Articles

Dementia risk after major elective surgery based on the route

of anaesthesia: a propensity score-matched population-based
cohort study
Mingyang Sun,a,i Wan-Ming Chen,b,c,i Szu-Yuan Wu,b,c,d,e,f,g,h,* and Jiaqiang Zhanga,**
a
Department of Anesthesiology and Perioperative Medicine, People’s Hospital of Zhengzhou University, Henan Provincial People’s
Hospital, Zhengzhou, Henan, China
b
Graduate Institute of Business Administration, College of Management, Fu Jen Catholic University, Taipei, Taiwan
c
Artificial Intelligence Development Centre, Fu Jen Catholic University, Taipei, Taiwan
d
Centre for Regional Anaesthesia and Pain Medicine, Wan Fang Hospital, Taipei Medical University, Taipei, Taiwan
e
Department of Food Nutrition and Health Biotechnology, College of Medical and Health Science, Asia University, Taichung, Taiwan
f
Big Data Centre, Lo-Hsu Medical Foundation, Lotung Poh-Ai Hospital, Yilan, Taiwan
g
Division of Radiation Oncology, Lo-Hsu Medical Foundation, Lotung Poh-Ai Hospital, Yilan, Taiwan
h
Department of Healthcare Administration, College of Medical and Health Science, Asia University, Taichung, Taiwan

Summary eClinicalMedicine
2023;55: 101727
Background Whether the route of anaesthesia is an independent risk factor for dementia remains unclear. Therefore,
we conducted a propensity score–matched (PSM) population-based cohort study to compare dementia incidence Published Online 4
November 2022
among surgical patients undergoing different routes of anaesthesia. https://doi.org/10.
1016/j.eclinm.2022.
Methods The inclusion criteria were being an inpatient >20 years of age who underwent major elective surgery, 101727
defined as those requiring GA without or with inhalation anaesthetics or regional anaesthesia, and being hospitalised
for >1 day between Jan 1, 2008 and Dec 31, 2019 in Taiwan. Patients undergoing major elective surgery were
categorised into three groups according to the type of anaesthesia administered: noninhalation anaesthesia, inha-
lation anaesthesia, and regional anaesthesia, matched at a 1:1 ratio. The incidence rate (IR) of dementia was
determined.

Findings PSM yielded 63,750 patients (21,250 in the noninhalation anaesthesia group, 21,250 in the inhalation
anaesthesia group, and 21,250 in the regional anaesthesia group). In the multivariate Cox regression analysis, the
adjusted hazard ratios (aHRs; 95% confidence intervals) of dementia for the inhalation and noninhalation anaesthesia
groups compared with the regional anaesthesia group were 20.16 (15.40–26.35; p < 0.001) and 18.33 (14.03–24.04;
p < 0.001), respectively. The aHR of dementia for inhalation anaesthesia compared with noninhalation anaesthesia
was 1.13 (1.03–1.22; p = 0.028). The IRs of dementia for the inhalation, noninhalation, and regional anaesthesia
groups were 3647.90, 3492.00, and 272.99 per 100,000 person-years, respectively.

Interpretation In this population based cohort study, the incidence of dementia among surgical patients undergoing
general anaesthesia was higher than among those undergoing regional anaesthesia. Among patients undergoing
general anaesthesia, inhalation anaesthesia was associated with a higher risk of dementia than noninhalation
anaesthesia. Our results should be confirmed in a randomised controlled trial.

Abbreviations: HR, hazard ratio; aHR, adjusted hazard ratio; CI, confidence interval; RCT, randomised controlled trial; ICD-9-CM, International
Classification of Diseases, Ninth Revision, Clinical Modification; IRs, incidence rates; IRRs, incidence rate ratios; PSM, propensity score matching;
NHIRD, National Health Insurance Research Database; GA, General anaesthesia; ASA, American Society of Anesthesiology; SD, standard deviation;
SMD, standardized mean difference; IQR, interquartile range; AD, Alzheimer disease; IPTW, inverse probability of treatment weighting
*Corresponding author. Associate Professor, College of Medical and Health Science, Asia University, Taichung, Taiwan; Director, Big Data Centre,
Lo-Hsu Medical Foundation, LotungPoh-Ai Hospital, Yilan, Taiwan; Attending Physician, Division of Radiation Oncology, Department of Medicine,
Lo-Hsu Medical Foundation, LotungPoh-Ai Hospital, Yilan, Taiwan; No. 83, Nanchang St., Luodong Township, Yilan County 265, Taiwan.
**Corresponding author. Department of Anesthesiology and Perioperative Medicine, Henan Provincial People’s Hospital, People’s Hospital of
Zhengzhou University, No. 7, Weiwu Road, Jinshui County, Zhengzhou, Henan, 450003, China.
E-mail addresses: [email protected] (S.-Y. Wu), [email protected] (J. Zhang).
i
These authors have contributed equally to this study (joint primary authors).

www.thelancet.com Vol 55 January, 2023 1

 Articles

Funding The study was partially supported by Lo-Hsu Medical Foundation, Lotung Poh-Ai Hospital (Funding
Number: 10908, 10909, 11001, 11002, 11003, 11006, and 11013).

Copyright © 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keywords: Anaesthesia; Dementia; Incidence rate; General anaesthesia; Regional anaesthesia

