Alzheimer’s Dementia after Exposure to Anesthesia and Surgery in the Elderly: A Matched Natural Experiment using Appendicitis

Abstract

Objective:

To determine if surgery and anesthesia in the elderly may promote Alzheimer’s Disease and Related Dementias (ADRD).

Background:

There is a substantial conflicting literature concerning the hypothesis that surgery and anesthesia promotes ADRD. Much of the literature is confounded by indications for surgery or has small sample size. This study examines elderly patients with appendicitis, a common condition that strikes mostly at random after controlling for some known associations.

Methods:

A matched natural experiment of patients undergoing appendectomy for appendicitis versus control patients without appendicitis using Medicare data from 2002 to 2017, examining 54,996 patients without previous diagnoses of ADRD, cognitive impairment, or neurological degeneration, who developed appendicitis between ages 68 through 77 years and underwent an appendectomy (the “Appendectomy” treated group), matching them 5:1 to 274,980 controls, examining the subsequent hazard for developing ADRD.

Results:

The hazard ratio (HR) for developing ADRD or death was lower in the Appendectomy group than controls: HR = 0.96 (95% CI 0.94, 0.98), P < 0.0001, (28.2% in Appendectomy versus 29.1% in controls, at 7.5 years). The HR for death was 0.97 (0.95, 0.99), P = 0.002, (22.7% versus 23.1% at 7.5 years). The HR for developing ADRD alone was 0.89 (0.86, 0.92), P < 0.0001, (7.6% in Appendectomy versus 8.6% in controls, at 7.5 years). No subgroup analyses found significantly elevated rates of ADRD in the Appendectomy group.

Conclusions:

In this natural experiment involving 329,976 elderly patients, exposure to appendectomy surgery and anesthesia did not increase the subsequent rate of ADRD.

MINI-ABSTRACT

There is concern that surgery and anesthesia in the elderly may promote Alzheimer’s Dementia. We utilize a natural experiment, examining 54,996 Medicare patients with appendicitis, a condition that strikes mostly at random, matching them to 274,980 controls, and tracking their subsequent occurrence of Alzheimer’s Dementia up to 15 years later.

INTRODUCTION

Animal studies have long suggested a link between anesthetic exposure and pathological findings that resemble those found in Alzheimer’s Disease.^1-6 In humans, there is both literature suggesting anesthesia, or surgery and anesthesia, may be associated with Alzheimer’s disease,^7-14 and studies failing to support this association.^15-18 To date, there have been no large-scale randomized trials. A randomized study of Alzheimer’s and surgery or anesthesia would face practical and ethical obstacles, not least the need for a very large trial with extensive follow-up for a long time to determine the rates of cognitive decline.

The present study is a natural experiment. Appendicitis occurs haphazardly in elderly individuals who are otherwise not necessarily extremely ill; moreover, with appropriate surgery, recovery is expected without continuing morbidity or treatment. In this sense, appendectomy is quite unlike surgery to remove a malignancy or surgery to transplant a kidney. The typical case of appendicitis in Medicare resembles the typical person in Medicare, but likely received about an hour of additional general anesthesia. There are a few known predictors of appendectomy---age, sex, race, and possibly some conditions of the colon --- but we match for these, so matched groups are comparable in terms of these risk factors as well. This study follows through time 54,996 elderly patients in the Medicare system who developed appendicitis and received an appendectomy (and are therefore exposed to surgery and anesthesia), and 274,980 (= 5x54,996) similar matched patients to these exposed patients who did not have an appendectomy with control non-appendicitis status determined by looking back three years from the date of the exposed patient’s appendectomy using Medicare claims.

METHODS

This research study was approved by the Children's Hospital of Philadelphia Institutional Review Board.

