Causality in  the health sciences: interpretation and rationale Federica Russo Universit é  catholique de Louvain
Causality  in  the sciences <ul><li>Interpreting  causality </li></ul><ul><ul><li>Epistemic theory </li></ul></ul><ul><ul>...
Overview:  interpreting causality <ul><li>Two types of evidence </li></ul><ul><ul><li>Probabilistic </li></ul></ul><ul><ul...
Probabilistic evidence <ul><li>Observed dependencies  </li></ul><ul><li>in a range of similar studies </li></ul><ul><li>Co...
Mechanistic evidence <ul><li>Biomedical mechanisms </li></ul><ul><ul><li>Chemical reactions, electric signals,  </li></ul>...
We need both types of evidence <ul><li>Semmelweis and puerperal fever </li></ul><ul><ul><li>He had statistics but the link...
A classic: Bradford Hill’s criteria <ul><li>Strength of association </li></ul><ul><li>Temporality </li></ul><ul><li>Consis...
Contemporary medicine: IARC <ul><li>IARC reviews published studies </li></ul><ul><li>Assessment of causality depends on: <...
Monistic accounts won’t do <ul><li>Mechanistic accounts </li></ul><ul><ul><li>Causal processes intersect  </li></ul></ul><...
Pluralistic accounts won’t do … either <ul><li>Uniformity of causal language: </li></ul><ul><li>A  single  notion of cause...
The way out: epistemic causality <ul><li>Rational causal beliefs: </li></ul><ul><ul><li>The agent’s evidence determines  <...
Constraints on causal beliefs <ul><li>The agent’s causal beliefs should account  </li></ul><ul><li>for all known dependenc...
An application: epistemic causality in cancer science <ul><li>Dataset of clinical observations of past patients </li></ul>...
To sum up <ul><li>The health sciences need and employ two types of evidence </li></ul><ul><li>Monistic accounts won’t do <...
To conclude <ul><li>There is a key distinction between </li></ul><ul><ul><li>Evidence  from which we draw causal </li></ul...
Overview:  the rationale of causality <ul><li>The rationale: measuring variations </li></ul><ul><li>Arguments for the vari...
The rationale of causality <ul><li>An epistemological question </li></ul><ul><li>The rationale: measuring variations </li>...
Smoking and lung cancer Socio economic status Asbestos exposure Cigarette smoking Lung cancer
The case for  the rationale of variation <ul><li>Empirical arguments </li></ul><ul><li>Methodological arguments </li></ul>...
Regularists accounts <ul><li>A heritage of Hume (1748): </li></ul><ul><ul><li>“ A cause is an object, followed by another,...
Contemporary  regularists accounts <ul><li>A token event  c  causes  </li></ul><ul><li>a token event  e  </li></ul><ul><li...
Invariance under intervention <ul><li>Woodward (2003) </li></ul><ul><ul><li>A theory of causal explanation </li></ul></ul>...
Smoking and lung cancer Socio economic status Asbestos exposure Cigarette smoking Lung cancer
<ul><li>Woodward (2003) </li></ul><ul><ul><li>A theory of causal explanation </li></ul></ul><ul><ul><li>Causal generalisat...
Contrast and compare: <ul><li>Variation rather than regularity </li></ul><ul><ul><li>Regularity of what? Of a variation </...
Therefore … <ul><li> Regularity and invariance are </li></ul><ul><li>constraints  on the causal relation </li></ul><ul><l...
Epidemiology aims  at establishing  variational causal claims
Goals <ul><li>Epidemiology studies  </li></ul><ul><li>the  variability  of disease  </li></ul><ul><li>due to  variation  i...
<ul><li>Jewell 2004,  Statistics for epidemiology </li></ul><ul><ul><li>In this book we describe the collection of data th...
Goals <ul><li>Susser 1973  </li></ul><ul><li>Causal thinking in the health sciences </li></ul><ul><ul><li>Epidemiologists ...
