Showing papers in "Emerging Themes in Epidemiology in 2005"

Dengue fever: new paradigms for a changing epidemiology.

Abstract: Dengue is the most important arthropod-borne viral disease of public health significance. Compared with nine reporting countries in the 1950s, today the geographic distribution includes more than 100 countries worldwide. Many of these had not reported dengue for 20 or more years and several have no known history of the disease. The World Health Organization estimates that more than 2.5 billion people are at risk of dengue infection. First recognised in the 1950s, it has become a leading cause of child mortality in several Asian and South American countries.This paper reviews the changing epidemiology of the disease, focusing on host and societal factors and drawing on national and regional journals as well as international publications. It does not include vaccine and vector issues. We have selected areas where the literature raises challenges to prevailing views and those that are key for improved service delivery in poor countries.Shifts in modal age, rural spread, and social and biological determinants of race- and sex-related susceptibility have major implications for health services. Behavioural risk factors, individual determinants of outcome and leading indicators of severe illness are poorly understood, compromising effectiveness of control programmes. Early detection and case management practices were noted as a critical factor for survival. Inadequacy of sound statistical methods compromised conclusions on case fatality or disease-specific mortality rates, especially since the data were often based on hospitalised patients who actively sought care in tertiary centres.Well-targeted operational research, such as population-based epidemiological studies with clear operational objectives, is urgently needed to make progress in control and prevention.

The Bradford Hill considerations on causality: a counterfactual perspective

Abstract: Bradford Hill's considerations published in 1965 had an enormous influence on attempts to separate causal from non-causal explanations of observed associations. These considerations were often applied as a checklist of criteria, although they were by no means intended to be used in this way by Hill himself. Hill, however, avoided defining explicitly what he meant by "causal effect". This paper provides a fresh point of view on Hill's considerations from the perspective of counterfactual causality. I argue that counterfactual arguments strongly contribute to the question of when to apply the Hill considerations. Some of the considerations, however, involve many counterfactuals in a broader causal system, and their heuristic value decreases as the complexity of a system increases; the danger of misapplying them can be high. The impacts of these insights for study design and data analysis are discussed. The key analysis tool to assess the applicability of Hill's considerations is multiple bias modelling (Bayesian methods and Monte Carlo sensitivity analysis); these methods should be used much more frequently.

The epidemiological impact of antiretroviral use predicted by mathematical models: a review

Abstract: This review summarises theoretical studies attempting to assess the population impact of antiretroviral therapy (ART) use on mortality and HIV incidence. We describe the key parameters that determine the impact of therapy, and argue that mathematical models of disease transmission are the natural framework within which to explore the interaction between antiviral use and the dynamics of an HIV epidemic. Our review focuses on the potential effects of ART in resource-poor settings. We discuss choice of model type and structure, the potential for risk behaviour change following widespread introduction of ART, the importance of the stage of HIV infection at which treatment is initiated, and the potential for spread of drug resistance. These issues are illustrated with results from models of HIV transmission. We demonstrate that HIV transmission models predicting the impact of ART use should incorporate a realistic progression through stages of HIV infection in order to capture the effect of the timing of treatment initiation on disease spread. The realism of existing models falls short of properly reproducing patterns of diagnosis timing, incorporating heterogeneity in sexual behaviour, and describing the evolution and transmission of drug resistance. The uncertainty surrounding certain effects of ART, such as changes in sexual behaviour and transmission of ART-resistant HIV strains, demands exploration of best and worst case scenarios in modelling, but this must be complemented by surveillance and behavioural surveys to quantify such effects in settings where ART is implemented.

Epidemiologic measures and policy formulation: lessons from potential outcomes

Abstract: This paper provides a critique of the common practice in the health-policy literature of focusing on hypothetical outcome removal at the expense of intervention analysis. The paper begins with an introduction to measures of causal effects within the potential-outcomes framework, focusing on underlying conceptual models, definitions and drawbacks of special relevance to policy formulation based on epidemiologic data. It is argued that, for policy purposes, one should analyze intervention effects within a multivariate-outcome framework to capture the impact of major sources of morbidity and mortality. This framework can clarify what is captured and missed by summary measures of population health, and shows that the concept of summary measure can and should be extended to multidimensional indices.

