PREDICTING MINS WITH FRAILTY AND BIOMARKERS IN GERIATRIC SURGERY

Overview

The primary objective of this study is to develop and validate a machine learning model that integrates preoperative clinical data, biomarkers, and modified frailty indices (mFI-5) to accurately predict myocardial injury after non-cardiac surgery (MINS) in geriatric patients ($\\ge$65 years) undergoing major orthopedic surgery and requiring postoperative intensive care. The research aims to compare the predictive performance of advanced algorithms, such as XGBoost and Random Forest, against traditional clinical risk scores like the Revised Cardiac Risk Index (RCRI), while specifically evaluating the impact of frailty on the model's area under the curve (AUC). Furthermore, by identifying the most critical preoperative predictors, this study seeks to establish an objective clinical decision support mechanism to guide clinicians in the early risk stratification of high-risk geriatric patients.

Myocardial injury after non-cardiac surgery (MINS) is defined as a troponin elevation occurring within the first 30 days following a surgical intervention, presumed to be caused by myocardial ischemia. Unlike the traditional diagnosis of myocardial infarction, MINS follows a "silent" course in more than 90% of cases, without ischemic symptoms or ECG changes. However, this silent progression is misleading; the 30-day postoperative mortality risk for patients who develop MINS is approximately 10 times higher than for those who do not. The geriatric orthopedic population, in particular, is in the highest risk group for this complication due to comorbidities and reduced physiological reserve. Currently, tools used in perioperative risk assessment, such as the Revised Cardiac Risk Index (RCRI) or ACS-NSQIP, focus primarily on chronic organ failures and remain insufficient in reflecting the dynamic physiological state of the geriatric patient. The low predictive success (AUC 0.54-0.62) of these scoring systems in the geriatric surgical group proves that clinicians require more precise tools for risk management.The Revised Cardiac Risk Index (RCRI), also known in the literature as the 'Lee Index,' is a widely used scoring system to predict perioperative major adverse cardiac events based on six clinical variables: high-risk surgery type, history of ischemic heart disease, congestive heart failure, history of cerebrovascular disease, preoperative insulin use, and a serum creatinine level above 2 mg/dL. However, RCRI focuses largely on the patient's existing chronic diagnoses; it does not account for the biological reserve loss that develops with aging, the depth of anemia, and specifically, the acute inflammatory response and fluid-electrolyte shifts triggered by orthopedic surgery. This situation significantly limits the sensitivity of RCRI in detecting silent myocardial injury (MINS) in the geriatric population. Given the high surgical urgency and stress in geriatric orthopedic patients, the early prediction of cardiovascular events has become a vital necessity.A review of the existing literature reveals that MINS prediction has focused either solely on clinical risk scores or on individual biomarkers (hs-cTnT, NT-proBNP). However, the concept of frailty, although it indicates the patient's biological reserve independent of chronological age, has not been sufficiently integrated into perioperative risk models. The combined effect of the "objective biological stress" data provided by biomarkers and the "physiological resilience" data provided by frailty indices has not yet been comprehensively modeled, specifically for orthopedic geriatrics. Traditional statistical methods struggle to capture the complex and non-linear relationships between these multidimensional data. There is a lack of a preoperative model in the literature where these variables are synthesized with machine learning algorithms.The primary objective of this study is to develop and validate a machine learning model that accurately predicts myocardial injury (MINS) following surgery in geriatric patients ($\\ge$65 years) undergoing major orthopedic surgery and followed in the postoperative intensive care unit, by integrating only preoperative clinical data, biomarkers, and modified frailty indices. In addition to the primary aim of the research, the study intends to: compare the predictive performance of advanced machine learning models (XGBoost, Random Forest) with traditional clinical risk scores (Revised Cardiac Risk Index) used widely in the literature; reveal the impact of adding validated frailty indices (mFI-5) to patients' existing comorbidities on the model's predictive power (AUC); rank the preoperative variables with the highest predictive value in determining MINS risk in geriatric orthopedic patients; and provide a risk classification based on objective data to guide clinicians in the preoperative identification of high-risk patients.