Goals <ul><li>Lilienfeld and Stolley 1994  </li></ul><ul><li>Foundations of epidemiology </li></ul><ul><ul><li>A relations...
Methods: observational studies <ul><li>Consider: </li></ul><ul><ul><li>(1) ‘Exposure does/does not cause disease’ </li></u...
Methods: observational studies <ul><li>Cohort studies:  </li></ul><ul><ul><li>compare exposed individuals  </li></ul></ul>...
Methods: observational studies <ul><li>Epidemiological methods  </li></ul><ul><ul><li>Make  comparisons </li></ul></ul><ul...
Methods: risks and odds <ul><li>Variables: Exposure E, Disease D </li></ul><ul><li>Risk: </li></ul><ul><li>Relative Risk: ...
Methods: risks and odds <ul><li>Epidemiologists are interested in  </li></ul><ul><li>ratios  between conditional probabili...
To sum up <ul><li> A rationale of causality is the principle that guides causal reasoning. </li></ul><ul><li>   We reaso...
As a general conclusion <ul><li>Biomedical research raises  </li></ul><ul><li>substantial philosophical issues </li></ul><...
Upcoming SlideShare
Loading in …5
×

Russo Madrid Medicine Oct07

695 views
640 views

Published on

Published in: Health & Medicine
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
695
On SlideShare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
6
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Russo Madrid Medicine Oct07

  1. 1. Causality in the health sciences: interpretation and rationale Federica Russo Universit é catholique de Louvain
  2. 2. Causality in the sciences <ul><li>Interpreting causality </li></ul><ul><ul><li>Epistemic theory </li></ul></ul><ul><ul><li>(“Interpreting causality in the health sciences”, ISPS, 21(2) 2007, with Jon Williamson) </li></ul></ul><ul><ul><li>Giving a rationale </li></ul></ul><ul><ul><li>Variation </li></ul></ul><ul><ul><li>(Variational casual claims in epidemiology, manuscript under submission) </li></ul></ul>
  3. 3. Overview: interpreting causality <ul><li>Two types of evidence </li></ul><ul><ul><li>Probabilistic </li></ul></ul><ul><ul><li>Mechanistic </li></ul></ul><ul><li>Against causal monism </li></ul><ul><li>Against causal pluralism </li></ul><ul><li>The case for epistemic causality </li></ul>
  4. 4. Probabilistic evidence <ul><li>Observed dependencies </li></ul><ul><li>in a range of similar studies </li></ul><ul><li>Coherent results </li></ul><ul><li>Tests for stability </li></ul><ul><li>in structural models </li></ul><ul><li>… </li></ul>
  5. 5. Mechanistic evidence <ul><li>Biomedical mechanisms </li></ul><ul><ul><li>Chemical reactions, electric signals, </li></ul></ul><ul><ul><li>alterations at the cellular level, … </li></ul></ul><ul><li>A plausible (physiological) link </li></ul><ul><li>from the cause to the effect </li></ul>
  6. 6. We need both types of evidence <ul><li>Semmelweis and puerperal fever </li></ul><ul><ul><li>He had statistics but the link </li></ul></ul><ul><ul><li>wasn’t accepted until backed </li></ul></ul><ul><ul><li>with the mechanism </li></ul></ul><ul><li>Helicobacter pylory </li></ul><ul><ul><li>The causal relation was hypothesised </li></ul></ul><ul><ul><li>based on probabilistic evidence and </li></ul></ul><ul><ul><li>accepted when backed with the mechanism </li></ul></ul>
  7. 7. A classic: Bradford Hill’s criteria <ul><li>Strength of association </li></ul><ul><li>Temporality </li></ul><ul><li>Consistency </li></ul><ul><li>Theoretical plausibility </li></ul><ul><li>Coherence </li></ul><ul><li>Specificity in the causes </li></ul><ul><li>Dose response relationship </li></ul><ul><li>Experimental evidence </li></ul><ul><li>Analogy </li></ul>
