Chapter 2 - Determination of Association Strength Between A - 2012 - Practical B

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Chapter 2

Determination of Association
Strength between an Exposure
Factor and an Event in
Observational Studies

The aim in intervention studies is to demonstrate the difference  generally


as a favorable outcome  between two groups. Alternatively, the goal in
observational studies is to measure the odds or the risk for the occurrence of
an event between two groups. Based on the observational study type, two
different approaches are possible.

2.1 CASECONTROL STUDIES


Odds ratio (OR) is an index for association strength determination between
an exposure factor and an event. In observational studies, it expresses the
ratio between the odds for the occurrence of an event in a group exposed to
a factor and the odds for the occurrence of the same event in a group
exposed to a different factor (or not exposed). OR may be used in studies of
epidemiological interest or in therapeutic observational studies. OR can also
derive the number needed to harm (discussed later).

2.1.1 Odds Ratio


Odds Ratio for Studies of Epidemiological Interest
As an example, a population of 100 individuals is divided into a group with lung
cancer (case) and a group without lung cancer (control), with the aim of measur-
ing the odds for the occurrence of lung cancer related to smoke exposure. Both
groups are divided into two subgroupssmokers and nonsmokersand retro-
spectively followed for up to 25 years (Figure 2.1 and Table 2.1).
Based on the results, we can infer the following:
G There is 4:1 odds of smokers presenting lung cancer (a/b).
G There is 1:4 odds of nonsmokers presenting lung cancer (c/d).

M. Suchmacher & M. Geller: Practical Biostatistics. DOI: 10.1016/B978-0-12-415794-1.00002-1


© 2012 Elsevier Inc. All rights reserved. 19
20 PART | II Observational Studies

Time

Investigation direction

Smokers (40)

Group with lung cancer


(50)

Non-smokers (10)

Initial
Smokers (10) population

Group without lung cancer


(50)

Non-smokers (40)

FIGURE 2.1 Schematic representation of an initial population of healthy and lung cancer
patients, under a casecontrol study type.

TABLE 2.1 Study Results for Odds Ratio Calculation

Results
Case Control
Smokers 40 (a) 10 (b)
Nonsmokers 10 (c) 40 (d )

Establishing the Odds Ratio

a=b
OR 5
c=d
40=10
OR 5 5 16
10=40

This result means that the odds of lung cancer occurrence is 16:1 for
smokers in relation to nonsmokers.
Chapter | 2 Determination of Association Strength 21

Odds Ratio for Therapeutic Studies


As an example, a population of 120 perimenopaused women is divided into
a group with perimenopausal symptoms (case) and a group without perimen-
opausal symptoms (control), with the aim of measuring the odds of the
occurrence of perimenopausal symptoms relative to regular ingestion of soy
isoflavones. The groups are each divided into two subgroups: subgroup A,
comprising women who regularly ingest soy isoflavones, and subgroup B,
comprising women who do not ingest soy isoflavones. Both groups are retro-
spectively followed for up to 10 years (Figure 2.2 and Table 2.2).

Time

Investigation direction

Subgroup A (20)

Group of symptomatic
women (70)

Subgroup B (50)

Initial
Subgroup A (30) population

Group of asymptomatic
women (50)

Subgroup B (20)

FIGURE 2.2 Schematic representation of a population of perimenopaused women, under a


casecontrol study type.

TABLE 2.2 Study Results for Odds Ratio Calculation

Results
Case Control
Subgroup A 20 (a) 30 (b)
Subgroup B 50 (c) 20 (d )
22 PART | II Observational Studies

Based on the results, we can infer the following:


G There is 0.6:1 odds of women who regularly ingest soy isoflavones pre-
senting perimenopausal symptoms (a/b).
G There is 2.5:1 odds of women who do not ingest soy isoflavones present-
ing perimenopausal symptoms (c/d).

Establishing the Odds Ratio

a=b
OR 5
c=d
20=30
OR 5 5 0:26
50=20

This result means that the odds of the occurrence of perimenopausal symp-
toms is 0.2:1 for women who regularly ingest soy isoflavones in relation to
women who do not ingest soy isoflavones.

2.1.2 Number Needed to Harm


Number needed to harm (NNH) corresponds to the number of individuals
who must be treated so that one individual presents an adverse reaction
accountable to the treatment. The main usefulness of NNH is to make the
OR data more practical to physicians and comprehensible for patients. Its
interpretation must be performed based on the physician’s practice experi-
ence and on NNHs established for other treatment modalities related to the
case. For example, a population of 180 individuals with recently treated lung
tuberculosis is divided into a group with drug-induced hepatitis (case) and a
group without drug-induced hepatitis (control), with the aim of measuring
the odds of the occurrence of isoniazid-related hepatitis. The groups are
divided into two subgroups: regimen A, treated with isoniazid, and regimen
B, not treated with isoniazid. Both groups are retrospectively followed until
the beginning of tuberculostatic regimens (Figure 2.3 and Table 2.3).
a=b
OR 5
c=d
65=35
OR 5 5 2:4
35=45

