Content uploaded by Yan Xie
Author content
All content in this area was uploaded by Yan Xie on Jun 10, 2019
Content may be subject to copyright.
Available via license: CC BY-NC 4.0
Content may be subject to copyright.
RESEARCH
thebmj
BMJ
2019;365:l1580 | doi: 10.1136/bmj.l1580 1
Estimates of all cause mortality and cause specific mortality
associated with proton pump inhibitors among US veterans:
cohort study
Yan Xie,1,2 Benjamin Bowe,1,3 Yan Yan,1,4 Hong Xian,1,3 Tingting Li,1,5 Ziyad Al-Aly1,2,5,6,7
ABSTRACT
OBJECTIVE
To estimate all cause mortality and cause specic
mortality among patients taking proton pump
inhibitors (PPIs).
DESIGN
Longitudinal observational cohort study.
SETTING
US Department of Veterans Aairs.
PARTICIPANTS
New users of PPIs (n=157 625) or H2 blockers
(n=56 842).
MAIN OUTCOME MEASURES
All cause mortality and cause specic mortality
associated with taking PPIs (values reported as number
of attributable deaths per 1000 patients taking PPIs).
RESULTS
There were 45.20 excess deaths (95% condence
interval 28.20 to 61.40) per 1000 patients taking
PPIs. Circulatory system diseases (number of
attributable deaths per 1000 patients taking PPIs
17.47, 95% condence interval 5.47 to 28.80),
neoplasms (12.94, 1.24 to 24.28), infectious
and parasitic diseases (4.20, 1.57 to 7.02), and
genitourinary system diseases (6.25, 3.22 to 9.24)
were associated with taking PPIs. There was a graded
relation between cumulative duration of PPI exposure
and the risk of all cause mortality and death due
to circulatory system diseases, neoplasms, and
genitourinary system diseases. Analyses of subcauses
of death suggested that taking PPIs was associated
with an excess mortality due to cardiovascular
disease (15.48, 5.02 to 25.19) and chronic kidney
disease (4.19, 1.56 to 6.58). Among patients without
documented indication for acid suppression drugs
(n=116 377), taking PPIs was associated with an
excess mortality due to cardiovascular disease (22.91,
11.89 to 33.57), chronic kidney disease (4.74, 1.53 to
8.05), and upper gastrointestinal cancer (3.12, 0.91 to
5.44). Formal interaction analyses suggested that the
risk of death due to these subcauses was not modied
by a history of cardiovascular disease, chronic kidney
disease, or upper gastrointestinal cancer. Taking
PPIs was not associated with an excess burden of
transportation related mortality and death due to
peptic ulcer disease (as negative outcome controls).
CONCLUSIONS
Taking PPIs is associated with a small excess of cause
specic mortality including death due to cardiovascular
disease, chronic kidney disease, and upper
gastrointestinal cancer. The burden was also observed
in patients without an indication for PPI use. Heightened
vigilance in the use of PPI may be warranted.
Introduction
Proton pump inhibitors (PPIs) are widely used either as
prescription or over-the-counter drugs.1 2 Several studies
suggest that taking PPIs is associated with a number
of serious adverse events including cardiovascular
disease, acute kidney injury, chronic kidney disease,
dementia, pneumonia, gastric cancer, Clostridium
dicile infections, and osteoporotic fractures.3 Some of
these adverse events are associated with an increased
risk of death. Recent studies described an excess risk
of all cause mortality among patients taking PPIs.4
However, a detailed quantitative analysis of the cause
specific mortality that is attributable to taking PPIs
is not available. We hypothesized that taking PPIs is
associated with an increased risk of cause specific
mortality that are mapped to well characterized adverse
events of PPIs. Identification of specific causes of death
attributable to taking PPIs will inform the public about
the risk of taking PPIs in the long term and could inform
risk stratification, risk mitigation strategies, and help
shape the development of deprescription interventions
to reduce unnecessary or un-indicated PPI use. In this
work, we built a longitudinal cohort of 214 467 United
States veterans that were new users of acid suppression
drugs— histamine H2 receptor antagonists (H2 blockers)
or PPIs—and developed analytic strategies to estimate
the all cause mortality and cause specific mortality
associated with taking PPIs.
Methods
Overall study design and specication of a target
trial
We designed the cohort, exposure definitions,
covariate choices, outcome definitions, and an
WHAT IS ALREADY KNOWN ON THIS TOPIC
Taking proton pump inhibitors (PPIs) is associated with several serious adverse
events and with an increased risk of all cause mortality
WHAT THIS STUDY ADDS
Taking PPIs is associated with an excess of mortality from cardiovascular disease
and chronic kidney disease
Patients without a documented indication for acid suppression drugs have an
excess mortality due to cardiovascular disease, chronic kidney disease, and
upper gastrointestinal cancer
Previous history of cardiovascular disease, chronic kidney disease, and upper
gastrointestinal cancer do not modify the relation between PPI use and the
risk of death due to cardiovascular disease, chronic kidney disease, and upper
gastrointestinal cancer, respectively
1Clinical Epidemiology Center,
Department of Veterans Aairs
St Louis Health Care System,
915 North Grand Boulevard, St
Louis, MO 63106, USA
2Veterans Research and
Education Foundation of St
Louis, St Louis, MO, USA
3Department of Biostatistics,
College for Public Health and
Social Justice, Saint Louis
University, St Louis, MO, USA
4Division of Public Health
Sciences, Department of
Surgery, Washington University
School of Medicine, St Louis,
MO, USA
5Department of Medicine,
Washington University School of
Medicine, St Louis, MO, USA
6Renal Section, Medicine
Service, Department of Veterans
Aairs Saint Louis Health Care
System, St Louis, MO, USA
7Institute for Public Health,
Washington University School of
Medicine, St Louis, MO, USA
Correspondence to: Z Al-Aly
zalaly@gmail.com
(or @zalaly on Twitter;
ORCID 0000-0002-2600-0434)
Cite this as: BMJ 2019;365:l1580
http://dx.doi.org/10.1136/bmj.l1580
Accepted: 20 March 2019
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
2 doi: 10.1136/bmj.l1580 |
BMJ
2019;365:l1580 | thebmj
analytic strategy based on the framework proposed by
Hernán and Robins.5 We emulated a target randomized
controlled trial of the comparative eect of new use of
PPIs versus H2 blockers on the risk of all cause and
cause specific mortality (details of the specified target
trial protocol are presented in supplemental table
1). We then employed causal inference strategies to
estimate the mortality attributable to PPI use (further
described in the methods and in supplemental
table 1).
Cohort design
We selected new users of acid suppression drugs
between 1 July 2002 and 30 June 2004 and followed
them for up to 10 years to examine the associations
between new use of PPIs and causes of death. New use
was defined as having no record of an acid suppression
drug prescription between 1 October 1999 and 30 June
2002.
There were 405 490 new users of PPIs. To reduce the
probability of misclassification, we further selected
from this cohort 201 557 patients who were prescribed
more than a 90 day supply of a PPI in the 180 day
period after new PPI use. Additionally, 24 061 patients
were excluded because they were taking H2 blockers
during the 180 day period, resulting in 177 496 new
users of PPI.
There were 212 735 new users of H2 blockers and
69 731 of them were prescribed more than a 90 day
supply in the 180 day period after new H2 blocker use.
In this group of patients, 9528 were excluded because
they were taking PPIs during the 180 day period,
resulting in 60 203 new users of H2 blockers.
In the two groups of new users of acid suppression
drugs, 234 950 patients had known sex, race, and
date of birth and were alive after 180 days of their
first record of prescription. We selected new users
whose prescribing physician had also prescribed PPIs
or H2 blockers to other new users within the one year
before the cohort patient’s first acid suppressant drug
prescription, to facilitate building an instrumental
variable. We further selected new users whose
prescribing facility and clinic are known, yielding a
final cohort of 214 467 patients (fig 1).
