Volume 10, Number 1; 37-48, February 2019

http://dx.doi.org/10.14336/AD.2017.1202
Original Article

Metformin and the Risk of Dementia in Type 2

Diabetes Patients
Chin-Hsiao Tseng1,2,3, *
1Department of Internal Medicine, National Taiwan University College of Medicine, Taipei, Taiwan
2Division of Endocrinology and Metabolism, Department of Internal Medicine, National Taiwan University

Hospital, Taipei, Taiwan

3Division of Environmental Health and Occupational Medicine of the National Health Research Institutes, Zhunan,

Taiwan
[Received October 20, 2017; Revised December 1, 2017; Accepted December 2, 2017]

ABSTRACT: This retrospective cohort study investigated dementia risk associated with metformin
2use in typepatients by using the reimbursement database of the Taiwan’s National Health
diabetes
Insurance.hadThenew-onset diabetes during 1999-2005 and were followed up until December 31, 2011.
patients
An unmatched
cohort of 147,729 ever users and 15,676 never users of metformin were identified, and a
matched-pair
15,676 cohortandof15,676 never users was created by propensity score (PS). Hazard ratios
ever users
wereregression
Cox estimated incorporated
by with the inverse probability of treatment weighting using PS.
Results showed that
in the unmatched cohort, 713 never users and 3943 ever users developed dementia with
respective
1029.20 andincidence
570.03 perof100,000 person-years. The overall hazard ratio was 0.550 (95%
confidence interval:
0.508-0.596). The hazard ratio for the first (<27.0 months), second (27.0-58.1 months) and
third (>58.1
tertile months) duration of metformin therapy was 0.975 (0.893-1.066), 0.554 (0.506-
of cumulative
0.607) and 0.286respectively. Analyses in the matched cohort showed an overall hazard ratio of
(0.259-0.315),
0.707the(0.632-0.791)
and hazard ratio for the respective tertile was 1.279 (1.100-1.488), 0.704 (0.598-0.829)
and 0.387 (0.320-0.468). In conclusion, metformin use is associated with a reduced dementia
risk.
Key words: dementia, diabetes mellitus, metformin, Taiwan

ISSN: 2152-5250
*Correspondence
Copyright:
which permits
© 2017 loss
nearly
en/).
unrestricted
should
TsengbeCH. ofAdistribution,
10
addressed
use,
This daily
call
is million
for
self-care
inflammation,
actions
new
to:open-access
an Dr. cases
Chin-Hsiaoability.
and will
research
and reproduction
article
Tseng, oxidative
It any
add affects
into
distributed priorities
stress,
themainly
growing
atherosclerosis,
tothe
Dementia can be caused
global by
burden ofDepartment
vascular in medium,
dementiaunder
etiologyof Internal
hasthe
or provided
terms
been Medicine,
of the
the
advocatedoriginal
Creative
National
author
Commons
Taiwan
following and
University
Attribution
source
37 are credited.
Hospital,
License, neurodegenerative
syndrome
people
older
The
growing
and
currently
of
pool
reduce
patients
World
estimated
Taipei, but
the
incidence
characterized
Health
around
may
Taiwan.(www.who.int/mediacentre/factsheets/fs362/
that
also
the
and
insulinemia,
Conference
dementia
probably
β
amyloid-
Organization
47disease
of
E-mail: the
deposition,
happen
million
First
Elevated
patients
dementia
by multifactorial
number
[2].
(Alzheimer’s
deterioration
World
in
on
and
inadvanced
blood
with
the
(2017)
in
of
Global
The
the
brain
dysregulation
Health
the
people
younger
diabetes
link
ccktsh@ms6.hinet.networld
glucose
has
world
Action
insulin
glycation
Organization
between
summoned
disease).
in
with
and
recognized
and
generation.
memory
population
may
have
Against
mechanisms
dementia
each
resistance
ofimpair
diabetes
an
end-products
in
lipid
Itincreased
and
year
March
the
Ministerial
Dementia
isis
cerebral
metabolism
may
aand
with
2015
involve
dementia
risk
(AGEs)
hyper-
[1].
function
[3,4].
of is
 Tseng CH
Metformin and dementia

