Effects of Volatile Anesthetics on Postoperative Ischemic Stroke Incidence.

Background Preclinical studies suggest that volatile anesthetics decrease infarct volume and improve the outcome of ischemic stroke. This study aims to determine their effect during noncardiac surgery on postoperative ischemic stroke incidence. Methods and Results This was a retrospective cohort study of surgical patients undergoing general anesthesia at 2 tertiary care centers in Boston, MA, between October 2005 and September 2017. Exclusion criteria comprised brain death, age <18 years, cardiac surgery, and missing covariate data. The exposure was defined as median age-adjusted minimum alveolar concentration of all intraoperative measurements of desflurane, sevoflurane, and isoflurane. The primary outcome was postoperative ischemic stroke within 30 days. Among 314 932 patients, 1957 (0.6%) experienced the primary outcome. Higher doses of volatile anesthetics had a protective effect on postoperative ischemic stroke incidence (adjusted odds ratio per 1 minimum alveolar concentration increase 0.49, 95% CI, 0.40-0.59, P<0.001). In Cox proportional hazards regression, the effect was observed for 17 postoperative days (postoperative day 1: hazard ratio (HR), 0.56; 95% CI, 0.48-0.65; versus day 17: HR, 0.85; 95% CI, 0.74-0.99). Volatile anesthetics were also associated with lower stroke severity: Every 1-unit increase in minimum alveolar concentration was associated with a 0.006-unit decrease in the National Institutes of Health Stroke Scale (95% CI, -0.01 to -0.002, P=0.002). The effects were robust throughout various sensitivity analyses including adjustment for anesthesia providers as random effect. Conclusions Among patients undergoing noncardiac surgery, volatile anesthetics showed a dose-dependent protective effect on the incidence and severity of early postoperative ischemic stroke.

Clinical Perspective

What Is New?

In this retrospective cohort study of 314 932 patients undergoing noncardiac surgery with general anesthesia, volatile anesthetics had a dose‐dependent protective effect on postoperative ischemic stroke incidence and severity; the magnitude of the protective effect was dependent on the postoperative timing of stroke occurrence.

Compared with patients undergoing propofol‐based anesthesia, patients receiving volatile anesthetics had a significantly lower risk of postoperative ischemic stroke.

Our findings were robust across several sensitivity analyses, including propensity score matching and adjustment, and in a mixed‐effects model adjusting for provider variability.

What Are the Clinical Implications?

This study supports the use of volatile anesthetics in patients who require general anesthesia and who are vulnerable to postoperative ischemic stroke.

Since the effects were dose dependent, clinicians should know that using higher doses of volatile anesthetics, as compared with lower doses or propofol‐based anesthesia, might be considered for stroke prevention.

Optimally, our study results would be confirmed in a randomized controlled trial.

In the United States, ≈60 000 patients undergo general anesthesia every day. With an aging surgical population, increasing surgical case volumes, and incidences of perioperative stroke ranging from 0.1% to 9.7%, perioperative stroke prevention has become a goal of increasing importance.

Intraoperative anesthetic management strategies have substantial consequences on a patient's susceptibility to ischemic stroke after surgery. , ,

Preclinical data suggest that volatile anesthetics have neuroprotective effects that may decrease the risk of stroke. , , However, recent studies lack conclusive evidence derived from large patient cohorts. ,

The primary aim of this study was to examine the effect of intraoperative volatile anesthetic dose on ischemic stroke within 30 days after noncardiac surgery in a large and diverse surgical cohort.

Methods

Study Design and Setting

This was a hospital registry study of patients undergoing surgery with general anesthesia at Beth Israel Deaconess Medical Center in Boston, Massachusetts, between October 2005 and September 2017, and at Massachusetts General Hospital in Boston, Massachusetts, between January 2007 and December 2015. The study was approved by the Beth Israel Deaconess Medical Center Institutional Review Board (protocol number 2019P000014) and the Partners Human Research Committee (reliance agreement [SMART IRB] number 1627). Requirement for informed consent was waived. Data were retrieved from data repositories at Beth Israel Deaconess Medical Center and Partners HealthCare, and subsequently combined into a deidentified data set (Data S1). The authors TTH and ME had full access to the data in this study and take responsibility for their integrity and analysis.

The data supporting our findings are available from the corresponding author upon reasonable request. This article adheres to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for reporting observational research (Data S2).