Data sources
We used Department of Veterans Aairs databases in
the study.6 The Department of Veterans Aairs operates
the largest integrated healthcare system in the US—a
veteran specific national health service—to discharged
veterans of the US armed forces.7 The Department of
Veterans Aairs provides a broad range of healthcare at
1400 healthcare facilities, including 143 Department
of Veterans Aairs hospitals and 1241 outpatient sites
of care of varying complexity to over 9 million veterans
(estimated in 2018) enrolled in the Department of
Veterans Aairs healthcare program.7-9 All enrolled
veterans have access to the Department of Veterans
Aairs’s comprehensive medical benefits package
including inpatient hospital care; outpatient services;
preventive, primary, and specialty care; prescriptions;
mental healthcare; home healthcare; geriatric and
extended care; medical equipment; and prosthetics.8 9
We used medical SAS datasets from the
Department of Veterans Aairs corporate data
warehouse, which provided data on inpatient and
outpatient encounters, to obtain information about
demographic characteristics, healthcare system
and clinic encounters, comorbidities, procedures,
and surgeries.10-17 We also collected demographic
information from the Department of Veterans Aairs
Vital Status databases.6 The Department of Veterans
Aairs Managerial Cost Accounting System Laboratory
Results from Department of Veterans Aairs corporate
data warehouse provided laboratory results of cohort
patients.10-14 17-20 The Department of Veterans Aairs
corporate data warehouse Outpatient Pharmacy
domain provided outpatient prescription records and
an identifier for the hospital and Veterans Integrated
Service Network in which the prescription was
provided.4 21-23 Additionally, we used National Death
Index data to collect information about date of death
and the underlying cause of death.24
Outcomes
Study outcomes were specific causes of death based
on national death index underlying cause of death
coded based on ICD-10 (international classification
of diseases, 10th revision).24 25 Causes of death were
categorized into the following categories: circulatory
system diseases; neoplasms; respiratory system
diseases; external causes; endocrine, nutritional,
and metabolism diseases; nervous system diseases;
digestive system diseases; mental and behavioral
disorders; genitourinary system diseases; infectious
and parasitic diseases; and other causes. Based on
results from causes of death, we further defined
subcauses of death within those categories which
exhibited statistical significance and for which there
existed well characterized evidence supporting a
relation between taking PPIs and adverse events which
could be associated with cause specific mortality.3
These subcauses included death due to cardiovascular
diseases, upper gastrointestinal cancer, Clostridium
dicile infections, and chronic kidney disease.3
Exposure
We applied an intention to treat design for new use of
acid suppressant drugs. Intention to treat with PPIs was
defined as a prescription of more than a 90 day supply
of a PPI in the 180 day period since first prescription,
and with no H2 blocker prescriptions within this
period. Intention to treat with H2 blockers, which
served as an active comparator control, was defined as
a prescription of more than a 90 day supply of an H2
blocker in the 180 day period since first prescription,
and with no PPI prescriptions within this period. Use
of an active comparator, compared with a non-user
control, allows for comparison to those who were
prescribed another drug with similar indications; this
approach might increase the overlap of characteristics
between groups, and might reduce the potential for
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
thebmj
BMJ
2019;365:l1580 | doi: 10.1136/bmj.l1580 3
confounding by indication.26 In all analyses, we used
days of supply as an indication of number days with a
prescription.
Covariates
We collected covariates within one year before the
first acid suppressant prescription. We selected
basic demographics, health service utilization
characteristics, and indications for prescription of
acid suppressant drugs based on previous knowledge
including age, sex, race, year of first prescription,
number of outpatient visits, total length of stay
in hospital, level of complexity of the hospital in
which the prescription was provided, type of clinic
in which the prescription was provided, location of
the hospital where the prescription was provided,
gastresophageal reflux disease, upper gastrointestinal
tract bleeding, ulcer disease, H pylori infection,
Barrett’s esophagus, achalasia, stricture, and
esophageal adenocarcinoma.416 17 27 28 Age, number of
outpatient visits and total length of stay hospital were
modeled as cubic spline functions. Level of hospital
complexity was defined as outpatient clinic, medical
center, and healthcare system. Clinic type was defined
as gastroenterology, primary care, and other. Location
of hospital was defined by the Veterans Integrated
Service Network.29 30 To more closely emulate our
target trial, which would have random assignment
of acid suppressant drug, and to reduce bias from
non-random assignment by reducing imbalances in
patient characteristics between PPIs and H2 blockers,
we built a high dimensional propensity score using
pre-exposure information in domains including
outpatient ICD-9 (international classification of
diseases, ninth revision) diagnoses, outpatient Current
Procedural Terminology codes, inpatient ICD-9
diagnoses, inpatient procedures, inpatient surgeries,
and outpatient pharmacy and laboratory records.31
We used physicians’ prescribing preference as an
instrumental variable to reduce the probability that
an observed association (between PPIs and causes of
death) is contributed by unmeasured confounders.32 33
Statistical analyses
Characteristics and outcome events of cohort patients
for the PPI and H2 blocker groups are reported as
number and percentage, mean and standard deviation,
or median and interquartile range, as appropriate.
Kaplan-Meier curves of all cause mortality in new users
of PPIs and H2 blockers are presented.
To more closely mimic a target trial where new use
of PPIs and H2 blockers is randomly assigned, we
used the approach developed by Schneeweiss and
colleagues to generate high dimensional propensity
scores. This approach selects potential confounders
among those included in our data domains based
on their relative association for new use of PPIs
compared with new use of H2 blockers.31 34 We then
used predefined covariables and algorithmically
selected covariates together to generate propensity
scores.35 36 We applied inverse treatment probability
weights based on the propensity scores to the cohort,
which results in a weighted pseudo cohort where
treatment assignment is independent of measured
confounders.37 For the PPIs and H2 blockers groups,
plots of the distributions of the propensity scores and
standardized dierences before and after weighting
are provided in supplemental figures 1a-c.
To reduce bias from unmeasured confounding, we
applied instrumental variable analyses using the two-
stage residual inclusion method to the weighted pseudo
cohort.32 33 38 We used physician-specific prescribing
No PPI use within 180 day period
>90 day supply of H2 blocker in 180 day period
New users of acid suppression drugs
First prescription between 1 July 2002 and 30 June 2004
New users of H2 blockers
69 731
60 203
212 735
New users of proton pump inhibitors (PPIs)
405 490
No H
2
blocker use within 180 day period
177 496
>90 day supply of PPI in 180 day period
201 557
618 225
With known sex, race, and date of birth, and alive aer 180 days of first prescription
234 950
With known prescribing physician and facility data
214 467
Fig1 | Flowchart for cohort building
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
4 doi: 10.1136/bmj.l1580 |
BMJ
2019;365:l1580 | thebmj
preference as the instrumental variable to account for
unmeasured confounders that might not be accounted
for in the high dimensional propensity score, which
could include severity of diseases and other treatment
indications.39 In the first stage, the residual between
the observed and predicted probability of receiving
the assigned treatment given instrumental variable
was computed from logistic regression weighted by
inverse treatment probability weights based on high
dimensional propensity scores. In the second stage,
we used the residual as an independent variable
indicating unmeasured confounders in the inverse
treatment probability weighted cause specific Cox
survival analyses and Fine and Gray survival analyses.
Physician prescription preferences in past patients
has been used as an instrumental variable in the
conduction of pharmacoepidemiologic studies.39 40 41
To assess the strength of our instrumental variable,
we conducted a logistic regression of the odds of being
prescribed PPIs versus H2 blockers. Results suggested
that a 10% increase in a physician prescribing
preference toward prescribing PPIs in past patients
was associated with a 35% (95% confidence interval
35% to 35%) increase in odds of the current patient
being prescribed PPIs compared with H2 blockers after
adjustment for patient characteristics at the time of
prescription. These results suggest that we do not have
a weak instrumental variable. Further discussion on
instrumental variable assumptions can be found in the
supplemental methods.
We also applied negative and positive controls to
examine if potential biases could have modified the
relation between PPI use and cause specific mortality.
We examined acute kidney injury as a positive outcome
control, where previous studies have shown a relation
with PPIs.22 We examined transportation related death
as a negative outcome control following the approach
described by Lipsitch and colleagues, where—based on
current knowledge—we assumed that there should be
no causal relation between PPI use and transportation
related mortality.42 The relation of this exposure-
outcome pair could share the same potential biases
with PPIs and other cause specific deaths including
unmeasured confounders, modeling algorithms,
exposure measurement, outcome ascertainments, and
other biases.42 We also applied death due to peptic ulcer
disease as an additional negative outcome control,
where, based on previous knowledge, we expect that
PPI users should not have a higher risk of death due to
peptic ulcer disease if treatment by indication has been
accounted for; the choice of this outcome control was
motivated by the fact that peptic ulcer disease is an
underlying indication for PPI use and that the relation
between this exposure-outcome pair could have the
same potential bias as PPIs and other outcomes in the
field of treatment by indication.43
In addition to the intention to treat design, since a
proportion of new users of H2 blockers used PPIs later
during follow-up, we conducted a sensitivity analyses
that examined PPI ever-use as a time varying exposure.
We also conducted a duration analysis in new users
of PPIs where cumulative exposure was defined as
the total number of days of PPI supply up to the last
day of prescription. To address immortal time bias,
the T0 in this analysis was set to be the end of the last
prescription.4
To further evaluate cause specific mortality, we
analyzed detailed subcauses of death (as detailed
in the outcomes section). In addition, to remove
potential confounding by gastrointestinal disease
severity, we repeated the analyses in a sub cohort
where patients had no documented gastrointestinal
indications for acid suppression drugs before receipt of
the first prescription. Moreover, we conducted formal
interaction analyses to examine whether the presence
of a baseline comorbid condition modified the relation
between new PPI use and its related cause specific
mortality.
Main results are reported as the event rate per 100
people in the PPIs and H2 blockers groups, and as
estimated excess burden associated with new use
PPI per 1000 people based on estimated cumulative
incidence rate probability at 10 years. To account
for the impact on variance that results from inverse
probability of treatment weighting and two stage
residual inclusion methods,33 44 we analyzed 1000
bootstrapped samples, and report the 2.5 and 97.5
centiles as 95% confidence intervals.
A 95% confidence interval that does not cross 0
for absolute risk and does not cross 1 for ratio was
considered statistically significant. Figure 2 and the
supplemental methods show a detailed description of
the analytic approach. All analyses were done using
SAS Enterprise Guide version 7.1 (SAS Institute, Cary,
NC). The study was approved by the Institutional
Review Board of the Department of Veterans Aairs St
Louis Health Care System, St Louis, MO.