Introduction In vivo and in vitro studies have indicated that

Dementia is characterised by a decline in cognition certain anaesthetics increase AD-related pathological
involving one or more cognitive domains that is severe changes in the brain.13–19 Exposure to 4.1% sevoflurane
enough to interfere with daily function and indepen- for 6 h induces apoptosis, alters amyloid precursor
dence.1 The most common form of dementia in adults is protein processing, and increases the production of Aβ
Alzheimer disease (AD); 60%–80% of people with de- in H4-APP human neuroglioma cells.18 An in vivo study
mentia have AD. Globally, the total economic cost of exposed naïve mice to 2.5% sevoflurane for 2 h and
dementia increased from US$279.6 billion in 2000 to observed increased levels of activated caspases, beta-site
US$948 billion in 2016, with an annual growth rate of amyloid precursor protein-cleaving enzyme, and Aβ
15.9%. The total economic cost includes informal care aggregates in the brain at 6, 12, and 24 h following
costs, which were $95.1 billion in 2000 and $401.9 anaesthesia.20 Animal studies have suggested that sur-
billion in 2016, with an annual growth rate of 21.50%.1 gery and anaesthetics may accelerate AD, with cognitive
According to the World Health Organization, the esti- changes such as postoperative cognitive dysfunction
mated total global societal cost of dementia was US$1.3 being common.14–19 Clinical studies have identified an
trillion in 2019, and this total is expected to surpass association between the adverse effects of anaesthesia/
US$2.8 trillion by 2030 as both the number of people surgery and cognitive impairment/dementia and exac-
living with dementia and care costs increase.2 To lower erbation of neurodegeneration in susceptible in-
this burden, the determination of preventable risk dividuals; this matter is worthy of attention.6,8–12 In most
factors for dementia is critical. This information can situations, whether cognitive changes are due to surgery
guide health policies in promoting the early detection of or anaesthesia, inflammation, or the natural course of
dementia or dementia risk. aging is difficult to determine. Previous studies have
Perioperative neurocognitive disorder (PND) and had limitations or design flaws, such as insufficient
postoperative cognitive dysfunction are common in sample size, insufficient follow-up time, inappropriate
older adults.3,4 Moller et al. published their landmark control groups, unknown comorbidities associated with
study on long-term PND in older adults in 1998,5 and up dementia, unclear duration of surgery, unclear surgical
to 30% of postoperative patients will experience wors- types, unclear route of anaesthesia, and unclear anaes-
ening dementia.4 Postoperative cognitive dysfunction is thetic use, making it challenging to determine the
relatively common in older adults, and it may at least impact of anaesthesia on dementia.6,8–12,21–23 Therefore,
partly be caused by anaesthetics.6 Inhalation anaes- we conducted a head-to-head propensity score–matched
thetics such as sevoflurane may activate caspases, in- study to balance the aforementioned covariates and
crease β-amyloid protein (Aβ) synthesis and evaluate the dementia risk in surgical patients under-
accumulation, induce the hyperphosphorylation of tau going anaesthesia through different routes.
proteins and structural changes in brain vascular
endothelial cells, and increase blood−brain barrier
permeability; all of these mechanisms can contribute to Methods
dementia.7 However, the relationship between anaes- Data sources
thesia and dementia remains unclear. Studies have Data were obtained from Taiwan’s National Health In-
revealed inconsistent findings, making it difficult to surance Research Database (NHIRD), which contains
conclude whether the route of anaesthesia—general the details of the original outpatient and inpatient
anaesthesia (GA) or regional anaesthesia—or a specific claims data of all the beneficiaries of Taiwan’s National
anaesthetic agent increases the risk of dementia.6,8–12 Health Insurance program (>99% of the population).24–26
Many studies have reported postoperative worsening All data in the NHIRD are encrypted and include patient
of memory in patients who have undergone surgery.6,8–13 identification number, birth date, sex, diagnostic codes
This has raised questions about whether certain anaes- according to the International Classification of Diseases,
thetics or routes of anaesthesia can cause dementia. So Ninth Revision, Clinical Modification (ICD-9-CM),
far, no evidence has conclusively demonstrated whether treatment information, medical costs, dates of hospital
anaesthetic agents can cause dementia. admission and discharge, and date of death.24–26

2 www.thelancet.com Vol 55 January, 2023

 Articles

Research in context
Evidence before this study Added value of this study
Whether the route of anaesthesia is an independent risk The patients undergoing major elective surgery were
factor for dementia remains unclear. Literature research was categorised into three groups according to whether they
done through PubMed, Embase, Cochrane Review, ISI Web of received GA with or without inhalation anaesthetics or
Science, and SCOPUS up to May 31, 2022. One population- regional anaesthesia, matched at a 1:1 ratio. We determined
based retrospective PSM cohort study demonstrated no the significant hazard ratios of dementia in surgical patients
association of dementia risk with general anaesthesia (GA) or undergoing GA compared with those undergoing regional
regional anaesthesia among surgical patients. However, this anaesthesia. Our findings suggested that surgical patients
study was limited by the inclusion of only elective surgery, a undergoing inhalation anaesthesia had a higher risk of
small sample size, and no data on surgical types. Another dementia than those undergoing noninhalation anaesthesia.
longitudinal duration of surgery study by using a sample
Implications of all the available evidence
cohort based on a nationwide population sample and
To lower this health economic burden, determination of
demonstrated a significant positive association between GA
preventable risk factors for dementia is critical. This
and dementia. However, the study had a small sample size in
information, if our findings can be confirmed in a randomised
the GA group, and the control group comprised those not
controlled trial, can guide health policies in promoting early
undergoing anaesthesia.
detection of dementia or dementia risk.

Informed consent was waived because the data sets are elective surgeries during follow-up or received emer-
covered under the Personal Information Protection Act. gency surgery; (3) underwent different routes of anaes-
The study protocols were reviewed and approved by the thesia administration during the follow-up duration; (4)
Institutional Review Board of Tzu-Chi Medical had respiratory, cardiovascular, or brain surgery; (5) had
Foundation (IRB109-015-B). an American Society of Anesthesiologists (ASA) class of
≥III; (6) had a diagnosis of cancer during follow-up; or
(7) died of any cause during follow-up. We identified
Participant selection cases of inhalation anaesthesia from the database and
The inclusion criteria were being an inpatient >20 years then used 1:1 propensity score matching (PSM) to
of age who underwent major elective surgery, defined as randomly select those undergoing noninhalation
those requiring GA without or with inhalation anaes- anaesthesia and regional anaesthesia from the remain-
thetics or regional anaesthesia, and being hospitalised ing cohort.
for >1 day between Jan 1, 2008 and Dec 31, 2019 in
Taiwan. The index date was the date of the surgery. GA
was induced using intravenous induction agents and PSM and covariates
maintained using intravenous (noninhalation anaes- After adjustment for confounders, we used a Cox pro-
thesia) or inhalational anaesthetic agents (inhalation portional hazards model to calculate the time from the
anaesthesia). Regional anaesthesia included spinal, index date to dementia diagnosis for surgical patients
epidural, or combined spinal–epidural anaesthesia, who received different routes of anaesthesia. To reduce
without sedation. The use of nerve block (including type the effects of potential confounders when comparing
and dose) was at the discretion of the professional dementia risk among different routes of anaesthesia,
anaesthesiologist. The primary endpoint was the inci- the participants were PSM at a ratio of 1:1 by using the
dence of dementia. We identified patients with de- greedy method with a caliper of 0.2.27 The variables used
mentia as those with a primary diagnosis of the for matching were age, sex, comorbidities (hyperten-
following ICD-9-CM codes: 290.0, 290.1, 290.2, 290.3, sion, hyperlipidemia, coronary artery disease, stroke,
290.4, 294.1, and 331.0. In Taiwan, dementia is diag- diabetes, depression, anxiety, hearing impairment,
nosed by a board-certified psychiatrist or neurologist obesity, head injury), medication use (opioids, gaba-
according to the Diagnostic and Statistical Manual of pentinoids, Z-drugs, and benzodiazepines), cigarette
Mental Disorders, Fourth Edition. To identify patients smoking, alcohol-related diseases, ASA class, type of
with dementia with sufficient accuracy, all dementia surgery (skin, breast, musculoskeletal, digestive, kidney,
cases had at least three records of outpatient visits or ureter, bladder, and gynecological surgery), and dura-
one admission diagnosis. We excluded patients who (1) tion of surgery (Table 1). Comorbidities were deter-
had a history of dementia before or within 1 year after mined according to the main diagnosis in inpatient
the index date (to remove any undiagnosed cognitive records (based on ICD-9-CM codes) or the particular
impairment preoperatively); (2) underwent ≥2 major diagnostic code recorded at ≥2 outpatient visits within 1