Patient Selection

Patients undergoing surgery with general anesthesia were included in the study cohort. Exclusion criteria comprised age <18 years, cardiac surgery, or an American Society of Anesthesiologists physical status classification of VI (brain death) (Data S1). Patients with missing covariate data were excluded from analyses to apply the complete case method.

Exposure and Primary Outcome

The primary exposure variable was defined as median age‐adjusted minimum alveolar concentration (MAC) of all minute‐by‐minute end‐tidal measurements of volatile anesthetics (ie, desflurane, sevoflurane, and isoflurane) throughout the surgical case (Data S1).

The primary outcome was postoperative ischemic stroke within 30 days, identified through International Classification of Diseases, Ninth/Tenth Revision (ICD‐9/ICD‐10) billing diagnoses (Table S1).

Covariate Model

Confounding factors were selected a priori utilizing a model for preoperative assessment of perioperative ischemic stroke risk developed by our group. Subsequently, we expanded the model for perioperative factors based on available literature , , as well as biological and clinical plausibility. The final model adjusted for patient demographics, comorbidities, and anesthesia‐ and procedure‐related factors (Data S1).

Primary Analysis

In the primary analysis, we applied multivariable logistic regression to examine the effect of intraoperative volatile anesthetic dose on ischemic stroke within 30 days after surgery.

Secondary Analyses

In secondary analyses, we assessed the time dependency of the observed effect using a Cox proportional hazards model adjusting for the covariates included in the primary model. Therefore, the exposure was dichotomized by the median MAC among all patients (MAC >0.73 versus ≤0.73). An interaction term (volatile anesthetic dose×time after surgery) was included in the model. In a post hoc logistic regression analysis based on findings from the Cox proportional hazards model, we compared the effect of volatile anesthetics on postoperative ischemic stroke occurring earlier (within days 1–17) versus later (within days 18–30 after surgery).

Further, we analyzed the effect of volatile anesthetic dose on postoperative transient ischemic attack (TIA) and all‐cause mortality within 30 days after surgery (Data S1).

Sensitivity Analyses

Multiple sensitivity analyses were performed to test the robustness of findings, discussed below.

Anesthetic Requirement

We created a propensity score for receiving high doses of volatile anesthetics (highest tertile) utilizing the covariates of the full primary logistic regression model (Data S1). , , , Subsequently, patients were categorized into equally sized tertiles of low, intermediate, and high propensity, respectively, and we repeated the primary analysis in each group.

Effects of Intraoperative Hypotension

To further address potential differences in patients' individual susceptibility to anesthetics, we assessed the maximum blood pressure decrease within 5 minutes of anesthesia induction with propofol (in percent from baseline before induction). Maximum mean arterial pressure (MAP) decrease was categorized (<10%, ≥10 and <20%, ≥20 and <30%, or ≥30%) and an interaction term (volatile anesthetic dose×maximum MAP decrease) was introduced into the primary model.

The primary analysis was also repeated in patients with no, short, intermediate, and prolonged duration of intraoperative hypotension, defined as 0, >0 and ≤2, >2 and ≤5, and >5 measurements of MAP <55 mm Hg during surgery, respectively.

Propensity Score Matching and Adjustment

To address potential residual confounding, 1:1 propensity score matching was performed for the variable high‐dose (higher than the cohort median) volatile anesthetics, utilizing the covariates of the primary model. A calculated caliper was used for matching, and variables were examined for residual imbalances. Further, we calculated the E‐value for our primary finding to quantify the potential impact of unmeasured confounding. As introduced by VanderWeele and Ding, the E‐value is defined as the minimal magnitude of association that unmeasured confounding would need to have with both exposure and outcome in order to fully explain away the observed effect.

Analysis Stratified by Baseline Stroke Risk

The full primary covariate model was utilized to estimate patients' baseline risk of postoperative ischemic stroke. The study cohort was then divided into equally sized tertiles of patients with low, intermediate, and high baseline risk, respectively, and we reran the primary analysis in every risk tertile.

