A thesis submitted to the Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, in fulfillment of the requirements for the degree of Doctor of Philosophy Johannesburg, October 2014 / Introduction Data on the burden of influenza-associated hospitalisation and mortality in relation to other aetiologies of pneumonia as well as risk groups for severe and complicated disease are important to guide influenza prevention policy. Materials and methods We estimated influenza-related deaths as excess mortality above a model baseline during influenza epidemic periods from monthly age-specific mortality data using Serfling regression models. For individuals aged ≥65 years from South Africa and the United States of America (US) we evaluated influenza-related deaths due to all causes, pneumonia and influenza (P&I) and other influenza-associated diagnoses for 1998-2005. For adults with acquired immune deficiency syndrome (AIDS) aged 25-54 years in South Africa (1998-2005) and the US (pre-highly active antiretroviral therapy (HAART) era: 1987-1994; HAART era: 1997-2005) we estimated deaths due to all-causes and P&I. We prospectively enrolled individuals with severe acute respiratory illness (SARI) at six hospitals in four provinces of South Africa from 2009-2012. Using polymerase chain reaction, respiratory samples were tested for ten respiratory viruses and blood for pneumococcal DNA. Cumulative annual SARI incidence was estimated at one site with available population denominators. Results Age-standardised excess mortality rates amongst seniors were higher in South Africa than in the US (545 vs. 133 per 100,000 for all-causes, p<0.001; 63 vs. 21 for P&I, p=0.03). The mean percent of winter deaths attributable to influenza was 16% in South Africa and 6% in the US, p<0.001. For all respiratory causes, cerebrovascular disease and diabetes age-standardised excess death rates were 4- to 8-fold greater in South Africa than in the US, and the percent increase in winter deaths attributable to influenza was 2- to 4-fold higher. In the US pre-HAART, influenza-related mortality rates in adults with AIDS were 150- (95% confidence interval (CI) 49-460) and 208- (95% CI 74-583) times greater than in the general population for all-cause and P&I respectively and 2.5- (95% CI 0.9-7.2) and 4.1- (95% CI 1.4-13) times higher than in seniors. Following HAART introduction, influenza-related mortality in adults with AIDS dropped 3-6 fold but remained elevated compared to the general population (all cause relative risk (RR) 44, 95% CI 16-12); P&I RR 73, 95% CI 47-113). Influenza-related mortality in South African adults with AIDS was similar to that in the US in the pre-HAART era. From 2009-2012 we enrolled 8723 children age <5 years with SARI. The human immunodeficiency virus (HIV) prevalence among tested children was 12% (705/5964). The overall prevalence of respiratory viruses identified was 78% (6517/8393), which included 26% (n=2216) respiratory syncytial virus (RSV) and 7% (n=613) influenza. The annual incidence of SARI hospitalisation in children age <5 years ranged from 2530-3173 per 100,000 and was 1.1-3-fold greater in HIV-infected than HIV-uninfected children. In multivariable analysis, compared to HIV-uninfected children, HIV-infected children were more likely to be hospitalised >7 days (odds ratio (OR) 3.6, 95% CI 2.8-5.0) and had a 4.2-fold (95% CI 2.6-6.8) higher case-fatality ratio. From 2009-2012, we enrolled 7193 individuals aged ≥5 years with SARI. HIV-prevalence was 74% (4663/6334) and 9% (621/7067) tested influenza positive. The annual incidence of SARI hospitalisation in individuals age ≥5 years ranged from 325-617 per 100,000 population and was 13 to 19-fold greater in HIV-infected individuals (p<0.001). On multivariable analysis, compared to HIV-uninfected individuals, HIV-infected individuals were more likely to be receiving tuberculosis treatment (OR 2.1, 95% CI 1.3-3.2), have pneumococcal infection (OR 2.2, 95% CI 1.6-2.9), be hospitalised for longer (>7 days rather than <2 days OR 2.4, 95% CI 1.8-3.2) and had a higher case-fatality ratio (8% vs. 5%; OR 1.6, 95% CI 1.2-2.2), but were less likely to be infected with influenza (OR 0.6, 95% CI 0.5-0.8). Influenza was identified in 9% (1056/11925) of patients of all ages enrolled in SARI surveillance from 2009-2011. Among influenza case-patients, 44% (358/819) were HIV-infected. Age-adjusted influenza-associated SARI incidence was 4-8 times greater in HIV-infected (186-228 per 100,000 population) than HIV-uninfected (26-54 per 100,000 population). On multivariable analysis, compared to HIV-uninfected individuals, HIV-infected individuals with influenza-associated SARI were more likely to have pneumococcal co-infection (OR 2.3, 95% CI 1.0-5.0), influenza type B than type A (OR 1.6, 95% CI 1.0-2.4), be hospitalised for 2-7 days (OR 2.8 95% CI 1.5-5.5) or >7 days (OR 4.5, 95% CI 2.1-9.5) and more likely to die (OR 3.9, 95% CI 1.1-14.1). Discussion and conclusions The mortality impact of seasonal influenza in the South African elderly may be substantially higher in an African setting compared to the US. Adults with AIDS in South Africa and the US experience substantially elevated influenza-associated mortality rates, which although lessened by widespread HAART treatment does not completely abrogate the heightened risk for influenza illness. HIV-infected children and adults also experience substantially elevated incidence of hospitalisation for influenza-associated SARI and have higher case-fatality ratios. Influenza is commonly detected amongst children (7%) and adults (9%) with SARI. Less frequent identification of influenza amongst HIV-infected than -uninfected individuals aged ≥5 years likely reflects increased relative burden and role of other opportunistic pathogens such as pnuemococcus and Pneumocystis jirovecii. Improved access to HAART for HIV-infected individuals and vaccination against influenza virus amongst HIV-infected individuals, young children and the elderly, where the influenza burden is great may reduce the high burden of hospitalisations and mortality associated with influenza.