www.thelancet.com Vol 55 January, 2023 3

 Articles

Noninhalation GA Inhalation GA RA (spinal or P value SMDa SMDb

epidural)
N % N % N %
21,250 21,250 21,250
Age (mean ± SD) 49.30 ± 16.55 49.46 ± 16.41 49.49 ± 16.51 0.458
Age group (years) 0.965 0.019 0.001
20–30 2628 12.4% 2631 12.4% 2633 12.4%
30–40 3881 18.3% 3882 18.3% 3887 18.3%
40–50 4618 21.7% 4617 21.7% 4707 22.2%
50–60 4239 20.0% 4241 20.0% 4223 19.9%
60–70 2927 13.8% 2930 13.8% 2820 13.3%
70–80 2118 10.0% 2113 9.9% 2110 9.9%
>80 839 4.0% 836 3.9% 870 4.1%
Sex 0.496 0.010 0.000
Women 11,838 55.7% 11,838 55.7% 11,733 55.2%
Men 9412 44.3% 9412 44.3% 9517 44.8%
Coexisting medical conditions
Hypertension 6996 32.9% 6997 32.9% 7224 34.0% 0.125 0.023 0.000
Hyperlipidemia 5791 27.3% 5792 27.3% 5892 27.7% 0.451 0.011 0.000
Coronary artery disease 4117 19.4% 4117 19.4% 4350 20.5% 0.104 0.028 0.000
Stroke 806 3.8% 819 3.9% 817 3.8% 0.939 0.000 0.003
Diabetes mellitus 4406 20.7% 4406 20.7% 4539 21.4% 0.177 0.016 0.000
Depression 4417 20.8% 4418 20.8% 4556 21.4% 0.163 0.016 0.000
Anxiety 1676 7.9% 1679 7.9% 1701 8.0% 0.886 0.016 0.001
Hearing impairment 268 1.3% 268 1.3% 269 1.3% 0.920 0.001 0.000
Obesity 844 4.0% 845 4.0% 843 4.0% 0.994 0.000 0.001
Head injury 168 0.8% 169 0.8% 167 0.8% 0.989 0.000 0.001
Medication use
Opioids 3188 15.0% 3188 15.0% 3189 15.0% 0.995 0.000 0.000
Gabapentinoids 855 4.0% 856 4.0% 857 4.0% 0.990 0.000 0.000
Z-drugs 5377 25.3% 5379 25.3% 5378 25.3% 0.987 0.000 0.000
Benzodiazepines 6524 30.7% 6523 30.7% 6524 30.7% 0.998 0.000 0.000
Cigarette smoking 1706 8.0% 1709 8.0% 1710 8.1% 0.871 0.003 0.002
Alcohol-related diseases 946 4.45% 946 4.45% 945 4.45% 0.939 0.000 0.001
ASA 0.996 0.001 0.000
I 12,756 60.03% 12,756 60.03% 12,758 60.04%
II 8494 39.97% 8494 39.97% 8492 37.96%
Type of surgery
Skin 637 3.0% 637 3.0% 630 2.96% 0.931 0.001 0.000
Breast 1708 8.0% 1708 8.0% 1718 8.08% 0.960 0.003 0.000
Musculoskeletal 5322 25.0% 5322 25.0% 5320 25.0% 0.999 0.000 0.000
Digestive 5904 27.8% 5904 27.8% 5909 27.8% 0.966 0.002 0.000
Kidney–ureter–bladder 4033 19.0% 4033 19.0% 4037 19.0% 0.996 0.001 0.000
Gynecological surgery 3646 17.2% 3646 17.2% 3636 17.1% 0.932 0.002 0.000
Duration of surgery (min, mean ± SD) 91.5 ± 21.5 91.9 ± 23.1 90.1 ± 18.8 0.587
Follow-up (years, mean ± SD) 7.2 ± 4.9 7.0 ± 4.6 6.9 ± 4.3 0.084
Follow-up (years, median (IQR)) 6.6(2.9–11.2) 6.6(2.4–6.7) 6.5(2.4–8.9) 0.071
Outcomes
Dementia <0.001
No 20,379 95.9% 20,166 94.9% 21,193 99.7%
Yes 871 4.1% 1084 5.1% 57 0.2%

Abbreviations: RA, regional anaesthesia; GA, general anaesthesia; ASA, American Society of Anesthesiology; SD, standard deviation; SMD, standardised mean difference;
IQR, interquartile range (Q1–Q3). aThe standardised mean difference of noninhalation GA and RA. bThe standardised mean difference of inhalation GA and noninhalation
GA.

Table 1: Characteristics of propensity score–matched surgical patients who received different routes of anaesthesia.

