A substantial and confusing variation exists in handling of baseline
covariates in randomized controlled trials: a review of trials published
in leading medical journals
Peter C. Austina,b,c,*, Andrea Mancad, Merrick Zwarensteina,c,e, David N. Juurlinka,f,
Matthew B. Stanbrooka,f
aInstitute for Clinical Evaluative Sciences, Toronto, Ontario, Canada
bDalla Lana School of Public Health, University of Toronto, Toronto, Ontario, Canada
cDepartment of Health Policy, Management and Evaluation, University of Toronto, Toronto, Ontario, Canada
dCentre for Health Economics, The University of York, York, United Kingdom
eCentre for Health Services Sciences, Sunnybrook Hospital, Toronto, Ontario, Canada
fDepartment of Medicine, University of Toronto, Toronto, Ontario, Canada
Accepted 16 June 2009
Objective: Statisticians have criticized the use of significance testing to compare the distribution of baseline covariates between treat-
ment groups in randomized controlled trials (RCTs). Furthermore, some have advocated for the use of regression adjustment to estimate the
effect of treatment after adjusting for potential imbalances in prognostically important baseline covariates between treatment groups.
Study Design and Setting: We examined 114 RCTs published in the New England Journal of Medicine, the Journal of the American
Medical Association, The Lancet, and the British Medical Journal between January 1, 2007 and June 30, 2007.
Results: Significance testing was used to compare baseline characteristics between treatment arms in 38% of the studies. The practice
was very rare in British journals and more common in the U.S. journals. In 29% of the studies, the primary outcome was continuous,
whereas in 65% of the studies, the primary outcome was either dichotomous or time-to-event in nature. Adjustment for baseline covariates
was reported when estimating the treatment effect in 34% of the studies.
Conclusions: Our findings suggest the need for greater editorial consistency across journals in the reporting of RCTs. Furthermore,
there is a need for greater debate about the relative merits of unadjusted vs. adjusted estimates of treatment effect. ? 2010 Elsevier
Inc. All rights reserved.
Keywords: Baseline covariates; Randomized controlled trial; Significance testing; Analysis of covariance; Regression adjustment; CONSORT statement;
In randomized controlled trials (RCTs), randomization
ensures that, on average, the distribution of baseline
covariates will be similar between the different treatment
arms. However, in any given RCT, there is likely to be
residual differences in baseline characteristics between the
treatment arms . Adherence to the CONSORT guidelines
on the reporting of RCTs requires that baseline demographic
and clinical characteristics be described in each treatment
group . With few exceptions, the statistical literature is
uniform in its agreement on the inappropriateness of using
hypothesis testing to compare the distribution of baseline
covariates between treated and untreated subjects in RCTs
[1,3e6]. Senn writes that, in an RCT, ‘‘over all the random-
izations the groups are balanced; and that for a particular
randomization they are unbalanced’’ . Thus, in an RCT,
the only reason to use a significance test would be to exam-
significance testofthe association between the covariateand
the treatment assignment is a test of the hypothesis that the
treatments are randomly distributed. In other words, it is
a test of a null hypothesis that is known to be true’’ . Sim-
ilarly, Altman writes that ‘‘performing a significance test to
compare baseline variables is to assess the probability of
* Corresponding author. Institute for Clinical Evaluative Sciences,
G1 06, 2075 Bayview Avenue, Toronto, Ontario M4N 3M5, Canada.
Tel.: þ416-480-6131; fax: þ416-480-6048.
E-mail address: email@example.com (P.C. Austin).
0895-4356/10/$ e see front matter ? 2010 Elsevier Inc. All rights reserved.
Journal of Clinical Epidemiology 63 (2010) 142e153
What is new?
? Substantial variation exists among leading medical
journals in the use of statistical significance testing
to compare baseline covariates between treatment
groups in randomized controlled trials (RCTs).
? Unadjusted estimates of treatment effect were
reported more frequently than adjusted analyses.
? Primary outcomes were more frequently either
binary or time-to-event in nature than continuous
in published RCTs.
What this adds to what was known?
? In the current era, there remain inconsistencies
between the reporting of analyses in RCTs and what
is considered the best statistical practice.
What is the implication and what should change now?
forcing the statistical proscription against the use of
covariates between treatment groups in RCTs.
? There is a need for an informed debate about the
relative interpretability and utility for clinical and
policy decision making of unadjusted vs. adjusted
measures of treatment effect for binary and
something having occurred by chance when we know that it
did occur by chance. Such a procedure is clearly absurd’’
. Although randomization will, on average, balance
covariates between treated and untreated subjects, it need
not do so in any particular randomization.
A related issuewhen estimating the effect of treatment on
outcomes in RCTs is adjusting for baseline covariates.
be used when estimating the effect of the treatment on the
ance in prognostically important baseline covariates. Fur-
thermore, the use of regression analysis can result in
increased precision when estimating the treatment effect
[8,9]. Importantly, both Senn and Lavori et al. argue that
the covariates should be selected a priori, and that the
selected covariates should be strongly prognostic of the
outcome [1,9]. The selection of covariates for inclusion in
comparing their distribution across arms of the trial.
It is not known to what extent the prohibition against
using significance testing to compare baseline covariates
between treatment arms and the prescription of the use of
regression adjustment are adhered to in the current medical
literature. The objective of the current study was to conduct
a review of the treatment of baseline covariates in published
RCTs in the general medical literature.
2. Methods for the systematic review
We conducted a systematic review of RCTs published in
four leading general medical journals: the New England
Journal of Medicine (NEJM), the Journal of the American
Medical Association (JAMA), The Lancet, and the British
Medical Journal (BMJ). We searched PubMed for articles
published in these four journals between January 1, 2007
and June 30, 2007. We limited our search to reports of
RCTs in humans. We excluded cluster randomization trials,
reports of subgroup analyses of prior RCTs, secondary
analyses of prior RCTs, studies in which no primary
outcome was identified, and cost-effectiveness analyses
conducted alongside RCTs.
The CONSORT statement requires that authors describe
the baseline demographic and clinical characteristics of
each group (item 15) . In an explanation and elaboration
of the CONSORT statement, it is suggested that baseline
information is ‘‘efficiently presented in a table’’ . We
examined whether each published report of an RCT
included a table describing baseline demographic and
clinical characteristics of each arm. When such a table
was not reported, we examined whether a comparison of
baseline characteristics was described in the text.
Many RCTs use stratified randomization. Because it is
commonly agreed that stratification factors should be
accounted for in the analysis estimating treatment efficacy
, an analysis was not considered an adjusted analysis if
it adjusted only for stratification factors. Similarly, many
studies adjust the effect of treatment on outcome for the
baseline value of the response variable. An analysis that
adjusted for the value of the response at baseline was not
considered an adjusted analysis. Thus, an adjusted analysis
had to adjust for at least one baseline covariate that was not
a stratification factor and that was not the value of the
response variable at baseline.
For each identified study, we abstracted the following
information: (1) the primary outcome; (2) the nature of the
ing baseline characteristics between treatment arms. When
such a table was not published, we examined whether
between-group comparisons of baseline characteristics were
examined whether it was reported that significance testing
hadbeen usedtocompare thedistributionofbaselinecovari-
ates across treatment arms; (5) whether an unadjusted analy-
sis of the effect of treatment on the primary outcome was
reported. If an unadjusted analysis was reported for a contin-
uous outcome, we examined whether the analysis had
adjusted for the baseline value of the response variable; (6)
whether an adjusted analysis of the effect of treatment on
143 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
the primary outcome was reported; and (7) how authors
reported selecting the covariates for adjustment when an
adjusted analysis was reported.
When multiple primary outcomes were listed, we
selected the primary outcome that was described as the
‘‘first’’ primary outcome. If none of the primary outcomes
was described as the ‘‘first’’ primary outcome, we selected
the primary outcome that was used in the sample-size cal-
culation. If multiple primary outcomes were listed and each
was used in a sample-size calculation, then we used all
identified primary outcomes.
When recording the method used for selecting baseline
covariates for use in an adjusted analysis, we allowed for
three possible options. First, we noted whether the authors
explicitly reported that covariates for adjustment were
selected before the analysis. Second, we noted whether
the selection of covariates was reported to be based on
a post hoc analysis (e.g., observed imbalance in baseline
covariates, statistical significant differences in baseline
covariates between arms of the study, or by the use of an
automated variable-selection method, such as backward
variable elimination). Third, in the absence of the reported
use of one of the first two methods for covariate selection,
we described the method of covariate selections as unclear.
