Use of a Polygenic Risk Score Improves Prediction of Myocardial Injury after Non-cardiac Surgery

Abstract

Background

While postoperative myocardial injury remains a major driver of morbidity and mortality, the ability to accurately identify patients at risk remains limited despite decades of clinical research. The role of genetic information in predicting myocardial injury after noncardiac surgery (MINS) remains unknown and requires large scale electronic health record and genomic datasets.

Methods

In this retrospective observational study of adult patients undergoing non-cardiac surgery, we defined MINS as new troponin elevation within 30 days following surgery. To determine the incremental value of polygenic risk score for coronary artery disease, we added the score to three models of MINS risk: Revised Cardiac Risk Index (RCRI), a model comprised entirely of preoperative variables, and a model with combined preoperative plus intraoperative variables. We assessed performance without and with polygenic risk scores via area under the receiver operating characteristic curve and Net Reclassification Index.

Results

Among 90,053 procedures across 40,498 genotyped individuals, we observed 429 cases with MINS (0.5%). Polygenic risk score for coronary artery disease was independently associated with MINS for each multivariable model created (odds ratio (OR) =1.12, 95% Confidence Interval (CI)= 1.02 to 1.24, p=0.023 in the RCRI-based model, OR=1.19, 95% CI=1.07–1.31, p=0.001 in the preoperative model, and OR=1.17, 95% CI=1.06–1.30, p=0.003 in the preoperative plus intraoperative model). The addition of clinical risk factors improved model discrimination. When polygenic risk score was included with preoperative and preoperative plus intraoperative models, up to 3.6% of procedures were shifted into a new outcome classification.

Conclusions

The addition of a polygenic risk score does not significantly improve discrimination but remains independently associated with MINS and improves goodness of fit. As genetic analysis becomes more common, clinicians will have an opportunity to use polygenic risk to predict perioperative complications. Further studies are necessary to determine if polygenic risk scores can inform MINS surveillance.

Introduction

Myocardial injury after noncardiac surgery (MINS) occurs after 8–16% of noncardiac procedures,^1,2,3 and is strongly associated with postoperative mortality, even in the 80% of patients who never experience ischemic symptoms.^1,2 Existing metrics for identifying at-risk patients have been either over-simplified to assign risk in a non-physiologic manner⁴ or distributed across disparate electronic health record (EHR) data sources unable to be synthesized at the point of clinical need.⁵ The expansion of the EHR has dramatically increased data types and sources that can be integrated into risk assessment algorithms. Yet, even relatively complex predictive metrics, such as the American College of Surgeons Surgical Risk Calculator, are limited to 21 variables requiring manual assessment and data entry.⁵ Next-generation risk assessment tools will incorporate diverse data from pharmaceutical records, laboratory results, vital signs, billing diagnoses from previous hospital encounters, and genotype to assist perioperative physicians with planning and care.

The number of patients with genetic information available at the time of surgery has grown exponentially,⁶ yet existing perioperative risk assessments fail to incorporate any genetic information, despite evidence that it may explain a substantial portion of the overall risk.⁷ Genome Wide Association Studies (GWAS) revealed polymorphisms associated with increased susceptibility to cardiovascular events in the perioperative period,⁸ however, the clinical utility of these data have been limited due to weak signal intensity and variable population prevalence. Genome-wide polygenic risk scores (PRS) attempt to overcome this issue by calculating a single risk score using millions of genetic polymorphisms validated across hundreds of thousands of patients.^7,9 PRS scoring has been used to identify subsets of the population at increased risk of developing coronary artery disease (CAD),⁷ but these methods have not previously been studied within the context of predictive modelling for perioperative outcomes, despite their amenability for being deployed in an automated clinical support system.^10,11

Here, we report a novel approach to perioperative cardiac risk assessment. The purpose of this article is to determine the utility of adding polygenic risk scoring, preoperative and intraoperative data to risk models predicting postoperative MINS. Our primary hypothesis was that the addition of polygenic risk scoring would improve our ability to predict MINS. Our secondary hypothesis was that a more diverse set of preoperative and intraoperative variables would lead to improved MINS modeling compared to a validated benchmark, the Revised Cardiac Risk Index.⁴