Patient and public involvement
No patients were involved in developing the hypothesis,
the specific aims, or the research questions, nor
were they involved in developing plans for design or
implementation of the study. No patients were involved
in the interpretation or writing up of results. There are
no plans to disseminate the results of the research to
study participants.
Results
Table 1 shows the demographic and health chara-
cteristics of the overall cohort of new users of acid
suppression drugs (n=214 467), by the type of acid
suppressant drug at the time of cohort entry (PPIs,
n=157 625; H2 blockers, n=56 842). In the overall
cohort, the average age was 65.10 (±12.25), 95.93%
were male, and 87.43% were white. Table 2 and
supplemental table 2 show that among new users of
PPIs, rabeprazole 20 mg once a day, omeprazole 20
mg once a day, and rabeprazole 20 mg twice a day
accounted for 58.78%, 21.66%, and 8.41% of all
initial PPI prescriptions, respectively. Over a median
duration of follow up of 10 years (interquartile range
6.95-10.00), there were 80 062 (37.33%) deaths.
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
thebmj
BMJ
2019;365:l1580 | doi: 10.1136/bmj.l1580 5
The most common causes of death were circulatory
system diseases (12.45%), neoplasms (9.72%),
and respiratory system diseases (4.80%). There
were more deaths among patients taking PPIs than
those taking H2 blockers (37.92% and 35.69%,
respectively). Table 1 shows that the median time-to-
death was 4.84 (interquartile range 2.35-7.38) and
4.96 (2.49-7.48) years in the PPIs and H2 blockers
groups, respectively. Kaplan-Meier curves for the PPIs
and H2 blockers new use groups are presented in
supplemental figure 2.
Development of a target trial and application of
positive and negative controls
To estimate the association between exposure to PPIs
and causes of death, we aimed to emulate a target
trial where patients would be randomly assigned to
new use of PPIs or H2 blockers for more than 90 days
(supplemental table 1). We followed the approach
outlined by Hernán and Robins of using big data to
emulate a target trial when a randomized trial is not
available5; we designed a multipronged approach
involving several strategies detailed in supplemental
table 1. To further resolve concerns about spurious
associations, we first applied a positive control to
examine the association between exposure to PPIs
and the risk of acute kidney injury where a priori
knowledge suggests an association is expected.3 4 21 22
Table 3 shows that the results suggested that exposure
to PPIs was associated with an increased burden of
acute kidney injury (number of attributable cases per
1000 PPI users 15.03, 95% confidence interval 1.17 to
29.85). We then tested two negative controls following
the approach outlined by Lipsitch and colleagues.42
We first examined the association between PPI use and
transportation related death where the relation of this
exposure-outcome pair could share the same potential
biases with PPIs and other cause specific deaths. Table
Outpatient data sources
Diagnosis Procedure Laboratory Pharmacy
Inpatient data sources
Diagnosis Procedure Laboratory Pharmacy
Process
Dataset
Result
Cohort
HDPS
2 stage residual inclusion
Result
COHORT
500 components with the highest relative risk (RR)
between confounder and exposure Predefined variables
Ever Sometime Oen
Inverse probability of treatment weighting
PSEUDO COHORT
Estimated probability
Excess burden
Estimated survival
probability when PPI = 0
Estimated survival
probability when PPI = 1
Fine and Gray Cox
Cause specific hazard ratio
Logistic regression
Treatment = IV
Instrumental
variables (IV)
Residual between observed
and predicted treatment
Bootstrapping
Physician prescribing preference
Subdistribution hazard ratio
and cumulative incidence rate
Key
Fig2 | Flowchart for analytic approach
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
6 doi: 10.1136/bmj.l1580 |
BMJ
2019;365:l1580 | thebmj
Characteristic Overall PPIs H2 blockers
Total 21 4467 157 625 (73.50) 56 842 (26.50)
Mean (SD) age (years) 65.10 (12.25) 65.51 (12.14) 63.97 (12.46)
Sex:
Men 20 5748 (95.93) 151 399 (96.05) 54 349 (95.61)
Women 8719 (4.07) 6226 (3.95) 2493 (4.39)
Race:
White 187 519 (87.43) 138 967 (88.16) 48 552 (85.42)
Black 20 089 (9.37) 14 193 (9.00) 5896 (10.37)
Other 6859 (3.20) 4465 (2.83) 2394 (4.21)
Admitted to hospital in past year 20 794 (9.70) 15 221 (9.66) 5573 (9.80)
Median (IQR) length of stay among inpatients (days) 7 (4 to 13) 7 (4 to 13) 7 (4 to 14)
Median (IQR) no of outpatient visits 4 (1 to 10) 4 (1 to 10) 5 (2 to 11)
Disease:
Diabetes mellitus 48 869 (22.79) 35 777 (22.70) 13 092 (23.03)
Hypertension 116 536 (54.34) 85 136 (54.01) 31 400 (55.24)
Chronic lung disease 29 517 (13.76) 21 588 (13.70) 7929 (13.95)
Peripheral artery disease 2475 (1.15) 1745 (1.11) 720 (1.28)
Cardiovascular disease 54 122 (25.24) 40 641 (25.78) 13 481 (23.72)
Dementia 4747 (2.21) 3420 (2.17) 1327 (2.33)
Hyperlipidemia 90 812 (42.34) 66 613 (42.26) 24 199 (42.57)
Hepatitis C 1953 (0.91) 1403 (0.89) 550 (0.97)
HIV 57 (0.03) 37 (0.02) 20 (0.04)
Cancer 9738 (4.54) 7465 (4.74) 2273 (4.00)
Any documented gastrointestinal indication for acid suppression
drugs
98 090 (45.74) 76 581 (48.58) 21 509 (37.84)
Gastresophageal reflux disease 83 904 (39.12) 64 602 (40.98) 19 302 (33.96)
Upper gastrointestinal tract bleeding 3356 (1.56) 3072 (1.95) 284 (0.50)
Ulcer disease 13 856 (6.46) 11 585 (7.35) 2271 (4.00)
H pylori infection 809 (0.38) 746 (0.47) 63 (0.11)
Barrett’s esophagus 597 (0.28) 588 (0.37) 9 (0.02)
Achalasia 66 (0.03) 60 (0.04) 6 (0.01)
Stricture 1288 (0.60) 1202 (0.76) 86 (0.15)
Esophageal adenocarcinoma 42 (0.02) 36 (0.02) 6 (0.01)
Drugs:
Angiotensin converting enzyme inhibitors or angiotensin receptor
blockers
84 832 (39.34) 62 306 (39.53) 22 076 (38.84)
Statins 86 546 (40.35) 64 440 (40.88) 22 106 (38.89)
Nonsteroidal anti-inflammatory drugs 56 346 (26.27) 38 945 (24.71) 17 401 (30.61)
Mean (SD) estimated glomerular ltration rate (mL/min/1.73m2)73.43 (21.03) 72.88 (21.14) 74.84 (20.69)
Median (IQR) HbA1C (%) 6.2 (5.6-7.3) 6.2 (5.6-7.2) 6.2 (5.6-7.3)
Mean (SD) systolic blood pressure (mmHg) 137.19 (19.64) 137.09 (19.64) 137.49 (19.63)
Mean (SD) diastolic blood pressure (mmHg) 76.17 (11.79) 75.97 (11.80) 76.74 (11.76)
Median (IQR) high density lipoprotein (mg/dL) 41.87(35.00-50.00) 41.60 (35.00-50.00) 42.00 (35.00-50.00)
Median (IQR) low density lipoprotein (mg/dL) 107.0 (86.0-131.4) 106.3 (85.0-131.0) 109.6 (88.0-133.0)
Smoking status:
Current 41 809 (19.49) 28 928 (18.35) 12 881 (22.66)
Former 44 216 (20.62) 34 247 (21.73) 9969 (17.54)
Never 128 442 (59.89) 94 450 (59.92) 33 992 (59.80)
Median (IQR) washout period (days) 728 (174-1584) 675 (158-1565) 889 (230-1631)
Median (IQR) years of follow-up 10.00 (6.59-10.00) 10.00 (6.45-10.00) 10.00 (6.95-10.00)
Median (IQR) days of PPI prescription during follow-up 1278 (354-2591) 1682 (682-2879) 123 (0-1288)
Median (IQR) days of H2 blocker prescription during follow-up 0 (0-270) 0 (0-0) 597 (270-1299)
Median (IQR) time-to-death (years) 4.87 (2.39-7.40) 4.84 (2.35-7.38) 4.96 (2.49-7.48)
All cause mortality 80 062 (37.33) 59 776 (37.92) 20 286 (35.69)
Cause specic mortality:
Circulatory system diseases 26 711 (12.45) 19 923 (12.64) 6788 (11.94)
Neoplasms 20 847 (9.72) 15 529 (9.85) 5318 (9.36)
Respiratory system diseases 10 294 (4.80) 7593 (4.82) 2701 (4.75)
External causes 3406 (1.59) 2483 (1.58) 923 (1.62)
Endocrine, nutritional, and metabolism diseases 3581 (1.67) 2628 (1.67) 953 (1.68)
Nervous system diseases 3391 (1.58) 2574 (1.63) 817 (1.44)
Digestive system diseases 3299 (1.54) 2552 (1.62) 747 (1.31)
Mental and behavioral disorders 2114 (0.99) 1664 (1.06) 450 (0.79)
Genitourinary system diseases 2373 (1.11) 1827 (1.16) 546 (0.96)
Infectious and parasitic diseases 2114 (0.99) 1664 (1.06) 450 (0.79)
Symptoms, signs, and abnormal clinical or laboratory results 789 (0.37) 588 (0.37) 201 (0.35)
Table1 | Demographic and health characteristics of overall cohort and by type of acid suppressant drug. Values are
numbers (percentages) unless stated otherwise
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
thebmj
BMJ
2019;365:l1580 | doi: 10.1136/bmj.l1580 7
3 shows that PPI exposure was not associated with
excess burden of transportation related death (−0.21,
−2.96 to 2.48). To verify that indication bias was
accounted for, we then estimated the mortality due to
peptic ulcer disease—an underlying indication for PPI
use—as an additional negative outcome control where
the relation between this exposure-outcome pair might
share the same potential biases as PPIs and outcomes
in the area of treatment by indication; the results
showed no excess of peptic ulcer disease related death
(−0.46, −2.43 to 0.27) suggesting that indication bias
might have been accounted for.
