Should you get rid of the nightlight in your child’s room?

1. No, keep nightlight - B Yes, get rid of nightlight - A Need more information - C Quinn et al. 1999 Does the use of nightlights in children cause them to have myopia (nearsightedness)? Prevalence ratio = Should you get rid of the nightlight in your child’s room?

2. Confounding and Interaction I • Confounding: one of the central problems in observational human subjects research • What is it? What does it do? • Positive and negative confounding • Use of counterfactuals to conceptualize origins of confounding • Definition of/criteria for a “confounder” • Historical -- narrative definitions • Modern -- directed acyclic graphs (DAGs) - “Common causes” are root source of confounding • Use of DAGs to: • Identify what to adjust for • Know what not to adjust for (“colliders”) • How to handle multiple causal pathways • Confounding is a substantive, not statistical issue

3. Matches and Lung Cancer • A tobacco company researcher believes that exposure to matches is a cause of lung cancer • He conducts a large case-control study to test this hypothesis • Exposure odds ratio = (820/180) / (340/660) = disease odds ratio • OR = 8.8 (95% CI: 7.2, 10.9) • Should we remove matches from the environment as a means of preventing lung cancer?

4. Smoking, Matches, and Lung Cancer Crude OR crude Stratified Smokers Non-Smokers OR CF+ = ORsmokers OR CF- = ORnon-smokers • Stratification produces two 2-by-2 tables • In each stratum, all subjects are homogeneous with respect to smoking • We have stratified, adjusted,controlled, or conditioned for smoking • ORcrude = 8.8 (7.2, 10.9) • ORsmokers = 1.0 (0.6, 1.5) • ORnon-smoker = 1.0 (0.5, 2.0) • ORadjusted = 1.0 (0.5, 2.0)

5. Confounding: Smoking, Matches, and Lung Cancer • Illustrates how confounding can create an apparent effect even when there is no actual true effect • Can also be opposite: confounding can mask an effect when one is truly present • Proper terminology • In the relationship between matches and lung cancer, smoking is a confounding factor or a confounder • Smoking confounds the relationship between matches and lung cancer • In research, confounding has very specific meaning

6. Estes continues to be confounding puzzle Ray RATTO SHAWN ESTES seemed loath to analyze his own performance last night, for fear that people would see the first three innings and use them to obscure the last four. But that's what made his outing so perfectly Estes-like -- an ongoing argument with himself that he eventually won. Well, an argument in which he held his own and his teammates won for him in the bottom of the ninth. Ramon Martinez lined a game-tying single with two outs, and Jeff Kent followed two batters later with a bases-loaded walk off Juan Acevedo to give the Giants a 2-1 victory against Colorado and move them to within 4 1/2 games of division leader Arizona. It was in many ways an eye-opening victory for a team that hadn't had one of this type for a while.

7. Confounding: Examples of Magnitude and Direction Crude (unadjusted) OR crude Potential Confounder Absent Stratified (adjusted) Potential Confounder Present OR CF+ OR CF-

8. Nightlights Let there be light!

9. Nightlights and Myopia • Quinn et al. Nature 1999 • Prevalence Ratio =

10. Insert picture with nightlight off Lights are off and the stumbling around begins.

11. No, keep nightlight - B Yes, get rid of nightlight - A Need more information - C Prevalence ratio = Should you get rid of the nightlight in your child’s room? Quinn et al. 1999

12. No, keep nightlight - B Yes, get rid of nightlight - A Need more information - C Prevalence ratio = Should you get rid of the nightlight in your child’s room? Quinn et al. 1999

13. Nightlights and Myopia • Two subsequent studies found no association and explained the prior result by confounding • Zadnik et al. and Gwiazda et al. Nature, 2000

14. How might confounding account for the apparent effect of the night light on childhood myopia? Child’s Myopia Night Light ?