Methods

Data Availability

Data were collected by the Michigan Genomics Initiative (MGI), a longitudinal biorepository within Michigan Medicine with genetic data linkable to medical phenotype and EHR information.¹² Participants were approached and enrolled prior to surgical procedures beginning in 2012. Using MGI data, polygenic risk scores for coronary artery disease (PRS_CAD) were generated following a previously published approach (http://www.broadcvdi.org/informational/data).⁷ Additional patient variables including demographic information, laboratory results, comorbidities, diagnostic codes, and intraoperative details were queried and collected from the electronic perioperative anesthesia database (Centricity®, General Electric Healthcare, Waukesha, WI) and EHR (Epic, Verona, WI). Because of the sensitive nature of some of the data collected for this study, requests to access the datasets from qualified researchers trained in human subject confidentiality protocols may be sent to the corresponding author.

All study procedures were approved by the Institutional Review Board (HUM00127909 and HUM00099605) and the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) reporting guidelines were followed.¹³ Informed patient consent was obtained with enrollment.

Full methods are available in the Supplemental Material.

Results

The study population had a median age of 55 years [Interquartile Range (IQR): 42–65]. Fifty-five percent were female, and 90% were white (self-reported race). Forty-eight percent of procedures were classified as American Society of Anesthesiologists (ASA) class 2 and 44% as ASA 3. Seventy-six percent of procedures were under general anesthesia, with a median case duration of 127 minutes [IQR: 76–205 minutes]. The most common classes of surgery were neurosurgical/otolaryngology (27%) and abdominal surgery (26%). Patients developed MINS following 429 (0.5%) of 90,053 procedures, across 40,498 genotyped participants. Troponin testing was performed following 3,112 of the included procedures (3.5%) with 2,683 assays not meeting criteria for MINS and 429 meeting criteria. MINS patients were older, more likely to be male, have a higher ASA physical status, and a higher frequency of most comorbidities. Intraoperatively, patients developed MINS more frequently following high risk surgery (RCRI definition),⁵ longer duration cases, transfusion, and higher vasoactive medication administration. Patient and procedural characteristics between MINS and non-MINS groups are shown in Table 1. Seventy percent of all troponin qualifying for a diagnosis of MINS in our study occurred within the first 4-days following surgery. Thirty-nine percent of MINS cases had a troponin value <0.3 ng/mL and 34% of MINS cases had a troponin ≥ 1.0 ng/mL. Full distributions for the timing of postoperative troponin assays and maximum troponin value for each MINS case are shown in Supplemental Figure 1.

Table 1.

