Gender Analysis
Week 3
1
Objectives
1. Understand how to practically conduct gender
analysis
2. Learn and apply measures of frequency and
association
2
100
75
50
25
Measures of Disease
Frequency and Association
3
Measures of Disease
Frequency
Need measurements that describe the existence or
development of disease
Number of cases is useful
If you want to assess if there's a problem, need more
information
• Excess vs expected
4
Measures of Disease
Frequency
Need:
frequency of disease expressed
per unit size of the population
in a specified time period
number of cases
in specified
time period
population
Various measures of frequency, most fall into either
prevalence and incidence
5
Prevalence (P)
• Refers to those with the disease at a given point in time (snapshot)
• =(# of existing cases)/(# in total population) at a point in time
• Interpretation: On DATE, X% of POPULATION in LOCATION had OUTCOME.
• Range of values: 0-1, or 0-100%
• Seen as the probability that an individual will have a disease at a given point in time
Cumulative incidence (CI, “incidence”)
• Refers to those who develop a disease among those who are at baseline free of disease and at risk, over
a given time period
• =(# of new cases)/(# at risk in population) in a specified time period
• Interpretation: Between DATE1 and DATE2, X% of POPULATION at risk in LOCATION developed
OUTCOME.
• Range of values: 0-1, or 0-100%
• Seen as the probability that an individual will have a disease within a specified time period
• Requires everyone followed for the same specified time period, else use incidence rate (IR)
6
Loss to follow-up
• Refers to subjects with missing portion of follow-up
• May still be at risk and not developed disease
• May have developed disease
Solution: incidence rate (IR)
• Rate new cases of disease occur in a population at risk for the disease
• =(# of new cases developing over study period)/(total person time)
• Range of values: 0-infinity
• Difficult to intuitively understand
7
Measures of Association
After measures of disease frequency in different groups, goal:
• Summarize into single measure
• Magnitude (strength) of the relationship (association)
between the exposure/risk factors and outcome
1. Summarize the data in table
2. Calculate measures of disease frequency for each
of the exposure groups
3. Combine and calculate measures of association
8
1. Summarize the Data
Disease
(outcome)
Yes No
Yes a b a+b
Exposure
No c d c+d
a+c b+d N
a=# exposed and have the disease
b=# exposed and do not have the disease
c=# not exposed and have the disease
d=# both not exposed and do not have the disease
9
Brain Cancer
Yes No
Yes a b a+b
Cell Phone
Use
No c d c+d
a+c b+d N
10
Brain Cancer
Yes No
Yes a - PTexposed
Cell Phone
Use
No c - PTunexposed
a+c - PTtotal
11
2. Calculate Measures of
Disease Frequency
Cohort study with individuals in the denominator (count data)
• CIexposed=(# exposed cases)/(# exposed)=a/a+b
• CIunexposed=(# unexposed cases)/(# unexposed)=c/c+d
Cohort study with person-time denominator (person-time data)
• IRexposed=(# exposed cases)/(exposed person-time)=a/PTexposed
• IRunexposed=(# unexposed cases)/(unexposed person-time)=c/
PTunxposed
12
3. Combine and calculate
measures of association
Various measures of association
• Divide measures of frequency
• Ratio measures (relative scale)
• Subtract the two values
• Difference measures (absolute scale)
Will depend upon the study design
13
Ratio Measures
Relative Risk: Generic term for risk ratio, rate ratio, or odds ratio
Risk ratio: cumulative incidence ratio
• CIexposed/CIunexposed in cohort study with count data
Rate ratio: incidence rate ratio
• IRexposed/IRunexposed in cohort study with person-time data
Odds ratio (OR)
• OR=ad/bc in case-control study
• Cannot calculate incidence rates, since there is no rate of development of the
disease
• Individuals are selected because they either do or do not have the disease
14
Ratio Measures
Risk Ratio~Rate Ratio~Odds Ratio, in meaning,
calculated differently
Range of values 0 to infinity, null value = no
association: RR = 1.0
• Relative risk >1.0 positive association: exposure is
associated with an increased risk of disease
• Relative risk <1.0 inverse association: exposure is
associated with a decreased risk of disease
15
To interpret any of these measures in a succinct sentence:
• The EXPOSED have X times the RISK/RATE/ODDS of having the
OUTCOME compared to the UNEXPOSED.
You can also interpret a relative risk using a percentage.
• When the relative risk is <1 and therefore protective: The EXPOSED
have [(1.00-X)*100]% DECREASED RISK/RATE/ODDS of having the
OUTCOME compared to the UNEXPOSED.
