Development and Validation of an Interpretable Risk Prediction Model for Perioperative Ischemic Stroke in Noncardiac, Nonvascular, and Nonneurosurgical Patients: A Retrospective Study

doi:10.21203/rs.3.rs-4683421/v1

Background

This study introduces an interpretable machine learning model, derived from patient data, to address the notable lack of perioperative stroke prediction tools for adults undergoing noncardiac, nonvascular, and nonneurosurgical procedures, thereby improving clinical decision-making.

Methods

A retrospective cohort study encompassed 106,328 patients aged 18 years or older who underwent non-cardiac, non-vascular, and non-neurosurgical surgeries in our institution. The training cohort included 74,429 patients with 140 perioperative stroke incidents, and the validation cohort comprised 31,899 patients with 59 incidents. Risk factors for perioperative stroke were identified using univariable logistic regression analysis. The Least Absolute Shrinkage and Selection Operator (LASSO) regression method screened variables, followed by the development, validation, and performance evaluation of the prediction model through multivariate logistic regression analysis.

Results

The established prediction model, leveraging 16 variables including demographic information, medical history, and pre- and post-operative data, demonstrated robust discriminatory capability in forecasting perioperative stroke (AUC = 0.919; 95% CI, 0.896–0.942). It also showed an excellent fit with the validation cohort (Hosmer–Lemeshow test, χ²=4.085, P = 0.906). Decision curve analysis affirmed the model's substantial net benefit.

Conclusion

Through the analysis of patients aged 18 and above undergoing specified surgeries, this study successfully identified risk factors for perioperative stroke. Subsequently, it developed and validated effective prediction models that exhibit notable predictive accuracy, thereby serving as a pivotal tool for clinicians in decision-making processes. These insights lay the groundwork for the prevention and enhanced perioperative management of stroke, marking a significant stride in patient care optimization.

Prediction model

Risk assessment

Perioperative stroke

Noncardiac

nonvascular

and nonneurosurgical procedures

General anesthesia

Perioperative stroke, a serious complication that can occur within 30 days after surgery, leads to significant neurological dysfunction, prolonged hospitalization, and increased mortality rates¹. The incidence of perioperative stroke varies across surgery types, ranging from 0.1–1.9% in noncardiac surgeries to 1.9–9.7% in cardiovascular surgeries^2,3. Although various risk factors for perioperative stroke have been identified, traditional statistical-based prediction models exhibit limitations in addressing nonlinearity and variable selection issues⁴. Recently, machine learning has been recognized as a promising tool for predicting perioperative stroke among patients undergoing noncardiac surgeries^5,6.

The aim of this study is to gather the preoperative and intraoperative data, including the relevant variables and demographics, of patients undergoing noncardiac surgeries to develop an interpretable machine learning model for perioperative stroke prediction. The results of the study will improve physicians’ understanding of the mechanisms of perioperative stroke so as to facilitate informed decision-making and stroke prevention and prediction.

Study design and study population

Data from the electronic medical records of patients treated in Henan Provincial People's Hospital from November 2014 to June 2021 were retrospectively studied. The informed consent was exempted by the Henan Provincial People's Hospital's Ethics Review Committee (Approval No. 2021 − 157). The committee determined that the study met the criteria for waiver of informed consent as per relevant guidelines. All methods were performed in accordance with the relevant guidelines and regulations. The study was approved by the Henan Provincial People's Hospital's Ethics Review Committee (Approval No. 2021 − 157). The data collected included demographic characteristics, laboratory results, and comorbidities. The inclusion criteria included noncardiac, nonvascular and nonneurosurgical surgery patients aged 18 years and older with an American Society of Anaesthesiologists (ASA) grade of I-III. Cardiac and major vascular surgery patients, neurosurgery patients, neurology and neuro-ICU inpatients, other critically ill ICU patients, Grade IV-V patients, and patients with insufficient clinical or laboratory data were excluded. The study included 106,328 noncardiac, nonvascular and nonneurosurgical surgery patients, but only those who met the inclusion criteria were included. For a detailed illustration of the processing steps, refer to Fig. 1.

Variables

This study included the clinical data of 106,328 patients who underwent noncardiac, nonvascular and noncardiac surgery and for whom perioperative demographic, clinical, imaging, and laboratory data were comprehensively collected. The study considered the following variables:

The demographic characteristics included age, sex, height, weight, and BMI.

The surgery-related factors, such as ASA grade, type and duration of surgery, anaesthesia duration and method, urine output, and blood loss volume, were also included.

The comorbidities included a history of smoking and alcohol consumption, ascites, hypertension, diabetes mellitus, various cardiac conditions, COPD, renal issues, and cerebrovascular diseases.

The laboratory test results included blood counts; serum creatinine levels; albumin levels; liver enzymes; and various coagulation profiles.

Preoperative medications included a range of antihypertensive, anticoagulant, antiplatelet, and other drugs, along with specific details such as the duration of discontinuation of certain medications.

