This document discusses common measures of association used in medical and epidemiological research, including odds, risk, odds ratios, and relative risk ratios. It explains how these measures are calculated using a 2x2 contingency table and how to interpret the results, including whether the exposure increases or decreases risk/odds. Confidence intervals are also discussed as they provide accuracy about the estimates. The document cautions about solely relying on p-values and advocates considering the magnitude and confidence in the measures of association.
Common measures of association in medical research handout
1. COMMON MEASURES OF
ASSOCIATION IN MEDICAL AND
EPIDEMIOLOGIC RESEARCH:
ODDS, RISK, & THE 2X2 TABLE
Patrick Barlow
PhD. Student in Evaluation, Statistics, & Measurement
The University of Tennessee
2. ON THE AGENDA
What are odds/risks?
The 2x2 table explained
Calculating measures of association
Odds Ratio
Risk Ratio
Interpreting measures of association
Magnitude of the relationship
Accuracy of the inference
The P-value fallacy
3. SOME TERMS
2x2 table
Proportion
Odds
Risk
Odds Ratio (OR)
Relative Risk Ratio (RR)
4. WHAT IS PROBABILITY?
The probability of a favorable event is the fraction of times you expect to
see that event in many trials. In epidemiology, a “risk” is considered a
probability.
For example…
You record 25 heads on 50 flips of a coin, what is the
probability of a heads?
Remember: a probability should never exceed 1.0
or 100%.
5. WHAT ARE ODDS?
An “odds” is a probability of a favorable event occurring vs. not
occurring.
For example…
What are the odds you will get a heads when flipping a fair
coin?
“The odds of flipping heads to flipping tails is 1 to 1”
In clinical and epidemiologic research, we use a ratio of two
odds, or Odds Ratio (OR) and Relative Risk Ratio (RR), to express
the strength of relationship between two variables.
6. RELATIVE RISK VS. ODDS RATIOS
Relative Risk (RR) is a more accurate measure of
incidence of an outcome of interest.
Used in prospective studies or when the total population
are known
What study designs would use RR?
An odds ratio (OR) provides researchers with an
estimate of RR in situations where the total
population is unknown.
What study designs would use ORs instead of RRs?
7. THE 2X2 TABLE
The basis of nearly every common measure of
association in medical and epidemiologic research
can be traced back to a 2x2 contingency table.
A B
C D
8. THE 2X2 TABLE
For every measure of association using the 2x2
table, your research question comes from the A
cell.
A B
C D
9. EXAMPLE
What is the risk of myocardial infarction (MI) if a
patient is taking aspirin versus a placebo?
Had MI No MI
Aspirin
A B
Placebo
C D
10. RELATIVE RISK ON A 2X2 TABLE
What is the risk of myocardial infarction (MI) if a
patient is taking aspirin versus a placebo?
Had MI No MI
Aspirin 50 1030
Placebo 200 1570
11. RELATIVE RISK ON A 2X2 TABLE
Had MI No MI
Aspirin 50 1030
Placebo 200 1570
What is the risk of MI if a patient is taking aspirin?
Risk of MI for aspirin = Number with MI / Number on Aspirin =
50 / 1080 = .048 or 4.8%
What is the risk of MI if a patient is taking placebo?
Risk of MI for placebo = Number with MI / Number on placebo
= 200 / 1770 = .11 or 11%
12. RELATIVE RISK ON A 2X2 TABLE
Had MI No MI
Aspirin 50 1030
Placebo 200 1570
So…
What is the risk of myocardial infarction (MI) if a patient
is taking aspirin versus a placebo?
RR = (A / A+B) / (C / C+D)
RR = Risk of MI for Aspirin / Risk of MI for Placebo
RR = .048 / .11 = .41 or 41%
13. YOUR TURN
Work in pairs to calculate the RRs for each of the
2x2 tables below.
No Lung
1 PE No PE
3 Lung Cancer
Cancer
DVT 79 157 Smoking Hx 190 450
No Smoking
No DVT 100 375 Hx 70 700
Glucose No DM Type
2 Tolerance
Improved
Tolerance not
Improved 4 DM Type II
II
Lap Band 35 170 BMI < 30 25 350
Gastric
Bypass 52 160 BMI > 30 65 200
14. YOUR TURN
Work in pairs to calculate the RRs for each of the
2x2 tables below.
RR = (79/79+157) / RR = (190/(190+450)) /
(100/100+375) = 1.59 (70/(70+700)) = 3.27
RR = (35/(35+170)) / RR = (25/(25+350)) /
(52/(52+160)) = .70 (65/(65+200)) = .27
15. ODDS RATIOS AND THE 2X2 TABLE
Recall…
Odds ratios are used to estimate RR when the true
population is unknown.
For discussion
Why can’t we just use RR all the time?
Will an OR and RR differ from one another? If so, how?
Odds ratios look at prevalence rather than
incidence of the event.
Remember:
OR = “Odds of having the outcome”
RR = “Risk of developing the outcome”
16. ODDS RATIOS AND THE 2X2 TABLE
Had MI No MI
Aspirin 50 1030
Placebo 200 1570
What are the odds of myocardial infarction (MI) if a
patient is taking aspirin versus a placebo?
OR = A*D / B*C
OR = 50*1570 / 1030 * 200 = .38 or 38%
17. YOUR TURN
Work in pairs to calculate the ORs for the same 2x2
tables as before. How do the ORs and RRs differ?