4 www.thelancet.com Vol 55 January, 2023

 Articles

year. Comorbidities that presented at 1 year before the test, p < 0.05 was considered significant. Poisson
index date were recorded. The prematched de- regression models were used to compare overall de-
mographics are presented in Supplemental Table S1. mentia incidence rates by estimating IRRs with 95%
Supplemental Table S4 and Supplemental Figure S1 CIs. Gray test was used to compare the curves of
present the full and final propensity-score model cumulative incidence functions of dementia risk
applied (including all covariates and their coefficients). undergo the different route of anaesthesia.29,31

Incidence rate and incidence rate ratios of Role of the funding source
dementia risk The funder had no role in study design; in the collec-
The primary endpoints were the incidence rate (IR) and tion, analysis, and interpretation of data; in the writing
IR ratios (IRRs) of dementia risk of surgical patients of the report; and in the decision to submit the paper for
undergoing different routes of anaesthesia determined publication. All authors confirm that they had full access
through Poisson regression by using 100,000 person- to all the data in the study and accept responsibility to
years as an offset. The Poisson regression model for submit for publication.
the adjusted IRRs of dementia risk was adjusted for the
aforementioned PSM variables.
Results
Sensitivity analysis PSM and study cohort
A sensitivity analysis of dementia risk was conducted PSM yielded 63,750 patients (21,250 per group); their
using inverse probability of treatment weighting (IPTW) characteristics are listed in Table 1. Because of PSM, no
for surgical patients undergoing inhalation or non- significant between-group differences were observed in
inhalation anaesthesia. This was conducted to clarify the age, sex, comorbidities, cigarette smoking, alcohol-
association of dementia with undergoing inhalation or related diseases, ASA class, type of surgery, and dura-
noninhalation anaesthesia among surgical patients tion of surgery. The crude dementia incidence in the
stratified by age and comorbidities. All analyses were inhalation anaesthesia group significantly differed from
adjusted for the covariates mentioned in Table 2. that in the noninhalation and regional anaesthesia
groups (p < 0.001; Table 1).

Statistical analysis
Continuous variables are presented as mean ± standard Predictors for dementia risk after multivariate cox
deviation, where appropriate. A Cox model was used to regression analysis
perform the regression of the variables of dementia risk Multivariate Cox regression analysis revealed that the
in the three groups of surgical patients (noninhalation inhalation anaesthesia group had the highest dementia
anaesthesia, inhalation anaesthesia, and regional risk, followed by the noninhalation anaesthesia group
anaesthesia), and a robust sandwich estimator was used and the regional anaesthesia group (Table 2). No sig-
to account for clustering within matched sets.28 Multi- nificant differences were observed in explanatory vari-
variate Cox regression analysis was performed to ables, except for age (>40 years), hypertension,
calculate hazard ratios with 95% confidence intervals hyperlipidemia, diabetes, coronary artery disease,
(CIs) to determine the potential independent predictors stroke, depression, anxiety, and medication use.
of dementia risk. In order to overcome the death bias, Compared with the regional anaesthesia group, the
competing risk analysis has been performed to estimate aHRs (95% CIs) of dementia risk of the inhalation and
correctly marginal probability of dementia. Some pa- noninhalation anaesthesia groups were 20.16
tients might die before a diagnosis of dementia during (15.40–26.35; p < 0.001) and 18.33 (14.03–24.04;
the follow-up period. With the release of SAS version 9.4 p < 0.001), respectively, whereas that of the inhalation
(SAS Institute, Cary, NC, USA), Fine and Gray’s sub- anaesthesia group compared with the noninhalation
distribution hazard model can be fitted by specifying anaesthesia group was 1.13 (1.03–1.22; p = 0.028;
event code option in PROC PHREG. “Proportional Table 3). Compared with those aged 20–30 years, the
subdistribution hazards model in PHREG” can be used aHRs (95% CIs) of dementia risk of those aged 31–40,
to assess the effect of competing events (death) on 41–50, 51–60, 61–70, 71–80, and >80 years were 1.12
outcome (dementia) which otherwise would have been (0.80–1.54), 1.55 (1.14–2.14), 3.60 (2.71–4.82), 9.40
censored.29 SAS syntax (competing risk death model) (7.11–12.50), 13.12, (17.44–17.64), and 15.42
was used for competing risk analysis and dementia (10.81–18.43), respectively (Table 2). The aHRs (95% CI)
incidence figure constructed in our study. of dementia risk of patients with hypertension, hyper-
All statistical analyses were performed using SAS lipidemia, diabetes, coronary artery disease, stroke,
version 9.4, and the matching procedure was imple- depression, and anxiety compared with those without
mented using PROC PSMATCH.30 In a two-tailed Wald the respective comorbidities were 1.19 (1.07–1.33), 1.04

