Epidemiology Risk and Causation Report

8/12/2019 Epidemiology Risk and Causation Report

1/24

Epidemiology, risk and

causationConceptual and methodological issues in public

health science

Alex Broadbent


2/24

BackgroundIn 2007 the PHG Foundation began funding the author of this report, Dr Alex Broadbent, to conduct

research into the conceptual and methodological issues arising in connection with epidemiology. These

issues include the nature of causation, methods for causal inference, the nature and communication of

risk, the proper use of statistical signicance testing, and the social determinants of health. The project

produced a number of academic articles and included a series of workshops held at Cambridge in 2010,contributions for which form the basis of a special section of the journal Preventive Medicine (2011) Volume

53, issues 4-5. A book on the philosophy of epidemiology is now under contract with Palgrave Macmillan.

AcknowledgementsThe author is especially grateful to Ron Zimmern for suggesting a philosophical project on epidemiology,

for continual interest and encouragement, and for his support in funding decisions. Thanks are also due to:

the steering committee for the workshops Philip Dawid, Stephen John, Tim Lewens, Sridhar Venkatapuram

and Ron Zimmern; the Department of History and Philosophy of Science at Cambridge for administrative

and academic support and for providing venues for the workshops; the Brocher Foundation in Geneva for

supporting some of this research; and a number of individuals at the PHG Foundation including Jane Lane,

Carol Lyon, Hilary Burton and Caroline Wright for various academic discussions and administrative support.

This report is available from

www.phgfoundation.org

Published by PHG Foundation

2 Worts CausewayCambridge

CB1 8RNUK

Tel: +44 (0)1223 740200Fax: +44 (0)1223 740892

November 2011

2011 PHG Foundation

ISBN 978-1-907198-09-0

The PHG Foundation is the working name of the Foundation for Genomics and Population Health, a charitable organisation(registered in England and Wales, charity no. 1118664; company no. 5823194) which works with partners to achieve better health

through the responsible and evidence-based application of biomedical science.


3/24

Epidemiology, risk and causation

1

Motivation ............................................................................................................2

Timeline ..............................................................................................................3

Outputs ...............................................................................................................3

Workshops ............................................................................................................4

1. Determinants of disease ....................................................................................4

2. Risk, probability and harm ..................................................................................4

3. Explanation and intervention ..............................................................................5

4. Causation in epidemiology ..................................................................................5

Analysis ...............................................................................................................7

Summary.......................................................................................................... 7

Topics .............................................................................................................7

Interfaces ........................................................................................................10

Practical signicance ..............................................................................................12

The need for conceptual clarity in the use of health statistics .........................................12

The need for clarity in the use of statistical signicance testing in particular.......................12

The difculty of causal inference and its continued resistance to formal methods..................12

The importance of distinguishing between internal and external validity .............................12

The importance of continued methodological development in epidemiology ........................12

Philosophical signicance .........................................................................................13

The complexity of the sorts of claims about general causation .........................................13

The complexity of causal inference .........................................................................13

The paradox of prevention as a genuine philosophical problem ......................................13

The relation between risks applying to populations and individual risks ..............................13

The apparent fact that epidemiology does not focus either on theory or experiment ..............13

The need for and interest of a specic focus on the philosophy of epidemiology....................14

Directions for further research ...................................................................................15

Philosophy of epidemiology ...................................................................................15

Epidemiology and law .........................................................................................15

Statistical signicance .........................................................................................15

References ..........................................................................................................16

Contents


4/24

2

Epidemiology is the core science of publichealth. The biomedical sciences have attractedthe attention of ethicists, political philosophersand social scientists. However, clinical medicine

attracts more attention than public health,and ethical issues attract more attention thanmethodological and conceptual questions. Thesequestions are the speciality of philosophers ofscience, epistemologists and metaphysicians. Yetto date, epistemologists and metaphysicians havenot paid systematic attention to epidemiology.Nonetheless, the philosophical questions thatarise in the theory and practice of epidemiologyare not limited to ethics. The same is true for thetranslation of research ndings into policy.

This project was conceived to identify and addressmethodological and conceptual questions arising

in epidemiology, with a view to focusing attentionof academics in relevant disciplines: in particular,epidemiology, philosophy and statistics.

The epidemiological literature suggests aneed for work in this area. Epidemiology is ayouthful scientic discipline, notwithstandingits accomplishments. Over several decades, thediscussion sections of epidemiology journals haveincluded insightful papers on topics which can onlybe described as philosophical: causal inference(Hill 1965), the nature of causation (Rothman1976; Susser 1991; Rothman and Greenland 2005),

the reliability of new scientic results (Ioannidis2005), and the use of epidemiological evidence inlitigation (Greenland and Robins 1988; Greenlandand Robins 2000).

This tendency is not abating; indeed, it maybe intensifying. Leading text books includeintroductory material on the philosophy of scienceand whole chapters on philosophical issues specicto epidemiology (notably Rothman, Greenland,and Lash 2008). As well as publishing a largenumber of individual papers on topics of this kind,epidemiological journals have begun setting asidespace for discussion of issues such as these: In 2004

Epidemiologydevoted space to discussion betweensome eminent epidemiologists and an equallyeminent philosopher on the value of risk factorepidemiology (Greenland, Gago-Dominguez, andCastelao 2004; Haack 2004), with specic emphasison nding a theoretical philosophical basis forcausal inference in that context. The September2008 issue of The Journal of Public Healthcontained a section of ve papers whose task wasto revisit, with fresh reasoning, the whole idea ofcausality in the public health sciences (Leung andGray 2008, 217).

In January 2010 the journal Epidemiologyagaindevoted a special section to a conceptual question,this time the neglected merits of absolutemeasures of risk, asking why epidemiologistshave come to rely almost entirely on relative

measures of effect (odds ratios, risk ratios andhazard ratios), even though this practice generatesconsiderable confusion, especially over interaction,effect modication, and the potential public

health benets associated with reported effects(Kaufman 2010, 2).

The editors motivations for devoting spaceto these matters are clearly practical, but theunderlying questions are conceptual.

The fact that epidemiological journals devotespace to issues of this kind demonstrates thatthese discussions are not considered closed by atleast a substantial and important subsection ofthe scientic community. This project set out toidentify the problems driving these discussions andprovide a forum for a more sustained treatment.

It also aimed to provide a forum for thoroughgoingphilosophical attention to these problems.

Although philosophers have dabbled inepidemiological discussions, no philosophers todate have specialised in epidemiology. One of theaims of this project was to discover whether sucha speciality made sense, both from a philosophicalperspective and for the purpose of helping toclarify and make progress with the conceptualissues clouding epidemiological practice.