NNH is calculated using the following formula:


1 2 ½PEER 3 ð1 2 ORÞ
ð1 2 PEERÞ 3 PEER 3 ð1 2 ORÞ
where PEER is the patient expected event rate.
Chapter | 2 Determination of Association Strength 23

Time

Investigation direction

Regimen A (65)

Group with drug-induced


hepatitis (100)

Regimen B (35)

Initial
Regimen A (35) population

Group without drug-induced


hepatitis (80)

Regimen B (45)

FIGURE 2.3 Schematic representation of a population of recently treated tuberculosis patients,


under a casecontrol study type.

TABLE 2.3 Study Results for Odds Ratio Calculation

Results
Regimen A 65 (a) 35 (b)
Regimen B 35 (c) 45 (d )

In this example, the chosen PEER corresponds to the proportion of regimen


B individuals belonging to the drug-induced hepatitis group: 30% (or 0.3).
1 2 ½0:3 3 ð1 2 2:4Þ
55
ð1 2 0:3Þ 3 0:3 3 ð1 2 2:4Þ
This result means that it would be necessary to treat five patients with
lung tuberculosis for one to present isoniazid drug-induced hepatitis.
Alternatively, it is possible to consult an NNH table (Tables 2.4 and 2.5).

2.1.3 Summary
Casecontrol studies allow group rather than individual-to-individual expo-
sure assessment because they afford less control over study conditions. For
this reason, this type of study does not allow for risk but, rather, odds status
24 PART | II Observational Studies

TABLE 2.4 NNH Table for Odds Ratio Calculation , 1.0

For OR , 1.0
0.9 0.8 0.7 0.6 0.5 0.4 0.3
PEER 0.05 209 104 69 52 41 34 29
0.10 110 54 36 27 21 18 15
0.20 61 30 20 14 11 10 8
0.30 46 22 14 10 8 7 5
0.40 40 19 12 9 7 6 4
0.50 83 18 11 8 6 5 4
0.70 44 10 13 9 6 5 4
0.90 101 46 27 18 12 9 4

TABLE 2.5 NNH Table for Odds Ratio Calculation . 1.0

For OR . 1.0
1.1 1.25 1.5 1.75 2 2.25 2.5
PEER 0.05 212 86 44 30 28 18 16
0.10 113 46 24 16 13 10 9
0.20 64 27 14 10 8 7 6
0.30 50 21 11 8 7 6 5
0.40 44 19 10 8 6 6 5
0.50 42 18 10 8 6 6 5
0.70 51 23 13 10 9 8 7
0.90 121 55 33 25 22 19 18

correlations only. This also explains why OR has limited usefulness in inter-
vention cohort studies.
Determination of OR values that suggest a significant relationship
between the exposure factor and the event is empirically based. Normally,
the following factors are taken into consideration:
G Influence of unknown variables and potential confounders: Casecontrol
studies are more prone to the occurrence of unknown variables and
Chapter | 2 Determination of Association Strength 25

confounders than are cohort studies. Therefore, there is a higher “toler-


ance” for the assumption of more elevated OR values.
G Event severity: The more severe the event, the lesser the “tolerance” for
the assumption of more elevated OR values.

2.2 COHORT STUDIES


Relative risk (RR) is an index for association strength determination between
an exposure factor and an event. It is defined as the ratio between the risk
for the occurrence of an event in a group exposed to a factor and the risk for
the occurrence of the same event in a group exposed to a different factor (or
not exposed). RR may be used in studies of epidemiological interest or in
therapeutic observational studies. By analogy with OR, RR can also derive
the NNH.

2.2.1 Relative Risk


Relative Risk for Studies of Epidemiological Interest
As an example, a population of 100 individuals is divided into an exposed
group (smokers) and a nonexposed group (nonsmokers), with the aim of
measuring the risk for the occurrence of lung cancer related to smoke expo-
sure. Both groups are prospectively followed for 15 years and then divided
into two subgroups each  individuals with lung cancer and individuals
without lung cancer (Figure 2.4 and Table 2.6).
Based on the results, we can infer the following:
G There is an 80% risk of smoke-exposed individuals presenting lung can-
cer [a/(a 1 b)].
G There is a 40% risk of non-smoke-exposed individuals presenting lung
cancer [c/(c 1 d)].

Establishing the Relation Between the Risks

½a=ða 1 bÞ
RR 5
½c=ðc 1 dÞ
½40=ð40 1 10Þ
RR 5 52
½20=ð20 1 30Þ

This result means that the risk of lung cancer is twice as high among
smoke-exposed individuals in relation to nonexposed individuals.

