall attack rate in those >75 years of age may have been an anomaly, because this age group had the smallest number of specimens and the prepandemic specimens were predominantly from NSW, while the postpandemic specimens were a combination of specimens from NSW, NZ, and NT. Other age groups had more similar compositions between pre- and postpandemic phases and so are likely to be more directly comparable. Alternative explanations include waning immunity in elderly persons in the months after seasonal influenza vaccination or limitations of study K, which observed the greatest decrease in titers in this age group.

Although several coexisting conditions have been found to be associated with severity of infection with A(H1N1)pdm09, most laboratory-confirmed cases across the Southern Hemisphere have occurred in persons without known risk factors (27). We found no increase in risk for postpandemic seropositivity among hemodialysis patients and a group of persons with generally well-controlled HIV infection. These results are consistent with the observation that HIV-infected patients admitted to a hospital for influenza have similar clinical outcomes as do non-HIV patients (28). Pregnant women represented ≈7%–9% of patients with laboratory-confirmed cases, severe infections, and admissions to intensive care units in the Southern Hemisphere (27). Although it has been postulated that this occurred because of the patients’ younger ages and close contact with children, our results suggest that pregnancy is associated with a lower likelihood of infection, possibly because pregnant women actively avoid infection. Therefore, pregnant women appear more susceptible to severe illness with A(H1N1)pdm09 infection, which may relate to lower levels of immunoglobulin G₂ (29). By contrast, our results suggest that the 6- to 7-fold higher rates of hospitalization seen in indigenous persons in our region are likely to be partially attributable to a higher attack rate in Australia (30). We did not find evidence for a higher attack rate among health care workers or military personnel, with levels of seropositivity comparable to those of the general community. A study in a Finnish garrison found that 22.3% of personnel had titers of >40, a level lower than the 33.9% observed in Singaporean military personnel postpandemic, even though this study was performed in response to a recent outbreak (31). Levels of seropositivity in those living in residential care appeared lower than in community-living persons of comparable age, both in a prepandemic comparison in NSW and in a postpandemic comparison in Singapore.

The unavoidable limitation to our comparisons is that they included data from multiple studies that used differing methods. Studies differed by epidemiologic approach, specimen type, and laboratory methods, and the jurisdictions studied exhibited different public health responses. We excluded from analysis data we considered to have been obtained with methods that were unlikely to give a population-wide estimate of serologic immunity, for example, 1 retrospective prepandemic collection from persons with postpandemic seropositivity (M) and the postintervention assessments from clinical trials (L, O, R, S). Several studies used convenience collections of specimens taken for clinical indications before routine discarding. These studies enabled population-based estimates but were subject to selection bias, given that conditions predisposing to influenza might increase the chance of being tested. By contrast, the use of blood donor specimens may select for a healthier sample. Cohort studies (E–H) were analyzed in the same manner as for cross-sectional surveys, although samples included in these datasets were determined by selection biases relating to original enrollment in the cohort as well as to enrollees dropping out. Previous evidence indicates that this is a valid approach to estimating population-wide immunity (32). Moreover, our analysis found no effect from differing study methods when comparing 2 pairs of studies performed in the same populations. Therefore, while the differences seen between the risk groups could have been caused by differences in study method, we found no evidence of this from the data available.

Whether the epidemiologic differences are due to differences in transmission in differing populations or because of the effectiveness of public health responses is difficult to gauge. In Australia, most jurisdictions moved from the Delay to the Contain phase on May 22 and from the Contain to the Protect phase on June 22. Only Victoria, which contributed 234 specimens to this pooled analysis, differed in the timing of its response phases (33). Although there were notable differences between public health management of the response to the outbreak in NZ and Singapore, the timing of the transition to containment was broadly similar, with NZ focusing primarily on containment from April 25 to June 21 (34), while Singapore began its transition to the Mitigation phase on June 29 (35).

Protocols for the HI assay may differ between laboratories in terms of specimen source and preparation (serum or plasma, erythrocyte adsorption), reagents (erythrocyte species, antigen preparation), procedure (incubation conditions), and controls. Furthermore, use of fresh erythrocytes for HI assays means inherent within-laboratory variability must be managed. To minimize variability between laboratory method and erythrocyte batches, control panels of serum samples were shared and results were standardized. A common source of virus antigen was also shared. Such comparative experiments were performed early in the pandemic between 3 of the 4 laboratories described in this analysis, with minimal variation seen. These 3 laboratories used a common source of A(H1N1)pdm09 antigen for at least 15 of the 19 datasets included. International standards were also available in 2009 for standardization of serologic assays around the world. Notably, the source of erythrocytes to detect influenza virus may vary, depending on the binding specificity of the hemagglutinin protein for each virus. A(H1N1)pdm09 virus recognized human, turkey, and guinea pig erythrocytes. This enabled laboratories to use cells that were available and that they were experienced in handling.

Although all studies used a titer of >40 as the cutoff for seropositivity, this is an over-simplification of the complex immune response to influenza infection, which includes both cellular and humoral components (36). Although a titer of 40 was achieved in 80%–90% of persons with PCR-confirmed infection with A(H1N1)pdm09 (37), in unpaired analysis, no single cutoff reliably determines past infection and subsequent immunity. Serologic studies that incorporated interviewing participants about symptoms of influenza-like illness suggest that as many as half of those with serum titers of >40 in the postpandemic phase do not have a history of a compatible illness (16,31). This finding partly reflects the fact that a proportion of those patients who were seropositive in the postpandemic phase were already seropositive in the prepandemic phase, but also suggests that some persons who seroconverted did not experience or report symptoms. Despite this, pre- and postpandemic cross-sectional serologic surveys are the most convenient and inclusive method for assessing population-wide serologic immunity.

Our results provide a broad picture of the effects of A(H1N1)pdm09 in the Southern Hemisphere during the winter of 2009. The absence of clear differences between estimates with different study methods suggests that pooling of data is likely to be useful in estimating the effects of the virus across population groups. We found greater levels of prepandemic seropositivity as patient’s age increased, particularly in those >75 years of age. By contrast, in the postpandemic period, school-aged children showed the greatest levels of immunity. Health care workers, military personnel, persons with HIV infection, and hemodialysis patients had levels of postpandemic seropositivity similar to those of the general community. Pregnancy and residential care appeared protective from infection, suggesting more severe disease in those infected. Despite recording the lowest prepandemic levels of immunity, Singapore retained comparatively lower levels of seropositivity after the pandemic.

Provides dates samples collected in serologic studies and dates used to define pandemic phases in serologic studies.