• When the relative risk is >1 and therefore harmful: The EXPOSED have
[(1.00-X)*100]% INCREASED RISK/RATE/ODDS of having the
OUTCOME compared to the UNEXPOSED.
• When the relative risk is ≥2, it is easier to understand the following: The
EXPOSED have (X*100)% the RISK/RATE/ODDS of having the
OUTCOME compared to the UNEXPOSED.
16
Example 1 (when the relative risk is <1): Individuals who wore
sunscreen on Crane Beach in July 2014 have 0.6 times the odds
(or 40% decreased odds) of developing a sunburn compared to
individual who didn’t wear any sunscreen.
Example 2 (when the relative risk is >1 but <2): Adults who ate 2
portions of red meat have 1.7 times the risk (or 70% increased
risk) of having a myocardial infarction compared to women you
ate 1 portion of red meat.
Example 3 (when the relative risk is ≥2): Women who used talc
powder have 2.2 times the risk (or 222% the risk) of having the
ovarian cancer compared to women you never used talc powder.
Always include as much information as you have available to you
(e.g., dates, exposure definitions, geographic location).
17
Risk Ratio Example
Lung Cancer
Yes No
Yes 36 39,964 40,000
Smoking
No 4 59,996 60,000
40 99,960 100,000
CIexposed=a/a+b=36/40,000=9/10,000
CIunexposed=c/c+d=4/60,000=0.7/10,000
Risk ratio=CI exposed/CIunexposed
=(36/40,000)/(4/60,000)
=13.5
Women who are smokers had 13.5 times the risk of lung cancer as
nonsmokers (or had a 1350% increased risk)
18
Rate Ratio Example
Breast Cancer
Yes No
35+ 49 - 26,757
Age at 1st person-years
birth 841,827
<35 782 -
person-years
831 -
IRexposed=a/PTexposed=49/26,757=183/100,000 person-years
IRunexposed=c/PTunxposed=782/841,827=93/100,000 person-years
Rate ratio=IRexposed/IRunexposed
=(49/26,757)/(782/841,827)
=1.97
Women who had their first birth after age 35 had 1.97 times the rate of breast
cancer as those who were <35 years of age at first birth (or had a 97%
increased rate).
19
Odds Ratio Example
Myocardial
Infarction
Yes No
Yes 23 304
Current OC
Use
No 133 2616
156 3120 3276
Odds ratio=OR
=ad/bc
=23 (2816)/133(304)
=1.6
Women who are current OC users had 1.6 times the odds of a myocardial
infarction as nonusers (or had a 60% increased odds)
20
Difference Measures
Risk difference: General term indicates amount of disease in the exposed group that can be
considered due to (attributable to) the exposure, by subtracting out the rate (risk) of disease in the
nunexposed group
Assumes causality has been established
Null is 0 (rather than 1 like in ratio measures)
Risk difference
• CIexposed-CIunexposed in cohort study with count data
Rate difference
• IRexposed-IRunexposed in cohort studies with person-time denominators
Can change to percent (“attributable risk percent”)
• (CIexposed-CIunexposed)/CIexposed
• (OR-1)/OR
21
Risk Difference Example
Lung Cancer
Yes No
Yes 36 39,964 40,000
Smoking
No 4 59,996 60,000
40 99,960 100,000
CIexposed=a/a+b=36/40,000=9/10,000
CIunexposed=c/c+d=4/60,000=0.7/10,000
Risk ratio=CI exposed-CIunexposed
=(36/40,000)-(4/60,000)
=8.3/10,000
Assuming smoking causes lung cancer, 8.3 cases of lung cancer per 10,000
smokers (exposed) is due to smoking, or could be eliminated among smokers if
22
smoking were eliminated.
Risk Difference Example
Lung Cancer
Yes No
Yes 36 39,964 40,000
Smoking
No 4 59,996 60,000
40 99,960 100,000
Attributable risk percent=(CIexposed-CIunexposed)/CIexposed
=[(36/40,000)-(4/60,000)]/(36/40,000)
=92.6%
If smoking causes lung cancer, 93% of the lung cancer among smokers is due to
smoking, or could be avoided among smokers if smoking were eliminated.
23
Measures of Frequency
and Association
Exercise
24
For each description below:
• Indicate which measure of disease frequency
(prevalence, cumulative incidence, or incidence
rate) best describes each of the following.
• Indicate the reason why.
25
Percentage of freshman girls who become pregnant
over the course of their high school years.
• Cumulative incidence: development, uniform
follow-up
26
The percentage of senior boys who are fathers at the
time of graduation.
• Prevalence: existing at a point in time
27
The proportion of women who experience depression
on the third postpartum day.
