Special Communication

Whats the Relative Risk?

A Method of Correcting the Odds Ratio
in Cohort Studies of Common Outcomes
Jun Zhang, MB, PhD; Kai F. Yu, PhD
Logistic regression is used frequently in cohort studies and clinical trials. When
the incidence of an outcome of interest is common in the study population
(10%), the adjusted odds ratio derived from the logistic regression can no
longer approximate the risk ratio. The more frequent the outcome, the more the
odds ratio overestimates the risk ratio when it is more than 1 or underestimates
it when it is less than 1. We propose a simple method to approximate a risk ra-
tio from the adjusted odds ratio and derive an estimate of an association or
treatment effect that better represents the true relative risk.
JAMA. 1998;280:1690-1691
RELATIVE RISK has become one of
the standard measures in biomedical re-
search. It usually means the multiple of
risk of the outcome in one group com-
pared with another group and is ex-
pressed as the risk ratio in cohort stud-
ies andclinical trials. Whentheriskratio
cannot be obtained directly (such as in a
case-control study), the odds ratio is cal-
culated and often interpreted as if it
were the risk ratio. Subsequently, the
term relative risk commonly refers to
either the risk ratio or the odds ratio.
However, only under certain conditions
does the odds ratio approximate the risk
ratio. The Figure shows that when the
incidenceof anoutcomeof interest inthe
studypopulationis low(10%), theodds
ratio is close to the risk ratio. However,
themorefrequent theoutcomebecomes,
the more the odds ratio will overesti-
mate the risk ratio when it is more than
1 or underestimate the riskratio whenit
is less than 1.
Logistic regression is a widely used
technique to adjust for confounders, not
onlyincase-control studies but also inco-
hort studies.
1
However, logistic regres-
sionyieldsanoddsratioratherthanarisk
ratio, even in a cohort study. Under the
same rule, when the outcome of interest
is common in the study population
(though it could be rare in the general
population), the adjustedodds ratio from
the logistic regression may exaggerate a
riskassociationoratreatment effect. For
instance, a previous study assessed the
performanceof neonatal unitsinHospital
Aand Hospital Bby comparing neonatal
mortality in very low birthweight neo-
nates between these 2 hospitals.
2
At first
glance, Hospital Ahad a lower mortality
rate than Hospital B (18% vs 24%, risk
ratio, 18%:24%[0.75]). However, afterad-
justing for clinical variables and initial
diseaseseverityusinglogisticregression,
the adjusted odds ratio of Hospital A vs
Hospital B was 3.27 (95% confidence in-
terval, 1.35-7.92). Can one therefore con-
clude that neonates with very lowbirth-
weight in Hospital Ahad 3 times the risk
of death than those in Hospital B? Prob-
ably not, because the outcome (neonatal
death) was common in this study popula-
tion. To provide a measure that more ac-
curately represents the concept of rela-
tive risk, correctionof the odds ratio may
be desirable.
A modified logistic regression with
special macro functions has been devel-
From the Division of Epidemiology, Statistics and
Prevention Research, National Institute of Child Health
and Human Development, National Institutes of Health,
Bethesda, Md.
Correspondingauthor: JunZhang, MB, PhD, Division
of Epidemiology, Statistics and Prevention Research,
National Institute of Child Health and Human Develop-
ment, Bldg 6100, Room 7B03, Bethesda, MD 20892-
7510 (e-mail: [email protected]).
Reprints not available from the authors.
10
3
5
2
1
0.7
0.5
0.3
0.1
1 5 10 20 30 40 50 60 70
Incidence Among the Nonexposed, %
O
d
d
s