Subgroup Analysis in Patients With High Procedure‐Related Stroke Risk

We further examined the primary association in a subgroup of patients with high procedural risk of postoperative ischemic stroke, defined as brain and vascular surgery (including carotid endarterectomy). ,

Impact of Provider Variability

Since the dosing of volatile anesthetics may depend on provider preference, we analyzed provider variability in our cohort and assessed its impact on the primary finding. Individual providers' preferences for using high doses were defined using a mixed‐effects logistic regression model with volatile anesthetic doses higher than the cohort median as the outcome. All covariates of the primary model were added as fixed effects while individual providers with a total experience of ≥100 procedures were added as random effect, resulting in the predicted probability of receiving high doses for each patient. For each provider, the adjusted preference of using doses higher than the median was calculated across all cases performed by the respective provider.

To assess the impact of provider variability on our primary finding, a mixed‐effects model was used on the primary logistic regression model, adding anesthesia providers as random effect.

Adjudicated Outcome Based on Chart Review

We reran the primary analysis using an adjudicated outcome variable based on medical record review performed by an interdisciplinary team as previously described by our group. , All cases with a billing diagnosis of ischemic stroke within 30 days after surgery were manually reviewed using brain scan reports, discharge summaries, and neurology consultation notes (Data S1). Additionally, stroke location according to the Oxfordshire Community Stroke Project , and National Institutes of Health Stroke Scales (NIHSS) were assessed.

Effects of Anesthesia Depth (Bispectral Index)

To discriminate between the effect of volatile anesthetic dose and the effect of the level of unconsciousness during anesthesia, the association of the median intraoperative bispectral index (BIS) and postoperative ischemic stroke was assessed in a subgroup of patients with available data. Further, an interaction term (BIS×volatile anesthetic dose) was included in the model.

Supplemental Sensitivity Analyses

Additional sensitivity analyses, including those taking into account missing follow‐up and covariate data, are provided in Tables S2 and S3).

Exploratory Analyses

We compared volatile anesthesia (while allowing for an induction bolus of propofol) to total intravenous anesthesia (TIVA) using propofol regarding the primary outcome.

Model fit was assessed using c‐statistics (ie, area under the receiver operating characteristics curve), as well as calibration plot and Brier score (squared difference between estimated and observed outcomes).

Additional post hoc analyses are described in Data S1. Briefly, to address a suggestion of a peer reviewer, we also included all patients undergoing cardiac surgery who had previously been excluded. Further, we explored compound‐specific effects of volatile agents in the respective subgroups.

Statistical Analyses

If not further specified, multivariable logistic regression utilizing the full covariate model was used for binary outcome variables. Analyses were selected a priori. Statistical significance was assumed at a 2‐sided P<0.05.

Analyses were performed using Stata (version 15; StataCorp LLC, College Station, TX) or Rstudio (version 1.1.442; Rstudio Inc., Boston, MA).

Results

Study Cohort

A total of 400 257 patients underwent surgery under general anesthesia at Beth Israel Deaconess Medical Center and Massachusetts General Hospital within the studied time frame (Figure 1). There were 23 719 patients excluded for being underage, brain‐dead, or undergoing cardiac surgery, and 61 606 patients were excluded from analysis because of missing covariate data. The final study cohort comprised 314 932 cases (Table 1). 298 505 patients (94.5%) received volatile anesthetics for anesthesia. The overall median MAC of volatile anesthetics was 0.73 (interquartile range, 0.50, 0.94).

Figure 1

Study flow.

*Multiple criteria may apply. ASA indicates American Society of Anesthesiologists physical status classification (as classified by the anesthesiologist); BMI, body mass index; ED95, median effective dose required to achieve a 95% reduction in maximal twitch response from baseline; and NMBA, neuromuscular blocking agent.