Conceptual Model

Although appendicitis occurs unpredictably, there are a few well-known associations with appendicitis that are not random. Therefore, we exactly matched controls to appendicitis patients for variables sometimes associated with appendicitis, such as year of birth, race, and sex^19-21 and chronic colon diseases such as inflammatory bowel disease ²² and diverticular disease.²³ We followed patients forward in time to identify ADRD, with follow-up ranging from five to 15 years. We also matched on conditions thought to potentially be associated with ADRD, such as diabetes,²⁴ hypertension,²⁵ and even the possibility that prior hospitalizations themselves add to the risk of subsequently developing ADRD.^{14, 26, 27}

Patient Population

We were granted access via the Centers for Medicare & Medicaid Services Virtual Research Data Center (VRDC) to patient beneficiary summary files, administrative claims (Inpatient, Outpatient, Carrier/Part B, and Hospice files) for fee-for-service Medicare beneficiaries and assessment files for nursing home residents. We analyzed patients ages 68 through 77 years who developed appendicitis and underwent an appendectomy along with matched controls and followed them for at least five years. After setting aside a 1% random sample to test claims-based definitions, we studied the remaining 99% random sample of all Medicare beneficiaries. We excluded patients if they lacked fee-for-service Medicare in the three years prior to inclusion or if they did not have adequate follow-up and look-back information. Details of the sample construction and ICD9-CM code lists are provided in Supplemental Digital Content 1, Section 1 (eTable 1.1). We also excluded patients with a previous history of cognitive decline, neuro-degeneration, dementia, or diagnoses related to ADRD (see Supplemental Digital Content 1, eTables 2.1 and 2.2).

Definitions

Patient characteristics were defined using a three-year look-back window from the time of the index appendectomy event, examining Inpatient, Outpatient, Carrier/Part B, Skilled Nursing Facility, Durable Medical Equipment, Home Healthcare, and Hospice files. A treated patient was defined as a patient who underwent an appendectomy associated with appendicitis between the ages of 68 through 77. ICD9-CM codes for defining an appendectomy and appendicitis are provided in Supplemental Digital Content 1, Section 1 (eTable 1.1).

ADRD was defined using Medicare claims prior to analyzing results of this study. The list of diagnosis codes reflecting ADRD is provided in Supplemental Digital Content 1, eTable 2.1. For each appendicitis exposed patient, and their matched controls, we scanned forward in time starting six months after the exposed surgical event and identified whether any codes for ADRD were observed, and whether the patient stayed at least a total of six months in a nursing home. If a patient had an ADRD code that was confirmed by either a second code separated by at least one year, or a total of six-months of nursing home stay, then we defined the patient as having developed ADRD. The time of onset was defined as when the first ADRD diagnosis was observed. Details of the development and validation of our ADRD claims-based definition are found in Supplemental Digital Content 1, Section 2.

Comorbidities were defined using ICD9-CM codes from previous work.^28-30 These definitions are provided in Supplemental Digital Content 1, Section 2, eTable 2.3.

Matching Methods

Matching was performed on a national scale in three-month increments from 2002 to 2012 (44 quarters). Note we had Medicare claims on all patients in the match, allowing for a minimum potential follow-up of five years for those undergoing appendectomy in 2012, and their matched controls. Moving forward in time, we identified all eligible patients in a time increment who underwent an appendectomy for appendicitis (hereon referred to as a “treated” or “exposed” patient) and found the best five controls per treated patient, controlling for both age and year of birth. Control patients did not have evidence of appendicitis and appendectomy prior to matching; that is, the match was a risk-set match.^{31, 32} We assigned patients to the control group if they did not have a principal procedure of appendectomy with a principal diagnosis of appendicitis. A small fraction (0.065% (=216/329,976)) of patients had a prior appendectomy in Medicare claims. We performed a stability analysis omitting matched pairs that displayed any prior appendectomies in Medicare claims and our results were unchanged. Of course, as is discussed later, there is always the possibility that patients underwent appendectomies prior to entering Medicare. The matching was implemented through the NETFLOW procedure in SAS.³³ Details of the matching algorithm can be found in Supplemental Digital Content 1, Section 3, eTables 3.1 and 3.2.

Matching was completed before outcomes were examined.³⁴ We aimed to attain standardized differences in covariate means below 0.1 standard deviations after matching, though below 0.2 is a traditional standard.^{35, 36} We also assessed balance using two-sample randomization tests, specifically the Wilcoxon rank-sum test for each continuous covariate and Fisher’s exact test for each binary one, thereby comparing the balance achieved by matching to the balance expected from complete randomization.³⁷

A three-year lookback prior to the beginning of the quarter was used to define comorbidities (Supplemental Digital Content 1, eTable 2.3). Demographic information was determined from the last quarter of the lookback (closest to the start of the matching quarter). We matched exactly on year of birth, race, sex, history of inflammatory bowel or diverticular disease, and other variables. The matching distance included history of liver disorders; colon cancer; gastrointestinal cancer; stroke or cerebrovascular diseases; the number of inpatient admissions during the three-year lookback period; and other variables. To facilitate matching an extremely large pool of controls who did not undergo appendectomy to treated appendectomy patients, we split potential controls into smaller subgroups based on the work of Zhang et al.³⁸ (see Supplemental Digital Content 1, Section 3). We also defined two geographic regions based on the Medicaid uptake of each state between 2005 and 2010³⁹ (see Supplemental Digital Content 1, eTable 3.3).