Metformin cognitive
is now considered
impairment the (oddsfirst-line
ratio 0.49, therapy
95% confidence
for
type 2 diabetesintervalmellitus.
0.25-0.95).
It reduces
In the blood
Australian
glucose clinical
level study,
by Moore
reducing hepaticet al.gluconeogenesis
showed that, among and increasing
subgroup participants with
muscular glucose
diabetes uptake
(n=104,
through
35 metformin
activation users
of theand 5'-
91 non-users),
adenosine monophosphate-activated
worse cognitive performance protein waskinase
observed in metformin
(AMPK) [5]. usersIn patients
(odds ratiowith type
2.23,2 95%diabetes
confidence
mellitus,
interval
in 1.05-
4.75)
addition to its glucose lowering effect, metformin has [13].
also
been shown to reduce Becausethe metformin
risk of is atherosclerotic
widely used inevents a large number
and cancers of anddiabetes
have an patients,
anti-agingthe effect
conflicting
[6]. findings of
metformin
Studies evaluating
on dementiathe riskeffect
and cognitive
of metformin function
on thewarrant more in-
risk
of dementia depthare still
researchrare.toFour
clarify
population-based
whether it can be beneficial or
observational harmful.
studies Therefore,
can be found the present
in the literature,
study aimed at
three
from Taiwan investigating
using the administrative
the risk of database
dementia of associated
the with
National Healthmetformin
Insurance
use in(NHI)
type and2 diabetes
one frompatients
the UK with careful
using the General
consideration
PracticeofResearch
potential Database.
bias andThe confounding
first
study by Hsu commonly
et al. encountered
from Taiwan in showed
pharmacoepidemiological
that users of
metformin only (n=1864, studies
hazard
usingratio
existing
0.76, administrative
95% databases.
confidence interval 0.58-0.98) and users of metformin
plus sulfonylureas (n=9257, hazard MATERIALS
ratio 0.65,AND METHODS
95%
confidence interval 0.56-0.74) had lower risk of dementia
while compared The toNational
diabetes Health
patients
Insurance
without (NHI)taking
was any
implemented in
antidiabeticTaiwandrugs since
(n=10519) March[7].
1995.TheItsecond
is a unique
study from healthcare
Taiwan by Cheng system
that et
covers
al. enrolled
99.6% of new-onset
the Taiwan’s type population
2 diabetes and has
patients who contracts
had beenwith using allsingle
in-hospitals
oral antidiabetic
and 93% of drug all medical
of metformin, settings
sulfonylureas
[14]. Alland records
thiazolidinediones
including disease diagnoses,
(TZDs), respectively
prescribed[8]. medications
When metformin
and performed
users were procedures are
treated as the keptreferent
as a database,
group, thewhich risk
canofbedementia
used forwas academic
significantly research
higherafter for users
approval
of TZDs
by ethics
but not review.
for users
The present
of
sulfonylureas [8]. The study
third wasstudy
granted
fromanTaiwan
approval by Kuan
number 99274.
et al. published Diabetes
recently wascompared
coded 250.XX
4651 metformin
according users to the
and a comparable
International
number of Classification
non-users matched of Diseases,
on Ninth Revision,
propensity scoreClinical(PS)Modification
[9]. They showed(ICD-9-CM).
a significantly
Dementia was coded
higher risk in as metformin
abridged codes users ofwith
A210an oradjusted
A222, or hazard
as ICD-9-CM
ratio of 1.66 codes
(95% of confidence
290.0, 290.1, interval
290.2,1.35-2.04).
290.4, 294.1, The 331.0–331.2,
UK
study by Imfeld et al. showed an increased riskorof331.7–331.9.
dementia associated The database
with metformin
was described
use (odds in ratio
more detail
1.71, in
95% confidence previously
intervalpublished
1.12-2.60) papers
by using
[15,16].a matched
Figure 1 shows the
Aging and Disease • Volume 10,
insulin
showed
the
up study
use aNumber
sensitizers
of
case-control 1,
statistical
conducted
either
2001
of
procedures February
new-onset
were
TZDs
inontrend
available
Singapore
the
design including 2019
(relative
diabetes
incidence
of risk
for
used to createand
risk
after
reduction
each
a of
0.75,
small
7086 incident 1999,
dementia,
patient,
the unmatched95%
clinical
cases to
associated
patients
Ye
of ensure
etwith acohort
original
38 Alzheimer’s
[10].
al.
with
interval
confidence
0.79,
cognitive
conducted
study
function
diabetes
Singapore
that Inmetformin
Conflicting
95%
a recent
0.56-1.00,
of
patients
confidence
function
in
Longitudinal
diagnosis
antidiabetic
and
1999-2005
diagnosis
Ever
cancer
diabetes
antidiabetic
first
included
because
423,949
Because
(n=2,062),
diagnosis
204
disease
Australia.
use
the
meta-analysis
findings
metformin
users
excluded
recruited
before
were
was
matched
of
patients
longitudinal
P=0.050)
patients
noted
and
interval
in
(n=26,720,
of
possible
2)
Aging
associated
drugs
also
Ng
the
drugs
in
7086
metformin
entry
missing
users
(n=183,837).
between
et
cohort
the
from
outpatient
diagnosed
or
observed
Study
with
evaluating
for
controls
0.62-1.01,
al.
before
or
inclusion
effect
metformin
versus
reimbursement
cancer
the
data
2with
compared
from
within
1)
1996
[12].
who
orpopulation-based
metformin
type
between
of
(n=425),
of
more
without
a161
the
clinics
had
Other
and
patients
the
P=0.064)
They
lower
6of
new-onset
metformin
(relative
1non-users
the
database.
received
months
1998
times
diabetes
distorted
impact
exclusion
adata
showed
dementia
cognitive
risk
with
was
3)
follow-
were
were
[11].
of
diagnosis
initiated
diabetes
from
of
on
risk
of
prescription
any
of
Amellitus
diabetes
not
excluded
follow-up
identified.
total
criteria
other
1996
included.
during
of
of
were
toany
time
of
 Tseng CH
Metformin and dementia

due to shortened
never lifespan
users wasand
created
possible
by matching
misdiagnosis
the PSof based on the
dementia dueGreedy
to clinical
8 →1-digit
presentations
match algorithm
of malignancy),
[17]. Logistic
4)
diagnosis of
regression
dementia was
before
usedentry
to create
(n=10,516),
the PS 5)
from
ageall
<25
years (n=9,322),
characteristics
6) age >75 (collected
years (n=22,860)
until the
and end
7) of follow-up)
listed
follow-up duration
in Table<180
1 plus
daysthe
(n=4,802).
date of entry.
As a result,
This matching method
147,729 everhas
users
beenand
used
15,676
in ournever
previous
users research
of metformin
and described in
were identified as the unmatched original detail
cohort.
elsewhere
The [15,16].
matched-pair cohort (the matched cohort) of ever and

Aging and
Figure 1.
matched-pairs Disease
Flowchart • Volume
of metformin
showing ever 10,
thearterial
eye Number
disease,
procedures
and never users 1, in
disease),
receptor
followed
from February
classified
stroke,antidiabetic
blocker,
ischemic
the as
creating 2019
detailed
calcium
heart
drugs
reimbursement
the elsewhere
disease
channel
(insulin,
unmatched
database of and
[18].
blocker,
original
the In brief,
cohort
Nationalstatin,
and
Healththe of
a cohort
39
1:1
Insurance. months
dose-response
demographic
major
obesity),
peripheral
meglitinide,
sulfonylureas
Cumulative
comorbidities
wasdiabetes-related
region
related
calculated
commonly
and
fibrate
obstructive
(angiotensin
living
data
acarbose,
analyses.
commonly
aspirin).
duration
(age,
was
diagnoses,
(hypertension,
encountered
classified
and
sex,
rosiglitazone
pulmonary
Potential
used
converting
of
The
categorized
complications
metformin
occupation
medications
living
head
comorbidities
as
disease,
confounders
injury
dyslipidemia
enzyme
Taipei,
region
and
therapy
into
andpioglitazone),
(nephropathy,
in
and
tobacco
inhibitor/angiotensin
tertiles
living
and
Northern,
diabetes
in
Parkinson’s
included
(chronic
occupation
andregion),
abuse,
for
patients
Central,
alcohol-
were
disease)
 Tseng CH
Metformin and dementia

Southern, and
369:Kao-Ping/Eastern.
blindness and lowOccupation
vision, 366.41:
was diabetic cataract,
classified as
andclass
365.44:
I (civil
glaucoma
servants,
associated
teachers,
with systemic
employees of
governmentalsyndromes),
or private stroke
businesses,
(430-438),
professionals
ischemic and
heart disease
technicians),
(410-414),
class IIperipheral
(people without
arterial
a specific
disease (250.7, 785.4,
employer,
self-employed
443.81
people
andor440-448),
seamen), chronic
class III
obstructive
(farmers or
pulmonary
fishermen) and
disease
class(aIVsurrogate
(low-income
for families
smoking; supported
490-496), tobacco abuse
by social welfare,
(305.1, 649.0
or veterans).
and 989.84),
The ICD-9-CM
alcohol-related
codes for
diagnoses (291,
the related 303,
diagnoses
535.3,are
571.0-571.3
provided below:
and 980.0),
hypertension
head injury (959.01)
(401-405), dyslipidemia (272.0-272.4), and Parkinson’s
obesity (278),
disease (332).
nephropathy (580-589), eye diseases (250.5: diabetes with
ophthalmic manifestations, 362.0: diabetic retinopathy,

Table 1. Characteristics of metformin never users and ever users in the unmatched original
cohort and in the propensity,
score matched cohort.

