The incidence of myocardial injury after non-cardiac surgery (MINS) is approximately 12-15% and is associated with an increased risk of 30-day mortality, 1-year mortality, and 2-year major vascular events. Using both traditional longitudinal analysis techniques and novel methods in machine learning, investigators will explore whether intraoperative and postoperative vital signs can enhance MINS surveillance by providing temporal prediction of MINS events.

The incidence of myocardial injury after non-cardiac surgery (MINS) is approximately 12-15% and is associated with an increased risk of 30-day mortality, 1-year mortality, and 2-year major vascular events. Since more than 90% of patients with MINS are asymptomatic, routine troponin monitoring is required for detection. The postoperative days 0, 1, and 2 accounts for approximately 40%, 40%, and 10% of MINS, respectively. Presently, the Canadian Cardiovascular Society (CCS) perioperative guidelines recommend patients identified to be at risk according to the Revised Cardiac Risk Index (RCRI) and BNP/NT-pro B-type Natriuretic Peptide (NT-pro BNP) receive daily postoperative troponin monitoring for three days to identify MINS events (5). European and American societies have similar recommendations for troponin monitoring to detect MINS.

Current risk stratification models have multiple limitations. Most importantly, they predict an elevated risk over the postoperative period but cannot pinpoint when MINS may happen in the postoperative course given the patients' changing condition. Moreover, prescription of troponin monitoring is not universal, infrequent and inconsistent troponin monitoring may lead to delayed detection and management, and the rise of troponin is delayed by 3-4 hours from the time of injury.

Retrospective cohort studies show association amongst intraoperative and postoperative derangements in vital signs and MINS. Vital signs are routinely available within the electronic medical record, and may serve as objective predictors (i.e. as opposed to free text and disease names that have higher risks of misclassification and errors).

Using both traditional longitudinal analysis techniques and novel methods in machine learning, investigators will explore whether intraoperative and postoperative vital signs can enhance MINS surveillance by providing temporal prediction of MINS events.

Objectives

1. To develop and internally validate a model that uses the duration and degree of intraoperative and postoperative hypotension to predict the daily maximum troponin level from postoperative days zero to two, in a high risk population where troponin monitoring was ordered.
2. To develop and internally validate a model that uses the duration and degree of intraoperative and postoperative hypotension to predict daily probability of MINS or death (binary outcome, according to the 2021 American Heart Association (AHA) definitions from postoperative days zero to two.
3. To evaluate how different definitions of hypotension affect the primary and secondary models above, and analyze the intraoperative and postoperative hypotension separately to determine whether intraoperative or postoperative hypotension alone are sufficient for prediction.
4. To explore whether other intraoperative and postoperative vital sign information (heart rate, oxygen saturation, and end-tidal carbon dioxide derangements at various definitions) add predictive value to the primary and secondary models above.
5. To use machine learning methods to perform exploratory analysis to determine 1) optimal methods for imputation for time series data; 2) visualization of time series data in the setting of prediction; 3) development and internal validation of machine learning models to use the time series data to predict troponin levels; and 4) to determine how many hours earlier than a binary MINS diagnosis were vital signs able to predict a MINS diagnosis (as determined by when in the time series prior to MINS diagnosis that the model achieved various thresholds of predicted probability).

Preoperative laboratory values are collected within 30 days before surgery. Preoperative vitals are collected within 24 hours before surgery. The total duration of subsequent data collection will be from the time of surgery up to postoperative day 3 (since the MINS protocol monitors for 3 days) or hospital discharge or death, whichever occurs first. Since 90% of MINS happen between postoperative days zero and two and the frequency of monitoring would likely decrease by postoperative day 3, investigators will model troponin from postoperative day 0 to postoperative day 2 for the primary analysis to balance utility of the temporal model and prediction data quality.

Inclusion Criteria:

• Patients who underwent inpatient noncardiac surgery, who had myocardial injury after non-cardiac surgery (MINS) protocol ordered for postoperative high sensitivity troponin monitoring based on the Canadian Cardiovascular Society (CCS) guidelines
• Patients who underwent inpatient non-cardiac surgery from January 2020 to June 2021.

Exclusion Criteria:

• Patients who did not have intraoperative and postoperative vital signs
• Patients who did not have any postoperative troponin measurements
• Patients who underwent repeat surgery
• Patients who were readmitted within the first 72 hours postoperatively
• Patients who had positive troponin on postoperative day 4

This study will include patients aged 45 years and older undergoing inpatient noncardiac surgery, who had MINS protocol ordered for postoperative high sensitivity troponin monitoring based on the CCS guidelines from January 2020 to June 2021