www.thelancet.com Vol 55 January, 2023 5

 Articles

Crude HR 95% CI P value aHRa 95% CI P value

Routes of anaesthesia (Ref. RA)
Noninhalation GA 16.42 (12.53–21.42) <0.001 18.33 (14.03–24.04) <0.001
Inhalation GA 14.93 (11.90–19.45) <0.001 20.16 (15.40–26.35) <0.001
Sex (Ref. Women)
Men 1.18 (1.08–1.32) <0.001 1.08 (1.00–1.10) 0.055
Age (years, Ref. 20–30)
30–40 1.13 (0.81–1.56) 0.494 1.12 (0.80–1.54) 0.480
40–50 1.67 (1.21–2.25) 0.001 1.55 (1.14–2.14) 0.005
50–60 4.16 (3.11–5.52) <0.001 3.60 (2.71–4.82) <0.001
60–70 11.51 (8.74–15.21) <0.001 9.40 (7.11–12.50) <0.001
70–80 15.12 (12.17–18.29) <0.001 13.12 (17.44–17.64) <0.001
>80 16.62 (12.61–25.26) <0.001 15.42 (10.81–18.43) <0.001
Coexisting medical conditions
Hypertension (Ref. no hypertension) 3.84 (3.53–4.22) <0.001 1.19 (1.07–1.33) 0.001
Hyperlipidemia (Ref. no hyperlipidemia) 2.10 (1.91–2.29) <0.001 1.04 (1.01–1.06) 0.043
Diabetes (Ref. no diabetes) 2.85 (2.62–3.11) <0.001 1.31 (1.21–1.46) <0.001
Coronary artery disease (Ref. no coronary artery diseases) 3.32 (3.03–3.64) <0.001 1.06 (1.02–1.21) 0.006
Stroke (Ref. no stroke) 4.00 (3.98–5.30) <0.001 1.58 (1.34–1.81) <0.001
Depression (Ref. no depression) 1.69 (1.51–1.84) <0.001 1.11 (1.07–1.22) 0.013
Anxiety (Ref. no depression) 1.93 (1.71–2.20) <0.001 1.67 (1.42–1.91) <0.001
Hearing impairment (Ref. no hearing impairment) 1.11 (0.63–1.42) 0.376 1.09 (0.73–1.40) 0.362
Obesity (Ref. no obesity) 1.07 (0.57–1.40) 0.503 1.04 (0.61–1.60) 0.505
Head injury (Ref. no head injury) 1.03 (0.44–1.70) 0.679 1.04 (0.45–1.67) 0.611
Medication use
Opioids (Ref. no opioid use) 1.09 (0.82–1.47) 0.338 1.03 (0.85–1.38) 0.382
Gabapentinoids (Ref. no gabapentinoid use) 1.05 (0.63–1.41) 0.327 1.04 (0.75–1.22) 0.539
Z-drugs (Ref. no Z-drug use) 1.07 (0.69–1.21) 0.214 1.05 (0.70–1.16) 0.244
Benzodiazepines (Ref. no benzodiazepine use) 1.04 (0.59–1.11) 0.368 1.02 (0.61–1.08) 0.263
Cigarette smoking (Ref. no smoking) 1.06 (0.59–1.41) 0.313 1.05 (0.56–1.34) 0.371
Alcohol-related diseases (Ref. no alcohol-related diseases) 1.11 (0.78–1.82) 0.279 1.01 (0.84–1.82) 0.233
ASA class (Ref. I)
II 1.16 (0.85–1.34) 0.199 1.15 (0.90–1.22) 0.163
Abbreviations: CI, confidence interval; HR, hazard ratio; aHR, adjusted hazard ratio; Ref., reference group; RA, regional anaesthesia; GA, general anaesthesia; ASA, American
Society of Anesthesiology. aAdjusted for all covariates in Table 2.

Table 2: Univariable and multivariable Cox proportional regression model of dementia for surgical patients who received different routes of
anaesthesia.

(1.01–1.06), 1.31 (1.21–1.46), 1.06 (1.02–1.21), 1.58 IR and IRRs of dementia among different routes of
(1.34–1.81), 1.11 (1.07–1.22), and 1.67 (1.42–1.91), anaesthesia
respectively. Overall, significant adjusted IRRs of dementia risk of
inhalation anaesthesia were identified (Table 4). The

Routes of anaesthesia Crude HR 95% CI P value aHRa 95% CI P value

Ref. RA
Inhalation GA 14.92 (11.92–19.49) <0.001 20.16 (15.40–26.35) <0.001
Ref. RA
Noninhalation GA 16.43 (12.58–21.46) <0.001 18.33 (14.03–24.04) <0.001
Ref. Noninhalation GA
Inhalation GA 1.10 (1.01–1.20) 0.037 1.13 (1.03–1.22) 0.028

Abbreviations: CI, confidence interval; HR, hazard ratio; aHR, adjusted hazard ratio; Ref., reference group; RA, regional anaesthesia; GA, general anaesthesia. aAdjusted for all
covariates in Table 2.

Table 3: Cox proportional regression model of dementia for surgical patients who received different routes of anaesthesia.

6 www.thelancet.com Vol 55 January, 2023

 Articles

Noninhalation GA (N = 21,250) Inhalation GA (N = 21,250)

Incidence of dementia % IRa Incidence of dementia % IRa Adjusted IRRb 95% CI P value
871 4.1% 3492.0 1084 5.1% 3647.9 1.11 (1.02–1.18) 0.022

RA (spinal or epidural) (N = 21,250) Inhalation GA (N = 21,250)

Incidence of dementia % IRa Incidence of dementia % IRa Adjusted IRRb 95% CI P value

57 0.3% 273.0 1084 5.1% 3647.9 16.48 (14.48–21.52) <0.001

Abbreviations: N, number; IR, incidence rate; IRR, incidence rate ratio; CI, confidence interval. aIncidence rate: per 100,000 person-years. bAll covariates presented in Table 2
were adjusted.

Table 4: Incidence rates and IRRs of dementia in surgical patients who received different routes of anaesthesia.