From the philosophical end, the timing of theproject has been good. Philosophers haverecently become more interested in a number ofapparently disparate issues which come togetherin epidemiology. These include: causal inference;the use of causal modelling techniques; generalcausation and its relation to singular causation;the viability of evidence hierarchies such asthose proposed under the banner evidence-basedmedicine; the nature of health and disease; andhealth as a property of populations (as opposed toindividuals). Epidemiology provides ample materialfor philosophical study of these topics.

Doubt is often expressed when philosophers

purport to contribute to the scientic enterprise.This project is premised on the rejection of suchscepticism. Philosophers have a reputation forbeing better at creating problems than solvingthem. Again, this project is premised on the falsityof this view. Good philosophy is not always useful,but it can be. Specically, it can be useful whenit is addressed at philosophical problems arisingin the course of a useful activity. Epidemiologyis a useful activity, and there is ample evidencein epidemiological journals and text books thatphilosophical problems arise in the course of doingit.

The motivation of this project was to identify someof these problems, and to begin the process ofsolving them.

Motivation


5/24


3

2007-8 - Postdoctoral research project,Philosophical Issues in Epidemiology. PostdoctoralFellow: Dr Alex Broadbent, University ofCambridge, Department of History and Philosophy

of Science. (Funding: PHG Foundation.)

2009-10 - Workshop series Epidemiology, Risk andGenomics, hosted at the University of Cambridge,Department of History and Philosophy of Science.(Funding: PHG Foundation.)

2011 - Report published. (Funding: PHGFoundation.)

2011 - Special section of Preventive Medicinepublished: Epidemiology, Risk and Causation.

2013 - Projected publication of book The

Philosophy of Epidemiology under contract.Published by Palgrave Macmillan in the series NewDirections in the Philosophy of Science.

Publications in academic journals

Broadbent, Alex. 2008a. For Analytic Bioethics.Clinical Ethics3: 185-188.

2008b. A note on epidemiological caution.Letterto the British Medical Journal. www.bmj.com/cgi/eletters/335/7630/1116-b#191569.

2009. Causation and models of disease inepidemiology. Studies in History and Philosophy ofBiological and Biomedical Sciences40: 302-311.

2011a. Causal inference in epidemiology:mechanisms, black boxes, and contrasts. InCausality in the Sciences, ed. Phyllis McKay Illari,Federica Russo, and Jon Williamson, 45-69. Oxford:Oxford University Press.

2011b. Dening neglected disease.BioSocieties6(1): 51-70.

2011c. Conceptual and methodological issues inepidemiology: An overview. Preventive Medicine53, 215-6 (special section: Epidemiology, Risk, andCausation).

2011d. What could possibly go wrong? A heuristicfor predicting population health outcomes ofinterventions. Preventive Medicine53, 256-9. (special section: Epidemiology, Risk, andCausation).

Epidemiological Evidence in Proof of SpecicCausation, Legal Theory, in press.

Special section of PreventiveMedicineThe epidemiology and public health journalPreventive Medicinepublished a special sectionEpidemiology, Risk and Causation, guest edited byAlex Broadbent, devoted to papers arising from the2010 workshops held in Cambridge:

Epidemiology, Risk and Causation. PreventiveMedicine53(2011).

Book

Broadbent, Alex. The Philosophy of Epidemiology.Under contract with Palgrave Macmillan in theirseries New Directions in the Philosophy of Science,ed. Steven French. Projected publication 2013.

Timeline Outputs


6/24

4

In 2010, the PHG Foundation funded four workshopson the conceptual and methodological challengesfacing epidemiology. Each workshop featuredfour invited speakers drawn from the elds of

epidemiology, statistics and philosophy. The topicsaddressed at each of the workshops were:

1. Determinants of disease

The scope of epidemiology has been expanding intwo different ways. Previously, parasite, deciencyand defect were the principle objects of study.The explosion ofgeneticknowledge since the latterpart of the twentieth century has opened up a newplace to look for determinants of disease, studiedbygenetic epidemiologists. At the same time, theinuence of socioeconomic factors on health hasattracted the attention of social epidemiologists.

These labels genetic and social are probablyan oversimplication; genes and environmentinteract, and few epidemiologists strictly connetheir attention to one or the other. Nevertheless,there is on occasion tension and even competitionbetween genetic (or more broadly speakingbiological) approaches, and social ones.

At present, genetic epidemiology tends to berelatively expensive due to genotyping costs; itmakes heavier use of novel biological knowledge;and it often seeks to identify biological

mechanisms. Social epidemiology is comparativelycheaper at present; it is less closely connectedto developments in molecular biology or otherlaboratory sciences; and it is concerned withidentifying psychosocial pathways to disease(Marmot 2006; Brunner and Marmot 2006).

With time, the relative costs of these twoactivities will change and may invert as the cost ofgenotyping comes down.

These approaches can look quite different: butare they in principle distinct? Or is there a singleepidemiological ethos motivating them both?

Can we generalise about which approach servespublic health better? Which promises to advanceunderstanding more?

Speakers Titles

Professor Sir

Michael Marmot

Social determinants of

healthProfessorJonathan Wolff

How should social policy

respond to the social

determinants of health?

Dr NadeemSarwar

Assessment of emerging

strategies to predict and

prevent coronary heart

Professor CarloBerzuini

Statistical methods and

causal inference

2. Risk, probability and harmThere are at least two ways that risks demand ourattention: by theprobabilityof their materialising,and by how bad would be the harmif they did.

But probability and harm are different kinds ofconcepts, and each is a focus of controversywhich working epidemiologists need to negotiate.Is probability a property of individuals (or can itbe)? This view may be appealing in interpretationsof quantum mechanics, where it is doubtfulwhether hidden variables can explain apparentlyprobabilistic phenomena (Price 1996). But even if

irreducible chances are the right interpretationof quantum mechanics, it is not obvious that therationale will carry over to higher level sciencessuch as epidemiology, where the existence ofhidden variables is beyond doubt.

Maybe, then, the appropriate understanding ofprobability for epidemiology is purely statistical,reducing to average frequency in a speciedpopulation. But then it is not clear what basiswe have for moving from the observedaveragefrequencies on which our probability estimates arebased, to the unobservedfrequencies which (inpractical applications) they estimate.

The concept of harm bears more subtly but no lessdirectly on epidemiology. For one thing, the kindofharm that epidemiology studies requires somethought, and is not necessarily dictated by clinicalmedicine.

To take the obvious example, suicide is a publichealth concern, but never a clinical one (even if

preventionof suicide may be). Moreover, the rangeof health conditions studied by epidemiology isincreasing.

The relation between harm and autonomy is alsocomplex, and poorly articulated in the publichealth context. Whereas an individual smokermight be able to refuse clinical treatment for lung

Workshops


7/24


5

cancer, the population of smokers generally cannotavoid public health initiatives on smoking. And it isnot clear how, if at all, their (various) desires canor should be taken into account when counting the

cost of smoking to public health.