Each identified article was read independently by two
study authors. One study author (P.C.A.) read all the iden-
tified articles. Two of the authors (A.M. and M.B.S.) each
read one-third of the identified articles. A fourth study
author (M.Z.) abstracted some items from the remaining
third, whereas the fifth study author (D.N.J.) abstracted
between abstractors were resolved by the two abstractors
reviewing the disputed study and arriving at a consensus.
As a prespecified analysis, we used a chi-square test to
test whether there were differences among the four journals
in the proportion of articles that reported using statistical
hypothesis testing to compare baseline characteristics
between randomization arms.
3. Results of the systematic review
Results of the review are summarized in Table 1 and are
described in the following subsections.
3.1. Included studies
The initial search strategy identified 135 published
reports of RCTs [12e146]. Of these, 21 were excluded
for the following reasons: ancillary study of prior RCT
(one study) , cluster randomization trials (seven stud-
ies) [37,50,60,125,130,135,144], cost-effectiveness studies
(one study) , no primary outcome identified (three
studies) [26,116,142], pooled subgroup analysis of two
RCTs (one study) , secondary analysis of an RCT
(six studies) [25,31,71,88,112,128], subgroup analysis of
an RCT (one study) , and substudy of an RCT (one
study) . This resulted in the inclusion of 114 published
reports of RCTs. Of these, 42 (36.8%) were published in
the NEJM, 25 (21.9%) in the JAMA, 28 (24.6%) in The
Lancet, and 19 (16.7%) in the BMJ.
3.2. Comparison of baseline characteristics between
Of the included 114 randomized trials, 110 (96.5%)
reported a table in which baseline demographic and clinical
characteristics were compared between treatment arms. Of
the 110 randomized trials reporting such a table, signifi-
cance testing was used in 42 (38.2%) to compare the distri-
bution of baseline covariates across treatment arms. When
we examined the prevalence of significance testing to com-
pare baseline covariates in each journal separately, the
results were as follows: 29 of 40 (72.5%) articles in the
NEJM, 11 of 24 (45.8%) articles in the JAMA, 2 of 28
(7.1%) articles in The Lancet, and 0 of 18 (0%) articles
in the BMJ reported using statistical hypothesis testing to
compare baseline covariates
groups. A chi-square test was used to compare the propor-
tion of studies in which significance was used to compare
baseline covariates between the four journals. The signifi-
cance level of the chi-square test was less than 0.0001,
indicating that there were differences in the probability of
using significance testing to compare baseline covariates
across these four journals. In inspecting the aforementioned
results, one notes that only two of the 46 studies (4.4%)
published in the two British journals (The Lancet and the
BMJ) used significance testing to compare baseline covari-
ates across treatment arms. In contrast, 40 of the 64
(62.5%) studies published in the two American journals
(NEJM and JAMA) used significance testing (P ! 0.0001
for the comparison between the two British journals and
the two American journals). Of the four studies that did
not report a table comparing baseline characteristics
between treatment groups, three provided brief qualitative
or quantitative comparisons of the baseline characteristics
between study groups. Of these three studies, one study
reported the result of a significance test comparing a base-
line characteristic between the study groups. One study did
not report any comparison of baseline characteristics
between randomization arms.
3.3. Nature of the primary outcome
In the 114 included studies, the nature of primary outcome
was as follows: continuous (29.0%), binary (28.1%), time-to-
event (28.1%), binary/time-to-event (8.8%), count (2.6%),
continuous/binary (0.9%), continuous/time-to-event (0.9%),
and ordinal (1.8%). In several studies, the primary outcome
was analyzedasbotha binaryand asa time-to-event outcome
[14,36,66,79,90,93,106,110,134,137]. Journal-specific preva-
lences of the different types of outcomes are reported in
144 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
3.4. Unadjusted and adjusted estimates of treatment
Of the 114 included articles, 107 (93.9%) presented
unadjusted analyses of treatment effect, six (5.3%) did
not report unadjusted analysis, and for one study (0.9%),
it was unclear whether an unadjusted analysis had been
conducted. Similarly, 39 (34.2%) presented adjusted analy-
ses of treatment effect, 72 (63.2%) did not report adjusted
analysis, and for three studies (2.6%), it was unclear
whetheran adjusted analysis
Journal-specific frequencies of unadjusted and adjusted
analyses are reported in Table 1. Of the reviewed articles,
34 (31.8%) explicitly presented both unadjusted and
adjusted estimates of treatment effect.
In 33 (29.0%) studies, the nature of the primary outcome
was only continuous. We excluded one study in which it
was unclear whether an unadjusted analysis had been
conducted. In the remaining 32 studies, 27 (84.4%)
reported using an unadjusted analysis, 11 (34.4%) reported
an adjusted analysis, and six (18.8%) reported both unad-
justed and adjusted analysis. In 74 (64.9%) studies, the
nature of the primary outcome was either binary, time-
to-event, or binary/time-to-event. We excluded the two
studies in which it was unclear whether unadjusted or
adjusted analyses had been conducted. In the 72 remaining
studies, 71 (98.6%) reported using an unadjusted analysis,
25 (34.7%) reported an adjusted analysis, and 25 (34.7%)
reported both unadjusted and adjusted analyses. In 32 of
these 74 studies, the primary outcome was binary. Among
had been conducted.
these 32 studies, 31 (96.9%) reported an unadjusted analy-
sis, 11 (34.4%) reported an adjusted analysis, and 11
(34.4%) reported both unadjusted and adjusted analyses.
In 32 of the aforementioned 74 studies, the primary
outcome was time-to-event in nature. Among these 32 stud-
ies, 32 (100%) reported an unadjusted analysis, 12 (37.5%)
reported an adjusted analysis, and 12 (37.5%) reported both
unadjusted and adjusted analyses. Finally, in eight of the
aforementioned studies, the primary outcome was both
binary and time-to-event in nature. In these eight studies,
eight (100%) reported an unadjusted analysis, two
(25.0%) reported an adjusted analysis, and two (25.0%)
reported both unadjusted and adjusted analyses.
There were 28 studies that reported an unadjusted
estimate for the effect of treatment on a primary outcome
that was only a continuous variable. Of these, 15 (53.6%)
reported adjusting for the baseline value of the response
variable (in one of these 15 studies, the continuous outcome
was also the change from baseline). Twelve (42.9%) studies
did not adjust for the baseline value of the response vari-
able. In these 12 studies, the outcome variable was the
change from baseline in four studies and the percent change
from baseline in one study. In one of the 28 studies, it was
unclear whether the analysis had adjusted for the baseline
value of the response variable.
3.5. Selection of covariates for use in adjusted analyses
Thirty-nine studies reported the results of an adjusted
analysis. Of these, six (15.4%) clearly reported that the
Overall and journal-specific results of review of handling of baseline covariates in reports of RCTs
Item BMJJAMALancet NEJM Overall
Comparison of baseline characteristics
Table comparing baseline
Reported significance testing
to compare characteristics in
18/19 24/2528/28 40/42 110/114
0/1811/24 2/28 29/40 42/110
Nature of primary outcome
Estimates of treatment effect
Unadjusted analysis reported
Adjusted analysis reported
Reported method for the selection of covariates for adjusted analyses
A priori selection
Post hoc selection
Abbreviations: RCTs, randomized controlled trials; BMJ, British Medical Journal; JAMA, Journal of the American Medical Association; NEJM, New
England Journal of Medicine.
145 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
variables used for adjustment had been selected a priori. In
two (5.1%) studies, it was reported that stepwise regression
was used to select the baseline characteristics for inclusion
in the adjusted model. In three studies (7.7%), the selection
of covariates was reported to be based on significance test-
ing, whereas in two (5.1%) studies, the selection of covari-
ates was based on a post hoc observation or decision.
Finally, in 26 (66.7%) of the studies, it was unclear how
the covariates for inclusion in the regression model were
selected. Journal-specific prevalences of the different
methods for selection covariates for adjustment are
reported in Table 1. One notes that the proportion of studies
in which it was reported that covariates were selected
before the analysis varied across the four journals.
The objective of the current study was to examine the
that authors describe the baseline demographic and clinical
characteristics of each group (item 15) . In an explanation
and elaboration of the CONSORT statement, it is sugg-
ested that baseline information is ‘‘efficiently presented in
als reported a table comparing baseline characteristics be-
tween the treatment arms. Brief quantitative or qualitative
comparisons of the similarity of baseline characteristics be-
tween treatment groups were reported in the text in three of
the four articles in which a table was not published. Several
pare the distribution of baseline covariates in RCTs [1,3e6].
We found that 38.2% of studies reported using significance
testing to compare baseline covariates between treatment
arms. The use of significance testing varied across journals.