Après des décennies d'amélioration, l'espérance de vie a stagné dans plusieurs pays à faible mortalité ces dernières années, avec, dans certains cas, quelques reculs. L’augmentation de la mortalité due à la grippe et aux surdoses de drogue, en particulier dans la génération des baby-boomers, a été le principal responsable de cette stagnation de l’espérance de vie. Cette découverte était inattendue, car il est considéré que la mortalité extrinsèque – par opposition à la mortalité intrinsèque due à des maladies dégénératives se déclarant souvent aux grands âges – joue un rôle négligeable dans les changements actuels d'espérance de vie. Pour la même raison, les tendances temporelles de la mortalité extrinsèque n’ont guère retenu l’attention des chercheurs. Les crises périodiques dues aux épidémies de grippe et à la crise des opioïdes sont considérées comme les principaux déterminants des variations de la mortalité extrinsèque. Cependant, des preuves récentes suggèrent que les effets de cohorte jouent un rôle important dans la modulation de la mortalité extrinsèque, mais que de telles influences sont encore mal connues. L'objectif principal de cette thèse est d'examiner le rôle des effets de cohorte sur l’évolution de la mortalité extrinsèque dans les dernières décennies, avec un accent particulier mis sur la grippe et les causes de décès comportementales. Plus spécifiquement, elle vise à (1) déterminer les différences par cohorte de mortalité par la grippe et l’influence des expositions précoces au virus sur cette mortalité; (2) analyser le désavantage de mortalité des baby-boomers au Canada et aux États-Unis en identifiant la contribution des causes comportementales à ce désavantage; et (3) développer un outil méthodologique permettant à la fois l'analyse visuelle de la dynamique temporelle des effets non linéaires d'âge, de période et de cohorte (APC) et la comparaison entre divers phénomènes ou populations. Pour ces analyses, nous utilisons des micro-données de mortalité provenant de systèmes de statistiques de l’état civil au Canada et aux États-Unis. Nous utilisons également les taux de mortalité et de fécondité de divers pays pour généraliser l'analyse visuelle des effets non linéaires à d'autres phénomènes démographiques que la mortalité. Les analyses ont été réalisées en appliquant des modèles de Serfling pour l’estimation de la mortalité par grippe, des mesures démographiques permettant une décomposition par cause des variations de la mortalité, des techniques de lissage pour identifier les tendances et des approches statistiques et visuelles sur des configurations de Lexis pour l’analyse des effets APC. Les résultats, sous la forme de trois articles scientifiques, montrent que malgré des fluctuations marquées au cours des années calendrier (période), les cohortes de naissance ont une influence indépendante et durable sur la mortalité liée à la grippe ou due au comportement. Les principaux résultats du premier article suggèrent que deux mécanismes modulent la mortalité grippale au fil des cohortes. Pour la population jeune et adulte, les risques de mortalité par cohortes dépendent du contraste en le premier virus auquel on est vraisemblablement exposé (le virus laissant« l’empreinte antigénique ») et le virus rencontré à l’âge adulte, au moment de l’épidémie sous observation. Des modifications significatives du risque de décès ont ainsi été observées lors d’épidémies de grippes pour les cohortes nées lors d'importants changements antigéniques (par exemple, une diminution significative du risque pour les cohortes nées entre 1957 et 1968). Pour les âges plus avancés, nous n’avons pas identifié de tels effets de cohorte « ponctuels », mais plutôt un effet de cohorte de plus longue haleine, qui aura conduit à un déclin progressif de la mortalité par grippe entre 1959 et 2016. En nous inspirant des théories dites de technophysio ou de cohort morbidity phenotype, nous attribuons ce déclin à des changements s’étant produit bien avant, c’est-à-dire à l’amélioration marqué des conditions sanitaires qui a eu lieu entre 1900 et 1930, au moment où les cohortes concernées venaient au monde et dont elles ont pu bénéficier. Les travaux du deuxième article de cette thèse révèlent que la plupart des excès de mortalité chez les baby-boomers au Canada et aux États-Unis sont dus à des causes comportementales. Le désavantage des baby-boomers résulte de plusieurs effets de cohortes sur des causes comportementales différentes, et non pas d'effets de période ponctuels affectant la même cohorte aux âges différents, un mécanisme alternatif qui pourrait expliquer la «pénalité des boomers». Les baby-boomers présentaient respectivement un risque d'hépatite C et de mortalité par drogue trois fois et deux fois plus élevé que les cohortes voisines. La contribution méthodologique des graphique de courbure APC, présentée dans le troisième article, nous a permis d'analyser la dynamique des effets non linéaires au fil du temps, à travers divers phénomènes et populations. Cette technique offre une plus grande flexibilité que les modèles statistiques ou autres graphiques de Lexis. Les résultats présentés dans cette thèse montrent l'importance d'analyser les effets de cohortes sur la mortalité extrinsèque. Nos résultats indiquent que même en présence de perturbations de période importantes affectant la mortalité extrinsèque à la plupart des âges, les effets de cohorte se sont maintenus au fil du