Incidence and risk factors of peri‐operative stroke in major non‐cardiovascular, non‐neurologic surgery—A retrospective register‐based cohort study

Peri‐operative stroke is a rare but serious surgical complication. Both overt and covert stroke, occurring in approximately 0.1% and 7% of cases, respectively, are associated with significant long‐term effects and increased morbidity.

stroke risk is unclear.This retrospective analysis presents incidence results for clinically evident stroke within 30 days of surgery from a Swedish multicenter cohort, and identifies associated case factors including advanced age and higher ASA class, being male and known cerebrovascular disease.

| INTRODUCTION
Despite advances, stroke continues to be a major cause of morbidity and mortality and was 2010 the second leading cause of death worldwide. 1In the peri-operative setting, stroke can be a catastrophic outcome, and carries a higher mortality compared with stroke in the non-surgical setting. 2 Although relatively rare in lowrisk patients and procedures (approximately 0.1%), 3 this complication is nevertheless substantial at a population level.Furthermore, the risk of peri-operative stroke in higher-risk patients increases to approximately 1.9%, even in relatively low-risk procedures. 3Additionally, the recent NeuroVISION trial investigating covert stroke (i.e., clinically unrecognised ischaemia seen on magnetic resonance imaging) in a relatively low-risk surgical population demonstrated a staggering 7% incidence of covert stroke.The same trial suggested that even a covert stroke can have a significant long-term effect on cognitive function and cerebrovascular risk. 4Taken together, these data indicate that efforts to reduce the incidence of peri-operative stroke could have a significant impact on public health.
Various pre-and intra-operative factors that increase the risk of stroke have been elucidated.In a 2011 study of 529,059 patients undergoing non-cardiac, non-neurosurgical procedures, age >61 years was identified as the strongest independent risk factor for perioperative stroke, followed by myocardial infarction within 6 months, acute renal failure or pre-existing dialysis, prior stroke/transient ischaemic attack, hypertension, chronic obstructive pulmonary disease and smoking. 3Other studies have supported these findings, 2,[5][6][7][8] as well as identifying diabetes mellitus and hyperglycaemia, 7,9 current or prior atrial fibrillation, 7,10,11 congestive heart failure, 2,7 valvular heart disease, 7 renal disease, 7 coronary artery disease, 12 cancer, patent foramen ovale 4,13 and migraine 14 as independent risk factors for peri-operative stroke. 2,7,8,13Additionally, the impact of preexisting medications, such as statins 15 and acetylsalicylic acid, 16,17 has also been studied in relation to peri-operative stroke with generally inconclusive results, along with ramifications of intra-operative management strategies, none of which clearly demonstrates superiority.
In summary, several potentially modifiable pre-and intraoperative risk factors appear to exist; however, robust evidence to guide clinical practice is still lacking.In light of previous research, our study sought to assess incidence and risk factors for peri-operative stroke in a large population of Swedish patients to lay a foundation for future research into potential areas of improvement in perioperative management leading to reduced peri-operative morbidity and, in turn, socio-economic benefits.

| Study design
This was a retrospective, register based multicentre cohort study including 23 hospitals in Sweden (centres specified in Table S4).All centres shared data voluntarily using their surgical planning software.
The study protocol was approved by the Regional Ethics Committee of Stockholm, Sweden, with protocol number 2014/1306-31/3, and the study was performed in accordance with the Declaration of Helsinki.Due to the retrospective, de-identified nature of the register data used in the study, acquisition of informed consent was neither necessary nor possible and this was approved by the Institutional Review Board.

| Data sources
Data were extracted de-identified from multiple sources for the period 2007-2014.Cross-matching between data sources was performed using the Swedish personal identification number assigned to all inhabitants at birth or immigration. 18The surgical planning software Orbit (Tietoevry, Oslo, Norway) was used to generate data relating to the surgical event.This was then cross-matched with data from registers belonging to the Swedish National Board of Health and Welfare: the national patient register (NPR) for both in-and outpatients, 19 the national cause of death register 20 and the prescribed drug register. 21Data were also linked to the national clinical quality register 'Riksstroke' which records high-resolution data from stroke cases across Sweden. 22A more detailed description of the various data sources and how they were used in the present study can be found in Table S6.

