From February 2009 through December 2012, a total of 21,792 patients hospitalized with lower respiratory tract infection were approached for enrollment in SARI surveillance. Of those, 16,005 (73%) were enrolled and 1,239 (8%) had positive influenza virus test results. Of the 5,876 patients who were approached but not enrolled, the most common reasons for not enrolling were unavailability of a legal guardian (among children <5 years of age; 1,452 [25%]), refusal (1,296 [22%]), and being confused or too ill (431 [7%]). Of the influenza-positive SARI cases, 463 (37%) were caused by influenza A(H3N2), 338 (27%) by influenza A(H1N1)pdm09, and 418 (34%) by influenza B viruses; 20 (2%) influenza A viruses could not be further subtyped because of low viral yield in the samples. Influenza epidemics occur annually during the colder months in South Africa (May–September), and little activity occurs during the rest of the year (Figure). The circulating types and subtypes varied between study years and within annual epidemics. During 2009, influenza virus activity occurred in 2 peaks; the first was caused by subtype A(H3N2) (194/379, 51%), which occurred earlier than in the other years, and the second was caused by subtype A(H1N1)pdm09 (160/379 42%) (Table 1 [an expanded version of this table is shown in the Technical Appendix]; Figure). The predominant influenza virus types or subtypes in the other years were as follows: B (164/273, 60%) in 2010, A(H1N1)pdm09 (140/362, 39%) in 2011, and A(H3N2) (99/205, 48%) and B (105/205, 51%) in 2012. Most (71%) case-patients were at Chris Hani-Baragwanath Academic Hospital, which reflects the higher number of SARI case-patients enrolled there. Of 12,494 SARI case-patients for whom treatment data were available, 7 (0.1%) received oseltamivir, 1 of whom had laboratory-confirmed influenza. Of 12,173 SARI case-patients for whom influenza vaccine histories were available, 19 (0.2%) reported having been vaccinated. HIV test results were available for 947 (76%) of influenza case-patients. Of those, 399 (42%) were positive for HIV: 377 (94%) from anonymized testing at the National Institute for Communicable Diseases and 22 (6%) from standard-of-care testing at the treating hospitals.

The age distribution of SARI case-patients with influenza was bimodal: most of the 1,239 influenza case-patients were <5 years of age (613 [49.5%]), followed by those 25–44 years of age (306 [24.7%]); few patients were ≥65 years of age (53 [4.3%]). This bimodal age distribution is repeated for each of the types and subtypes (Table 1) except that the first wave of A(H1N1)pdm09 infection disproportionately affected those 5–24 years of age (Table 2). According to univariate analysis, case-patients infected with influenza A(H1N1)pdm09 virus were less likely than case-patients infected with influenza A(H3N2) virus to be co-infected with another virus (crude RRR [cRRR] 0.6, 95% CI 0.4–0.8), and case-patients infected with influenza B virus were more likely to be infected with HIV (cRRR 1.7, 95% CI 1.2–2.3), have stridor (cRRR 2.1, 95% CI 1.2–3.6), have symptoms >3 days before admission (cRRR 1.6, 955 CI 1.2–2.1), and to have been hospitalized for >2 days (cRRR 1.6, 95% CI 1.2–2.2), and were less likely to have a measured fever of >38°C (cRRR 0.5, 95% CI 0.4–0.7) (Table 1). In the multivariate analysis model, only age and year remained statistically significant (Table 1). We found no statistical difference in case-fatality rates between virus types and subtypes (2.8% for A[H3N2], 1.5% for A[H1N1]pdm09, and 3.9% for B) and no difference in other markers of severity, such as admission to an intensive care unit, need for mechanical ventilation, need for supplemental oxygen, or prolonged hospitalization (Table 1).

To further explore the association between influenza types and characteristics such as HIV status, we conducted a univariate analysis and constructed a multivariable logistic regression model comparing influenza B virus with influenza A (both A[H3N2] and A[H1N1]pdm09) viruses. Except for co-infection with any virus other than influenza, the same variables were significant on this univariate analysis as were significant on the previous analysis. According to multivariate analysis, only year and HIV status remained statistically significant and were retained in the final model. Because age group was not significantly associated with virus type and did not have an interaction with HIV infection in the multivariate model, we did not include age in the final model. When we controlled for year, this model showed that case-patients with influenza B virus infection were more likely than patients with influenza A virus infection to also be infected with HIV (adjusted odds ratio 1.4, 95% CI 1.02–1.80).