  8. 8. Contemporary medicine: IARC <ul><li>IARC reviews published studies </li></ul><ul><li>Assessment of causality depends on: </li></ul><ul><ul><li>Presence of a plausible mechanism </li></ul></ul><ul><ul><li>Probabilistic evidence </li></ul></ul><ul><ul><li>(e.g. frequencies, risks) </li></ul></ul>
  9. 9. Monistic accounts won’t do <ul><li>Mechanistic accounts </li></ul><ul><ul><li>Causal processes intersect </li></ul></ul><ul><ul><li>with each other in interactive forks </li></ul></ul><ul><li>Probabilistic accounts </li></ul><ul><ul><li>Causes make a difference </li></ul></ul><ul><ul><li>in the probability of the effect </li></ul></ul><ul><ul><li>(ceteris paribus) </li></ul></ul><ul><li>Problem: </li></ul><ul><li>Neither can handle the dual aspect </li></ul><ul><li>of causal epistemology </li></ul>
  10. 10. Pluralistic accounts won’t do … either <ul><li>Uniformity of causal language: </li></ul><ul><li>A single notion of cause is used </li></ul><ul><li>The pluralist rebuts: </li></ul><ul><ul><li>A mechanistic cause 1 </li></ul></ul><ul><ul><li>A probabilistic cause 2 </li></ul></ul><ul><ul><li>Different meanings of cause </li></ul></ul><ul><ul><li>But each refers to a single concept! </li></ul></ul><ul><li>Therefore, the pluralist has </li></ul><ul><li>twice as the problems of the monist! </li></ul>
  11. 11. The way out: epistemic causality <ul><li>Rational causal beliefs: </li></ul><ul><ul><li>The agent’s evidence determines </li></ul></ul><ul><ul><li>which beliefs to adopt </li></ul></ul><ul><ul><li>A causal relation is the set </li></ul></ul><ul><ul><li>of causal beliefs that an agent </li></ul></ul><ul><ul><li>with total evidence should adopt </li></ul></ul>
  12. 12. Constraints on causal beliefs <ul><li>The agent’s causal beliefs should account </li></ul><ul><li>for all known dependencies </li></ul><ul><li>that are not already accounted for </li></ul><ul><li>by non-causal dependencies </li></ul><ul><li>The agent’s causal beliefs should be </li></ul><ul><li>compatible with other knowledge </li></ul><ul><li>The agent should not have causal beliefs </li></ul><ul><li>that are not warranted by her evidence </li></ul>
  13. 13. An application: epistemic causality in cancer science <ul><li>Dataset of clinical observations of past patients </li></ul><ul><li>Dataset of observations at molecular level </li></ul><ul><li>Probabilistic evidence </li></ul><ul><li>Knowledge of biological mechanisms </li></ul><ul><li>Mechanistic evidence </li></ul><ul><li>Knowledge of semantic relations between variables </li></ul><ul><li>Pro/contra mechanistic evidence </li></ul>
  14. 14. To sum up <ul><li>The health sciences need and employ two types of evidence </li></ul><ul><li>Monistic accounts won’t do </li></ul><ul><li>Pluralistic accounts won’t do either </li></ul><ul><li>The epistemic account succeeds </li></ul>
  15. 15. To conclude <ul><li>There is a key distinction between </li></ul><ul><ul><li>Evidence from which we draw causal </li></ul></ul><ul><ul><li>conclusions and the concept of cause </li></ul></ul><ul><li>A single epistemic concept suits </li></ul><ul><li>the case of the health sciences </li></ul>
  16. 16. Overview: the rationale of causality <ul><li>The rationale: measuring variations </li></ul><ul><li>Arguments for the variation rationale </li></ul><ul><ul><li>Variation, not regularity </li></ul></ul><ul><ul><li>Variation, not invariance </li></ul></ul><ul><li>The variation rationale in epidemiology </li></ul><ul><ul><li>Goals </li></ul></ul><ul><ul><li>Methods: Observational studies </li></ul></ul><ul><ul><li>Odds & Risks </li></ul></ul>
  17. 17. The rationale of causality <ul><li>An epistemological question </li></ul><ul><li>The rationale: measuring variations </li></ul><ul><li>Variation: the bottom-line concept </li></ul>