In general, the practice was rare in the two British journals
(BMJ and The Lancet), whereas it occurred in most of the
We speculate that the observed differences between journals
may reflect editorial preferences, with the two British
journals having an editorial proscription against the use of
the decision to individual authors or associate editors.
tests comparing baseline characteristics between study arms
is the result of confusion between three different questions:
(1) Has the randomization been properly conducted? (2)
Could imbalance in baseline characteristics cause chance
bias? (3) Should the analysis be adjusted for baseline vari-
ables? . The European Agency for the Evaluation of
Medical Products states: ‘‘Statistical testing for baseline
imbalance has no role in a trial where the handling of the
line summaries with respect to the main covariates should be
presented and discussed from a clinical point of view,
irrespective of whether a statistical test indicated a ‘‘statisti-
process of allocating patients to treatment has, in fact, not
been random then any resulting bias cannot be corrected by
any statistical adjustment’’ . Fayers and King provide
a brief summary of methods for handling baseline character-
istics in RCTs .
Given the strong condemnation in the statistical and
methodological literature of the use of significance testing
to compare baseline characteristics between treatment
arms, our study suggests the need for greater consistency
in editorial policies across journals concerning the report-
ing of RCTs. We suggest that journal editors adopt a policy
against the use of significance testing to compare baseline
covariates between treatment arms. As Roberts and Torger-
son suggest, authors may be incorrectly selecting covari-
ates for inclusion in a regression model to estimate an
adjusted treatment effect based on the results of statistical
tests comparing baseline characteristics between treatment
groups . Senn argues that the selection of covariates
for inclusion in a regression model should be based on
an a priori judgment of which variables are strongly
prognostic of the outcome . An inappropriate selection
of covariates for adjustment may result in biased estimates
of treatment effect. Pocock et al. suggest that there is often
inadequate prior knowledge as to which baseline character-
istics are related to prognosis, and that further research is
needed in variable selection algorithms . Furthermore,
there is the danger that ad hoc variable-selection methods
can lead to biased inferences . In a specific setting,
Pocock et al. suggest that if the correlation between a single
covariate and a continuous outcome is weak, then even
statistically significant covariate imbalance is unimportant,
whereas if a covariate is strongly correlated with the
outcome, then it is important to adjust for that cova-
riate, regardless of the statistical significance of the
Several authors have suggested that, rather than present-
ing unadjusted estimates, regression adjustment be used to
estimate treatment effects [1,7e9,150]. Advantages to this
approach include the ability to adjust for chance imbalance
in prognostically important baseline covariates between
treatment arms, and greater precision in estimating the
treatment effect. We found that approximately one-third
of studies reported an adjusted estimate of treatment effect.
In most of the studies, only unadjusted estimates of treat-
ment effect were reported.
The suggestion for reporting adjusted estimates has typ-
ically been made in the context of a linear treatment effect
and a continuous outcome variable. In the current study, we
found that 34.4% of the studies in which the outcome was
continuous reported an adjusted estimate of treatment
effect. We found that adjustment for baseline covariates
was used in the medical literature for dichotomous out-
comes and for time-to-event (survival) outcomes as well
as for continuous outcomes. Of studies in which the
146 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
primary outcome was binary or time-to-event in nature,
34.7% reported an adjusted analysis. However, there are
theoretical limitations to the use of regression adjustment
when estimating the effect of treatment on dichotomous
and time-to-event outcomes. Gail et al. found that the unad-
justed treatment effect in a randomized clinical trial can
lead to an attenuated estimate of treatment effect compared
with the adjusted treatment effect when the outcome is
dichotomous or time-to-event . When estimating the
effect of treatment on a binary or time-to-event outcome,
the unadjusted estimate of the treatment effect will be
systematically closer to the null value compared with the
adjusted treatment effect, even in the presence of random-
ization. However, for linear treatment effects on continuous
outcomes, the adjusted and unadjusted treatment effects
will, on average, coincide. This phenomenon is related to
the issue of the collapsibility of an estimator that has been
discussed by Greenland et al. [152,153]. An estimator is
collapsible when the population average or marginal effect
is the same as the conditional, adjusted, or subject-specific
effect. Differences in means and risk differences are
collapsible. The unadjusted difference in means is equal
to the subject-specific difference in means. For binary out-
comes, regression adjustment would usually entail the use
of a logistic regression model. However, the odds ratio is
not a collapsible estimator. Thus, the marginal odds ratio
will differ from the conditional or subject-specific odds ra-
tio. For this and other reasons, the use of the odds ratio in
prospective studies is discouraged [152,154]. For binary
outcomes, risk differences and relative risks (assuming
a uniform relative risk) are collapsible estimators .
However, their use precludes the use of regression adjust-
ment (although an adjusted relative risk can be estimated
using a log-binomial generalized linear model, its use in
practice may be problematic owing to difficulties with esti-
mation and convergence). When outcomes are binary or
time-to-event, adjusted and unadjusted estimates can be
expected to differ. There is disagreement in the literature
as to which estimate (the marginal/unadjusted/crude esti-
mate vs. the conditional/adjusted) is more informative to
clinicians, policy makers, and health care funders. Hauck
et al. argue that the conditional estimate is more meaningful
from a clinical perspective . In contrast, Martens et al.
suggest that the marginal treatment effect is better defined
and appears to be of greater interest . This issue
requires further attention, because in most of the studies
examined, the primary outcome was either binary or
time-to-event in nature: settings in which the conditional
and marginal estimates do not coincide. In only a few stud-
ies (29.0%) was the primary outcome continuous in nature.
In two recent articles, methods for estimating absolute
risk reductions and numbers needed to treat from adjusted
logistic regression models and adjusted survival models
have been proposed [157,158]. These methods are applica-
ble both in the context of RCTs, in which one wants to
adjust for potential imbalance in prognostically important
baseline covariates, and in observational studies, in which
there may be systematic differences between treated and
untreated subjects. Increased use of these methods in RCTs
with binary or time-to-event outcomes would allow for the
reporting of risk differences, which are collapsible: the pop-
ulation-average risk difference is equivalent to the subject-
specific risk difference. Furthermore, absolute measures of
treatment effect may be more useful for clinical decision
making than relative measures, such as the odds ratio.
Our study suggests the need for an informed debate
about the relative merits of the interpretation of unadjusted
vs. adjusted estimates of treatment effect in settings with
binary or time-to-event outcomes. In particular, the relative
merits of adjusted vs. unadjusted estimates of treatment
effect for clinicians, policy makers, and health care funders
require further examination. If marginal or population-
average treatment effects are more meaningful from either
a clinical or a policy perspective, then solely reporting
adjusted estimates of treatment effect may result in a misun-
derstanding about the magnitude of the benefit of treatment
at the population level .
A second criticism of the advocacy of regression adjust-
ment in RCTs is that it only allows adjustment for measured
have argued that if measured prognostic variables are imbal-
anced, it is possible that prognostically important unmea-
sured variables are also imbalanced . Regression
adjustment using measured baseline variables may result in
the unmeasured baseline variables remaining imbalanced
to investigators and readers.
We found that six of the 114 (5.3%) studies did not report
implications of presenting only adjusted estimates of treat-
ment effect. Meta-analyses pool estimates of treatment
effects across RCTs examining the same exposure and out-
come. The pooling of unadjusted treatment effects is rela-
tively simple from a conceptual perspective, because the
nature of the treatment effect is the same across all studies-
dan unadjusted or population-average treatment effect.
However, interpreting the results of a meta-analysis that
pooled adjusted treatment effects is more difficult. This is
because each study may have adjusted for different baseline
covariates. Thus, each study reports a different adjusted
exist multiple conditional treatment effects: adjusting for
a different set of covariates results in a different conditional
treatment effect . From a conceptual perspective, it is
unclear how to interpret the pooling of adjusted treatment
effects, when different sets of covariates were controlled
for in each study. Additionally, pooling adjusted and unad-
justed treatment effects is problematic when outcomes are
ing estimates of conditional and marginal treatment effects.
Because of the noncollapsibility of the odds ratio and the
147P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
hazard ratio, these are estimating different quantities. For
of unadjusted estimates is strongly encouraged. These prob-
lems are exacerbated when the analysis of summary data
involves the use of multiparameter evidence synthesis or
mixed treatment-comparison models .
error was the use of significance testing to compare baseline
balance in covariates. Altman and Dore reviewed the reports
of 80 RCTs published in 1987 and 1988 in four general
medical journals (including three that were considered in
our review) . They found that hypothesis tests were used
to compare baseline variables in 58% of the trials (Altman
and Dore note that, of the 600 hypothesis tests conducted
in these 46 trials, 4% were significant at the 5% level). How-
ever,they didnotreportjournal-specificresultsforthe useof
significance testing to compare baseline characteristics.
