Lectures 16 & 17: Measures of Association & Effect

Question 1

What is the difference b/w measures of association vs statistical significance?

Answer 1

Measures of association
-measure the magnitude/ strength of the relationship b/w an exposure and outcome
-ie strength, might tell us if causal relationship exists

Statistical significance
-reflect the probability of having an association as great or greater than observed if the null hypothesis was true
-p-values, how likely, but only tells us the probability of having an association

Question 2

What is the difference b/w measures of association and measures of effect?

Answer 2

Association
-Measure the magnitude/strength of the relationship b/w an exposure and outcome as a relative effect (the RATIO of two measures of disease frequency)

Effect
-Express the effect of exposure on an outcome as an absolute effect (DIFFERENCE b/w two measures of disease frequency)
-Sub one from the other and can be + or - just looking for the absolute effect

Question 3

What are the 3 types of measures of ASSOCIATION?

Answer 3

1. Risk ratio (RR)
2. Odds ratio (OR)
3. Incidence Rate Ratio (IRR)

Question 4

What is risk ratio and how do you calculate it? (Also called relative risk)

Answer 4

-Proportion so the numerator is part of the denominator
RR= risk of disease in exposed group /
Risk of disease in non-exposed group
or [a/(a+b)] / [c/(c+d)]
-EX if the RR is 2, this means that the risk of the disease in the exposed group is 2 times the risk of disease in non-exposed group (the exposed are twice as likely to be diseased as non-exposed)

Question 5

How does the null value relate to RR?

Answer 5

-RR is a ratio therefore the null value is 1
-a RR=1 means that there is NO association b/w the exposure (E) and outcome (O)
-RR>1: exposure is pos associated with o (those exposed have a greater chance of having the outcome)
-RR<1: exposure is neg associated with o (those not exposed would be highly associated with having the outcome) usually for vax or protective factors
-same as last class if CIs for RR includes 1, this means the association is NOT stat. sig

Question 6

What happens when RR= <1?

Answer 6

-Usually for vax or protective factors
-When RR < 1 take the reciprocal and flip the exposure groups
EX if RR= 0.25 do 1/0.25=4 and then can do E- (not get vax) compared to E+ (did get vax)

the risk of influenza in individuals who did not get the flu vaccine (E-) its 4 times the risk of influenza in those who did get the vaccine (E+)

Question 7

What is odds ration (OR) and how do you calculate it?

Answer 7

-Numerator is NOT part of the denominator
-OR are interpreted nearly the same as RR but more advantageous than RR bc they can be used in studies where RR may not be applicable
-Odd of disease in exposed group/ odds of disease in non-exposed group
or
-AD/BC

Question 8

How do we interpret OR?

Answer 8

-When we are looking at disease (ex cohort studies) if OR=2.6 the odds of disease in the exposed group is 2.6 times the odds of the disease in the non-exposed group

-When we are looking at exposure (case-control studies) if OR=3 the odds of exposure in the diseased group is 3 times the odds of the exposure in the non-diseased group

Question 9

What happens when the OR is <1?

Answer 9

-Same as in RR we can take the OR and divide it by 1
ex1/0.4= 2.5
This means the odds of measles for individuals not vaccinated against measles is 2.5 times the odds of measles amongst individuals vaccinated against measles

Question 10

How does rare disease change for RR and OD?

Answer 10

-With rare diseases (<5%) the OR closely approximates the RR
-so RR will be similar to OR with rare diseases so formula similar to RR
-RR ~ (a/b)/(c/d) ~ OR

Question 11

What is incidence rate ratio (IRR)?

Answer 11

-Different than RR or OR bc now we have a time component
-Measuring rate: measure of speed at which the disease develops in a population
IRR= incidence rate of disease in exposed group / incidence rate of disease in non-exposed group
or
(a/b) / (c/d)

ex IRR= 2 the incidence rate of disease in the exposed group is 2 times the incidence rate in the non-exposed group

Question 12

How can we interpret measures of association?

Answer 12

large range of 0-infinity
-RR, OR, IRR = 1 NO ASSOCIATION
-RR, OR, IRR, > 1: exposure is positively associated with the outcome (possible causal)
-RR, OR, IRR < 1: exposure is negatively associated with the outcome (possibility protective/sparing)

*possibly bc association doesn’t equal causation

Question 13

What are the 2 types of measures of effect and how can they be broken down?

Answer 13

In the exposed group (how much of the total risk of disease is actually due to the exposure of interest)
1. Risk or Rate difference (RD)
2. Attributable proportion exposed (AP)

In populations
1. Population attributable risk (PAR)
2. Population Attributable Fraction (PAF)

Question 14

What are measures of effect?

