Key goals of global risk assessment
An effective risk assessment must have a well-defined scope, which in turn depends on the purpose of the analysis. For example, an evaluation of emissions from a particular industrial facility is likely to concentrate on their health effects on local populations. In contrast, a project to set national environmental priorities may be much broader in scope, covering such factors as emissions of greenhouse gases and ozone-depleting substances. Some trade-offs will inevitably be required. Governments and ministries of health oversee overall population health and so, at the broadest level, need information from risk assessments that are comprehensive as well as being reliable, relevant and timely. Because the range of risks to health is almost limitless, it is essential for governments to have a quantitative approach to gauging their importance. Risks need to be defined and studied comprehensively irrespective of factors such as their place in a causal chain or the methods used (from the disciplines of the physical, natural, health, and social sciences) for their analysis. The following sections outline some of the different dimensions that should be considered.
Standardized comparisons and common outcome measures
Ideally, the impact of each risk factor should be assessed in terms of a "common currency" that incorporates loss of quality of life as well as loss of life years. The principal metric used in this report is the DALY (disability-adjusted life year) -- one DALY being equal to the loss of one healthy life year (13).
A key initial question when assessing the impact of a risk to health is to ask "compared to what?" This report employs an explicit counterfactual approach, in which current distributions of risk factors are compared with some alternative, or counterfactual, distribution of exposure. Many different counterfactuals are potentially of interest. To enhance comparability across risk factors, the basis for the results in Chapter 4 is the theoretical minimum risk distribution, that is exposure levels that would yield the lowest population risk (for example, no tobacco use by any members of a population). For the analysis of the costs and effects of interventions to reduce risk in Chapter 5, a related counterfactual is used -- based on the burden that would exist in the absence of relevant interventions. Risk factor distributions that are plausible, feasible and cost-effective will lie somewhere between the current risk factor levels and the related theoretical minimum. The envisaged shift from current to counterfactual scenarios has been termed the distributional transition (see Figure 2.1).
In many instances, the counterfactual of most relevance will involve small to moderate distributional transitions (for example, 10%, 20% or 30%), as these are most likely to be feasible and cost-effective. These estimates are also less susceptible to the influence of arbitrary choices of theoretical minima, and are likely to be the most reliable, as the dose--response is often least certain at low exposure levels.
Assessing protective as well as hazardous factors
Factors that affect risk of disease or injury are, of course, not all harmful. Risk factor does have a negative connotation, but ideally a risk assessment should include a range of protective as well as hazardous risk factors. For example, this report considers the protective benefits of fruit and vegetable intake and physical activity by assessing people with low levels of these factors. The important role of protective factors in adolescent health is outlined in Box 2.2.
Box 2.2 Protective factors
A growing body of cross-cultural evidence indicates that various psychological, social and behavioural factors are protective of health in adolescence and later life. Such protection facilitates resistance to disease, minimizes and delays the emergence of disabilities, and promotes more rapid recovery from illness.
Among the psychosocial factors that have been linked to protection in adults are: an optimistic outlook on life with a sense of purpose and direction, effective strategies for coping with challenge, perceived control over life outcomes, and expressions of positive emotion. Epidemiological studies have shown reduced morbidity and delayed mortality among people who are socially integrated. The quality of social relationships in the home (parent--child relations and spousal ties) and the workplace (employer--employee relations and coworker connections) are now recognized as key influences on physical and mental health. A growing literature underscores the protective health benefits associated with persistently positive and emotionally rewarding social relationships. Positive health behaviours (e.g., proper diet and adequate exercise, and avoiding cigarettes, drugs, excessive alcohol and risky sexual practices) are also influenced by psychosocial factors.
The presence of psychosocial factors in understanding positive human health points to new directions for research and practice. The biological mechanisms through which psychosocial and behavioural factors influence health are a flourishing area of scientific inquiry: investigations in affective neuroscience are relating emotional experience to neural structures, function, dynamics and their health consequences. There is a need for greater emphasis in policy and practice on interventions built around the growing knowledge that psychosocial factors protect health.
Adolescence is a critical life stage when lifestyle choices are established, including health-related behaviours with impacts throughout life. Recent research has begun to focus on the role of protective factors in youth behaviour, complementing previous approaches concerned only with problems and risk taking.
Evidence from 25 developing countries, 25 European countries, Canada, Israel and the United States shows that adolescents who report having a positive connection to a trusted adult (parent or teacher) are committed to school, have a sense of spirituality and exhibit a significantly lower prevalence of risky behaviours. This is in addition to being more socially competent and showing higher self-esteem than adolescents without such a connection. Studies in the US have shown that these protective factors also predict positive outcomes (remaining connected to school, engaging in more exercise and having healthy diets) while diminishing negative behaviour (problem drinking, use of marijuana and other illicit drugs, and delinquent behaviour).