Unmatched original cohort Matched cohort

Never users Ever users Never users Ever users

Variable
(n=15676) (n=147729) P SD (n=15676) (n=15676) P SD
n % n % n % n %

Demographic data
Age* (years) 63.4±10.4 61.6±10.0 <0.01 -17.83 63.4±10.4 63.5±9.9 0.30 1.67
Sex (men) 9009 57.47 80123 54.24 <0.01 -7.18 9009 57.47 8985 57.32 0.78 -0.56
Occupation
I 6142 39.18 58066 39.31 <0.01 6142 39.18 6084 38.81 0.92
II 3122 19.92 34112 23.09 8.28 3122 19.92 3153 20.11 0.51
III 3301 21.06 30600 20.71 -0.65 3301 21.06 3316 21.15 0.45
IV 3111 19.85 24951 16.89 -8.58 3111 19.85 3123 19.92 0.02
Living region
Taipei 5211 33.24 46512 31.48 <0.01 5211 33.24 5296 33.78 0.31
Northern 1586 10.12 16577 11.22 3.71 1586 10.12 1499 9.56 -1.86
Central 2769 17.66 27115 18.35 1.88 2769 17.66 2691 17.17 -1.27
Aging and Disease
Southern 2746 17.52 • Volume 10, Number
25098 16.99 1, February
-1.49 2746 17.52 2780 201917.73 0.65
40 diseases
Hypertension
Dyslipidemia
Obesity
Nephropathy
Eye
Stroke
Ischemic
Peripheral
Insulin
Kao-Ping
4996
424
1282
heart
arterial
5356
and
12804
11299
2.70
2942
31.87
8.18
disease
Eastern
34.17
6676
18.77
81.68
72.08
42101
disease
34527384
3364
4.52
40101
47803
2.34
120731
122549
28.50
3550
47.10
21.46
<0.01
Major
27.14
Antidiabetic
<0.01
32.36
<0.01
22.65
81.72
82.96
67018
9.96
32427
comorbidities
-29.90
<0.01
<0.01
-8.08
37742
0.89
<0.01
424drugs
45.37
21.95
-17.32
31.61
1282
4996
0.27
28.63
2.70
25.55
<0.01
8.18
1.35
5356
2942
31.87
12804
389
11299
<0.01
-3.60
1048
3364
2.48
34.17
18.77
4885
81.68
6.98
72.08
7384
6.69
21.46
0.21
5296
2662
31.16
3550
12836
11290
47.10
<0.01
-1.29
3410
33.78
16.98
0.18
22.65
81.88
72.02
7348
-6.59
21.75
0.47
<0.01
-1.49
3550
0.64
46.87
0.91
-1.31
0.81
-4.94
22.65
0.69
0.261.00
0.68 -0.30-0.12
 Tseng CH
Metformin and dementia

Sulfonylurea 11468 73.16 107896 73.04 0.75 5.13 11468 73.16 11768 75.07 <0.01 5.32
Meglitinide 1283 8.18 5767 3.90 <0.01 -19.17 1283 8.18 1259 8.03 0.62 -0.50
Acarbose 1743 11.12 8088 5.47 <0.01 -20.10 1743 11.12 1697 10.83 0.41 -1.42
Rosiglitazone 464 2.96 7388 5.00 <0.01 10.94 464 2.96 467 2.98 0.92 -0.06
Pioglitazone 387 2.47 3943 2.67 0.14 -20.10 387 2.47 398 2.54 0.69 -1.42
Commonly encountered comorbidities
COPD 7675 48.96 71044 48.09 0.03 -2.12 7675 48.96 7807 49.80 0.14 1.86
Tobacco abuse 442 2.82 5943 4.02 <0.01 6.83 442 2.82 469 2.99 0.36 1.06
Alcohol-related diagnoses 1231 7.85 10490 7.10 <0.01 -4.16 1231 7.85 1087 6.93 <0.01 -3.77
Head injury 538 3.43 5524 3.74 0.05 1.44 538 3.43 523 3.34 0.64 -0.63

Parkinson’s disease 504 3.22 3349 2.27 <0.01 -6.26 504 3.22 504 3.22 1.00 0.10

Commonly used medications in diabetes patients

ACEI/ARB 10854 69.24 107911 73.05 <0.01 8.81 10854 69.24 10879 69.40 0.76 -0.30
Calcium channel blocker 9771 62.33 88083 59.62 <0.01 -5.62 9771 62.33 9767 62.31 0.96 0.41
Statin 8428 53.76 97358 65.90 <0.01 26.59 8428 53.76 8300 52.95 0.15 0.06
Fibrate 5338 34.05 63817 43.20 <0.01 20.06 5338 34.05 5170 32.98 <0.05 -1.41
Aspirin 8871 56.59 90400 61.19 <0.01 9.87 8871 56.59 8797 56.12 0.40 -2.07
*Age is expressed as mean ± standard deviation.
SD: standardized difference.
COPD: chronic obstructive pulmonary disease, ACEI/ARB: angiotensin converting enzyme inhibitor/angiotensin receptor
blocker. Refer to “Materials and Methods” for the classification of
occupation

