The n e w e ng l a n d j o u r na l of m e dic i n e
Geographic Variations in Controlled Trials
To the Editor: In their exploration of multi
national clinical trial analysis, Yusuf and Wittes
(Dec. 8 issue)1 appropriately emphasize the neces-
sity of distinguishing true heterogeneity across
countries from chance variation. We wish to point
out additional limitations of testing for country-
level heterogeneity.
A positive heterogeneity test could be driven
by especially low efficacy (i.e., harm) in some
countries rather than by high efficacy in others.
Trialists should consider supplementing such
tests with shrinkage estimation analysis, a statis-
tical tool that refines subgroup estimates with
the use of data beyond the subgroup. This tech-
nique provides more accurate estimates of effi-
cacy in subgroups (with confidence intervals) by
pulling subgroup findings toward the overall
mean in proportion to the uncertainty underly-
ing the results in that subgroup.2,3
Even when the estimate for one country indi-
cates unusually high efficacy and chance is not
the cause, the benefits may not materialize if the
trial intervention is implemented nationally, be-
cause study sites are only a small nonrandom
sample within the country. The study sites may
not represent the care system, study population,
enrollment practices, and other elements in the
country more broadly. Trialists should proceed
with caution.
AaronL. Schwartz, Ph.D.
Harvard Medical School
Boston, MA
AriB. Friedman, M.D., Ph.D.
Beth Israel Deaconess Medical Center
Boston, MA
arib@alumni.upenn.edu
No potential conflict of interest relevant to this letter was re-
ported.
1. Yusuf S, Wittes J. Interpreting geographic variations in results
of randomized, controlled trials. N Engl J Med 2016;375:2263-71.
2. James W, Stein C. Estimation with quadratic loss. In: Pro-
ceedings of the Fourth Berkeley Symposium on Mathematical
Statistics and Probability. Vol. 1. Berkeley:University of Califor-
nia Press, 1961:361-79.
3. Willan AR, Pinto EM, OBrien BJ, et al. Country specific cost
comparisons from multinational clinical trials using empirical
Bayesian shrinkage estimation: the Canadian ASSENT-3 econom-
ic analysis. Health Econ 2005;14:327-38.
DOI: 10.1056/NEJMc1700529
1196 n engl j med 376;12nejm.org March 23, 2017
The New England Journal of Medicine
Downloaded from nejm.org on March 31, 2017. For personal use only. No other uses without permission.
Copyright 2017 Massachusetts Medical Society. All rights reserved.
To the Editor: Yusuf and Wittes report exam-
ples of regional differences in the results of trials
and provide interpretation regarding whether such
differences are likely to be due to chance. A recent
and striking example of such regional differences
concerns the cardiovascular safety of glucagon-
like peptide 1 (GLP-1) analogues and inhibitors
of sodiumglucose cotransporter 2 (SGLT2). Over-
all, trials evaluating these drugs have shown a
benefit with regard to cardiovascular outcomes,
yet a subgroup meta-analysis from four trials1-4
that enrolled 25,725 patients reveals significant
differences in cardiovascular outcomes according
to region (Fig. 1). The global effect size is driven
by Latin America, Africa, and Asia, whereas the
effects in Europe and North America are extreme-
ly small or nonexistent. How should we interpret
these variations?
Matthieu Roustit, Pharm.D., Ph.D.
Universit Grenoble Alpes
Grenoble, France
MRoustit@chu-grenoble.fr
Charles Khouri, Pharm.D.
CHU de Grenoble
Grenoble, France
Rmy Boussageon, M.D., Ph.D.
Universit de Poitiers
Poitiers, France
Figure 1 (facing page). Forest Plot of an Inverse
Variance Random Effect Meta-Analysis of Trials
That Assessed GLP-1 Analogues or SGLT-2 Inhibitors
in Patients with Type 2 Diabetes.
The trials included in the meta-analysis evaluated glu-
cagon-like peptide 1 (GLP-1) analogues and inhibitors
of sodiumglucose cotransporter 2 (SGLT2) in patients
with type 2 diabetes who were receiving treatment with
a standard-of-care regimen. The primary outcome was
cardiovascular death, nonfatal myocardial infarction, or
nonfatal stroke. Subgroups were based on geographic
region and included North America; Europe; and Latin
America, Africa, and Asia; only studies from which re-
gional data were available were included. Empagliflozin
data are from Zinman et al.,1 lixisenatide data are from
Pfeffer et al.,2 liraglutide data are from Marso et al.,3 and
semaglutide data are from Marso et al.4 RevMan soft-
ware, version 5.3 (Cochrane Collaboration), was used
for analysis. CI denotes confidence interval.
 Correspondence

