A novel predictive model of intraoperative blood loss in patients undergoing elective lumbar surgery for degenerative pathologies

BACKGROUND CONTEXT: Intraoperative blood loss (IOBL) is unavoidable during surgery; however, high IOBL is associated with increased morbidity and increased risk for requiring allogenic blood transfusion, itself associated with poorer outcomes. PURPOSE: Here we sought to develop and validate a predictive calculator for IOBL that could be used by surgeons to estimate likely blood loss. STUDY DESIGN/SETTING: Retrospective cohort. PATIENT SAMPLE: Series of consecutive patients who underwent elective lumbar spine surgery for degenerative pathologies over a 27-month period at a single tertiary care center. OUTCOME MEASURES: Primary outcome was IOBL. Secondary outcome was the occurrence of “major intraoperative bleeding,” defined as IOBL exceeding 1 L. METHODS: Charts of included patients were reviewed for medical comorbidities, preoperative laboratory data, surgical plan, and anesthesia records. Univariate linear regressions were performed to find significant predictors of IOBL, which were then subjected to a multivariate analysis to identify the final model. Model training was performed using 70% of the included cohort and external validation was performed using 30% of the cohort. Results of the model were deployed as a freely available online calculator. RESULTS: We identified 1,281 patients who met inclusion/exclusion criteria. Mean age was 60±15 years, mean Charlson Comorbidity score was 1.1±1.6, and 51.8% were male. There were no significant differences between the training and validation cohorts with regard to any of the demographic variables or intraoperative variables; tranexamic acid use and surgical invasiveness were also similar in both cohorts. Multivariate analysis identified body mass index (βₙ=7.14; 95% confidence interval [3.15, 11.13]; p<.001), surgical invasiveness (βₙ=29.18; [24.62, 33.74]; p<.001), tranexamic acid use (βₙ=−0.093; [−0.171, −0.014]; p=.02), and surgical duration (βₙ=2.13; [1.75, 2.51]; p<.001) as significant predictors of IOBL. The model had an overall fit of r=0.693 in the validation cohort. Construction of a receiver-operating curve for predicting major IOBL showed a C-statistic of 0.895 within the validation cohort. CONCLUSION: Here we identify and validate a model for predicting IOBL in patients undergoing lumbar spine surgery. The model was a moderately strong predictor of absolute IOBL and was demonstrated to predict the occurrence of major IOBL with a high degree of accuracy. We propose it may have future utility when counseling patients about surgical morbidity and the probability of requiring transfusion.