Finally, there is a question as to whether riskis itselfa kind of harm, so that exposing someoneto an increased risk of lung cancer by passivesmoking is harming them even if they do not in factdevelop lung cancer. This relates in turn to legalquestions about how causation is proved in toxictort cases, where the correct presentation andinterpretation of epidemiological evidence is ofparamount importance.

Speakers Titles

Professor SanderGreenland

The art and (pseudo?)

science of epidemiologic

risk analysis

Dr MarkParascandola

Causing harm versus

causing risk

Dr Stephen John Is a principle of savingthe most at risk

defensible?

Professor DavidSpiegelhalter

Visualising and analysing

uncertainty about

benets and harms

3. Explanation and intervention

The currency of contemporary epidemiology isthe risk factor. But what is a risk factor? The termarises in part from the focus of epidemiology,unusually (but not uniquely) among sciences,onpublic policy intervention. This suggestsinteresting questions about the relation betweenexplanation and successful intervention. How muchdo we need to understand before we can fruitfullyintervene? Risk factors seem to offer a way todesign interventions in circumstances of incompleteknowledge, either about the circumstances or thenature of the cases of illness in question. But asJacob Henle pointed out, cataloguing the causesof disease is not sufcient for a scientic approachto medicine. It is no more scientic, Henle claims,than if a physicist were to identify as the removalof boards and beams, the cutting of ropes, theopening up of holes, and so forth as causes offalling (Henle 1844; cited in Carter 2003, 24).Identifying these risk factors for falling would notby itself bring the physicist any nearer to a theoryof gravity.

This objection appears to be borne out by thesubsequent history of epidemiology: the mostdramatically successful interventions have tended

to be closely linked to improvements in ourunderstanding. But is that historical claim correct?If so, it would seem that our ability to explainwhyillness occurs is linked to our ability to cure

and preventit. But how? Do risk factors encourageepidemiologists, and their colleagues studyingfunctional genomics, knock-out models, andsimilar, to seek explanations?

Or do they allow potentially dangerousmisunderstandings about the scientic andpractical import of the results they represent? Isit a nave mistake to suggest that public healthinterventions might be best served by seekinggeneral explanations on the model of the physicalsciences? The answers to these questions beardirectly on the future direction of epidemiologyand how it can appropriately employ its conceptual

tools.

Speakers Titles

Dr Michael Joffe Difference-making,discovery, mechanism and

effectiveness

Professor PhilipDawid

How much do we need

to understand in order to

decide what to do?

Professor Alfredo

Morabia

Until the lab takes it away

from epidemiology

ProfessorAlexander Bird

Intervention in medicine

and inference to the best

explanation

4. Causation in epidemiology

There has been a rash of epidemiological literature

on the traditionally philosophical topic of the

nature of causation (a small selection: Rothman

1976; Robins and Greenland 1989; Susser 1991;

M. Parascandola and Weed 2001; Rothman and

Greenland 2005). Why is this? Epidemiologists often

seek to infercausation, but then so do many other

scientists. There are at least three (and probably

more) more likely reasons that causation is an issue

in epidemiology.

First, it is a science relying heavily on observationalstudies; this means that the difference betweencause and coincidence cannot be easily reducedto a difference between repeatability andcoincidence.


8/24

6

Second, epidemiology deals with general causalclaims, such as Smoking causes lung cancer.These claims are hard to interpret. Do they expressa relation between two properties, smoking

and lung cancer? Or are they (exception-ridden)generalisations over individual cases of smokingcausing lung cancer? These questions relate tothe interpretation of probability, discussed in thesecond workshop, since causal generalisationsin epidemiology are usually probabilistic. Thereis a related question here about whetherepidemiology deals with laws of nature. Are thereepidemiological laws, or are the relevant lawsall biological or even physical? And what is therelation between each of the aforementionedconcepts of general causation and ceteris paribuslaws?

Finally, the identication of causes in manysciences is tied up with the process of explanation.In epidemiology, however, causes may not beexplanatory. An epidemiological study mightestablish that smoking causes lung cancer, tothe satisfaction of governments and the generalpublic, without explaining why people get lungcancer. Intuitively, this is because the link betweensmoking and lung cancer is not illuminated byestablishing that smoking causes lung cancer.

This last point links back to the topic of thethird workshop, since it is unclear how the

identication of causes for the purpose of devisinginterventions relates to causal explanation (whichhas traditionally received a great deal moreattention from philosophers of science). Moreover,the kindof cause that we focus our attentionon may bear both on the chances of a goodexplanation and of devising a good intervention.For example, identifying readily manipulablesocioeconomic causes of illness may offer promisinginterventions but little understanding; with geneticcauses, the reverse may be true. These are, ofcourse, disputable claims, relating to the topic ofthe rst workshop.

This nal workshop therefore tied together some ofthemes of the previous three.

Speakers Titles

Dr Olaf Dekkers Causation inepidemiology

Professor NancyCartwright

The long road from it-

works-somewhere to it-

will-work-for-us

Professor DanHausman

How can practical causal

generalisations guide

practice

Professor JohnWorrall

Causality in medicine:

getting back to the Hill

top


9/24


7

Summary

The most important point to arise out of thisproject is the most obvious one: that there is

considerable scope for academic research intothe conceptual and methodological foundations ofepidemiology, and that this research is of practicalimportance to the development of the disciplineand to the interpretation and deployment of itsresults in clinical and public health contexts.

In particular, there is scope for further research inthe following areas:

Causation and causal inference in epidemiology

Risk: interpretations, communication, andrelation between population and individual

The treatment of uncertainty in both academicand policy contexts

None of these topics is new, either to philosophersor to epidemiologists, but the need to focusspecically on epidemiology in studying each ofthese topics has not been clearly recognised, and iselaborated further below.

In addition to topics, it has been possible toidentify a number of interfaces between differentdomains across which epidemiological researchtravels, and which stand in need of attention:

The epistemic/normative interface

The conceptual/applied interface

The research/policy interface

Again, the contribution of this project lies not inany novelty claimed for these distinctions, but forthe recognition of a need to focus specically onepidemiology at each of these interfaces. This need

is elaborated further below.

Topics

Causation and causal inferenceThe nature of causation is an old and difcultphilosophical problem. One way to formulate it isby contrasting causal and non-causal sequencesof events. You see the footballer swing his foottowards the ball. When it makes contact, the ballies into the back of the goal. At the same time,you also see the referee scratch his nose. When hishand touches his nose, the ball ies into the backof the goal. The rst sequence of events is causal:the footballers kick causedthe ball to y into theback of the net. The second it not: the refereesscratch did not cause the ball to y into the backof the net.