15. Parental Myopia Child’s Myopia Night Light ? How might confounding account for the apparent effect of the night light on childhood myopia? Positive or negative confounding? Positive

16. Insert picture with nightlight on again Let there be light (again)!

17. AZT to Prevent HIV After Needlesticks • Case-control study of whether post-exposure AZT use can prevent HIV seroconversion after needlestick (NEJM 1997) Crude ORcrude = 0.61 (95% CI: 0.26 - 1.4)

18. Sex of provider - B Severity of exposure - D Age of provider - A Age of patient - C Time of exposure - E AZT Use HIV acquisition ? What is the most important confounder at play here?

19. Severity of exposure - D Sex of provider - B Age of provider - A Age of patient - C Time of exposure - E Severity of Exposure AZT Use HIV Acquisition ? What is the most important confounder at play here?

20. Severity of Exposure AZT Use HIV Acquisition ? Negative or positive confounding? Positive confounding - B Negative confounding - A Need more information - C

21. Severity of Exposure AZT Use HIV Acquisition ? Negative or positive confounding? Negative confounding - A Positive confounding - B Need more information - C

22. Adjustment for Confounder Crude • Potential confounder: severity of exposure ORcrude = 0.61 Stratified Minor Severity Major Severity ORadjusted = 0.30 (95% CI: 0.12 – 0.79) Negative Confounding “Confounding by Indication”

23. Exposed to night lights Go back in time Unexposed to night lights time Counterfactuals: Conceptualizing Why Confounding Occurs Night lights and myopia • Ideal study: evaluate children exposed to night lights for several years and directly compare them to the SAME children not exposed to night lights • Result (e.g., risk ratio) is called the “causal effect measure” of night lights • Assuming no measurement or sampling error, the “causal effect measure” must be true. • However, since time has passed and children are older it is impossible to assess them without night lights • Hence, the ideal is “counterfactual” – contrary to the fact. It is unobservable. It cannot happen.

24. Counterfactuals: Conceptualizing Why Confounding Occurs Nights and Myopia • Because we cannot perform the counterfactual ideal (SAME population studied under 2 conditions), we must evaluate TWO distinct populations (exposed to night lights and unexposed) to study the problem Exposed to night lights Other influences Unexposed to night lights time • Result (e.g. risk ratio): a “measure of association” • The TWO distinct populations may be subject to different influences OTHER than just the night light • If these otherinfluences cause the disease under study, any difference in the risk ratio between the SAME population study (effect measure) and the TWO population study (measure of association) is what is known as confounding • Confounding occurs because of these other influences, a mixing of effects

25. Striving for the Counterfactual In the real (observable) world • All of our strategies in analytic studies are striving to simulate the counterfactual • We strive for our TWO distinct populations (exposed & unexposed) to be “exchangeable” • i.e., identical in the other influences upon them • Whenever the TWO distinct populations are “non-exchangeable”, confounding will occur • Our strategies to manage confounding are attempts to make our populations exchangeable

26. Why Strive for the Counterfactual?Causal Inference • Counterfactual ideal would allow certainty in knowing whether a numerical/statistical association between an exposure and outcome is causal • Knowing whether a relationship is causal is the “holy grail” in human subjects research • If a relationship is causal, interventions that change the exposure will change the outcome

27. Counterfactuals • Also known as “potential outcomes” • “Potential” refers to something that could happen (although actually never does) • Useful concept for understanding: • Origins of confounding • Advanced approaches to managing confounding • See BIOSTAT 215

28. Definition of/Criteria for a Confounder: A History The Narrative Era • Confounding occurs because of mixing between exposures of interest and unwanted extraneous factors. These extraneous factors are termed confounders. • Simplest Definition • A confounder is a variable which is associated with both exposure and disease. • Traditional Definition A confounding factor must: 1. Be associated with the exposure under study in the source population 2. Be a risk factor for the outcome 3. Not be affected by (caused by) the exposure or the outcome • Refined Definition 2. Factor must affect the outcome Sensitive but not fully specific Also not fully specific