  18. 18. Smoking and lung cancer Socio economic status Asbestos exposure Cigarette smoking Lung cancer
  19. 19. The case for the rationale of variation <ul><li>Empirical arguments </li></ul><ul><li>Methodological arguments </li></ul><ul><li>Foundational arguments </li></ul><ul><li>Objections </li></ul><ul><li>Methodological consequences </li></ul><ul><li>Foundational consequences </li></ul><ul><li>… </li></ul>
  20. 20. Regularists accounts <ul><li>A heritage of Hume (1748): </li></ul><ul><ul><li>“ A cause is an object, followed by another, </li></ul></ul><ul><ul><li>and where all objects similar to the first </li></ul></ul><ul><ul><li>are followed by objects similar to the second” </li></ul></ul><ul><ul><li>Causality is in our psychological habit of witnessing </li></ul></ul><ul><ul><li>effects that regularly follow causes </li></ul></ul><ul><ul><li>in time and space </li></ul></ul>
  21. 21. Contemporary regularists accounts <ul><li>A token event c causes </li></ul><ul><li>a token event e </li></ul><ul><li>if </li></ul><ul><li>events of type E regularly follow </li></ul><ul><li>events of type C </li></ul><ul><li>Jack’s smoking caused him lung cancer </li></ul><ul><li>if lung cancer regularly follows smoking </li></ul>
  22. 22. Invariance under intervention <ul><li>Woodward (2003) </li></ul><ul><ul><li>A theory of causal explanation </li></ul></ul><ul><ul><li>Causal generalisations are change-relating </li></ul></ul><ul><ul><li>Change-relating relations are explanatory: </li></ul></ul><ul><ul><li>they are invariant under a large class of interventions or environmental changes </li></ul></ul>
  23. 23. Smoking and lung cancer Socio economic status Asbestos exposure Cigarette smoking Lung cancer
  24. 24. <ul><li>Woodward (2003) </li></ul><ul><ul><li>A theory of causal explanation </li></ul></ul><ul><ul><li>Causal generalisations are change-relating </li></ul></ul><ul><ul><li>Change-relating relations are explanatory: </li></ul></ul><ul><ul><li>they are invariant under a large class of interventions or environmental changes </li></ul></ul>Invariance under intervention
  25. 25. Contrast and compare: <ul><li>Variation rather than regularity </li></ul><ul><ul><li>Regularity of what? Of a variation </li></ul></ul><ul><li>Variation rather than invariance </li></ul><ul><ul><li>Invariance of what? Of a variation </li></ul></ul>
  26. 26. Therefore … <ul><li> Regularity and invariance are </li></ul><ul><li>constraints on the causal relation </li></ul><ul><li> Variation conceptually precedes </li></ul><ul><li>regularity and invariance </li></ul>
  27. 27. Epidemiology aims at establishing variational causal claims
  28. 28. Goals <ul><li>Epidemiology studies </li></ul><ul><li>the variability of disease </li></ul><ul><li>due to variation in exposure </li></ul><ul><li>Appeal to regularity is </li></ul><ul><li>virtually absent </li></ul>
  29. 29. <ul><li>Jewell 2004, Statistics for epidemiology </li></ul><ul><ul><li>In this book we describe the collection of data that speak to relationships between the occurrence of disease and various descriptive characteristics in individuals in a population. Specifically, we want to understand whether and how differences in individuals might explain patterns of disease distribution across a population. </li></ul></ul>Goals