There are, of course, many visible differencesbetween these two sequences of events. Theproblem is that none of these visible differencesis causation: so philosophers try to say what the

difference is. For example, we might note that ifthe footballer had not kicked then the goal wouldnot have been scored; but that if the referee hadnot scratched his nose, the goal would still havebeen scored. Or we might note that kicks of thissort are always or often followed by goals, whereasnose-scratches are not. But whatever we say aboutthe difference, we must do more than simply pointto our experience of those events. This means thatboth the nature of causation, and our knowledge ofit, are topics of philosophical interest.

Both the metaphysics and epistemology ofcausation are also topics of epidemiological

interest. There are a number of differences,however, between philosophical andepidemiological approaches. Most importantly,philosophers have focussed primarily on singularcausation, that is, causation between single events(such as the footballers kick and the ball enteringthe goal). Epidemiologists, however, focus primarilyon general causation, that is, the relation (thoughtechnically it may not be a relation) between twovariables, factors, or classes of events that holdswhen it is true to say that one causes the other(such as holds between the inhalation of asbestosand the disease mesothelioma, for example).

It is well known to both philosophers andepidemiologists that causation is a topic of interestin each of their elds. What this project has shown,however, is just how different the topic looks ineach context. It is very tempting for philosophersin particular to doubt this (as I did at the beginningof this project). We are apt to suppose that theproblems epidemiologists address are eitherrephrasings of familiar philosophical concerns, orelse not philosophical concerns at all. But oftenthey are neither, implying that they are new, or atleast unstudied, philosophical problems.

The point can be illustrated with two examples,one concerning the metaphysics (or nature) ofcausation and one concerning its epistemology(causal inference). As I have already indicated,philosophers interested in the metaphysicsof causation have focussed largely, thoughnot exclusively, on singular causation, whileepidemiologists focus on general causation.Philosophers have a lot to learn about generalcausation in this context, and in particular aboutits complexity.

Analysis


10/24

8

The most widely-used philosophical exampleof a general causal claim is Smoking causescancer, but this is at best a distant descendentof the output of actual epidemiological work on

smoking. In particular, epidemiologists seek tomeasure the strength of the effects they identify,and to compare the strength of these effects.This is simply not something on the radar of mostphilosophers working on causation (an exception isSober 1988). This oversimplication is illustrated bythe fact that many, though not all, philosophers usethe term general causation interchangeably withcausal generalisation. Epidemiology shows thatthis is an error, since it deals neither in singularcausal claims nor in generalisations about singularcausal relations, and especially not exceptionlessor universal generalisations (as Dan Hausmanemphasised in his presentation at Workshop 4).

This suggests that an instance of general causationand a causal generalisation are not the same thing,any more than a general interest is the same asan interesting generalisation. Other philosophershave suggested that general causation is a relationbetween types of event (Eells 1991). But thisview does not help make sense of the quantitativeaspect of causal claims in epidemiology. Moreoverit is hard to reconcile with the fact that many suchclaims are relative to a particular population, in away that claims about abstracted types are not.

The second example of an unstudied philosophicalproblem in this area concerns causal inference.Philosophers are accustomed to the idea thatcausation is not directly perceptible and thus thatwe often, perhaps always, infer that it is present ina given case.

Again, the study of epidemiology suggests thatthere is more complexity to causal inference,especially at the general level, than might besuspected. In particular the distinction betweeninternal and external validity, familiar toepidemiologists, has philosophical ramications.

A study is internally valid when its conclusions holdfor the studied group. It is externally validwhen its conclusions hold for some target groupthat is partly or wholly outside the studied group.When a study is internally valid and has a causalconclusion, it may nevertheless fail to be externallyvalid. Thus a study may warrant a causal inferencefor the group studied, but not for some target orwider group.

Philosophers have not generally paid muchattention to these two components of causalinference. Yet they are clearly important in a

public health context. Just because somethingworks somewhere does not mean it will work forus, to use Nancy Cartwrights phrase.

The insensitivity of extant philosophical theory tothese two steps is related to the oversimpliedtreatment of causation itself: there is simplyno ready philosophical toolkit for framing a

distinction between the assertion that X causesY in a studied group, and the assertion that Xcauses Y in some target group. The underlyingassumption is that either X causes Y or it does not.Tautological though this may sound, a study ofepidemiology suggests that it is false. X may causeY in the studied group but the study may fail to beexternally valid, and it may turn out that X doesnot cause Y in some context that is either moregeneral or more pertinent to the aims of the study.

These two examples are quite specic, and, alongwith similar examples, they receive their fullexpression in academic activities and outputs of

the project, especially discussion at the workshops,articles in the special issue of Preventive Medicine,and the authors forthcoming book.

In the course of this project, a number of moregeneral points have emerged, which also deserveemphasis. Simply conrming that causation andcausal inference are still live issues in philosophyand epidemiology is important, especially inepidemiology. It is important that epidemiologists,and epidemiology students, appreciate thatthere are methodological problems in theirdiscipline which have not been solved, and where

conceptual work is still necessary to reach eventhe most pragmatic goals. The importance ofthis appreciation can be seen in the commonoccurrence of various misunderstandings, whichwere the subject of much discussion at theworkshops.

For example, Hills famous viewpoints fordeciding whether an association is causal arefrequently elevated to the status of criteria,despite his explicit (and correct) insistencethat they must not be treated as such. He wasclear that they are only aids for answering thefundamental question - is there any other wayof explaining the set of facts before us, is thereany other answer equally, or more, likely thancause and effect? (Hill 1965, 299). This was apoint emphasised by both Alexander Bird and JohnWorrall at the workshops.

Relatedly, the attempt to substitute statisticalsignicance testing for causal inference, criticisedin Hills paper, was a focus of considerablediscussion at the workshops (especially by SanderGreenland and Mark Parascandola). It is importantto continue to stress that there is no formal oralgorithmic method of causal inference, and that

causation itself is not well-conceptualised orunderstood. Any claims to the contrary are false,and, in the context of epidemiology, dangerous.


11/24


9

Risk

Risk is a topic of growing interest in politicalphilosophy and the philosophy of science. The

word risk also occurs frequently in epidemiology.Possibly the simplest point to make here is thatthe epidemiological use of the word risk oftendoes not correspond to risk as it is commonlyunderstood, or as it is understood by philosophersstudying risk.

In epidemiology, a risk is a kind of measure:it is the number of new cases of a disease orother health event occurring within a given timeperiod. Philosophers, on the other hand, typicallyconceptualise risk as the product or other functionof the probability of a harm occurring and howharmful it is if it does occur. The epidemiologicmeasure takes no account of the harmfulness ofthe outcome and is not related to any probabilitieswithout further assumptions; it is simply a fraction.This is a terminological matter, but it has clearpotential to cause confusion, especially in thepublic reporting of epidemiological results andin feeding epidemiological results into policydecisions.

