Philosophy of Epidemiology

A. Broadbent

• Publisher: Palgrave Macmillan
• Year: 2013
• ISBN:
  9781137315601
• Topic: Filosofia
• Subtopic: Filosofia ed etica nella scienza

Epidemiology is one of the fastest growing and increasingly important sciences. This thorough analysis lays out the conceptual foundations of epidemiology, identifying traps and setting out the benefits of properly understanding this fascinating and important discipline, as well as providing the means to do so.

1

Why Philosophy of Epidemiology?

Introducing philosophy of epidemiology

Epidemiology makes headlines more often than most other sciences. Every time you hear that something is good for you or bad for you, either there is an epidemiological study involved, or there is an epidemiological study being planned to check whether the claim is true. (If neither of these is the case, you should get your news somewhere else.) Yet many people have never heard of epidemiology.

Epidemiology is traditionally defined as the study of the distribution and determinants of disease and other health states in human populations for the purpose of improving the health of those populations (for similar definitions see Rothman, Greenland, and Lash 2008, 32; Last 1995). More exact definitions tend to include the manner of study – that is, the use of group comparisons – to emphasise the fact that epidemiology does not merely study the health of individuals or groups but makes comparisons between groups and draws inferences from these comparisons (more on this when epidemiological study designs are examined in Chapter 2). Accordingly the definition we will work with is this:

Epidemiology is the study of the distribution and determinants of disease and other health states in human populations by means of group comparisons for the purpose of improving population health.

No doubt this could be improved, but it covers the central features of the science.

Epidemiology is not usually taught at school or at undergraduate level, except as a component of medicine, where it typically occupies only a small proportion of the syllabus. Even well-educated, scientifically literate people are often hard pressed to say what epidemiology is, unless their Greek is good enough for them to guess.

Maybe this explains why philosophers of science have neglected epidemiology. It is true that there are some philosophers who have thought about epidemiology and many more who have used epidemiological examples without identifying them as such (some of them will be discussed in later chapters). There are also a number of epidemiologists who have taken courses in the philosophy of science at some stage during their training, then sought to apply what they learned to their own discipline (these too will be discussed). But there have been no extended efforts to apply philosophical inquiry to the science of epidemiology in a thoroughgoing way, as has been done with physics, biology, psychology, and a number of other sciences. Although a few philosophers have studied epidemiology, there have been no philosophical studies of epidemiology.

Epidemiology is as philosophically interesting and as worthy of philosophical study as physics, biology, or psychology. As in physics, biology, and psychology, the philosophical issues arising in epidemiology are a mix: some are fresh slants on old problems, and some are specific to the discipline, thrown up in the course of its work. There is no sharp distinction between philosophy of the special sciences and philosophy of science in particular, nor between philosophy of science and philosophy more generally. So this book is not an exercise in intellectual territorialism. Nonetheless, philosophers in general, and philosophers of science in particular, need material to work with, and one way to get that material is to focus on the conceptual and methodological challenges that a particular science faces. That is the spirit in which this study is conducted.

Epidemiology involves a lot of statistics, and statistics is philosophically interesting in its own right. One purpose of this book is to identify philosophical problems in epidemiology that are not primarily problems in the theory of statistics. This is partly for pragmatic reasons: the philosophy of statistics is its own topic and is dealt with admirably in other places. But there is also a threat posed by the growth of statistics. Its increasing importance to a number of scientific disciplines threatens to dwarf concerns that belong to or arise from the domain of those disciplines. Epidemiology uses statistics, but it is not only statistics, and its conceptual challenges are not settled by solving conceptual problems in statistics. This book is meant to illustrate these claims more than to argue for them and, in doing so, to reinforce the sense of epidemiology as a discipline in its own right – something a philosophical treatment of a science ought to do.

This book does not pretend to survey the entire discipline of epidemiology or to identify every philosophical question associated with epidemiology. But it does aim to identify at least some of the big ones and establish some links between them.

Themes

What sort of conceptual and methodological challenges does epidemiology face? What makes epidemiology philosophically interesting? There is no point attempting an exhaustive list, but six features of this young science are salient.