The vital signs monitored intraoperatively included blood pressure, heart rate, temperature, BIS value, and end-tidal carbon dioxide.

Intraoperative medications ranged from inhaled anaesthetics to various drugs, including diuretics and anticoagulants.

Intraoperative fluids and blood transfusions included colloids, crystalloids, and various blood components.

Intraoperative vasopressor drugs encompassed a range of medications from ephedrine to phenylephrine.

Postoperative medications included statins, anticoagulants, antiplatelets, and other drugs, along with details regarding their dosages and durations.

Statistical analysis

In this study, a variety of statistical methods were employed to analyse and describe both continuous and categorical variables. Continuous variables are presented as medians and interquartile ranges (25th to 75th percentiles), and differences between them were assessed using the Mann‒Whitney U test. Categorical variables are presented in terms of frequencies and percentages, with differences evaluated using the chi-square test or Fisher's exact test.

Logistic regression analysis was conducted using the glmnet package in R, applying the LASSO method for variable selection. Significant factors were incorporated into the logistic regression model via a forward stepwise selection procedure to minimize the Akaike information criterion (AIC) value and ensure minimal variable inclusion.

Multivariate logistic regression analysis was used to assess multicollinearity among variable combinations by calculating the variance inflation factor (VIF). VIF values less than 10 suggested an absence of multicollinearity concerns. For variables significantly correlated with the main outcomes, odds ratios (ORs) and 95% confidence intervals (CIs) were computed.

The model's discriminative ability was evaluated through the construction of a diagnostic chart that included important variables. Discriminative ability was assessed using the receiver operating characteristic (ROC) curve, with the area under the curve (AUC) quantifying performance. Calibration was evaluated using calibration plots and Spiegelhalter's Z test, which indicated good calibration when the predicted values closely matched the actual risk.

A sample of 70% of the cohort generated randomly using a seed was applied for the training cohort; the remaining 30% was used for the validation cohort. The rms package in R was used for graphic evaluation. All tests were two-sided, with a p value less than 0.05 indicating statistical significance. The statistical analysis was conducted using R version 4.3.1.

A total of 106,328 adult patients met the inclusion criteria during the study period, and the study cohort consisted of 70,718 patients after application of exclusion criteria. In our study, we analysed the data of 199 perioperative stroke patients, allocating 140 to the training cohort and 59 to the validation cohort. Tables 1 and 2 provide descriptive statistical analyses, where Table 1 compares patients with and without ischemic stroke and Table 2 contrasts the training and validation cohorts. Comparative analysis revealed no statistically significant differences in clinical information between the training cohort and the validation cohort (P > 0.05).