There are also conceptual difculties in expressingrisk, and especially in communicating it. DavidSpiegelhalter discussed several of these in hispresentation at Workshop 2.

It is not at all clear how to measure risk in ameaningful way. One option is to pick a riskyexposure, such as the eating of bacon sandwiches,and quantify other risks in terms of that risk.Thus the riskiness of paragliding every weekendmight be expressed in terms of the number ofbacon sandwiches one would have to eat toreduce ones life expectancy to that of a personwho paraglides every weekend. However, as thisexample illustrates, it is not always easy to ndsuitable measures. Moreover the method relies onan ability to conceptualise the riskiness of eating abacon sandwich in the rst place, an ability which,

arguably, many of us do not possess.

Another option is to pick an outcome, such asdeath, which is objective and whose harmfulnesscan be assumed to be held reasonably constantacross evaluations. We can then seek to expressthe risk of an activity in terms of the number ofdeaths per unit of activity. However it is clearthat policy makers, and perhaps many members ofthe public, do not hold the value of death equalacross such evaluations. Professor Spiegelhalternoted government initiatives to encourage youngpeople to countenance a reasonable and controlled

degree of risk in the context of sporting andoutdoor activities. He noted, however, that onsome measures, horse-riding and taking ecstasyare about equally risky. The government does not

encourage taking ecstasy, and in fact forbids it.This suggests either a governmental view that itis worse to die from taking ecstasy than falling offa horse, or ignorance of the relative riskiness of

these activities.

Whichever it is, there is a need for conceptualclarication in the deployment of statistics inpublic decisions concerning how dangerous agiven exposure is and how its riskiness relates toattempts to regulate it. Epidemiology is implicatedbecause it supplies many of the relevant statistics,and it has a scientic and civic duty to render themas clear and as t for purpose (whatever that maybe) as it can.

A particularly important area in which risk posesconceptual difculties concerns the bearing of

population risks on individual risks. At a populationlevel, risk is (in epidemiology) a clearly denedterm. It tells us how many new cases of the diseaseor other harm occur within a specied time period,as a proportion of the population size at the startof the time period.

But what does this mean for an individual patient,or a consumer, or a doctor seeking to treat anindividual patient? An individual either developsthe disease or does not. So the risk for thepopulation consisting of that individual alone iseither 100% or 0%. What the individual will want to

know, arguably, is herprobabilityof developing thedisease.

The interpretation of probability, especially of theoccurrence of an individual event, is a philosophicaltopic, and one which has direct bearing on the useof data derived from population studies in clinicalsettings, consumer decisions, court rulings, andother occasions where a decision must be takenabout an individual in light of population-leveldata. To ignore such data is surely wrong; but thecorrect way to accommodate it is far from obvious.

UncertaintyUncertainty is a feature of most, if not all, of ourknowledge. There is little, if anything, that weknow without room for doubt, nor with certainty soabsolute that greater certainty is inconceivable.


12/24

10

Epidemiology is a particularly uncertain activity,because epidemiologists are often studyingassociations about which little is known. Theunderlying process giving rise to an association

may be entirely or partially mysterious, and mayremain so even after epidemiological evidencesuggests that the association is robust.For example,we know that smoking causes cancer even thoughwork to identify and understand the action of thecarcinogens in tobacco smoke is ongoing.As AlfredoMorabia argued in his presentation to Workshop 4,epidemiology actively investigates a phenomenononly until it is sufciently well-understood forthe laboratory sciences to take over. After that,epidemiologists may be called upon to makepredictions about the spread of a disease, butwork on identifying and understanding the causes(analytic epidemiology) ceases. This means that,by its nature, (analytic) epidemiology is at theforefront of our knowledge, where it inevitablyconfronts uncertainty.

Considerable discussion at the workshops focusedon the proper treatment of uncertainty. Inparticular, the use of statistical signicancewas discussed. Sander Greenland and MarkParascandola both discussed limitations on the useof p-values. Two such limitations are particularlyimportant.

First, p-values test at most one kind of error: the

probability that an observed association arose bychance, where the latter phrase indicates that thevariables in question occur randomly with respectto each other, and that the association observedon this occasion is the result of that randomoccurrence. (Errors of this kind are often referredto asfalse positives or type 1 errors). This meansthat p-values are, at best, a guard against wronglyinferring a false hypothesis. They are no guard atall against failing to infer a true hypothesis (oftenreferred to asfalse negativesor type 2 errors).If p-values are not low enough then, on standardmethodologies, no inference will be made - but thehypothesis under test may nonetheless be true. We

lack a widely accepted method for guarding againstthis sort of error, even though it is potentiallyjust as harmful. Moreover, an excessive p-valuedoes not even mean that a causal inference isunwarranted. That will depend on methodologicalfactors that cannot be meaningfully reduced tothat statistical measure.

The other important limitation of the use ofp-values is that the null hypothesis is rarelysubjected to the same scrutiny as the hypothesisunder test. Mark Parascandola pointed out that ifthe null hypothesis is applied to studies or meta-

analyses of large numbers of variables, it amountsto the hypothesis that everything is due to chance,a presupposition which, if seriously entertained,would undercut the scientic enterprise altogether.

Sander Greenland pointed out that the use ofthe p-values in many contexts amounts to anunwarranted bias towards the null, by encouragingthe view that the null hypothesis is to be held

until disproved. In fact, we cannot regard thenull hypothesis as proven by a failure to nd astatistically signicant association. We cannotconclude that there is no association. All we canimmediately conclude, in the absence of furtherevidence or repeated studies, is that we still dontknowwhether there is an association.

The misuse of statistical tests in the contextof quantifying uncertainty by estimating theprobability of error is related to the incorrectsubstitution of signicance testing for causalinference, mentioned previously. In both cases,the emerging picture is the same: the existence

of mathematical tools sometimes misleadsepidemiologists into thinking that they havemore (or less) certainty than they really do; andsometimes it leads them into substantive errors.The lesson is likewise the same: that appreciation ofthe conceptual foundations of the discipline, and ofthe persistently informal, unalgorithmic nature ofkey components of its methodology, is essential toits success.

Interfaces

The epistemic / normativeinterfaceEpidemiology, or its outputs, move across anumber of interfaces, in ways that are conceptuallyinteresting and sometimes challenging. Mostobvious among these is the interface betweenepistemic activities, concerning the gathering ofknowledge, and normative activities, concerningevaluation and recommendation. Epidemiologyemploys the methods of natural sciences, but it isso closely tied with human concerns that in someforms it might also be seen as a social science.

In his presentation at Workshop 1, Michael Marmotargued that medical professionals are duty-boundto identifying and eliminating causes of ill-health,even when these include poverty or inequalityof social status things normally considered thepreserve of politics or perhaps economics, not ofmedicine, and certainly not of the natural sciences.Jonathan Wolff, in his presentation, sought to workout what, if anything, should be done if ProfessorMarmot is correct that socioeconomic inequalitycauses ill health.