| Study participants
Patients over 18 years undergoing surgery between 1 January 2007 and 31 December 2014 were included in the study.Cases of neurologic, cardiac, vascular, obstetric, minor and ambulatory care surgery were excluded along with cases of missing surgical code data in Orbit and cases with missing ASA classification.A detailed list of included and excluded surgical procedures is provided in Table S7.
Pre-existing comorbidities were extracted on an individual level from the NPR for a period of 5 years by retrieving all registered ICD-diagnoses from entries with a discharge date within the 5 years preceding the index surgery.Additionally, by using the Prescribed Drug Register to find prescriptions of Metformin or Insulin dispensed in the 5 years preceding the index surgery, the presence of diabetes solely treated in primary care was established.Subsequent categorisation of specific comorbidities into organ groups was performed and details of this were provided in Table S5.

| Outcomes
The primary outcome of the study was the incidence of peri-operative stroke within 30 days from the index surgery.A peri-operative stroke was considered to have occurred if there was an entry of ischaemic stroke (ICD-code I63) in the high-resolution 'Riksstroke'-database on or within 30 days of an index surgery.The temporal detail of 'Riksstroke' is limited to whole days and therefore orthopaedic cases (surgery codes beginning with N) occurring on the same day as a stroke were excluded to reduce the risk of including surgeries occurring after rather than before the occurrence of stroke.Subsequently, cases from the NPR were added if they met the following criteria: both admission and discharge occurring between postoperative Days 1 and 30; and either a primary diagnosis of stroke (ICD-code I63) with no other records of stroke within 30 days or a secondary diagnosis of stroke (ICD-code I63) with no other records of stroke within 5 years.
Secondary outcomes were mortality within 30, 90 and 365 days, along with the composite outcome 'days alive and at home within 30 days after surgery' (DAH 30).DAH 30 is a composite outcome consisting of the total number of days spent out of hospital during the first 30 days following surgery along with mortality.Therefore, a lower DAH 30 corresponds to a poorer postoperative outcome, and death within 30 days from surgery always results in a DAH 30 of 0. 23

| Statistical and regression analyses
Statistical analysis of the absolute incidence of peri-operative stroke was performed using a χ 2 test for categorical variables, and a nonparametric Mann-Whitney U test for continuous variables.
To investigate the contribution of various pre-operative characteristics on the risk of subsequent peri-operative stroke, a logistic regression model was constructed with the dependent variable set as stroke within 30 days from surgery.Initially, a univariate approach was applied, followed by a full multivariable model including all variables reaching statistical significance in the univariate model.Further model testing was done using both a backwards stepwise logistic regression method as well as a Least Absolute Shrinkage and Selection Operator penalised logistic regression.Finally, however, the full multivariable model was judged to be most appropriate and employed.
Data were summarised in a forest plot.
F I G U R E 1 Flow-diagram of cases included in the study population.Initially, 1,125,434 potential cases were extracted from the Orbit software.After exclusion of cases not fulfilling various data quality requirements (see figure), a total of 434,516 cases remained.Finally, exclusion of cardiac, obstetric, neurologic, vascular and minor surgery resulted in 318,017 cases included in the study.
The effect of a peri-operative stroke on the composite outcome DAH 30 was assessed using a proportional odds logistic regression model after stratification of DAH 30 scores into similarly sized subgroups.The model was expanded integrating interaction terms identified as significantly influencing the risk of perioperative stroke in the above-mentioned multivariable logistic regression model.The final model was subject to diagnostic assessment, where the Brant test was used to examine the assumption of proportional odds, indicating this assumption was fulfilled for the dependent variable DAH 30.The final model was visualised as an effects plot.
Finally, mortality estimates were calculated by using a multivariable logistic regression model with all-cause mortality set as the dependent variable and model adjustment performed for known covariates identified in the above-mentioned logistic regression model.
Unfortunately, data quality issues precluded the use of more detailed temporal mortality analysis with Cox-regression modelling, since no reliable estimate of the exact time point of an event (i.e., stroke) could be extracted.Attempts to implement a Cox-regression model for mortality between postoperative Days 31 and 365 were also hindered by clear violation of the assumption of proportional hazards and were therefore not included in the study.
All statistical analysis was performed using R Statistics version 4.2.2 through RStudio. 24Packages used in the final analysis are summarised in Table S3.

| Patient characteristics
Following exclusion of neurologic, cardiac and major vascular surgery, a total of 318,017 cases were included in the analyses (Figure 1).
A slight majority of cases were female and approximately one third of cases were non-elective.Descriptive data, including past medical history for a period of 5 years, is summarised in Table 1.