Table 1

Characteristics of the Study Population by Postoperative Ischemic Stroke Status

Characteristics	No Ischemic Stroke Within 30 d (n=312 975)	Ischemic Stroke Within 30 d (n=1957)
Demographics
Age, y	53.7±16.5	64.1±15.2
Sex, male	137 662 (44.0%)	1094 (55.9%)
Body mass index, kg/m2	28.4±6.9	27.6±6.0
Comorbidities*
Arterial hypertension	125 363 (40.1%)	1416 (72.4%)
Atrial fibrillation	21 311 (6.8%)	444 (22.7%)
Carotid stenosis	6976 (2.2%)	679 (34.7%)
Chronic kidney disease	20 187 (6.5%)	352 (18.0%)
Diabetes mellitus	45 628 (14.6%)	553 (28.3%)
Dyslipidemia	96 622 (30.1%)	1066 (54.5%)
Ischemic stroke	7639 (2.4%)	1357 (69.3%)
Malignancy	92 111 (29.4%)	579 (29.6%)
Migraine	11 373 (3.6%)	87 (4.4%)
Patent foramen ovale without closure	2767 (0.9%)	135 (6.9%)
Peripheral vascular disease	12 222 (3.9%)	303 (15.5%)
Smoking	51 969 (16.6%)	516 (26.4%)
Transient ischemic attack	4691 (1.5%)	399 (20.4%)
Valvular heart disease	26 679 (8.5%)	732 (37.4%)
Beta‐blocker prescription within 28 d prior	44 342 (14.2%)	1096 (56.0%)
Charlson Comorbidity Index 30	1 (0, 3)	4 (2, 6)
ASA physical status	2 (2, 3)	3 (2, 3)
Surgical factors
Emergency surgery	13 994 (4.5%)	182 (9.3%)
Inpatient surgery	199 951 (63.9%)	1868 (95.5%)
Duration of surgery, min	155±108	194±134
Work relative value units	14.7±9.8	19.0±12.9
Surgical service
Burn	1976 (0.6%)	22 (1.1%)
Emergent–urgent	10 721 (3.4%)	164 (8.4%)
General	57 773 (18.5%)	86 (4.4%)
Gynecology/obstetrics	31 200 (10.0%)	33 (1.7%)
Neurosurgery	21 899 (7.0%)	531 (27.1%)
Oral/maxillofacial	3297 (1.1%)	7 (0.4%)
Orthopedic	72 620 (23.2%)	160 (8.2%)
Other (dermatology, etc)	4463 (1.4%)	58 (3.0%)
Otolaryngology	8808 (2.8%)	9 (0.5%)
Plastic	18 716 (6.0%)	19 (1.0%)
Radiology	1728 (0.6%)	71 (3.6%)
Surgical oncology	15 003 (4.8%)	19 (1.0%)
Thoracic	19 986 (6.4%)	98 (5.0%)
Transplant	5704 (1.8%)	23 (1.2%)
Urology	22 176 (7.1%)	59 (3.0%)
Vascular	10 402 (3.3%)	410 (21.0%)
Anesthetic factors
Use of volatile anesthetic	296 714 (94.8%)	1791 (91.5%)
MAC of volatile anesthetic	0.72±0.35	0.51±0.34
MAC of nitrous oxide	0.07 (0, 0.41)	0.21 (0, 0.56)
Total opioid dose (oral morphine equivalents)	50.5 (31.3, 79.5)	62.5 (37.5, 103.3)
Total propofol dose, mg	200 (150, 260)	170 (110, 250)
Total neuromuscular blocking agent ED95 dose	1.86 (0, 3.06)	2.82 (1.69, 4.4)
Total vasopressor dose, mg (norepinephrine equivalents)	0.01 (0, 0.1)	0.17 (0.03, 0.53)
Total fluid volume administered, mL	2000 (1000, 3000)	1350 (750, 2500)
Administration of packed red blood cells	9118 (2.9%)	141 (7.2%)
Neuraxial anesthesia	10 271 (3.3%)	65 (3.3%)
Minutes with MAP <55 mm Hg	0 (0, 2)	1 (0, 3)

Values provided as frequency (prevalence in %), mean±SD, or median (interquartile range [25th–75th percentile], values separated by comma). ASA indicates American Society of Anesthesiologists; ED95, median effective dose required to achieve a 95% reduction in maximal twitch response from baseline; MAC, minimum alveolar concentration; and MAP, mean arterial pressure.

For comorbidity definitions, refer to Table S1.

Of all patients in the study cohort, 1957 patients (0.6%) had an ischemic stroke within 30 days after surgery. The median time to ischemic stroke was 4 days (interquartile range, 1, 14) (Figure S1).

A higher MAC of volatile anesthetics had a significant protective effect on ischemic stroke within 30 days after surgery in unadjusted (odds ratio per 1 MAC increase 0.16, 95% CI, 0.14–0.19, P<0.001) as well as adjusted analyses (adjusted odds ratio [aOR] per 1 MAC increase 0.49, 95% CI, 0.40–0.59, P<0.001) (Table 2; Table S4).