Analyses were Sensitivity

conducted analyses
in both thewereunmatched
conducted after excluding
original cohortpatients
and the
who matched
receivedcohort
consecutive
to examine
prescriptions
the of
consistencymetformin
of the findings.
spanning Student’s
more than 4t months
test compared
and 6 months,
the
difference respectively.
of age betweenBecause
never andthe ever
Bureauusers
of the
and NHI
Chi-allows at
square test mostwas used
3 months
for other
of drug variables.
prescriptions
Standardized
for the patients in
difference proposed
each outpatient
by Austin visit,
and Stuart
these analyses
as a testmightfor have excluded
balance diagnostics
most of the waspatients
calculated withforpoor
alladherence
covariates, and did not
and
a value >10%receive
might indicate
regular drugpotential
refill. confounding
Incretin-based from therapies
the variable were
not[19].
reimbursed by the NHI until after 2009 in Taiwan.
Aging and Disease • Volume Incidence
10,
diagnosisNumberBecause
density
1,
ofdecades
been
estimateda recent
dementia, of
thedementia
February
introduced
and for
or study
2019 suggested
onguidelines
patients
into was
the clinical
date calculated
enrolled
of
for that
death
practice
their
in sitagliptin
oruse
each with
the
during
specific
have evolveduse
lastwas
the year
41 regards
subgroups:
cumulative
of
The
which
last
ever
referent toto
reimbursement
incorporated
weighting
this
differences
newly the
associated
denominator
Hazard
users
method
ended use
diagnosed
(IPTW)
never
duration.
and
therapies,
over
1999
[21],
two
these
software,
excluding
ratios
on
never
reduces
in
from
with of
record.
December
for metformin
with
characteristics
Analyses
was
In
tousers,
using
to
years,
users
the
and
dementia
each an
consideration
2005
the
avoid
The
version
patients
inverse
sensitivity
their
the
ever
the
was
PS.
the improvement
in
person-years
31,
tertile
were
potential
in
the
were
numerator the
identified
2011,
PS-weighted
95%
users
estimated
9.3
conducted
probability
potential
who
considered
As
[20].
unmatched
of
based following
confidence
that
(SAS
at
proposed
confounding
analysis
happened
and
was
cumulative
matched
ofmore
the in cognitive
Institute,
therapy
the
during
by
using
follow-up,
impact
hazard
time
original
statistically
Cox
of
tertiles
case
by
was
antidiabetic
cohort,
to
intervals
treatment
SAS
follow-up.
of
duration
regression
Austin,
from
during
receive
also
of
ratios
number
aCary, function
statistical
cohort
incretin-based
new
the
respectively.
of
conducted
follow-up.
for
were
an
in
NC).
significant.
drugs
and
incretin-
also
Pthe
after
have
< 0.05
 Tseng CH
Metformin and dementia

Table 2. Incidence rates of dementia and hazard ratios by metformin

exposure
Metformin use n N Person-year Incidence rate (per 100,000 person-years) HR 95% CI P value

Unmatched original cohort

Never users 713 15676 69277.31 1029.20 1.000
Ever users 3943 147730 691712.02 570.03 0.550 (0.508-0.596) <0.0001
Tertiles of cumulative duration of metformin therapy (months)
Never users 713 15676 69277.31 1029.20 1.000
<27.0 1657 48645 168899.36 981.06 0.975 (0.893-1.066) 0.5819
27.0-58.1 1363 48872 237111.30 574.84 0.554 (0.506-0.607) <0.0001
>58.1 923 50213 285701.36 323.06 0.286 (0.259-0.315) <0.0001
Matched cohort
Never users 713 15676 69277.31 1029.20 1.000
Ever users 531 15676 72593.50 731.47 0.707 (0.632-0.791) <0.0001
Tertiles of cumulative duration of metformin therapy (months)
Never users 713 15676 69277.31 1029.20 1.000
<26.6 226 5171 17707.20 1276.32 1.279 (1.100-1.488) 0.0014
26.6-57.8 180 5175 24707.24 728.53 0.704 (0.598-0.829) <0.0001
>57.8 125 5330 30179.07 414.19 0.387 (0.320-0.468) <0.0001
n: incident case number of dementia, N: case number followed
HR: hazard ratio (weighted for propensity score), CI: confidence
interval

RESULTS
analysis but was slightly higher with a significant p-
value in the matched cohort.
Table 1 compares Sensitivity
the characteristics
analyses between
conducted never
in theand unmatched
ever users of
cohort
metformin.
after excluding
In the unmatched
patientsoriginal
who had not received
cohort,
age and sex regular
differedrefillsignificantly.
of metformin The (i.e.,
mean age periods
was between two
older
(63.4±10.4 consecutive
vs. 61.6±10.0 prescriptions
years, P<0.01)of metformin
and the spanning >4
proportion of
months
men was
and higher
>6 months,(57.47%
respectively)
vs. 54.24%,or patients who
P<0.01) in never
happenedusers.to beAlltreated
other variables,
with incretin-based
except therapies
hypertension,
during
sulfonylureas,
follow-up did pioglitazone
not changeand thehead
conclusions of the
injury,
also differed significantly in the original cohort. study (Table 3).
Aging and Disease • Volume 10,
variables
diseases,
first
However,
Number
tertile,
were
cohort1,
insulin,
diabetes
not
the February
Table
and
different
sulfonylureas,
risk
patients
the
4 was
shows2019
matched
neutral
significantly
was
the associated
cohort,
alcohol-related
hazard
in theratios
respectively.
(except
unmatched
with for
a for
significantly
patients
It is
in the matched cohort, age and sex were similar and most
42 eye more
and
diagnoses
differences
metformin
hazard
dementia
unmatched
analyses,
dose-response
for
consistently
cohort
fibrate).
Theratios
incidence
associated
than
exposure
cohort
the
evident
risk
with
lower
more
in
that
The
dose-response
analyses
inshowed
While
suggested
hazard
2pattern.
each
the
findings
of
metformin
years
than
enrolled
the
ordementia,
of
matched
is
specific
examining
with
athe
use
lower
(Table
ratios
2dementia
shown
in
significantly
Patients
years
awas
matched
suggested
metformin
the
pattern
use
significantly
risk
cohort,
3).
in
suggested
not
year
especially
second
(Table
was
the
and
Table
of
The
cohort.
who
affected
from
and
not
standardized
dementia
that
the
none
use
lower
had
and
reduced
2).
2.
was
affected
a1999
hazard
in
metformin
when
had
The
lower
In
reduced
used
third
The
by
consistent
risk
either
enrollment
tertile
to
associated
arisk
overall
risk.
the
it
metformin
ratios
value
risk
tertiles
2005
ofbyhad
risk
year
DISCUSSION
the
reduction
use
dementia
the
For
of
in
been
>10%.
in
by
in
of
in
(Table
year
the
the
with
sensitivity
aenrollment.
type
used
unmatched
associated
of
showed
metformin
4).
2for a
 Tseng CH
Metformin and dementia

Table 3. Sensitivity analyses estimating hazard ratios for dementia for ever versus never
users of metformin in the original cohort.