29. C3 C1 C2 C1 D E ? D E ? Definition of/Criteria for Confounding: A History The Modern Era: Graphical Confounding occurs if there is a factor C that is a “Common Cause” of both E and D Depicted in a Directed Acyclic Graph (DAG) C part of a “non-causal” (aka “biasing”, “backdoor”, “confounding”) path from E to D. C E D ? C2

30. Directed Acyclic Graph (DAG) History • Humans have been drawing diagrams to depict relationships since they learned how to write. • Formal rules for such graphs started in the fields of engineering, computer science, & artificial intelligence • Adapted for epidemiology in 1990’s by computer scientist Judea Pearl (father of journalist Daniel Pearl) Y W X E D ? Basic Rules • Consist of nodes or vertices (variables) & edges (lines with arrowheads) • “Directed”: all edges have one-way direction which depict causal relationships • “Acyclic”: there is never a complete circle around any node (i.e. no factor can cause itself) • Hashed edge with ? sometimes used to depict relationship under study (purpose of the study) • Other edges placed based on prior causal knowledge

31. WARNING • S + N text does not follow contemporary rules of DAG depiction (requirement of uni-directional edges, etc.), but we will follow these rules in lecture and in problem sets • We don’t expect to see any two-headed edges (connections) in the problem sets

32. Directed Acyclic Graph (DAG) Some more terminology • All nodes immediately caused by another node are called “child” node; the proximal node is called the “parent” Parent of K and E L J K H G C F A B D E ? Descendants of E • All nodes along a path with edges in same direction are “descendants”; all proximal nodes are “ancestors”

33. Connections Between Variables A B D E • Path – any series of edges, regardless of direction, between two nodes (e.g., between E and D) C F D E G H D E D E J K L Types of Paths • Directed path – aka “causal” path • Series of edges are all going in the same direction • All edges have a head-to-tail sequence • “one way street” • Depicts a causal relationship • Undirected path – aka “non-causal” path • Any other series of edges between 2 nodes • “Backdoor path” – an undirected path where the initial edge points towards the initial node

34. How do we use DAGs? D E ? Major Use • For causal/etiologic research: Does E cause D? • Decide what factors to contend with in order to manage (i.e., eliminate or preclude) confounding when investigating the causal relationship between E and D C Process • Establish exposure and outcome • Draw all non-causal paths between exposure and outcome • Via either study design and/or analysis, develop a plan that closes (“blocks”) all the non-causal paths • If there remains an association between E and D in your data, this suggests a causal relationship (although measurement error and chance are still possible explanations)

35. How do we use DAGs? • For causal/etiologic research: Does E cause D? Smoking Lung Cancer Matches ? • Smoking is a “common cause” of matches & lung cancer • Controlling for smoking blocks the non-causal (backdoor) path and unconfounds relationship

36. How do we use DAGs? Genetic Factor (not measured) History of birth defects Birth Defects Multivitamin Use ? • For causal/etiologic research: Does E cause D? • A genetic factor is a “common cause” of multivitamins use & birth defects • Controlling for history of birth defects blocks the non-causal (backdoor) path and unconfounds relationship

37. How do we use DAGs? • For causal/etiologic research: Does E cause D? General Health (self-reported) Biologic Factors (not measured) Sexual Activity Mortality ? • An unmeasured biologic factor is a “common cause” of sexual activity & mortality • Controlling for self-reported general health blocks the non-causal (backdoor) path and unconfounds relationship

38. Confounding is root source of one type of a non-causal path D E ? A Note on Terminology • C is the common cause, but A, B, X, and Z in addition to C are known as “confounders” C B A X Z • Adjusting (e.g., stratifying) for A, B, X, or Z (or C) will block this non-causal path and eliminate confounding • DAGs focus on the process of confounding (and how to eliminate) rather than on confounders per se