R
a
t
i
o
RR = 0.75
RR = 1.25
RR = 1.5
RR = 1.75
RR = 2.0
RR = 2.5
RR = 3.0
RR = 0.5
RR = 0.3
The relationship between risk ratio (RR) and odds ratio by incidence of the outcome.
1690 JAMA, November 18, 1998Vol 280, No. 19 Whats the Relative Risk?Zhang & Yu
1998 American Medical Association. All rights reserved.
Downloaded From: http://jama.jamanetwork.com/ on 12/05/2013
oped to address this issue.
3
However, it
is mathematically complex and uses a
General Linear Interactive Modeling
System (Numerical Algorithms Group,
Oxford, England). Consequently, this
method is rarely used. Another alterna-
tive is to use the Mantel-Haenszel
method,
4
which can adjust for 1 or 2 con-
founders and still provide a risk ratio in
a cohort study. However, this method
becomes inefficient when several fac-
tors, especiallycontinuousvariables, are
being adjusted for simultaneously. We
herein propose an easy approximation
withasimpleformulathat canbeapplied
not only in binary analysis
5
but also in
multivariate analysis.
In a cohort study, P
0
indicates the in-
cidence of the outcome of interest in the
nonexposed group and P
1
in the exposed
group; OR, odds ratio; and RR, risk
ratio: OR=(P
1
/1 P
1
)/(P
0
/1 P
0
); thus,
(P
1
/P
0
) =OR/[(1 P
0
) +(P
0
OR)]. Since
RR=P
1
/P
0
, the corrected
RR =
OR
(1 P
0
) + (P
o
OR)
.
We can use this formula to correct the
adjusted odds ratio obtained fromlogis-
tic regression and derive an estimate of
an association or treatment effect that
better represents the true relative risk.
It can also be used to correct the lower
and upper limits of the confidence inter-
val byapplyingthis formula to the lower
and upper confidence limits of the ad-
justed odds ratio. In the above example,
after the odds ratio is corrected (where
OR=3.27 and P
0
=0.24), the risk ratio
becomes 2.12 (95% confidence interval,
1.25-2.98), ie, very low birthweight
neonates in Hospital A had twice the
risk of neonatal death than those in
Hospital B.
To examine the validity of this correc-
tion method in various scenarios, we
simulated a series of hypothetical co-
horts based on predetermined risk ra-
tios (called true RR). Each cohort con-
sists of 1000 subjects with 1 binary out-
come (0,1), 1 exposure variable (0,1), and
2 confounders. Both confounders have 3
levels (1,2,3). The true risk ratio is kept
constant across strata of the confound-
ers. As expected, with an increase in in-
cidence of outcome and risk ratio, the
discrepancybetweenriskratio andodds
ratio increases (Table). The corrected
risk ratio, which is calculated based on
the odds ratio from logistic regression
after having adjusted for the confound-
ers, is very close to the true risk ratio.
This procedure can be applied to both
unmatched and matched cohort studies.
It can further be used in cross-sectional
studies, in which the prevalence ratio
rather than the risk ratio will be gener-
ated. It enables us to obtain a corrected
prevalence ratio very close to the one
obtained from a complex statistical
model
6
(data not shown).
Due to the differences in underlying
assumptions between Mantel-Haenszel
riskratioandlogisticregressionoddsra-
tio, some discrepancy between the Man-
tel-Haenszel riskratioandthecorrected
risk ratio is expected (detailed discus-
sion of which is beyond the scope of this
work). More importantly, the validity of
thecorrectedriskratiorelies entirelyon
the appropriateness of logistic regres-
sionmodel, ie, onlywhenlogistic regres-
sionyields anappropriateodds ratiowill
the correction procedure provide a bet-
ter estimate. Therefore, in a cohort
study, whenever feasible, the Mantel-
Haenszel estimate should be used.
In summary, in a cohort study, if the
incidence of outcome is more than 10%
andtheodds ratiois morethan2.5or less
than0.5, correctionof theodds ratiomay
be desirable to more appropriately in-
terpret the magnitude of an association.
References
1. Hosmer DW, LemeshowS. Applied Logistic Re-
gression. New York, NY: John Wiley & Sons Inc;
1989.
2. Tarnow-Mordi W, OgstonS, WilkinsonAR, et al.
Predictingdeathfrominitial diseaseseverityinvery
lowbirthweight infants: amethodforcomparingthe
performance of neonatal units. BMJ. 1990;300:1611-
1614.
3. Wacholder S. Binomial regression in GLIM: es-
timating risk ratios and risk differences. AmJ Epi-
demiol. 1986;123:174-184.
4. Mantel N, Haenszel W. Statistical aspects of the
analysis of data from retrospective studies of dis-
ease. J Natl Cancer Inst. 1959;22:719-748.
5. Sinclair JC, Bracken MB. Clinically useful mea-
sures of effect in binary analyses of randomized tri-
als. J Clin Epidemiol. 1994;47:881-889.
6. Lee J. Odds ratio or relative risk for cross-sec-
tional data? Int J Epidemiol. 1994;23:201-202.
Correcting the Odds Ratio in Cohort Studies of Common Outcomes in a Series of Simulated Cohorts*
True
RR
Crude RR
(95% CI)
Crude OR
(95% CI)
Logistic OR
(95% CI)
M-H RR
(95% CI)
Corrected RR
(95% CI) P0, %
7.4 6.5 (4.4-9.7) 9.4 (5.9-15.2) 14.1 (7.8-27.5) 8.0 (5.2-12.2) 8.3 (5.4-11.4) 5
4.2 3.6 (2.8-4.8) 5.6 (3.9-8.1) 8.7 (5.5-14.3) 4.3 (3.2-5.8) 4.6 (3.6-5.6) 12
3.0 2.9 (2.5-3.2) 25.0 (17.2-35.7) 27.4 (17.2-45.8) 3.0 (2.6-3.4) 2.9 (2.8-3.0) 32
2.0 0.93 (0.73-1.2) 0.90 (0.66-1.2) 4.5 (2.7-7.8) 2.0 (1.6-2.5) 2.3 (1.8-2.7) 27
0.37 0.34 (0.27-0.42) 0.23 (0.17-0.32) 0.25 (0.17-0.37) 0.37 (0.28-0.48) 0.36 (0.25-0.49) 40
0.14 0.13 (0.09-0.17) 0.08 (0.06-0.11) 0.09 (0.05-0.14) 0.14 (0.10-0.20) 0.14 (0.08-0.21) 40
*RR indicates risk ratio; CI, condence interval; OR, odds ratio; logistic OR, odds ratio from logistic regression;
M-H RR, risk ratio from Mantel-Haenszel estimate; P0, incidence of outcome of interest in the nonexposed group;
and corrected RR, risk ratio corrected by the above formula using logistic OR.
Due to the sample sizes used in the simulation and the need to round numbers to integers, the M-H RR differs
from the true RR.
JAMA, November 18, 1998Vol 280, No. 19 Whats the Relative Risk?Zhang & Yu 1691
1998 American Medical Association. All rights reserved.
Downloaded From: http://jama.jamanetwork.com/ on 12/05/2013