We obtained data on underlying causes of death for persons >5 years of age during 1998–2009 from Statistics South Africa (15). We used codes from the International Classification of Diseases, 10th Revision (ICD-10), to compile an age-specific (5–19, 20–44, 45–64, 64–74, and >75 years) monthly mortality data time series for all-cause (ICD-10: any); all respiratory (ICD-10: J00–J99); all circulatory (ICD-10: I00–I99); and pneumonia and influenza (P&I) (ICD-10: J10–J18), a subset of all respiratory deaths. During the study period, underreporting of deaths was estimated to be <5% (13). Population denominators were obtained from Statistics South Africa (16); we obtained year- and age-specific estimates of HIV prevalence in the population and HAART coverage among HIV-positive older children and adults from the Actuarial Society of South Africa AIDS and Demographic Model (12).

For the study period before 2002, we obtained influenza virus data, including types and subtypes, from influenza-like illness surveillance implemented by the National Institute for Communicable Diseases, a division of the National Health Laboratory Service, South Africa (17), and RSV data from a cohort study (10). For the years after 2002, we acquired virologic data on influenza and RSV from the National Health Laboratory Services corporate data warehouse, a national database that includes all patients tested for respiratory viruses in the public sector in South Africa. We considered an influenza type or subtype to be dominant during the influenza season if it represented >50% of the circulating viruses.

We conducted a 2-stage analysis. In the first stage, we estimated the annual number of deaths associated with influenza and RSV in South Africa; in the second stage, we estimated the proportion of these deaths that were experienced by HIV-positive and HIV-negative persons. During the first stage, to estimate the number of deaths associated with seasonal and pandemic influenza and RSV, we fitted age-specific generalized regression models with a Poisson distribution and an identity link to the number of monthly deaths as previously described (14). The full model (model 1) included co-variates for time trends, seasonal variation, and proxies for viral circulation. Model specification, selection procedures and sensitivity analysis are provided in the Technical Appendix. Separate models were fitted for each age group and cause of death.

In South Africa, a diagnosis of AIDS is rarely coded on the death certificate (13), which hinders direct estimation of respiratory virus–associated deaths by HIV status. To assess the proportionate number of deaths associated with influenza and RSV among HIV-positive and HIV-negative persons, we used a previously developed methodology (14) that leverages the increasing trend in HIV prevalence in South Africa. The rationale is that if HIV is a risk factor for influenza- or RSV– associated death, influenza- or RSV– related mortality rates should increase over time proportionately to the observed increase in HIV prevalence. Our approach also controls for increasing HAART coverage, which may decrease the severity of influenza or RSV in HIV-positive patients (18). In this second-stage analysis, we fitted regression models (model 2) to annual estimates of influenza- and RSV-associated excess mortality rates by age group, including data for HIV prevalence and HAART coverage, time trends, and dominant influenza subtypes (for the influenza model) (14). Model specifications are provided in the Technical Appendix.

Subsequently, we obtained mortality rates associated with HIV status by dividing the estimated influenza-associated deaths by HIV status from model 2 by the mid-year population estimates within each category (i.e., HIV-positive and HIV-negative persons). Mortality rates were expressed per 100,000 person-years. We used log-binomial regression (14) to estimate age-specific and age-adjusted relative risk for influenza- and RSV-associated death related to HIV infection by comparing influenza- or RSV-associated mortality rates among HIV-positive and HIV-negative persons. We used STATA version 12 (StataCorp, College Station, Texas, USA) to implement the statistical analysis.

In contrast to previous studies (3,4), our models estimated that influenza, but not RSV, was associated with excess deaths in South Africa among persons aged >45 years. We hypothesized that differences in the timing of the RSV and influenza seasons may explain these discrepancies because RSV and influenza rarely co-circulate in South Africa, but they do in other temperate settings (3,4). This difference could potentially confound burden of illness estimates for these pathogens. We implemented a sensitivity analysis to explore this hypothesis by artificially shifting the RSV surveillance time series so that it overlapped with the influenza season and repeating model 1 calculations (Technical Appendix).

This analysis used only publicly available mortality data and deidentified and aggregated laboratory data. Therefore, the study was considered to be exempt from human subjects ethics review.