  30. 30. Goals <ul><li>Susser 1973 </li></ul><ul><li>Causal thinking in the health sciences </li></ul><ul><ul><li>Epidemiologists in search for causes want to make asymmetrical statements that have direction. They seek to establish that an independent variable X causes changes in the dependent variable Y and not the reverse. </li></ul></ul><ul><ul><li>The central problem of cohort studies is to cope with the change that occurs with the passage of time. The study of cause involves the detection of change in a dependent variable by change in an independent variable . </li></ul></ul>
  31. 31. Goals <ul><li>Lilienfeld and Stolley 1994 </li></ul><ul><li>Foundations of epidemiology </li></ul><ul><ul><li>A relationship is considered causal whenever evidence indicates that the factors form part of the complex circumstances which increase the probability of occurrence of disease and that a diminution of one or more of these factors decreases the frequency of disease. </li></ul></ul>
  32. 32. Methods: observational studies <ul><li>Consider: </li></ul><ul><ul><li>(1) ‘Exposure does/does not cause disease’ </li></ul></ul><ul><ul><li>(2) ‘The risk of disease is x times greater among exposed persons than unexposed persons </li></ul></ul><ul><li>To establish (1) we need to establish (2) first </li></ul><ul><li>Causal relations are established </li></ul><ul><li>through comparative statements </li></ul>
  33. 33. Methods: observational studies <ul><li>Cohort studies: </li></ul><ul><ul><li>compare exposed individuals </li></ul></ul><ul><ul><li>with non-exposed individuals </li></ul></ul><ul><li>Case control studies: </li></ul><ul><ul><li>compare diseased individuals </li></ul></ul><ul><ul><li>with non-diseased individuals </li></ul></ul><ul><li>Cross-sectional studies: </li></ul><ul><ul><li>compare various </li></ul></ul><ul><ul><li>individual characteristics </li></ul></ul><ul><ul><li>at a specific point of time </li></ul></ul>
  34. 34. Methods: observational studies <ul><li>Epidemiological methods </li></ul><ul><ul><li>Make comparisons </li></ul></ul><ul><ul><li>Aim at establishing variational causal claims </li></ul></ul><ul><li>But </li></ul><ul><ul><li>Do not aim at establishing whether </li></ul></ul><ul><ul><li>disease regularly and invariably </li></ul></ul><ul><ul><li>follows exposure </li></ul></ul>
  35. 35. Methods: risks and odds <ul><li>Variables: Exposure E, Disease D </li></ul><ul><li>Risk: </li></ul><ul><li>Relative Risk: </li></ul><ul><li>Odds: </li></ul><ul><li>Odds ratio: </li></ul>
  36. 36. Methods: risks and odds <ul><li>Epidemiologists are interested in </li></ul><ul><li>ratios between conditional probabilities </li></ul><ul><li>i.e., in quantifying how and to what extent </li></ul><ul><li>probability of disease varies according to </li></ul><ul><li>variations in the exposure </li></ul><ul><li>They do not measure whether </li></ul><ul><li>disease regularly follows exposure in time </li></ul><ul><li>But how the proportion of diseased individuals </li></ul><ul><li>changed in different exposure conditions </li></ul><ul><li>in a given lapse of time </li></ul>
  37. 37. To sum up <ul><li> A rationale of causality is the principle that guides causal reasoning. </li></ul><ul><li> We reason about variations </li></ul><ul><li> Epidemiology establishes </li></ul><ul><li>variational causal claims </li></ul><ul><ul><li>As stated in its goals </li></ul></ul><ul><ul><li>As shown by the comparative character of observational studies and risks & odds </li></ul></ul>
  38. 38. As a general conclusion <ul><li>Biomedical research raises </li></ul><ul><li>substantial philosophical issues </li></ul><ul><ul><li>About methods </li></ul></ul><ul><ul><li>About notions </li></ul></ul><ul><ul><li>About actions </li></ul></ul><ul><li>Substantial differences between </li></ul><ul><li>Evidence – Interpretation – Rationale </li></ul>

×