Baseline Patient and Procedural Characteristics

	No MINS (Control)	MINS (Case)	t-test	χ2
	N=89,624 median [IQR] or n (%)	N=429 median [IQR] or n (%)
Age at Surgery (y)	55 [42 – 65]	65 [58 – 73]	<0.001
Female Sex	49,335 (55.1)	169 (39.4)		<0.001
WHO BMI Classification				0.066
Underweight	1,527 (1.7)	7 (1.7)
Normal	21,755 (24.7)	78 (18.5)
Overweight	27,535 (31.2)	146 (34.7)
Obese I	19,458 (22.1)	92 (21.9)
Obese II	10,184 (11.5)	59 (14.0)
Obese III	7,804 (8.8)	39 (9.3)
Race				0.653
Caucasian	80,483 (89.8)	382 (89.0)
African American	4,589 (5.1)	25 (5.8)
Asian/Pacific Islander	1,117 (1.3)	4 (0.9)
American Indian/Alaskan	301 (0.3)	3 (0.7)
Other	3,134 (3.5)	15 (3.5)
ASA Status				<0.001
ASA 1	5,674 (6.3)	0 (0.0)
ASA 2	42,684 (47.7)	51 (11.9)
ASA 3	39,032 (43.6)	298 (69.6)
ASA 4	2,182 (2.4)	77 (18.0)
ASA 5	15 (0.0)	2 (0.5)
Emergent Surgery	2,102 (2.4)	45 (10.5)		<0.001
Admission Type				<0.001
Outpatient	52,503 (58.6)	44 (10.3)
Admit	30,343 (33.9)	273 (63.6)
Inpatient	6,778 (7.6)	112 (26.1)
RCRI Risk Factors
High Risk Surgery	14,048 (15.7)	140 (32.6)		<0.001
Ischemic Heart Disease	1,764 (2.0)	112 (26.1)		<0.001
CHF	3,600 (4.0)	134 (31.2)		<0.001
Cerebrovascular Disease	4,485 (5.0)	32 (7.5)		0.020
Insulin for Diabetes	5,188 (5.8)	62 (14.5)		<0.001
Preop Cr > 2.0 mg/dL	1,892 (2.1)	43 (10.0)		<0.001
Preoperative Medications
Beta-blocker	19,875 (22.2)	209 (48.7)		<0.001
ACE-inhibitor	15,088 (16.8)	125 (29.1)		<0.001
ARB	7,252 (8.1)	69 (16.1)		<0.001
OCP	3,505 (7.1% of females)	5 (1.2% of females)		0.002
Laboratory:
WBC	7.3 [5.7 – 9.5]	7.5 [5.9 – 9.8]	0.198
Hct	40.1 [36.7 – 43.0]	38.8 [33.4 – 41.8]	<0.001
Plt	239.0 [195.0 – 288.0]	222.0 [171.0 – 270.0]	<0.001
Cr	0.9 [0.7 – 1.0]	1.0 [0.8 – 1.3]	<0.001
BNP	150.5 [52.0 – 340.0]	190.0 [83.0 – 545.0]	0.074
Albumin	4.2 [4.0 – 4.5]	4.1 [3.7 – 4.3]	<0.001
Comorbidities
Anemia (blood loss)	947 (1.1)	13 (3.0)		<0.001
Anemia (iron deficiency)	2,602 (2.9)	42 (9.8)		<0.001
Cardiac Arrhythmias	10,869 (12.1)	273 (63.6)		<0.001
Coagulopathy	2,437 (2.7)	70 (16.3)		<0.001
COPD	10,562 (11.8)	110 (25.6)		<0.001
Fluid/Electrolyte Disorders	6,011 (6.7)	183 (42.7)		<0.001
Hypertension	26,615 (29.7)	332 (77.4)		<0.001
Liver Disease	4,414 (4.9)	74 (17.3)		<0.001
Metastatic Cancer	7,302 (8.2)	62 (14.5)		<0.001
Peripheral Vascular Disorders	4,115 (4.6)	103 (24.0)		<0.001
Pulmonary Circulation Disorder	1,997 (2.2)	67 (15.6)		<0.001
Valvular Diseases	2,240 (2.5)	64 (14.9)		<0.001
Unexpected Weight Loss	3,340 (3.7)	54 (12.6)		<0.001
Procedural Class (CPT)				<0.001
Neurosurgical/ENT	24,234 (27.0)	85 (19.8)
Thoracic	11,312 (12.6)	85 (19.8)
Abdominal	23,598 (26.3)	153 (35.7)
Pelvic	11,067 (12.4)	15 (3.5)
Orthopedic/Extremities	15,108 (16.9)	47 (11.0)
Radiological Procedures	3,208 (3.6)	43 (10.0)
Obstetric	993 (1.1)	1 (0.2)
Burn Excision/Debridement	64 (0.1)	0 (0.0)
Other	38 (0.0)	0 (0.0)
Intraoperative Details
General Anesthetic	67,684 (75.5)	361 (84.2)		<0.001
Case Duration (min)	126 [76 – 204]	271 [169 – 402]		<0.001
Estimated Blood Loss (mL)	25 [6 – 100]	150 [30 – 550]		<0.001
Total Crystalloid (L)	1.0 [0.6 – 1.6]	1.9 [1.0 – 3.0]		<0.001
Urine Output (mL)	0 [0 – 0]	0 [0 – 250]		<0.001
Any pRBC Transfusion	1,140 (1.3)	67 (15.6)		<0.001
Any FFP Transfusion	365 (0.4)	31 (7.2)		<0.001
Dopamine Use	0 (0.0)	7 (1.6)		<0.001
Ephedrine Use	16,384 (18.3)	119 (27.7)		<0.001
Epinephrine Use	338 (0.4)	37 (8.6)		<0.001
Norepinephrine Use	275 (0.3)	25 (5.8)		<0.001
Phenylephrine Use	31,486 (35.1)	272 (63.4)		<0.001
Vasopressin Use	1,320 (1.5)	54 (12.6)		<0.001
Any MAP < 50 mmHg	11,727 (13.1)	163 (38.0)		<0.001

Data are presented as frequency (percentage) or as median [25th percentile - 75th percentile]

Subsequent multivariable models treated these rare events as continuous variables with both zero and non-zero values included.