Table 1

Baseline demographic and clinical characteristics of included patients diagnosed with or without ischemic stroke.
			Ischemic stroke
Variables		Total (n = 106328)	No (n = 106129)	Yes (n = 199)	P
Emergency operation, n (%)					< 0.001
	No	92972 (87)	92818 (87)	154 (77)
	Yes	13356 (13)	13311 (13)	45 (23)
Sex, n (%)					< 0.001
	male	46481 (44)	46369 (44)	112 (56)
	female	59847 (56)	59760 (56)	87 (44)
Age, (Median [Q1, Q3]), yr		52 (40, 63)	52 (40, 63)	68 (62, 75)	< 0.001
Duration of the procedure, (Median [Q1, Q3]), min		140 (100, 210)	140 (100, 210)	170 (120, 245)	< 0.001
Duration of anesthesia, (Median [Q1, Q3]), min		156 (110, 230)	155 (110, 230)	190 (133, 265)	< 0.001
ASA classification, n (%)					< 0.001
	I	10816 (10)	10816 (10)	0 (0)
	II	79995 (75)	79897 (75)	98 (49)
	III	15517 (15)	15416 (15)	101 (51)
Methods of anesthesia, n (%)					0.106
	Simple general anesthesia	72209 (68)	72085 (68)	124 (62)
	General anesthesia combined with nerve block	34119 (32)	34044 (32)	75 (38)
Urine output, (Median [Q1, Q3]), ml		300 (100, 500)	300 (100, 500)	300 (200, 500)	0.115
Amount of blood loss, (Median [Q1, Q3]), ml		50 (20, 200)	50 (20, 200)	100 (30, 200)	< 0.001
History of smoking, n (%)					0.881
	No	97817 (92)	97635 (92)	182 (91)
	Yes	8511 (8)	8494 (8)	17 (9)
History of drinking, n (%)					0.002
	No	95825 (90)	95659 (90)	166 (83)
	Yes	10503 (10)	10470 (10)	33 (17)
Ascites, n (%)					1
	No	101938 (96)	101747 (96)	191 (96)
	Yes	4390 (4)	4382 (4)	8 (4)
Hypertension, n (%)					< 0.001
	No	93361 (88)	93227 (88)	134 (67)
	Yes	12967 (12)	12902 (12)	65 (33)
Diabetes, n (%)					< 0.001
	No	99138 (93)	98983 (93)	155 (78)
	Yes	7190 (7)	7146 (7)	44 (22)
Coronary heart disease, n (%)					< 0.001
	No	102624 (97)	102448 (97)	176 (88)
	Yes	3704 (3)	3681 (3)	23 (12)
Stenocardia, n (%)					< 0.001
	No	106155 (100)	105961 (100)	194 (97)
	Yes	173 (0)	168 (0)	5 (3)
Valvular heart disease, n (%)					0.014
	No	105583 (99)	105389 (99)	194 (97)
	Yes	745 (1)	740 (1)	5 (3)
Myocardial infarction, n (%)					< 0.001
	No	105968 (100)	105776 (100)	192 (96)
	Yes	360 (0)	353 (0)	7 (4)
Heart failure, n (%)					0.02
	No	106214 (100)	106017 (100)	197 (99)
	Yes	114 (0)	112 (0)	2 (1)
Arrhythmia, n (%)					0.132
	No	105290 (99)	105095 (99)	195 (98)
	Yes	1038 (1)	1034 (1)	4 (2)
Atrial fibrillation, n (%)					0.134
	No	105987 (100)	105790 (100)	197 (99)
	Yes	341 (0)	339 (0)	2 (1)
Intracoronary stent implantation, n (%)					0.027
	No	105728 (99)	105533 (99)	195 (98)
	Yes	600 (1)	596 (1)	4 (2)
Cardiac surgery, n (%)					< 0.001
	No	13284 (12)	13284 (13)	0 (0)
	Yes	93044 (88)	92845 (87)	199 (100)
Peripheral vascular disease, n (%)					0.013
	No	99429 (94)	99252 (94)	177 (89)
	Yes	6899 (6)	6877 (6)	22 (11)
COPD, n (%)					1
	No	105543 (99)	105345 (99)	198 (99)
	Yes	785 (1)	784 (1)	1 (1)
Dialysis, n (%)					0.594
	No	104364 (98)	104167 (98)	197 (99)
	Yes	1964 (2)	1962 (2)	2 (1)
Renal insufficiency, n (%)					0.638
	No	105685 (99)	105486 (99)	199 (100)
	Yes	643 (1)	643 (1)	0 (0)
History of cerebrovascular disease, n (%)					< 0.001
	No	102270 (96)	102128 (96)	142 (71)
	Yes	4058 (4)	4001 (4)	57 (29)
TIA, n (%)					1
	No	106261 (100)	106062 (100)	199 (100)
	Yes	67 (0)	67 (0)	0 (0)
Stroke, n (%)					< 0.001
	No	102725 (97)	102579 (97)	146 (73)
	Yes	3603 (3)	3550 (3)	53 (27)
Paraplegia, n (%)					< 0.001
	No	106134 (100)	105939 (100)	195 (98)
	Yes	194 (0)	190 (0)	4 (2)
Cancer, n (%)					0.128
	No	93815 (88)	93632 (88)	183 (92)
	Yes	12513 (12)	12497 (12)	16 (8)
Preoperative white blood cell count, (Median [Q1, Q3]), 10^9/L		6.35 (5.1, 8.2)	6.35 (5.1, 8.2)	6.73 (5.16, 8.54)	0.139
Preoperative red blood cell count, (Median [Q1, Q3]), 10^12/L		4.21 (3.8, 4.61)	4.22 (3.8, 4.61)	3.99 (3.54, 4.4)	< 0.001
Preoperative platelet count, (Median [Q1, Q3]), 10^9/L		226 (183, 274)	226 (183, 274)	215 (165.5, 271)	0.04
Preoperative hemoglobin levels, (Median [Q1, Q3]), g/L		126 (112, 139)	126 (112, 139)	120 (106.5, 134)	< 0.001
Preoperative serum creatinine, (Median [Q1, Q3]), µmol/L		55 (47, 66)	55 (47, 66)	56 (48, 72)	0.024
Preoperative serum albumin levels, (Median [Q1, Q3]), g/L		40.7 (37, 44)	40.7 (37, 44)	37 (34.05, 40.5)	< 0.001
Preoperative alanine aminotransferase, (Median [Q1, Q3]), U/L		18 (12.6, 29)	18 (12.6, 29)	18 (12.5, 31)	0.681
Preoperative aspartate aminotransferase, (Median [Q1, Q3]), U/L		19.4 (16, 25.7)	19.4 (16, 25.7)	21.5 (16.65, 30)	0.01
Preoperative serum sodium, (Median [Q1, Q3]), mmol/L		141 (139, 142)	141 (139, 142)	140.7 (138, 142)	0.013
Preoperative blood potassium, (Median [Q1, Q3]), mmol/L		4.17 (3.9, 4.43)	4.17 (3.9, 4.43)	4.09 (3.74, 4.38)	0.001
Preoperative blood calcium, (Median [Q1, Q3]), mmol/L		2.24 (2.14, 2.33)	2.24 (2.14, 2.33)	2.18 (2.08, 2.28)	< 0.001
Preoperative thrombin time, (Median [Q1, Q3]), s		16.5 (15.5, 17.6)	16.5 (15.5, 17.6)	16.6 (15.6, 17.5)	0.644
Preoperative plasma activated partial thromboplastin time, (Median [Q1, Q3]), s		34.1 (30.4, 37.5)	34.1 (30.4, 37.5)	34.1 (29.65, 37.5)	0.905
Preoperative plasma prothrombin time, (Median [Q1, Q3]), s		12.1 (11.5, 12.9)	12.1 (11.5, 12.9)	12.3 (11.4, 13.1)	0.059
Preoperative international normalized ratio, (Median [Q1, Q3])		0.93 (0.87, 1.02)	0.93 (0.87, 1.02)	0.95 (0.88, 1.08)	0.006
Abbreviations: ASA, American Society of Anesthesiologists; COPD, chronic obstructive pulmonary disease; TIA, transient ischemic attack.