One thing that Professor Marmots work showsis how epidemiology can exert an expansivepressure on medicine. Epidemiological methodscan be applied without much sensitivity to subjectmatter. It is possible to assess the effect of socialinequality on health just as long as it is possible to


13/24


11

measure poverty and health. This paves the way forarguments such as Marmots, explicitly advocatingan expansion of the scope of medical attention.

Another point at which epidemiology potentiallycrosses the border between epistemic andnormative is in the handling of uncertainty,discussed previously. The choice of a value forstatistical signicance is arbitrary in relationto the statistical signicance tests themselves.It depends on a judgement as to how seriousthe consequences of error will be, and also, asdiscussed, on weighing the relative perils of fallinginto different kinds of error.

The conceptual / applied interfaceEpidemiology deals in both practical and

theoretical matters. But the distinction betweenthe practical and theoretical is not very precise.We can split it into two more precise distinctions.The rst is between conceptual and applied;the second is between research and policy.Epidemiology works across both these interfaces.

The interface between conceptual and appliedwork in epidemiology is evident in goodepidemiological text-books (e.g. Szklo and Nieto2007; Rothman, Greenland, and Lash 2008). Thesefocus on teaching epidemiological methods with aview to their application to the sorts of problems

that epidemiologists need to solve. Yet they oftenalso devote space, sometimes considerable space,to conceptual issues more specically, to the wayin which conceptual issues can arise during theattempted application of epidemiological methods.Thus the conceptual sections of epidemiology textbooks the chapters on causation, for example donot simply set down a conceptual framework forthe student to absorb. They describe the lack ofsuch a framework (if they are honest) and indicatehow this lack can make an epidemiologists workdifcult.

What this means is that epidemiologists cannotbe simply technicians. They must also have someappreciation of the conceptual foundations of thediscipline, and in particular of the issues arisingin the topic areas described above. If they donot then they are likely to commit errors, such asexhibiting an unwarranted bias towards the nullhypothesis, failing to properly balance the riskof different kinds of errors, making bad causalinferences, or failing to make good ones.

The research / policy interfaceThe other dimension of the more general interface

between practical and theoretical work is thedistinction between scientic research and healthpolicy making. As noted above, epidemiology haselements in common with both natural and social

sciences. Its methods may be scientic, but itsobjectives are often thoroughly human.

One particularly interesting example of the

interface between research and policy is inGeoffrey Roses paradox of prevention (Rose1992). The paradox arises from the fact that alarge number of people at low risk can generatea larger number of adverse outcomes than asmall number at high risk. This means that, incases where individuals are distributed alonga risk prole (for example, across a range ofcholesterol levels as a risk factor for heartdisease), it may be more effective to interveneon the majority at a moderate to low risk thanto intervene on those at high risk. The paradoxis that most of these individuals will receive nobenet from the intervention, but will presumably

suffer the attendant harms, if only the harm ofinconvenience. John Worrall emphasises the factthat many (perhaps 95%) of those who take statinsreceive no benet, but do suffer the attendantrisks of taking the drug.

However, effective prevention strategies oftenrequire individuals to make exactly this sort ofsacrice. Vaccination is the best-known example.At Workshop 4, Stephen John used this feature ofpreventive strategies to bring out philosophicalquestions about the direct and indirect benetsthat an individual can derive from a preventive

strategy.

Dr John also sought a clear formulation ofthe paradox of prevention. Arriving at a clearformulation is of practical as well as academicinterest, since individuals choosing whetherto participate in preventive programs such asvaccination will inevitably ask themselves what arethe likely risks and benets for they themselves.


14/24

12

The following points of practical signicance aresalient in the foregoing analysis.

The need for conceptual clarity in the use of

health statistics

A number of workshop contributions highlightedways in which both statistical analysis and relianceon statistics in policy may rest on implicit andunwarranted assumptions. Greater conceptualclarity on the part of those performing andusing statistical analyses, and a critical approachto statistical warrant for policy decisions, are

necessary to prevent this.

The need for clarity in the use of statisticalsignicance testing in particular

A particularly prominent case where conceptualclarity is often lacking is in the use of statisticalsignicance testing. The underlying message isthat statistical signicance must not be confusedwith signicance simpliciter. An association mightfail to be statistically signicant due to the sizeof the study, yet the nature of the study mightnevertheless provide convincing evidence foran inference. Moreover statistical signicancetests do not estimate the probability of wronglyfailing to make an inference, and do not estimatethe practical costs of wrongly failing to make aninference. A result might be signicant becauseit suggests a serious hazard even though it is

not statistically signicant. Morever, statisticalsignicance is no guarantee of truthfulness.For a p-value of 0.05 we would expect 1 in 20independent studies of null associations to wronglypronounce that there is an association. Statisticalsignicance is thus neither necessary nor sufcientfor signicance more generally and must always besupplemented by methodological reasoning and,where appropriate, consideration of the practicalimplications of both Type 1 and Type 2 errors.

The difculty of causal inference and itscontinued resistance to formal methods

Causal inference is difcult. Statistical toolscan help. However, there is no algorithm ormathematical tool which can answer the mostimportant question: is there a causal relationshipunderlying the observed association, and if sowhat are its properties? Moreover the applicationof formal tools often requires some prior causalinterpretation. This does not prevent themfrom being useful, but it does show that causalinference remains prior to any statistical or othermathematical analysis. Contrary to what some haveasserted, causation has not yet been mathematized(cf. Pearl 2000, xiii).

The importance of distinguishing betweeninternal and external validity, and the difcultiesattendant on applying or generalising the resultsof a study to a target or wider group

Even if the foregoing cautions are all taken intoaccount, further thought is required before usingthe results of a study, or even several studies, tomake claims about other populations. There aremany reasons why a drug or a policy might work onone occasion but not another; and many reasonswhy an exposure might cause ill health on oneoccasion but not another. As Nancy Cartwrightemphasizes in her contribution to the special issueof Preventive Medicine, ensuring that evidencemeets strict criteria is only one part of establishinga sound basis for policy.

Another, equally important, part is nding somereason to think that the study provides a guide forwhat will happen in the circumstances with whichthe policy is concerned. This point has sometimesbeen neglected due to an overriding emphasis onquality of evidence, as opposed to relevance or

applicability.

The importance of continued methodologicaldevelopment in epidemiology, and thatepidemiologists and policy makers have someunderstanding of the live methodological andconceptual debates within the discipline

Epidemiology is a young science, with a numberof central conceptual questions unresolved or onlypartially resolved. It is also an unusual science,in that the body of epidemiological knowledgeis not primarily factual, but methodological.Epidemiology is in large part a collection ofmethods for nding things out on the basis ofscant evidence, and this by its nature is difcult. Ifworking epidemiologists and policy-makers possesssome understanding of the live conceptual debatesin epidemiology then they will be better able toavoid erroneous certainty.