Influenza associated excess mortality in Germany, 1985 – 2001

Abstract: Influenza-associated excess mortality is widely used to assess the severity of influenza epidemics. In Germany, however, it is not yet established as a routine component of influenza surveillance. We therefore applied a simple method based on the annual distribution of monthly relative mortality (relative mortality distribution method, RMDM) to a time-series of German monthly all-cause mortality data from 1985–2001 to estimate influenza-associated excess mortality. Results were compared to those obtained by cyclical regression. Both methods distinguished stronger from milder influenza seasons, but RMDM gave the better fit (R2 = 0.80). For the years after reunification, i.e. 1990/91 through 2000/01, RMDM yielded an average of 6900 (conservative estimate) to13600 influenza-asssociated excess deaths per season (crude estimate). The most severe epidemics occurred during subtype A/H3N2 seasons. While German all-cause mortality declined over the study period, the number of excess deaths displayed an upward trend, coinciding with an increase of the proportion of the elderly population.

Assessing observational studies of medical treatments

Abstract: Background Previous studies have assessed the validity of the observational study design by comparing results of studies using this design to results from randomized controlled trials. The present study examined design features of observational studies that could have influenced these comparisons.

Biodemographic perspectives for epidemiologists

Abstract: A new scientific discipline arose in the late 20th century known as biodemography. When applied to aging, biodemography is the scientific study of common age patterns and causes of death observed among humans and other sexually reproducing species and the biological forces that contribute to them. Biodemography is interdisciplinary, involving a combination of the population sciences and such fields as molecular and evolutionary biology. Researchers in this emerging field have discovered attributes of aging and death in humans that may very well change the way epidemiologists view and study the causes and expression of disease. In this paper, the biodemography of aging is introduced in light of traditional epidemiologic models of disease causation and death.

AIDS, conflict and the media in Africa: risks in reporting bad data badly.

Abstract: Background Conflict, poverty and HIV disproportionately affect people in sub-Saharan Africa. The manner in which governments, national and international organisations and the media report on the HIV epidemic in situations of conflict, post-conflict and reconstruction can have unintended and negative consequences for those affected populations. The media in particular has a huge influence on how the world observes and reacts to the HIV epidemic among conflict-affected and displaced populations.

Epidemiologists and causation in an intricate world.