Table 2

Baseline demographic and clinical characteristics of included patients between training cohort and validation cohort
Variables		Total (n = 106328)	Training cohort (n = 74429)	Validation cohort (n = 31899)	P
Ischemic stroke, n (%)					0.975
	No	106129 (100)	74289 (100)	31840 (100)
	Yes	199 (0)	140 (0)	59 (0)
Emergency operation, n (%)					0.894
	No	92972 (87)	65087 (87)	27885 (87)
	Yes	13356 (13)	9342 (13)	4014 (13)
Sex, n (%)					0.572
	male	46481 (44)	32494 (44)	13987 (44)
	female	59847 (56)	41935 (56)	17912 (56)
Age, (Median [Q1, Q3]), yr		52 (40, 63)	52 (40, 63)	52 (40, 63)	0.316
Duration of the procedure, (Median [Q1, Q3]), min		140 (100, 210)	140 (100, 210)	140 (100, 210)	0.56
Duration of anesthesia, (Median [Q1, Q3]), min		156 (110, 230)	156 (110, 230)	155 (110, 230)	0.582
ASA classification, n (%)					0.875
	I	10816 (10)	7579 (10)	3237 (10)
	II	79995 (75)	55964 (75)	24031 (75)
	III	15517 (15)	10886 (15)	4631 (15)
Methods of anesthesia, n (%)					0.47
	Simple general anesthesia	72209 (68)	50495 (68)	21714 (68)
	General anesthesia combined with nerve block	34119 (32)	23934 (32)	10185 (32)
Urine output, (Median [Q1, Q3]), ml		300 (100, 500)	300 (100, 500)	300 (100, 500)	0.938
Amount of blood loss, (Median [Q1, Q3]), ml		50 (20, 200)	50 (20, 200)	50 (20, 200)	0.128
History of smoking, n (%)					0.508
	No	97817 (92)	68444 (92)	29373 (92)
	Yes	8511 (8)	5985 (8)	2526 (8)
History of drinking, n (%)					0.388
	No	95825 (90)	67038 (90)	28787 (90)
	Yes	10503 (10)	7391 (10)	3112 (10)
Ascites, n (%)					0.698
	No	101938 (96)	71344 (96)	30594 (96)
	Yes	4390 (4)	3085 (4)	1305 (4)
Hypertension, n (%)					0.859
	No	93361 (88)	65343 (88)	28018 (88)
	Yes	12967 (12)	9086 (12)	3881 (12)
Diabetes, n (%)					0.416
	No	99138 (93)	69365 (93)	29773 (93)
	Yes	7190 (7)	5064 (7)	2126 (7)
Coronary heart disease, n (%)					0.367
	No	102624 (97)	71811 (96)	30813 (97)
	Yes	3704 (3)	2618 (4)	1086 (3)
Stenocardia, n (%)					0.816
	No	106155 (100)	74306 (100)	31849 (100)
	Yes	173 (0)	123 (0)	50 (0)
Valvular heart disease, n (%)					1
	No	105583 (99)	73907 (99)	31676 (99)
	Yes	745 (1)	522 (1)	223 (1)
Myocardial infarction, n (%)					0.454
	No	105968 (100)	74170 (100)	31798 (100)
	Yes	360 (0)	259 (0)	101 (0)
Heart failure, n (%)					0.728
	No	106214 (100)	74347 (100)	31867 (100)
	Yes	114 (0)	82 (0)	32 (0)
Arrhythmia, n (%)					0.245
	No	105290 (99)	73720 (99)	31570 (99)
	Yes	1038 (1)	709 (1)	329 (1)
Atrial fibrillation, n (%)					0.332
	No	105987 (100)	74199 (100)	31788 (100)
	Yes	341 (0)	230 (0)	111 (0)
Intracoronary stent implantation, n (%)					0.687
	No	105728 (99)	74004 (99)	31724 (99)
	Yes	600 (1)	425 (1)	175 (1)
Cardiac surgery, n (%)					0.457
	No	13284 (12)	9336 (13)	3948 (12)
	Yes	93044 (88)	65093 (87)	27951 (88)
Peripheral vascular disease, n (%)					0.573
	No	99429 (94)	69621 (94)	29808 (93)
	Yes	6899 (6)	4808 (6)	2091 (7)
