Relationship between Surgery under General Anesthesia and the Development of Dementia: A Systematic Review and Meta-Analysis.

Objective To investigate the association between exposure to general anesthesia and the development of Alzheimer's disease (AD) and dementia by reviewing and integrating the evidence from epidemiological studies published to date. Methods We searched MEDLINE, EMBASE, and Google Scholar to identify all relevant articles up to April 2018 reporting the risk of AD/dementia following exposure to general anesthesia and finally updated in February 2020. We included patients older than 60 or 65 years who had not been diagnosed with dementia or AD before the study period. The overall pooled effect size (ES) was evaluated with a random-effect model. Subgroup analyses were conducted and possibility of publication bias was assessed. Results A total of 23 studies with 412253 patients were included in our analysis. A statistically significant positive association between exposure to general anesthesia and the occurrence of AD was detected in the overall analysis (pooled ES = 1.11, 95%confidence interval = 1.07–1.15), but with substantial heterogeneity (pχ2 < 0.001, I2 = 79.4). Although the overall analysis revealed a significant association, the results of the subgroup analyses were inconsistent, and the possibility of publication bias was detected. Conclusion s. This meta-analysis demonstrated a significant positive association between general anesthesia and AD. However, considering other results, our meta-analysis must be interpreted with caution. Particularly, it should be considered that it was nearly impossible to discriminate the influence of general anesthesia from the effect of surgery itself on the development of AD. Further, large-scale studies devised to reduce the risk of bias are needed to elucidate the evidence of association between general anesthesia and AD. Trial registration. PROSPERO International prospective register of systematic reviews CRD42017073790.


Introduction
With the development of medical technology and healthcare systems, life expectancy has increased worldwide. Accordingly, the number of surgeries performed in elderly patients has also been increasing [1]. However, the high morbidity and mortality following surgery are major concerns to the elderly themselves, surgeons, anesthesiologists, and policymakers.
Dementia is a progressive neurodegenerative disease characterized by multiple cognitive impairments that represent a decline from one's previous level of functioning. The global prevalence rate of dementia in people over the age of 60 ranges from 5-7%, and increases rapidly with age, to 20% in people over the age of 85 years [2]. The Global Burden of Disease Study 2015 reported that the global prevalence of dementia has increased from 21.7 million in 1990 to 46.0 million in 2015 [3], and it is expected to be more than 100 million by 2050 [4]. Dementia has become a substantial public health concern because of its progressive, irreversible course and high prevalence in the elderly [5].
Alzheimer's disease (AD) is the most common cause of dementia, accounting for approximately 60-80% of dementia cases [6]. The pathophysiology of AD is still unclear but is mainly associated with accumulation of extracellular betaamyloid plaques and intracellular neurofibrillary tangles, which lead to neuronal cell death and degeneration [6]. There is no single method for the definitive diagnosis of AD. The prevailing guidelines include the Diagnostic and Statistical Manual of Mental Disorders (DSM) and the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA) criteria [7]. However, some physicians diagnose AD based on their clinical knowledge and not on these guidelines [8][9][10][11][12][13].
Previous studies have suggested advanced age, female sex, family history of AD, cardiovascular disease, head trauma, depression, and lower educational level as potential risk factors [6,12,14,15]. Despite these reports, the precise causes and risk factors for AD still remain unclear because of its multifactorial and complex features. Recently, it has been suggested that previous general anesthesia exposure may act as a risk factor for AD in the elderly [16,17]. Although some experimental data obtained in cells and animal models have suggested a significant association between general anesthetics, especially volatile agents [18], and AD, evidence from human studies is controversial.
There have been several attempts to integrate the evidence from various epidemiological studies, such as the reanalysis of eight case-control studies [19] and two metaanalyses [20,21]. Even though they consistently showed no evidence of an association between general anesthesia and dementia/AD, an updated meta-analysis is needed because the previous meta-analyses incorporated non-peer-reviewed articles and studies using excessively broad outcome measures. Moreover, two of the previous meta-analyses [19,21] included only case-control studies, which are prone to bias. We critically reviewed and synthesized the current evidence to determine the association between administration of general anesthesia and development of AD and to verify whether general anesthesia acts as a risk factor for AD in the elderly.

Methods
The protocol for this review has been registered in the PROS-PERO network (registration number: CRD42017073790) and published in a peer-reviewed journal [22]. This systematic review and meta-analysis of the association between general anesthesia and the development of AD were performed according to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines [23] and reported according to the guidelines of Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) [24].

