Approximately 20% of RBC units are used for surgeries. Currently, hospitals estimate the risk of perioperative transfusions using historical charts known as the maximum surgical blood ordering schedule (MSBOS) for specific procedures. A new machine learning model, named the Surgical Personalized Anticipation of Transfusion Hazard (S-PATH), has been developed to provide more accurate estimations of perioperative transfusion risks. S-PATH considers historical rates of transfusion for each surgery, as well as patient specific characteristics such as age, weight, sex, relevant comorbidities, and pre-operative laboratory values. In a recent retrospective cohort study at 45 US hospitals, S-PATH was validated using data from over 3.28 million surgical patients between 2020 and 2021 (median age, 57 years; 53% female; obstetric and nonoperative cases excluded). Notably, 1.5% of these patients required perioperative transfusions. While MSBOS recommended type and screen orders for a median of 51.6% of the patients at each hospital, S-PATH recommended type and screen orders for 32.5% of patients—a reduction of 17.9%–while maintaining a 96% sensitivity. This improvement enhances resource allocation and reduces health care costs. Overall, the receiver operating characteristics curve for S-PATH exceeded 0.91 in 75% of the hospitals studied; however, smaller hospitals exhibited greater variation than larger teaching hospitals. S-PATH has the potential to enhance the estimation of perioperative transfusion risks and contribute to cost reductions.
Reference:
Leave a Reply