Models n N HR 95% CI P value

Excluding two consecutive prescriptions of metformin spanning more than 4 months

Never users 713 15676 1.000

Ever users 1046 49704 0.467 (0.425-0.514) <0.0001

Tertiles of cumulative duration of metformin therapy (months)

Never users 713 15676 1.000

<27.0 392 16043 0.937 (0.826-1.064) 0.3174

27.0-58.1 325 13665 0.528 (0.463-0.603) <0.0001

>58.1 329 19996 0.261 (0.229-0.298) <0.0001

Excluding two consecutive prescriptions of metformin spanning more than 6 months

Never users 713 15676 1.000

Ever users 1448 65976 0.469 (0.429-0.513) <0.0001

Tertiles of cumulative duration of metformin therapy (months)

Never users 713 15676 1.000

<27.0 512 19446 0.966 (0.860-1.085) 0.5606

27.0-58.1 482 19013 0.541 (0.482-0.607) <0.0001

>58.1 454 27517 0.259 (0.230-0.292) <0.0001

Excluding patients treated with incretin-based therapies during follow-up

Never users 692 14750 1.000

Ever users 3615 113090 0.655 (0.604-0.711) <0.0001

Tertiles of cumulative duration of metformin therapy (months)

Never users 692 14750 1.000

<27.0 1580 41031 1.072 (0.979-1.173) 0.1318

Aging and Disease • Volume could10,explain
Number
stress1, are
such February
a beneficial 2019effect.
characteristic pathophysiological
Metformin changes in
27.0-58.1 1241 37153 0.650 (0.592-0.713) <0.0001
43 dementia
fully
gluconeogenesis
inhibits
activating
through
complex
AMPK-dependent
neuroprotective
Although
investigated,
1inhibiting
>58.1 794 34906 0.349 (0.315-0.387)
n:
HR: associated
[5].
metabolism
the
impaired
insulin
function
the
Metformin
activity
crossover
brain
the
liver
Studies
pathway
cerebrospinal
inof
effects
<0.0001
incident
hazard ratio
casemechanisms
the
receptor
with
some
insulin
[25].
[24].
patients
kinase
improves
trial
isin
numberliver
mitochondrial
suggested
[22].
observed
(weighted
ofmetformin
biological
the
Increased
Afor
showed
B1
signaling
expression
pilot
of
and
fluid
with
brain
Insulin
insulin
(LKB1)/AMPK
the
dementia,lowers
that
in
randomized
that
dementia
use
with
reduced
actions
exerts
inflammation
propensity
N: patients
respiratory-chain
activation
resistance
and
resistance
have
and
casemetformin
blood
improvement
decreased
pathway
score),improving
[3].
risk
not
of
number placebo-controlled
CI:with
glucose
metformin
been
Evidence
of
with
and
bydementia
was
followed glucose
in
increasing
tyrosine
oxidative
measurable
bycognitive
confidence suggested
[23].
interval kinase
in
 Tseng CH
Metformin and dementia

that metformin
glycemic
may protect
controlthe[33].
cardiac
Metformin
and vascular
can reduce the
system fromformation
oxidativeofstress
AGEs through
and inflammation
improving viaglycemic control
AMPK-dependent
and and
additionally
-independent
it has
pathways
been shown
[26]. that
In metformin may
line with such
exert
findings,
a scavenging
animaleffect
studies
on supported
AGEs [34].thatDysregulation of
treatment with
lipid
metformin
metabolism
improved
[4] andcognitive
gut microbiota
functiondysbiosis
in [35]
rats with a have
significant
also beenreduction
implicated
in inflammation
as potential andlinks between
oxidative stress
diabetes
in the
and brain
dementia.
[27,28].
Metformin
Upregulation
may reverse
of insulin
the
mammalian target
resistance,
of rapamycin
improve(mTOR)
insulin pathway
signaling
has and correct lipid
also been implicated
dysmetabolism
as a[24].
majorRecent
pathological
studies process
also suggested that
leading to Alzheimer’s
metformin maydisease
change [29].
the composition
Metformin is of well
gut microbiota
known for its
with
inhibitory
an increase
effect
in Akkermansia
on mTOR via species
activation
leading
of to
LKB1/AMPK [24].
improvement
Althoughinaninsulin
early laboratory
resistance study
and reduction in tissue
suggested that
inflammation
metformin [36].
increased
The United
the biogenesis
Kingdom Prospective
of
amyloid-β inDiabetes
neuronal
Study
tissues,
supported
whichthat
might
metformin
be might have a
potentially
harmful to neuronal
cardioprotective
cells, this
effect
sameresulting
study showed
in reduced
that
metformin inatherosclerotic
combination with events
insulin
in obese
reducedpatients
amyloid-
with type 2
β levels [30].
diabetes
Moremellitus
recent studies,
[37]. It onhasthe
been
contrary,
well recognized that
suggested that
atherosclerosis
metformin was plays
neuroprotective
an important against
role in the
development
amyloid-β-induced
of vascular
mitochondrial
dementia. dysfunction
Therefore, metformin
in human may also
neuronal stem
reduce
cellsthe
viarisk
an AMPK-dependent
of dementia through
pathwayits[31]
anti-atherogenic
and that metformin
action onalleviated
the vascular
apoptosis
system. induced
Taken bytogether, metformin
amyloid-β via
may suppressing
exert its beneficial
the c-Jun effect
N-terminal
on dementia
protein via either
kinases/mitogen-activatedvascular
proteinprotection
kinase pathway
or neuronal
in protection.
culture hippocampal neurons [32]. AGEs can be
responsible for dementia in diabetes patients with poor

Table 4. Hazard ratios for dementia for ever versus never users of metformin estimated for
each specific year from 1999 to 2005

Year Ever users Never users HR 95% CI P value

n N n N
Unmatched original cohort
1999 793 21033 73 1335 0.562 (0.442-0.714) <0.0001
2000 704 21309 70 1473 0.588 (0.460-0.752) <0.0001
2001 639 22089 85 1726 0.514 (0.410-0.645) <0.0001
Aging2002
and 535
Disease
21624 • Volume
100 211010,0.485
Number(0.392-0.600)
1, February <0.0001
2019
44 2004
1999
2000
2001
2002
2003
2003 55
2004
2005 90
84
96
59
69
502
397
376
2005
2304
2329
2420
2231
2223
2142
21997
20430
19247
49Matched
2027
73 157
70
85
100
104
124104
124
1335
1473
1726
2110
2411
2873
cohort
157
2411
2873
3748
0.588
0.659
0.725
3748
0.529
0.697
0.611
0.511
0.456
0.5070.624
(0.432-0.801)
(0.480-0.906)
(0.541-0.971)
(0.383-0.730)
(0.514-0.945)
(0.445-0.840)
(0.414-0.632)
(0.373-0.558)
(0.420-0.611)
n: incident
HR: hazard (0.453-0.861)
ratio
case number0.0309
0.0008
0.0101
0.0001
0.0200
0.0024
of <0.0001
(weighted 0.0041
dementia,
for propensity
N: case
score),
numberCI:followed
confidence interval
 Tseng CH
Metformin and dementia