Causes of death among patients taking PPIs
We then used our analytic approach to estimate the
excess cause specific mortality burden associated with
new use of PPIs. Details of the model construction
are presented in figure 2 and supplemental table 1.
Our results suggest that over the course of 10 years of
follow-up there were 45.52 (95% confidence interval
28.20 to 61.40) excess deaths per 1000 PPI users.
Table 4 shows that over the follow-up period of 10
years, causes of death which exhibited significant
excess mortality associated with PPI use included
circulatory system diseases (number of attributable
deaths per 1000 PPI users 17.47, 95% confidence
interval 5.47 to 28.80), neoplasms (12.94, 1.24 to
24.28), genitourinary system diseases (6.25, 3.22 to
9.24), and infectious and parasitic diseases (4.20, 1.57
to 7.02). Notably, taking PPIs was not associated with
increased mortality due to digestive system diseases
(0.43, −3.72 to 4.07). The results were consistent in
sensitivity analyses where exposure was treated as
time varying (supplemental table 3).
Table 5 shows that in analyses evaluating the
relation between cumulative duration of exposure and
the risks of all cause and cause specific mortality, there
was a graded relation between duration of exposure
and risks of all cause mortality, death due to circulatory
system diseases, neoplasms, and genitourinary
system diseases. The risk of death due to infectious
and parasitic diseases was not related to duration of
exposure.
Subcauses of death
Because our results showed excess deaths due to
circulatory system diseases, neoplasms, genitourinary
system diseases, and infectious and parasitic diseases,
we further examined excess death in subcauses of
these conditions which could be mapped to adverse
events of PPIs, which are well characterized. These
subcauses included death due cardiovascular diseases,
upper gastrointestinal cancer, Clostridium dicile
infections, and chronic kidney disease.3 Table 6 shows
that we observed excess deaths due to cardiovascular
disease (number of attributable deaths per 1000 PPI
users 15.48, 95% confidence interval 5.02 to 25.19)
and chronic kidney disease (4.19, 1.56 to 6.58), but
not due to upper gastrointestinal cancer (1.72, −0.15
to 3.74) or Clostridium dicile infections (0.65, −0.18
to 1.70).
We examined the association between PPI use and
the four subcauses in patients without documented
gastrointestinal indication for acid suppression drugs.
Table 7 shows that there is an excess of cause specific
mortality associated with taking PPIs for cardiovascular
diseases (number of attributable deaths per 1000 PPI
users 22.91, 95% confidence interval 11.89 to 33.57),
upper gastrointestinal cancer (3.12, 0.91 to 5.44), and
chronic kidney disease (4.74, 1.53 to 8.05).
Formal interaction analyses were undertaken
to evaluate whether the presence of baseline
cardiovascular disease, upper gastrointestinal cancer,
or chronic kidney disease modified the association
between PPI use and the related subcauses of death.
Results suggest no significant interaction for death due
to cardiovascular diseases (P=0.22 for interaction),
upper gastrointestinal cancer (P=0.54 for interaction),
and chronic kidney disease (P=0.10 for interaction).
Interaction analyses between PPI use and history of
Clostridium dicile infections could not be conducted
because no patients in the H2 blockers group with
Table2 | Top three proton pump inhibitor (PPI) and H2 blocker prescriptions
Rank
PPIs H2 blockers
Prescription N (%) Prescription N (%)
1Rabeprazole 20 mg once a day 92 650 (58.78) Ranitidine 150 mg twice a day 42 349 (74.50)
2Omeprazole 20 mg once a day 34 149 (21.66) Ranitidine 150 mg once a day 82 25 (14.47)
3Rabeprazole 20 mg twice a day 13 250 (8.41) Ranitidine 300 mg twice a day 3156 (5.55)
Table1 | Continued
Characteristic Overall PPIs H2 blockers
Musculoskeletal system diseases 342 (0.16) 267 (0.17) 75 (0.13)
Blood diseases 287 (0.13) 223 (0.14) 64 (0.11)
Skin and subcutaneous diseases 113 (0.05) 75 (0.05) 38 (0.07)
Congenital malformations 44 (0.02) 33 (0.02) 11 (0.02)
Ear and mastoid diseases 4 (0.00) 3 (0.00) 1 (0.00)
Eye diseases 2 (0.00) 2 (0.00) 0 (0.00)
Nonspecic* 21 (0.01) 18 (0.01) 3 (0.01)
PPIs=proton pump inhibitors; IQR=interquartile range
*Underlying cause of death missing
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
8 doi: 10.1136/bmj.l1580 |
BMJ
2019;365:l1580 | thebmj
history of Clostridium dicile infections experienced
death due to Clostridium dicile.
Discussion
We examined the causes of death associated with new
use of PPIs in a longitudinal observational cohort of US
veterans. Overall, there were 45.20 attributable deaths
per 1000 PPI users; 38.65% were related to circulatory
system diseases, 28.63% to neoplasms,13.83% to
genitourinary system diseases, and 9.29% to infectious
and parasitic diseases. Analyses by subcauses of death
suggest that new use of PPIs is associated with an
excess of mortality from cardiovascular disease and
chronic kidney disease; this pattern was consistent
in analyses considering those without documented
indication for acid suppression drugs. Increased risk
of death due to upper gastrointestinal cancer was
more evident in those without documented indication
for acid suppression drugs. The risk of cause specific
mortality was not modified by the presence of previous
history of cardiovascular disease, chronic kidney
disease, or upper gastrointestinal cancer.
Contextual evaluation of ndings
PPIs are often used without indication and for
much longer than needed.45-52 Evidence from the
past several years suggests that taking PPIs is
associated with serious adverse events including
cardiovascular disease, pneumonia, osteoporotic
fractures, Clostridium dicile infections, acute kidney
injury, chronic kidney disease, dementia, and upper
gastrointestinal cancer.3 We previously described an
excess risk of all cause mortality among PPI users.4
In this report, we used advanced causal inference
methods to identify and estimate the excess of cause
specific mortality associated with taking PPIs. Our
findings suggest that although PPI use might be
associated with many serious adverse events, excess
mortality was only mapped to a few specific causes
including cardiovascular disease, chronic kidney
disease, and upper gastrointestinal cancer. Notably,
excess burden was present in those without underlying
documented indications for PPI use, that is, patients
who may not benefit from PPI exposure. Our results,
however, should not preclude prescription and use
Table3 | Positive and negative outcome controls
Outcome
Event rate per 100 (95% CI) Excess burden per 1000
(95% CI)
Hazard ratio (95% CI)
PPIs H2 blockers Fine and Gray Cox
Acute kidney injury* 11.34
(10.25 to 12.55)
9.83
(9.51 to 10.20)
15.03
(1.17 to 29.85)
1.16
(1.01 to 1.33)
1.20
(1.05 to 1.38)
Transportation related
death†
0.29
(0.23 to 0.44)
0.31
(0.19 to 0.54)
−0.21
(−2.96 to 2.48)
0.93
(0.45 to 2.34)
0.96
(0.46 to 2.44)
Peptic ulcer disease
related death‡
0.04
(0.03 to 0.06)
0.08
(0.03 to 0.28)
−0.46
(−2.43 to 0.27)
0.45
(0.11 to 1.91)
0.47
(0.12 to 1.99)
*Positive outcome control. First acute kidney injury during follow up dened by ICD-9 584.
†Negative outcome control. Dened by ICD-10 V00-V99.