Practical signicance


15/24


13

In the foregoing analysis, the following points of

philosophical signicance are salient.

The complexity of the sorts of claims about

general causation that epidemiology makes, as ascience primarily interested in general causation

Philosophers have tended to work with very simpleclaims, such as smoking causes lung cancer, whenthinking about general causation. Epidemiologiststend to make much more complex claims. Thisproject has suggested that there are importantphilosophical issues lurking in the complexities; butit is impossible to say what they are in advance of athorough philosophical treatment.

The complexity of causal inference, especiallyto general causation, as exemplied by the

distinction between internal and externalvalidity

Philosophers have given a considerable thoughtto the way in which we make causal inferences.They have focused, naturally enough, on simpleexamples and idealised cases. They have alsofocused on experimental scenarios, in which anintervention is made (Semmelweiss work is aclassic in philosophy of science, for instance).Philosophers have not yet caught up with theadvent of sophisticated observational methods. Forexample there has been no thorough and extendedphilosophical analysis of the cohort study, thecase-control study, or indeed many other centralepidemiological methods. Nor have philosophersgiven much explicit thought to the question of howthe results of a study may be generalised: theyhave tended to focus on the problem of inferringcausation from an experimental result, and not onthe question of how if causation can be inferred that conclusion might be applied elsewhere. Theseare clear gaps in the philosophical literature andit is to be hoped that philosophers of science willattend to them soon.

The paradox of prevention as a genuine

philosophical problem

The paradox of prevention (a term coined byGeoffrey Rose (1992)) arises when reducing therisk of persons in medium- to low-risk groupshas a larger impact on the overall risk in thepopulation than reducing the risk of persons inhigh-risk groups only. The paradox arises from thefact that interventions on persons on these groupswill typically offer little or no benet to thoseindividuals (or even incur costs), despite the effecton the health of the population. Vaccination andthe reduction of cholesterol levels are two well-known examples. In his contribution to the special

issue of Preventive Medicine, Stephen John arguesthat this is not a mere pragmatic awkwardnessbut a real philosophical paradox (at least on somecontractualist views of the source of political

obligation). As such it deserves to be studied byphilosophers.

The relation between risks applying to

populations and individual risks

As epidemiologists use the term, risk has noapplication to the individual. Yet ignoring therisks derived from studying populations in makingdecisions about individuals would be foolish. Inthe clinical setting, the evidence-based medicinemovement has advocated attaching very greatweight to evidence derived from studies of largepopulations. One criticism they have faced isthat the individual case may differ in importantways which render the evidence irrelevant.More generally there is an open philosophicalquestion about what the exact rational bearingof population-level data on an individual is.The problem has also been explored in legalcontexts, when plaintiffs have sought to relyon epidemiological evidence. This problem is ageneral theme of many efforts to use the output ofepidemiological work and it raises questions whicha philosopher can help to answer.

The apparent fact that epidemiology does notfocus either on theory or experiment, meaningthat most philosophical characterisations ofscience are of limited or no applicability toepidemiology

It is evident from even a cursory acquaintancewith epidemiology that experiment does notplay the central role that it does in some othersciences. Less immediately evident is the factthat epidemiologists do not develop theory inthe way that other scientists do; or rather, thetheory they develop is of a different sort,and not what a philosopher would call theory.Epidemiologists develop methods for nding outabout things. They do not, however, develop anoverarching view of the way that things are inthe domain they study. Epidemiologists employtheoretical knowledge of this kind from other

disciplines biology, medicine, and others. Butepidemiological ndings, for example that smokingcauses lung cancer, are not assimilated to a bodyof epidemiological facts, in the way that medicalor biological facts might be. This is becauseepidemiology focuses on nding things out, anddeveloping better methods for doing so. This makesit very suitable for philosophical study. Moreover,the lack of theory (as philosophers use the term)and experiment makes it hard to see how some ofthe central philosophical views of science applyto epidemiology. This suggests that they needreconsidering.

Philosophical signcance


16/24

14

The need for and interest of a specic focuson the philosophy of epidemiology within thephilosophy of science

If so many promising topics of philosophical interestcan arise in the course of this project, it seemsvery likely that a systematic philosophical study ofepidemiology will prove worthwhile. No such studyhas yet been undertaken, and there is not as yet awidely-recognised sub-area within the philosophyof science devoted to epidemiology. This projectsuggests that there should be philosophy of

epidemiology.


17/24


15

The project has thrown up a large number ofresearch questions, and they differ dependingon the interests of the parties involved and theircontact with the project. Rather than seek to list

them, this report identies three important generalareas for further research.

Philosophy of epidemiology

This project suggests that there is scope for adistinct focus on epidemiology within philosophyof science. There is no established philosophyof epidemiology, as there is of physics, biology,economics, medicine, and so forth. This projecthas demonstrated the richness and diversity ofintellectual problems that epidemiology bringstogether. It has also demonstrated the practicalneed for academic study of these problems. Thesimplest and most direct way to satisfy this needis to promote the philosophy of epidemiology as adistinct focus within the philosophy of science.

The author is writing a book on this topic and has

obtained funding for further academic meetings.

Epidemiology and law

A second direction for more research concernsthe role of epidemiological evidence in litigation.This is a topic that Mark Parascandola and SanderGreenland discussed in Workshop 3 as well as in

a variety of prior publications (Greenland andRobins 1988; Greenland and Robins 2000; Robinsand Greenland 1989; Greenland 2004; MarkParascandola 1998). It is also something that legalacademics have discussed (see especially Wright1988; Wright 2008).

There are a number of difculties in usingepidemiological evidence in litigation. Two areparticularly salient. The rst is the application ofstatistical tests of signicance to either bolsteror refute causal inferences which are not, in fact,bolstered or refuted by the data in question (atopic treated thoroughly by Sander Greenland in

his contribution to the special issue of PreventiveMedicine). The second is the question of whetherand if so how epidemiological ndings can beapplied to individual litigants, especially in proofof causation. Both of these are pressing concerns,because epidemiological evidence is sometimesthe only evidence available for proof of thecausal element of liability. As Professor Morabiaemphasized at Workshop 3, epidemiology is activelyengaged in researching a disease only when thatdisease is not tractable to laboratory scientists.Legal controversy is also much more likely to arisewhen the grounds of a claim are novel. This is a

systemic reason for the fact that epidemiologysometimes becomes embroiled in legal controversy,and provides a reason for further academic workon the proper use of epidemiological evidence in

litigation.