Table 2

Primary and Secondary Outcomes in Patients Receiving Low Versus High Doses of Volatile Anesthetics

Outcome	Low‐Dose Volatile Anesthetics (Lowest Tertile)		High‐Dose Volatile Anesthetics (Highest Tertile)		High vs Low Doses		Odds Ratio (95% CI)*
	Outcome Rate (%)	Estimated Risk (%, 95% CI)	Outcome Rate (%)	Estimated Risk (%, 95% CI)	aARD (%)	RRR (%)	Unadjusted	Adjusted
Ischemic stroke	1.1	0.08 (0.07–0.10)	0.28	0.05 (0.04–0.06)	−0.03	37.5	0.16 (0.14–0.19)	0.49 (0.40–0.59)
TIA	0.38	0.03 (0.02–0.04)	0.07	0.01 (0.009–0.018)	−0.02	66.7	0.12 (0.10–0.16)	0.35 (0.25–0.49)
Death	1.0	0.1 (0.08–0.11)	0.49	0.05 (0.04–0.06)	−0.05	50.0	0.36 (0.32–0.41)	0.48 (0.41–0.56)

Table 2 depicts results from primary and secondary analyses in patients receiving the lowest and highest tertile of volatile anesthetic dose across the study cohort, respectively. All outcomes were assessed within 30 days after surgery. Analyses were adjusted for all covariates included in the primary model. Estimated risk and risk differences were calculated using Stata packages “predict” and “margins.” Rates are rounded to 2 decimal places. aARD indicates adjusted absolute risk difference; RRR, relative risk reduction; and TIA, transient ischemic attack.

P<0.001 for all results listed in Table 2.

In the Cox proportional hazards regression, we found a time‐dependent effect of volatile anesthetic dose and postoperative ischemic stroke (P for interaction volatile anesthetics×days after surgery <0.001): Patients receiving high volatile anesthetic doses showed a significantly lower hazard rate of ischemic stroke for up to 17 days after surgery (postoperative day 1: hazard ratio [HR], 0.56; 95% CI, 0.48–0.65; versus postoperative day 17: HR, 0.85; 95% CI, 0.74–0.99). The protective effect was found to be no longer statistically significant (P>0.05) at postoperative day 18 (HR, 0.88; 95% CI, 0.75–1.02). One thousand five hundred ninety‐four of 1957 ischemic strokes (81.5%) occurred within 17 days after surgery. At postoperative day 23, volatile anesthetics were no longer found to have a protective effect (HR, 1.00; 95% CI, 0.83–1.21) (Figure 2). A significant protective effect of higher volatile anesthetic doses was confirmed for ischemic strokes within 17 days (aOR, 0.41; 95% CI, 0.33–0.51, P<0.001) but not for ischemic strokes within days 18 to 30 (aOR, 0.98; 95% CI, 0.67–1.45, P=0.94) in subsequent multivariable logistic regression analyses. For details of patients receiving low versus high doses, see Table S5.

Figure 2

Hazard ratio for ischemic stroke per postoperative day.

Results of the Cox proportional hazards regression regarding the effect of volatile anesthetics higher than minimum alveolar concentration=0.73 (cohort median) on postoperative ischemic stroke, with a hazard ratio of 1.00 shown as bold line in the graph. Hazard ratios are presented per postoperative day. Patients receiving higher doses of volatile anesthetics showed significantly lower hazard of ischemic stroke for up to 17 days after surgery.

Estimated risks of ischemic stroke were 0.5 for every 1000 patients receiving high doses of volatile anesthetics (highest tertile, mean [SD] MAC 1.1 [0.23]) and 0.8 for every 1000 patients receiving low doses (lowest tertile, mean [SD] MAC 0.35 [0.19]). In comparison to patients receiving low doses, patients receiving high doses were found to have a significantly lower risk of ischemic stroke after surgery (adjusted absolute risk difference, −0.03%; 95% CI, −0.04 to −0.02, P<0.001; relative risk reduction, 37.5%) (Table 2). For results of all other secondary analyses, please see Table 2 and Data S1.

The primary effect was found to be robust across tertiles of volatile anesthetic dose (Table 3). The decreasing odds ratio for tertiles of intermediate and high doses, respectively, highly suggest a dose dependency of the primary effect (P for trend <0.001).