Protective factors promote positive behaviours and inhibit risk behaviours, hence mitigating the impacts of exposure to risk. Current efforts to reduce risks in the lives of adolescents should be broadened to include the strengthening of protective factors.
Including proximal and distal causes
Risks to health do not occur in isolation. The chain of events leading to an adverse health outcome includes both proximal and distal causes -- proximal factors act directly or almost directly to cause disease, and distal causes are further back in the causal chain and act via a number of intermediary causes (see Figure 2.2). The factors that lead to someone developing disease on a particular day are likely to have their roots in a complex chain of environmental events that may have begun years previously, which in turn were shaped by broader socioeconomic determinants. For example, society and culture are linked to certain drinking patterns, which in turn influence outcomes such as coronary heart disease via physiological processes such as platelet aggregation. Clearly, there are risks over which an individual has at least some control (for example, inactivity) and risks that mostly or entirely rest at a population or group level (for example, ambient air pollution). It is essential that the whole of the causal chain is considered in the assessment of risks to health. Indeed, many risks cannot be disentangled in order to be considered in isolation, as they act at different levels, which vary over time. An appropriate range of policies can be generated only if a range of risks is assessed.
There are many trade-offs between assessments of proximal and distal causes. As one moves further from the direct, proximal causes of disease there can be a decrease in causal certainty and consistency, often accompanied by increasing complexity. Conversely, distal causes are likely to have amplifying effects -- they can affect many different sets of proximal causes and so have the potential to make very large differences (20). In addition, many distal risks to health, such as climate change or socioeconomic disparity, cannot appropriately be defined at the individual level. A population's health may also reflect more than a simple aggregation of the risk factor profile and health status of its individual members, being a collective characteristic and a public good that in turn affects the health status of its members (21).
Research into the different levels of risks should be seen as complementary. There is considerable importance in knowing the population-level determinants of major proximal risks to health such as smoking. Similarly, there is value in knowing the mechanisms through which distal determinants operate. Understanding both proximal and distal risks requires contributions from different scientific traditions and different areas of health impact: environmental, communicable, noncommunicable, injury, and so on, and as a result different intellectual tools and methods, including those of health, physical and social sciences. This in turn requires consideration of the context of particular risks: some are likely always to have negative health effects (for example, tobacco use) while others may have a role that changes from setting to setting (for example, breastfeeding protects against diarrhoeal disease, to an extent that depends on the prevalent patterns of diarrhoea). Also, the same risk can be measured and quantified at various levels depending on measurement technology and policy needs. For example, measuring iodine levels in food and in the environment requires different tools and the results have different implications.
When distal exposures operate through different levels of risk factors, their full impact may not be captured in traditional regression analysis methods in which both proximal and distal variables are included. More complex multilevel models and characterization of causal webs of interactions among risk factors may lead to more appropriate estimates, as well as facilitating estimation of the effect of simultaneous changes in two or more risk factor distributions. Some examples are shown later in the report.
Risk factors can also be separated from outcomes in time, sometimes by many decades. Box 2.3 shows how disadvantage can be accumulated across the life course.
Box 2.3 Risks to health across the life course
In recent years, a life-course approach to the study of health and illness -- which suggests that exposure to disadvantageous experiences and environments accumulates throughout life and increases the risk of illness and premature death -- has helped to explain the existence of wide socioeconomic differentials in adult morbidity and mortality rates.
Chronic illness in childhood, more common among children of manual workers, can have long-term consequences both for health and socioeconomic circumstances in later life. Slow growth in childhood (short stature for age and sex) is an indicator of early disadvantage. Early material and psychosocial disadvantage may also have an adverse impact on psychological and cognitive development, which in turn may affect health and labour-market success later in life. The impact of living and working environments -- and lifestyle factors such as smoking -- on health inequalities has long been recognized. Cumulative differential lifetime exposure to health-damaging or health-promoting environments appears to be the main explanation for observed variations in health and life expectancy by socioeconomic status.
Disadvantage may begin even before birth: low birth weight is associated with increased rates of coronary heart disease, stroke, hypertension and non-insulin-dependent diabetes. These associations extend across the normal range of birth weight and depend on lower birth weights in relation to the duration of gestation rather than the effects of premature birth. The associations may be a consequence of "programming", whereby a stimulus or insult at a critical, sensitive period of early life has permanent effects on structure, physiology and metabolism. Programming of the fetus may result from adaptations invoked when the maternal--placental nutrient supply fails to match the fetal nutrient demand. Although the influences that impair fetal development and programme adult cardiovascular disease remain to be defined, there are strong pointers to the importance of maternal body composition and dietary balance during pregnancy.