Pocock et al. examined the reports of 50 RCTs published in
1997 in the same four general medical journals that were
esis tests were used to compare baseline variables in 48% of
the trials. In comparison, we found that 38.2% of studies
reported using significance testing to compare baselinechar-
acteristics. The study by Pocock et al. did not report journal-
specific results. Thus, journal-specific comparisons cannot
be made with our study. The use of significance testing to
to have decreased substantially across the eras examined by
Lavori et al. (1978 and 1979), by Altman and Dore (1987
and 1988), by Pocock et al. (1997), and by the current
Altman and Dore found that when estimating treatment
effects, 64% of studies reported either unadjusted estimates
or estimates that accounted for change from baseline. Only
26% of the studies reported using statistical modeling that
adjusted for baseline covariates when estimating the treat-
ment effect. In the study by Altman and Dore, journal-
specific rates of reporting the use of an adjusted analyses
were 40%, 10%, 20%, and 35% for the Annals of Internal
Medicine, the BMJ, The Lancet, and the NEJM, respec-
tively. In the current review, it was found that 34.2% of
studies reported an analysis that adjusted for baseline cova-
riates. Thus, there was a modest increase in the use of
regression adjustment for estimating treatment effects
between 1987 and 2007. In the current review, journal-
specific rates of reporting an adjusted analysis were
36.8%, 40.0%, 42.9%, and 23.8% for the BMJ, the JAMA,
The Lancet, and the NEJM, respectively. Therefore, the
reported use of an adjusted analysis increased in both the
BMJ, and The Lancet between the era examined by Altman
and Dore (1987/1988) and that considered in the current
review (2007). However, the use of adjusted analyses
decreased in the NEJM between these two periods.
There are several limitations to the current study. First,
we only examined RCTs published in four leading general
medical journals over a 6-month period in 2007. We did not
examine RCTs published in other journals. Thus, we do not
know the extent to which our findings are generalizable to
other journals. However, despite restricting our review to
leading general medical journals, we did observe substan-
tial between-journal heterogeneity in some aspects of the
reporting of the results of RCTs. Second, we were only able
to examine what authors reported, and not what was
actually done. A prior study found that in a substantial
proportion of studies, there were unacknowledged discrep-
ancies between the published reports and the study
protocols . It is conceivable that there were discrep-
ancies between what was done and what was reported in
the studies that we examined.
In summary, we found that the statistical treatment of
baseline covariates in leading general medical journals
differs from the guidance provided in the statistical litera-
ture. In the NEJM and in the JAMA, statistical hypothesis
testing was routinely used to compare baseline covariates
between treatment arms, whereas the practice was rare in
the BMJ and The Lancet. Adjustment for potential imbal-
ance in baseline covariates was only performed in a few
studies. In most of the RCTs examined, the primary
outcomes were either binary or time-to-event in nature.
Further examination of the relative merits of adjusted and
unadjusted estimates of treatment effects in this context is
The Institute for Clinical Evaluative Sciences (ICES) is
supported in part by a grant from the Ontario Ministry of
Health and Long Term Care. The opinions, results, and
conclusions are those of the authors and no endorsement
by the Ministry of Health and Long-Term Care or by the
Institute for Clinical Evaluative Sciences is intended or
should be inferred. Dr. Austin is supported in part by
a Career Investigator Award from the Heart and Stroke
Foundation of Ontario, and Dr. Juurlink was supported by
a New Investigator Award from the Canadian Institutes
for Health Research.
 Senn S. Testing for baseline balance in clinical trials. Stat Med
 Moher D, Schulz KF, Altman D. The CONSORT statement: revised
randomized trials. JAMA 2001;285:1987e91.
 Begg CB. Significance tests of covariate imbalance in clinical trials.
Control Clin Trials 1990;11:223e5.
 Altman DG. Comparability of randomised groups. Statistician
 Rothman KJ. Epidemiologic methods in clinical trials. Cancer
 Altman DT, Dore CJ. Randomisation and baseline comparisons in
clinical trials. Lancet 1990;335:149e53.
148 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
 Senn SJ. Covariate imbalance and random allocation in clinical
trials. Stat Med 1989;8:467e75.
 Altman DG, Dore CJ. Baseline comparisons in randomized clinical
trials. Stat Med 1991;10:797e802.
 Lavori PW, Louis TA, Bailar JC III, Polansky M. Designs for
experimentsdparallel comparisons of treatment. N Engl J Med
 Altman DG, Schulz KF, Moher D, Egger M, Davidoff F,
Elbourne D, et al. for the CONSORT Group. The revised CON-
SORT statement for reporting randomized trials: explanation and
elaboration. Ann Intern Med 2001;134:663e94.
 Cook TD, DeMets DL, editors. Introduction to statistical methods
for clinical trials. Boca Raton, FL: Chapman & Hall/CRC; 2008.
 Savoye M, Shaw M, Dziura J, Tamborlane WV, Rose P,
Guandalini C, et al. Effects of a weight management program on
body composition and metabolic parameters in overweight children:
a randomized controlled trial. JAMA 2007;297:2697e704.
 Cole BF, Baron JA, Sandler RS, Haile RW, Ahnen DJ, Bresalier RS,
et al. Polyp Prevention Study Group. Folic acid for the prevention of
colorectal adenomas: a randomized clinical trial. JAMA 2007;297:
 Dorsey G, Staedke S, Clark TD, Njama-Meya D, Nzarubara B,
Maiteki-Sebuguzi C, et al. Combination therapy for uncomplicated
falciparum malaria in Ugandan children: a randomized trial. JAMA
 Leslie T, Mayan MI, Hasan MA, Safi MH, Klinkenberg E, Whitty CJ,
et al. Sulfadoxine-pyrimethamine, chlorproguanil-dapsone, or chloro-
quine for the treatment of Plasmodium vivax malaria in Afghanistan
and Pakistan: a randomized controlled trial. JAMA 2007;297:2201e9.
 Ebbeling CB, Leidig MM, Feldman HA, Lovesky MM, Ludwig DS.
Effects of a low-glycemic load vs low-fat diet in obese young adults:
a randomized trial. JAMA 2007;297:2092e102.
of physical activity on cardiorespiratory fitness among sedentary,
overweight or obese postmenopausal women with elevated blood
pressure: a randomized controlled trial. JAMA 2007;297:2081e91.
 Erne P, Schoenenberger AW, Burckhardt D, Zuber M, Kiowski W,
Buser PT, et al. Effects of percutaneous coronary interventions in
silent ischemia after myocardial infarction: the SWISSI II random-
ized controlled trial. JAMA 2007;297:1985e91.
 Mebazaa A, Nieminen MS, Packer M, Cohen-Solal A, Kleber FX,
for patients with acute decompensated heart failure: the SURVIVE
randomized trial. JAMA 2007;297:1883e91.
 Morrow DA, Scirica BM, Karwatowska-Prokopczuk E, Murphy SA,
Budaj A, Varshavsky S, et al. MERLIN-TIMI 36 Trial Investigators.
non-ST-elevation acute coronary syndromes: the MERLIN-TIMI 36
randomized trial. JAMA 2007;297:1775e83.
 Tardif JC, Gr? egoire J, L’Allier PL, Ibrahim R, Lesp? erance J,
Heinonen TM, et al. Effect of rHDL on Atherosclerosis-Safety and
lipoprotein infusions on coronary atherosclerosis: a randomized
controlled trial. JAMA 2007;297:1675e82.
 Alexander JH, ReynoldsHR,
Harrington RA, Van de Werf F, et al. TRIUMPH Investigators. Ef-
fect of tilarginine acetate in patients with acute myocardial infarc-
tion and cardiogenic shock: the TRIUMPH randomized controlled
trial. JAMA 2007;297:1657e66.
 Treanor JJ, Schiff GM, Hayden FG, Brady RC, Hay CM, Meyer AL,
et al. Safety and immunogenicity of a baculovirus-expressed hemag-
glutinin influenza vaccine: a randomized controlled trial. JAMA
 Halonen J, Halonen P, Ja ¨rvinen O, Taskinen P, Auvinen T,
Tarkka M, et al. Corticosteroids for the prevention of atrial fibrilla-
tion after cardiac surgery: a randomized controlled trial. JAMA
Stebbins AL, DzavikV,
 Rossouw JE, Prentice RL, Manson JE, Wu L, Barad D,
Barnabei VM, et al. Postmenopausal hormone therapy and risk of
cardiovascular disease by age and years since menopause. JAMA
 Brandes JL, Kudrow D, Stark SR, O’Carroll CP, Adelman JU,
O’Donnell FJ, et al. Sumatriptan-naproxen for acute treatment of
migraine: a randomized trial. JAMA 2007;297:1443e54.