ACE-inhibitor, Angiotensin-converting-enzyme inhibitor; ARB, Angiotensin II receptor blockers; OCP, Oral Contraceptive Pill; WBC, White Blood Cells; Hct, Hematocrit; Plt, Platelets; Cr, Creatinine; BNP, B-type Natriuretic Peptide; COPD, Chronic Obstructive Pulmonary Disease; CPT, Current Procedural Terminology; ENT, Ear, Nose, and Throat; pRBC, packed red blood cells; FFP, fresh frozen plasma; MAP, mean arterial pressure.

Characteristics of PRS_CAD

Patients developing MINS following surgery had a higher standard PRS_CAD score than the non-MINS group (median: 0.18, 95% CI: −0.54 to 0.81 versus median: −0.03, 95% CI: −0.64 to 0.65, p=0.004) (Figure 1A). The population with PRS_CAD > 1 (calculated risk greater than 1-standard deviation above the mean) (odds ratio (OR) =1.36: 95% CI=1.06 to 2.40, p=0.016), and PRS_CAD > 2 (calculated risk greater than 2-standard deviations above the mean) (OR=2.11, 95% CI=1.26 to 3.55, p=0.005) were both associated with MINS on univariate analysis (Figure 1B).

Figure.

Density Distributions and Risk for Developing MINS as a Function of Genome-wide polygenic risk scores for coronary artery disease (PRSCAD) A. Distribution of PRS_CAD in patients developing MINS compared to those without MINS. For MINS group, mean was 0.14 (SD=1.03) and median was 0.18 [95% CI: −0.54 to 0.81]. For non-MINS group mean was −0.01 (SD=0.99) and median was −0.03 [95% CI: −0.64 to 0.65], p=0.004. B. Risk for MINS according to PRS_CAD in MGI population. Patients with PRS_CAD > 1 standard deviation above the mean have a 1.36 fold greater risk of developing MINS [95% CI=1.06–2.40, p=0.016]. Patients with PRS_CAD > 2 standard deviations above the mean have a 2.11 fold greater risk of developing MINS [95% CI=1.26–3.55, p=0.005].

MINS, Myocardial Injury after Noncardiac Surgery; PRS_CAD, Polygenic Risk Score for Coronary Artery Disease; SD, Standard Deviation, CI, Confidence Interval; MGI, Michigan Genomics Initiative.

Multivariable Logistic Regression Models

Model 1 - PRS_CAD

After adjustment for age, sex, and race, PRS_CAD was associated with a significantly increased odds for developing MINS (adjusted OR (aOR) = 1.23, 95% CI=1.11 to 1.37, p<0.001), which signified that for every 1 point increase in PRS there was 1.23 times the likelihood of MINS. The area under the receiver operating curve (c-statistic*) for Model 1 was 0.720 ± 0.011. Full results of Model 1 can be seen in Supplemental Table 1.

Models 2 and 3 – RCRI Variables without and with PRS_CAD

On multivariable regression, high risk surgery, history of ischemic heart disease, history of congestive heart failure, insulin therapy for diabetes, and preoperative Cr > 2.0 mg/dL were significantly associated with increased likelihood of MINS. History of cerebrovascular disease was the only traditional RCRI variable that was not significant in Model 2. Area under the receiver operator characteristic curve (c-statistic) for the RCRI-based regression was 0.786 ± 0.013.

When we added PRS_CAD (Model 3), we found that higher PRS_CAD was associated with greater likelihood of MINS (adjusted OR (aOR) =1.12, 95% CI=1.02 to 1.24, p=0.023). While the addition of PRS_CAD did not improve model discrimination (c-statistic for Model 2 = 0.786 ± * Interpretation of the c-statistic: 0.5 or less for a poor model, over 0.7 for a good model, over 0.8 for a strong model, 1.0 for a perfect model 0.013, Model 3 = 0.793 ± 0.014, p= 0.314) (Supplemental Figure 2A), it did improve the model relative goodness of fit as assessed by both AIC (Model 2: 4766; Model 3: 4465) and BIC (Model 2: 4841; Model 3: 4550) (Table 2). At a threshold that optimizes both specificity and sensitivity, the addition of PRS_CAD to the model did not improve sensitivity (NRI_UP -- MINS = 0.00%) but did improve specificity (NRI_DOWN -- No MINS = 0.65%) (Table 2).¹⁴ The NRI values across the entire range of 1-specificity values are shown in the Supplemental Figure 2A. Full results of Models 2 and 3 can be seen in Supplemental Table 2.