Abstract: The delivery of treatments for diseases has always been the central mission of medicine as the delivery of preventative or control measures has always been the mission of public health. This capacity to deliver interventions has been present over the history of both disciplines, independent of the degree of knowledge of causes of diseases or effectiveness of treatment or preventative measures. The development and reaffirmation of epidemiology as a scientific discipline has been closely associated with its ability to search for causes of health-related events. Causal reasoning was always part of human thinking and of philosophical concerns. However, the first to transform philosophical causal concerns in an organized logical system from which causal relationships could be inferred was the philosopher J. Stuart Mill in the mid-19th century. He formulated the so called Mill's Canons [1]. At the same time, bacteriologists excited with their newly-emerging discipline were looking for more specific systems to help them to check etiological associations between infectious agents and specific diseases, and to attend this need the Henle-Kock postulates were formulated. After that, it would take nearly a century for new causal formulations related to health-related events to be developed. The rise of the chronic diseases as the main component of the morbidity and the mortality in rich countries and the consequent changes that this had upon the causal thinking in epidemiology (until then dominated by infectious disease) was the stimulus for the development of the widely known Hill criteria of causality [2]. In order to address problems raised by newly discovered infectious agents that did not behave according to the Henle-Kock postulates and in an effort to unify the causal verification of infectious and non-infectious health problems, Evans [3] presented a new set of postulates. To make the lives of epidemiologists difficult, the causes of health-related events are multiple and very heterogeneous – varying from a micro-infectious agent to a macro-social factor. Ideas of multicausality have long been part of epidemiological thinking, and multi-causal models have been built-up. However, the common way epidemiology search for causes continue to be throughout the test of one-by-one potential causes, even if it is part of a multi-causal complex where grouping all the others causes are grouped under the general label of confounders. It is not by chance that the classical study designs used in epidemiology are well suited for this task. During the 1990s several leading epidemiologists intensified alerts about the insufficiency and limits of such strategies used in epidemiology [4-8]. While the focus of their criticisms varied from theoretical or paradigmatic questions to more applied ones, together they raised serious concerns about the role of epidemiology to fulfill its mission of producing sound knowledge and informing effective actions compatible with present and future population health needs. As a consequence, this is a special moment in which the clarification of aspects related with causation could help epidemiology to overcome its epistemological difficulties and keep on track with its mission, including the study of the determinants of health-related states or events. It is worth noting that this is not a crisis restricted to epidemiology; it also affects different disciplines involved in the understanding of human and social events, and the roots of their causes. We are living in a moment when crucial philosophical and scientific questions are being debated. More than ever words like chaos, complexity, dynamic models, etc. have been present in the philosophical, scientific and lay literatures. In this context, the causality debate has been intensified, but now in connection with the complex ways that human and social events are now perceived. Causality is understood as a relational phenomenon, which has theoretical, but also practical implications. A cause can be the presence or the absence of an action depending of the position of the observer. A cause can increase or decrease the occurrence of a health-related event – causation and prevention are different faces of the same coin. A cause is always the outcome of other causes. But, as far prevention is concerned, when suppression or activation of a specific cause is feasible and generates the desired prevention, there is no immediate need to understand its own causes. A cause is an analyzable factor but some times it is also consequence of a deliberate intervention – the quality of environment affects health, but a clean environment can be a natural occurrence in forests or can be the consequence of an intervention directed to decrease pollution in urban places. Similarly, housing is an important factor related to health-events, but it also consequence of programs implemented to change the housing situation of a given population. Epidemiology has used two different approaches to study causes of the health-related events: the experimental and the observational. Experimental studies (frequently randomized community trials) evaluate a cause by comparing similar groups of individuals exposed and non-exposed to an intervention targeted to suppress or stimulate this cause. This characteristic means that randomized trials are considered by many to be the ultimate standard for definition of a causal association. For some radical minds, the observational approach cannot even be considered in causal discussion! However, for several reasons, including operational and ethical ones, a great part of the causal knowledge accumulated in epidemiology comes from observational studies where very often the comparison groups are not similar or even do not exist. Consider the situation where the vaccine X is a "cause". While it is possible to test its effect using observational studies there is a great consensus that this is best done by a randomized trial. However, if the situation of a vaccination program using the same vaccine, a new causal problem is born, and use of a randomized trial is not as feasible as before. And for a great number of causal problems, experimental studies are totally unfeasible. In some special cases, the two approaches could be used complementarily: for instance, the effect of vitamin A deficiency on child morbidity (diarrhea and acute lower respiratory infections) was verified using observational studies and later, the effect of vitamin A supplementation of deficient children on morbidity was tested experimentally. The questions put forward by Greenland in the paper published here [9] is part of a great effort made by him and others [10-12] to understand the complex nature of causation. In this intricate world, advancements in the understanding of causality is a task that needs to conjugate very wise philosophical (but not metaphysical) and empirical (but not empiricist) perspectives. Epidemiologists have a difficult time as they must see causality as a dynamic and multivariate process, but without losing the opportunities for prevention and without getting lost in the web of causation.

Are social causes so different from all other causes? A comment on Sander Greenland