‡Negative outcome control. Dened by ICD-10 K20, K211, K226, K250-K289
Table4 | Causes of death associated with proton pump inhibitor (PPI) use during 10 years of follow-up
Cause of death
ICD-10 cause of
death
Event rate per 100 (95% CI) Excess burden per
1000 (95% CI)
Hazard ratio (95% CI)
PPIs H2 blockers Fine and Gray Cox
All Any 38.74
(38.19 to 39.31)
34.22
(33.04 to 35.46)
45.20
(28.20 to 61.40)
1.17
(1.10 to 1.24)
1.17
(1.10 to 1.24)
Circulatory system
diseases
I00-I99 13.10
(12.73 to 13.49)
11.35
(10.54 to 12.25)
17.47
(5.47 to 28.80)
1.17
(1.05 to 1.29)
1.19
(1.07 to 1.33)
Neoplasms C00-D49 10.20
(9.81 to 10.64)
8.90
(8.16 to 9.75)
12.94
(1.24 to 24.28)
1.15
(1.01 to 1.31)
1.18
(1.03 to 1.35)
Respiratory system
diseases
J00-J99 4.87
(4.68 to 5.07)
4.65
(4.23 to 5.19)
2.25
(−4.84 to 8.14)
1.05
(0.90 to 1.2)
1.09
(0.94 to 1.24)
External causes V00-Y99 1.50
(1.38 to 1.66)
1.92
(1.48 to 2.66)
−4.17
(−12.7 to 1.85)
0.78
(0.52 to 1.12)
0.81
(0.54 to 1.18)
Endocrine, nutritional,
and metabolism
diseases
E00-E89 1.61
(1.51 to 1.71)
1.82
(1.53 to 2.19)
−2.11
(−6.51 to 1.63)
0.88
(0.70 to 1.11)
0.91
(0.72 to 1.14)
Nervous system
diseases
G00-G99 1.68
(1.53 to 1.87)
1.39
(1.16 to 1.72)
2.84
(−1.61 to 6.83)
1.21
(0.91 to 1.59)
1.25
(0.94 to 1.67)
Digestive system
diseases
K00-K99 1.54
(1.44 to 1.66)
1.50
(1.24 to 1.83)
0.43
(−3.72 to 4.07)
1.03
(0.79 to 1.33)
1.06
(0.82 to 1.37)
Mental and
behavioral disorders
F00-F99 1.22
(1.11 to 1.36)
1.04
(0.85 to 1.27)
1.82
(−1.45 to 4.96)
1.18
(0.89 to 1.58)
1.23
(0.93 to 1.66)
Genitourinary system
diseases
N00-N99 1.35
(1.21 to 1.54)
0.72
(0.59 to 0.91)
6.25
(3.22 to 9.24)
1.87
(1.35 to 2.58)
1.94
(1.41 to 2.68)
Infectious and
parasitic diseases
A00-B99 1.12
(1.01 to 1.26)
0.70
(0.55 to 0.88)
4.20
(1.57 to 7.02)
1.61
(1.18 to 2.28)
1.66
(1.21 to 2.35)
Other causes* D50-D89, H00-H95,
L00-M99, O00-R99
0.81
(0.72 to 0.95)
0.60
(0.44 to 0.85)
2.11
(−1.26 to 5.02)
1.35
(0.85 to 2.16)
1.40
(0.88 to 2.23)
*Death from symptoms, signs, and abnormal clinical or laboratory result; musculoskeletal system diseases; blood diseases; skin and subcutaneous
diseases; congenital malformations; ear and mastoid diseases; eye diseases; and nonspecic death
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
thebmj
BMJ
2019;365:l1580 | doi: 10.1136/bmj.l1580 9
of PPIs where medically indicated; nevertheless, the
findings emphasize the need to promote awareness of
potential adverse events of long term PPI use, for better
pharmacovigilance, and the need to limit prescription
of PPIs to patients where the benefits outweigh potential
risks.1 Identification of those at high risk of adverse
events attributable to taking PPIs is an important
knowledge gap and could inform risk stratification and
risk mitigation strategies. Future research should also
investigate the best way to implement deprescription
programs to reduce the unnecessary or un-indicated
use of PPIs.51 53
We designed this study to evaluate the research
question using a cohort from a time when the
prevalence of PPI prescriptions was lower; and the
doses prescribed were lower. Over 80% of new users of
PPIs in our cohort had an initial dose that is equivalent
to over-the-counter doses (table 2). That and the
findings of increased risk of cause specific mortality
with prolonged duration of exposure suggests
that prescription PPI use should be limited to well
documented indications (where patients may derive
benefit), and for a well defined duration—defined by
the underlying indication. Over-the-counter use of PPIs
should only be for a brief duration of time (generally
not to exceed 14 days).54 Eorts to target and reduce
prolonged use of prescription PPIs without indications
and to curtail extended use of over-the-counter PPIs
might be a good approach.
Evidence suggests that taking PPIs is associated
with an increased risk of cardiovascular disease and
chronic kidney disease1 21 22; the finding in our study
that taking PPIs is associated with an excess mortality
due to cardiovascular disease and chronic kidney
disease suggests that beyond the occurrence of the
adverse events, excess PPI use is linked to loss of life.
Furthermore, the results of formal interaction analyses
show that the relation between taking PPIs and cause
specific mortality (death due cardiovascular disease and
chronic kidney disease) is not modified by the presence
of related baseline comorbid condition, suggesting
that the presence of baseline cardiovascular disease
or chronic kidney disease does not increase the risk of
PPI related cause specific mortality. The pathways in
which exposure to PPIs leads to excess cause specific
mortality is likely mediated by either worsening of the
underlying baseline disease (cardiovascular disease or
chronic kidney disease) or the occurrence of de novo
disease (cardiovascular disease or chronic kidney
disease) which would then heighten the risk of cause
specific mortality. Experimental evidence from Yepuri
and colleagues suggested a “unifying mechanism”
for the association of PPI use with an increased
risk of cardiovascular and kidney morbidity and
mortality.55 The investigators reported that long term
exposure to PPIs blunted lysosomal acidification and
proteostasis in endothelial cells, increased oxidative
stress, impaired endothelial function, and accelerated
human endothelial senescence by reducing telomere
length.55 Our findings of excess mortality due to
cardiovascular disease and chronic kidney disease are
consistent with this proposed biologic mechanism,
Table5 | Causes of death associated with cumulative exposure to proton pump inhibitors (PPIs) during 10 years of
follow-up. Values are hazard ratios (95% condence intervals)
Duration (days) Deaths
Circulatory
system diseases* Neoplasms†
Genitourinary
system diseases‡
Infectious and
parasitic diseases§
0-120 1 (ref) 1 (ref) 1 (ref) 1 (ref ) 1 (ref)
121-240 1.23 (1.12 to 1.34) 1.13 (0.97 to 1.31) 1.09 (0.93 to 1.29) 1.03 (0.62 to 1.71) 0.90 (0.57 to 1.43)
241-360 1.47 (1.34 to 1.60) 1.34 (1.15 to 1.55) 1.19 (1.01 to 1.39) 1.20 (0.72 to 1.99) 0.94 (0.59 to 1.49)
361-480 1.63 (1.49 to 1.79) 1.37 (1.17 to 1.59) 1.25 (1.06 to 1.48) 1.30 (0.77 to 2.18) 0.96 (0.60 to 1.55)
481-600 1.71 (1.56 to 1.87) 1.46 (1.25 to 1.70) 1.25 (1.06 to 1.48) 1.48 (0.88 to 2.48) 0.90 (0.56 to 1.45)
P value for trend <0.001 <0.001 0.002 0.005 0.93
Analysis conducted in new users of PPIs. T0 was set to be the end of the last PPI prescription.
*ICD10 I00-I99
†ICD10 C00-D49
‡ICD10 N00-N99
§ICD10 A00-B99
Table6 | Subcauses of death associated with proton pump inhibitor (PPI) use during 10 years of follow-up
Cause of death Subcause of death
Event rate per 100 (95% CI) Excess burden
per 1000
(95% CI)
Hazard ratio (95%CI)
PPIs H2 blockers Fine and Gray Cox
Circulatory system
diseases
Cardiovascular disease* 8.87
(8.54 to 9.23)
7.33
(6.65 to 8.08)
15.48
(5.02 to 25.19)
1.22
(1.07 to 1.40)
1.25
(1.10 to 1.44)
Neoplasms Upper gastrointestinal
cancer†
0.63
(0.57 to 0.72)
0.46
(0.34 to 0.6)
1.72
(−0.15 to 3.74)
1.38
(0.97 to 2.09)
1.41
(1.00 to 2.15)
Genitourinary system
diseases
Chronic kidney disease‡ 0.86
(0.75 to 1.01)
0.44
(0.34 to 0.60)
4.19
(1.56 to 6.58)
1.95
(1.26 to 2.89)
2.02
(1.31 to 3.00)
Infectious and
parasitic diseases
Clostridium dicile
infections§
0.12
(0.09 to 0.21)
0.06
(0.03 to 0.12)
0.65
(−0.18 to 1.70)
2.09
(0.84 to 5.73)
2.18
(0.86 to 6.04)
Subcauses are subcategories of causes of death which exhibited signicant association with PPI use and for which there was well characterized evidence
supporting a relation between taking PPIs and adverse events which may be associated with cause specic mortality.