 Gheorghiade M, Konstam MA, Burnett JC Jr, Grinfeld L,
Maggioni AP, Swedberg K, et al. Efficacy of Vasopressin Antago-
nism in Heart Failure Outcome Study with Tolvaptan (EVEREST)
Investigators. Short-term clinical effects of tolvaptan, an oral
vasopressin antagonist, in patients hospitalized for heart failure:
the EVEREST Clinical Status Trials. JAMA 2007;297:1332e43.
 Konstam MA, Gheorghiade M, Burnett JC Jr, Grinfeld L,
Maggioni AP, Swedberg K, et al. Efficacy of Vasopressin Antago-
nism in Heart Failure Outcome Study with Tolvaptan (EVEREST)
Investigators. Effects of oral tolvaptan in patients hospitalized for
worsening heart failure: the EVEREST Outcome Trial. JAMA
 Nissen SE, Nicholls SJ, Wolski K, Howey DC, McErlean E,
Wang MD, et al. Effects of a potent and selective PPAR-alpha
agonist in patients with atherogenic dyslipidemia or hypercholes-
terolemia: two randomized controlled trials. JAMA 2007;297:
 Crouse JR III, Raichlen JS, Riley WA, Evans GW, Palmer MK,
O’Leary DH, et al. METEOR Study Group. Effect of rosuvastatin
on progression of carotid intima-media thickness in low-risk
individuals with subclinical atherosclerosis: the METEOR Trial.
and minor ECG abnormalities in asymptomatic women and risk of
cardiovascular events and mortality. JAMA 2007;297:978e85.
 Gardner CD, Kiazand A, Alhassan S, Kim S, Stafford RS,
Balise RR, et al. Comparison of the Atkins, Zone, Ornish, and
LEARN diets for change in weight and related risk factors among
overweight premenopausal women: the ATO Z Weight Loss Study:
a randomized trial. JAMA 2007;297:969e77.
 Schnurr PP, Friedman MJ, Engel CC, Foa EB, Shea MT, Chow BK,
et al. Cognitive behavioral therapy for posttraumatic stress disorder
in women: a randomized controlled trial. JAMA 2007;297:820e30.
 van Dijk D, Spoor M, Hijman R, Nathoe HM, Borst C, Jansen EW,
et al. Octopus Study Group. Cognitive and cardiac outcomes 5 years
after off-pump vs on-pump coronary artery bypass graft surgery.
 Zacharski LR, Chow BK, Howes PS, Shamayeva G, Baron JA,
Dalman RL, et al. Reduction of iron stores and cardiovascular
outcomes in patients with peripheral arterial disease: a randomized
controlled trial. JAMA 2007;297:603e10.
 Stone GW, Bertrand ME, Moses JW, Ohman EM, Lincoff AM,
Ware JH, et al. ACUITY Investigators. Routine upstream initiation
vs deferred selective use of glycoprotein IIb/IIIa inhibitors in acute
coronary syndromes: the ACUITY Timing trial. JAMA 2007;297:
 Thiam S, LeFevre AM, Hane F, Ndiaye A, Ba F, Fielding KL, et al.
Effectiveness of a strategy to improve adherence to tuberculosis
treatment in a resource-poor setting: a cluster randomized controlled
trial. JAMA 2007;297:380e6.
 Lesp? erance F, Frasure-Smith N, Koszycki D, Lalibert? e MA, van
Zyl LT, Baker B, et al. CREATE Investigators. Effects of citalopram
and interpersonal psychotherapy on depression in patients with
coronary artery disease: the Canadian Cardiac Randomized Evalua-
tion of Antidepressant and Psychotherapy Efficacy (CREATE) trial.
 Oettle H, Post S, Neuhaus P, Gellert K, Langrehr J, Ridwelski K,
et al. Adjuvant chemotherapy with gemcitabine vs observation in
patients undergoing curative-intent resection of pancreatic cancer:
a randomized controlled trial. JAMA 2007;297:267e77.
149 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
 Armstrong PW, Granger CB, Adams PX, Hamm C, Holmes D Jr,
O’Neill WW, et al. APEX AMI Investigators. Pexelizumab for acute
ST-elevation myocardial infarction in patients undergoing primary
percutaneous coronary intervention: a randomized controlled trial.
 StrasserH, MarksteinerR,
Mitterberger M, Frauscher F, et al. Autologous myoblasts and fibro-
blasts versus collagen for treatment of stress urinary incontinence in
women: a randomised controlled trial. Lancet 2007;369:2179e86.
 Paavonen J, Jenkins D, Bosch FX, Naud P, Salmero ´n J,
Wheeler CM, et al. HPV PATRICIA Study Group. Efficacy of
a prophylactic adjuvanted bivalent L1 virus-like-particle vaccine
against infection with human papillomavirus types 16 and 18 in
young women: an interim analysis of a phase III double-blind,
randomised controlled trial. Lancet 2007;369:2161e70.
 Darboe MK, Thurnham DI, Morgan G, Adegbola RA, Secka O,
Solon JA, et al. Effectiveness of an early supplementation scheme
of high-dose vitamin A versus standard WHO protocol in Gambian
mothers and infants: a randomised controlled trial. Lancet
 Solomon SD, Janardhanan R, Verma A, Bourgoun M, Daley WL,
Purkayastha D, et al. Valsartan In Diastolic Dysfunction (VALIDD)
Investigators. Effect of angiotensin receptor blockade and antihyper-
tensive drugs on diastolic function in patients with hypertension and
diastolic dysfunction: a randomised trial. Lancet 2007;369:
 Khan MS, Dar O, Sismanidis C, Shah K, Godfrey-Faussett P. Im-
provement of tuberculosis case detection and reduction of discrep-
ancies between men and women by simple sputum-submission
instructions: a pragmatic randomised controlled trial. Lancet
 von Hertzen H, Piaggio G, Huong NT, Arustamyan K, Cabezas E,
Gomez M, et al. WHO Research Group on Postovulatory Methods
omised controlled equivalence trial. Lancet 2007;369:1938e46.
 Gilligan D, Nicolson M, Smith I, Groen H, Dalesio O, Goldstraw P,
cell lung cancer: results of the MRC LU22/NVALT 2/EORTC 08012
multicentre randomised trial and update of systematic review. Lancet
 Andang’o PE, Osendarp SJ, Ayah R, West CE, Mwaniki DL,
De Wolf CA, et al. Efficacy of iron-fortified whole maize flour on
iron status of schoolchildren in Kenya: a randomised controlled
trial. Lancet 2007;369:1799e806.
 Chan FK, Wong VW, Suen BY, Wu JC, Ching JY, Hung LC, et al.
Combination of a cyclo-oxygenase-2 inhibitor and a proton-pump
inhibitor for prevention of recurrent ulcer bleeding in patients at
very high risk: a double-blind, randomised trial. Lancet 2007;369:
 Griffiths C, Sturdy P, Brewin P, Bothamley G, Eldridge S,
Martineau A, et al. Educational outreach to promote screening for
tuberculosis in primary care: a cluster randomised controlled trial.
 Kuse ER, Chetchotisakd P, da Cunha CA, Ruhnke M, Barrios C,
Raghunadharao D, et al. Micafungin Invasive Candidiasis Working
Group. Micafungin versus liposomal amphotericin B for candidae-
mia and invasive candidosis: a phase III randomised double-blind
trial. Lancet 2007;369:1519e27.
 Mochizuki S, Dahlo ¨f B, Shimizu M, Ikewaki K, Yoshikawa M,
Taniguchi I, et al. Jikei Heart Study Group. Valsartan in a Japanese
population with hypertension and other cardiovascular disease
(Jikei Heart Study): a randomised, open-label, blinded endpoint
morbidity-mortality study. Lancet 2007;369:1431e9.
 Sherman DG, Albers GW, Bladin C, Fieschi C, Gabbai AA,
Kase CS, et al. PREVAIL Investigators. The efficacy and safety of
enoxaparin versus unfractionated heparin for the prevention of
venous thromboembolism after acute ischaemic stroke (PREVAIL
Study): an open-label randomised comparison. Lancet 2007;369:
 Grinsztejn B, Nguyen BY, Katlama C, Gatell JM, Lazzarin A,
Vittecoq D, et al. Protocol 005 Team. Safety and efficacy of the
HIV-1 integrase inhibitor raltegravir (MK-0518) in treatment-exper-
ienced patients with multidrug-resistant virus: a phase II randomised
controlled trial. Lancet 2007;369:1261e9.
