Measures of Association
Intermediate Epidemiology
Objectives
• To play “name that rate”
• To discuss case-control study designs
• To discuss and calculate measures of association
• To discuss the design and measures of association of a study
Name that Rate
• What’s the numerator
• What’s the denominator
• What is the type of rate
0.2
3.2
0.9
0.1
N=0N=2
N=84N=237N=58
AmericanIndian/
Alaska Native
WhiteNot Hispanic
BlackNot Hispanic
Hispanic Asian/PacificIslander
0
1
2
3
4
5
AIDS Rates per 100,000 Children <13 Years of Ageby Race/Ethnicity, Reported in 1998*, United States
Race/Ethnicity*US Rate=0.7/100,000 N=382
Rate
per
10
0,0
00
4.7
13.5
VI 29.6PR 44.3
<55 - 14.915+
AIDS Rates per100,000 PopulationReported in1998
Rate per 100,000
DE
MARICTNJ
MDDC
NH
26.320.3
23.4
189.131.9
15.012.9
3.5
11.1
13.88.0
7.9
36.5
16.9
2.6
10.8
2.6
4.87.1
21.8
2.5
7.3
4.0
15.1
8.1
3.3
4.314.8
12.0
47.9
10.4
0.9
6.18.2
8.5
6.2
14.5
20.3
2.0
12.8
20.1
6.6
14.7
7.8
4.7
3.9
17.3
1.2
VT3.4
0
10
20
30
40
50
60
70
Year of Report
American Indian/Alaska Native
Black, not Hispanic
Hispanic
Asian/Pacific Islander
White, not Hispanic
1985 1987 1989 1991 1993 1995 1997
Proportion of AIDS Cases, by Race/Ethnicity and Year of Report,1985-1998, United States
Perc
en
t of
Case
s
AIDS Cases in Adult/Adolescent Women by Race/Ethnicity per 100,000 Population, Reported in 1998,
United States
Race/Ethnicity
White, not Hispanic
Black, not Hispanic
Hispanic
Asian/Pacific Islander
American Indian/Alaska NativeTotal*
Cases
2,031
6,775
2,055
59
3010,998
Rate
2
50
17
1
410
*Includes 48 women whose race/ethnicity is unknown.
Born1993-March 1998 in 29 States‡, United Stateswho Received or whose Mothers Received any ZDV*
0
20
40
60
80
100
* Any ZDV=Prenatal, intrapartum, or neonatal receipt of Zidovudine to reduce perinatal HIV transmission
Quarter-Year of Birth
1993N=1313
1994N=1271
1995N=1315
1996N=1260
1997N=1291
1998N=281
‡ Includes 29 areas that have conducted pediatric HIV Surveillance since 1993; data reported through March 1999
Percent of Perinatally HIV Exposed or Infected ChildrenPerc
ent
Rece
ivin
g Z
idovudin
e
NortheastN=14,399
North CentralN=4,317
South N=19,474
WestN=8,121
Metropolitanarea 50,000-
499,999 population
Non-metropolitanarea
AIDS Cases by Region and Size of Place of ResidenceReported in 1998, United States
%
93.1 4.5 2.4
76.3 14.4 9.3
75.6 13.5 10.9
86.7 8.2 5.0
% %
Metropolitanarea 500,000population
71%
12%
13%
AIDS Cases N=48,269*
PopulationN=274,766,000
AIDS Cases Reported in 1998 and Estimated 1998 Population, by Race/Ethnicity, United States
White, not HispanicBlack, not Hispanic
Hispanic
Asian/Pacific IslanderAmerican Indian/ Alaska Native
33%
45%20%
*Includes 211 persons with unknown race/ethnicity
<1%1%
1%4%
Exposure Category N=15020-2913-19
N=2,01530-49N=7,683
50+N=1,150
Injection drug use 9
% % % %
19 34 19
Heterosexual contact 41 46 35 39
Transfusion recipient 2 1 1 3
Other/not identified* 48 34 30 39
Age at Diagnosis (in years)
*Includes patients whose medical record review is pending; patients who died, were lost to follow-up, or declined interview; and patients with other or undetermined modes of exposure
AIDS in Women, by Exposure Category andAge at Diagnosis, Reported in 1998, United States
Case-control study
• Persons are categorized by disease status and then compared for their exposure status
• The odds ratio is the measure of association (an estimate of relative risk)
• The assumption is that the cases and controls originate from the same hypothetical source cohort.
• When this assumption is not met, the results are susceptible to selection bias
• Traditional method is case-based case-control study
Nested case-control or case-cohort
• When cases are identified within a well defined cohort
• Also known as a hybrid or ambi-directional study
Alternatives for selection of controls and a case-control study
• Case-cohort - When the controls are randomly selected from a defined cohort at baseline.
• Nested case-control – based on incidence density sampling (matching cases and controls on duration of follow-up) or risk-set sampling (or comparison of cases with a subset of the cohort members as risk of being cases at the time when each case occurs)
Case-crossover
• Compares the exposure status of a case immediately before it’s occurrence with that of the same case at some other prior time
• Appropriate for the study of acute (brief) exposures that produce transient change in risk of an acute condition (eg, MI and episodes of anger or sexual activity)
Matching
• Performed to reduce confounding
• Can be individually matched or frequency matched
Advantages of Matching
• May be the only way to control some degrees of confounding (especially very strong confounding)
• Tends to increase the statistical power
• Logistically straight forward
Disadvantages of Matching• May be impossible to find a control• When matching is done, the association between the
matched variables and the outcome can no longer be assessed
• Also eliminates the ability to assess additive interaction between the matched variable and the exposure of interest
• Special statistical tests need to be used• By matching you remove the representativeness of your
controls• Once matching is done, it cannot be undone• No statistical power is gained if the matching variables are
weak confounders
Relative Risk for a disease exposure
CVD No CVDObesity 75 25 100No Obesity 25 75 100
100 100 200
RR = 75/100 = 3.00 25/100
C.I. (2.10 - 4.29)
Relative Risk for preventive intervention
Disease No DiseaseCounseling 25 75 100No Counseling 50 50 100
75 125 200
RR = 25/100 = .50 50/100
C.I. (.39-.79)