Table 3

Results of Sensitivity Analyses

Type of Analysis	Subgroup of Patients	aOR, 95% CI	P Value
Tertiles of volatile anesthetic dose	Receiving low doses	1.0 (reference level)
	Receiving intermediate doses	0.78, 0.68–0.89	<0.001
	Receiving high doses	0.61, 0.51–0.72	<0.001
Anesthetic requirement	Low propensity of receiving high doses	0.48, 0.35–0.65	<0.001
	Intermediate propensity of receiving high doses	0.50, 0.36–0.70	<0.001
	High propensity of receiving high doses	0.60, 0.40–0.91	0.02
Effects of intraoperative hypotension	No hypotension	0.54, 0.41–0.72	<0.001
	Short duration of hypotension	0.48, 0.32–0.72	<0.001
	Intermediate duration of hypotension	0.49, 0.28–0.87	0.02
	Prolonged duration of hypotension	0.38, 0.23–0.64	<0.001
Propensity score matching	Propensity of doses higher vs lower than median	0.66, 0.57–0.75	<0.001
Propensity score adjustment	Propensity of doses higher vs lower than median (vs without propensity score adjustment)	0.64, 0.57–0.72 (0.68, 0.60–0.78)	<0.001 (<0.001)
Analysis stratified by baseline stroke risk	Low baseline risk of stroke	0.99, 0.27–3.55	0.98
	Intermediate baseline risk of stroke	0.82, 0.32–2.07	0.67
	High baseline risk of stroke	0.46, 0.38–0.56	<0.001
Subgroup analysis in patients with high procedure‐related stroke risk	Undergoing brain or vascular surgery	0.58, 0.43–0.79	<0.001
Impact of provider variability	Mixed‐effects model adjusting for anesthesia provider	0.70, 0.60–0.83	<0.001
Adjudicated outcome based on chart review	Full study cohort	0.39, 0.30–0.50	<0.001

Table 3 depicts results from the sensitivity analyses performed in order to test the robustness of the primary finding. All sensitivity analyses use the primary outcome (ischemic stroke within 30 days after surgery). Analyses were adjusted for all covariates included in the primary model. aOR indicates adjusted odds ratio.

Repeating the primary analysis in each tertile of the propensity score for receiving high‐dose volatiles confirmed the primary finding: Higher doses of volatile anesthetics had a significant protective effect on ischemic stroke in patients with the highest propensity of high doses as well as in patients with intermediate and low propensity of high doses (Table 3).

In 163 241 patients with available data, no effect modification by maximum MAP decrease from baseline after induction was found (≥10 and <20%: P for interaction=0.58; ≥20 and <30%: P for interaction=0.62; ≥30%: P for interaction=0.81).

The primary finding was substantiated in all categories of intraoperative hypotension: Patients without intraoperative MAP <55 mm Hg (n=172 089), as well as with short (n=72 565), intermediate (n=36 980), and prolonged duration of intraoperative hypotension (n=32 973) (Table 3).

With a calculated caliper of 0.359, a total of 159 692 (50.7%) patients higher and lower than the cohort median dose of volatile anesthetics were matched according to their propensity for receiving high‐dose volatile anesthetics. In the matched cohort, the primary finding was confirmed.

Since propensity matching omitted unmatched individuals from analysis, we further performed propensity score adjustment. A model was fitted including the aforementioned exposure (doses higher versus lower than median) as well as a linear and squared term of the propensity score. Our primary finding was confirmed (Table 3).

In our primary analysis, we observed an aOR of 0.49 (95% CI, 0.40–0.59). An unmeasured confounder would have to be associated with both the exposure and the outcome, respectively, with an aOR (adjusted for all measured confounders) of 3.5 (E‐value) each in order to fully explain away the observed effect. To move the 95% CI such that the observed effect would no longer be statistically significant, an unmeasured confounder of the same nature would have to have an aOR of 2.78 for the primary association. Weaker confounding could not explain away the observed association.

The primary finding was also confirmed in patients within the highest tertile of baseline risk of postoperative ischemic stroke but not in patients within the intermediate‐ and low‐risk group (n=104 977 each; Table 3).

Further, the protective effect of volatile anesthetics on postoperative ischemic stroke incidence was robust in a subgroup of 24 195 patients (7.7%) with high procedural risk of stroke, such as brain and vascular surgery (Table 3).

After excluding cases performed by anesthesiologists with a total experience of <100 cases, 70 277 patients remained for analysis. In this cohort, 862 individual providers were documented. The predicted preference for individual providers to use volatile anesthetic doses higher than the cohort median ranged from 3.1% to 93.9%, demonstrating high provider variability (Figure S2). The primary finding remained robust when adjusting for provider variability in a mixed‐effects model (Table 3).