 Clotet B, Bellos N, Molina JM, Cooper D, Goffard JC, Lazzarin A,
et al. POWER 1 and 2 Study Groups. Efficacy and safety of daruna-
vir-ritonavir at week 48 in treatment-experienced patients with
HIV-1 infection in POWER 1 and 2: a pooled subgroup analysis
of data from two randomised trials. Lancet 2007;369:1169e78.
 Yokoyama M, Origasa H, Matsuzaki M, Matsuzawa Y, Saito Y,
Ishikawa Y, et al. Japan EPA Lipid Intervention Study (JELIS) Inves-
hypercholesterolaemic patients (JELIS): a randomised open-label,
blinded endpoint analysis. Lancet 2007;369:1090e8.
 Franc ¸ois B, Bellissant E, Gissot V, Desachy A, Normand S,
Boulain T, et al. Association des R? eanimateurs du Centre-Ouest
(ARCO). 12-h pretreatment with methylprednisolone versus placebo
for prevention of postextubation laryngeal oedema: a randomised
double-blind trial. Lancet 2007;369:1083e9.
 Marson AG, Al-Kharusi AM, Alwaidh M, Appleton R, Baker GA,
Chadwick DW, et al. SANAD Study Group. The SANAD study of
effectiveness of valproate, lamotrigine, or topiramate for generalised
and unclassifiable epilepsy: an unblinded randomised controlled
trial. Lancet 2007;369:1016e26.
 Marson AG, Al-Kharusi AM, Alwaidh M, Appleton R, Baker GA,
Chadwick DW, et al. SANAD Study Group. The SANAD study of
effectiveness of carbamazepine, gabapentin, lamotrigine, oxcarba-
zepine, or topiramate for treatment of partial epilepsy: an unblinded
randomised controlled trial. Lancet 2007;369:1000e15.
 Sazawal S, Black RE, Ramsan M, Chwaya HM, Dutta A,
Dhingra U, et al. Effect of zinc supplementation on mortality in
children aged 1-48 months: a community-based randomised place-
bo-controlled trial. Lancet 2007;369:927e34.
 Stone GW, White HD, Ohman EM, Bertrand ME, Lincoff AM,
McLaurin BT, et al. Acute Catheterization and Urgent Intervention
Triage strategy (ACUITY) Trial Investigators. Bivalirudin in
patients with acute coronary syndromes undergoing percutaneous
coronary intervention: a subgroup analysis from the Acute Catheter-
ization and Urgent Intervention Triage strategy (ACUITY) trial.
 Nadel S, Goldstein B, Williams MD, Dalton H, Peters M,
Macias WL, et al. REsearching severe Sepsis and Organ dysfunction
in children: a gLobal perspective (RESOLVE) Study Group. Drotre-
cogin alfa (activated) in children with severe sepsis: a multicentre
phase III randomised controlled trial. Lancet 2007;369:836e43.
 Hirsch A, Windhausen F, Tijssen JG, Verheugt FW, Cornel JH,
de Winter RJ Invasive versus Conservative Treatment in Unstable
coronary Syndromes (ICTUS) Investigators. Long-term outcome
after an early invasive versus selective invasive treatment strategy
in patients with non-ST-elevation acute coronary syndrome and
elevated cardiac troponin T (the ICTUS trial): a follow-up study.
 Ratcliff A,SiswantoroH,
Wuwung RM, Laihad F, et al. Two fixed-dose artemisinin combina-
tions for drug-resistant falciparum and vivax malaria in Papua,
 Heijnen EM, Eijkemans MJ, De Klerk C, Polinder S, Beckers NG,
Klinkert ER, et al. A mild treatment strategy for in-vitro fertilisa-
tion: a randomised non-inferiority trial. Lancet 2007;369:743e9.
 Gray RH, Kigozi G, Serwadda D, Makumbi F, Watya S,
Nalugoda F, et al. Male circumcision for HIV prevention in men
in Rakai, Uganda: a randomised trial. Lancet 2007;369:657e66.
150 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
 Bailey RC, Moses S, Parker CB, Agot K, Maclean I, Krieger JN,
et al. Male circumcision for HIV prevention in young men in Kisu-
mu, Kenya: a randomised controlled trial. Lancet 2007;369:643e56.
 Coombes RC, Kilburn LS, Snowdon CF, Paridaens R, Coleman RE,
Jones SE, et al. Intergroup Exemestane Study. Survival and safety of
exemestane versus tamoxifen after 2-3 years’ tamoxifen treatment
(Intergroup Exemestane Study): a randomised controlled trial.
 Zongo I, Dorsey G, Rouamba N, Tinto H, Dokomajilar C,
Guiguemde RT, et al. Artemether-lumefantrine versus amodiaquine
plus sulfadoxine-pyrimethamine for uncomplicated falciparum
malaria in Burkina Faso: a randomised non-inferiority trial. Lancet
 Malhotra-Kumar S, Lammens C, Coenen S, Van Herck K,
Goossens H. Effect of azithromycin and clarithromycin therapy on
pharyngeal carriage of macrolide-resistant streptococci in healthy
volunteers: a randomised, double-blind, placebo-controlled study.
 Durga J, van Boxtel MP, Schouten EG, Kok FJ, Jolles J, Katan MB,
et al. Effect of 3-year folic acid supplementation on cognitive func-
tion in older adults in the FACIT trial: a randomised, double blind,
controlled trial. Lancet 2007;369:208e16.
 Conter V, Valsecchi MG, Silvestri D, Campbell M, Dibar E,
to intensive chemotherapy for children with intermediate-risk acute
lymphoblastic leukaemia: a multicentre randomised trial. Lancet
 Smith I, Procter M, Gelber RD, Guillaume S, Feyereislova A,
Dowsett M, et al. HERA Study Team. 2-year follow-up of trastuzu-
mab after adjuvant chemotherapy in HER2-positive breast cancer:
a randomised controlled trial. Lancet 2007;369:29e36.
 Home PD, Pocock SJ, Beck-Nielsen H, Gomis R, Hanefeld M,
Jones NP, et al. RECORD Study Group. Rosiglitazone evaluated
for cardiovascular outcomesdan interim analysis. N Engl J Med
 Manson JE, Allison MA, Rossouw JE, Carr JJ, Langer RD, Hsia J,
et al. WHI and WHI-CACS Investigators. Estrogen therapy and
coronary-artery calcification. N Engl J Med 2007;356:2591e602.
 Sundar S, Jha TK, Thakur CP, Sinha PK, Bhattacharya SK. Inject-
able paromomycin for visceral leishmaniasis in India. N Engl J
 Manzoni P, Stolfi I, Pugni L, Decembrino L, Magnani C, Vetrano G,
et al. Italian Task Force for the Study and Prevention of Neonatal
Fungal Infections. Italian Society of Neonatology. A multicenter,
randomized trial of prophylactic fluconazole in preterm neonates.
N Engl J Med 2007;356:2483e95.
 Reboli AC, Rotstein C, Pappas PG, Chapman SW, Kett DH,
Kumar D, et al. Anidulafungin Study Group. Anidulafungin versus
fluconazole for invasive candidiasis. N Engl J Med 2007;356:
 Dember LM, Hawkins PN, Hazenberg BP, Gorevic PD, Merlini G,
Butrimiene I, et al. Eprodisate for AA Amyloidosis Trial Group.
Eprodisate for the treatment of renal disease in AA amyloidosis.
N Engl J Med 2007;356:2349e60.
 Hudes G, Carducci M, Tomczak P, Dutcher J, Figlin R,
Kapoor A, et al. Global ARCC Trial. Temsirolimus, interferon
alfa, or both for advanced renal-cell carcinoma. N Engl J Med
 Weinstein JN, Lurie JD, Tosteson TD, Hanscom B, Tosteson AN,
Blood EA, et al. Surgical versus nonsurgical treatment for lumbar
degenerative spondylolisthesis. N Engl J Med 2007;356:2257e70.
 Peul WC, van Houwelingen HC, van den Hout WB, Brand R,
Eekhof JA, Tans JT, et al. Leiden-The Hague Spine Intervention
Prognostic Study Group. Surgery versus prolonged conservative
treatment for sciatica. N Engl J Med 2007;356:2245e56.
 Albo ME, Richter HE, Brubaker L, Norton P, Kraus SR,
Zimmern PE, et al. Urinary Incontinence Treatment Network. Burch
colposuspension versus fascial sling to reduce urinary stress incon-
tinence. N Engl J Med 2007;356:2143e55.
 Papi A, Canonica GW, Maestrelli P, Paggiaro P, Olivieri D, Pozzi E,
et al. BEST Study Group. Rescue use of beclomethasone and
albuterol in a single inhaler for mild asthma. N Engl J Med
 Peters SP, Anthonisen N, Castro M, Holbrook JT, Irvin CG,
Smith LJ, et al. American Lung Association Asthma Clinical
Research Centers. Randomized comparison of strategies for
reducing treatment in mild persistent asthma. N Engl J Med
 Garland SM, Hernandez-Avila M, Wheeler CM, Perez G,
Harper DM, Leodolter S, et al. Females United to Unilaterally
Reduce Endo/Ectocervical Disease (FUTURE) I Investigators.