South Africa had a population of ≈44.8 million persons >5 years of age in 2009; persons 5–64 years of age accounted for 85% of this population. During the study period, a mean of 463,594 deaths occurred annually among persons from South Africa >5 years of age, of which 101,450 (22%) were attributable to respiratory and 112,716 (24%) to circulatory causes of death (Technical Appendix Table 1). The mean annual mortality rate for all-cause death increased from 112 for persons 5–19 years of age to 9,732 for persons >75 years of age. Similar patterns were observed regarding the other underlying causes of death evaluated in this study. Among persons 20–44 years of age, for which the HIV burden is greatest (24% HIV prevalence in 2009 [12]), the annual mortality rate for all respiratory deaths increased from 78 in 1998 to 310 in 2004, subsequently declining to 233 in 2009 (monthly trends provided in the Figure, panel A). In contrast, no evident secular trend for all respiratory mortality rates was observed among persons >75 years of age, for whom HIV incidence is lowest (<0.01% HIV prevalence in 2009 [12]) (Figure, panel B).

A mean of 3,403 (range 227–15,321) and 1,810 (range 578–5,247) respiratory specimens were tested annually for influenza virus and RSV, respectively. The mean annual number of specimens that tested positive was 937 (27%) for influenza virus and 356 (20%) for RSV. During the study period, the influenza season peaked between May and August (winter in South Africa); 10 of the 12 years showed peak activity during June–July (Figure, panel C). In 2009, an epidemic of influenza A(H3N2) peaked in June, and influenza A(H1N1)pdm09 activity peaked in August. RSV peak activity was observed during March and April (autumn in South Africa) in 8 of the 12 years. Early or late peaks were observed in February or May in the remaining years.

During 1998–2009, the estimated annual number of all-cause seasonal influenza-associated deaths in persons >5 years of age (model 1) ranged from 6,450 to 11,012 (rate 16.7–24.5) (Technical Appendix Table 2). In the same population, estimated annual all-cause RSV-associated deaths ranged from 292 to 626 (rate 0.7–1.4) (Technical Appendix Table 2).

The estimated mean seasonal influenza–associated mortality rate for all-cause deaths increased progressively from 0.8 for the 5–19-year age group to 379.2 for the >75-year age group. Similar trends were observed for the other causes of death evaluated in this study (Table 1). Overall, the estimated mean seasonal influenza–associated mortality rate for all-cause deaths was higher for HIV-positive persons than for those who were HIV-negative (age-adjusted relative risk [aRR] 7.9, 95% CI 7.1–8.9). Overall, 28% (2,564/9,093) of estimated all-cause seasonal influenza–associated deaths occurred among HIV-positive persons.

In 2009, we estimated 4,113 (rate 9.2) all-cause influenza A(H1N1)pdm09–associated deaths among persons >5 years of age (Table 2). The mortality rate associated with influenza A(H1N1)pdm09 in 2009 in persons 5–19 years of age was 5.4 times higher than the mean for prepandemic years. In contrast, persons >75 years of age experienced ≈100 times lower influenza A(H1N1)pdm09 rates than expected in typical nonpandemic years. A similar trend was observed for other causes of death evaluated in this study.

The estimated RSV-associated mortality rate for all-cause deaths was 0.4 for the 5–19-year age group and 2.4 for the 20–44-year age group (Table 3). However, no RSV-associated deaths were estimated among persons >45 years of age. Among persons 5–44 years of age, the RSV-associated mortality rate for all causes of death was considerably higher for HIV-positive than HIV-negative persons (aRR 66.1, 95% CI 26.0–167.8). Similar trends were observed for the RSV-associated mortality rate among all respiratory and P&I deaths. A nonsignificant RSV-associated mortality rate (mean annual deaths: 8) was identified among circulatory deaths only in the 5–19-year age group. Overall, 89% (455/511) of all-cause RSV-associated death occurred among HIV-positive persons.

On sensitivity analysis, we applied an artificial incremental shift of the RSV laboratory surveillance time series to make the influenza and RSV seasons more synchronous, resulting in a progressive increase in estimated RSV-associated deaths (Technical Appendix Table 3). We found that annual all-cause RSV-associated deaths peaked at 3,661 among persons >45 years of age (compared with 0 in the main analysis) when the RSV and influenza seasons coincided in most years (2 months’ incremental shift of the RSV season). Thereafter, the estimated mean annual all-cause RSV-associated deaths decreased again to 0 when the peak circulation of the 2 pathogens were farther apart (5-month incremental shift of the RSV season). This trend was observed for all causes of death evaluated in this study. Conversely, the estimated influenza-associated deaths remained stable throughout sensitivity analyses, without regard to the shift in RSV season. Specifically, the estimated all-cause influenza-associated deaths among persons >45 years of age remained within 10% of its original value (Technical Appendix Table 3).