Statistical Methods

The primary outcome was ADRD or death, as identified by claims. We also examined separately the specific hazards of mortality and ADRD using Cox’s proportional hazards model and cumulative hazard plots. We modeled the hazard ratio associated with treatment status after each match using the matched version of the proportional hazards model.⁴⁰ These models were also used to examine treated vs. control hazard ratios for patients in specific subsets, such as by sex, by race, and by history of colon disease. When examining outcomes at 5, 7.5, and 10 years, results reflect the weighted estimates of rates, with each matched set receiving weights equal to the number of eligible appendectomy patients in that set.⁴¹ P-values reflect results from performing the Mantel-Haenszel test stratified by matched set.⁴² Rates at 5, 7.5, and 10 years represent the number of events out of the total number of people who had the potential for complete records during the respective entire time period. These were individuals in which after 5, 7.5, or 10 years would still have the potential for records to be available (records analyzed are through 2017). Individuals that did not have complete Part A and B or joined an HMO during the 5, 7.5, or 10 years were not part of the yearly analysis.

Role of the Funding Source

The National Institute on Aging had no role in the study design; collection, analysis, or interpretation of the data; the writing of the report; or the decision to submit for publication.

RESULTS

When describing the results of the study, we will first report the entire exposed or treated group, and compare these patients to their matched controls.

The Quality of the Matched Sets

Table 1 reports the quality of the matches for selected covariates, while results on all matching variables and variables not specified in the match are displayed in eTable 3.2 in Supplemental Digital Content 1. After matching, we observed no standardized differences larger than 0.2 standard deviations, and most were below 0.1 standard deviations. Variables that were exactly matched, such as year of birth, race and sex had zero standardized differences between treated and control groups. Variables that were not always exactly matched, but were included in the matching algorithm, had standardized differences less than 0.1 SD’s, such as Cardiac Disorders, displaying 42.29% in the treated group and 42.94% in the controls, with a standardized difference between groups of −0.01 and a P-value of 0.005. Despite the small differences in means and rates between appendectomy and matched control groups, some significant differences were observed. However, it should be noted that the sample sizes were very large, so even small and acceptable standardized differences between groups may achieve statistical significance.

Table 1: Comparison of Selected Patient Characteristics between the Appendectomy Treated and Matched Control Groups.

The full list of patient covariates (with expanded definitions) is provided in the Supplemental Digital Content 1, Table e3.2.

Variable	Mean Appendectomy	Mean Control	Standardized Difference
N Patients	54,996	274,980
Year of birth	1934.24	1934.24	0.00
Age at match (years)	72.60	72.60	−0.00
Male (%)	50.41	50.41	0.00
Female (%)	49.59	49.59	0.00
White non-Hispanic (%)	91.65	91.65	0.00
Black (%)	3.68	3.68	0.00
Hispanic (%)	1.05	1.05	0.00
Dyslipidemia (%)	80.10	81.63	−0.04
Hypertension (%)	77.04	78.49	−0.03
Cardiac disorders (%)	42.29	42.94	−0.01
Coronary artery disease (%)	31.62	32.46	−0.02
Arrhythmias (%)	29.75	29.66	0.00
Valvular heart diseases (%)	20.45	20.77	−0.01
Heart failure (%)	6.19	5.60	0.03
Lung disorders (%)	35.84	35.75	0.00
Diabetes (%)	28.33	28.35	0.00
Colon disorders (%)	25.29	25.29	0.00
Endocrine disorders (%)	23.65	22.46	0.03
Cancer (%)	21.33	19.77	0.04
Cerebrovascular disorders (%)	12.46	12.45	0.00
Liver disorders (%)	8.06	7.58	0.02
Kidney disorders (%)	7.43	6.90	0.02
Physician office visit count+	18.97	18.55	0.03
Anesthesia minutes+	33.26	34.98	−0.02
Inpatient admission count+	0.62	0.61	0.01
Total procedure count+	0.36	0.37	−0.01
ED visit count+	0.28	0.28	0.00
Principal procedure count+	0.23	0.24	−0.01
Median income ($, mean)	61,514	61,160	0.01
Neighborhood high school graduation (mean of Census %)	88.13	88.20	−0.01
Neighborhood percent below poverty line (mean of Census %)	10.09	9.95	0.02
Dual eligible (%)+	8.23	7.74	0.02