Antidiabetic
Subgroup Drug Placebo Weight Hazard Ratio (95% CI)
no. of patients %
Latin America, Africa, and Asia
Empagliflozin, Africa 211 102 2.0 0.86 (0.451.65)
Empagliflozin, Asia 897 450 5.1 0.70 (0.491.01)
Empagliflozin, Latin America 721 360 4.4 0.58 (0.390.86)
Liraglutide, Asia 360 351 2.9 0.62 (0.371.04)
Liraglutide, world except Asia, 1,268 1,218 9.4 0.83 (0.681.03)
Europe, and North America
Lixisenatide, Africa, Near East 154 142 2.3 0.66 (0.361.20)
Lixisenatide, Asia, Pacific 374 329 3.1 0.99 (0.601.63)
Lixisenatide, Latin America 972 972 7.7 0.86 (0.671.10)
Semaglutide, world except 752 776 5.3 0.68 (0.480.98)
Europe and North America
Subtotal 5,709 4,700 42.2 0.77 (0.690.86)
Heterogeneity: 2 =0.00; 2 =6.03,
8 df (P=0.64); I2 =0%
Test for overall effect: Z=4.46 (P<0.001)
Europe
Empagliflozin, Europe 1,926 959 8.3 1.02 (0.811.28)
Liraglutide, Europe 1,639 1,657 9.9 0.82 (0.680.98)
Lixisenatide, eastern Europe 776 811 7.5 1.19 (0.921.54)
Lixisenatide, western Europe 354 377 4.6 1.45 (0.992.12)
Semaglutide, Europe 326 306 2.3 0.62 (0.341.13)
Subtotal 5,021 4,110 32.6 1.01 (0.801.26)
Heterogeneity: 2 =0.04; 2 =12.07,
4 df (P=0.02); I2 =67%
Test for overall effect: Z=0.06 (P=0.95)
North America
Empagliflozin, North America (with Australia 932 462 6.0 0.89 (0.651.21)
and New Zealand)
Liraglutide, North America 1,401 1,446 10.0 1.01 (0.841.22)
Lixisenatide, North America 404 403 5.2 0.95 (0.671.35)
Semaglutide, United States 570 567 4.0 0.87 (0.571.34)
Subtotal 3,307 2,878 25.2 0.96 (0.841.10)
Heterogeneity: 2 =0.00; 2 =0.73,
3 df (P=0.87); I2 =0%
Test for overall effect: Z=0.56 (P=0.58)
Total 14,037 11,688 100.0 0.88 (0.790.97)
Heterogeneity: 2 =0.02; 2 =28.69,
17 df (P=0.04); I2 =41%
Test for overall effect: Z=2.61 (P=0.009)
Test for subgroup differences: 2 =7.95,
2 df (P=0.02); I2 =74.9%
0.50 0.75 1.00 1.50 2.00

Antidiabetic Drug Placebo Better

Better

No potential conflict of interest relevant to this letter was re-
ported.
1. Zinman B, Wanner C, Lachin JM, et al. Empagliflozin, car-
diovascular outcomes, and mortality in type 2 diabetes. N Engl
J Med 2015;373:2117-28.
2. Pfeffer MA, Claggett B, Diaz R, et al. Lixisenatide in patients
with type 2 diabetes and acute coronary syndrome. N Engl J Med
2015;373:2247-57.
3. Marso SP, Daniels GH, Brown-Frandsen K, et al. Liraglutide
and cardiovascular outcomes in type 2 diabetes. N Engl J Med
2016;375:311-22.
4. Marso SP, Bain SC, Consoli A, et al. Semaglutide and cardio-
vascular outcomes in patients with type 2 diabetes. N Engl J Med
2016;375:1834-44.
DOI: 10.1056/NEJMc1700529

n engl j med 376;12 nejm.org March 23, 2017 1197

The New England Journal of Medicine
Downloaded from nejm.org on March 31, 2017. For personal use only. No other uses without permission.
Copyright 2017 Massachusetts Medical Society. All rights reserved.