Quadrivalent vaccine against human papillomavirus to prevent
anogenital diseases. N Engl J Med 2007;356:1928e43.
 FUTURE II Study Group. Quadrivalent vaccine against human
papillomavirus to prevent high-grade cervical lesions. N Engl J
 Jacobson AM, Musen G, Ryan CM, Silvers N, Cleary P, Waberski B,
et al. Diabetes Control and Complications Trial/Epidemiology of
Diabetes Interventions and Complications Study Research Group.
Long-term effect of diabetes and its treatment on cognitive function.
N Engl J Med 2007;356:1842e52.
 Sezer M, Oflaz H, Go ¨ren T, Okc ¸ular I, Umman B, Nis xanci Y, et al.
Intracoronary streptokinase after primary percutaneous coronary
intervention. N Engl J Med 2007;356:1823e34.
 Black DM, Delmas PD, Eastell R, Reid IR, Boonen S, Cauley JA,
et al. HORIZON Pivotal Fracture Trial. Once-yearly zoledronic acid
for treatment of postmenopausal osteoporosis. N Engl J Med
 SachsGS, NierenbergAA,
Wisniewski SR, Gyulai L, et al. Effectiveness of adjunctive antide-
pressant treatment for bipolar depression. N Engl J Med 2007;356:
 Lau JY, Leung WK, Wu JC, Chan FK, Wong VW, Chiu PW, et al.
Omeprazole before endoscopy in patients with gastrointestinal
bleeding. N Engl J Med 2007;356:1631e40.
 Lacroix J, H? ebert PC, Hutchison JS, Hume HA, Tucci M, Ducruet T,
et al. TRIPICU Investigators; Canadian Critical Care Trials Group;
Pediatric Acute Lung Injury and Sepsis Investigators Network.
Transfusion strategies for patients in pediatric intensive care units.
N Engl J Med 2007;356:1609e19.
 Kastelein JJ,van Leuven
Kuivenhoven JA, Barter PJ, et al. RADIANCE 1 Investigators. Ef-
fect of torcetrapib on carotid atherosclerosis in familial hypercholes-
terolemia. N Engl J Med 2007;356:1620e30.
 Cuba IPV Study Collaborative Group. Randomized, placebo-con-
trolled trial of inactivated poliovirus vaccine in Cuba. N Engl J
 Keime-Guibert F, Chinot O, Taillandier L, Cartalat-Carel S,
Frenay M, Kantor G, et al. Association of French-Speaking Neuro--
Oncologists. Radiotherapy for glioblastoma in the elderly. N Engl J
 Larsen CM, Faulenbach M, Vaag A, Vølund A, Ehses JA, Seifert B,
et al. Interleukin-1-receptor antagonist in type 2 diabetes mellitus.
N Engl J Med 2007;356:1517e26.
 Boden WE, O’Rourke RA, Teo KK, Hartigan PM, Maron DJ,
Kostuk WJ, et al. COURAGE Trial Research Group. Optimal
medical therapy with or without PCI for stable coronary disease.
N Engl J Med 2007;356:1503e16.
 Fawzi WW, Msamanga GI, Urassa W, Hertzmark E, Petraro P,
Willett WC, et al. Vitamins and perinatal outcomes among
 Cox G, Thomson NC, Rubin AS, Niven RM, Corris PA, Siersted HC,
et al. AIR Trial Study Group. Asthma control during the year after
bronchial thermoplasty. N Engl J Med 2007;356:1327e37.
SI, BurgessL,Evans GW,
151 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
 Nissen SE, Tardif JC, Nicholls SJ, Revkin JH, Shear CL,
Duggan WT, et al. ILLUSTRATE Investigators. Effect of torcetra-
pib on the progression of coronary atherosclerosis. N Engl J Med
 Tonetti MS, D’Aiuto F, Nibali L, Donald A, Storry C, Parkar M,
et al. Treatment of periodontitis and endothelial function. N Engl
J Med 2007;356:911e20.
 Shrestha MP, Scott RM, Joshi DM, Mammen MP Jr, Thapa GB,
Thapa N, et al. Safety and efficacy of a recombinant hepatitis E
vaccine. N Engl J Med 2007;356:895e903.
 Nagot N, Ou? edraogo A, Foulongne V, Konat? e I, Weiss HA,
Vergne L, et al. ANRS 1285 Study Group. Reduction of HIV-1
RNA levels with therapy to suppress herpes simplex virus. N Engl
J Med 2007;356:790e9.
 Calverley PM, Anderson JA, Celli B, Ferguson GT, Jenkins C,
Jones PW, et al. TORCH Investigators. Salmeterol and fluticasone
propionate and survival in chronic obstructive pulmonary disease.
N Engl J Med 2007;356:775e89.
 Belshe RB, Edwards KM, Vesikari T, Black SV, Walker RE,
Hultquist M, et al. CAIV-T Comparative Efficacy Study Group. Live
attenuated versus inactivated influenza vaccine in infants and young
children. N Engl J Med 2007;356:685e96.
 Cahen DL, Gouma DJ, Nio Y, Rauws EA, Boermeester MA,
Busch OR, et al. Endoscopic versus surgical drainage of the pancre-
atic duct in chronic pancreatitis. N Engl J Med 2007;356:676e84.
 NewburgerJW, SleeperLA,
Gersony W, Vetter VL, et al. Pediatric Heart Network Investigators.
Randomized trial of pulsed corticosteroid therapy for primary treat-
ment of Kawasaki disease. N Engl J Med 2007;356:663e75.
et al. CNTO 1275 PsoriasisStudy Group. A human interleukin-12/23
monoclonal antibody for the treatment of psoriasis. N Engl J Med
 Legro RS, Barnhart HX, Schlaff WD, Carr BR, Diamond MP,
Carson SA, et al. Cooperative Multicenter Reproductive Medicine
Network. Clomiphene, metformin, or both for infertility in the
polycystic ovary syndrome. N Engl J Med 2007;356:551e66.
 Lautrette A, Darmon M, Megarbane B, Joly LM, Chevret S,
Adrie C, et al. A communication strategy and brochure for relatives
of patients dying in the ICU. N Engl J Med 2007;356:469e78.
 Kuhle J, Pohl C, Mehling M, Edan G, Freedman MS, Hartung HP,
et al. Lack of association between antimyelin antibodies and
progression to multiple sclerosis. N Engl J Med 2007;356:371e8.
et al. Committee of the Randomized Trial of Embolization versus
Surgical Treatment for Fibroids. Uterine-artery embolization versus
surgery for symptomatic uterine fibroids. N Engl J Med 2007;356:
 Cornely OA, Maertens J, Winston DJ, Perfect J, Ullmann AJ,
Walsh TJ, et al. Posaconazole vs. fluconazole or itraconazole
prophylaxis in patients with neutropenia. N Engl J Med 2007;356:
 Ullmann AJ, Lipton JH, Vesole DH, Chandrasekar P, Langston A,
Tarantolo SR, et al. Posaconazole or fluconazole for prophylaxis in
severe graft-versus-host disease. N Engl J Med 2007;356:335e47.
9 to 11 years of age. N Engl J Med 2007;356:248e61.
 Lockman S, Shapiro RL, Smeaton LM, Wester C, Thior I,
Stevens L, et al. Response to antiretroviral therapy after a single,
peripartum dose of nevirapine. N Engl J Med 2007;356:135e47.
 Escudier B, Eisen T, Stadler WM, Szczylik C, Oudard S, Siebels M,
et al. TARGET Study Group. Sorafenib in advanced clear-cell
renal-cell carcinoma. N Engl J Med 2007;356:125e34.
 Motzer RJ, Hutson TE, Tomczak P, Michaelson MD, Bukowski RM,
Rixe O, et al. Sunitinib versus interferon alfa in metastatic renal-cell
carcinoma. N Engl J Med 2007;356:115e24.
McCrindleBW, Minich LL,
 Hickson M, D’Souza AL, Muthu N, Rogers TR, Want S,
Rajkumar C, et al. Use of probiotic Lactobacillus preparation to pre-
vent diarrhoea associated with antibiotics: randomised double blind
placebo controlled trial. BMJ 2007;335:80.
 Farmer A, Wade A, Goyder E, Yudkin P, French D, Craven A, et al.