⁺During prior 3 years

Comparing Outcomes Between Appendectomy Treated and Control Patients

Table 2 provides the main results of the study across all matched sets, comparing the hazards of death and ADRD for the appendectomy patients and their matched controls. In this study, censoring occurred for one of two reasons: the end of the follow-up period, December 31, 2017, or the individual entered an HMO, where we do not observe fee-for-service data to identify ADRD claims. P-values refer to the hazard ratio coefficient from Cox’s stratified proportional hazards model.

Table 2: Comparison of outcomes between treated and control patients.

Treated patients (N = 54,996) underwent appendectomy. Controls (N = 274,980) were matched on clinical characteristics in the three years prior to the appendectomy event.

Outcome#	Cox Model Results			At Year 5++			At Year 7.5			At Year 10
	Hazard Ratio+	(95% CI)	P-value	Appendix Treated N=47,876	Matched Controls N=237,431	P-value	Appendix Treated N=36,361	Matched Controls N=179,841	P-value	Appendix Treated N=25,098	Matched Controls N=124,285	P-value
ADRD or Death	0.958	(0.942, 0.975)	<.0001	17.07%	17.60%	0.0032	28.19%	29.11%	0.0002	40.07%	41.91%	<.0001
Death	0.972	(0.954, 0.990)	0.0022	13.47%	13.60%	0.4085	22.69%	23.13%	0.0465	33.28%	34.52%	0.0001
ADRD	0.891	(0.863, 0.920)	<.0001	4.28%	4.86%	<.0001	7.64%	8.59%	<.0001	11.42%	12.87%	<.0001

⁺The hazard ratio represents the Cox model stratified by the matched sets, with the reference group being control (non-appendectomy or untreated) group. P-value reflects significance of the hazard ratio coefficient from the Cox model.

⁺⁺These rates reflect status at 5, 7.5, and 10 years (they are the weighted estimates of rates, with each matched set receiving weights equal to the number of eligible appendectomy patients in that set). N at top of each column provides the numbers who are still eligible for analysis after 5, 7.5, and 10 years.

Note

^#:In the Cox model output (first three columns).

ADRD or Death is the hazard of either ADRD or DEATH prior to going to HMO or losing Part A or B or prior to reaching the end of 2017 data.

Death is the hazard of dying prior to going to an HMO or losing Part A or B, or reaching the end of 2017 data.

ADRD is the hazard of getting ADRD prior to death, going to HMO or losing Part A or B, or reaching the end of 2017 data.

For the yearly 5, 7.5, and 10-year rates, the outcomes are defined as follows:

ADRD or Death is the rate of ADRD or DEATH for all people who did not go into HMO or lose Part A or B or reach the end of 2017 data before 5, 7.5, and 10 years.

Death is the rate of death for all people who did not go into HMO or lose Part A or B or reach the end of 2017 data before 5, 7.5, and 10 years.

ADRD is the rate of ADRD for all people who did not go into HMO or lose Part A or B or reach the end of 2017 data before 5, 7.5, and 10 years.

We also report results as rates at 5, 7.5, and 10 years for the subset of individuals observed without censoring for these periods. As can be seen, we always see slightly lower ADRD events in the appendectomy group, so there is no indication that the anesthesia accompanying appendectomy is promoting ADRD.

For all patients, the hazard ratio (HR) for developing ADRD or death was slightly lower in the Appendectomy group than controls: HR = 0.96 (95% CI 0.94, 0.98), P < 0.0001, (28.2% in Appendectomy versus 29.1% in controls, at 7.5 years). The HR for death was 0.97 (0.95, 0.99), P = 0.002, (22.7% in Appendectomy versus 23.1% in controls, at 7.5 years). The HR for developing ADRD alone was 0.89 (0.86, 0.92), P < 0.0001, (7.6% in Appendectomy versus 8.6% in controls, at 7.5 years).