Search Strategy. A search was performed by two different investigators independently in MEDLINE, EMBASE, and
Google Scholar for articles up to April 2018 using search terms related to AD, dementia, and general anesthesia and updated in February 2020. The search terms used in MED-LINE and EMBASE are presented in Appendix S1. In order to identify all relevant articles, we scanned the reference lists of the original papers until no further relevant references could be found. No language or date restrictions were applied.

Selection Criteria.
Our selection criteria are as follows.

Study Design.
Peer-reviewed cohort and case-control studies including nested case-control studies were eligible for inclusion. We excluded data from proceedings, letters to the editor, posters, commentaries, laboratory science studies, and any other nonrelevant studies.

Population.
Inclusion criteria for study populations were as follows: (1) the elderly (defined as more than 60 or 65 years old) from all countries and (2) those who had not been diagnosed with dementia or AD before the beginning of the study period. If the study's definition of the elderly was other than being older than 60 or 65 years of age, an attempt was made to contact the study authors to obtain the relevant information. When unsuccessful, we performed a pooled analysis including the data of that study first, and then, we performed a sensitivity analysis excluding the data. No restrictions were applied in terms of sex, race/ethnicity, or socioeconomic status.

2.2.3.
Exposure. Exposure to general anesthesia for surgery, usually using inhalation anesthetics, was included. Intravenous anesthesia, spinal anesthesia, epidural anesthesia, and regional anesthesia were excluded. If an article reported on general anesthesia including intravenous, spinal, epidural, or regional anesthesia along with inhalation anesthesia, we tried to contact the study authors to obtain information on general anesthesia using inhalation anesthetics. When unsuccessful, we first analyzed the data on general anesthesia, including intravenous, spinal, epidural, or regional anesthesia, and then we performed sensitivity analysis excluding the data. The source of exposure assessment was also collected.

Comparison.
Comparison groups included individuals with no history of general anesthesia. If a study only reported previous anesthesia history during the study period, we tried to contact the study authors for further information about previous anesthesia history prior to the study period. When unsuccessful, the reported information was used for our analysis. If a study investigated the associations of AD and general anesthesia using two or more comparison groups and reported each outcome separately, pooled estimates of associations for these groups were calculated and used for analysis.

Outcome Measures.
To cover as many AD cases as possible, we included not only AD cases but also dementia cases, of which AD cases comprise the largest portion, diagnosed by standard criteria such as the DSM or clinically diagnosed by a professional physician. Studies reporting effect size (ES) as odds ratio (OR), relative risk (RR), or hazard ratio (HR) of dementia/AD to general anesthesia exposure were included. In some studies, which reported only the number of individuals with and without AD instead of ES, we obtained the ES by calculation from the data provided. When the study reported only dementia cases without distinguishing AD [25][26][27][28], we attempted to contact the study authors to obtain the classified data. If the attempt was unsuccessful, we conducted a pooled analysis and examined the difference by subgroup analysis. The titles and abstracts identified through the search strategy were scanned independently by two investigators. To minimize data duplication as a result of multiple reporting, papers from the same author were compared. For reports determined to be eligible based on the title or abstract, the full paper was retrieved. Potentially relevant studies chosen by at least one investigator were retrieved and evaluated in full-text versions. Articles meeting the inclusion criteria were assessed separately by two investigators, and any discrepancies were resolved through discussion. In cases where agreement could not be reached, the disputes were resolved with the help of a third investigator.
2.4. Data Extraction. Using a standardized extraction form, the following data were extracted independently by two reviewers: study name (along with the name of the first author and year of publication); region where the study was conducted; study design; source from which subjects were selected; age of subjects; exposure definition; method of data collection (self-reported vs. medical records); outcome definition; ES such as OR, RR, and HR with 95% confidence intervals (CIs); methods for controlling covariates and the confounding variables controlled for; number of cases/controls or cohort groups; and total number of participants. If information was missing, an attempt was made to contact the study authors to obtain the relevant information. When unsuccessful, missing information was calculated if possible from the relevant data in the study. As the ES was not reported or needed to be integrated because of the multiple groups, it was calculated in six studies [8,12,13,[29][30][31] using the relevant data. The reference lists were divided in half, and two reviewers completed the data extraction for each half of the list. Then, data extraction forms were cross-checked to verify the accuracy and consistency of extracted data.