Impact of self monitoring of blood glucose in the management of
patients with non-insulin treated diabetes: open parallel group
randomised trial. BMJ 2007;335:132.
 Goodyer I, Dubicka B, Wilkinson P, Kelvin R, Roberts C, Byford S,
et al. Selective serotonin reuptake inhibitors (SSRIs) and routine
specialist care with and without cognitive behaviour therapy in
adolescents with major depression: randomised controlled trial.
 Gohel MS, Barwell JR, Taylor M, Chant T, Foy C, Earnshaw JJ,
et al. Long term results of compression therapy alone versus
compression plus surgery in chronic venous ulceration (ESCHAR):
randomised controlled trial. BMJ 2007;335:83.
 Montgomery AA, Emmett CL, Fahey T, Jones C, Ricketts I,
Patel RR, et al. DiAMOND Study Group. Two decision aids for
mode of delivery among women with previous caesarean section:
randomised controlled trial. BMJ 2007;334:1305.
 Ranson MK, Sinha T, Chatterjee M, Gandhi F, Jayswal R, Patel F,
et al. Equitable utilisation of Indian community based health
insurance scheme among its rural membership: cluster randomised
controlled trial. BMJ 2007;334:1309.
 Ronco G, Cuzick J, Pierotti P, Cariaggi MP, Dalla Palma P,
Naldoni C, et al. Accuracy of liquid based versus conventional
cytology: overall results of new technologies for cervical cancer
screening: randomised controlled trial. BMJ 2007;335:28.
 Duffy M, Gillespie K, Clark DM. Post-traumatic stress disorder in
the context of terrorism and other civil conflict in Northern Ireland:
randomised controlled trial. BMJ 2007;334:1147.
 Kang JH, Cook N, Manson J, Buring JE, Grodstein F. Low dose
aspirin and cognitive function in the women’s health study cognitive
cohort. BMJ 2007;334:987.
 Holland R, Brooksby I, Lenaghan E, Ashton K, Hay L, Smith R,
et al. Effectiveness of visits from community pharmacists for
patients with heart failure: HeartMed randomised controlled trial.
 de Groot M, de Keijser J, Neeleman J, Kerkhof A, Nolen W,
Burger H. Cognitive behaviour therapy to prevent complicated grief
among relatives and spouses bereaved by suicide: cluster rando-
mised controlled trial. BMJ 2007;334:994.
 Salter C, Holland R, Harvey I, Henwood K. ‘‘I haven’t even phoned
my doctor yet.’’ The advice giving role of the pharmacist during
consultations for medication review with patients aged 80 or more:
qualitative discourse analysis. BMJ 2007;334:1101.
 Hutchings J, Gardner F, Bywater T, Daley D, Whitaker C, Jones K,
et al. Parenting intervention in Sure Start services for children at risk
of developing conduct disorder: pragmatic randomised controlled
trial. BMJ 2007;334:678.
 Edwards RT, C? eilleachair A, Bywater T, Hughes DA, Hutchings J.
Parenting programme for parents of children at risk of developing
conduct disorder: cost effectiveness analysis. BMJ 2007;334:682.
 Alho OP, Koivunen P, Penna T, Teppo H, Koskela M, Luotonen J.
Tonsillectomy versus watchful waiting in recurrent streptococcal
 Howden-Chapman P,Matheson
Cunningham M, Blakely T, et al. Effect of insulating existing houses
on health inequality: cluster randomised study in the community.
 Mutrie N, Campbell AM, Whyte F, McConnachie A, Emslie C,
Lee L, et al. Benefits of supervised group exercise programme for
women being treated for early stage breast cancer: pragmatic rand-
omised controlled trial. BMJ 2007;334:517.
 Campbell IA, Bentley DP, Prescott RJ, Routledge PA, Shetty HG,
Williamson IJ. Anticoagulation for three versus six months in
A, Crane J,ViggersH,
152 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153
patients with deep vein thrombosis or pulmonary embolism, or both:
randomised trial. BMJ 2007;334:674.
 Bech BH, Obel C, Henriksen TB, Olsen J. Effect of reducing
caffeine intake on birth weight and length of gestation: randomised
controlled trial. BMJ 2007;334:409.
 Reyburn H, Mbakilwa H, Mwangi R, Mwerinde O, Olomi R,
Drakeley C, et al. Rapid diagnostic tests compared with malaria
microscopy for guiding outpatient treatment of febrile illness in
Tanzania: randomised trial. BMJ 2007;334:403.
 PetrieKJ,Mu ¨ller JT, Schirmbeck F, DonkinL, BroadbentE, EllisCJ,
et al. Effect of providing information about normal test results on pa-
tients’ reassurance: randomised controlled trial. BMJ 2007;334:352.
 Banu SH, Jahan M, Koli UK, Ferdousi S, Khan NZ, Neville B. Side
effects of phenobarbital and carbamazepine in childhood epilepsy:
randomised controlled trial. BMJ 2007;334:1207.
 Koh TH, Butow PN, Coory M, Budge D, Collie LA, Whitehall J,
et al. Provision of taped conversations with neonatologists to
mothers of babies in intensive care: randomised controlled trial.
 Sazawal S, Dhingra U, Dhingra P, Hiremath G, Kumar J, Sarkar A,
et al. Effects of fortified milk on morbidity in young children in
north India: community based, randomised, double masked placebo
controlled trial. BMJ 2007;334:140.
 Henderson M, Wight D, Raab GM, Abraham C, Parkes A, Scott S,
et al. Impact of a theoretically based sex education programme
(SHARE) delivered by teachers on NHS registered conceptions
and terminations: final results of cluster randomised trial. BMJ
 Brown CT, Yap T, Cromwell DA, Rixon L, Steed L, Mulligan K,
et al. Self management for men with lower urinary tract symptoms:
randomised controlled trial. BMJ 2007;334:25.
 Zar HJ, Cotton MF, Strauss S, Karpakis J, Hussey G, Schaaf HS,
et al. Effect of isoniazid prophylaxis on mortality and incidence
of tuberculosis in children with HIV: randomised controlled trial.
 Roberts C, Torgerson DJ. Understanding controlled trials: baseline
imbalance in randomised controlled trials. BMJ 1999;319:185.
Committee for Proprietary Medical Products (CPMP). Points to
consider on adjustment for baseline covariates. Available at:
Accessed October 3, 2008.
 Fayers P, King M. The baseline characteristics did not differ signif-
icantly. Qual Life Res 2008;17:1047e8.
 Pocock SJ, Assmann SE, Enos LE, Kasten LE. Subgroup analysis, co-
variate adjustment and baseline comparisons in clinical trial reporting:
current practice and problems. Stat Med 2002;21:2917e30.
 Gail MH, Wieand S, Piantadosi S. Biased estimates of treatment
effect in randomized experiments with nonlinear regressions and
omitted covariates. Biometrika 1984;7:431e44.
 Greenland S. Interpretation and choice of effect measures in
epidemiologic analyses. Am J Epidemiol 1987;125:761e8.
 Greenland S, Robins JM, Pearl J. Confounding and collapsibility in
causal inference. Stat Sci 1999;14:29e46.
 Newcombe RG. A deficiency of the odds ratio as a measure of effect
size. Stat Med 2006;25:4235e40.
 Hauck WW, Anderson S, Marcus SM. Should we adjust for
covariates in nonlinear regression analyses of randomized trials?
Control Clin Trials 1998;19:249e56.
 Martens EP, Pestman WR, Klungel OH. Conditioning on the
propensity score can result in biased estimation of common
measures of treatment effect: a Monte Carlo study. (p n/a) by Peter
C. Austin, Paul Grootendorst, Sharon-Lise T. Normand, Geoffrey
M. Anderson, Stat Med, Published Online: June 16, 2006. DOI:
10.1002/sim. 2618. Stat Med 2007;26:3208e10.
totreat canbeobtainedfroma logistic regressionmodel.J ClinEpide-
 Austin PC. Absolute risk reductions and numbers needed to treat can
J Clin Epidemiol. 2010;63:46e55.
 Shrier I, Platt RW, Steele RJ. Re: ‘‘Variable selection for propensity
score models’’. Am J Epidemiol 2007;166:238e89.
 Ades AE, Sutton AJ. Multiparameter evidence synthesis in
epidemiology and medical decision-making: current approaches.
J R Stat Soc Ser A 2006;169:5e35.
 Chan AW, Hrobjartsson A, Jorgensen KJ, Gotzsche PC, Altman DG.
Discrepancies in sample size calculations and data analyses reported
in randomised trials: comparison of publications with protocols.
BMJ 2008;337:a2299. DOI: 10.1136/bmj.a2299.
153 P.C. Austin et al. / Journal of Clinical Epidemiology 63 (2010) 142e153