Answer 14

-Measures of association are useful for investigating causation BUT are NOT as helpful for informing practice (ex clinical or public health practice)
-ex could have high RR but if the exposure is rare it won’t contribute much to disease in population (high individual but low population)
-Helpful to know how much disease could be eliminated by removing a certain exposure

Question 15

What is the baseline/ background risk?

Answer 15

-Baseline risk of disease= risk of disease in the non-exposed group
-Ex lunch cancer and smoking vs non smokers, not all lung cancer is due to smoking so the indigence in the non-smoking group is the background or baseline risk

Question 16

What is the risk or rate difference (RD)?

Answer 16

-The difference risk or rate of the outcome on the exposed and non-exposed groups

Risk difference
[a/(a+b)] - [c/c+d)]

Rate difference
(a/b)-(c/d)

Question 17

How do you interpret RD?

Answer 17

RD= 0: exposure has no effect- the risk/rate of disease is same in E+ and E- (null hypothesis)

RD>0: exposure is positively associated with the disease

RD<0: exposure is negatively associated with the disease (possibly protective ie vax again)

Question 18

How do the null values differ with measures of association and measures of effect?

Answer 18

Association: null=1 and only 0 to infinity CIs= 1 than not stat sig
Effect: null= 0 and can have negative as long as confident interval doesn’t inlace null 0

Question 19

How do you interpret RD?

Answer 19

-Same calculation as RR except instead of dividing them you subtract bc we want the difference
EX [a/(a+b)] - [c/(c+d)] = 0.151 means that the EXCESS RISK of avian influenza in flocks with poor biosecurity that is due to having poor biosecurity is 0.15
-Another way to look at RD is: the risk of avian influenza is increased by 0.15 for flocks with poor biosecurity compared to those with good biosecurity

EX another written interpretation: the EXCESS RISK of lung cancer in smokers that is due to smoking is 0.64

Question 20

What does excess risk mean for Risk difference (RD)?

Answer 20

-rememeber RD is a measure of effect which is looking a the absolute difference in risk or rate b/w groups after accounting for baseline risk or rate
-so excess risk is that BEYOND the baseline

Question 21

What is the attributable proportion (exposed) (AP)?

Answer 21

-The proportion of risk or rate of the outcome in the exposed group that is attributable to the exposure

Risk data:
RD (risk difference) / [a/(a+b)]

Rate dta:
[(a/b-c/d)] / (a/b)

Question 22

How do you interpret attributable proportion (AP)?

Answer 22

-The proportion of disease in the exposed group that is due to the exposure
-So its the proportion of disease in the exposed group that we could prevent if the exposure was eliminated
EX: AP= 0.918 means that 91.8% of Avian influenza in flocks with poor biosecurity IS DUE TO having poor biosecurity
EX AP= 95.5% Means that 95.5% of lung cancer cases in smokers was due to them smoking

Question 23

How are these concepts different for measures of effect in populations rather than exposed group?

Answer 23

-Now measure apply to whole group NOT just the exposed
-Important for public health for distributing scares resources and important for determining heat exposures most important to reduce or eliminate I order to enhance general population health

-Very similar to effect (RD and AP) but use incidence of disease in total population not just exposed group

Question 24

What is the population attributable risk?

Answer 24

-Amount of disease in entire population that is due to the exposure
-Difference measure: prevalence in pop n - baseline level of disease
or
[(a+c)/n] - [c/(c+d)]

for rate data
[(a+c) / (b+d)] - (c/d)]

Question 25

How do you interpret PAR (population attribute risk)?

Answer 25

PAR= 0: exposure has no effect (ie null hypothesis)
PAR > 0: exposure is positively associated with the disease (possibly causal)
PAR<0: exposure is negatively associated with the disease (possibility protective)

*null value is 0 bc its a difference not a ratio

EX PAR= 0.075 so the EXCESS RISK of avian influenza in ALL flocks that is due to poor biosecurity is 0.075

EX PAR= 0.096 So the excess risk of lung cancer in the POPULATION that is due to smoking is 0.096

Question 26

What is the population attributable fraction (PAF)?

Answer 26

-Proportion of disease in entire population that is due to the exposure
PAR / [(a+c)/n]
or
[(a+c)/n] - [c/(c+d)] / [(a+c)/n]

Rate data
[[(a1+a0)/(t1+t0)] -(a0/t0)] / [(a1+a0)/(t1+t0)]

EX PAF= 84.8 say that 85% of avian influenza in all flocks is due to poor biosecurity

EX PAF= 76.8% say 76.8% of lung cancer in the population is due to smoking or approximently 77% of lung cancer cases in the population could be avoided if smoking was eliminated

Question 27

How can we interpret measures of effect values?

Answer 27

• 0: exposure has no effect- the risk or rate of disease is the same in E+ and E- groups (or entire population) ie the null
• > 0: exposure is positively associated with the disease
  -<0: exposure is =negatively associated with the disease (possibly protective)