Statistical signicance

A third important direction for further researchconcerns statistical signicance testing. This wasa recurrent theme at the workshops. There isconsiderable discontent among epidemiologistsand statisticians at the way in which statisticalsignicance testing is taught and employed in manycontexts, including by eminent statisticians. Thissuggests that there is scope for further research inat least two directions. First, the use of statisticalmethods for testing for false negatives as wellas false positives might be more widely taught.Second, procedural rules might be sought fordeciding whether a statistical signicance test

is being correctly applied. Whether these areachievable goals is not something that the authoris in a position to judge. But it is clear that muchof the confusion surrounding the use of statisticscomes from the difculty of understanding it.Since it is not feasible for everyone to become astatistician, the development of simple modelsand heuristics for understanding statistics such asthose being developed by Professor Spiegelhalter

must be a priority.

Directions for further research


18/24

16

Broadbent, Alex. 2008a. For Analytic Bioethics.

Clinical Ethics3: 185-188.

Broadbent, Alex. 2008b. A note on

epidemiological caution. Letter to the BritishMedical Journal. http://www.bmj.com/cgi/

eletters/335/7630/1116-b#191569.

Broadbent, Alex. 2009. Causation and models of

disease in epidemiology. Studies in History and

Philosophy of Biological and Biomedical Sciences

40: 302-311.

Broadbent, Alex. 2011a. Causal inference in

epidemiology: mechanisms, black boxes, and

contrasts. In Causality in the Sciences, ed. Phyllis

McKay Illari, Federica Russo, and Jon Williamson,

45-69. Oxford: Oxford University Press.

Broadbent, Alex. 2011b. Dening neglected disease.

BioSocieties6 (1): 51-70.

Brunner, Eric, and Michael Marmot. 2006.

Social organization, stress, and health. In Social

Determinants of Health, ed. Michael Marmot and

Richard Wilkinson, 6-30. Second. Oxford: Oxford

University Press.

Carter, K. Codell. 2003. The Rise of Causal

Concepts of Disease. Aldershot: Ashgate.

Eells, Ellery. 1991. Probabilistic Causality.

Cambridge: Cambridge University Press.

Greenland, Sander. 2004. The Need for Critical

Appraisal of Expert Witnesses in Epidemiology and

Statistics. Wake Forest Law Review39 (2): 291-310.

Greenland, Sander, and James Robins. 1988.

Conceptual Problems in the Denition and

Interpretation of Attributable Fractions.American

Journal of Epidemiology128 (6): 1185-1197.

Greenland, Sander, and James Robins. 2000.

Epidemiology, Justice, and the Probability ofCausation.Jurimetrics40: 321.

Greenland, Sander, Manuela Gago-Dominguez, and

Jose Esteban Castelao. 2004. The Value of Risk-

Factor (Black Box) Epidemiology.Epidemiology15

(5): 529-535.

Haack, Susan. 2004. An Epistemologist Among the

Epidemiologists.Epidemiology15 (5): 521-522.

Henle, Jacob. 1844. Medicinische Wissenschaft und

Empirie. Zeitschrift fur rationelle Medizin1: 1-35.

Hill, Austin Bradford. 1965. The environment and

disease: association or causation? Proceedings of

the Royal Society of Medicine58: 259-300.

Ioannidis, John P.A. 2005. Why Most Published

Research Findings Are Wrong. PLoS Medicine2 (8):

e124.

Kaufman, Jay S. 2010. Towards a MoreDisproportionate Epidemiology. Epidemiology21

(1): 1-2.

Leung, Gabriel M., and Selena Gray. 2008.

Translational public health: rehearsing the evidence

until the task is done. The Journal of Public Health

30 (3): 217-218.

Marmot, Michael. 2006. Health in an unequal

world: social circumstances, biology, and disease.

Clinical Medicine6 (6): 559-572.

Parascandola, M., and D. L Weed. 2001. Causationin epidemiology.Journal of Epidemiology and

Community Health55: 905912.

Parascandola, Mark. 1998. What is Wrong with the

Probability of Causation?Jurimetrics39: 29-44.

Pearl, Judea. 2000. Causality: Models, Reasoning

and Inference. Cambridge: Cambridge University

Press.

Price, Huw. 1996. Times Arrow and Archimedes

Point. Oxford: Oxford University Press.

Robins, James, and Sander Greenland. 1989. TheProbability of Causation under a Stochastic Model

for Individual Risk. Biometrics45: 1125-1138.

Rose, Geoffrey. 1992. The Strategy of Preventive

Medicine. Oxford: Oxford University Press.

Rothman, Kenneth J. 1976. Causes.American

Journal of Epidemiology104 (6): 587592.

Rothman, Kenneth J, and Sander Greenland. 2005.

Causation and Causal Inference in Epidemiology.

American Journal of Public Health (Supplement 1)

95 (S1): S144S150.

Rothman, Kenneth J, Sander Greenland,

andTimothy L Lash. 2008. Modern Epidemiology.

3rd ed. Philadelphia: Lippincott Williams & Wilkins.

Sober, Elliot. 1988. Apportioning Causal

Responsibility. The Journal of Philosophy85 (6):

303-318.

Susser, Mervyn. 1991. What is a cause and how do

we know one?American Journal of Epidemiology

133 (7): 635--647.

References


19/24


17

Szklo, Moyses, and F. Javier Nieto. 2007.

Epidemiology: Beyond the Basics. 2nd ed. Boston,

Toronto, London, Singapore: Jones and Bartlett

Publishers.

Wright, Richard. 1988. Causation, Responsibility,

Risk, Probability, Naked Statistics, and Proof:

Pruning the Bramble Bush by Clarifying the

Concepts. Iowa Law Review73: 1001-1077.

Wright, Richard. 2008. Liability for Possible Wrongs:

Causation, Statistical Probability, and the Burden of

Proof. Loyola of Los Angeles Law Review41: 1295-

1344.


20/24

18


21/24


19


22/24

20


23/24

About the PHG FoundationThe PHG Foundation is an independent, non-prot organisation based inCambridge, UK, with the mission making science work for health. We identifythe best opportunities for 21st century genomic and biomedical scienceto improve health and tackle disease in ways that are rapid and effective,

equitable and responsible. This entails work to promote the prompt translationof scientic innovation into medical and public health policy and practice.

We provide knowledge, evidence and ideas to stimulate and direct well-informed debate on the potential and pitfalls of key genomic and biomedicaldevelopments, and to inform and educate stakeholders policy makers, healthprofessionals, patients and the public. We also provide expert research, analysis,health services planning and consultancy services for governments, healthsystems, and other non-prot organisations.


24/24

PHG Foundation is the trading name of the Foundation for Genomics and Population Health

A charitable company registered in England and Wales

Company Number: 5823194 Charity Number: 1118664

Date post:	03-Jun-2018
Category:	Documents
Upload:	junaida-rahmi
View:	221 times
Download:	0 times

Epidemiology Risk and Causation Report

Documents