Table 2.

Comparison of Model Fit and Performance

	RCRI		Preoperative		Preoperative + Intraoperative
	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
AUC ROC	0.786 +/− 0.013	0.793 +/− 0.014	0.910 +/− 0.006	0.912 +/− 0.006	0.918 +/− 0.006	0.921 +/− 0.006
AIC	4766	4465	4023	3755	3949	3682
BIC	4841	4550	4098	3838	4116	3857
	Specificity	Sensitivity	Specificity	Sensitivity	Specificity	Sensitivity
Optimal Threshold (Euclidean Distance)	0.797	0.679	0.822	0.856	0.847	0.847
NRI_up -- MINS		0.0000		0.0051		0.0077
NRI_up -- No MINS		0.0006		0.0091		0.0069
NRI_down -- MINS		0.0025		0.0128		0.0128
NRI down -- No MINS		0.0065		0.0121		0.0095
Maximal Value (NRI_up -- MINS)		0.1219		0.0332		0.0332
Specificity and Sensitivity at Max Value	0.719	0.731	0.987	0.246	0.976	0.395
Maximal Value (NRI_up -- No MINS)		0.2777		0.0194		0.0153
Specificity and Sensitivity at Max Value	0.712	0.731	0.241	0.998	0.298	0.997
Maximal Value (NRI_down -- MINS)		0.1318		0.0332		0.0357
Specificity and Sensitivity at Max Value	0.080	0.985	0.969	0.414	0.956	0.566
Maximal Value (NRI_down -- No MINS)		0.3580		0.0285		0.0242
Specificity and Sensitivity at Max Value	0.080	0.985	0.155	0.998	0.274	0.997

C-statistic, area under the receiver operating characteristic curve; RCRI, Revised Cardiac Risk Index; AIC, Akaike information criterion; BIC, Bayesian information criterion; MINS, Myocardial Injury after Noncardiac Surgery; NRI, Net Reclassification Index.

Models 4 and 5 – Pre-induction Model without and with PRS_CAD

Model 4 using non-genetic variables known prior to the induction of anesthesia had excellent discrimination (c-statistic = 0.910 ± 0.006) for MINS when including age, admission type (admit and inpatient versus outpatient reference), RCRI score, history of a cardiac arrhythmia, fluid or electrolyte disorder, and hypertension (Supplemental Table 2).

When we added PRS_CAD to the pre-induction model (Model 5), we found that higher PRS_CAD was associated with increased likelihood of MINS (aOR=1.19, 95% CI=1.07 to 1.31, p=0.001). While the addition of PRS_CAD did not significantly improve model discrimination (c-statistic for Model 4 = 0.910 ± 0.006, Model 5 = 0.912 ± 0.006, p=0.529), it was associated with better model relative goodness of fit (AIC for Model 4: 4023 vs. Model 5: 3755; BIC for Model 4: 4098 vs Model 5: 3838) (Table 2). At Youden’s Point, the addition of PRS_CAD led to 0.51% of MINS patients being reclassified up and 1.21% of non-MINS patients being reclassified down (Table 2). The maximum value for NRI_UP -- MINS is 3.32% and the maximum value for NRI_DOWN -- No MINS is 2.85% (Table 2). Full results of Models 4 and 5 can be seen in Supplemental Table 2 and NRI values across the entire range of 1-specificity values can be visualized in Supplemental Figure 2B.