COPD, n (%)					0.755
	No	105543 (99)	73884 (99)	31659 (99)
	Yes	785 (1)	545 (1)	240 (1)
Dialysis, n (%)					1
	No	104364 (98)	73054 (98)	31310 (98)
	Yes	1964 (2)	1375 (2)	589 (2)
Renal insufficiency, n (%)					0.769
	No	105685 (99)	73975 (99)	31710 (99)
	Yes	643 (1)	454 (1)	189 (1)
History of cerebrovascular disease, n (%)					0.574
	No	102270 (96)	71605 (96)	30665 (96)
	Yes	4058 (4)	2824 (4)	1234 (4)
TIA, n (%)					0.709
	No	106261 (100)	74384 (100)	31877 (100)
	Yes	67 (0)	45 (0)	22 (0)
Stroke, n (%)					0.988
	No	102725 (97)	71906 (97)	30819 (97)
	Yes	3603 (3)	2523 (3)	1080 (3)
Paraplegia, n (%)					0.227
	No	106134 (100)	74285 (100)	31849 (100)
	Yes	194 (0)	144 (0)	50 (0)
Cancer, n (%)					0.149
	No	93815 (88)	65600 (88)	28215 (88)
	Yes	12513 (12)	8829 (12)	3684 (12)
Preoperative white blood cell count, (Median [Q1, Q3]), 10^9/L		6.35 (5.1, 8.2)	6.36 (5.1, 8.2)	6.32 (5.09, 8.19)	0.136
Preoperative red blood cell count, (Median [Q1, Q3]), 10^12/L		4.21 (3.8, 4.61)	4.21 (3.8, 4.61)	4.22 (3.8, 4.61)	0.408
Preoperative platelet count, (Median [Q1, Q3]), 10^9/L		226 (183, 274)	225 (183, 273)	226 (183, 274)	0.173
Preoperative hemoglobin levels, (Median [Q1, Q3]), g/L		126 (112, 139)	126 (112, 139)	126 (112, 139)	0.412
Preoperative serum creatinine, (Median [Q1, Q3]), µmol/L		55 (47, 66)	55 (47, 66)	55 (48, 66)	0.623
Preoperative serum albumin levels, (Median [Q1, Q3]), g/L		40.7 (37, 44)	40.7 (37, 44)	40.7 (37, 44)	0.228
Preoperative alanine aminotransferase, (Median [Q1, Q3]), U/L		18 (12.6, 29)	18 (12.6, 29)	18 (12.8, 29)	0.162
Preoperative aspartate aminotransferase, (Median [Q1, Q3]), U/L		19.4 (16, 25.7)	19.4 (16, 25.7)	19.4 (16, 25.7)	0.875
Preoperative serum sodium, (Median [Q1, Q3]), mmol/L		141 (139, 142)	141 (139, 142)	141 (139, 143)	0.078
Preoperative blood potassium, (Median [Q1, Q3]), mmol/L		4.17 (3.9, 4.43)	4.17 (3.9, 4.43)	4.17 (3.91, 4.43)	0.685
Preoperative blood calcium, (Median [Q1, Q3]), mmol/L		2.24 (2.14, 2.33)	2.24 (2.14, 2.33)	2.24 (2.14, 2.33)	0.736
Preoperative thrombin time, (Median [Q1, Q3]), s		16.5 (15.5, 17.6)	16.5 (15.5, 17.6)	16.5 (15.5, 17.6)	0.588
Preoperative plasma activated partial thromboplastin time, (Median [Q1, Q3]), s		34.1 (30.4, 37.5)	34 (30.4, 37.5)	34.1 (30.4, 37.5)	0.775
Preoperative plasma prothrombin time, (Median [Q1, Q3]), s		12.1 (11.5, 12.9)	12.1 (11.5, 12.9)	12.1 (11.5, 12.9)	0.732
Preoperative international normalized ratio, (Median [Q1, Q3])		0.93 (0.87, 1.02)	0.93 (0.87, 1.02)	0.93 (0.87, 1.02)	0.603
Abbreviations: ASA, American Society of Anesthesiologists; COPD, chronic obstructive pulmonary disease; TIA, transient ischemic attack.

We employed the LASSO logistic regression algorithm to identify predictive factors and to develop a corresponding model. Out of 179 factors, 31 with nonzero coefficients were selected based on an optimal lambda value of 0.0003226 (Fig. 2).