It is interesting
initial period thatofpatients
“immortal in the
time”.firstThe tertile
immortal time
of
short-term metformin
between diabetes
use showed diagnosis
a significantly
and the start higher of the use of
risk of dementia
antidiabetic
in the matched
drugs was cohort
actually
analysis
not calculated
(Table in the
2).
Because obesity
follow-up
is one person-years.
of the majorLévesque
risk factorset al. [39] pointed out
associated
with an increased
anotherrisk potential
of dementia
source[38] of immortal
and metformin
time that is can be
strongly indicated
introduced for during
diabetes thepatients
waiting withperiod obesity
between the
[37],
the increased prescription
risk in theand first
dispense
tertile of might
medications
have been when patients
carried over arefrom
discharged
patientsfrom withtheobesity
hospital.
who wereIt isfirst
worthy to note
initiated withthat
this metformin
would not treatment.
happen in the present study because all
Pharmacoepidemiological
patients were enrolled studiesfrom evaluating
the outpatient clinics. Even
if
clinical outcomes
the patients
relatedwere to medications
enrolled from using
the hospitals, neither
administrativewoulddatabases
this immortalmay suffer
time occur
from methodological
in Taiwan because all
limitations.discharge
These include
medications
prevalentcan user
be obtained
bias, immortal
directly from the
time bias and confounding hospitals
by indication.
when theBasically,
patients are thesedischarged.
potential limitationsIt is haveworthy been
to carefully
point out addressed
that immortal in time might
the
present study. be
introduced when the cumulative duration increased
The problem
becauseof the
prevalent
patients user
should
bias have
has been
livedavoided
long enough
by enrollingwithout
patients development
with newlyofdiagnosed
dementia diabetes
up to theand time of the
new users ofcumulative
metformin.duration.
The potential
Lévesque impacts
et al. resulted
pointed out that
from the usethere
isofa other
“direct antidiabetic
relation between
drugs before
the immortal
metformin period and
was initiated thewere also
magnitude of avoided
the bias” by including
[39]. Therefore,
only the magnitude of
patients
who had beenthe treated
hazardwith ratiosmetformin
in the as second
the first
and third tertiles (Table
antidiabetic2) should
drug in be ever
interpreted
users (Figure
more cautiously
1). In and the dose-
consideration
that the exclusion
response of effect
thesecould
patients
not be might
fully
introduce
clarified in the
present bias, secondary analyses were
another selection study.
conducted withoutConfounding
excluding these by indication
patients.could The overall
be much reduced
hazard ratiobyfor demonstrating
the unmatched thecohort
beneficial
was 0.508
effects(0.471-
of metformin in
0.549), and boththe hazard
the unmatched
ratios for original
the respective
cohort andtertiles
the PS-matched
of
cumulative duration
cohort (Table
of metformin
2), by modeling
therapy were with 0.894
Cox regression
(0.823-0.971), incorporated
0.511 (0.470-0.556)
with IPTW (Tableand 0.2612), (0.239-
and by showing a lack
0.285). For of thepotential
matched cohort,
residualthe confounding
overall hazardby calculating
ratio the
was 0.661 (0.590-0.742)
standardized and differences
the hazard andratios
none of forthethecovariates had a
respective tertiles were value1.210
>10%(1.037-1.411),
in the matched0.717 cohort (Table 1).
(0.610-
0.842) and 0.312Small (0.254-0.385).
sample sizes, Therefore,
prevalent theuser
resultsbias,ofimmortal
the study were timerobust
bias, and confounding
would notbybeindication,
affected bylack the of dose-
inclusion orresponse
exclusion analysis,
of theseand patients.
inadequate control group can be
Inappropriate
seen in earlier
assignment studies.
of treatment
For example,
statusthe andstudy by Hsu et
Aging and Disease • Volume 10, Number
follow-up time 1,
response
study
al. [7] February
has analysis
may compared
introducethe 2019
limitationsand potential
of
immortal small
risk oftime sample
risk
dementia
bias by of
in immortal
sizes, lacktime
subgroups of
of
45 including
period
calculation
study,
diabetes
antidiabetic
status
because
Taiwan
whole
Therefore,
avoided
treatment
follow-up
inappropriate
Furthermore,
period
during
it
of
and
the
misdiagnosis
byperiod
the
is
treatment
status.
control
patients
diabetes
all
the
were
included
bias
metformin
antidiabetic
Furthermore,
users
onset
group
sulfonylureas
released
dose-
of
enrolling
NHI
and
unlikely
drugs
since
so-called
which
assignment
the
prescription
approach
is
not
confounding
of
of
diabetes
the
group
follow-up
afor
the
who
patients
the
was
<180
well
very
sulfonylureas
diabetes
LHID2000
universal
of
only
exclusion
toimplementation
only
2had
immortal
outcome
also
used
diabetes
drugs
prevalent
without
include
or
days
of
addressed.
small
(n=796)
and
patients
those
used
more
information
follow-up
by
with
period
less
in
patients
database
(Figure
sulfonylurea
might
healthcare
numbers
indication.
of
cannot
the
the
solely
time
ambiguous
who
times
and
or
the
likely
user
or
patients
identified
[39].
present
The
have
TZDs
use
had
misclassification
TZDs
use
(the
of
1)
time
without
by
happen)
(Figure
of
bias
metformin
was
study
of
been
misclassified
the
has
included
(n=28)
the
of
In
system
new-onset
to
follow-up
diagnosis
during
plus
Kuan
any
with
kept
study
and
sulfonylureas
the
NHI.
avoided
metformin
Bureau
prescribed
by
from
ever
in
1).
antidiabetic
metformin
immortal
et
and
afor
present
in
Cheng
(n=1033),
has
the
an
The
the
al.
use
diabetes
compared
ofof
the
inappropriate
of
et
users.
cohort
included
oftime
NHIal.
to
any
only,
anddrugs.
a[8]
bias
This
of new-
the
 Tseng CH
Metformin and dementia