Models 6 and 7 - Case Completion Model without and with PRS_CAD

In Model 6, the list of variables was expanded to include intraoperative details with the goal of demonstrating how this improves model discrimination. We found that longer case duration (hours) (aOR=1.13, 95% CI=1.06 to 1.13, p<0.001), red cell transfusion (units) (aOR=1.14, 95% CI=1.07 to 1.020, p<0.001), total crystalloid resuscitation (L) (aOR=1.11, 95% CI=1.01 to 1.21, p=0.025), and total epinephrine dose (100 mcg) (aOR=1.01, 95% CI=1.00 to 1.01, p=0.033) were associated with increased likelihood of MINS. Adding intraoperative data without PRS_CAD, improved model discrimination when compared to the preoperative model without PRS_CAD (Model 4 = 0.910 ± 0.006, Model 6 = 0.918 ± 0.006, p=0.027) (Supplemental Table 2).

When PRS_CAD was added to the preoperative plus intraoperative model (Model 7), we found that PRS_CAD was associated with increased likelihood of MINS (aOR=1.17, 95% CI=1.06 to 1.30, p=0.003). While, PRS_CAD did not improve model discrimination (c-statistic for Model 6 = 0.918 ± 0.006, Model 7 = 0.921 ± 0.006, p= 0.361) (Table 2), it was associated with increased model relative goodness of fit: AIC (Model 6: 3949; Model 7: 3682) and BIC (Model 6: 4116; Model 7: 3857) (Table 2). ROC curves for Models 6 and 7 are compared in Supplemental Figure 2B.

At Youden’s point, the addition of PRS_CAD led to 0.77% of MINS patients being reclassified up and 0.95% of non-MINS patients be reclassified down. The maximum value for NRI_UP -- MINS is 3.32% and the maximum value for NRI_DOWN -- No MINS is 2.42% (Table 2). Full results of Models 6 and 7 can be seen in Supplemental Table 2 and NRI values across the entire range of 1-specificity values can be visualized in Supplemental Figure 2C.

Additional Comparison to RCRI Model

In the baseline genetic risk population (patients with PRS_CAD between −1 and 1), we found patients with RCRI = 0 had a 0.2% risk of developing MINS; RCRI = 1, 0.8%; RCRI = 2, 2.7%; RCRI = 3, 6.4%, and RCRI = 4, 6.8% (Supplemental Table 3). Genetic susceptibility had a non-uniform impact at different RCRI classes. Notably, the high genetic risk patients with RCRI = 3, had greater odds of developing MINS than the baseline genetic risk with RCRI = 4 patients (8.4% versus 6.8%). Furthermore, the addition of genetic risk reclassified a greater number of MINS cases as the RCRI score increased from 1 through 4 (16%, 18%, 26%, and 29%, respectively - Supplemental Table 3). The odds of developing MINS and NRI_UP --MINS and NRI_DOWN --MINS at each RCRI and genetic risk classification can be visualized in Supplemental Figure 3.

Discussion

In a generalizable cohort of adult non-cardiac surgical patients at a single academic quaternary care center without surveillance troponin screening, MINS was detected in 0.5% of cases. Patients developing MINS following surgery had a higher standard PRS_CAD score than those that did not develop MINS. We also found that a more diverse set of preoperative and intraoperative variables improves modeling of MINS compared to a validated benchmark (RCRI), a model based on genetics alone, or a model created entirely of preoperative variables. Furthermore, we demonstrate that integration of genetic data with the EHR dramatically improves predictive value compared to what can be achieved by polygenic risk score alone (c-statistic = 0.921 ± 0.006 versus 0.720 ± 0.011, p<0.001) or a model created from RCRI variables (0.921 ± 0.006 versus 0.786 ± 0.013, p<0.001). We demonstrate that genetic information (PRS_CAD) is independently associated with MINS and is included with models selected for parsimony in RCRI, preoperative, and preoperative plus intraoperative classes according to AIC and BIC selection criteria when compared to models without PRS_CAD. This study demonstrates a novel application of polygenic risk scoring specific to the perioperative period, showing that breakthroughs in understanding the genetics of general medical coronary artery disease risk may also have utility in the acute care perioperative period.

