Clinical Epidemiology: Evidence of Risk and Harmgalton.uchicago.edu/~thisted/courses/315/lectures/1397.pdf · Health Studies 315 1 Clinical Epidemiology: Evidence of Risk and Harm

Health Studies 315Health Studies 315

1

Clinical Epidemiology:

Evidence of Risk and Harm

Clinical Epidemiology:

Evidence of Risk and Harm

2

Patients encounter (possibly) risky exposuresPatients encounter (possibly) risky exposures

• Alcohol during pregnancy (fetal risk)

• Electromagnetic fields (cancer risk)

• Vasectomy (prostate cancer risk)

• Oral contraceptives (thromboembolism risk)

• Alcohol during pregnancy (fetal risk)

• Electromagnetic fields (cancer risk)

• Vasectomy (prostate cancer risk)

• Oral contraceptives (thromboembolism risk)

To examine such risksTo examine such risks

• Evaluate validity of data

• Evaluate strength of association between risk and outcome

• Evaluate relevance to particular individuals

• Evaluate validity of data

• Evaluate strength of association between risk and outcome

• Evaluate relevance to particular individuals

3

Most studies of risk are observational studiesMost studies of risk are observational studies

• These studies are nonrandomized

• Need:

• Basic rules of evidence for nonrandomized studies

• Pitfalls to which observational studies are prone

• These studies are nonrandomized

• Need:

• Basic rules of evidence for nonrandomized studies

• Pitfalls to which observational studies are prone

4

Types of observational studiesTypes of observational studies

• Cohort studies

• Case-control studies

• In each case, we need

• Well-defined outcome

• Well-defined exposure

• Clearly identified comparison groups

• Similarity between groups wrt all other factors that might affect the outcome

• Cohort studies

• Case-control studies

• In each case, we need

• Well-defined outcome

• Well-defined exposure


• Similarity between groups wrt all other factors that might affect the outcome

5

Compare to randomized trialCompare to randomized trial

• Well-defined outcome: Primary endpoint

• Well-defined exposure: Active treatment

• Clearly identified comparison groups: Control group for comparison

• Similarity between groups wrt all other factors that might affect the outcome: Randomization

• Well-defined outcome: Primary endpoint

• Well-defined exposure: Active treatment

• Clearly identified comparison groups: Control group for comparison

• Similarity between groups wrt all other factors that might affect the outcome: Randomization

6

Cohort studiesCohort studies

• Prospective

• Useful when pts cannot be randomized to exposure

• Identify groups of exposed and non-exposed

• Follow forward to determine rate of outcome in each group

• Prospective

• Useful when pts cannot be randomized to exposure

• Identify groups of exposed and non-exposed

• Follow forward to determine rate of outcome in each group

7

Example of question for cohort studyExample of question for cohort study

• Do operating room personnel suffer higher rates of miscarriage than do others?

• Outcome

• Exposure


• Similarity between groups re outcome

• Do operating room personnel suffer higher rates of miscarriage than do others?

• Outcome

• Exposure



8

Problems with cohort studiesProblems with cohort studies

• Primary issue: self-selection to group• Same factors that affect selection might also

affect outcome• Example: relationship of NSAIDs to GI bleeding

• Increased age associated with increased use of NSAIDs

• Increased age associated with increased GI bleeding

• Age here is a “confounding variable”• Document differences

• Adjust for statistically (regression)

• Primary issue: self-selection to group• Same factors that affect selection might also

affect outcome• Example: relationship of NSAIDs to GI bleeding

• Increased age associated with increased use of NSAIDs

• Increased age associated with increased GI bleeding

• Age here is a “confounding variable”• Document differences

• Adjust for statistically (regression)

9

Relative strength of evidence: RCT & CohortRelative strength of evidence: RCT & Cohort

• Can never rule out the presence of unidentified confounders

• A good cohort study requires a great deal of investigator ingenuity

• Identify possible biases

• Document and/or adjust for potential biasing factors

• Thus, cohort studies are inherently less convincing than well-conducted clinical trials

• Can never rule out the presence of unidentified confounders

• A good cohort study requires a great deal of investigator ingenuity

• Identify possible biases

• Document and/or adjust for potential biasing factors

• Thus, cohort studies are inherently less convincing than well-conducted clinical trials