| Primary outcome
During the study period, a total of 687 patients suffered a stroke within 30 days following surgery, corresponding to an incidence of 0.216% (Table 1).The incidence of postoperative stroke was associated with male gender, increasing age, non-elective procedures and higher ASA-classification, along with a history of comorbidities.

| Multivariable analysis of risk factors associated with peri-operative stroke
To investigate the effect of pre-operative characteristics on the risk of subsequent postoperative stroke a logistic regression was performed and adjusted odds ratios (ORs) of peri-operative stroke within 30 days were calculated (Figure 2).cerebrovascular disease was associated with an adj.OR of 2.72 (95% CI: 2.25-3.27)and both male gender and non-elective surgery were also shown to be significant risk factors.Additional specific comorbidities were analysed as independent predictors and are summarised in Table S2.

| Outcomes following peri-operative stroke
The impact of peri-operative stroke on subsequent morbidity and mortality was assessed by using a proportional odds logistic regression investigating the composite outcome DAH 30.Age, sex, nonelective surgery and ASA Physical Status were included as interaction terms influencing the main predictor of interest: stroke within 30 days following surgery.The model demonstrated with statistical significance that patients suffering a peri-operative stroke have a higher probability of a low DAH 30 score compared with patients without a peri-operative stroke (Figure 3).
The impact of peri-operative stroke on mortality was also assessed by using a multivariable logistic regression and demonstrated statistically significant increased adjusted odds of mortality in patients suffering a peri-operative stroke after adjusting for age, sex, ASA Physical Status and non-elective surgery (Table 2).S2.

Covariate
No Yes The effect of peri-operative stroke on DAH 30.A proportional-odds logistic regression was employed to investigate the effect of a peri-operative stroke within 30 days on the composite outcome 'Days alive and at home within 30 days after surgery' (DAH 30, see Section 2.4) and summarised in an effect plot.The model was statistically significant at a significance level of .05 and demonstrated a lower DAH 30 score in patients suffering a stroke within 30 days from surgery.DAH 30, 30 days after surgery.