39. Genetic Factor (not measured) History of birth defects Birth Defects Multivitamin Use ? Attraction of DAGs for Management of Confounding • Abstract: The Traditional Criteria for Confounding 1. Be associated with the exposure under study in the source population 2. Be a risk factor for the outcome 3. Not be affected by (caused by) the exposure or the outcome • More tangible: DAGs • Draw the system • Look for “common causes” of exposure and disease, and their attendant non-causal paths • Control for something on the non-causal path

40. The Central Challenge in Confounding DAGs provide the framework However, to avoid confounding, you need to be a subject matter expert to draw the DAG (in addition to being a methodologic expert) Confounding is mainly a substantive rather than statistical issue Advice: before planning a study, spend several weeks in the library to understand the surrounding system.

41. Confounding Paths are Not the Only Type of Non-Causal Paths i.e., DAGs can tell us a lot more about bias

42. RQ: Does lack of folate cause birth defects? Folate Intake Birth Defects ? Should we control for (e.g., stratify) stillbirths (spontaneous abortions)? Slone Epidemiology Unit Birth Defects Study • Stillbirths are associated with folate intake, even among infants without birth defects: OR = 0.50 (folate intake is protective against stillbirths) • Stillbirths are associated with birth detects: OR = 15.2 • Stillbirths are not on the causal pathway between folate and birth defects • In the past, other investigators have commonly adjusted for stillbirths in analyses. Hernan AJE 2002

43. No - B Yes - A Need more information - C RQ: Does lack of folate cause birth defects? • Stillbirths are associated with folate intake, even among infants without birth defects: OR = 0.50 (folate intake is protective) • Stillbirths are associated with birth detects: OR = 15.2 • Stillbirths are not on the causal pathway between folate and birth defects Folate Intake Birth Defects ? Should we adjust for (e.g., stratify) stillbirths? Hernan AJE 2002

44. RQ: Does lack of folate cause birth defects? • Stillbirths are associated with folate intake, even among infants without birth defects: OR = 0.50 (folate intake is protective) • Stillbirths are associated with birth detects: OR = 15.2 • Stillbirths are not on the causal pathway between folate and birth defects Folate Intake Birth Defects ? Should we adjust for (e.g., stratify) stillbirths? No - B Yes - A Need more information - C Hernan AJE 2002

45. Adjustment for Stillbirths Slone Epidemiology Unit Birth Defects Study Crude ORcrude = 0.65 (95% CI 0.45 – 0.95) Stratified No stillbirth Stillbirth ORadjusted = 0.80 (95% CI: 0.53 – 1.2) Apparent positive confounding Public health implication: No reason for women to supplement diet with folate Hernan AJE 2002

46. DAGs Identify What Not to Control For RQ: Does lack of folate intake cause birth defects? Undirected edge (interpret as going either direction) Stillbirths Folate Intake Birth Defects ? Stillbirths are a “common effect” of both exposure & disease – not a common cause. Common effects are called “colliders” Adjusting for colliders OPENS paths. Will actually result in bias. It is harmful. Hernan AJE 2002

47. Colliders are the Basis of the Other Type of Non-Causal Path Colliders are the common effect of 2 parents Undirected edge (interpret as going either direction). M E D ? Controlling for a collider induces a numerical relationship between the parents Collider as part of a backdoor path B A M D E ?

48. ? Conditioning (Stratifying) upon a Collider • A pair of dice (die A and die B) • We know they are independent • What if we stratify upon the sum of the dice? • This is stratifying for a collider • e.g., in stratum where sum = 7 • Now, if you know A, you know B • Stratifying has induced a relationship between A and B that otherwise does not exist. Sum of A and B Die A Die B

49. Confounding - B Mediator variable - D Nuisance causal pathway - A Selection Bias - C Some other answer - E Subject’s desire to participate in the research study Exposure Disease ? What does this DAG depict?

50. Subject’s desire to participate in the research study Exposure Disease ? What does this DAG depict? Selection Bias - C Confounding - B Mediator variable - D Nuisance causal pathway - A Some other answer - E