Subgroup Analyses

We explored various subgroups to determine if there were some effects suggesting the appendectomy group was associated with worse ADRD outcomes. In Table 3 we provide results by age group (68 through 72 versus 73 through 77); sex; race (non-Hispanic White versus Black); and in eTable 4, we provide results for history of colon disease (see eTable 4.4); surgical presentation (see Table 4.5: complicated appendectomy versus uncomplicated); and the period of the appendectomy procedure (Early versus Late, see eTable 4.6): Early was defined as an appendectomy performed from January 1, 2002 to June 30, 2007. Late was defined as an appendectomy performed from July 1, 2007 to December 31, 2012. In all subgroups, we never observed a significantly higher hazard of ADRD for the Appendectomy group versus their controls. For all subgroups we examined the same balance table as presented in Table 1 and found, as in our primary analysis, that no standardized differences for any variables exceeded 0.09 SDs across treatment and control groups. As this demonstrates, an advantage of our natural experiment and matching approach is that subgroups exhibit the same excellent comparability as our overall analysis, so comparisons within subgroups can proceed without concern.

Table 3: Subgroup Analyses: Comparison of outcomes between treated and control patients inside five subgroups.

Treated patients underwent appendectomy. Controls were matched on clinical characteristics in the three years prior to the appendectomy event. Expanded results are provided in Supplemental Digital Content 1, Table e4.1a through Table e4.6b.

Subgroup/Outcome	Cox Model Results				At Year 7.5
	Subgroup N	Hazard Ratio	(95% CI)	P-value	Appendix Treated N	Matched Controls N	Appendix Treated	Matched Controls	P-value
AGE at Appendectomy
ADRD or Death Age 68-72	186,012	0.957	(0.933, 0.981)	0.0005	20,037	99,323	23.06%	23.81%	0.0203
ADRD or Death Age 73-77	143,964	0.960	(0.937, 0.983)	0.0006	16,324	80,518	34.49%	35.61%	0.0034
Death Age 68-72	186,012	0.972	(0.946, 0.998)	0.0348	20,037	99,323	19.01%	19.16%	0.6002
Death Age 73-77	143,964	0.972	(0.948, 0.997)	0.0264	16,324	80,518	27.20%	28.01%	0.0234
ADRD Age 68-72	186,012	0.870	(0.829, 0.914)	<.0001	20,037	99,323	5.43%	6.33%	<.0001
ADRD Age 73-77	143,964	0.907	(0.869, 0.946)	<.0001	16,324	80,518	10.36%	11.36%	0.0002
SEX
ADRD or Death Males	166,332	1.002	(0.979, 1.025)	0.8906	18,464	91,495	32.73%	32.53%	0.6241
ADRD or Death Females	163,644	0.907	(0.884, 0.931)	<.0001	17,897	88,346	23.51%	25.58%	<.0001
Death Males	166,332	1.011	(0.987, 1.035)	0.3776	18,464	91,495	27.89%	27.48%	0.2784
Death Females	163,644	0.919	(0.893, 0.946)	<.0001	17,897	88,346	17.32%	18.65%	<.0001
ADRD Males	166,332	0.948	(0.905, 0.994)	0.0269	18,464	91,495	7.14%	7.73%	0.0071
ADRD Females	163,644	0.847	(0.811, 0.884)	<.0001	17,897	88,346	8.16%	9.48%	<.0001
RACE
ADRD or Death White	302,430	0.957	(0.940, 0.974)	<.0001	33,721	166,822	27.92%	28.83%	0.0004
ADRD or Death Black	12,156	0.962	(0.879, 1.052)	0.3919	1,129	5,640	38.44%	39.42%	0.4028
Death White	302,430	0.969	(0.951, 0.988)	0.0014	33,721	166,822	22.56%	22.98%	0.0766
Death Black	12,156	1.006	(0.913, 1.108)	0.9059	1,129	5,640	31.27%	30.90%	0.9654
ADRD White	302,430	0.888	(0.859, 0.918)	<.0001	33,721	166,822	7.46%	8.44%	<.0001
ADRD Black	12,156	0.906	(0773, 1.061)	0.2200	1,129	5,640	11.43%	12.40%	0.3838

Subset results for history of colon disease, appendicitis presentation, and early versus late period of appendicitis are provided in the Appendix, eTable 4.4 (a,b), 4.5 (a,b) and 4.6 (a,b).