| DISCUSSION
In this study, we examined the incidence of peri-operative stroke and associated pre-operative risk factors in a large cohort of noncardiovascular, non-neurologic patients collected between 2007 and 2014.Our reported incidence of approximately 0.22% is in line with previous research and clearly indicates stroke remains an important peri-operative complication to consider.This is especially pertinent given the findings in the 2019 NeuroVISION study 4 which demonstrated a significant detrimental effect on patient-centred outcomes even after a covert stroke (i.e., silent cerebral ischaemic event), an event with an incidence between one and two orders of magnitude greater than an overt stroke.Although the NeuroVISION study did not attempt to analyse the effect of pre-operative risk factors on the risk of subsequent covert stroke, the same cardiovascular risk factors identified as increasing the risk of overt stroke in our study could presumably also increase the risk of covert stroke.
In the multivariable logistic regression analysis performed, ASA Physical Status was one of the strongest pre-operative risk factors for peri-operative stroke.While this is not surprising, the stark increase in odds (an almost 10-fold increase when comparing ASA 4 to ASA 1) provides a potential for individualised peri-operative management.
This finding is also in line with previous studies showing a markedly increased risk of stroke in higher-risk patients, even when the surgical intervention remains relatively low risk. 3 addition to investigating the incidence and risk factors of perioperative stroke, we also examined its impact on both mortality and DAH 30, an important patient-centred outcome.Our results indicate peri-operative stroke significantly impacts both short-and long-term mortality, and furthermore markedly decreases DAH 30, thus leading to poorer patient outcomes.While not surprising, this result underlines the importance of efforts to reduce the incidence of this perioperative complication.
Our study has several inherent limitations.First, the data used in the study were generated between 2007 and 2014, making most of the data over a decade old.Potential developments in anaesthetic and surgical techniques, monitoring, diagnostics, risk-stratification and treatments implemented during the last decade may have had a significant impact on the incidence of peri-operative stroke, which would not be captured in our data.However, given our reported incidence of stroke is in line with both older 3 and more recent estimates of peri-operative stroke, 13,25 it is unlikely the magnitude of this effect is substantial.Additionally, the nature of our data presents several limitations.
The classification of postoperative stroke was based on a combination of Naturally, this introduces the risk of underestimating the incidence of stroke in orthopaedic surgery; however, since manual examination of patient records was neither possible nor ethically permitted, we feel this is an inherent and unresolvable limitation of our data.
Importantly, the current study does not attempt to evaluate intraoperative risk factors for peri-operative stroke, as such research would necessitate access to more detailed intraoperative monitoring.
There is, however, a significant need for such research since many of the risk factors mentioned in the introduction are modifiable.7][28] This mirrors results from larger studies on the general detrimental effects of IOH. 29,30Given studies such as Joshi et al. from 2012 31 place the lower limit of cerebral autoregulation at a mean arterial pressure (MAP) of around 66 mmHg, and transient drops in MAP under these levels frequently occur, there is a clinical rationale for future studies of intraoperative hypotension in relation to peri-operative stroke.
Additionally, cerebral autoregulation is known to be influenced by partial pressures of carbon dioxide in blood 32 -a direct effect of hypo-/ hyperventilation.In line with this, Vlisides et al. found an increased odds of peri-operative stroke in patients subjected to intraoperative hyper-or hypocapnia. 28The existence of other potentially modifiable risk factors such as intraoperative hyperglycaemia, atrial fibrillation, and anaemia is likely; however, these remain insufficiently characterised to guide clinical practice.
Despite inherent limitations, our study also has several notable strengths.Importantly, the data consist of cases from 23 different centres ranging from county to university hospitals.Furthermore, the size of the study cohort is large and on par with some of the largest retrospective studies in the field, permitting inference with high statistical power and adjustment for multiple covariates.The quality of the data is also a significant strength.By linking several registers using patients' person-number, 18 cross-validation of the data is made possible, along with simultaneous use of high-precision registers, such as Riksstroke and registers with near-universal coverage, such as the NPR.We therefore believe our estimated incidence of peri-operative stroke is highly accurate and closely reflects the true incidence in the entire Swedish population.Additionally, our study is, to our knowledge, the first analysis of peri-operative stroke performed on the general surgical population of Sweden.Finally, our study also provides an opportunity to evaluate the relatively new composite outcome DAH 30, adding an important patient-centred dimension to our results.
In conclusion, increasing ASA-class and age was clearly associated with an increased risk of peri-operative stroke, which in turn was associated with increased morbidity and mortality.Detailed preoperative risk stratification and individualised peri-operative management could potentially improve patient-centred outcomes and, in turn, have positive implications for public health.
higher and lower resolution data (Riksstroke and NPR, respectively), thus limiting the possibility of characterising the sub-type of stroke.The primary outcome consists of the incidence of ICD-10-classification I63, corresponding to a cerebral infarction, without the possibility of distinguishing specific aetiology (thromboses vs. emboli, pre-cerebral vs cerebral vessel occlusions, or arterial vs venous occlusions).In the lowerresolution data collected from the NPR, diagnoses were made on the date of discharge from hospital, limiting detailed temporal analysis.A clear example of the insufficient temporal detail of our data occurs in the case of admission under both a diagnosis of stroke and hip-fracture, where clinical reasoning suggests it being more likely that the hip-fracture is secondary to a post-stroke fall rather than the stroke being per-or postoperative.Exclusion of such cases was therefore necessary and based on clinical judgement, which resulted in 28 cases of stroke being excluded.
The strongest significant risk factors included increased age and ASA Physical Status, with age >90 having an adj.OR of 21 (95% confidence interval [CI]: 10.6-48.1)and ASA 4 having an adj.OR of 7.82 (95% CI: 5.03-12.5).A history of T A B L E 1 Patient characteristics.
Note: Baseline patient characteristics and demographics for the entire study cohort and the sub-group suffering a peri-operative stroke within 30 days.Comorbidities were available for a period of 5 years prior to the date of surgery.The absolute incidence of stroke was found to be 0.216% in the study cohort.Statistical significance was determined using a χ 2 test with p values <.05 considered significant.Additional patient characteristics can be found in TableS1.Abbreviation: COPD, chronic obstructive pulmonary disease.a χ 2 test.
T A B L E 2 Impact of stroke within 30 days on mortality.To investigate the impact of a peri-operative stroke within 30 days on patient mortality, a multivariable logistic regression was used and adjusted for covariates identified in the previous logistic regression (Figure2).Abbreviations: CI, confidence interval; OR, odds ratio.Adjusted for age, ASA-class, sex and non-elective surgery.Model: Multivariable logistic regression.b ORs for the predictor 'Stroke within 30 days'. a