Abstract: Sander Greenland's elegant paper [6] raises deep questions about the way in which we choose and evaluate public health actions. We agree with the main thrust of his argument with respect to public health policy. We have some concerns, however, about his treatment of social causes, and on this point we focus our critique. Greenland proposes a counterfactual definition of a cause which is framed in terms of alternative actions with different potential outcomes. He suggests that, under this framework, social conditions – such as socioeconomic status, sex, or race – present a quandary for causal inference. They cannot be considered as causes unless they can be reframed in terms of alternative actions. Greenland's approach to this problem is, thankfully, not to dismiss social causes, but rather to "identify potential causes within ordinary events". He suggests that we consider alternative actions to change social conditions, and their different potential outcomes. We do not believe it is warranted to single out the identification of social causes or characteristics as posing an especially severe problem for causal inference. Even under this framework of alternative actions, similar problems pertain to all kinds of exposures in observational studies. It is important to acknowledge this similarity, because if social exposures are perceived as being the most problematic for studies of causation, investigations of these causes and their remediation may be put at a disadvantage. First, in our identification of their causal effects, most of the exposures we study are more akin to conditions than actions. To infer a causal relationship for an exposure, we imagine that we could remove that exposure while "all other things remained equal", and compare the outcomes under the exposed and unexposed conditions. In other words, we compare what happened under the condition of exposure with what would have happened under the condition of no exposure with "all else held constant".1 This is not truly equivalent to a comparison of the outcomes of alternative actions, such as removing or not removing the exposure. If we had actually removed the exposure, all other things would not have remained equal. Most of the causes we study are similar to social causes in this respect. Consider the classic example of smoking cigarettes. If people were unable to smoke cigarettes, they might as a result drink more alcohol, have more episodes of depression, or gain more weight. All other things would not remain equal. Thus, when we construct a counterfactual that compares smoking with no smoking, we are not truly comparing the potential outcomes of alternative actions. We are constrained to comparing the outcomes under two alternative conditions, one of which is necessarily counterfactual. Second, most of the exposures we examine can be seen as the consequence of a previous action, and as a possible mediator of its effect on health. As noted earlier, Greenland suggests that we could reframe social causes (e.g. years of education) as potential outcomes of previous alternative actions (e.g. better schools to improve educational outcomes), which may in turn improve health. But with equal legitimacy, most other exposures could be reframed in the same way. We could reframe physical activity as a potential outcome of previous alternative actions to increase physical activity, which may in turn improve health. For all these exposures, we continually strive to better understand both the antecedents which lead to the exposures, and the biological mechanisms which connect them to disease outcomes. Third, although some conditions turn out to be more manipulable than others, we do not often know beforehand which ones they will be. A researcher's judgment about what conditions are and are not manipulable tends to be influenced more by values than by scientific empirical data. We certainly have no empirical data to support the view that social causes are generally less manipulable than others. On first impression, cigarette smoking may appear to be a readily manipulable action. But it has turned out to be extremely difficult to reduce the smoking epidemic worldwide over the past half century. While cigarette consumption per person has declined in some high income countries over the past few decades, the global impact on health continues to accumulate, with a predicted rise in smoking-related illness and death in low and middle income countries, where the vast majority of the world's more than one billion smokers now live [1,2]. On the other hand, raising levels of education, which may at first seem a more difficult task, has actually been achieved throughout much of the globe over the same period. With regard to "fixed" characteristics such as sex, we seek to modify their relationship to health and disease, by manipulating biological and social experiences alike. Notwithstanding these differences, we concur with Greenland on his central point about the formulation of public health policy. The effects of public health actions and policies "do not correspond to simple cause removal" {editor, citation to Greenland}. Therefore, we should clearly differentiate two endeavors: the identification of causes and the evaluation of interventions to remove these causes. Suppose we initiate a public health action to reduce smoking in a population. Whatever action that may be (e.g. banning the production and sale of cigarettes), it is not plausible to think that it could result in a population in which smoking cigarettes had been reduced while all other things remained equal. The population will change in its composition and historical time; the decline of the cigarette industry may lead to unemployment and consequent ill health in some regions, the opening of new markets for cigarettes in other areas, and so on. Epidemiologic studies of causes provide crucial clues to the design of preventive interventions, but they do not provide good estimates of the impact of these interventions. Like Greenland, we advocate studies that directly compare the effects of alternative public health actions (one of which may be inaction). We also believe that, insofar as possible, these studies should compare the effects of alternative actions across multiple health domains rather than only a single domain. Finally, we suggest that although epidemiologists most often study conditions rather than actions, counterfactual reasoning is applicable to our discipline. The inclusion of conditions as causes has a long tradition under counterfactual reasoning. In an early contribution to counterfactual reasoning about causation, the philosopher Mackie [3] argued that we must consider conditions as well as actions to be potential causes (chapter 2). In the field of psychology, Shadish et al. [4] adopt the counterfactual approach to defining causes, and explicitly state that the causes so defined include nonmanipulable as well as manipulable events. From epidemiology, we quote Rothman and Greenland: "We can define a cause of a specific disease as an antecedent event, condition, or characteristic that was necessary for the occurrence of the disease at the moment it occurred, given that other conditions are fixed." [5] (p. 8, our italics).