Medical record review verified 686 of 1957 (35.1%) ischemic strokes billed through ICD‐9/10 codes within 30 days after surgery, which translates to an ischemic stroke incidence of 0.2%. Of these verified strokes, 127 (18.5%) patients had a partial anterior circulation and 170 (24.5%) a total anterior circulation stroke. One hundred seventy‐eight (26.0%) patients had a posterior circulation infarct, while 38 (5.5%) patients showed lacunar infarctions. One hundred seventy‐three (25.2%) patients had an ischemic stroke of unclassifiable location. Repetition of the primary analysis utilizing the verified ischemic stroke outcome confirmed our primary findings (Table 3).

Fifty patients (7.3%) with verified ischemic stroke presented missing NIHSS data. Four hundred fifty‐nine (66.9%) patients presented with mild (NIHSS ≤5) and 177 (25.8%) with moderate–severe neurological symptoms (NIHSS >5). The median NIHSS among all verified ischemic strokes with complete information was 3 points (interquartile range, 1, 6.5). In a post hoc exploratory analysis, every 1‐unit increase in MAC of volatile anesthetics was associated with a 0.006‐unit decrease in NIHSS (β −0.006; 95% CI, −0.01 to −0.002, P=0.002). Further, in a multinomial logistic regression model, volatile anesthetics had a dose‐dependent protective effect on both patients with mild neurological symptoms (β −0.82; 95%, CI −1.18 to −0.46, P<0.001) and patients with moderate–severe neurological presentation (β −0.58; 95% CI, −1.17 to −0.01, P=0.054). There was no significant interaction between volatile anesthetics and intraoperative hypotension in either of the 2 groups (P for interaction=0.53 in patients with mild symptoms, and 0.76 in patients with moderate–severe symptoms).

Effects of Anesthesia Depth (BIS)

BIS was not associated with postoperative ischemic stroke in a subgroup of 14 862 patients with available data (aOR, 0.995; 95% CI, 0.97–1.02, P=0.75). No significant interaction of volatile dose and BIS was found regarding the primary outcome (P for interaction=0.70).

In our cohort, 204 522 (64.9%) patients received volatile anesthesia with the option of propofol as induction bolus. Some patients (13 918; 4.4%) underwent TIVA using propofol without volatile anesthetics. 96 492 patients in our cohort (30.6%) received a combined or different type of anesthesia and, thus, were not considered in this exploratory analysis. In comparison to TIVA, patients undergoing volatile anesthesia had significantly lower odds of experiencing the primary outcome (aOR, 0.71; 95% CI, 0.55–0.90, P=0.005).

The covariate model for postoperative ischemic stroke showed excellent discriminative ability independent of the exposure with an area under the receiver operating characteristics curve of 0.95 (Figure S3).

A reliability plot demonstrated excellent calibration of the covariate model (Figure S4). The Brier score for the covariate model was 0.007 and reliability was 0.002, reflecting excellent accuracy and calibration.

Discussion

In this cohort of 314 932 adult patients undergoing noncardiac surgery with general anesthesia, volatile anesthetics were found to have a dose‐dependent protective effect on postoperative ischemic stroke incidence and severity.

In our study, the protective effect of volatile anesthetics on ischemic stroke was stronger in patients who developed an early postoperative stroke, which supports a pharmacologically plausible effect: Volatile anesthetic preconditioning may prevent early cascades of brain ischemia from evolving into a clinically relevant ischemic stroke. A plausible mechanism is the dose‐dependent frequency inhibition of cortical spreading depolarizations, which have been reported to occur for several days after ischemia. , , Preclinical literature describes protective effects of volatile preconditioning for up to 3 days. In this study, significant effects are seen until postoperative day 17. It is possible that anti‐inflammatory and anti‐thrombogenic effects, as well as remote preconditioning, are at play as well. , ,

This hypothesis is in line with the results of a retrospective cohort study by Sivasankar et al, who reported that the use of volatile anesthetics led to significantly lower degrees of poststroke disability in a cohort of patients undergoing revascularization procedures after stroke. Our study adds to these findings that a protective effect may also be relevant for stroke prevention.