*ICD10 I21-I24.0, I24.2-I25.2, I25.8-I25.9, I60-I69
†ICD10 C15.0-C17.0, D00.1-D00.2, D13.0-D13.2, D37.1
‡ICD10 N18-N19
§ICD10 A04.7
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
10 doi: 10.1136/bmj.l1580 |
BMJ
2019;365:l1580 | thebmj
but do not exclude the possible contributions of other
mechanisms including those mediated by activation of
the heme oxygenase-1 enzyme in endothelial cells and
microbiome perturbation.56-58
Analyses by subcauses within the death due to
neoplasm category suggested that mortality due to
upper gastrointestinal cancer was more evident in
those without gastrointestinal indication for use of
acid suppression drugs, likely a manifestation of
lower baseline risk. The findings are consistent with
emerging evidence suggesting that long term exposure
to PPIs increases the risk of gastric malignancy.59-67 A
recent study by Cheung and colleagues examined the
risk of gastric cancer in a cohort of 63 397 patients
and reported excess burden among long term users
of PPI.66 The investigators reported that the adjusted
absolute risk dierence for PPI use versus non-PPI use
for excess gastric cancer was 4.29 (95% confidence
interval 1.25 to 9.54) per 10 000 person years.66
Wan and colleagues conducted a meta-analysis of
926 386 patients and found that long term PPI use
was associated with a twofold risk of gastric cancer
(odds ratio 2.10, 95% confidence interval 1.10 to
3.09).64 The underlying mechanism(s) by which long
term exposure to PPIs might increase the risk of gastric
cancer is hypothesized to involve gastrin mediated
trophic stimulus of gastric mucosa, gastric atrophy,
and alteration of gut microbiota and gastric mucosal
immunology.58 64 68
In our analyses, we observed a graded relation
between duration of exposure and the risk of mortality
due to chronic conditions including circulatory system
diseases, neoplasms, and genitourinary system
diseases. Notably, there was no relation between the
duration of exposure and the risk of death due to
infectious and parasitic diseases, most likely due to the
acuteness of the clinical condition where the relation
might be idiosyncratic.69
Given the observational nature of this study, we
carefully considered potential biases which could
result in false relations and designed a multipronged
causal inference analytic approach to emulate a target
randomized trial that would answer the research
question. We considered the following strategies: first,
we employed a new user design to enhance balance in
comparison groups based on pretreatment status, and
an active comparator control to reduce the chance of
confounding by indication. Second, to avoid capturing
reverse causation and to ensure the temporal direction
between exposure and diseases that lead to cause
specific mortality, we removed all events which
occurred within 180 days after first exposure. Third,
we applied inverse probability of treatment weighting
based on high dimensional propensity scores to create
a pseudo cohort whose treatment assignment was
independent of measured confounders.70 Fourth, to
reduce the probability that an observed association
between PPIs and causes of death is contributed
by unmeasured confounding, we employed an
instrumental variable method.71 Results from two
negative controls which showed no association
between PPI use and transportation mortality, and no
association between PPI use and death due to peptic
ulcer disease, lessen concerns about unmeasured
confounding and other biases. In particular, results
from our negative control analysis of death due to
peptic ulcer disease are consistent with those of
multiple randomized controlled trials.43 Furthermore,
the finding that PPI use was not associated with excess
mortality due to digestive system diseases further lends
validity to our approach. Taken together, the findings
suggest that subjecting our approach to the scrutinous
application of negative controls yielded results
consistent with a priori expectations and results from
randomized controlled trials, suggesting no observable
biases in analyses of established relations.
Strengths and limitations of study
The study has several limitations. The cohort included
US veterans who were mostly older, white, and
male, which might limit the generalizability of the
study results to a broader population. Although our
application of several inclusion and exclusion criteria
could have introduced selection bias, these criteria are
needed for more accurate cohort definition (and new
user definition) which will optimize the successful
emulation of a target trial. In our analyses, we defined
drug exposure based on Department of Veterans
Aairs prescription records and by days of supply
which might not necessarily be equivalent with days
of use since patients can obtain PPIs through over-the-
counter purchase, by other means, or may not adhere
Table7 | Subcauses of death associated with taking proton pump inhibitors (PPIs) in patients without indication for acid
suppression drugs at baseline (n=116 377)
Cause of death Subcause of death
Event rate per 100 (95% CI) Excess burden per
1000 (95% CI)
Hazard ratio (95%CI)
PPIs H2 blockers Fine and Gray Cox
Circulatory system
diseases
Cardiovascular
disease*
10.31
(9.84 to 10.79)
8.02
(7.38 to 8.71)
22.91
(11.89 to 33.57)
1.30
(1.15 to 1.48)
1.34
(1.19 to 1.53)
Neoplasms Upper gastrointestinal
cancer†
0.69
(0.59 to 0.85)
0.38
(0.27 to 0.52)
3.12
(0.91 to 5.44)
1.83
(1.18 to 2.99)
1.89
(1.21 to 3.09)
Genitourinary system
diseases
Chronic kidney
disease‡
1.08
(0.93 to 1.31)
0.61
(0.48 to 0.80)
4.74
(1.53 to 8.05)
1.78
(1.19 to 2.66)
1.86
(1.24 to 2.80)
Infectious and
parasitic diseases
Clostridium dicile
infections§
0.12
(0.09 to 0.18)
0.07
(0.03 to 0.15)
0.49
(−0.50 to 1.38)
1.71
(0.65 to 4.92)
1.78
(0.68 to 5.08)
*ICD10 I21-I24.0, I24.2-I25.2, I25.8-I25.9, I60-I69
†ICD10 C15.0-C17.0, D00.1-D00.2, D13.0-D13.2, D37.1
‡ICD10 N18-N19
§ICD10 A04.7
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
thebmj
BMJ
2019;365:l1580 | doi: 10.1136/bmj.l1580 11
to a Department of Veterans Aairs prescription.
Although we applied high dimensional propensity
scores, and used instrumental variable methods, our
overall approach is inherently limited by the validity
of the underlying assumptions.40 Furthermore, while
we used an active comparator design, and developed
strategies to reduce confounding, channeling bias,
and other forms of residual confounding might not be
completely eliminated.72 To obtain sucient follow-
up, we built a cohort of new users in 2003; as PPI
use became more prevalent over the last decade, we
anticipate that the proportion of patients where the risk
of taking PPIs might outweigh potential benefit could
have increased over time and as such our results could
have underestimated the true burden of cause specific
mortality. The study has several strengths, including
the use of national large scale data from a network
of integrated health systems, which were captured
during routine medical care that minimizes selection
bias. We employed a new user design with an active
comparator control from a time when H2 blockers were
commonly used, applied instrumental variable and
high dimensional propensity score method, and tested
positive and negative controls to more accurately
estimate the burden of cause specific mortality.
Conclusions
The results show a consistent excess of cause specific
mortality even among patients without documented
gastrointestinal indications for acid suppression
drugs—an alarming finding which might help guide the
design and implementation of deprescription programs
to reduce the number of unnecessary or un-indicated
PPI prescriptions. The evidence that mortality due to
cardiovascular disease, chronic kidney disease, and
upper gastrointestinal cancer was not modified by the
presence of baseline cardiovascular disease, chronic
kidney disease, or upper gastrointestinal cancer,
respectively, suggests the need for heightened vigilance
among those with and—with equal importance—
those at risk of these conditions. The evidence from
all available studies suggests that long term PPI use is
associated with serious adverse events, including an
increased risk of all cause mortality, and our results
specifically suggest an increased mortality due to
cardiovascular disease, chronic kidney disease, and
upper gastrointestinal cancer. Because of the high
prevalence of PPI use, the findings have public health
implications and underscore the important message
that PPIs should be used only when medically indicated
and for the minimum duration necessary.
The contents do not represent the views of the United States
Department of Veterans Aairs or the United States Government.
Contributors: YX, BB, TL, HX, YY, and ZAA developed the research
area and study design. YX and BB acquired the data. YX, BB, TL, HX,
YY, and ZAA analyzed and interpreted the data. YX and BB performed
the statistical analysis. ZAA supervised and mentored the team. Each
author contributed important intellectual content during manuscript
draing or revision and accepts accountability for the overall work
by ensuring that questions pertaining to the accuracy or integrity of
any portion of the work are appropriately investigated and resolved.
All authors had full access to the data in the study and can take
responsibility for the integrity of the data and the accuracy of the data
analysis. ZAA is the guarantor. The corresponding author attests that
all listed authors meet authorship criteria and that no others meeting
the criteria have been omitted.
Funding: This research was funded by the United States Department
of Veterans Aairs and the Institute for Public Health at Washington
University in St Louis, MO, USA (ZAA). The funders of this study had
no role in study design; collection, analysis, and interpretation of
data; writing the report; and the decision to submit the report for
publication.
Competing interests: All authors have completed the ICMJE uniform
disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no
support from any organization for the submitted work; no nancial
relationships with any organizations that might have an interest in the
submitted work in the previous three years; no other relationships or
activities that could appear to have influenced the submitted work.
Ethical approval: This research project (study# 1163689) was
reviewed and approved by the Institutional Review Board of the
Department of Veterans Aairs Saint Louis Health Care System.
Data sharing: All data are available through the United States
Department of Veterans Aairs.
The lead author (ZAA) arms that the manuscript is an honest,
accurate, and transparent account of the study being reported; that
no important aspects of the study have been omitted; and that any
discrepancies from the study as planned have been explained.