Subsequent logistic regression analysis, along with AIC comparison, led to the final model comprising 16 factors: sex, age, alcohol history, diabetes status, angina status, valvular heart disease status, cerebrovascular disease history, intraoperative MAP ≤ 60 mmHg for 5–10 minutes, preoperative serum sodium, preoperative ACEI use, perioperative nonsteroidal drug use, postoperative intravenous thrombolysis, postoperative statins, anticoagulants, antiplatelet medications, and butylphthalein (Table 3). The multiple logistic regression model successfully identified independent predictors of perioperative stroke risk in noncardiac, nonvascular and nonneurosurgical surgery patients.

Table 3

Multivariate logistic regression analysis for risk factors associated with acute ischemic stroke in perioperative noncardiac, nonvascular, and nonneurosurgical surgery patients.
Variables	B	SE	OR	CI	Z	P
Sex[female]	-0.408	0.199	0.67	0.45–0.98	-2.046	0.041
Age	0.057	0.008	1.06	1.04–1.08	7.145	0
Drinking history [Yes]	0.618	0.248	1.86	1.14–3.02	2.49	0.013
Diabetes [Yes]	0.547	0.218	1.73	1.13–2.65	2.501	0.012
Angina pectoris [Yes]	2.021	0.562	7.55	2.51–22.71	3.597	0
Valvular heart disease [Yes]	1.214	0.529	3.37	1.19–9.49	2.296	0.022
History of cerebrovascular disease [Yes]	0.782	0.223	2.19	1.41–3.38	3.509	0
Intraoperative MAP ≤ 60 mmHg for 5–10 minutes [Yes]	0.955	0.439	2.6	1.1–6.14	2.177	0.03
Preoperative blood sodium level	-0.05	0.023	0.95	0.91-1	-2.197	0.028
Preoperative ACEI drugs [Yes]	0.681	0.24	1.98	1.23–3.16	2.833	0.005
Perioperative non-steroidal drugs [Yes]	1.085	0.478	2.96	1.16–7.55	2.27	0.023
Postoperative intravenous thrombolysis [Yes]	4.303	1.224	73.94	6.71-814.22	3.517	0
Postoperative statins [Yes]	0.659	0.282	1.93	1.11–3.36	2.337	0.019
Postoperative anticoagulants [Yes]	1.126	0.217	3.08	2.02–4.72	5.199	0
Postoperative antiplatelet drugs [Yes]	0.57	0.262	1.77	1.06–2.96	2.175	0.03
Postoperative butylphthalides [Yes]	3.915	0.282	50.13	28.85–87.13	13.866	0
Abbreviations: B, regression coefficient; SE, standard error; OR, odds radio; CI, credibility interval.

After validating the training cohort and the validation cohort in R Studio, we determined the model's AUC values for both. Specifically, in the training cohort, the model's AUC was 0.919 (95% CI 0.896–0.942, P < 0.05), and in the validation cohort, it was 0.909 (95% CI 0.870–0.949, P < 0.05) (Fig. 3).

Calibration curve analysis revealed a consistent relationship between the predicted and actual probabilities in both sets, as indicated by Brier scores of 0.002. Additionally, the Hosmer–Lemeshow test results indicated a good model fit in both sets (training cohort: χ² = 5.202, P = 0.816; validation cohort: χ² = 4.085, P = 0.906) (Fig. 3).

Decision curve analysis was performed to establish the threshold probabilities of 0–93% and 0–92% for the training cohort and the validation cohort, respectively, indicating the potential clinical utility of the model for noncardiac, nonvascular and nonneurosurgical surgery patients (Fig. 4).

Our thorough validation process supported the model's effectiveness in clinical settings. We constructed and validated a logistic regression model based on patient characteristics and stroke risk factors using diverse graphical methods for evaluation.

Perioperative strokes in noncardiac and nonneurological surgery patients are relatively rare, occurring in approximately 0.1–0.8% of patients. Notably, among patients older than 65 years, the incidence is as high as 7%, with many strokes remaining undetected⁷. Although less frequently diagnosed following noncardiac, noncarotid, or nonneurological surgeries, these strokes are significant postoperative complications. These conditions are linked to increased perioperative mortality and morbidity risks, longer hospital stays, and higher health care costs^3,8.

Due to complex and diverse risk factors, preventing perioperative strokes is challenging. Common concurrent conditions in these patients include atrial fibrillation, diabetes, and hypertension, each of which is a risk factor for major stroke ⁹. Additionally, stress responses during the perioperative period, medication adjustments, and the surgery itself can increase the stroke risk¹.

Clinical prediction models are crucial for accurately predicting the occurrence of perioperative stroke. These models, which integrate factors such as age, medical history, and type of surgery, effectively identify high-risk patients and outperform traditional assessment methods¹⁰. Clinicians must develop personalized treatment plans to enhance the treatment efficacy and reduce the frequency of unnecessary interventions. However, the effectiveness of the treatment depends on the quality of the data and requires adaptation to various scenarios¹¹.

Applying these models in resource-limited settings is challenging, and therefore requires continuous improvements to ensure its relevancy in clinical practice and to bridge the clinical-research gap^10,11.