Assessing influenza-related mortality: Comment on Zucs et al.

Abstract: Influenza is an important source of mortality and morbidity, and an important public health priority. Measuring the health burden imposed by influenza viruses is an important, and still controversial, question. Some authors argue that influenza is directly or indirectly responsible for the majority of seasonal excess deaths in temperate countries [1], while others argue that they trigger only a small minority [2]. Retrospective cohort studies have shown a surprisingly large protective effect of influenza vaccination against deaths from any cause [3-5], and one author has provocatively suggested that increased influenza vaccination of the elderly could halve the total mortality rate [6]. The population-level interpretation of these cohort studies is not clear, however [7], and studies in the Netherlands [8] and Britain [9] have found substantially lower protection. The present contribution [10] is a welcome addition to the data base that can be used to address this important question. But much more remains to be done to standardize and improve methods, and to reconcile the results obtained from different approaches. The impact of influenza is difficult to measure, because there is a great deal of influenza-like illness (ILI) in the world, caused by a large number of viruses, and only a very small percentage of cases is confirmed virologically [11]. Deaths triggered by influenza may be attributed to a number of final causes, including pneumonia, heart disease and stroke, and may occur weeks after initial infection [1,12,13]. The work of [10] is based on a long tradition of estimating influenza deaths by inference from seasonal patterns in death series. This method was pioneered by Serfling [14], and further developed by workers including Simonsen and colleagues [7,15]. While this work is valuable, and has produced apparently robust results, it is largely disconnected from quantitative virologic data about influenza. Thus, attribution of health effects to influenza is based strongly on assumptions about underlying death trends. Recently, Thompson and colleagues have used virologic surveillance data to estimate influenza mortality [13] and hospitalizations [16], using a weekly, seasonal regression model. These models are the first to link quantitative virologic information to measures of influenza burden at the population level. The estimates produced are consistent with those of Serfling-like estimates [7,17] and are robust to the addition of estimates of respiratory syncytial virus (RSV) prevalence to the regressions. Questions remain about these estimates, however. The work of Thompson and colleagues removes a sinusoidal trend (fit at the same time as influenza prevalence), but does not take into account issues of autocorrelations; or the possibility of seasonal confounding between influenza prevalence, morbidity and mortality, and such factors as day length, temperature or school terms; or the likelihood that deaths caused by influenza infection in a given week may not occur until several weeks later. Keatinge, Donaldson and colleagues [2,18], also used simple regression methods that ignored autocorrelations and seasonal confounding to study the causes of winter mortality in Europe. Unlike Thompson and colleagues, they lacked virological data and instead used proxies for influenza, but included temperature data, and found that temperature rather than influenza explained most of the excess deaths in their models. Approaching a consensus on the health and mortality burden of influenza, and on the cause of winter excess mortality in general, is an important scientific and public-policy goal. For this to happen, further progess is needed in several areas. • Employing virological data. When possible, analyses of influenza burden should be tied to estimates of laboratory-confirmed influenza cases. In some cases, such measures can be combined with ILI surveillance to improve estimates. Efforts should be made to increase the amount of viral surveillance information available in the public domain, with spatial and temporal break downs. • More sophisticated statistical analyses. Time series methods that address issues of seasonal confounding and autocorrelation are available [19,20], but have been little used in analyses of seasonal mortality. Helfenstein [21] analyzed paired pre-whitened mortality series and inferred a "hidden relation" underlying heart disease in women and men. Much more needs to be done to investigate the relationship between mortality (or morbidity) and co-factors including weather, air pollution and epidemics of influenza and other viruses, while accounting for seasonal confounding and autocorrelations. Methods should evaluate multiple risk factors and consider the possibility of interactions between them. • Discuss and define time scales. An important, and usually unasked, question in comparing results from different estimation approaches is the time scale on which influenza deaths are being measured. Everybody dies, so what is being measured as the mortality burden of influenza (or of weather) is deaths that are hastened by the cause in question. The question is whether these deaths are being hastened by weeks, months or years. Regressions that use a weekly time frame are expected to count deaths hastened by even a few weeks, while traditional methods of summing excess deaths over a season of 3 months or longer will be measuring at a different time scale. • Quantitative spatial comparisons. As regional surveillance data, and data from different countries, become more available, analyses that explicitly incorporate risk factors and health outcome variables from various localities have the potential to greatly increase statistical power and shed light on unravelling the contribution of influenza and other risk factors to mortality and morbidity. The contribution of influenza to morbidity and mortality – and, more broadly, cataloging the causes of daily and seasonal excess deaths and hospitalizations – remain as unresolved questions with important scientific and public-health implications. There is a pressing need for more communication between researchers studying different causes, places and time scales, and for application of appropriate, powerful statistical methods to these questions.