temps. Ces résultats suggèrent également que les politiques publiques peuvent améliorer considérablement la santé de la population en formulant des politiques qui prennent en compte la sensibilité différentielle des cohortes aux facteurs de risque et en fournissant un soutien social aux cohortes les plus vulnérables. / After decades of improvement, life expectancy momentarily declined during 2014-15 in several high income countries, with subsequent reversals in some cases. The main sources of this stagnation have been increases in mortality from influenza and drug overdoses, mainly for the baby-boomer generation. This trend is unexpected because it has long been assumed that extrinsic mortality, which is due to causes originating outside the body – in opposition to intrinsic mortality from degenerative diseases at old ages –, plays a negligible role in life expectancy changes. For this reason, the temporal patterns of extrinsic mortality have received little attention in demographic research. Period crises such as influenza epidemics and the opioid crisis are considered the main determinants of variations of extrinsic mortality. However, despite recent evidence suggesting that cohort effects have an important role in modulating extrinsic mortality, little is known about this relationship. The main objective of this dissertation is to help fill this gap by examining cohort influences on extrinsic mortality change, with a particular emphasis on influenza and behavioral causes. More specifically, we aim (1) to quantify cohort differences in mortality from influenza and the influence of early life exposures to the virus on subsequent influenza mortality; (2) to analyze the baby boomers’ disadvantage in mortality in Canada and the United States, while identifying the contributions of behavioral causes to this disadvantage; and (3) to develop a methodological tool that can be used to both conduct visual analysis of the temporal dynamics of nonlinear Age-Period-Cohort (APC) effects, and compare these dynamics across various phenomena or populations. To achieve these goals, we use micro-level mortality data from vital statistics in Canada and the United States. We also employ death and fertility rates from various countries to generalize the visual analysis of nonlinear effects to other demographic phenomena. The analyses were conducted by applying Serfling models for the estimation of influenza mortality, demographic measures for the decomposition of cause-specific mortality changes, smoothing techniques for the identification of trends, and statistical and visual approaches on the Lexis configuration for the analysis of APC effects. The results, in the form of three scientific articles, show that despite marked fluctuations over calendar years (periods), birth cohorts have an independent and sustained influence on influenza and mortality from behavioral causes. The main results from the first paper suggest that two mechanisms modulated influenza mortality over cohorts. For the young and adult population, the mortality risks over cohorts depend of the contrast between the first virus to which individuals were exposed (the virus producing an antigenic imprinting) and the virus encountered in adulthood during the observed epidemic. For this age segment, significant changes in risk were found during influenza epidemics among cohorts born during important antigenic shifts (e.g., a decrease in risk for cohorts born between 1957 and 1968). For older ages, we did not identify such “punctual” cohort effects but rather a smooth and monotonic change in cohort effects that might have driven a progressive decline in influenza mortality between 1959 and 2016. Inspired by so-called cohort morbidity phenotype and technophysio evolution theories, we attributed this decline to changes produced earlier, i.e., to the sharp sanitary improvements occurred between 1900 and 1930, when the concerned cohorts were born and when they could have benefited. Findings from the second paper revealed that most of the baby boomers’ excess mortality in Canada and the United States is driven by behavioral causes of death. The “boomer disadvantage” resulted from multiple cohort effects on behavioral-related mortality, and not from punctual period effects affecting the same cohort at different ages. Among the baby boomers, the risk of dying from hepatitis C was almost three times higher, and the risk of dying from drug-related causes was almost two times higher, than among the adjacent cohorts. These results were obtained using an innovative methodology developed in the third paper, which allowed us to analyze the dynamics of nonlinear effects over time through APC curvature plots. This technique provides greater flexibility than statistical models or other Lexis plots, and it has been shown to be applicable to other demographic phenomena, such as fertility. The findings presented in this dissertation offer evidence of the importance of analyzing cohort effects on extrinsic mortality. Our results indicate that even in the presence of substantial period disturbances affecting extrinsic mortality at most ages, cohort effects were sustained over time. These findings also suggest that public policies can significantly improve the health of the population by formulating policies that take into account the differential sensitivity of cohorts to risk factors and by providing social support to the most vulnerable cohorts.