adjusted IRR (95% CI) of dementia risk of the inha- Discussion

lation anaesthesia group compared with the non- The association between the risk of dementia and the
inhalation anaesthesia group was 1.11 (1.02–1.18), and route of anaesthesia remains controversial.22,23,32 One
that compared with the regional anaesthesia group population-based retrospective PSM cohort study
was 16.48 (14.48–21.52). The IRs of dementia for the demonstrated no association of dementia risk with GA
inhalation, noninhalation, and regional anaesthesia or regional anaesthesia among surgical patients.22
groups were 3647.90, 3492.00, and 272.99 per 100,000 However, this study was limited by the inclusion of
person-years, respectively. Supplemental Table S2 only elective surgery, a small sample size (7499 per
presents the time interval between anaesthesia and group after PSM), short follow-up time (<5 years), and
dementia. Dementia risk was persistently higher in no data on duration of surgery.22 Sohn et al. performed a
the inhalation anaesthesia and noninhalation anaes- longitudinal duration of surgery study by using a sample
thesia groups than in the regional anaesthesia groups, cohort based on a nationwide population sample and
regardless of time interval. demonstrated a significant positive association between
GA and dementia.23 However, their study had a small
sample size in the GA group, and the control group
Kaplan–Meier curve of cumulative dementia comprised those not undergoing anaesthesia23; because
incidence among different routes of anaesthesia critical clinical problems are usually corrected surgically,
Fig. 1 presents the cumulative dementia risk in our the study lacked information to allow for comparisons of
cohort. The cumulative dementia risk was significantly the association of dementia risk between surgical and
higher in the inhalation anaesthesia group, followed by nonsurgical patients.23 We cannot stop surgery and
the noninhalation anaesthesia group and the regional anaesthesia for surgical patients with critical in-
anaesthesia group (p < 0.001). dications; therefore, no valuable reference or change in
clinical practice can be obtained for patients undergoing
surgery.
Sensitivity analysis of dementia for inhalation and A randomised controlled trial (RCT) concluded that
noninhalation anaesthesia groups (stratified by age patients who received sevoflurane, isoflurane, or halo-
and comorbidities) thane during spinal surgery were more likely to exhibit a
A sensitivity analysis of dementia for the inhalation progression of preexisting mild cognitive impairment
and noninhalation anaesthesia groups based on the than controls and patients who received propofol or
significant risk factors mentioned in Table 2 was epidural anaesthesia, although the role of inhalation
performed. A stratified analysis of distinct groups anaesthetics such as sevoflurane in the long-term
stratified by age and comorbidities based on IPTW sequela of dementia risk has been unclear in patients
was performed, and the results are presented as a undergoing GA.33 Our current study is the largest head-
forest plot in Fig. 2. In the inhalation anaesthesia to-head PSM study, with a sufficient follow-up time to
group, the patients aged >65 years or those having evaluate the dementia risk in surgical patients under-
hypertension, hyperlipidemia, diabetes, coronary arte- going different routes of anaesthesia. Our findings
rial diseases, stroke, depression, or anxiety had aHRs indicate that in patients undergoing major elective sur-
(95% CIs) of 1.15 (1.03–1.29), 1.11 (0.99–1.25), 1.22 gery, inhalation anaesthesia had the highest dementia
(1.05–1.41), 1.21 (1.04–1.41), 1.25 (1.08–1.45), 1.36 risk, followed by noninhalation anaesthesia and regional
(1.04–1.77), 1.35 (1.13–1.61), and 1.20 (0.92–1.56), anaesthesia, regardless of age and comorbidities.
respectively, indicating that a significantly higher risk Further RCTs are warranted to validate our results.
of dementia than the corresponding subgroups in the Among surgical patients undergoing anaesthesia,
noninhalation anaesthesia group. risk factors for dementia include age, sex, comorbidities

www.thelancet.com Vol 55 January, 2023 7

 Articles

Fig. 1: Cumulative incidence of dementia in propensity score–matched surgical patients who received different routes of anaesthesia
(general anaesthesia with or without inhalation anesthetics and regional anaesthesia).

(hypertension, hyperlipidemia, coronary artery disease, inference. Ours is the leading study to use well-designed
stroke, diabetes, depression, anxiety, hearing impair- PSM to investigate the effects of different routes of
ment, obesity, head injury), medication use, cigarette anaesthesia on dementia. However, PSM modeling
smoking, and alcohol-related diseases. We matched the from database-collected information cannot replace an
three anaesthesia groups for all these factors. We also RCT.
adjusted for the possible risk factors for the underlying Anaesthesia and surgery are associated with a
diseases of surgical type, anaesthetic exposure time, and modest acceleration in the rate of cognitive decline in
ASA class (Table 1). After PSM, all covariates were older patients.21 How this acceleration might be
balanced between the groups. We used a robust PSM- magnified in patients already on a steep trajectory of
based design to ensure homogeneity between the case cognitive decline, such as those with presymptomatic
and control groups in terms of potential confounding AD and related dementia, remains unknown. Anaes-
variables. Performing an RCT to evaluate surgical pa- thetics may increase cerebral β-amyloid deposits, a
tients undergoing different routes of anaesthesia is hallmark of AD.21,32 A 2014 study including 24,901 pa-
challenging because the routes cannot be corrected tients aged ≥50 years observed an increase in the inci-
through a tangible intervention.34 Balancing the con- dence of dementia and a reduced interval to dementia
founding factors of dementia in surgical patients un- diagnosis after anaesthesia and surgery.32 In addition,
dergoing different routes of anaesthesia—a main although the knowledge base for this clinical scenario is
requirement of an RCT design—is difficult.34 A PSM- limited, if the association between anaesthesia and
based design can resolve this problem by maintaining surgery and the acceleration of neurocognitive decline
a balance between the case and control groups in terms in a patient with a vulnerable brain is proven, whether
of confounding factors in the absence of bias. Moreover, the dementia risk differs among patients undergoing
PSM is recommended for estimating the effects of GA or regional anaesthesia requires clarification.
covariates in studies where potential bias may be pre- Moreover, these patients should be organised around
sent.27 Although the main advantage of PSM is the more factors such as comorbidities, the underlying condition
precise estimation of covariate effects, it cannot control necessitating surgery, the duration of surgery, or the
for factors not accounted for in the model. Moreover, surgical type.32
PSM is predicated on an explicit selection bias for those Our study is the largest head-to-head PSM study with
who can be matched, meaning that individuals who a long-term follow-up to match the aforementioned
cannot be matched are excluded from the scope of comorbidities, underlying conditions necessitating

8 www.thelancet.com Vol 55 January, 2023

 Articles

Fig. 2: Sensitivity analysis of age and comorbidities conducted using inverse probability of treatment weighting for surgical patients
who received noninhalation or inhalation general anaesthesia.

surgery, duration of surgery, and surgical type. Our re- noninhalation anaesthesia after sensitivity analysis,
sults revealed that the inhalation anaesthesia or non- regardless of patient age and comorbidity. The potential
inhalation anaesthesia groups had a higher risk of mechanisms might be that inhalation anaesthetics can
dementia than the regional anaesthesia group (Tables 2 cause neuronal apoptosis by activating caspase and Aβ
and 3). protein aggregation.7
With respect to the connection between anaesthesia This study has the largest sample size and the
and brain health, studies have reported conflicting longest follow-up period of any cohort study investi-
findings for regional anaesthesia and GA in surgical gating the association between dementia risk and sur-
patients.3,6-10,19–21 We demonstrated that GA is associated gical patients undergoing different routes of
with a higher risk of dementia than regional anaesthesia anaesthesia. In the current study, we used a head-to-
after adjustment for age, sex, comorbidities, cigarette head PSM design, mimicking an RCT, to eliminate
smoking, alcohol-related diseases, ASA class, types of potential bias. We matched the groups according to age,
surgery, and duration of surgery. Few studies have sex, comorbidities, cigarette smoking, alcohol-related
evaluated whether a specific anaesthetic agent affects diseases, ASA class, type of surgery, and duration of
cognitive outcomes after GA. An RCT indicated that surgery and adjusted for covariates to determine the
inhalation anaesthetics for lumbar spine surgery accel- effect of different routes of anaesthesia on dementia in
erated cognitive decline but recommended further surgical patients.
studies with a larger sample size and longer follow-up This study has some limitations. First, this study
period to validate this relationship.33 However, another recruited only participants of Asian ethnicity; therefore,
RCT indicated that the choice of inhalational anaesthetic caution should be exercised when extrapolating our re-
did not result in any significant difference in the inci- sults to non-Asian populations. Second, the diagnoses of
dence of postoperative cognitive dysfunction between all comorbidities were based on ICD-9-CM codes.
two groups.35 However, short-term postoperative cogni- Nevertheless, the NHIRD reviews charts and interviews
tive dysfunction with inhalation anaesthetics cannot be patients to verify the accuracy of such diagnoses, and
representative of long-term dementia risk. Our study is hospitals with outlier charges or practices may be
the largest PSM study with sufficient follow-up time to audited and subsequently heavily penalised if malprac-
demonstrate that inhalation anaesthesia is associated tice or discrepancies are identified. Accordingly, to
with a significantly higher risk of dementia relative to obtain crucial information on population specificity and