No Lung
1 PE No PE
3 Lung Cancer
Cancer
OR = (79*375)79(157*100) =
/ OR = (190*700) / (450*70) =
DVT 157 Smoking Hx 190 450
1.89 4.22
No Smoking
No DVT 100 375 Hx 70 700
Glucose No DM Type
2 Tolerance
Improved
Tolerance not
Improved 4 DM Type II
II
OR = (35*160) / (170*52) = .63 OR = (25*200) / (350*65) = .21
Lap Band 35 170 BMI < 30 25 350
Gastric
Bypass 52 160 BMI > 30 65 200
18. YOUR TURN
Work in pairs to calculate the ORs for the same 2x2
tables as before. How do the ORs and RRs differ?
OR = (79*375) / (157*100) = OR = (190*700) / (450*70) =
1.89 4.22
OR = (35*160) / (170*52) = .63 OR = (25*200) / (350*65) = .21
19. INTERPRETING ORS AND RRS: THE BASICS
Odds/Risk ratio ABOVE 1.0 = Your exposure
INCREASES risk of the event occurring
For OR/RRs between 1.00 and 1.99, the risk is
increased by (OR – 1)%.
For OR/RRs 2.00 or higher, the risk is increased OR
times, but you could also still use (OR – 1)%.
Example:
Smoking is found to increase your odds of breast
cancer by OR = 1.25. What is the increase in odds?
You are 25% more likely to have breast cancer if you are a
smoker.
Smoking is found to increase your risk of developing
lung cancer by RR = 4.8. What is the increase in risk?
You are 4.8 times more likely to develop lung cancer if you are
a smoker vs. non-smoker.
20. INTERPRETING ORS AND RRS: THE BASICS
Odds/Risk ratio BELOW 1.0 = Your exposure
DECREASES risk of the event occurring
The risk is decreased by (1 – OR)%
Often called a PROTECTIVE effect
Example:
Addition of the new guidelines for pacemaker/ICD
interrogation produced an OR for device interrogation of
OR = .30 versus the old guidelines. What is the
reduction in odds?
(1 – OR) = (1 – .30) = 70% reduction in odds.
21. INTERPRETING ORS AND RRS: THE BASICS
So for our example…
OR = .39
What is the reduction in odds?
So: “Taking aspirin provides a 61% reduction in the odds of
having an MI compared to a placebo.”
RR = .41
What is the reduction in risk?
So: “Taking aspirin provides a 59% reduction in risk of MI
compared to a placebo.”
22. INTERPRET THE FOLLOWING OR/RRS
OR = 3.00
OR = .39
RR = 1.50
OR = 1.00
RR = .22
RR = 18.99
OR = .78
What does the OR/RR say about the strength of
relationship?
23. OR/RR AND CONFIDENCE INTERVALS
The magnitude of the OR/RR only provides the
strength of the relationship, but not the accuracy
95% Confidence intervals are added to any OR/RR
calculation to provide an estimate on the accuracy
of the estimation.
95% of the time the true value will fall within a given
range
Wide CI = weaker inference
Narrow CI = stronger inference
CI crosses over 1.0 = non-significant
An OR/RR is only as important as the confidence
interval that comes with it
24. INTERPRET THESE 95% CIS
OR 2.4 (95% CI 1.7 - 3.3)
OR 6.7 (95% CI 1.4 - 107.2)
OR 1.2 (95% CI .147 - 1.97)
OR .37 (95% CI .22 - .56)
OR .57 (95% CI .12 - .99)
OR .78 (95% CI .36 – 1.65)
25. THE P-VALUE FALLACY
What is a p-value?
The probability that the observed statistics would occur
due to chance.
Alpha, usually set to .05
Values below .05 indicate a statistically significant
relationship exists.
What influences p-values?
Sample size
Chance
Effect size
Statistical power
Is a p-value of .001 a more significant relationship
than a value of .03?
26. GOING BEYOND THE P-VALUE
The OR/RR provides a far more vivid description of
the magnitude of the relationship.
Can you say an OR of 4.30 is stronger than an OR of
1.50?
What about RR = .25 vs. RR = .56?
The 95% CI provides far more information on the
accuracy of the inference.
Which is more accurate?
OR = 2.5 (95% CI = 1.2 – 10.0) vs. OR = 2.5 (95% CI = 1.2 –
3.1)
Hinweis der Redaktion
No matter what the situation, you can easily rearrange the table so that the research question you want is in the A cell. There are 4 different research questions for each 2x2 table, and you can change the values and labels around to answer the question you’re asked.
The OR consistently OVER-estimates the risk
Alternatively, the second example could be interpreted as: “Smoking increases your risk of lung cancer by 380% vs. non-smoking”
Assume (until I can find literature examples) that all of these are for generic “Exposure vs. Non-exposure” and “Disease vs. non-disease”
What influences p-values?Sample size: Larger sample sizes increases your likelihood of finding a statistically significant difference. Theoretically, the tiniest difference could be shown to be statistically significant if you have enough people. Many public health studies have massive sample sizes, so the statistically significant findings are very small practical differences.Chance: If Alpha is set to .05, then you have a 5% chance of making a Type I error, or, a false positive result. That is, you would conclude that a difference/association is statistically significant when you shouldn’t have.Effect size: Larger effect sizes (i.e. bigger odds ratios or higher correlation coefficients) are easier to find statistically significant because the association is stronger.Statistical power: Large sample sizes increase the “power” of your test to find a statistically significant difference. Low power increases you chance for a Type II error, or, “missing” the significant relationship.Is a p-value of .001 a more significant relationship than a value of .03? There is a lot of discussion about this, and it is more of a debate rather than a certainty, but what is trending now is to favor the effect size (e.g. OR or correlation coefficient) and 95% confidence intervals instead of just a p-value.
As you can see, it is much easier to claim an OR of 4.3 is HIGHER than one of 1.50, and it makes practical and intuitive sense to see this. Likewise, Saying we’re 95% confidence the difference lies between 1.2 and 3.1 is MORE accurate than between 1.2 and 10.0.