First and foremost, epidemiology focuses on causation: so much so, that some epidemiologists have complained about it (e.g. Lipton and Ødegaard 2005). When epidemiologists seek to identify “determinants” of disease, the determinants they are primarily interested in are not features of the local spatial geometry or truths of logic: they are causes. And when they seek to understand the distribution of disease, that is partly because studying distribution can help in the “hunt” for causes (to borrow Nancy Cartwright’s figure). Epidemiologists do other things, too, but hunting for causes is an overriding characteristic of the most famous episodes in epidemiology and finding them is a characteristic of its most famous successes. The discovery that drinking water contaminated with excrement from cholera sufferers causes cholera; the discovery that pellagra is not an infection afflicting those living in poverty but is caused by diet; the discovery that smoking causes lung cancer – these are epidemiologic milestones, and they all involve the identification of causal connections between what epidemiologists refer to as “exposures” and “outcomes”. Many sciences do this, of course. But scientists may put their data to various uses: discovering “laws of nature” (whatever they may be), developing grand theoretical frameworks, measuring constants, or whatever it may be. Epidemiologists are not concerned with these things at all. They are more or less exclusively concerned with finding causation. This central concern with causation means that epidemiologists think and write about it: about what it is and about how we find out about it. And since philosophers do, too, it would be positively surprising if there were no areas of mutual interest.

The second and third striking features of epidemiology arise from its nonconformity to standard philosophical images of science. Neither experiment nor theory feature prominently in epidemiology (these facts being striking features two and three, respectively). This deprives philosophers of science of two of the most obvious handles by which to get a grip on their subject matter and makes it awkward to use standard philosophical materials for teaching the basis of epidemiological methodology. The awkwardness is usually glossed over in epidemiological textbooks, probably because the authors’ humility leads them to believe that any cracks in the veneer arise from their own misunderstanding. But in fact, philosophical thinking about science – at least the big picture that most philosophers of science seem to operate with – is at fault, because it is a poor fit for epidemiology.

Epidemiology makes central use of “observational” methods, meaning methods that do not involve controlled experiments.¹ Two of the most important families of study – the cohort study and the case-control study – do not involve any intervention on the part of the investigator. In its classic form, the case-control study involves identifying a group of people – “cases” – with the outcome in question and then comparing the prevalence of the exposure among cases to a suitable group of “controls”. For example, Austin Bradford Hill and Richard Doll identified hospital admissions with lung cancer and compared the smoking habits of those patients with smoking habits among patients admitted for other diseases. In a cohort study, information is gathered about exposures in a study group; the group is then followed over a period of time, and outcomes are observed. For example, in the cohort study which followed their case-control study, Doll and Hill sent short questionnaires assessing smoking habits to nearly 60,000 British doctors and collected health information, especially on cause of death (as well as continuing to collect information about ongoing smoking habits). In neither study did they intervene, in the sense in which philosophers and scientists use the term: they did not (deliberately) make anyone smoke or make anyone stop smoking. Philosophers have made much of the role of intervention in science, and it surely characterises experiments, as the term is usually understood. Yet in epidemiology, inferences are often drawn without intervention and thus without experiment.

This is not to say that intervention is entirely absent from epidemiological studies. Epidemiology does have some methods that are typically, and reasonably, described as “experimental”, in the sense that they involve intervention. In particular, randomised controlled trials are a kind of experiment commonly used to assess new medical treatments, especially pharmaceutical ones. In a randomised controlled trial, subjects are divided at random into two groups, one of which receives the treatment, while the other receives a placebo (in the classic version). If the trial is (successfully) double-blinded, then none of the parties (patients, physicians, researchers) know which is which until after the trial is completed.

Randomised controlled trials may be counted as experiments since they involve intervention; but they are still not controlled experiments despite their name. In a controlled experiment, at least in the Millian ideal, there are as few differences as feasible between the control and the item being studied. The investigator seeks, often unsuccessfully, to ensure that among the many inevitable differences between control and test apparatus, there are none that matter to the outcome. In a randomised controlled trial, on the other hand, many differences persist between the subjects of the study.

What is more, these differences are often relevant to the outcome of the study. Subjects differ in age, race, diet, habits of thought, genetic material, and numerous other factors that could be relevant – even among relatively homogenous groups. Some of these characteristics might, in their own right, cause or prevent the outcome being studied, irrespective of the treatment; some might interact with the treatment to produce or prevent the studied outcome. The randomised controlled trial does not literally control these things, as they would be controlled in a laboratory controlled experiment. Rather, the design uses randomisation as a surrogate for control. The aim is to distribute these factors evenly among the treatment and control groups so that their effects get cancelled out. This is not the same as actually controlling relevant variables, even if it achieves the same effect. And whether randomisation does achieve epistemic equivalence with the truly controlled experiment is an important methodological question – one which has attracted some philosophical attention (Worrall 2002, 2007, 2010 2011; Howick 2011).

The absence of controlled experiments from epidemiology, coupled with the focus on finding out about causation, surely helps explain why causal inference is an active topic of debate among epidemiologists. It also means that we cannot involve experiment in any general characterisation of science or of scientific activity unless we are prepared to deny that observational epidemiological studies are scientific. Perhaps that is not news: astronomy is an ancient science and does not centrally involve experiment (although nor does it focus so heavily on causal inference). But epidemiology offers a third, deeper challenge to general accounts of science: its lack of theory.