Author's response to comments on "Epidemiologic measures and policy formulation".

Abstract: I see no important disagreement between me and the commentators, but disagreement does exist on these topics. I would classify opposing views into two major categories. On one side, some writers assert that causal inference can and should be done without counterfactuals. In my view, Dawid [1,2] is a moderate on that side, as he deploys devices that are isomorphic to counterfactuals, so the difference seems mostly one of labeling and emphasis (which is not always unimportant). More radical counterfactual deniers include Shafer [3], who appears mostly upset because counterfactual models continue to weave into the foundation of statistics, econometrics, sociology, and health sciences, while his approach [4] appears destined for the dustbin, along with other truly noncounterfactual theories of causation. Pearl [5] gives a succinct account of the failings of these theories (reference 5, section 7.5), noting (as do Greenland and Brumback [6]) that the causal models entering into scientific teaching and application (including causal graphs, causal "pies," and structural equations) have mappings into counterfactual formalisms. The target of my article was the other side, those who take counterfactuals uncritically or superficially, without paying enough attention to what these hypothetical quantities are supposed to mean. My article originated as a chapter in a WHO volume [7]. This volume arose from a conference which seemed a festival of counterfactual abuse, rife with talk of cause-of-death removal as if it were an intervention. It is disheartening if not frightening to witness discussion of global health policy framed in such terms. In this context, the concerns expressed by Dawid [1,2] about counterfactuals seem reserved. Counterfactual abuse can be diminished by connecting potential outcomes to interventions. Susser and Schwartz [8] point out that this connection is needed for lifestyle risk factors (smoking, physical inactivity, etc.) just as for social factors. I agree; risk-factor epidemiology could better serve public health if it addressed what could be done, rather than estimating effects of the unattainable (like removal of all tobacco exposure), without regard to how change is brought about. Becoming more realistic involves more than just operationalizing the exposure (right-hand) side of the structural equation; one also needs to expand the left-hand side to consider the full spectrum of intervention effects, such as all effects of smoking cessation (e.g., weight gain, depression). The traditional narrow focus on a few prominent endpoints (like cancer and cardiovascular disease), encouraged by the case-control viewpoint, has discouraged grappling with the multivariate complexity of outcomes as well as exposures. Worse, in the smoking context, there may be a bias against acknowledging that one of our most damaging population exposures (tobacco) may bring worthwhile benefits to a non-negligible portion of the population – not just medical benefits like Parkinsonism prevention, but also psychologic benefits like enhanced sense of well-being, which are hard to measure and weigh against costs. If we really believe in informed consent, then we must inform the public about how lifestyle choices are not just about lifespan maximization, but are also choices of how to live and die. This view will not sit well with those for whom good sensations are evil if the sensations do not come from sanctified sources like religious faith, licensed entertainment, or prescription drugs.