The focus of this study is ischaemic stroke, particularly the risk factors associated with perioperative stroke. Through statistical analysis and comprehensive factor evaluation, the model's reliability in predicting the risk of perioperative stroke in noncardiac surgery patients was confirmed^10,11. The model can assist clinicians in devising early, tailored treatment strategies.

Our model included 16 predictive factors¹²: sex, age, alcohol history, diabetes status, angina status, valvular heart disease status, history of cerebrovascular disease, preoperative serum sodium levels, preoperative ACE inhibitor use, intraoperative MAP ≤ 60 mmHg for 5–10 minutes, perioperative nonsteroidal anti-inflammatory drug use, postoperative venous thrombolysis, statins, anticoagulants, antiplatelet therapy, and postoperative dantrolene use. These factors significantly influence perioperative stroke risk, thus enhancing the clinical relevance of our model.

Preoperatively, stroke risk is positively correlated with factors such as male sex, advanced age, alcohol consumption, diabetes, angina, valvular heart disease, and cerebrovascular disease history¹³. Additionally, preoperative low serum sodium levels and cessation of ACE inhibitors during the perioperative period are associated with an increased risk^14,15. During surgery, prolonged low MAP and nonsteroidal anti-inflammatory drug use are key risk factors^12,13.

Postoperatively, the use of statins may reduce the risk, but the impact of anticoagulants and antiplatelet drugs requires further investigation¹². Postoperative butylphthalin use may increase the risk of HELLP syndrome.

In related research, Wu and Fang (2020) developed a stroke prediction model targeting individuals older than 60 years of age utilizing SMOTE technology for data balance correction and identified sex, LDL cholesterol, blood glucose, hypertension, and uric acid as key predictors¹⁶. In another study involving patients older than 65 years who underwent noncardiac surgery, the authors used diverse machine learning techniques to evaluate factors, including age, history of chest pain, heart failure symptoms, high-risk surgeries, intraoperative blood pressure, serum creatinine levels, left ventricular ejection fraction, and perioperative transfusions, ultimately developing a model to predict the occurrence of adverse cardio-cerebral events¹⁷. These findings are significant, particularly when applying machine learning techniques for risk factor analysis.

Unlike similar studies, our research included adult patients of all ages and employed advanced machine learning techniques. Utilizing comprehensive imputations and logistic and LASSO regression for selecting predictive factors, our model demonstrated remarkable AUC values and distinct clinical relevance as a result of internal validation at our institution¹⁸.

Nevertheless, our study has several limitations. Because this was a single-centre investigation, the conclusions require external validation¹⁹. Moreover, the low stroke incidence and small sample size may introduce selection bias, despite randomized sampling, thus necessitating extensive multicentre studies for thorough validation. Although our model exhibited significant AUC values and clinical relevance, accessing some data was challenging, thereby impacting its practicality. Finally, considering the retrospective nature of our study, future research should be prospective to further explore potential risk factors for perioperative stroke in noncardiac, nonvascular and nonneurosurgical surgery patients, with the aim of creating models with higher predictive accuracy.

This study identified the risk factors for perioperative stroke in noncardiac, nonvascular and nonneurosurgical surgery patients, which provided a basis for better management and prevention of perioperative stroke. A new model for predicting the risk of perioperative stroke was constructed and validated. It showed good prediction performance and may be used as a decision-making assistant tool for clinicians. These data lay a foundation for future large-scale prospective clinical trials.

Data availability statement

The original contributions presented in this study are included in the article, further inquiries can be directed to the corresponding author/s.

Ethics statement

The studies involving human participants were reviewed and approved by the Henan Provincial People's Hospital's Ethics Review Committee. The ethics committee waived the requirement of written informed consent for participation.

Author contributions

Xuhui Cong wrote the manuscript with contributions from all authors. Jiaqiang Zhang designed the study. Xuli Zou, Ruilou Zhu, Yubao Li and Lu Liu were responsible for data extraction and acquisition. Xuhui Cong, Ruilou Zhu, Yubao Li and Lu Liu designed and conducted the statistical analyses. All authors critically reviewed the report and approved the final version.

Funding

This study was supported by grants from the National Natural Science Foundation of China (grant nos. 82071217).

Acknowledgments

We would like to thank Linlin Xu and Tongyan Sun of Hangzhou Le9 Healthcare Technology Co., Ltd., for their assistance in the clinical data extraction during the study.

Competing interests

The authors declare no competing interests.