10

Case-control studiesCase-control studies• Retrospective• Useful in assessing risks for very rare outcomes• Useful when time between exposure and risk is

long• in utero exposure to DES, clear-cell adenocarcinoma of the

vagina

• asbestos, mesothelioma

• Identify persons with the outcome of interest (“cases”)

• Identify similar individuals (“controls”) who• do not have outcome

• are similar wrt all other factors associated with outcome

• Compare frequency of exposure between groups

• Retrospective• Useful in assessing risks for very rare outcomes• Useful when time between exposure and risk is

long• in utero exposure to DES, clear-cell adenocarcinoma of the

vagina

• asbestos, mesothelioma

• Identify persons with the outcome of interest (“cases”)

• Identify similar individuals (“controls”) who• do not have outcome

• are similar wrt all other factors associated with outcome

• Compare frequency of exposure between groups11

Example of question for case-control studyExample of question for case-control study

• What exposures predispose to lung cancer?

• Outcome

• Exposure



• What exposures predispose to lung cancer?

• Outcome

• Exposure



12

Problems with case-control studiesProblems with case-control studies

• Susceptiblity to unmeasured confounders

• Outcome = identified with outcome of interest

• Ascertainment bias

• Exposure = retrospective documentation of exposure

• Recall bias

• Interviewer bias

• Strength of C-C evidence is inherently more limited than cohort study

• Susceptiblity to unmeasured confounders

• Outcome = identified with outcome of interest

• Ascertainment bias

• Exposure = retrospective documentation of exposure

• Recall bias

• Interviewer bias

• Strength of C-C evidence is inherently more limited than cohort study 13

Second primary guide:Second primary guide:

• Were the exposures and outcomes measured in the same way in the groups being compared?

• RCT, cohort: ascertainment of outcome

• Case-control: ascertainment of exposure

• Were the exposures and outcomes measured in the same way in the groups being compared?

• RCT, cohort: ascertainment of outcome

• Case-control: ascertainment of exposure

14

Secondary guidesSecondary guides

• Is the temporal relationship correct?

• Is there a dose-response gradient?

• Is the temporal relationship correct?

• Is there a dose-response gradient?

15

Measuring the association between exposure and outcomeMeasuring the association between exposure and outcome

• Most common and useful: relative risk = p1/p2

• Exposure to encainide was associated with a 2.6-fold elevation in risk of death

• Can obtain relative risk from cohort studies or RCT (prospective)

• Most common and useful: relative risk = p1/p2

• Exposure to encainide was associated with a 2.6-fold elevation in risk of death

• Can obtain relative risk from cohort studies or RCT (prospective)

16

Case-control studies: No RR estimateCase-control studies: No RR estimate

17

• CC studies cannot produce an estimate of the relative risk

• RR = a/(a+b) / c/(c+d)

• OR = a/b / c/d = ad / bc

• When outcome is rare, OR RR

• CC studies cannot produce an estimate of the relative risk

• RR = a/(a+b) / c/(c+d)

• OR = a/b / c/d = ad / bc

• When outcome is rare, OR RR

aa bb

cc dd

18

Assessing applicabilityAssessing applicability

• Recall that the number needed to treat (NNT) depends on the absolute risk difference, not the relative risk

• Need to know something about the prevalence of the bad outcome

• Recall that the number needed to treat (NNT) depends on the absolute risk difference, not the relative risk

• Need to know something about the prevalence of the bad outcome

19

Case ControlCase Control

OR= 3.7p = 0.50OR= 3.7p = 0.50

Exposed

Unexposed

Exposed

Unexposed

Outcome: CC Adeno CAExposure: Maternal SmokingOutcome: CC Adeno CAExposure: Maternal Smoking

77 2121

11 1111

20


OR= 16.2p = 0.01OR= 16.2p = 0.01

Exposed

Unexposed

Exposed

Unexposed

Outcome: CC Adeno CAExposure: Any prior

pregnancy loss

Outcome: CC Adeno CAExposure: Any prior

pregnancy loss

66 55

22 2727

21


OR 300-400 p < 0.0001OR 300-400 p < 0.0001

Exposed

Unexposed

Exposed

Unexposed

Outcome: CC Adeno CAExposure: In utero estrogenOutcome: CC Adeno CAExposure: In utero estrogen

77 00

11 3232

Clinical Epidemiology: Evidence of Risk and Harmgalton.uchicago.edu/~thisted/courses/315/lectures/1397.pdf · Health Studies 315 1 Clinical Epidemiology: Evidence of Risk and Harm

Documents