Relative Risk for Therapeutic Studies


As an example, a population of 120 perimenopaused women is divided into
group A (women who regularly ingest soy isoflavones) and group B (women
26 PART | II Observational Studies

Time

Investigation direction

Individuals with lung


cancer (40)

Exposed individuals (50)

Individuals without lung


cancer (10)

Initial
population
Individuals with lung
cancer (20)
Non-exposed individuals
(50)

Individuals without lung


cancer (30)

FIGURE 2.4 Schematic representation of a population of smoking-exposed and -nonexposed


individuals, under a cohort study type.

TABLE 2.6 Study Results for Risk Calculation

Results
With Lung Cancer Without Lung Cancer Total
Exposed 40 (a) 10 (b) 50
Nonexposed 20 (c) 30 (d ) 50

who do not ingest soy isoflavones), with the aim of measuring the risk for
the occurrence of perimenopausal symptoms relative to regular ingestion of
soy isoflavones. Both groups are prospectively followed for 3 years and then
divided into two subgroups each  symptomatic women and asymptomatic
women (Figure 2.5 and Table 2.7).
Based on the results, we can infer the following:
G There is a 28% risk of women who regularly ingest soy isoflavones pre-
senting perimenopausal symptoms [a/(a 1 b)].
G There is a 60% risk of women who do not ingest soy isoflavones present-
ing perimenopausal symptoms [c/(c 1 d)].
Chapter | 2 Determination of Association Strength 27

Time

Investigation direction

Symptomatic women
(20)

Group A (70)

Asymptomatic women
(50)
Initial
population
Symptomatic women
(30)

Group B (50)

Asymptomatic women
(20)

FIGURE 2.5 Schematic representation of a population of perimenopaused women, under a


cohort study type.

TABLE 2.7 Study Results for RR Calculation

Results
Symptomatic Asymptomatic Total
Group A 20 (a) 50 (b) 70
Group B 30 (c) 20 (d ) 50

Establishing the Relation Between the Risks

½a=ða 1 bÞ
RR 5
½c=ðc 1 dÞ
½20=ð20 1 50Þ
RR 5 D0:5
½30=ð30 1 20Þ

This result means that the risk of the occurrence of perimenopausal


symptoms is 0.5 for women who regularly ingest soy isoflavones in relation
to women who do not ingest soy isoflavones.
28 PART | II Observational Studies

2.2.2 Number Needed to Harm


NNH corresponds to the number of individuals who must be treated so that
one of them presents an adverse reaction accountable to the treatment.
The main usefulness of NNH is to make RR data more practical to physi-
cians and comprehensible for patients. Its interpretation must be performed
based on physician’s own practice experience and on NNHs established
for other treatment modalities related to the case. For example, a population
of 180 individuals with diagnosis of lung tuberculosis is divided into two
groups treated with different regimens, with the aim of measuring the risk
for the occurrence of isoniazid drug-induced hepatitis: regimen A, isoniazid
included, and regimen B, isoniazid not included. Both groups are prospec-
tively followed for 1 year and then divided into two subgroups each 
individuals with drug-induced hepatitis and individuals without drug-induced
hepatitis (Figure 2.6 and Table 2.8).
NNH is calculated using the following formula:
½a=ða 1 bÞ 2 ½c=ðc 1 dÞ
½65=ð65 1 35Þ 2 ½35=ð35 1 45Þ 5 0:22.22
This result means that it is necessary to treat 22 patients with lung tuber-
culosis so that one presents isoniazid drug-induced hepatitis.

Time

Investigation direction

Individuals with drug-


induced hepatitis (65)

Regimen A (100)

Individuals without drug-


induced hepatitis (35)

Initial
population
Individuals with drug-
induced hepatitis (35)

Regimen B (80)

Individuals without drug-


induced hepatitis (45)

FIGURE 2.6 Schematic representation of a population of tuberculosis patients, under a cohort


study type.
Chapter | 2 Determination of Association Strength 29

TABLE 2.8 Study Results for NNH Calculation

Results
With Hepatitis Without Hepatitis Total
Regimen A 65 (a) 35 (b) 100
Regimen B 35 (c) 45 (d ) 80

2.2.3 Summary
Cohort studies allow individual-to-individual rather than group exposure
assessment because they afford more control over study conditions. For this
reason, this type of study allows risk status correlations. An intervention
study can also derive a cohort study if its adverse reaction results (often the
endpoint for which RR is more useful in this type of study) are submitted to
this kind of approach.
Determination of RR values that suggest a significant relationship
between the exposure factor and the event is empirically based. Normally,
two factors are taken into consideration:
G Influence of unknown variables and potential confounders: Cohort studies
are less prone to the occurrence of unknown variables and confounders
than are casecontrol studies. Therefore, there is a smaller tolerance for
the assumption of more elevated RR values.
G Event severity: The more severe the event, the lesser the tolerance for the
assumption of more elevated RR values.

You might also like