defined users ofInmetformin

summary, as theany present
use ofstudy at least supports
90 days a beneficial
(n=4651) andeffect
non-users
of metformin
as never on usethe of prevention
metformin of dementia in type
(n=4651) during
2 diabetes
the baseline
patients.yearThe of findings
2000 [9].give The rationale for
LHID2000 database
conductingwas formed
clinical by trials
a cohort to ofprove1 million
such an effect. Given
insurants who thatjoined
metformin
the NHIis insafe theand year cheap
2000and andwould
does not cause
not include hypoglycemia
any one who was whenborn
usedoraswho monotherapy,
joined the its NHI usefulness
after the yearfor 2000.
the prevention
Therefore,ofthe dementia
contamination
in both of thethe diabetes
use of otherpatients
antidiabetic
and non-diabetes
drugs for users people andisnon-users
worthy of in-depth
of
metformin at baseline was unavoidable during the investigation. long
follow-up period to December 31, 2010. The matched
case-control study by Imfeld et al. includedAcknowledgments 7086
incident
cases of Alzheimer’s disease diagnosed between 1998 and
2008 and a Thecomparable
study isnumberbased of in controls
part on data without from the National
dementia andHealth
matchedInsurance
on age, Research
sex, general Database practice,
provided by the
calendar time Bureau
and years
of National
of history
Health in Insurance,
the UK General Department of
Practice Research
Health Database
and managed [10]. by Because
NationalofHealth the cross-
Research
sectional nature
Institutes.
of the The case-control
interpretation design, andonlyconclusions
odds contained
ratios couldherein
be estimated,
do not represent
and it was those not of possible
Bureau to of National
Health
completely exclude
Insurance, the Department
potential risk of Health of prevalent
or National user Health
bias, immortalResearch
time biasInstitutes.
and confounding
The studybywas indication
supportedinby the
this study because
Ministry these
of Science
had not and beenTechnology
well addressed.(MOST 103-2314-B-
The
Singaporean 002-187-MY3)
study by Ng et of al.
Taiwan.
showing The an funders
improvement
had no role in
in cognitivestudyfunction
design, in users
data collection
of metformin and[12]analysis,
and thedecision to
Australian clinical studypublish, showing or a significantly
preparation ofhigher the manuscript.
risk of dementia associated with metformin use [13] were
not population-based studies. Furthermore, Conflicts both enrolled of interest
very small sample sizes and evaluated cognitive function
rather than dementia risk. The They
author bothdeclares
certainly no conflicts
might of interest.
suffer
from the potential risk of bias and confounding commonly
seen in large pharmacoepidemiological studies. References
While compared to previous studies, the present
study has a [1] Shah H,strength
combined Albanese of E, including
Duggan C, Rudan large I,samples
Langa KM,
of metformin users and Carrillo MC, etwith
patients al. dementia,
(2016). Researchaddressing priorities to
reduce the
most of the methodological global burden
limitations of dementia
associated with by 2025. Lancet
Neurol, 15: 1285-94.
pharmacoepidemiological studies and investigating the
[2] Li X, Song D, Leng SX (2015). Link between type 2
potential effect of dose-response in a follow-up design.
diabetes and Alzheimer's disease: from epidemiology
The study has additional meritsand
to mechanism of treatment.
using a nationwideClin Interv Aging, 10:
Aging and Disease • Volume
less 10, Number
likely 1,
because February
the drug 2019
cost-sharing
database that covers >99% of the population. is low in the 549-60.
Therefore,
46 theAGEs
population.
reduced
Detection
NHI
with
veterans
disease.
biochemical
confounders
alcohol
pattern,
Apo
of Taiwan
findings
E4
The
certain
genotype).
the
drinking,
for
study
orand
family
bias
The
potential
receiving
data
like
analyses.
[3]
[6]
[7]
[5]can
conditions
[4]which
limitations
due
use
history
beFurthermore,
and
anthropometric
lifestyle,
Huynhto readily
of
can
lack
biases
different
prescription
the
mechanism
and
Espinosa
profiles
CC,
et always
like
and
Rojas-Gutierrez
Wang
Hsu
Rena Incidence of
Alzheimer's
from
YW,
G,K,
al.
reducedmay
medical
generalized
genetic
Wahlqvist
Pearson
He related
low-income
measurement
nutritional
oxidative
Martins include
be
we
indications.
(2017).
SJ,
B,
in
of
ofby
E, socioeconomic
factors,
Chavez
diabetes
Feng
ER,
disease
action waived
did
dementia
RN,
ML,
the records
refills
parameters
to
Muñoz-Arenas
Metformin:
stress
Sakamoto
X,
Meikle
Lee
use
R,
Drug
of not
aDiabetologia,
Cheng
and
andto
self-reporting.
household,
lack
in
data
smoking,
status,
have
afor
the
significantly
patients
of
status
whole
chronic
(such
the
some
dietary
neurodegeneration.
Rojas
MS,
is
and
metformin:
of Des
metabolic
PJ
dementia.
Kincreased
sulfonylureas
J,
Tsai
inflammation
(2013).
(2017).
G,
K,
review
Devel
LuoTreviño
et
HN YT,
oldJdata
al. was
as
(2011).
Ther,
syndrome:
Molecular
of
in
Lipidomic
59:4
56:
Alzheimers
Tian
or
(2017).
its
type
S,
to
and
Synapse.
1898-906.
11:
new
33-44.
L,
potential
2metformin.
2421-9.
Adiabetes
insights?
link
Dis,
 Tseng CH
Metformin and dementia