Study Quality
Assessment. The quality of the studies was independently assessed by two investigators using the Risk of Bias Assessment Tool for Nonrandomized Studies (RoBANS) [32]. The quality of each study was evaluated according to the following six domains: the selection of participants, confounding variables, the measurement of exposure, the blinding of the outcome assessments, incomplete outcome data, and selective outcome reporting. The methodology of each study was graded as "high," "low," or "unclear" to indicate high risk of bias, low risk of bias, and unclear risk of bias. Any discrepancies were resolved through discussion.
If an agreement could not be reached, the dispute was resolved with the help of a third investigator.

Study Selection and Characteristics.
A total of 2784 articles were obtained after searching the databases and references and through the manual search ( Figure 1). After excluding the duplicates (n = 8), we reviewed the remaining articles (n = 2776); 2730 of the articles did not meet the selection criteria. The remaining 46 articles were selected for review of their whole content. The kappa value for selecting articles between the two reviewers was 0.756. Of the 46 studies selected for review of their whole content, 23 were excluded for the following reasons: One was a conference proceeding [36], 9 were reviews [17,[37][38][39][40][41][42][43][44], 3 were editorials [45][46][47], 1 was a consensus statement [48], 2 were meta-analyses [19,20], and 7 did not present the appropriate data [49][50][51][52][53][54][55]. Some of the studies required additional consultation among the reviewers to determine whether they should be included [8,9]. One study was about comparing the risk of AD following coronary artery bypass graft (CABG) and percutaneous transluminal coronary angioplasty (PTCA) [9]. Considering that CABG is a surgery requiring general anesthesia and PTCA is not, we decided to include this study in our pooled analysis. Another study investigated whether spine surgery contributes to the development of AD [8]. There were no details about the anesthesia procedure. However, the spine surgery ranged from discectomy to complex spine fusion procedures, all typically performed under general anesthesia. Therefore, we also included this study in our research. Therefore, 23 studies involving a total of 412253 patients were included in our analysis. To obtain more relevant information than reported, we tried to contact some authors of the included studies. Among the contacted authors, one [27] provided additional data about the study population sufficient for our complete analysis.
3.2. Study Quality Assessment. Overall risks of bias evaluated using the RoBANS are shown in Table 2. One study was assessed as having unclear risk of bias in selection of participants because of the lack of mention about the process of evaluating the cognitive function of controls with Parkinson's disease and nondegenerative neurological disease or verifying the absence of dementia [29]. We rated the studies which used self-or surrogate-reported data as "unclear" in the domain of measurement of exposure. As we could not find any protocol from the studies to compare, we graded every study as having unclear risk of bias in selective outcome reporting.

Studies included in quantitative synthesis
(meta-analysis) (n = 23)          As the heterogeneity for ES was considerable, metaregression was conducted to determine the origin of heterogeneity. According to metaregression, the study design, region, exposure assessment, case ascertainment, definition of dementia or AD, and year of publication were not likely to be a source of heterogeneity (Table 4).  13 BioMed Research International I 2 = 74:8) showed a significantly higher risk of AD in people who had been exposed to general anesthesia, but with substantial heterogeneity (Figure 2). 95% prediction interval was 0.93-1.27 and 0.81-1.20 in cohort studies and case-control studies, suggesting any further study could alter the significance of association between general anesthesia and AD.
3.5. Sensitivity Analysis. Sensitivity analysis was performed by excluding one study at a time; no change in statistical significance occurred.