The Corresponding Author has the right to grant on behalf of all
authors and does grant on behalf of all authors, a non-exclusive
license (the corresponding author is a US Government employee) on a
worldwide basis to the BMJ Publishing Group Ltd to permit this article
(if accepted) to be published in BMJ editions and any other BMJPGL
products and sublicences such use and exploit all subsidiary rights, as
set out in our license.”
This is an Open Access article distributed in accordance with the
Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license,
which permits others to distribute, remix, adapt, build upon this work
non-commercially, and license their derivative works on dierent
terms, provided the original work is properly cited and the use is non-
commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
1 Schoenfeld AJ, Grady D. Adverse Eects Associated With Proton
Pump Inhibitors. JAMA Intern Med2016;176:172-4. doi:10.1001/
jamainternmed.2015.7927
2 Xie Y, Bowe B, Li T, Xian H, Balasubramanian S, Al-Aly Z. Proton Pump
Inhibitors and Risk of Incident CKD and Progression to ESRD. J Am Soc
Nephrol2016;27:3153-63. doi:10.1681/ASN.2015121377
3 Li T, Xie Y, Al-Aly Z. The association of proton pump inhibitors
and chronic kidney disease: cause or confounding?Curr
Opin Nephrol Hypertens2018;27:182-7. doi:10.1097/
MNH.0000000000000406
4 Xie Y, Bowe B, Li T, Xian H, Yan Y, Al-Aly Z. Risk of death among users
of Proton Pump Inhibitors: a longitudinal observational cohort
study of United States veterans. BMJ Open2017;7:e015735.
doi:10.1136/bmjopen-2016-015735
5 Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial When
a Randomized Trial Is Not Available. Am J Epidemiol2016;183:758-
64. doi:10.1093/aje/kwv254
6 VIReC Research User Guide: Veterans Health Administration Decision
Support System Clinical National Data Extracts H, IL, U.S. Department
of Veterans Aairs. VA Information Resource Center September 2009.
https://www.virec.research.va.gov/Resources/RUGs.asp
7 Oliver A. The Veterans Health Administration: an American
success story?Milbank Q2007;85:5-35. doi:10.1111/j.1468-
0009.2007.00475.x
8 Medical Benets Package VA. Secondary Medical Benets Package
2019. https://www.va.gov/healthbenets/resources/publications/
hbco/hbco_medical_benets_package.asp.
9 VA. National Center for Veterans Analysis and Statistics. Secondary
National Center for Veterans Analysis and Statistics 2019. https://
www.va.gov/vetdata/report.asp.
10 Bowe B, Xie Y, Li T, Yan Y, Xian H, Al-Aly Z. Particulate Matter Air
Pollution and the Risk of Incident CKD and Progression to ESRD. J Am
Soc Nephrol2018;29:218-30. doi:10.1681/ASN.2017030253
11 Bowe B, Xie Y, Xian H, Balasubramanian S, Al-Aly Z. Low levels of
high-density lipoprotein cholesterol increase the risk of incident
kidney disease and its progression. Kidney Int2016;89:886-96.
doi:10.1016/j.kint.2015.12.034
12 Bowe B, Xie Y, Xian H, Balasubramanian S, Zayed MA, Al-Aly Z. High
Density Lipoprotein Cholesterol and the Risk of All-Cause Mortality
among U.S. Veterans. Clin J Am Soc Nephrol2016;11:1784-93.
doi:10.2215/CJN.00730116
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
12 doi: 10.1136/bmj.l1580 |
BMJ
2019;365:l1580 | thebmj
13 Bowe B, Xie Y, Xian H, Li T, Al-Aly Z. Association between Monocyte
Count and Risk of Incident CKD and Progression to ESRD. Clin J Am
Soc Nephrol2017;12:603-13. doi:10.2215/CJN.09710916
14 Bowe B, Xie Y, Xian H, Lian M, Al-Aly Z. Geographic Variation
and US County Characteristics Associated With Rapid Kidney
Function Decline. Kidney Int Rep2016;2:5-17. doi:10.1016/j.
ekir.2016.08.016
15 Bowe B, Xie Y, Li T, et al. Associations of ambient coarse
particulate matter, nitrogen dioxide, and carbon monoxide with
the risk of kidney disease: a cohort study. The Lancet Planetary
health2017;1:e267-e76. doi:10.1016/S2542-5196(17)30117-1
16 Xie Y, Bowe B, Li T, et al. Blood urea nitrogen and risk of insulin
use among people with diabetes. Diabetes and vascular disease
research2018;15:409-16. doi:10.1177/1479164118785050
17 Xie Y, Bowe B, Li T, Xian H, Yan Y, Al-Aly Z. Higher blood urea nitrogen
is associated with increased risk of incident diabetes mellitus. Kidney
Int2018;93:741-52. doi:10.1016/j.kint.2017.08.033
18 Xie Y, Bowe B, Xian H, Balasubramanian S, Al-Aly Z. Rate of Kidney
Function Decline and Risk of Hospitalizations in Stage 3A CKD. Clin J
Am Soc Nephrol2015;10:1946-55. doi:10.2215/CJN.04480415
19 Xie Y, Bowe B, Xian H, Balasubramanian S, Al-Aly Z. Renal Function
Trajectories in Patients with Prior Improved eGFR Slopes and Risk
of Death. PLoS One2016;11:e0149283. doi:10.1371/journal.
pone.0149283
20 Xie Y, Bowe B, Xian H, Balasubramanian S, Al-Aly Z. Estimated GFR
Trajectories of People Entering CKD Stage 4 and Subsequent Kidney
Disease Outcomes and Mortality. Am J Kidney Dis2016;68:219-28.
doi:10.1053/j.ajkd.2016.02.039
21 Xie Y, Bowe B, Li T, Xian H, Balasubramanian S, Al-Aly Z. Proton Pump
Inhibitors and Risk of Incident CKD and Progression to ESRD. J Am Soc
Nephrol2016;27:3153-63. doi:10.1681/ASN.2015121377
22 Xie Y, Bowe B, Li T, Xian H, Yan Y, Al-Aly Z. Long-term kidney outcomes
among users of proton pump inhibitors without intervening acute
kidney injury. Kidney Int2017;91:1482-94. doi:10.1016/j.
kint.2016.12.021
23 Bowe B, Xie Y, Li T, et al. The 2016 global and national burden
of diabetes mellitus attributable to PM2.5 air pollution. The
Lancet Planetary health2018;2:e301-e12. doi:10.1016/S2542-
5196(18)30140-2.
24 Centers for Disease Control and Prevention. Use of the National
Death Index in Health Research. https://www.cdc.gov/nchs/data/ndi/
ndi_bibliography_health_research.pdf.
25 Centers for Disease Control and Prevention. International
Classication of Diseases,Tenth Revision (ICD-10). https://www.cdc.
gov/nchs/icd/icd10.htm.
26 Yoshida K, Solomon DH, Kim SC. Active-comparator design
and new-user design in observational studies. Nat Rev
Rheumatol2015;11:437-41. doi:10.1038/nrrheum.2015.30
27 Bowe B, Xie Y, Li T, et al. Changes in the us burden of chronic kidney
disease from 2002 to 2016: An analysis of the global burden of
disease study. JAMA Netw Open2018;1:e184412. doi:10.1001/
jamanetworkopen.2018.4412
28 Xie Y, Bowe B, Mokdad AH, et al. Analysis of the Global Burden of
Disease study highlights the global, regional, and national trendsof
chronic kidney disease epidemiology from 1990 to 2016. Kidney
Int2018;94:567-81. doi:10.1016/j.kint.2018.04.011
29 Al-Aly Z, Balasubramanian S, McDonald JR, Scherrer JF, O’Hare
AM. Greater variability in kidney function is associated with an
increased risk of death. Kidney Int2012;82:1208-14. doi:10.1038/
ki.2012.276
30 Al-Aly Z, Zeringue A, Fu J, et al. Rate of kidney function decline
associates with mortality. J Am Soc Nephrol2010;21:1961-9.
doi:10.1681/ASN.2009121210
31 Franklin JM, Eddings W, Glynn RJ, Schneeweiss S. Regularized
Regression Versus the High-Dimensional Propensity Score for
Confounding Adjustment in Secondary Database Analyses. Am J
Epidemiol2015;182:651-9. doi:10.1093/aje/kwv108
32 Angrist D, Imbens GW, Rubin DB. Identication of Causal Eects Using
Instrumental Variables. Journal of the American Statistical Association.
1996;434:444-55. doi:10.1080/01621459.1996.10476902
33 Terza JV, Basu A, Rathouz PJ. Two-stage residual inclusion estimation:
addressing endogeneity in health econometric modeling. J Health
Econ2008;27:531-43. doi:10.1016/j.jhealeco.2007.09.009
34 Schneeweiss S, Rassen JA, Glynn RJ, Avorn J, Mogun H,
Brookhart MA. High-dimensional propensity score adjustment
in studies of treatment eects using health care claims
data. Epidemiology2009;20:512-22. doi:10.1097/
EDE.0b013e3181a663cc
35 Rassen JA, Glynn RJ, Brookhart MA, Schneeweiss S. Covariate
selection in high-dimensional propensity score analyses of treatment
eects in small samples. Am J Epidemiol2011;173:1404-13.
doi:10.1093/aje/kwr001
36 Rassen JADM, Huang W, Schneeweiss S. Pharmacoepidemiology
Toolbox. Secondary Pharmacoepidemiology Toolbox. http://www.
hdpharmacoepi.org.