In the subgroup of males, as displayed in Table 3, we see that the hazard ratio (HR) for developing ADRD or death was not different in the Appendectomy group than controls: HR = 1.002 (95% CI (0.979, 1.025)), P = 0.89, (32.73% in the Appendectomy group versus 32.53% in controls, at 7.5 years). The HR for death was 1.011 (0.987, 1.035), P = 0.3776, (27.89% versus 27.48% at 7.5 years). The HR for developing ADRD alone was 0.948 (0.905, 0.994), P= 0.027, (7.14% versus 7.73% at 7.5 years). However, for females, there were slightly lower rates of ADRD outcomes in the Appendectomy group. The HR for ADRD or death was 0.907 (95% CI (0.884, 0.931)), P < 0.0001, (23.51% versus 25.58% at 7.5 years). The HR for death was 0.919 (0.893, 0.946), P < 0.0001, (17.32% versus 18.65% at 7.5 years). The HR for developing ADRD alone was 0.847 (0.811, 0.884), P < 0.0001, (8.16% versus 9.48% at 7.5 years). Expanded analyses for these subgroups and all other subgroups analyzed are displayed in Supplemental Digital Content 1, eTable 4.1a to eTable 4.6b) In none of our subgroup analyses did we find any significant evidence of higher ADRD rates in the appendectomy group.

Figure 1 displays three Kaplan-Meier plots of death or ADRD. Figure 1a includes all patients in the study. Figure 1b includes only males. Figure 1c includes only females. For the treated and control groups, Figures 2a, 2b and 2c compare: (i) the cumulative hazard of death-or-ADRD (top curves), (ii) death-before-ADRD (middle curves), and (iii) ADRD (bottom curves). As seen in Figures 1 and 2, the treated and control curves are close in all cases, with slightly better results for the appendectomy group. Furthermore, whatever differences we observe between appendectomy treated and control patients is mostly in females, as male curves are almost overlapping.

Figure 1:

Kaplan-Meier plots of survival without ADRD post appendectomy event.

Figure 2: Cumulative Hazard Plots:

(i) the cumulative hazard of dying or developing ADRD for both the treated and control groups; (ii) the cumulative hazard of dying without developing ADRD and (iii) the cumulative hazard of developing ADRD in those who did not die. All plots also censor on loss to end of study or managed care (which do not report claims used for the ADRD definition).

Stability Analyses

We were careful not to look into the future when defining controls, so it was always possible that some controls “crossed over” and experienced appendicitis with appendectomy after being matched to a treated patient earlier. The incidence of crossovers was very small. There were only 71 crossovers out of 274,980 controls (71/274,980 = 0.00026). When we reran our results omitting these 71 matched sets, the HR for all three outcomes were unchanged for all overall and subgroup analyses.

Using a separate definition of ADRD from Medicare’s Chronic Condition Warehouse,⁴³ we found very stable results. The HR for ADRD was 0.93 (0.91, 0.95), again similar to our primary results HR for developing ADRD in the Appendectomy group versus controls (0.89 (0.86, 0.92)). We further explored the hazard for utilizing a nursing home, HR = 0.94 (0.90, 0.97) between appendectomy and control groups. All stability results were similar to our primary results (see Supplemental Digital Content 1, Sections 5.1-5.3).

Sensitivity Analyses

We observed a small difference in time to ADRD or death in the overall group analysis and for the women subgroup (Figure 1c and Table 3), with the treatment group doing better than controls. Should this improvement be taken as evidence of a protective effect of appendectomy? We do not believe it should. If the differences in Figure 1a and Figure 1c could be produced by very small biases, we would commonly understand them not to be important treatment effects in an observational study. We therefore asked about the magnitude of unmeasured bias in the occurrence of appendicitis that would need to be present to produce the difference in ADRD or death at 7.5 years in the overall analysis and in the subgroup analysis for women. Because the difference in outcomes overall and in women is very small, it may just have reflected a small unmeasured bias in the risk of developing appendicitis. An unobserved confounder linked to appendicitis by an odds ratio of gamma = 1.03 could produce the difference seen in the overall rates at 7.5 years. And an odds ratio of 1.1 could produce the difference seem for women.⁴⁴ Details on this calculation are provided in Supplemental Digital Contents Section 6. Therefore, the sensitivity analysis confirms our sense that that there was no convincing evidence that undergoing an appendectomy for appendicitis in the elderly is beneficial in preventing ADRD.