Discussion
The effect of general anesthetics, especially inhalation agents, on neurocognitive function is currently controversial; it is unknown whether general anesthetics are neurotoxic or neuroprotective [64]. Regarding neurotoxicity, numerous in vitro and in vivo studies with cells, tissues, animals, and biomarkers have suggested that volatile anesthetics may contribute to the neuropathogenesis of AD. However, evidence from human studies is very weak. This systematic review and meta-analysis were designed to investigate the association between general anesthesia and the risk of AD in human studies. By integrating the conflicting evidence documented to date, we detected a statistically significant association between exposure to general anesthesia and the occurrence of AD. This significant positive association was consistent in both cohort studies and case-control studies. In addition, the result of subgroup analysis of studies based on medical records, a trustworthy data source, supported the significant positive association between general anesthesia and AD. Notably, this outcome was contrary to the results of two previous meta-analyses [20,21] and a reanalysis [19], which reported no evidence of a significant relationship between general anesthesia and dementia. The differences between our investigation and the previous studies may explain the contradictory results. First, our meta-analysis includes recently published literature. The meta-analysis published in 2011 [21] included only case-control studies, which are susceptible to bias. Although another meta-analysis of epidemiological studies up to April 2017 was conducted lately [20], a new large-scale population-based prospective cohort study was published in April 2018 [27]. Including this study, we have evaluated six more relevant studies [26,27,31,60,62,63] than the previous meta-analysis. Second, there were some differences in study selection between our analysis and the others. We included only peer-reviewed cohort and casecontrol studies to ensure the reliability of the evidence. Therefore, some of the studies which had been included in the previous meta-analyses were excluded because they were non-peer-reviewed articles [36,44]. Several other studies have been excluded from our analysis because they did not meet our inclusion criteria: One study was about the risk of AD following occupational exposure to anesthetic gases [54], one included not only dementia but also mild cognitive impairment as outcomes [53], and one had a completely different study purpose than ours [55].
Given the various designs and methods used in the involved studies, we performed subgroup analyses to better understand the results of our meta-analysis. In the subgroup analysis based on exposure assessment, an increased risk of AD following exposure to general anesthesia was observed in studies using medical records. Because medical records are more objective and reliable data source than selfreported data collected through interview, this result 14 BioMed Research International enhances the validity of the positive association between general anesthesia and AD observed in overall analysis. In addition, as shown in Figure 3, total weight of the studies based on medical record was about 90%. Thus, the positive association detected from the studies based on medical records seems to have contributed significantly to the overall outcome. Although a low risk of AD was observed in the subgroup of studies based on self-or surrogate-reported data, it might be questionable since interview-based data that relies on per-son's memory is prone to involve recall bias. Elderly subjects, especially those who have memory decline, and their family members may have difficulty in remembering the details of medical history. However, according to the subgroup analysis based on case ascertainment, a negative association was observed in studies using standard criteria. Standard criteria are usually considered more reliable diagnostic tool because it is less likely to involve physician's subjectivity compared to clinical  15 BioMed Research International diagnosis. This finding suggests that the results of the overall analysis must be interpreted with caution.
Considering that AD cases comprise the largest portion of dementia cases (60-80% in one study) [6], we included studies on all-cause dementia in order to collect as many AD cases as possible. In subgroup analysis according to outcome definition, a stronger positive association was detected among the studies that defined cases as all-cause dementia, compared to the total analysis. On the other hand, studies that collected only AD cases showed a negative association between previous anesthesia exposure and AD. In the current diagnostic criteria, AD is diagnosed by excluding evidence of vascular dementia. Recently, however, the concept of mixed dementia (mixed vascular-Alzheimer dementia) has been emerging [65], and currently used diagnostic criteria may yield lower incidence rate of AD than the actual value. Recent  Figure 4: Forest plot for subgroups according to outcome definition: Alzheimer's disease and all-cause dementia. An increased risk of dementia following general anesthesia was detected among all-cause dementia group, but with considerable heterogeneity. However, decreased risk of AD with general anesthesia exposure was detected among the studies with outcome definition limited to only Alzheimer's disease. 16 BioMed Research International studies have suggested that about half of older adults with dementia have pathological evidence of more than one cause of dementia [6]. Besides, under certain circumstances, surgery and anesthesia are risk factors for cerebral ischemia [66]. These factors may have increased the differences between the risk of AD and all-cause dementia including AD and mixed vascular-Alzheimer dementia following surgery under general anesthesia. Additionally, these factors might have contributed to the statistically negative associa-tion between general anesthesia and AD observed in studies using standard diagnostic criteria.
