A Novel Scoring System Predicting Red Blood Cell Transfusion Requirements in Patients Undergoing Invasive Spine Surgery

Abstract

Background: Access to blood products is crucial for patient safety during the perioperative course. However, reduced donations and seasonally occurring blood shortages pose a significant challenge to the healthcare system, with surgeries being postponed. The German Blood Transfusion act requires that RBC packages become assigned to an individual patient, resulting in a significant reduction in the available blood products, further aggravating shortages. We aimed to develop a scoring system predicting transfusion probability in patients undergoing spine surgery to reduce assignment and, thus, increase the availability of blood products. Methods: The medical records of 252 patients who underwent spine surgery were evaluated and 18 potential predictors for RBC transfusion were tested to construct a logistic-regression-based predictive scoring system for blood transfusion in patients undergoing spine surgery. Results: The variables found to be the most important included the type of surgery, vertebral body replacement, number of stages, and pre-operative Hb concentration, indicating that surgical specification and the extent of the surgical procedure were more influential than the pre-existing patient condition and medication. Conclusions: Our model showed a good discrimination ability with an average AUC [min, max] of 0.87 [0.6, 0.97] and internal validation with a similar AUC of 0.84 [0.66, 0.97]. In summary, we developed a scoring system to forecast patients' perioperative transfusion needs when undergoing spine surgery using pre-operative predictors, potentially reducing the need for RBC allocation and, thus, resulting in an increased availability of this valuable resource.

Keywords: blood bank; blood shortage; storage.

Figure 1

Resampling approach for identifying the best-performing logistic model with respect to the AUC (training cohorts are denoted by $D_j, j=1, \dots , 100$; test cohorts are denoted by $D_j^{*}, j=1, \dots , 100$).

Figure 2

Nested resampling approach for evaluating the performance of the model selection algorithm (derivation cohorts are denoted by $D^{\left(k\right)}, k=1, \dots , 100$; validation cohorts are denoted by $D^{\left(k\right)*}, k=1, \dots , 100$).

Figure 3

Relative frequencies of score values in the complete cohort (n = 251) (green bars) and estimated probability of RBC transfusions (blue line). The dashed green line represents the observed frequency of RBC transfusions in the subgroups of patients reaching the respective score according to the scoring system.