Epidemiology has developed fast over the last few decades. Yet it has not accumulated theory in the way that other disciplines do. The expertise of an epidemiologist is methodological. Epidemiology discovered that smoking causes lung cancer. But if that claim were now overturned, that would not cause a rift in any great fabric of epidemiological theory. It would certainly have an indirect effect: it might cause epidemiologists to question the methods by which the earlier conclusion had been reached and to question other conclusions reached with those methods, and it would represent a substantial shift in the framework of current biomedical knowledge within which epidemiologists operate. But these are only indirect effects. Imagine what could happen if a central piece of biological knowledge were overturned; for example, suppose that Lamarckian inheritance turned out to be broadly correct for humans (to a much greater extent than currently known epigenetic effects), so that we pass on many more of our acquired characteristics to our offspring genetically than was previously thought. This would have implications: it would not fit with other biological theory, and that theory would need to be changed. Evolutionary biology, cellular biology, our understanding of DNA – these would all need altering.

Epidemiology, on the other hand, does not have a proper domain of theory. The theory behind the claim that smoking causes lung cancer belongs to other branches of the biomedical sciences; epidemiology discovered the causal link and passed it on to them to fit into a theoretical framework. An error might cause epidemiologists to question the methods they used, but there is no grand epidemiological theory on a par with Darwinian evolution by natural selection or general relativity into which they would try to fit this new fact.

The fourth striking feature of epidemiology is the relative domain insensitivity of its methods. In essence, epidemiologists count things and then draw conclusions by comparing the results of different counting exercises. They are interested in counting things that are relevant to health, but since many more things can be counted, this puts direct pressure on the notions of health, disease, and related concepts, as will be explored in Chapter 10. It would be incorrect to claim that epidemiological methods are completely domain-insensitive: implicit or explicit assumptions may underlie the use of particular methods in particular circumstances. Nonetheless, it is clear that modern epidemiology has exerted an expansive pressure on medical science and that this arises from the fact that its core methods can be applied beyond the limits of traditional medical concern. The methods of epidemiology are as suited to examining the relation between levels of internet use and suicide rates as they are to studying how the prevalence of the BRCA1 gene relates to the incidence of breast cancer. This has contributed to a broadening of the scope of health states that are regarded as medically interesting and to a broadening of the range of causes of health states that are thought to be within the scope of medical attention. Obesity is perhaps the best example of a condition which is being hauled into the medical arena by a series of epidemiological studies on both environmental and genetic risk factors. The remedies for obesity may also be unconventional. Thus in some places doctors will be consulted for a weight problem and will prescribe exercise. Epidemiology has played a central role in this expansion. But since the limits of what can be counted and compared are well outside the limits of what is medically significant, there is an interesting philosophical question as to how and where this expansion is to be curbed.

The fifth striking feature is certainly not unique: it is the centrality of population thinking. This term is familiar from other contexts, notably the philosophy of biology. In epidemiology, its importance consists in the idea that populations may be thought of as bearing health-related properties. This is sometimes counterintuitive, since it is individuals who suffer diseases. But measuring the level of a disease in a population is central to epidemiology, and it requires thinking of populations as entities which can bear properties. There are a number of philosophically interesting problems here. Some are general: for example, we can ask whether population thinking is merely instrumental or whether populations really are property bearers. Some are more specific to epidemiology: for example, how population properties relate to individual ones, what inferences from one to the other are licensed, and even, in some contexts, what inferences from population to individual are ethically or legally warranted. This issue will be explored in a legal context in Chapter 11.

The sixth and most obvious feature of epidemiology is that the stakes are high. This has epistemic as well as moral significance. The cost of failing to make a correct inference may be as high as the cost of making an incorrect inference. This is in contrast to many other sciences, where the cost (at least, the immediate cost) of failing to make a correct inference is merely slowed progress. In epidemiology, wrongly failing to pronounce HIV a cause of AIDS, for example, when it was a cause, could have been just as dangerous as wrongly pronouncing it a cause. This generates interesting questions about scientific attitudes to epistemic risk.

This is a survey, n...

Table of contents

• Cover
• Title
• 1  Why Philosophy of Epidemiology?
• 2  Philosophical and Epidemiological Basics
• 3  The Causal Interpretation Problem
• 4  Causal Inference, Translation, and Stability
• 5  Stable Causal Inference
• 6  Prediction
• 7  Making and Assessing Epidemiological Predictions
• 8  Puzzles of Attributability
• 9  Risk Relativism, Interaction, and the Shadow of Physics
• 10  Multifactorialism and Beyond
• 11  Epidemiology and the Law
• 12  Conclusion: Thinking Is Good for You
• Notes
• References
• Index