Most blood transfusions are given in the operating room. Adoption of the Maximum Surgical Blood Ordering Schedule in the 1970s reduced the amount of blood unnecessarily cross-matched, but the national cross-match-to-transfusion ratio remains at approximately two-to-one. We tested the ability of a patient-specific blood ordering system (PSBOS) to more accurately predict potential operative transfusion. All adult patients who had blood cross-matched before surgery (February through June 1999) for elective operative procedures at the University of Michigan Hospital were identified. Complex surgeries were excluded. Surgeons estimated the expected blood loss for their surgeries, and the expected postoperative hematocrit was calculated using the patient's blood volume, the surgeon-defined expected blood loss, and preoperative hematocrit. Lowest tolerated hematocrit was set at 21% except in patients with coronary artery disease or who were ASA physical status III or more (28%). Sensitivity, specificity, positive predictive value, and negative predictive value of the PSBOS were calculated. Our analysis included 178 cases in which blood was cross-matched before surgery, representing 69 different surgeries and 42 surgeons. Only 16% of patients received an intraoperative transfusion. Of the 156 patients that PSBOS predicted would not require an operating room transfusion, 139 were not transfused. Of the 21 patients PSBOS predicted would be transfused, 11 were. The sensitivity of the algorithm as tested was 41%, the specificity 93%, the positive predictive value was 55%, and the negative predictive value was 89%. We conclude that PSBOS, which includes patient and surgeon variables in transfusion prediction, is more accurate than the Maximum Surgical Blood Ordering Schedule, which uses only surgical procedure.
Implications: Currently, many units of blood set aside for surgery are never required, resulting in extra work and expense for blood banks. A formula that included patient weight and hematocrit and typical surgery blood loss was used to predict who would require transfusions. We reduced the predicted number of patients who had blood set aside from 178 to 21.