Assessing population-wide risks as well as high-risk individuals
Many risks to health are widely distributed in the population, with individuals differing in the extent of their risk rather than whether they are at risk or not. Binary categorization into "exposed" and "unexposed" can substantially underestimate the importance of continuous risk factor--disease relationships. Consequently, much of this report estimates the effects of shifting distributions of exposures by applying a counterfactual approach, that is, by comparing the burden caused by the observed risk factor distribution with that expected from some alternative, or counterfactual, distribution. This approach allows assessment of population-wide interventions (see Box 2.4 and Figure 2.3).
Box 2.4 Population-wide strategies for prevention
"It makes little sense to expect individuals to behave differently from their peers; it is more appropriate to seek a general change in behavioural norms and in the circumstances which facilitate their adoption." -- Geoffrey Rose, 1992.
The distribution and determinants of risks in a population have major implications for strategies of prevention. Geoffrey Rose observed, like others before and since, that for the vast majority of diseases "nature presents us with a process or continuum, not a dichotomy". Risk typically increases across the spectrum of a risk factor. Use of dichotomous labels such as "hypertensive" and "normotensive" are therefore not a description of the natural order, but rather an operational convenience. Following this line of thought, it becomes obvious that the "deviant minority" (e.g. hypertensives) who are considered to be at high risk are only part of a risk continuum, rather than a distinct group. This leads to one of the most fundamental axioms in preventive medicine: "a large number of people exposed to a small risk may generate many more cases than a small number exposed to high risk". Rose pointed out that wherever this axiom applies, a preventive strategy focusing on high-risk individuals will deal only with the margin of the problem and will not have any impact on the large proportion of disease occurring in the large proportion of people who are at moderate risk. For example, people with slightly raised blood pressure suffer more cardiovascular events than the hypertensive minority. While a high-risk approach may appear more appropriate to the individuals and their physicians, it can only have a limited effect at a population level. It does not alter the underlying causes of illness, relies on having adequate power to predict future disease, and requires continued and expensive screening for new high-risk individuals.
In contrast, population-based strategies that seek to shift the whole distribution of risk factors have the potential to control population incidence. Such strategies aim to make healthy behaviours and reduced exposures into social norms and thus lower the risk in the entire population. The potential gains are extensive, but the challenges are great as well -- a preventive measure that brings large benefits to the community appears to offer little to each participating individual. This may adversely affect motivation of the population at large (known as the "prevention paradox").
Although most often applied to cardiovascular disease prevention, a population-wide approach is often relevant in other areas. For example, a high-risk strategy for melanoma prevention might seek to identify and target individuals with three or more risk factors (such as a number of moles, blonde or auburn hair, previous sunburn, and a family history of skin cancer). However, only 24% of cases of melanoma occur in this 9% of the population, so a targeted approach would succeed in identifying those at high risk but would do little for population levels of melanoma -- 75% of cases occur in the 58% of the population with at least one risk factor. A population-wide strategy would seek to make sun protection a social norm, so that the whole population is less exposed to risk.
These approaches are complementary: a population approach can work to improve and extend the coverage of a high-risk approach. A key challenge is finding the right balance between population-wide and high-risk approaches. Rose concluded that this will require a wider world view of ill-health, its causes and solutions, and will lead to acknowledgement that the primary determinants of disease are mainly economic and social, and therefore remedies must also be economic and social.
Including risks that act together to cause disease
Many risks to health act jointly to cause disease or injury, and this has important implications for prevention opportunities, as outlined in Box 2.5. This report presents estimates of the individual effects of different selected risks to health, followed by analyses of the joint effect of selected clusters of risks.
Box 2.5 Multiple causes of disease
The impact of a single risk factor on disease is often summarized as the proportion of disease caused by, or attributable to, that risk factor. The fact that diseases and injuries are caused by the joint action of two or more risk factors means that the sum of their separate contributions can easily be more than 100%. Consider a hypothetical situation of deaths from car crashes on a hazardous stretch of road. Studies may have shown that they could be reduced by 20% by using headlights in daytime, 40% by stricter speed limits, 50% by installing more traffic lights, and 90% by creating speed bumps.
As a further example consider a smoker, also a heavy drinker, who develops throat cancer. The cancer would not have developed on that particular day if the person had not smoked or drunk heavily: it was very likely caused by both tobacco and alcohol. There are three possible scenarios for throat cancer, each with a different set of causes that must be present for the disease to occur. In the first scenario, smoking and alcohol work together with other environmental and genetic causes to result in the disease ("environmental" can be taken as all non-genetic causes). The second scenario is the same, except that throat cancer develops in a non-drinker. In the third, we do not know what caused the cancer, other than genetic and some unknown environmental causes. This simplified model illustrates the following important issues.
