CHAPTER 11 CASE CONTROL STUDIES. Copyright 2008 Health Administration Press. All rights reserved.

Transcription

1 CHAPTER 11 CASE CONTROL STUDIES 1

2 CASE CONTROL STUDIES o Epidemiological studies o Experimental studies o Observational studies o Case control studies o Case comparison o Case referent o Retrospective study 2

3 WHEN TO USE CASE CONTROL STUDIES o Exposure data are difficult/expensive to obtain o Disease is rare o Disease has long induction/latent period o Little is known about disease o Underlying population is dynamic 3

4 CLASSIFICATION OF STUDIES o When disease is identified o When exposure/treatment is recognized o When analysis is conducted 4

5 CLASSIFICATION OF CASE CONTROL STUDY o When disease is identified PRESENT o When exposure/treatment is recognized PAST o When analysis is conducted PRESENT 5

6 6

7 CASE CONTROL ASSUMPTIONS o Frequency disease population is small o Cases/controls are representative o Can t calculate relative risk (RR) directly 7

8 CHANGING VIEW OF CASE CONTROL STUDIES o Traditional (pre-1980s) o Case control as inferior alternative to cohort o Cohort: cause effect o Case control: cause effect, therefore backward o Called TROHOC studies (cohort spelled backward) Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 8

9 CHANGING VIEW OF CASE CONTROL STUDIES o Modern view o Case control study as a method of sampling population that gave rise to cases o Control group provides info of exposure distribution in population that gave rise to cases o Can calculate relative risk directly o Exposed/unexposed numerators come from cases o Exposed/unexposed denominators come from sample of population (controls) + cases Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 9

10 SELECTING CASES o Cases usually have the disease o Define cases specifically o Signs/symptoms o Clinical exams o Diagnostic tests o Err on being restrictive rather than inclusive 10

11 SOURCES OF CASE IDENTIFICATION o Clinic patient rosters o Death certificates o Surveys o Cancer/birth defect registries o Use incident, not prevalent, cases when possible why? 11

12 SOURCES OF CONTROLS o Population controls o Tax lists, voter registration, driver s license rosters, telephone directories, random digit dialing o Come from same population base but time consuming and expensive o Hospital/clinic controls o Illnesses in control must be unrelated to exposure o Illnesses in control must have same referral pattern to healthcare facility o Dead controls when some cases deceased o Friend/spouse/relative controls o Likely to share same socioeconomic status, race, etc., but cases reluctant to nominate and may share same habits Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 12

13 THE BALANCE IN SELECTING CONTROLS Matching versus finding controls 13

14 EXPOSURE o Characteristics/events that increase or decrease probability of disease/disability/death o Protective and malicious exposures 14

15 SOURCES OF EXPOSURE INFORMATION o In-person/telephone interview o Questionnaires o Medical/pharmacy/registry/employm ent/ insurance/birth/death records o Biological specimens biomarkers Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 15

16 CASE CROSSOVER STUDY o o o When brief exposure causes transient change in risk of a rare acute-onset disease o o o Risk of acute myocardial infarction (AMI) immediately following exertion Risk of motor vehicle (MV) collision while using cell phones Risk of unsafe sex following consumption of alcohol Cases serve as their own controls Exposure frequency during hazard period compares to nonhazard period (control) o o o Hazard period defined above as one hour before AMI o Get data on physical exertion inside and outside the one-hour window Hazard period defined above as ten minutes before MV accident o Compare phone activity during ten-minute window to a control period of time Get risk of exposure during hazard period compared to control period, e.g., risk of sex without and with alcohol consumption Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 16

17 RELATIVE RISK IN CASE CONTROL Relative risk = number of times more likely cases are to get disease than controls given exposure 17

18 MEASURING RELATIVE RISK The 2 2 table Disease No Disease Exposure (cases) (controls) With factor a b Without factor c d Odds ratio estimate of relative risk = ad/bc 18

19 Design 19

20 20

21 21

22 22

23 THE ODDS RATIO (OR) o An estimate of relative risk o When cases/controls are representative o If disease prevalence is small o If > 1.0, then increased risk o If < 1.0, then decreased risk (protective) 23

24 ATTRIBUTABLE FRACTION o How much is risk factor responsible for disease? o Proportion of disease attributed to risk factor o Attributable fraction (AF) = p(or 1) 100% p(or 1) + 1 p = proportion with risk factor OR = odds ratio 24

25 ATTRIBUTABLE FRACTION EXAMPLE o A study of the risk factors for lung cancer (Kreuzer et al. 1998) o ORs of 15.9 and 29.9 for males and females, respectively o Assume smoking prevalence: men (36.8 percent), women (21.5 percent) o AF (males) = ( 14. 9) % ( 14. 9) 1 o AF (females) = ( 28. 9) % ( 28. 9) 1 o o Male smokers: 15.9 times as likely to get lung cancer; smoking responsible for 84.5 percent of disease Female smokers: 29.9 times as likely to get lung cancer; smoking responsible for percent of disease

26 ATTRIBUTABLE FRACTION OTHER EXAMPLES o Human papillomavirus and cervical cancer in Morocco: OR = 61.6 and AF = 92 percent o Why is AF low? o Low odds ratio o Low prevalence of the risk factor o Induced abortion and breast cancer (Daling et al. 1996) o OR = 1.4, 30 percent get abortion, AF = 10.7 percent 26

