A Clinically Applicable Prediction Model for the Risk of Transfusion in Women Undergoing Myomectomy

Abstract

We sought to identify the variables independently associated with intra/postoperative blood transfusion at the time of myomectomy. We further hoped to develop an accurate prediction model using preoperative variables to categorize an individual's risk of blood transfusion during myomectomy. Case-control study. Not applicable to this study, which used the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database. Women who underwent an open/abdominal or laparoscopic (robotic or conventional) myomectomy between 2014 and 2017 at participating ACS-NSQIP sites. The primary dependent variable was occurrence of intra/postoperative bleeding requiring blood transfusion. Patient demographics, clinical characteristics, preoperative comorbidities, intraoperative variables, and additional 30-day postoperative outcomes were compared at the bivariable level. For the prediction-model development, only variables that can be reasonably known before surgery were included. Variables associated with intra/postoperative bleeding were entered into 2 separate multivariable logistic regression models. Validation of our prediction model was performed internally using 250 bootstrapped iterations of 50% subsamples drawn from the overall population of myomectomy cases from the ACS-NSQIP database. We identified 6387 myomectomies performed during the defined study period. The most common race in our population was black/African American (45.7%), and most of the patients (57.5%) received an open/abdominal route of myomectomy. A total of 623 patients who underwent myomectomy (9.8%) experienced intraoperative/postoperative bleeding with a need for blood transfusion. At the bivariable level, we identified several variables independently associated with the need for blood transfusion at the time of myomectomy. In using only those variables that can be reasonably known before surgery to develop our prediction model, additional multivariable logistic regression elucidated black race, need for preoperative blood transfusion, planned abdominal/open route of surgery, and preoperative hematocrit value as independently associated with blood transfusion. We identified a number of perioperative variables associated with intraoperative or postoperative bleeding requiring blood transfusion at the time of myomectomy. We subsequently created a model that accurately predicts individual bleeding risk from myomectomy, using variables that are reasonably apparent preoperatively. Making this prediction model clinically available to gynecologic surgeons will serve to improve the care of women undergoing myomectomy.