- Causes can add to more than 100%. If the scenarios were equally common, 66.6% of throat cancer would be attributable to smoking, 33.3% to alcohol, 100% to genetic causes, and 100% to unknown environmental causes, making a total of 300%. Causes can, and ideally should, total more than 100%; this is an inevitable result of different causes working together to produce disease, and reflects the extent of our knowledge of disease causation.
- Multicausality offers opportunities to tailor prevention. If these scenarios were numerically correct, throat cancer could be reduced by up to two-thirds with smoking cessation, by up to one-third with reduced alcohol intake, or by up to two-thirds with less marked decreases in both smoking and alcohol consumption. Further reductions could also take place if research led to additional preventive strategies based on genetic or other environmental causes. The key message of multicausality is that different sets of interventions can produce the same goal, with the choice of intervention being determined by such considerations as cost, availability and preferences. Even the most apparently single-cause conditions are on closer inspection multicausal; the tubercle bacillus may seem to be the single cause of tuberculosis but, as improved housing has been shown to reduce the disease, living conditions must also be considered a cause.
- Prevention need not wait until further causes are elucidated. In the foreseeable future we will not know all the causes of disease, or how to avoid all the disease burden attributable to genetic causes. Nonetheless, multicausality means that in many cases considerable gains can be achieved by reducing the risks to health that are already known.
Using best available evidence to assess certain and probable risks to health
It is important in any risk assessment to review quantitatively the best available evidence for both "definite" and "probable" risks. Estimation of the potential impact of a health hazard can never wait until perfect data are available, since that is unlikely to occur. Timeliness is essential. This area can be a source of tension between scientists and policy-makers. However, arguments are often clouded by the use of dichotomies -- assertions of uncertainty or certainty when, in fact, there are different degrees of uncertainty and disagreement about tolerable thresholds. Similarly, it may be asserted that there are no data when some indirect data are available, or at least the range of levels in other parts of the world is known. For example, in estimating fruit and vegetable intake for countries with no known surveys on this topic, upper and lower ranges can be estimated from surveys undertaken elsewhere, and food sales and agricultural data can be used to produce indirect estimates that occupy a narrower range. Internal consistency can help put ranges on uncertainty: for example, mortality rates, population numbers and birth rates should be internally consistent, and reliable estimates for some of these components will put bounds on the uncertainty of the others. However, as outlined earlier, the sum of causes is unbounded and so internal consistency checks cannot be performed in assessments of different risks to health. Strategies to minimize this problem include full documentation of data sources, methods and assumptions, extensive peer review, explicit assessments of causality, and quantitative estimates of other uncertainty.
Extrapolations and indirect methods are often justified where there are implications in delaying estimates of health impacts and subsequent policy choices. If decisions await improved estimates, then not producing best current estimates (with appropriate indications of uncertainty) may mean inappropriate inaction. Alternatively, decisions may be made with other even more uncertain information, where the uncertainty will often be implicit. Nonetheless, there can be costs in making incorrect estimates and, ultimately, it is largely a matter of judgement to decide when data are adequate.
Whenever possible, the level of uncertainty should be reported explicitly in risk assessments. There is still considerable debate about how this is best done in a policy-relevant way, given the inevitable play of chance and uncertainties in both the likelihood of causality and the validity of the estimation methods. Major uncertainty should result in calls for more data. In particular, data are often absent or scanty in the developing countries, where many risks are highest and more information could produce the greatest gains in knowledge. The management of highly uncertain risks and the use of the precautionary principle are discussed in Chapter 6.
Assessing avoidable as well as attributable burden
Risk assessments to date have typically used only attributable risk estimates, basically addressing the question "what proportion of current burden is caused by the accumulated effects of all prior exposure?" However, often a more policy-relevant question is "what are the likely future effects of partial removal of current exposure?" Two key developments are therefore needed: an explicit focus on future effects and on less-than-complete risk factor changes. This report presents estimates of attributable burden (current burden due to past exposure) and of avoidable burden (the proportion of future burden avoidable if current and future exposure levels are reduced to those specified by some alternative, or counterfactual, distribution). When the time between exposure and disease or death is short, the distinction between attributable and avoidable burden is not critical. However, for risk factors such as tobacco and some occupational exposures, a long time lag between exposure and health outcome may result in a major difference between attributable and avoidable burden. The distinction between attributable and avoidable burden is shown graphically in Figure 2.4.