Vlisides, P. & Mashour, G. A. Perioperative stroke. Can J Anaesth 63, 193-204, https://doi.org/10.1007/s12630-015-0494-9 (2016).
Marcucci, M., Chan, M. T. V., Smith, E. E., Absalom, A. R. & Devereaux, P. J. Prevention of perioperative stroke in patients undergoing non-cardiac surgery. Lancet Neurol 22, 946-958, https://doi.org/10.1016/s1474-4422(23)00209-0 (2023).
Ng, J. L., Chan, M. T. & Gelb, A. W. Perioperative stroke in noncardiac, nonneurosurgical surgery. Anesthesiology 115, 879-890, https://doi.org/10.1097/ALN.0b013e31822e9499 (2011).
Chahine, Y. et al. Machine Learning and the Conundrum of Stroke Risk Prediction. Arrhythm Electrophysiol Rev 12, e07, https://doi.org/10.15420/aer.2022.34 (2023).
Abraham, J. et al. Integrating machine learning predictions for perioperative risk management: Towards an empirical design of a flexible-standardized risk assessment tool. J Biomed Inform 137, 104270, https://doi.org/10.1016/j.jbi.2022.104270 (2023).
Fernandes, M. P. B., Armengol de la Hoz, M., Rangasamy, V. & Subramaniam, B. Machine Learning Models with Preoperative Risk Factors and Intraoperative Hypotension Parameters Predict Mortality After Cardiac Surgery. J Cardiothorac Vasc Anesth 35, 857-865, https://doi.org/10.1053/j.jvca.2020.07.029 (2021).
Lindberg, A. P. & Flexman, A. M. Perioperative stroke after non-cardiac, non-neurological surgery. BJA Educ 21, 59-65, https://doi.org/10.1016/j.bjae.2020.09.003 (2021).
Lewis, D. J., Al-Ghazawi, S. S., Al-Robaidi, K. A. & Thirumala, P. D. Perioperative stroke associated in-hospital morbidity and in-hospital mortality in common non-vascular non-neurological surgery. J Clin Neurosci 67, 32-39, https://doi.org/10.1016/j.jocn.2019.06.034 (2019).
Reinert, N. J. et al. Perioperative stroke-related mortality after non-cardiovascular, non-neurological procedures: A retrospective risk factor evaluation of common surgical comorbidities. J Perioper Pract 31, 80-88, https://doi.org/10.1177/1750458920911830 (2021).
Goeller, J. K. & Bartels, K. Improving prediction to prevent perioperative morbidity. Br J Anaesth 127, 671-674, https://doi.org/10.1016/j.bja.2021.08.004 (2021).
Woo, S. H., Marhefka, G. D., Cowan, S. W. & Ackermann, L. Development and Validation of a Prediction Model for Stroke, Cardiac, and Mortality Risk After Non-Cardiac Surgery. J Am Heart Assoc 10, e018013, https://doi.org/10.1161/jaha.120.018013 (2021).
Vlisides, P. E. & Moore, L. E. Stroke in Surgical Patients. Anesthesiology 134, 480-492, https://doi.org/10.1097/aln.0000000000003664 (2021).
Pai, S. L., Wang, R. D. & Aniskevich, S. Perioperative stroke: incidence, etiologic factors, and prevention. Minerva Anestesiol 83, 1178-1189, https://doi.org/10.23736/s0375-9393.17.11976-0 (2017).
Khan, F. W. et al. Hyponatremia: An Overlooked Risk Factor Associated With Adverse Outcomes After Cardiac Surgery. Ann Thorac Surg 112, 91-98, https://doi.org/10.1016/j.athoracsur.2020.08.030 (2021).
Roshanov, P. S. et al. Withholding versus Continuing Angiotensin-converting Enzyme Inhibitors or Angiotensin II Receptor Blockers before Noncardiac Surgery: An Analysis of the Vascular events In noncardiac Surgery patIents cOhort evaluatioN Prospective Cohort. Anesthesiology 126, 16-27, https://doi.org/10.1097/aln.0000000000001404 (2017).
Wu, Y. & Fang, Y. Stroke Prediction with Machine Learning Methods among Older Chinese. Int J Environ Res Public Health 17, https://doi.org/10.3390/ijerph17061828 (2020).
Wu, X., Hu, J. & Zhang, J. Machine learning-based model for predicting major adverse cardiovascular and cerebrovascular events in patients aged 65 years and older undergoing noncardiac surgery. BMC Geriatr 23, 819, https://doi.org/10.1186/s12877-023-04509-6 (2023).
Liew, B. X. W., Kovacs, F. M., Rügamer, D. & Royuela, A. Machine learning versus logistic regression for prognostic modelling in individuals with non-specific neck pain. Eur Spine J 31, 2082-2091, https://doi.org/10.1007/s00586-022-07188-w (2022).
Riley, R. D. et al. Minimum sample size for external validation of a clinical prediction model with a binary outcome. Stat Med 40, 4230-4251, https://doi.org/10.1002/sim.9025 (2021).

No competing interests reported.

Development and Validation of an Interpretable Risk Prediction Model for Perioperative Ischemic Stroke in Noncardiac, Nonvascular, and Nonneurosurgical Patients: A Retrospective Study

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Introduction

Materials and methods

Statistical analysis

Results

Discussion

Conclusion

Declarations

References

Additional Declarations

Status:

Version 1