37 Austin PC, Stuart EA. Moving towards best practice when using inverse
probability of treatment weighting (IPTW) using the propensity score
to estimate causal treatment eects in observational studies. Stat
Med2015;34:3661-79. doi:10.1002/sim.6607
38 Brookhart MA, Wang PS, Solomon DH, Schneeweiss S. Instrumental
variable analysis of secondary pharmacoepidemiologic
data. Epidemiology2006;17:373-4. doi:10.1097/01.
ede.0000222026.42077.ee
39 Brookhart MA, Wang PS, Solomon DH, Schneeweiss S. Evaluating
short-term drug eects using a physician-specic prescribing
preference as an instrumental variable. Epidemiology2006;17:268-
75. doi:10.1097/01.ede.0000193606.58671.c5
40 Rassen JA, Brookhart MA, Glynn RJ, Mittleman MA, Schneeweiss
S. Instrumental variables I: instrumental variables exploit
natural variation in nonexperimental data to estimate causal
relationships. J Clin Epidemiol2009;62:1226-32. doi:10.1016/j.
jclinepi.2008.12.005
41 Davies NM, Smith GD, Windmeijer F, Martin RM. COX-2 selective
nonsteroidal anti-inflammatory drugs and risk of gastrointestinal
tract complications and myocardial infarction: an instrumental
variable analysis. Epidemiology2013;24:352-62. doi:10.1097/
EDE.0b013e318289e024
42 Lipsitch M, Tchetgen Tchetgen E, Cohen T. Negative controls:
a tool for detecting confounding and bias in observational
studies. Epidemiology2010;21:383-8. doi:10.1097/
EDE.0b013e3181d61eeb
43 Scally B, Emberson JR, Spata E, et al. Eects of gastroprotectant
drugs for the prevention and treatment of peptic ulcer disease and
its complications: a meta-analysis of randomised trials. Lancet
Gastroenterol Hepatol2018;3:231-41. doi:10.1016/S2468-
1253(18)30037-2
44 Austin PC. Variance estimation when using inverse probability
of treatment weighting (IPTW) with survival analysis. Stat
Med2016;35:5642-55. doi:10.1002/sim.7084
45 Forgacs I, Loganayagam A. Overprescribing proton pump inhibitors.
BMJ2008;336:2-3. doi:10.1136/bmj.39406.449456.BE
46 Choudhry MN, Soran H, Ziglam HM. Overuse and inappropriate
prescribing of proton pump inhibitors in patients with Clostridium
dicile-associated disease. QJM: monthly journal of the Association
of Physicians2008;101:445-8. doi:10.1093/qjmed/hcn035
47 Zink DA, Pohlman M, Barnes M, Cannon ME. Long-term use of
acid suppression started inappropriately during hospitalization.
Aliment Pharmacol Ther2005;21:1203-9. doi:10.1111/j.1365-
2036.2005.02454.x
48 Strid H, Simrén M, Björnsson ES. Overuse of acid suppressant
drugs in patients with chronic renal failure. Nephrol Dial
Transplant2003;18:570-5. doi:10.1093/ndt/18.3.570
49 Qato DM, Alexander GC, Conti RM, Johnson M, Schumm P, Lindau
ST. Use of prescription and over-the-counter medications and
dietary supplements among older adults in the United States.
JAMA2008;300:2867-78. doi:10.1001/jama.2008.892
50 Katz MH. Failing the acid test: benets of proton pump inhibitors may
not justify the risks for many users. Arch Intern Med2010;170:747-
8. doi:10.1001/archinternmed.2010.64
51 Linsky A, Simon SR. Reversing gears: discontinuing medication
therapy to prevent adverse events. JAMA Intern Med2013;173:524-
5. doi:10.1001/jamainternmed.2013.4068
52 Grady D, Redberg RF. Less is more: how less health care can result
in better health. Arch Intern Med2010;170:749-50. doi:10.1001/
archinternmed.2010.90
53 Yachimski PS, Farrell EA, Hunt DP, Reid AE. Proton pump inhibitors
for prophylaxis of nosocomial upper gastrointestinal tract bleeding:
eect of standardized guidelines on prescribing practice. Arch Intern
Med2010;170:779-83. doi:10.1001/archinternmed.2010.51
54 FDA. Over-The-Counter (OTC) Heartburn Treatment. https://www.fda.
gov/drugs/resourcesforyou/consumers/ucm511944.htm.
55 Yepuri G, Sukhovershin R, Nazari-Shai TZ, Petrascheck M, Ghebre YT,
Cooke JP. Proton Pump Inhibitors Accelerate Endothelial Senescence.
Circ Res2016;118:e36-42. doi:10.1161/CIRCRESAHA.116.308807
56 Becker JC, Grosser N, Waltke C, et al. Beyond gastric acid reduction:
proton pump inhibitors induce heme oxygenase-1 in gastric and
endothelial cells. Biochem Biophys Res Commun2006;345:1014-
21. doi:10.1016/j.bbrc.2006.04.170
57 Nath KA. Heme oxygenase-1 and acute kidney injury. Curr
Opin Nephrol Hypertens2014;23:17-24. doi:10.1097/01.
mnh.0000437613.88158.d3
58 Imhann F, Bonder MJ, Vich Vila A, et al. Proton pump inhibitors
aect the gut microbiome. Gut2016;65:740-8. doi:10.1136/
gutjnl-2015-310376
59 Lai SW, Lai HC, Lin CL, et al. Proton pump inhibitors and risk of gastric
cancer in a case-control study. Gut2018;68:765-7. doi:10.1136/
gutjnl-2018-316371
60 Gueta I, Halkin H, Markovits N, et al. Proton pump inhibitors and the
risk for gastric cancer: possible confounding by serum vitamin B 12.
Gut2017;67:1904. doi:10.1136/gutjnl-2017-315695
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
RESEARCH
No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe: http://www.bmj.com/subscribe
61 Peng YC, Huang LR, Lin CL, et al. Association between proton pump
inhibitors use and risk of gastric cancer in patients with GERD.
Gut2018;68:374-6. doi:10.1136/gutjnl-2018-316057
62 Suissa S, Suissa A. Proton-pump inhibitors and increased gastric
cancer risk: time-related bias. Gut2018;67:2228-9. doi:10.1136/
gutjnl-2017-315729
63 Suzuki H, Matsuzaki J. Helicobacter pylori eradication failure may
have confounded the recent large-scale health database study
that showed proton pump inhibitors increase gastric cancer risk.
Gut2017;67:2071-2. doi:10.1136/gutjnl-2017-315698
64 Wan QY, Wu XT, Li N, et al. Long-term proton pump inhibitors use
and risk of gastric cancer: a meta-analysis of 926 386 participants.
Gut2018;2018;68:762-4. doi:10.1136/gutjnl-2018-316416
65 Brusselaers N, Wahlin K, Engstrand L, Lagergren J. Maintenance
therapy with proton pump inhibitors and risk of gastric cancer:
a nationwide population-based cohort study in Sweden. BMJ
Open2017;7:e017739. doi:10.1136/bmjopen-2017-017739
66 Cheung KS, Chan EW, Wong AYS, Chen L, Wong ICK, Leung WK. Long-
term proton pump inhibitors and risk of gastric cancer development
aer treatment for Helicobacter pylori: a population-based study.
Gut2018;67:28-35. doi:10.1136/gutjnl-2017-314605
67 Poulsen AH, Christensen S, McLaughlin JK, et al. Proton pump
inhibitors and risk of gastric cancer: a population-based cohort study.
Br J Cancer2009;100:1503-7. doi:10.1038/sj.bjc.6605024
68 Laterza L, Scaldaferri F, Gasbarrini A. Risk factors for gastric cancer:
is it time to discard PPIs?Gut2018;68:176-7. doi:10.1136/
gutjnl-2017-315621
69 Strom BL. Pharmacoepidemiology, Fourth Edition ed. Wiley,2005.
70 Schneeweiss S. Automated data-adaptive analytics for
electronic healthcare data to study causal treatment eects. Clin
Epidemiol2018;10:771-88. doi:10.2147/CLEP.S166545
71 Nørgaard M, Ehrenstein V, Vandenbroucke JP. Confounding in
observational studies based on large health care databases:
problems and potential solutions - a primer for the clinician. Clin
Epidemiol2017;9:185-93. doi:10.2147/CLEP.S129879
72 Lobo FS, Wagner S, Gross CR, Schommer JC. Addressing the issue
of channeling bias in observational studies with propensity scores
analysis. Res Social Adm Pharm2006;2:143-51. doi:10.1016/j.
sapharm.2005.12.001
Supplementary materials: Supplemental tables 1, 2,
and 3
Supplementary materials: Supplemental figure 1a-c
Supplementary materials: Supplemental figure 2
Supplementary materials: Supplemental methods
on 10 June 2019 by guest. Protected by copyright.http://www.bmj.com/BMJ: first published as 10.1136/bmj.l1580 on 29 May 2019. Downloaded from
