Content uploaded by Viktor Oskarsson
Author content
All content in this area was uploaded by Viktor Oskarsson on Dec 20, 2017
Content may be subject to copyright.
Letters to the Editors
Letter: coffee consumption and gallstone
disease –a cautionary note on the
assignment of exposure values in
dose–response meta-analyses
A. Crippa*
,†
, A. Discacciati*
,†
, N. Orsini*
,†
&
V. Oskarsson*
*Unit of Nutritional Epidemiology, Institute of Environmental
Medicine, Karolinska Institutet, Stockholm, Sweden.
†
Unit of Biostatistics, Institute of Environmental Medicine, Karolinska
Institutet, Stockholm, Sweden.
E-mail: viktor.oskarsson@ki.se
doi:10.1111/apt.13428
SIRS, We read with interest the dose–response meta-anal-
ysis by Zhang et al., where the authors reported a non-
linear inverse association between coffee consumption
and risk of gallstone disease (P
non-linearity
<0.05).
1
This
analysis was based on four prospective cohort studies
that presented results for several categories of coffee con-
sumption.
2–4
According to common practice, Zhang
et al. assigned the mid-point of each category as a proxy
for the exposure level. For the highest open-ended cate-
gory, however, the mid-point was set at 1.5 times the
lower boundary, which is quite unusual and may lead to
too high values of coffee consumption (up to 9 cups/
day). In this letter, we would like to discuss the conse-
quences of that exposure assignment and, by doing so,
draw attention to the importance of sensitivity analyses
in dose–response meta-analyses.
Data were reanalysed using different methods of expo-
sure assignment. In addition to that of Zhang et al.
(which referred to a meta-analysis on egg consumption),
we used a method proposed in two recent meta-analyses
on coffee consumption.
5, 6
This method is different
in only one respect; it assumes that the highest open-
ended category has the same amplitude as the preceding
one, which may lead to more reasonable values of coffee
consumption (up to 6.5 cups/day). For comparison, we
obtained the actual median values directly from the
authors of the original studies (up to 7 cups/day). Statis-
tical analyses were conducted with the R package dosres-
meta.
7
Compared to the median values, Zhang et al.’s mid-
points overestimated the coffee consumption in the high-
est categories by at least 22%. In contrast, the alternative
mid-points were much closer to the median values
(8%). As expected, we observed a non-linear inverse
association between coffee consumption and risk of gall-
stone disease when using Zhang et al.’s mid-points
(P
non-linearity
<0.05) (Figure 1). However, there was no
evidence of a nonlinear association when using the other
methods of exposure assignment, neither the alternative
mid-points (P
non-linearity
=0.60) nor the median values
(P
non-linearity
=0.69). Furthermore, using either of these
two methods, the inverse association was stronger than
that reported by Zhang et al. For example, the hazard
ratio (95% confidence interval) for 6 cups/day vs. 0
cups/day was 0.66 (0.51–0.85) when we used the median
values [previously reported to be 0.75 (0.64–0.88)].
The issues we have discussed in this letter do not
change Zhang et al.’s conclusion that coffee consumption
is related to a decreased risk of gallstone disease. How-
ever, they do change the interpretation of the dose–
response results, that is, the shape and magnitude of the
exposure-disease association (which is highly relevant
given the widespread consumption of coffee and the high
incidence of gallstones).
More broadly, this letter exemplifies the general
importance of exposure assignment in dose–response
meta-analyses. Whenever possible, we encourage meta-
analysts to retrieve information on the exposure distribu-
tion directly from the original authors. We also recom-
mend that sensitivity analyses are routinely performed to
examine whether the assigned exposure values have a
strong influence on the dose–response results.
AP&T invited editorial and correspondence columns are restricted to letters discussing papers that have been published
in the journal. A letter must have a maximum of 500 words, may contain one table or figure, and should have no more
than 10 references. It should be submitted electronically to the Editors via http://mc.manuscriptcentral.com/apt.
ª2015 John Wiley & Sons Ltd166
Alimentary Pharmacology and Therapeutics
ACKNOWLEDGEMENTS
The authors thank Michael Leitzmann, Diane Feskanich and Caro-
line Nordenvall for providing original data on coffee consumption.
Declaration of personal and funding interests: None.
REFERENCES
1. Zhang YP, Li WQ, Sun YL, Zhu RT, Wang WJ. Systematic
review with meta-analysis: coffee consumption and the risk
of gallstone disease. Aliment Pharmacol Ther 2015; 42:
637–48.
2. Nordenvall C, Oskarsson V, Wolk A. Inverse association
between coffee consumption and risk of cholecystectomy in
women but not in men. Clin Gastroenterol Hepatol 2015; 13:
1096–102.e1.
3. Leitzmann MF, Willett WC, Rimm EB, et al. A prospective study
of coffee consumption and the risk of symptomatic gallstone
disease in men. JAMA 1999; 281: 2106–12.
4. Leitzmann MF, Stampfer MJ, Willett WC, Spiegelman D, Colditz
GA, Giovannucci EL. Coffee intake is associated with lower risk
of symptomatic gallstone disease in women. Gastroenterology
2002; 123: 1823–30.
5. Discacciati A, Orsini N, Wolk A. Coffee consumption and risk of
nonaggressive, aggressive and fatal prostate cancer—a dose–
response meta-analysis. Ann Oncol 2014; 25: 584–91.
6. Crippa A, Discacciati A, Larsson SC, Wolk A, Orsini N. Coffee
consumption and mortality from all causes, cardiovascular
disease, and cancer: a dose-response meta-analysis. Am J
Epidemiol 2014; 180: 763–75.
7. Crippa A, Orsini N. Dosresmeta: performing multivariate dose-
response meta-analysis. Available at: http://CRAN.R-
project.org/package=dosresmeta (accessed 18 September 2015).
Letter: coffee consumption and gallstone
disease –a cautionary note on the
assignment of exposure values in dose–
response meta-analyses. Authors’reply
Y.-P. Zhang
1
, W.-Q. Li
1
, Y.-L. Sun, R.-T. Zhu & W.-J. Wang
Department of Hepatobiliary and Pancreatic Surgery, School of
Medicine, Institute of Hepatobiliary and Pancreatic Diseases, The First
Affiliated Hospital of Zhengzhou University, Zhengzhou University,
Zhengzhou, China.
E-mail: ylsun@zzu.edu.cn.
doi:10.1111/apt.13442
1
These authors contributed equally to this work.
SIRS, We would like to thank Crippa and colleagues for
their letter regarding our recent article.
1, 2
They raised
the point that different exposure assignment values
might change the shape of the dose–response curve in
meta-analyses and, thus, encouraged direct retrieval of
the exposure distribution data from the original authors.
We whole-heartedly agree with this point. However, dis-
cussion of this issue is warranted.
In assigning a single value between boundaries in each
category of exposure, the summary slope of individual
dose–response studies might be sensitive to the method
of assignment.
3
According to Berlin et al.,
4
assigning a
mean value to each category may reduce the apparent
0.6
0.7
0.8
0.9
1.0
Coffee consumption (cups/day)
Hazard ratio
0123456789
Figure 1 | Pooled dose–response association between coffee consumption and risk of gallstone disease when using different
exposure assignments for the highest open-ended category: assigning a mid-point of 1.5 times the lower boundary (dotted
line); assigning a mid-point under the assumption that the category had the same amplitude as the preceding one (dashed
line); and assigning the actual median value (solid line). The hazard ratios are plotted on the log scale.
Aliment Pharmacol Ther 2016; 43: 166–180 167
ª2015 John Wiley & Sons Ltd
Letters to the Editors