One of the factors important in outcome assessment is lag time, which means the latency period before the diagnosis of disease. In studies without including the lag time, the patients with incidental dementia not severe enough to be diagnosed may be misinterpreted as having dementia caused by general anesthesia exposure. As highlighted in the metaanalysis conducted to verify the association between  Figure 5: Forest plots for subgroups based on case ascertainment: standard criteria and clinical diagnosis. Whereas the pooled effect size of the studies, which clinically diagnosed dementia, showed a significant positive association between general anesthesia and dementia, inversely negative association was observed among the studies using standard diagnostic criteria of dementia.
17 BioMed Research International dementia and benzodiazepine which was published in 2018 [67], analysis considering lag time is important enough to influence the statistical significance of the findings. However, among the 23 studies included in our analysis, there was only one study introduced the lag time to their methodology [31]. Although the cohort studies excluded the patients with diagnosis of dementia or AD prior to enrollment, it was difficult to find the details about ensuring that dementia or AD was not present before the exposure of general anesthesia in the case-control studies. Thus, even though the positive association was detected from our findings, it is insufficient to verify the causality between general anesthesia and AD because of the uncertainty of time sequence and the lack of considering lag time.   Figure 6: Contour-enhanced funnel plots of total studies (a) and case-control studies (b). Asymmetry was observed in funnel plots of both total studies and case-control studies. 18 BioMed Research International Therefore, although a significant association between general anesthesia exposure and AD was observed, the results of this meta-analysis should be interpreted with caution. The possibility of publication bias was demonstrated by contourenhanced funnel plots and Egger's test. The statistical significance disappeared after trim and fill analysis. In addition, the 95% prediction interval computed from overall studies implied that any future study could alter the statistical significance of association between general anesthesia and AD. These findings support the need for prudence to properly interpret the results of this meta-analysis.
Our study has some limitations. First, substantial heterogeneity was observed among the included studies. Although various approaches were attempted to reveal the origin of heterogeneity, they were not successful. Second, because of insufficient details on the context of anesthesia including number of exposure, the agent used for induction and maintenance, dose and duration of exposure, and intraoperative events or perioperative complications such as hemodynamic instability and hypoxia, we could not conduct analysis adjusting for these variables. Moreover, surgery itself might contribute to an increased risk of AD. Although there are no data on the risk of AD, high-risk surgeries such as cardiac surgery have been reported as raising the risk of cognitive impairment such as delirium and postoperative cognitive dysfunction [42,68]. However, it is nearly impossible to analyze the effect of anesthesia and surgery separately, since anesthesia is not performed alone. Third, contour-enhanced funnel plots and Egger's test suggested the possibility of publication bias derived from studies that are not published in the current literature because of null results or small sample size.
According to the limitations discussed, additional welldesigned studies are necessary to clarify the relationship between general anesthesia and dementia or AD. Future studies are recommended to be conducted with the following considerations: large-scale study with adequate statistical power; prospective cohort studies with long-term follow-up including lag time, ensuring the subjects free of dementia before exposure of anesthesia; using reliable data source such as medical records; reporting sufficient details on the characteristics of exposure such as number of exposure, agents, and doses used; adequate adjustments for confounding variables; and outcome assessment with standard diagnostic tool of dementia. Recent review published by American scholars reported a wide variety of diagnostic methods for dementia used in large cohort studies, while emphasizing the need for the development of well-described and reproducible methods for diagnosing dementia in epidemiologic studies [69]. We believe that following standardized and reproducible process will reduce the heterogeneity and enhance the comparability of studies.
Despite the limitations above, our systematic review and meta-analysis demonstrate the strengths of a rigorous methodology based on a published, preplanned protocol to provide evidence of the relationship between general anesthesia exposure and risk of AD. Furthermore, our study has a value of suggesting general anesthesia exposure as a potential risk factor of AD and raising the necessity of further related stud-ies for better management of the elderly. Recently, there was a meta-analysis reported the significantly high mortality rates of the patients with dementia after undergoing hip fracture surgery, emphasizing the importance of perioperative care for dementia [70]. Although dementia is not a fatal condition, it can seriously degrade one's quality of life and even deteriorate the prognosis after surgery. In an aging society, further research for better understanding of the risk factors of dementia and perioperative managing of dementia patients are becoming valuable [71,72].

Conclusion
In conclusion, we observed a significant association between exposure to general anesthesia and an increased risk of AD. However, considering the substantial heterogeneity, evidence of publication bias, and inconsistent results of the subgroup analyses, the results of our meta-analysis should be interpreted with caution. Moreover, it was nearly impossible to discriminate the influence of general anesthesia from the effect of surgery itself on the development of AD. Further, large-scale prospective cohort studies designed to reduce the risk of bias considering lag time, using standardized methods and reliable data with adequate adjustments of confounding factors, are needed to elucidate the evidence of an association between general anesthesia and AD.

Data Availability
The study data supporting this systematic review and metaanalysis are from previously reported studies and datasets, which have been cited. The processed data are available in the article.

Conflicts of Interest
The authors declare no conflicts of interest.