Development of a machine learning algorithm model to predict intraoperative hypotension in elderly patients undergoing thoracic and abdominal surgeries

Comprehensive Summary

This article looks at the risk of intraoperative hypotension in elderly patients undergoing general anesthesia by looking at a secondary analysis of 1720 elderly patients. Hypotension during operations can lead to serious negative outcomes including myocardial injury, stroke risk, and mortality. Additionally, using mean arterial pressure (MAP), as a predictor, it was seen that levels below 80 leads to potential end organ damage. The aim of this study was to look at key identifying factors of patients soon to undergo general anesthesia and input these data points into machine learning programs to predict which patients are at the greatest risk of hypotension during these operations. Included in this study were patients aged 60-90 and hypotension was determined as a SBP below 80 mmHg or a 20% reduction from baseline for over 1 minutes. Of the data points analyzed (55 including age, sex, education years, cognitive scores, and co-morbidities), 16 key points were identified as being the most important which included data of the anesthesia protocol, co-morbidities present, intraoperative medications, preoperative MAP, and surgical site. By far the most important was the Charlson comorbidity index which had an importance of near 50, while the next closes was MAP at around a 30 importance. These data points were then entered into eight machine learning models, and the random forest model showed the most promise, achieving the most area under the curve compared to the rest. This model was then externally validated to ensure accuracy, and a SHAP test was used to generate an importance ranking plot which ranked the 16 key points again when looking at intraoperative hypotension. This was used to analyze the model’s interpretation of the data and to look at key levels within each data points where there was a disproportional jump.

Outcomes and Implications

If this machine learning model can be used to accurately predict which patients may be at an increased risk for intraoperative hypotension IOH), this information can then be used to alter the approach to anesthesia to minimize the risk of harm to the patient. It can be used to weight the risks and rewards of the surgery and potentially alter how the surgery is performed in order to minimize the time under anesthesia for a patient of high risk. This information can also be incorporated seamlessly into hospitals, as it can take existing data points from recent previous hospital encounters to form a risk assessment of the patient, and this can be discussed with the patient, both for pre-operative and post-operative prophylaxis of negative events associated with IOH such as preparing for an increase stroke risk after the surgery. It can also be used to have extra materials and personnel on standby in the case of an emergent event, such as blood if the blood pressure begins to drop too low. Overall, this machine learning model can be used to increase the overall safety of the patient during the operation in addition to post-operative while maximizing existing data points, allowing it to be incorporated into hospitals with ease.