Journal of Public Health Advance Access originally published online on July 11, 2008
Journal of Public Health 2008 30(3):232-233; doi:10.1093/pubmed/fdn056
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Rejoinder
Ian McDowell, Professor of Epidemiology
Department of Epidemiology and Community Medicine, Faculty of Medicine, University of Ottawa, 451 Smyth Rd, Ottawa, Ontario, Canada
Address correspondence to Ian McDowell, E-mail: mcdowell{at}uottawa.ca
Professor Bhopal argues that the apparent complexity of some health problems may be a mirage; the solutions may be staring us in the face. This is a really good challenge: a sort of Dr Seuss debate between those who would over-complicate the world and others who would over-simplify matters. I argue that both are needed, for distinct but complementary purposes; I am saddened by academic disciplinary battles that seem to pit one against the other. The simplifiers can often provide remarkably accurate estimates of useful parameters such as how many cases of cancer the local hospital can anticipate next year; this is the tradition of survey research. But the complexifier in me argues that this represents a surface description and not an understanding. Having observed family members struggle with body weight (to use Professor Bhopal's example), I am far from satisfied that obesity should be seen in terms of the balance of energy intake and expenditure; there is so much individual variability that a simple presentation of the process seems inadequate. While energy in and energy out may be central, I argue that there are a number of nonlinear processes that we understand neither at the metabolic, nor at the behavioural, nor at the social level.
Professor Bhopal argues from a firmly epidemiological perspective. He proposes, for example, that the risk factors have more explanatory value at the population than the individual level. I am arguing that the risk factors can describe, rather than necessarily explain, at the population level. This may be useful, but is ultimately not sufficient. I am also uncomfortable with his contrast between populations, in which ‘the risk of disease lies between 0 and 100%...’ and individuals for whom the ‘risk of disease is either 0 or 100%, never in between.’ If space permitted, we could debate the correspondence between the population prevalence of a disease and the notion of its risk; but here let me just comment on Professor Bhopal's binary view of disease. For virtually every health condition, a binary division is convenient, but is arbitrary and limits understanding. For example, the extension of dementia studies to include mild cognitive impairment has offered significant epidemiological insights,1 and even an apparently discontinuous event such as a motor vehicle collision could better be viewed as part of a pattern of risky driving practices and near-misses. The binary view tends to encourage needless disputes over prevalence and incidence figures that can depend critically on the precise case definition; it also reduces the predictive power of risk equations. Virtually every disease can benefit from being viewed dimensionally, along continua of time, intensity and invasion (or spread within the person). The binary approach may be sufficient for descriptive epidemiology but it places serious constraints on progress towards a fuller understanding.
In effect, Dr Schooling's comment that findings from one population often cannot be generalized to others echoes my concern over an exclusive focus on an epidemiological approach. Her comment recalls the dilemma of what is ‘a population’, the definition of which is often side-stepped in the epidemiological literature. Perhaps, when we come to view a population as an organism rather than merely as a denominator,2 important dimensions of etiological processes emerge. Dr Schooling and Professor Diez Roux both warn that the world moves on and will not wait for academics to provide the ultimate explanation: preventive approaches have to be developed, and fast. I strongly concur, and also agree with Dr Schooling that care is needed to ensure that these are based on adequately sophisticated causal analysis that is yet not needlessly detailed. However, I do think that most individuals are exceedingly interested in ‘why one individual rather than another gets a disease’, even though this may not be the focus of public health. There will be a strong market for such understanding and we should continue work on blending contributions from our various disciplines to provide it.
I found myself agreeing virtually with everything that Professor Woodward said; his tongue-in-cheek representation of risk factor research as that without lateral vision suggests that he, too, is looking for a research paradigm that can blend inputs from several disciplines. But he makes the important point that the precise nature of the question—he takes the example of motor vehicle crashes—will influence the blend of disciplinary inputs. Although this is correct, we should not forget that the choice and framing of the question may reflect our disciplinary orientation and so we have to try to step back to ensure that we really are asking the most insightful questions. However, Professor Woodward's proposal that our difficulties may end if we re-define disease categories to make them specific (‘cigarette cancer’) will surely not resolve the complexity of the up-stream etiological questions.
Professor Diez Roux gives an excellent summary of the issues under debate. She quibbles with my suggestion that we look for causes of cases at the individual level and the causes of patterns of disease at higher levels. I fully agree that we cannot explain a case only in terms of individual-level factors; we need the multiple levels to understand what gave rise to the individual circumstances. My perhaps imperfectly stated point was slightly different: we cannot use individual-level data to explain patterns of cases. It is like compositional versus contextual variables in analyses of place: prevalence is composed of individual cases, but the driver for explaining prevalence must come from outside, from the context. Even the quintessentially personal level, genetics, is ultimately derived from geographical and social patterns of selection and intermixing of gene pools. We could view the individual factors as distorting mirrors that reflect the higher-level factors, with some alteration, back onto the population for which we calculate statistical rates and proportions. Of course, the individual factors remain critical, but they are triggered by environmental factors, so ultimately the explanation for those characteristics comes from outside.
Hence, I agree with Professor Diez Roux that Rose's analysis is equivalent to a regression equation with an intercept that represents the base rate in the population, plus variables that account for individual deviations from this base-rate. Both are needed to understand the individual case. But I argue that the relationship between the two is perhaps more subtle than Rose portrays. Although this may not be the goal of an epidemiologist, achieving a complete etiological picture for a case would involve a causal chain for each individual-level variable (smoking, body weight and so forth) leading outwards to the higher-level factors. Furthermore, these causal chains are dynamically inter-linked, influencing each other. Although these processes may be summarized using individual-level variables, the move to understanding and explanation requires more, including an analysis of complex interrelationships between individual and population layers.
But this is all terribly academic, and all four commentators seem to agree that, for practical research purposes, we will make individual measurements and use these as predictive or explanatory variables in our routine analyses. Parsimony is elegant; removing a pump handle can halt a cholera outbreak and save lives, and Professor Woodward gives other examples. But it is intellectually insufficient, not to mention causing irritating interruptions to the water supply. I agree with Professor Diez Roux that the next challenge will be to develop analytic approaches capable of handling greater dynamic complexity. My goal was to propose some criteria by which new analytic approaches could ultimately be judged. In the meantime, I agree with all of the commentators that we absolutely need to get on with developing and testing interventions, even in the absence of full comprehension of the processes involved.
 |
References |
|---|
 TOP References |
|---|
- Tuokko H, McDowell I. An overview of mild cognitive impairment. Chapter 1. In: Mild Cognitive Impairment: International Perspectives.—Tuokko HA, Hultsch DF, eds. (2006) New York: Taylor and Francis. 3–28.
- McDowell I, Spasoff RA, Kristjansson B. On the classification of population health measurements. Am J Public Health (2004) 94:388–93.
[Abstract/Free Full Text]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||