According to univariate analysis, case-patients in the second wave of the A(H1N1)pdm09 pandemic were less likely than case-patients in the first wave to have had a measured fever of >38°C (crude odds ratio [cOR] 0.2, 95% CI 0.1–0.4) and more likely to have been co-infected with respiratory syncytial virus (cOR 6.4, 95% CI 1.4–29.6), have had symptoms for >3 days at admission (cOR 2.0, 95% CI 1.2–3.1), and to have needed supplemental oxygen (cOR 2.6, 95% CI 1.6–4.2; Table 2). According to multivariable logistic regression, only age group and surveillance site remained statistically significant (Table 2). Severity of hospitalization, as measured by admission to an intensive care unit, need for mechanical ventilation, need for supplemental oxygen, or prolonged hospitalization, did not differ between waves (Table 2). In addition, case-fatality rates did not differ between the first (1.3%) and second (1.5%) waves.

The influenza virus types and subtypes that circulated during the annual winter influenza seasons in South Africa varied from 2009 (the year of the A(H1N1)pdm09 pandemic) to 2012. Characteristics of patients hospitalized with SARI differed by infection with different influenza types and subtypes, particularly with regard to age and co-infection with HIV. In South Africa, the age distribution of those hospitalized with influenza during the second wave of the A(H1N1)pdm09 pandemic was more similar to the age distribution of those infected by seasonal influenza types and subtypes (a bimodal distribution with a peak in young adults 25–44 years of age) than to that of those who experienced severe disease during the first wave of the A(H1N1)pdm09 pandemic (18). This age distribution of respiratory influenza infection in South Africa is driven by the high prevalence of HIV infection among young adults in South Africa because HIV-infected adults are at increased risk for severe disease from influenza virus infection (18). In South Africa in 2009, the prevalence of HIV infection among the total population was 11% (23) and the prevalence among women attending antenatal care was 29% (24). In other settings, infection with influenza B virus is associated with less severe disease than is infection with influenza A(H3N2) virus (4–6). We found that hospitalization with influenza B virus infection was associated with HIV infection. This finding suggests that underlying immunosuppression can trigger severe influenza illness requiring hospitalization for infection caused by virus types, such as influenza B, that can cause milder illness in immunocompetent persons.

Unlike case-fatality rates and disease severity previously reported from South Africa and other countries, we found no differences in case-fatality rates or severity in South Africa during the years studied among the virus types and subtypes or between the first and second waves of the A(H1N1)pdm09 pandemic. Previous excess death models have suggested increased deaths in years when influenza A(H3N2) virus circulated in South Africa (7). The contrast between case-fatality and severity found in this analysis and that observed in previous studies in South Africa might be the result of different methods or different study periods. Although our study was conducted over fewer years and might have had less power to detect differences at a population level, we were able to look at markers of severity in individual cases and to compare different waves of A(H1N1)pdm09 virus infection.

This study has several limitations. We compared influenza types and subtypes across 4 years, so some associations might have resulted from changes in prevalence of other diseases such as HIV over the same period. We do not have data on nonrespiratory influenza disease, which might have different associations with influenza virus types and subtypes than respiratory influenza disease. Although obesity and pregnancy have been associated with infection with influenza A(H1N1)pdm09 virus, we identified few case-patients who were pregnant, and obesity was not included in our analysis because so few obese case-patients were identified by surveillance. Other factors and conditions, such as neuromuscular disorders that are associated with severe influenza disease, might be associated with specific types and subtypes, but we were unable to evaluate this association because of the small number of patients with these conditions. Patients were not enrolled on weekends, which could introduce bias if patients had more or less severe disease on weekends than patients enrolled during the week. Last, most patients were identified at a single surveillance site, so the results might more strongly reflect differences observed at that site.

Vaccination remains the best way to prevent influenza infection. Influenza vaccination coverage is very low in South Africa (25). In that country, influenza vaccination is recommended for HIV-infected persons (26), and efforts should be made to encourage higher vaccine coverage. Although differences exist between infection with different influenza types and subtypes, particularly with regard to age distribution and co-infections, it can be difficult for the clinician to differentiate infection by different types and subtypes for individual patients. Current treatment recommendations do not differ according to the subtype with which a patient is infected, in part because it is not common to type and subtype the virus in individual patients in time for clinical decision-making.

Additional modeling of influenza A and B virus, HIV status, and patient age and characteristics of patients hospitalized with influenza-associated severe acute respiratory illness, by virus type and subtype, 4 sites, South Africa, 2009–2012.