www.thelancet.com Vol 55 January, 2023 9

 Articles

disease occurrence, a large-scale RCT comparing care- comorbidities. Our results may guide shared decision-
fully selected surgical patients undergoing different making between surgical patients and physicians.
routes of anaesthesia may be necessary. Third, the
prevalence of dementia was relatively low, likely because Contributors
our patients were comparatively younger than those in Mingyang Sun and Wan-Ming Chen were responsible for manuscript
other datasets (Table 1).6,8–12 If incidence of the endpoint preparation. Szu-Yuan Wu, and Jiaqiang Zhang have verified the un-
(dementia) is extremely low in patients aged <65 years, a derlying data. All authors confirm that they had full access to all the data
in the study and accept responsibility to submit for publication.
large sample size would be required to reach statistical
significance. However, the sample size in our study was
sufficient to reach statistical significance. In addition, Data sharing statement
We used data from Taiwan’s National Health Insurance Research
our results could address the issue of dementia risk in Database. The authors confirm that for approved reasons, some access
surgical patients of a wide range of ages undergoing restrictions apply to the data underlying the findings. The data used in
different routes of anaesthesia, not only in elderly pa- this study cannot be made available in the manuscript, the supplemental
tients. The dementia risk was persistently higher in the files, or in a public repository due to the Personal Information Protec-
tion Act in force in Taiwan since 2012. Requests for data can be sent as a
inhalation anaesthesia and noninhalation anaesthesia
formal proposal to obtain approval from the ethics review committee of
groups than in the regional anaesthesia group, regard- the appropriate department of the Government of Taiwan. Specifically,
less of age group (Supplemental Table S3). Fourth, un- data requests may be sent via the following links: http://nhird.nhri.org.
diagnosed cognitive impairment preoperatively might tw/en/Data_Subsets.html#S3 and http://nhis.nhri.org.tw/point.html.
constitute a bias. However, we excluded dementia
diagnosed within 1 year after the index date (to remove Editor note
any undiagnosed cognitive impairment preoperatively). The Lancet Group takes a neutral position with respect to territorial
Moreover, the most enrolled patients (approximately claims in published maps and institutional affiliations.
80%) were <65 years old with less prevalence of cogni-
tive impairment. We think the bias of undiagnosed
Declaration of interests
cognitive impairment preoperatively can be ignored The authors have no potential conflicts of interest to declare.
because of the aforementioned reasons and the large
sample size in the current study were based on the law
of large numbers. Fifth, the actual inhalation anaes- Acknowledgements
thetics used in Taiwan are usually crossovers and The authors would like to thank the medical staff who assisted us in the
study. The study was partially supported by Lo-Hsu Medical Foundation,
combined with sevoflurane, isoflurane, or desflurane.
Lotung Poh-Ai Hospital (Funding Number: 10908, 10909, 11001, 11002,
The details of the prescribed anaesthetics are not avail- 11003, 11006, and 11013).
able in the NHIRD. Providing some data on the actual
anaesthetic used within each of the three main cate-
gories and analyzing whether the type of drug has any Appendix A. Supplementary data
impact on outcome would be difficult for the afore- Supplementary data related to this article can be found at https://doi.
org/10.1016/j.eclinm.2022.101727.
mentioned reasons. Sixth, there is no indication of the
initial intent in using different routes of anaesthesia in
an observational study. However, we matched the types
References
of surgery to avoid the bias of major or minor surgeries 1 Xu J, Zhang Y, Qiu C, Cheng F. Global and regional economic
in the current study. Finally, the NHIRD does not costs of dementia: a systematic review. Lancet. 2017;390:S47.
contain information on dietary habits, education data, 2 World Health Organization. Dementia: World Health Organization.
Regional Office for the Eastern Mediterranean; 2019.
family history, or laboratory data, which may be risk 3 Eckenhoff RG, Maze M, Xie Z, et al. Perioperative neurocognitive
factors for dementia. disorder: state of the preclinical science. Anesthesiology.
Despite these limitations, a major strength of this 2020;132(1):55–68.
4 Evered LA, Goldstein PA. Reducing perioperative neurocognitive
study is its use of a nationwide population-based registry disorders (PND) through depth of anesthesia monitoring: a critical
with detailed baseline information. Lifelong follow-up review. Int J Gen Med. 2021;14:153–162.
5 Moller JT, Cluitmans P, Rasmussen LS, et al. Long-term post-
was possible through the linkage of the registry with operative cognitive dysfunction in the elderly ISPOCD1 study.
the National Cause of Death database. Considering the ISPOCD investigators. International Study of Post-Operative
magnitude and statistical significance of the effects Cognitive Dysfunction. Lancet. 1998;351(9106):857–861.
6 Hussain M, Berger M, Eckenhoff RG, Seitz DP. General anesthetic
observed in the current study, the aforementioned lim- and the risk of dementia in elderly patients: current insights. Clin
itations are unlikely to have influenced our results. Interv Aging. 2014;9:1619–1628.
The incidence of dementia was higher among sur- 7 Jiang J, Jiang H. Effect of the inhaled anesthetics isoflurane, sev-
oflurane and desflurane on the neuropathogenesis of Alzheimer’s
gical patients undergoing GA than those undergoing disease (review). Mol Med Rep. 2015;12(1):3–12.
regional anaesthesia. Moreover, inhalation anaesthesia 8 Chen CW, Lin CC, Chen KB, Kuo YC, Li CY, Chung CJ. Increased
risk of dementia in people with previous exposure to general
was associated with a higher risk of dementia than anesthesia: a nationwide population-based case-control study. Alz-
noninhalation anaesthesia, regardless of age or heimers Dement. 2014;10(2):196–204.