We created a multivariable regression using the six variables included in the RCRI model as a benchmark for subsequent models.⁵ This validation was important as our methodology differed from RCRI in two important ways: (i) our data collection was automated from the EHR, as opposed to a structured evaluation by an anesthesiologist and (ii) our outcome was MINS as opposed to major cardiac complications including: myocardial infarction, pulmonary edema, ventricular fibrillation, primary cardiac arrest, and complete heart block.⁵ The rate of MINS at each RCRI score in our population (0.2%, 0.7%, 2.8%, 5.9%, and 6.9%) compares to the rate of cardiac complications in the published validation of the RCRI (0.4%; 0.9%, 7%, and 11% at RCRI ≥ 3).⁴ Additionally, we found that at low RCRI risk classes (0 and 1), the genetic influence is smaller than that of an additional RCRI risk factor. However, at high RCRI risk classes (3 and 4), genetic factors appear to play a larger role. Notably, a patient with a RCRI=3 and a high genetic risk has a 8.4% odds of developing MINS while a patient with a RCRI=4 and baseline genetic risk only has a 6.8% risk. Furthermore, a patient with a RCRI=3 and low genetic risk and a patient with a RCRI=2 and baseline genetic risk (PRS_CAD between −1 and 1) both have a 2.7% risk of developing MINS. At RCRI class 3, genetic risk (set to a threshold of 1 standard deviation above or below the mean) essentially adds or subtracts a RCRI point. Additionally, consideration of genetic risk would correctly identify between 16–29% of MINS cases that would not be detected using the traditional RCRI metric.

In this study, we capture a wider number and variety of data than any existing assessment algorithms.^4,5,15 We quantified the individual and combined contributions of our three major data classes (genetic, preoperative, and intraoperative variables). Our preoperative model outperforms a traditional metric (Model 4 versus 2). Intraoperative data further improve model discrimination (Model 6 versus 4). The addition of genetic information fails to improve discrimination in models built from RCRI variables alone (Model 3 versus 2), preoperative variables alone (Model 5 versus 4), or a combination of preoperative plus intraoperative variables (Model 7 versus 6). The independent associations between PRS_CAD and MINS, as well as improvement in discrimination with the addition of intraoperative variables (such as case duration and transfusion) and MINS reinforces the role these data types have in predicting MINS. Our models include a variable: History of Ischemic Heart Disease, which is curated from EHR-based diagnostic codes for myocardial infarction, ischemic heart disease, as well as, billing codes for coronary interventions and surgeries. This covariate, defined with a more stringent definition than coronary artery disease, was found to be a stronger predictor of MINS than PRS_CAD (aOR=7.83 compared to 1.12 in Model 3), however, the effect size of a binary variable (History of Ischemic Heart Disease) cannot be directly compared against that of a continuous variable (PRS_CAD). PRS_CAD and history of ischemic heart disease are both independent risk factors for MINS, even with adjusting for the effect of the other. This independent effect may be partially explained by patients who have not had a prior ischemic event or are unaware of their coronary artery disease.

While our composite model which integrates EHR (preoperative plus intraoperative) and genetic scores (Model 7) improves prediction compared to a model created from RCRI variables (Model 2) (0.921 ± 0.006 versus 0.786 ± 0.013, p<0.001) or a model based on genetics alone (Model 1) (0.921 ± 0.006 versus 0.720 ± 0.011, p<0.001), significant complexity must be added for relatively modest improvement in discrimination. Providers may determine that a c-statistic of 0.910 ± 0.006 for Model 4 constructed entirely of preoperative variables, without genetic information, provides adequate assessment of cardiac risk without the complexity of adding intraoperative variables (c-statistic = 0.918 ± 0.006) or genetic variables (c-statistic=0.912 ± 0.006). Additionally, the complexity of each model makes rough estimation of risk difficult to conceptualize, limiting utility without an integrated system. The ability of PRS_CAD to predict MINS in the absence of all but basic demographic factors (c-statistic=0.720 ± 0.011), highlights the role genotyping can play in perioperative risk assessment. Identification of genes associated with MINS may also lead to medicines or anesthetic management targeting those genes to lessen the risk of MINS.

At a threshold that optimizes both specificity and sensitivity, 0.77% of all MINS cases would be identified in Model 7 that were not identified in Model 6. If the clinician were to select a threshold with higher specificity and lower sensitivity, as many as 3.32% new MINS patients and 2.42% of non-MINS patients would be correctly identified. This means that 30 patients would need to undergo genetic testing and scoring to identify a single new MINS case (number needed to treat = 30).