DISCUSSION

The findings in our study provide a new window into the risks of surgery and anesthesia in the elderly and these results are reassuring. Through the use of a natural experiment, our comparison of appendectomy patients and their matched controls is less likely to be severely biased by confounding than comparisons involving surgery for chronic health problems. Rather than studying ADRD after an elective or required surgical procedure that may be needed because of a health problem that could also be related to developing ADRD, this study examines a non-trivial operative procedure that is generally thought to be unrelated to other medical conditions, after controlling for age, sex, race, and possibly diseases of the colon. Our results provide no indication that surgery with anesthesia in the elderly undergoing an appendectomy between ages 68 through 77 accelerates the rate of ADRD. Furthermore, had there been an increase in ADRD after an appendectomy, our sample size, including over 325,000 patients, was large enough to detect clinically meaningful differences between groups.

There are limitations to our study. We based our outcomes on Medicare claims data, without neurocognitive testing for all of the 329,976 study patients. However, our claims-based definition of ADRD (see Supplemental Digital Content 1, Section 2) was validated on neurocognitive testing data in its development. Furthermore, given the natural experiment of appendectomy, the expected measurement accuracy of our definition based on claims should work equally well for both the appendectomy and control patients.

A further limitation of our study was that we did not have data on Medicare patients prior to age 65. An older patient who had appendicitis at age 64 might be in the matched control group. In fact, because appendicitis is very rare at this age, we estimate that there are very few such instances. Using the age and sex specific rates of appendicitis as reported by Addiss et al.¹⁹ (Addiss Table 5), we can estimate that there could have been up to 1603 control patients out of the 274,980 controls (1603/274,980 = 0.00583) who may have had appendicitis in the 15 years prior to being matched (see Supplemental Digital Contents Section 7 and eTable 7.1). It would be impossible for these half-of-one percent of our matched controls to have materially altered the conclusions of this study.

There are many procedures which may resemble appendectomy with respect to surgery and anesthesia for which our natural experiment’s results may be relevant. In our study we found that the typical appendectomy procedure anesthesia time for our appendectomy treated patients as reflected through the anesthesia bill was approximately 90 minutes. In our previous work,⁴⁵ we have also found that an appendectomy takes approximately 90 minutes, a herniorraphy approximately 100 minutes and a laparoscopic cholecystectomy about 112 minutes. Therefore, we would suggest that from the perspective of ADRD risk from anesthesia, there is reason to believe that our reassuring findings for appendectomy should also apply to these other procedures.

Our study is premised on the idea that appendicitis is a comparatively haphazard, almost random event requiring surgery and anesthesia. And, indeed, this is largely true. Surgery for appendicitis is not elective, so there is no ambiguity from self-selection, as there would be with many forms of orthopedic surgery. Unlike most diseases that may require surgery, appendicitis does not disproportionately target people who have serious, enduring health problems, so it does not come packaged with major risk factors for ADRD. There is, however, a small limitation in this claim. To have appendicitis, one must have an appendix. A person who, in the distant past, had an incidental appendectomy as part of some other surgery for some other health problem, cannot be in our appendectomy group, but might end up in our control group. Because our data start with the start of Medicare coverage, we cannot identify and remove patients who had an incidental appendectomy at earlier ages. Incidental appendectomies are rare in men, a little less rare in women. Prior to age 65, women have over a five-fold higher rate of incidental appendectomy than men.¹⁹ Presumably, this issue does not obscure the comparison for men, but might have a small impact on the comparison for women. Figures 1b and 1c might be reexamined with this in mind.

CONCLUSION

Analyzing over 325,000 patients while taking advantage of the natural experiment of an appendectomy event and our ability to identify claims associated with ADRD on a vast scale throughout the United States, our study found no evidence that surgery and anesthesia in the elderly promotes the subsequent diagnosis of Alzheimer’s Disease and Related Dementias. This study should be reassuring for elderly patients without the past diagnosis of ADRD, cognitive impairment, or neurological degeneration, when needing surgery similar to the exposure we studied using appendectomy.