10 www.thelancet.com Vol 55 January, 2023

 Articles

9 Acharya NK, Goldwaser EL, Forsberg MM, et al. Sevoflurane and 23 Sohn JH, Lee JJ, Lee SH, et al. Longitudinal study of the association
Isoflurane induce structural changes in brain vascular endothelial between general anesthesia and increased risk of developing de-
cells and increase blood-brain barrier permeability: possible link to mentia. J Pers Med. 2021;11(11).
postoperative delirium and cognitive decline. Brain Res. 24 Sun M, Chang CL, Lu CY, Zhang J, Wu SY. Effect of opioids on
2015;1620:29–41. cancer survival in patients with chronic pain: a propensity score-
10 Seitz DP, Shah PS, Herrmann N, Beyene J, Siddiqui N. Exposure to matched population-based cohort study. Br J Anaesth.
general anesthesia and risk of Alzheimer’s disease: a systematic 2022;128(4):708–717.
review and meta-analysis. BMC Geriatr. 2011;11:83. 25 Wu SY, Chang CL, Chen CI, Huang CC. Comparison of acute and
11 Jiang J, Dong Y, Huang W, Bao M. General anesthesia exposure chronic surgical complications following robot-assisted, laparo-
and risk of dementia: a meta-analysis of epidemiological studies. scopic, and traditional open radical prostatectomy among men in
Oncotarget. 2017;8(35):59628–59637. Taiwan. JAMA Netw Open. 2021;4(8):e2120156.
12 Lee JJ, Choi GJ, Kang H, et al. Relationship between surgery under 26 Cheng CL, Kao YH, Lin SJ, Lee CH, Lai ML. Validation of the
general anesthesia and the development of dementia: a systematic national health insurance research database with ischemic
review and meta-analysis. BioMed Res Int. 2020;2020:3234013. stroke cases in Taiwan. Pharmacoepidemiol Drug Saf.
13 Bittner EA, Yue Y, Xie Z. Brief review: anesthetic neurotoxicity in 2011;20(3):236–242.
the elderly, cognitive dysfunction and Alzheimer’s disease. Can J 27 Austin PC. Optimal caliper widths for propensity-score matching
Anaesth. 2011;58(2):216–223. when estimating differences in means and differences in pro-
14 Ikeda Y, Ishiguro K, Fujita SC. Ether stress-induced Alzheimer-like portions in observational studies. Pharm Stat. 2011;10(2):150–161.
tau phosphorylation in the normal mouse brain. FEBS Lett. 28 Austin PC. The performance of different propensity score
2007;581(5):891–897. methods for estimating marginal hazard ratios. Stat Med.
15 Planel E, Richter KE, Nolan CE, et al. Anesthesia leads to tau 2013;32(16):2837–2849.
hyperphosphorylation through inhibition of phosphatase activity by 29 Fine JP, Gray RJ. A proportional hazards model for the
hypothermia. J Neurosci. 2007;27(12):3090–3097. subdistribution of a competing risk. J Am Stat Assoc.
16 Bianchi SL, Tran T, Liu C, et al. Brain and behavior changes in 12- 1999;94(446):496–509.
month-old Tg2576 and nontransgenic mice exposed to anesthetics. 30 Yuan Y, Yung Y-F, Stokes M. Propensity score methods for causal
Neurobiol Aging. 2008;29(7):1002–1010. inference with the PSMATCH procedure. Proceedings at the SAS
17 Kim JH, Jung H, Lee Y, Sohn JH. Surgery performed under pro- Global Forum 2017 Conference. Cary, NC: SAS Institute; 2017.
pofol anesthesia induces cognitive impairment and amyloid pa- 31 Dignam JJ, Kocherginsky MN. Choice and interpretation of statis-
thology in ApoE4 Knock-in mouse model. Front Aging Neurosci. tical tests used when competing risks are present. J Clin Oncol.
2021;13:658860. 2008;26(24):4027–4034.
18 Dong Y, Zhang G, Zhang B, et al. The common inhalational 32 Chen PL, Yang CW, Tseng YK, et al. Risk of dementia
anesthetic sevoflurane induces apoptosis and increases beta- after anaesthesia and surgery. Br J Psychiatry. 2014;204(3):
amyloid protein levels. Arch Neurol. 2009;66(5):620–631. 188–193.
19 Whittington RA, Virag L, Marcouiller F, et al. Propofol directly 33 Liu Y, Pan N, Ma Y, et al. Inhaled sevoflurane may promote pro-
increases tau phosphorylation. PLoS One. 2011;6(1):e16648. gression of amnestic mild cognitive impairment: a prospective,
20 Xie Z, Culley DJ, Dong Y, et al. The common inhalation anesthetic randomized parallel-group study. Am J Med Sci. 2013;345(5):
isoflurane induces caspase activation and increases amyloid beta- 355–360.
protein level in vivo. Ann Neurol. 2008;64(6):618–627. 34 Deaton A, Cartwright N. Understanding and misunderstanding
21 Vacas S, Cole DJ, Cannesson M. Cognitive decline associated with randomized controlled trials. Soc Sci Med. 2018;210:2–21.
anesthesia and surgery in older patients. JAMA. 2021;326:863–864. 35 Rortgen D, Kloos J, Fries M, et al. Comparison of early cognitive
22 Velkers C, Berger M, Gill SS, et al. Association between exposure to function and recovery after desflurane or sevoflurane anaesthesia in
general versus regional anesthesia and risk of dementia in older the elderly: a double-blinded randomized controlled trial. Br J
adults. J Am Geriatr Soc. 2021;69(1):58–67. Anaesth. 2010;104(2):167–174.

www.thelancet.com Vol 55 January, 2023 11