• Prevalence: existing at a point in time
28
The percentage of infants with congenital heart
defects at birth.
• Prevalence: existing at a point in time
29
The percentage of workers who were all followed for
10 years after beginning employment in a coal mine,
who developed lung cancer during that period.
• Cumulative incidence: development, uniform
follow-up
30
The number of first myocardial infarctions (heart
attacks) occurring in middle-aged men in 10,000
person-years of observation.
• Incidence rate: development, variable follow-up
31
The percentage of potential army recruits rejected at
their initial enrollment physical exam because of poor
vision.
• Prevalence: existing at a point in time
32
A study was conducted among 400 Canadian women
diagnosed with breast cancer.
The goal of the study was to assess whether the type
of hospital at which the woman was treated
(provincial versus teaching) was related to her
survival from the disease over a 10-year period.
The table below shows the outcome variable (the
number of women surviving as of March 1999), the
exposure variable (type of hospital [provincial or
teaching] at which the woman was treated), as well
as the extent of her disease at diagnosis.
33
Type of Hospital Where Treated
Provincial Hospital Teaching Hospital
Extent of
Breast N N N N
% %
Cancer treated survived treated survived
Local 240 80 144 30 30 21
Regional 30 10 9 40 40 20
Metastatic 30 10 3 30 30 6
34
What is the study design?
• Cohort study
• Exposure = type of hospital where treated
• Outcome = survival
35
Set up a 2x2 table with total breast cancer deaths as
the outcome of interest and hospital type (provincial
or teaching) as the exposure of interest.
Died
Hospital
Type Yes No
Provincial 144
156
300
Teaching 53
47
100
197
203
400
36
Calculate the cumulative incidence ratio (risk ratio) of
total breast cancer deaths in the provincial hospitals
relative to the teaching hospitals.
Died
Hospital
Type Yes
Yes No
(144/300)
Provincial 144No 156 300 IR= (53/100) =0.9
Teaching 53156 47 100
197 203 400
Interpret this value in words: Women treated in a
provincial hospital had 90% of the risk of dying (or 10%
less risk of dying) from breast cancer over 10 years as
those treated in a teaching hospital.
37
Calculate the attributable risk percent among the exposed
(those treated in provincial hospitals).
ARE %=
( R1 − R0 )
*100 =
(144 / 300 − 53 /100)
= −10%
R1 (144 / 300)
Died
Hospital
Type Yes No
Provincial 144 156 300
Teaching 53 47 100
197 203 400
Interpret this value in words: Assuming that being treated
at a provincial hospital is causally related to a reduction in
mortality, 10% of deaths from breast cancer among those
treated at the provincial hospital were prevented because
they were treated at the provincial hospital.
38
Using data provided in the original table, suggest an
alternative explanation as to why this reduction in mortality
could have been observed, besides that being treated at
a provincial hospital is causally related to greater survival.
• Confounding by extent of disease at diagnosis
• 80% of those treated at the provincial hospitals had
local disease at diagnosis
• As compared with only 30% at the teaching hospitals,
and local disease has a better survival rate than, for
example, metastatic disease
39
To evaluate possible risk factors for breast cancer, investigators
conducted a case-control study
Cases: 1,000 women with breast cancer who had previously given birth
using a tumor registry list from Massachusetts
Controls: 1,000 women without breast cancer who had previously given
birth were selected at random from voter registration lists
All subjects were interviewed on a variety of exposures including
reproductive characteristics, demographic information, and past medical
history
525 of those with breast cancer reported having an age at first birth of 35
years or younger
635 women without breast cancer reported having their age at first birth
as 35 years or younger
40
Construct a 2x2 table that summarizes these data on the
association between age at first birth of 35 years or younger
versus older than 35 years, and breast cancer.
Breast Cancer
Age at first
birth (years) Yes No
≤ 35 525 635 1160
> 35 475 365 840
1000 1000 2000
41
What measure of association would you use to
describe the relationship between age at first birth
and breast cancer?
• Odds ratio
42
Calculate that measure of association for this study.
Breast Cancer
Age at first
Yes No
birth (years)
≤ 35 525 635 1160 OR =
(525 * 365)
= 0.64
(475 * 635)
> 35 475No 365 840
635
1000 1000 2000
43
Interpret the meaning of this value in words.
• Women who have an early age at first birth have
0.64 times the odds (or 64% of the odds, or 36%
less odds) of developing breast cancer compared
with those who have a late age at first birth.
44
© Brittany M. Charlton 2016
45
MIT OpenCourseWare
http://ocw.mit.edu
WGS.151 Gender, Health, and Society
Spring 2016
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
