Case-Control Study Design
Two groups are selected, one of people with the disease (cases), and the other of people with the same general characteristics but without the disease (controls)
Compare the past exposures of both groups
1
Case Control Study Design
Exposed Diseased
(Cases)
Not Exposed Exposed
Target Population Not Diseased
(Controls)
Not Exposed
2
Case-Control Study Design
Limitations:
Cannot yield incidence rates because
subjects are selected on outcome
An estimate of the ratio of incidence rates or risks (RR) is obtained by calculating an
odds ratio (OR)
3
Odds Ratio Calculation
Outcome
Exposure Cases Controls
Exposed Not Exposed
A C
B D
Odds of exposure for cases Odds of exposure for controls
A/C B/D
= Odds Ratio (estimates the relative risk)
4
Comparing Odds Ratios and Relative Risks
Outcome Exposure Exposed Not Exposed Cases 70 30
100
Controls 300 700
1000
370 730
1100
OR = AD/BC = 5.44
RR = Ie/In = 4.41
5
Stating your results
OR = 5.44
Those with the disease are 5.44 times as likely to have had the exposure compared to those without the disease
RR = 4.41
Those with the exposure are 4.41 times as likely to develop the disease compared to those without the exposure
6
Summary of Strengths and Limitations of Prospective Cohort and Case-Control Studies
Prospective Cohort Case-Control
Strengths: Strengths: Opportunity to measure risk factors Useful for rare disease before disease occurs Relatively inexpensive Can study multiple disease Relatively quick results outcomes Limitations: Can yield incidence rates as well Possible bias in measuring risk as relative risk estimates factors after disease has occurred Limitations: Possible bias in selecting control Useful for rare disease group Relatively inexpensive Identified cases may not Relatively quick results represent exposure of all cases
7
Randomized Clinical Trials (RCT) The Gold Standard Cohort Study
8
Schematic diagram of a clinical trial
Study Population
Non-participants Participants
Randomization
Treatment arm Intervention or new treatment
Control arm Control
Improved
Not
Improved
Not Improved
Improved
9
Crossover Design
Subjects are randomized to a sequence of two or more treatments Each subject serves as his own control
Group I II Sequence A, B B,A Period 1 A B Washout Period 2 B A
10
Factorial Design
Two or more treatments are evaluated simultaneously in the same set of subjects using varying combinations of treatments
Randomization
Treatment A
Placebo
Treatment B
Placebo
Treatment B
Placebo
11
How do we evaluate whether cancer studies are valid?
Understand bias and confounding
Testing for a true association
Examine the methodology for bias Examine the analysis for confounding Examine the results for statistical significance
13
Examine the study design for Bias
Selection Bias
Errors in the process of identifying the study population and selecting the subjects
Information/Observation Bias
Errors in measurements of exposure or disease status
14
Confounding
Confounding is an apparent association
between disease and exposure caused by a third factor not taken into consideration
15
Examples of Confounders
Study A found an association between
gambling and lung cancer. The study may
be confounded by smoking.
Study B found a larger crude death rate in Florida than in Alaska. The rate may be confounded by differences in the population age structure.
16
Testing for Confounding
1. Calculate the crude rate
2. Calculate a rate adjusted for the confounding
variable
3. Compare the two measures
The two measures will be different if the variable
is a confounder (in practice, when the crude and adjusted measures differ by at least 10%)
17
Age
0-18 19-64 65+
Cancer Deaths
Population at risk
ASR
1980 U.S. Standard Population
Expected Number of Deaths
(1)
5 10 100
(2)
5,000 25,000 15,000
(1) / (2) = (3)
1.00 per 1000
(4)
60,500,000
(3) x (4) = (5)
60,500 56,120 171,419 288,039
0.40 per 1000 140,300,000 6.67 per 1000
25,700,000
Total
115
45,000
xxx
226,500,000
Crude Rate (115 / 45,000) x 1000 2.56 per 1,000
Not equal
Age-Adjusted Rate (288,039 / 226,500,000) x 1000 1.27 per 1,000
AGE IS A CONFOUNDER FOR DEATH FROM CANCER
18
Evaluating Statistical Significance
The probability that you would get your results by chance alone is the p-value A low p-value ( < 0.05 ) says that chance is not likely to explain your results A 95% confidence interval (CI) is the range of values in which the true value will be found 95% of the time Large samples yield small confidence intervals Small samples yield large confidence intervals
19
Evaluating Results
RR = 1: No difference in disease between exposed and unexposed groups OR = 1: No difference in exposure between cases and controls
Examples:
RR = 1.8 (1.6, 2.0) is statistically significant RR = 1.8 (0.8, 2.9) is not statistically significant OR = 0.7 (0.6, 0.8) is statistically significant OR = 0.7 (0.4, 1.2) is not statistically significant
20
How do we assess whether associations between cancer and risk factors are causal?
Understand criteria for causality
To Show Cause
Chronic disease and complex conditions require the use of Hill’s Postulates
Strength of association Consistency of association Specificity of association Temporality Biologic gradient
5. 6. 7. 8.
1. 2.
3. 4. 5.
Plausibility Coherence Experiment Analogy
22
How much of the morbidity and mortality from cancer might be prevented by interventions?
Understand the impacts of education and screening programs
Validity
Principles of Screening
Sensitivity: correctly identify those with disease
Specificity: correctly identify those without disease
+ Predictive Value: proportion of correct positive tests
- Predictive Value: proportion of correct negative tests
Reliability: ability of test to give consistent results
Yield: amount of unrecognized disease brought to treatment due to screening
24
Calculating Measures of Validity
True Diagnosis
Test Result Positive Negative Total
Disease a c a+c
No Disease b d b+d
Total a+b c+d a+b+c+d
Sensitivity = a/(a+c) Specificity = d/(b+d)
Positive Predictive Value = a/(a+b) Negative Predictive Value = d/(c+d)
25
Example: Breast Cancer Screening
Mammogram Results Disease No Disease Total
Positive Negative Total
132 45 177
983 63,650 64,633
1,115 63,695 64,810
Sensitivity = 132/177 = 74.6% Specificity = 63,650/64,633 = 98.5% Positive Predictive Value = 132/1,115 = 11.8%
Negative Predictive Value = 63,650/63,695 = 99.9%
26
Keys to Screening
Sensitivity: detect a sufficient number of preclinical cases to be useful
Prevalence: screen high-risk populations
Frequency: one-time screening does not allow for differences in individual risk or differences in onset
Participation: tests unacceptable to the target population will not be utilized Follow-up: those with positive tests need to be provided with a plan of action
27
Advice for Reading the Literature
Identify the study design Understand how subjects are selected Understand how exposure is defined
Evaluate potential bias and confounding Determine if the statistical evaluation is appropriate Make decisions about whether the outcome measures are statistically significant and/or clinically important Use good judgment
28
End