Our study is the first nationally representative study fromAfrica to estimate seasonal influenza-associated excess mortality. Overall, rates of seasonal influenza-related excess mortality among adults aged
65 years in South Africa were substantially greater than those observed in the United States. These patterns were consistent for a variety of death outcomes that have been used in the past as indicators of influenza disease burden, including all causes, pneumonia and influenza, all respiratory diseases, cerebrovascular disease, and diabetes [3
]. Standardization for demographic characteristics, as well as proxies of death-coding practices and baseline health, accounted for some but not all of these differences. These data support increased efforts for control of seasonal influenza in elderly individuals in South Africa and other low- and middle-income countries, where the excess seasonal mortality burden could be greater than previously thought.
Studies from several northern hemisphere, temperate countries, as well as from urbanized tropical areas such as Hong Kong and Singapore, have yielded estimates of influenza-related excess mortality in elderly persons remarkably similar to those from the United States [2
]. By contrast, unusually severe outbreaks of influenza A/H3N2 occurred in Madagascar and the Democratic Republic of Congo in 2002, suggesting that populations in African countries may experience increased risk of severe outcomes following influenza infection [15
]. Casefatality ratios for these outbreaks were high, possibly because of limited access to medical care, high prevalence of underlying illness and malnutrition, and crowded living conditions. Remote areas with less frequent influenza virus activity could experience higher influenza-related mortality [15
], although serological studies suggest that rural African populations may be exposed to influenza each season [32
]. While there are several published reports describing influenza circulation from various locales in Africa, surveillance remains limited, especially in rural areas; thus, the epidemiology and seasonality of interpandemic influenza virus activity remain unclear in this region [34
A previous study evaluated influenza-associated excess mortality in elderly individuals occurring at a single South African hospital in 1997–1999 [14
] and found lower excess mortality rates than ours (82–221 deaths per 100,000 population aged
65 years). Given that only 50% of deaths occur in the hospital, this study may have substantially underestimated the influenza mortality burden [14
]. In addition, the hospital study covered only a short period and was conducted in an urban community with relatively good access to health care, limiting the generalizability of the findings [14
Our study had several potential limitations. Some deaths in South Africa were not registered [19
]; however, reporting improved during the study period, and under-reporting is unlikely to vary with season. Approximately 14% of deaths in South Africa were nonspecifically coded, which is higher than in the United States (0.8%–1.2%). If anything, this difference should lead to underestimation of the excess mortality in South Africa, especially for disease-specific mortality indicators. Standardization for baseline should account for differences in coding practices or baseline mortality between settings and did not eliminate differences in the excess death rate between countries. We also observed a slight increase in the proportion of deaths not specifically coded in winter in South Africa (
1%). However, given that the seasonal component explained only 9% of the variance in nonspecifically coded deaths (P
=.01), we did not adjust further for this factor. We also note that our analyses of all-cause mortality are not prone to biases in the proportion of nonspecific codes.
All-cause excess mortality is less likely to be subject to differences in coding practices, and the most pronounced and consistent differences between South Africa and the United States were for this mortality outcome. We also found similar between-country differences for other diagnoses, such as pneumonia and influenza, all respiratory deaths, cardiovascular disease, and diabetes. Although not significant, because of small sample size, we found elevated excess mortality in seasons when influenza A/H3N2 viruses predominated in South Africa, as has been seen in previous studies [2
]. These patterns support the hypothesis that excess deaths were due to influenza and not to other causes, such as temperature variation. Moreover, the observation of elevated mortality in South Africa is robust because our study period saw fewer influenza A/H3N2-dominated seasons in South Africa (4 of 8 seasons), compared with the United States (6 of 8 seasons) [2
]. Overall, our data suggest that South African populations experience risk factors similar to those of other countries, including older age, and that the impact of influenza is worse in seasons dominated by the A/H3N2 subtype.
This study highlights that seasonal influenza-related excess mortality in Africa is likely greater than that observed in wealthier countries. Possible contributing factors may include socioeconomic differences, as well as variability in the contributing role of cocirculating viral and bacterial respiratory pathogens. Studies have suggested that much of the morbidity and mortality associated with influenza infection may be due to secondary bacterial infection, particularly Streptococcus pneumoniae
]. In addition, the relative contribution of other respiratory viruses, such as respiratory syncytial virus, to excess mortality in elderly individuals has been debated in the literature. The type of modeling we have used in this study is thought to account for the relatively constant year-to-year mortality from respiratory syncytial virus in the baseline [3
]. Our approach, however, cannot evaluate the relative contribution of secondary bacterial infections to influenza mortality in the United States and South Africa, and this should be explored in further studies.
The observed differences in excess mortality are not explained by differences in vaccination coverage between US (~65%) and South African (15% of insured population in one study) elderly individuals, because influenza-associated mortality in US elderly individuals has remained constant since the 1980s, despite increased vaccination coverage [2
]. Human immunodeficiency virus (HIV)/AIDS comorbidity is also probably not a contributing factor in persons aged
65 years, unless high HIV prevalence in younger populations is associated with increased intensity of influenza transmission in the community. It has been suggested that pandemic influenza-related mortality may be greater in lower socioeconomic settings [12
]. In 2001, it was estimated that >40% of South Africans resided in a rural settings and >50% were living in poverty [40
]. Although it is difficult to predict how representative estimates from South Africa are of the situation elsewhere in Africa, it is likely that nonspecific factors related to poverty may have contributed at least in part to the observed increased excess mortality.
The mortality burden of the 2009 influenza A/H1N1 pandemic clearly differs from that of seasonal influenza, with the highest apparent mortality risk in younger adults with underlying health conditions [42
]. We have demonstrated elevated influenza-associated mortality in South African elderly individuals. Our study is consistent with the concept that African populations in general may be at higher risk for severemortality from pandemic influenza [12
], which may be confirmed as more-robust data become available from the African continent. Future studies are needed to estimate influenza disease burden in younger African populations and to evaluate the impact of underlying host susceptibility, socioeconomic factors, and cocirculating bacterial and viral pathogens. Such data are key to predict the potential impact of influenza pandemics and to provide support for influenza vaccination programs and other control measures in developing countries. The model presented here could be applied to any near-real-time data available from South Africa, to estimate the burden in each wave as the pandemic progresses.