We detected a substantially lower rate of MINS in our population when compared to The Vascular events in Noncardiac Surgery Patients Cohort Evaluation (VISION) which found 8% of prospectively screened patients had a troponin elevation within the first 3 days following surgery.³ In our population, troponin levels were obtained based on clinical practice, while VISION screened all high risk patients. Further study is needed to determine the efficacy of PRS scoring, and using pre- and intraoperative data when moderate and high risk patients are routinely screened with troponin levels. Although myocardial infarction confirmed by electrocardiography or wall motion imaging is a clinically relevant outcome to explore in further studies, we selected MINS because troponin values can be easily obtained, and troponin elevation predicts mortality independently of other ischemic features included in the Universal Definition of Myocardial Infarction.^16–18 We expect that adding PRS to predictive models could identify high risk patients who experience MINS in the absence of ischemic features, increasing the detection rate. Improved identification and prospective troponin screening for patients at high risk for MINS could lead to improved classification of cases and controls, potentially improving the predictive power of the model and further highlighting the clinical utility of our assessment methodology.

Limitations

This study has multiple limitations. Although PRS_CAD for the MGI population were generated using a previously validated methodology, a major limitation of our study is that the application of these PRS_CAD scores to predict risk for MINS was performed in a single center without external validation. As a retrospective analysis, we cannot determine if prospective identification of high-risk patients leads to clinically meaningful alterations in patient care. Establishing a more granular, detailed, and precise phenotype would potentially lead to stronger genetic associations and more meaningful clinical impact. Furthermore, the pathophysiology of MINS is more varied than that of our genetic proxy, coronary artery disease. Despite evidence that Type 2 myocardial infarction is a more frequent cause of perioperative MI than non-operative MI, PRS_CAD score was independently associated with MINS in all models.^19,20 While we focused our initial work on coronary artery disease, because the genetic associations are better understood and more completely characterized than MINS, further study is needed to develop and validate a PRS specifically for MINS and compare this to the PRS_CAD.

Another limitation is that we cannot determine whether associated intraoperative variables (such as transfusion or total epinephrine dose) trigger subsequent myocardial injury or are secondary to myocardial injury that has already occurred; however, the strength of the association suggests utility for incorporation within a real-time clinical support system.^10,11 As troponin elevation following surgery predicts mortality even in the majority of patients who remain asymptomatic;^17,18 correctly identifying patients at high risk for postoperative myocardial injury may enable diagnosis, treatment and interventions that decrease mortality. Additionally, the EHR data for this retrospective study were collected for the purposes of clinical care and not for research, therefore, misclassification of EHR comorbidities is possible. Another weakness of our study is that troponin measurements were collected based upon clinical need, at non-fixed time points. In comparison to the VISION Study, which found 87% of MINS cases occurred within the first 2-days after surgery, 47% of MINS cases in our population occurred within 2-days of surgery and 70% occurred within 4-days following surgery (Supplemental Figure 1).³ Other limitations to our study are the small number of MINS cases (429) and a purely laboratory-based diagnosis, without clinician adjudication of ischemic etiology or exclusion of non-ischemic conditions, such as pulmonary embolism, sepsis, or cardioversion.

Conclusion

As genetic testing becomes more common the ability to use polygenic risk scores to predict perioperative complications will become more feasible. We found that the addition of polygenic risk score incrementally improves the ability to predict MINS. We also found that a more diverse set of preoperative and intraoperative variables improves MINS modeling compared to a validated benchmark, the Revised Cardiac Risk Index⁴ or a model comprised entirely of preoperative variables. These data may support the incremental value of incorporating PRS_CAD into a perioperative MINS risk stratification model when already available through prior genetic testing; however, they do not support obtaining genetic information for the sole purpose of risk stratification. This work could inform the design of future prospective trials identifying MINS in asymptomatic patients and preventing MINS through modifications in intraoperative management.

Nonstandard Abbreviations and Acronyms

MINS

Noncardiac surgery

RCRI

Revised Cardiac Risk Index

PRS

Polygenic Risk Score

HER

Electronic Health Record

GWAS

Genome Wide Association Studies

CAD

Coronary Artery Disease

MGI

Michigan Genomics Initiative

TRIPOD

Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis

IQR

Interquartile Range

ASA

American Society of Anesthesiologists

OR

Odds Ratio

aOR

Adjusted Odds Ratio