Our data further suggest that volatile anesthesia is associated with lower odds of postoperative ischemic stroke when compared with TIVA using propofol. This corresponds with a randomized controlled trial of patients undergoing carotid endarterectomy by Kuzkov et al showing that, compared with propofol, sevoflurane suppressed intraoperative asymmetry of cerebral oxygenation and improved postoperative cognition. Similarly, in a randomized controlled trial of 128 patients undergoing cardiac surgery with cardiopulmonary bypass, Schoen et al found that sevoflurane‐based anesthesia (compared with propofol) attenuated the effects of intraoperative desaturation on postoperative neurocognition.

Strengths and Limitations

A major strength of this study is the generalizability of results derived from a large, diverse, and multicentric surgical cohort. While residual confounding cannot be ruled out because of the study's observational nature, the covariate model was shown to have excellent discriminative ability with an area under the receiver operating characteristics curve of 0.95. Misclassification of the outcome based on billing codes and varying coding practices between hospitals was likely random and unrelated to the exposure. Potential bias was addressed by medical record review of the outcome variable, which did not change our conclusions.

One might speculate that younger and healthier patients are more likely to receive higher doses of volatile anesthetics while having an overall low baseline risk of stroke and, thus, low outcome rates. Therefore, we examined subgroups of patients with varying probability of receiving higher doses. We observed the dose‐dependent effect of volatile anesthetics on ischemic stroke to be robust across patients with different propensity of receiving high doses. Another marker of anesthetic requirement may be the hemodynamic response to an induction dose of an intravenous anesthetic, typically propofol. Our data showed that the patient's hemodynamic susceptibility to anesthetics did not modify the observed primary effect.

This study focuses on the preconditioning effect of volatile anesthetics and is not designed to explore the effects of volatile anesthetics on patients with ongoing stroke. While we are enthusiastic about our results, we would caution extrapolation of these results to other settings, such as endovascular stroke treatment.

Clinical Implications

The early postoperative period corresponds with a particularly high risk of having an ischemic stroke. Additionally, recognition of stroke during inpatient stays may be delayed because of comorbidities and hospital practice, while treatment options early after surgery may be limited. , Our data support the use of volatile anesthetics to prevent clinically relevant ischemic strokes during this highly vulnerable period.

This study found that the protective effect of volatile anesthetics might be of particular relevance in patients requiring general anesthesia while carrying a high baseline risk of postoperative ischemic stroke. In anticipation of potential ischemic events in such patients, prophylactic neuroprotection should be considered. We believe that certain patient populations who have a high risk of surgery‐related ischemic stroke (such as patients undergoing vascular or neurosurgical procedures) may have their risk mitigated with the use of volatile anesthetics. Simultaneously, our data encourage the preferential use of volatile anesthesia over TIVA using propofol in patients at particular risk of postoperative ischemic stroke. Since the observed effect was dose dependent, clinicians should be aware that, compared with lower doses or propofol‐based anesthesia, using higher doses of volatile anesthetics might be helpful for stroke prevention.

Future randomized controlled trials, especially in patients undergoing high‐risk procedures such as neurosurgery and vascular surgery, will be needed to confirm our findings.

In summary, volatile anesthetics had a dose‐dependent protective effect on the incidence and severity of ischemic stroke within 30 days after noncardiac surgery in this diverse cohort of 314 932 adult patients. The effect was found to be of specific importance during the early postoperative period.

Sources of Funding

This work was supported by Philanthropic Donors Jeffrey and Judy Buzen in an unrestricted grant to Matthias Eikermann. The funders had no role in the design or conduct of the study; the handling of data; the preparation of the manuscript, or the decision to submit it for publication.

Disclosures

Dr Hanafy reports a grant by the NIH (grant no. R01NS109174). Dr Houle reports grants from the National Institute of Neurological Disorders and Stroke (NINDS; PI), grants from the National Institute of General Medical Sciences (NIGMS), personal fees from Headache, personal fees from Anesthesiology, and personal fees from Cephalalgia, outside the submitted work. Dr Eikermann received funding for investigator‐initiated trials from Merck and honorarium for giving advice to Merck, and a grant by the NIH (grant no. UG3HL140177); he is an Associate Editor of the British Journal of Anaesthesia. The remaining authors have no disclosures to report.

Datas S1–S2

Tables S1–S5

Figures S1–S4

References ,

Click here for additional data file.