Transfusion Probability as a Novel Measure for Lab-Guided Medical Decision-Making

The clinical decision to transfuse is strongly influenced by laboratory results. Analysis of transfusion decision-making based on pre-transfusion laboratory results (e.g. pre-transfusion hemoglobin) is a common yet misleading approach to study lab-guided transfusion practice. We introduce "Transfusion Probability" as a novel method which overcomes many limitations of pre-transfusion lab result analyses. Under this approach, we estimate the probability of transfusion after results at a specific value (e.g. hemoglobin 7.4 g/dL) or in a range of values (e.g. 7.0-7.9 g/dL) using the proportion of tests followed by transfusion. We provide statistical methodology for causal inference on the effect of patient conditions and apply our method to a large multi-center dataset. Analyses using pre-transfusion and transfusion probability were compared using data from a large longitudinal cohort of hospitalized patients (N=525,032 patients). We found red blood cell transfusion probabilities of 76.2% in the 6.0-6.9 g/dL, 18.9% in the 7.0-7.9 g/dL, and 4.5% in the 8.0-8.9 g/dL hemoglobin range. After confounder adjustment, patients with gastrointestinal bleeding patients were more likely to be transfused across all ranges, with risk differences ranging from 6.6% in the 8.0-8.9 g/dL range to 13.8% in the 6.0-6.9 g/dL range. Pre-transfusion hemoglobin results showed minimal differences between gastrointestinal bleeding patients and other patients in unadjusted (0.00 g/dL) and adjusted analyses (-0.20 g/dL). In contrast to pre-transfusion result analysis, transfusion probability offers a nuanced account of transfusion practice and allows for natural comparisons between patient groups. Wider adoption of transfusion probability analysis may provide direct and actionable insights for clinical decision-making. KEY POINTS O_LIPre-transfusion lab results are a widely used method for studying lab-guided transfusion but subject to many limitations C_LIO_LITransfusion probability analysis is a novel and superior approach C_LI