J Alzheimers Metformin
Dis, 24: 485-93.
- a future therapy for neurodegenerative
[8] Cheng C, Lin CH, Tsai YW, Tsai CJ, Chou PH, diseases. Lan Pharm Res.
TH (2014).
[23] Chen
TypeY,2 Zhang
diabetesJ, Zhang
and antidiabetic
B, Gong CX (2016). Targeting
medications in insulin
relationsignaling
to dementia for the diagnosis.
treatmentJ of Alzheimer's
Gerontol A Biol Sci MedDisease. Sci, 69:Curr 1299-305.
Top Med Chem, 16: 485-92.
[9] Kuan YC, [24]
HuangViollet
KW, LinB,CL, Guigas
Hu CJ, B, Kao SanzCHGarcia
(2017). N, Leclerc J, Foretz
Effects of metformin
M, Andreelli exposure
F (2012).on neurodegenerative
Cellular and molecular
diseases in elderly
mechanisms patients
of metformin:
with typean2 overview.diabetes Clin Sci
mellitus. Prog Neuropsychopharmacol Biol (Lond), Psychiatry, 122: 253-70.
79(Pt[25] B): Koenig
77-83. AM, Mechanic-Hamilton D, Xie SX, Combs
[10] Imfeld P, BodmerMF, M, Cappola
Jick SS,AR, MeierXie CR L, (2012).
et al. (2017). Effects of the
Metformin, other insulin antidiabetic
sensitizerdrugs, metformin and risk
in Alzheimer
of disease:
Alzheimer's disease:
pilot data a population-based
from a randomizedcase-control placebo-controlled
study. J Am Geriatr crossover Soc,study.
60: 916-21.Alzheimer Dis Assoc Disord, 31: 107-
[11] Ye F, Luo YJ, Xiao J, Yu NW, Yi G (2016). Impact of 13.
insulin [26]sensitizers
Nesti L, Natali on theA incidence
(2017). Metformin of dementia: effects a on the
meta-analysis. heartDement
and Geriatr
the cardiovascular
Cogn Disord,system: 41: 251- A review of
60. experimental and clinical data. Nutr Metab Cardiovasc
[12] Ng TP, Feng L, Yap KB, Lee TS, Tan CH, Winblad BDis, 27: 657-69.
(2014).[27]Long-term
Mostafa DK, metformin
Ismail usageCA, Ghareeb and cognitive
DA (2016).
function among Differential
older adults metformin
with diabetes. dose-dependent
J effects on
Alzheimers Dis, cognition
41: 61-8. in rats: role of Akt. Psychopharmacology
[13] Moore EM, Mander AG, Ames D, Kotowicz MA, (Berl), 233: 2513-24.
Carne[28] RP, Pandey
BrodatyS,H,Garabadu
et al. (2013).
D (2017). Increased
Piracetam risk facilitates
of
cognitive impairment
the anti-amnesic
in patients but with not anti-diabetic
diabetes is activity of
associated with metformin
metformin.in experimentally
Diabetes Care,induced 36: 2981-7. type-2 diabetic
[14] Introduction toencephalopathic
the National Health rats.Insurance
Cell Mol Neurobiol,
(in 37: 791-802.
Chinese).
[29] Wang C, Yu JT, Miao D, Wu ZC, Tan MS, Tan L
https://www.nhi.gov.tw/Content_List.aspx?n=5C56D
(2014). Targeting the mTOR signaling network for
AE8D685EA0E&topn=FB01D469347C76A7
Alzheimer's disease therapy. (last Mol Neurobiol, 49: 120-
accessed November 28, 2017). 35.
[15] Tseng CH[30] (2017).
Chen Metformin
Y, Zhou K,and Wang lung R, cancer
Liu Y, risk Kwak in YD, Ma T, et
patients withal. type (2009).
2 diabetes
Antidiabetic
mellitus. drug
Oncotarget,
metformin 8:
41132-42. (GlucophageR) increases biogenesis of Alzheimer's
[16] Tseng CH (2017).amyloid Metformin peptides
is associated
via up-regulating
with a BACE1
lower risk oftranscription.
colorectal cancer Proc in Natl Taiwanese
Acad ScipatientsU S A, 106: 3907-
with type 12.
[31] 2Chiang
diabetes:MC, Cheng
a retrospective
YC, Chen SJ, cohort
Yen CH, analysis.
Huang RN
Diabetes Metab, (2016).
43: 438-45.
Metformin activation of AMPK-dependent
[17] D'Agostino, RB, Jr pathways
(1998) is Tutorial
neuroprotective
in Biostatistics:
in human neural stem
Propensity Score cellsMethods
againstfor amyloid-beta-induced
Bias Reduction in the mitochondrial
Comparison of a Treatmentdysfunction. to a Non-Randomized Exp Cell Res, 347: 322-31.
Aging and Disease • Volume 10, Control
Number 1,
[32]Group, February
Chendiabetologist's
B,Statistics 2019
Teng Y, Zhangin perspective.
Medicine,
X, Lv X,17, Acta
Yin Diabetol, 54: 417-
2265-2281
Y (2016).
47 [18] Markowicz-Piasecka
[19]
[20]
[21]
[22] Tseng
Austin
Isik
cancer:
Eur
practice
weighting
estimate
studies.
propensity
hazard
of
functions
Alzheimer's
192-8.
Skupień
AT,
CHPC
sitagliptin,
J[35]
PC,
[36]
[34]
[33]
(2012).
Endocrinol,
Soysal
(2013).
ratios.
AStat
Stuart
A,
when
causal
Jiang
Hur
Bonnefont-Rousselot
population-based
(IPTW)
in
Simó
score
Mikiciuk-Olasik
disease.
cultured
microbiota
24.
Metformin
suppression
emerging
AGE
Biol,
metformin
P,
elderly
(2017).
KY,
Diabetes,
using
Med,
EA
The
Stat
R,
C,
treatment
aYay
M,
therapeutics:
using
methods
Lee
167:
(2015).
DPP-4
Ciudin
Li
195:
34:
performance
Sikora
inverse
Med,
A,diabetic
Diabetes
hippocampal
Cognitive
G,
MS
complication
alleviated
action:
409-16.
the
3661-79.
Usarel
and
391-8.
metformin
inhibitor,
Huang
(2015).
of
for
32:
A,Moving
effects
J,cohort
propensity
E,
Alzheimer's
JNK łtowards
probability
Simó-Servat
estimating
2837-49.
Szyd
DHuttunen
gut.
CP,
patients
Res
evaluation
(Article
Focusing
of
impairment
MAPK
(2001).
(2017).
New
neurons.
Liu
owska
Aβ-induced
use,
study
in
different
Clin
Jof
onmechanisms
signaling
Diabetes
observational
Z,
type
score
cognitive
and
with
Pract,
KM
disease.
A,
Antioxidant
in
on
in
of
marginal
The
O,
Zhao
best
and
(2017).
Biomed
and
2Hernández
French
mitochondria
colon
Taiwan.
treatment
to
effects
or
diabetes-The
BInvestig,
apoptosis
perspectives.
123:
dementia:
pathway
without
(2017).
ofJRes
with
1421430.
Alzheimers
abstract).
58:
and
CInt,
1-15.
in
English
The
via
anti-
and
6:
a new
2016:
600-9.
gut
J Dis,
theSoc
 Tseng CH
Metformin and dementia

[37] UK Prospective Diabetes

review and
Study
meta-regression
(UKPDS) Groupanalysis of 589,649 men
(1998). Effectandofwomen
intensive
followed
blood-glucose
in longitudinal
control
studies.
with
metformin on complicationsAlzheimers
in overweight
Dementpatients
(Amst), 8: 165-78.
with [39]
type Lévesque
2 diabetes
LE,(UKPDS
Hanley34).
JA, Lancet,
Kezouh A, 352:
Suissa
854- S (2010).
65. Problem of immortal time bias in cohort studies:
[38] Albanese E, Launer
example
LJ, Egger
using M,
statins
Princefor
MJ,preventing progression of
Giannakopoulos P, Wolters FJ, et al. diabetes.
(2017). BMJ,
Body 340: b5087.
mass index in midlife and dementia: Systematic

Aging and Disease • Volume 10, Number 1, February 2019

48