The authors reply: We agree with the recom-
mendations from Schwartz and Friedman about
the usefulness of shrinkage estimates to assess
the effects of treatments in subgroups. The broad-
er issue of the expected effect from implement-
ing the results of an intervention proven to be
effective in specific countries within a trial will
depend on a large number of factors that go be-
yond interpretation of subgroup results within
trials, including some that Schwartz and Fried-
man have raised.
Roustit et al. provide an example of an appar-
ent benefit with GLP-1 analogues and SGLT2
inhibitors being confined to patients enrolled
from Africa, Asia, and Latin America but not
from Europe or North America. In our view, this
is probably due to chance. First, Asians, Latin
Americans, and Africans are highly heteroge-
neous in their genetics, lifestyles, and risks of
diabetes, and so there is no biologic rationale for
putting them into a single group. The decision
to group them for this analysis was probably
data-derived. Second, we know of no rationale for
combining GLP-1 analogues or SGLT2 inhibitors,
since their mechanisms of action are quite differ-
ent from one another. Third, some trials of di-
peptidyl peptidase 4 inhibitors have not reported
such results.1 Inclusion of the results from these
trials may negate the apparent interaction accord-
ing to region that was presented by Roustit et al.
Salim Yusuf, D.Phil.
Population Health Research Institute
Hamilton, ON, Canada
yusufs@mcmaster.ca
Janet Wittes, Ph.D.
Statistics Collaborative
Washington, DC
Since publication of their article, the authors report no fur-
ther potential conflict of interest.
1. Green JB, Bethel MA, Armstrong PW, et al. Effect of sita-
gliptin on cardiovascular outcomes in type 2 diabetes. N Engl J
Med 2015;373:232-42.
DOI: 10.1056/NEJMc1700529
Correspondence Copyright 2017 Massachusetts Medical Society.
instructions for letters to the editor
Letters to the Editor are considered for publication, subject
to editing and abridgment, provided they do not contain
material that has been submitted or published elsewhere.
Letters accepted for publication will appear in print, on our
website at NEJM.org, or both.
Please note the following:
Letters in reference to a Journal article must not exceed 175
1198 n engl j med 376;12nejm.org March 23, 2017
The New England Journal of Medicine
Downloaded from nejm.org on March 31, 2017. For personal use only. No other uses without permission.
Copyright 2017 Massachusetts Medical Society. All rights reserved.
notices
words (excluding references) and must be received within
3 weeks after publication of the article.
Letters not related to a Journal article must not exceed 400
words.
A letter can have no more than five references and one figure
or table.
A letter can be signed by no more than three authors.
Financial associations or other possible conflicts of interest
must be disclosed. Disclosures will be published with the
letters. (For authors of Journal articles who are responding
to letters, we will only publish new relevant relationships
that have developed since publication of the article.)
Include your full mailing address, telephone number, fax
number, and e-mail address with your letter.
All letters must be submitted at authors.NEJM.org.
Letters that do not adhere to these instructions will not be
considered. We will notify you when we have made a decision
about possible publication. Letters regarding a recent Journal
article may be shared with the authors of that article. We are
unable to provide prepublication proofs. Submission of a
letter constitutes permission for the Massachusetts Medical
Society, its licensees, and its assignees to use it in the Journals
various print and electronic publications and in collections,
revisions, and any other form or medium.
notices
Notices submitted for publication should contain a mailing
address and telephone number of a contact person or depart-
ment. We regret that we are unable to publish all notices
received.
CANCER METASTASIS THROUGH THE
LYMPHOVASCULAR SYSTEM: BIOLOGY & TREATMENT
The 7th international symposium will be held in San Fran-
cisco, April 2022. It is sponsored by the Sentinel Node Oncol-
ogy Foundation.
Contact the Sentinel Node Oncology Foundation, 62 Richard-
son Rd., Novato, CA 94949; or e-mail joefner@cancermetastasis
.org; or see http://cancermetastasis.org.
5TH ANNUAL MEETING OF THE INTERNATIONAL
CYTOKINE AND INTERFERON SOCIETY
The meeting, entitled Looking Beyond the Horizon of Inte-
grated Cytokine Research, will be held in Kanazawa, Japan,
Oct. 29Nov. 2.
Contact the International Cytokine and Interferon Society,
297 Kinderkamack Rd., Suite 348, Oradell, NJ 07649; or call
(800) 947-1960; or fax (201) 322-1818; or see http://icis2017japan
.com.
the journals web and e-mail addresses
To submit a letter to the Editor: authors.NEJM.org
For information about the status of a submitted manuscript:
authors.NEJM.org
To submit a meeting notice: meetingnotices@NEJM.org
The Journals web pages: NEJM.org