27 CROSS-SECTIONAL STUDY (PREVALENCE STUDY) o Usually, survey done at a particular point in time snapshot of the population o Both exposure and disease outcome determined simultaneously o Cases of disease are prevalent (not incident) cases 27

28 28

29 HOW TO KNOW STUDY IS VALID o Eliminate alternative explanations o Bias o Confounding o Random error 29

30 BIAS o Definition: systematic error that leads to incorrect/invalid estimate of association o Easier to avoid bias than remove or fix bias o Types of bias o Selection bias o Observation (or misclassification) bias o How to evaluate for bias? o Identify source o Estimate magnitude o Assess direction Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 30

31 TOO MANY NAMES o More than 100 different kinds of bias with all sorts of names o I like to think of two broad categories o Selection bias how groups (disease or exposure) are selected o Misclassification bias whether a subject is incorrectly classified on the basis of disease or exposure 31

32 SELECTION BIAS o Definition: error due to systemic differences in characteristics between those selected for study and those not (Last) o How subjects are chosen o Case control if different criteria related to exposure o Retrospective cohort if selection of exposed or unexposed group related to outcome Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 32

33 SELECTION BIAS WITH CASE CONTROL o If using hospitalized patients, bias is called Berkson bias o Suppose you are trying to associate exposure (X) with disease A using disease B as controls o Bias occurs if A, B, and X have different probabilities of admission to hospital o No bias if: o Can t be hospitalized simply because have X; or o Rate of admission to hospital is same for A and B 33

34 EXAMPLE OF BERKSON BIAS Suppose you want to test if depression (X) is a risk factor for breast cancer (A) using pneumonia patients (B) as controls 34

35 HOW TO PREVENT SELECTION BIAS? o Define study population independent of disease, not after cases appear (if possible, define study population prior to follow-up) o o o Get same information from cases and controls (selection criteria the same) Don t let disease influence the availability of information, e.g., in occupational cohort study if those who develop disease have records sent to their physicians and they are misplaced Don t let disease influence loss of subjects to follow-up, e.g., in cohort study if those who don t develop disease drop out of study more than those who do 35

36 INFORMATION BIAS o Definition: the means of obtaining information about subjects is inadequate incorrect o Misclassification bias o Bias in abstracting records o Bias in interviewing o Bias from surrogate interviews o Surveillance bias o Recall bias o Reporting bias 36

37 MISCLASSIFICATION BIAS o Misclassification by disease or exposure o Four types: o Nondifferential misclassification bias of disease o Differential misclassification bias of disease o Nondifferential misclassification bias of exposure o Differential misclassification bias of exposure o Differential bias: opposite directions o Nondifferential bias: same direction 37

38 NONDIFFERENTIAL MISCLASSIFICATION BIAS OF DISEASE o Definition of disease gap too narrow or too wide o Bias underestimates effect o Shifts relative risk toward

39 DIFFERENTIAL MISCLASSIFICATION BIAS OF DISEASE o Classification as cases/controls depends on exposure o Exposed more likely to be cases o Tendency without blinding o Fen-Phen example o Bias unpredictable in terms of effect on odds ratio 39

40 NONDIFFERENTIAL MISCLASSIFICATION BIAS OF EXPOSURE o Misrepresent exposure status o Cases/controls both overestimate and underestimate o Bias underestimates effect o Shifts relative risk (or odds ratio) toward

41 DIFFERENTIAL MISCLASSIFICATION BIAS OF EXPOSURE o Misrepresentation of exposure different for cases/controls o Either cases/controls misrepresent or bias is in different directions o Also called recall bias 41

42 EXAMPLE OF DIFFERENTIAL MISCLASSIFICATION BIAS OF EXPOSURE o Breast cancer and abortion in the Netherlands o Two regions where Roman Catholics are 63 percent of population o Two regions where Roman Catholics are 28 percent of population o Willingness to report abortion Source: Rookus and Van Leeuwen (1996). 42

43 Overall adjusted OR for both areas =

44 HOW TO INTERPRET THIS STUDY o Odds ratio in Roman Catholic areas exaggerated o Cases report exposure o Controls misrepresent exposures o Effect exaggerated: relative risk shifted 44

45 CONFOUNDING o Definition: mixing of effects between exposure, outcome, and third variable (confounder) o Factors other than exposure that risk of disease o Failure of comparison group to reflect exposed group (as if it were unexposed) o Can lead to biased results Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 45

46 CRITERIA FOR CONFOUNDER o Associated with exposure? o Associated with disease? o Not intermediate step in causal pathway Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 46

47 CONFOUNDING Exposure or risk factor Disease Confounder 47

48 CONFOUNDING: A MATTER OF STRENGTH (MAGNITUDE) AND DIRECTION o Magnitude of confounding = (RR crude RR adjusted)/rr adjusted o Size bias depends on degree of association o Effect of multiple confounders may be large o Direction of confounding o Exaggerate (positive confounding) o Hide (negative confounding) Source: Aschengrau, A., and G. R. Seage Essentials of Epidemiology. Boston: Jones and Bartlett. 48

49 Confounding Example 1 Positive Confounding Oral contraceptive use + Myocardial infarction + Smoking + Source: Elwood (1998). 49

50 Confounding Example 2 Negative Confounding Oral contraceptive use + Myocardial infarction - + Obesity Source: Elwood (1998). 50