Background/Case Studies: Physicians more likely transfuse red cells (RBC) in patients with hemoglobin (Hb) levels just below leading digit thresholds (e.g. 6.9 g/dL vs. 7.1 g/dL). While such differences in Hb are clinically negligible, this tendency enables a natural experiment for clinically similar patients to face different transfusion probabilities. We sought to study the leading digit effect for transfusion rates and associated clinical outcomes.
Study
Design/Methods: Retrospective analysis of the GEMINI database from December 1, 2016 to June 30, 2022 representing approximately 50% of patients hospitalized on a General Internal Medicine ward in the province of Ontario. Primary analysis compared the probability of 1) RBC transfusion within 24h, and 2) a composite outcome of hospital mortality or 30-day readmission between patients with a Hb level just below (6.8 or 6.9 g/dL) and just above the 7.0 g/dL threshold (7.0 or 7.1 g/dL). Secondary analyses examined mortality, post-transfusion Hb increment, and individual physician propensity for threshold-based transfusion. Analyses were adjusted for age, gender, and Charlson Comorbidity Index.
Results/Findings: A total of 528,543 hospitalizations were included across 24 hospitals, of which 52,292 (9.9%) involved RBC transfusion. Patients were attended by 1,886 unique physicians over 4,962,810 inpatient days. Primary analysis found that patients just below the 7.0 g/dL threshold (n = 2,265) had a higher rate of RBC transfusion within 24h (66.5% [CI: 64.6, 68.4]) compared to patients just above (38.6% [CI: 36.9, 40.3]). Unlike the sharp change observed for transfusion rates at 7.0 g/dL, the odds of hospital mortality or 30-day readmission did not statistically differ across this threshold in unadjusted (OR 0.92 [CI: 0.82, 1.03]) or adjusted analyses (OR 0.92 [ CI: 0.82, 1.04]; Figure 1). Among transfused patients, a pre-transfusion Hb just below threshold was associated with a higher post-transfusion increment (+0.12 g/dL [CI: 0.04, 0.19]) compared to just above threshold. Among physicians who encountered at least 100 Hb values in the range of interest (6.8-7.1 g/dL) (N=165), we found substantial variation in their probability of ordering transfusion above (37.3%, IQR 32.4-43.7%) and below the threshold (72.8%, IQR 64.2-77.7%). There were minimal differences in baseline patient characteristics at either side of Hb thresholds. Conclusions: In this large multi-center retrospective cohort, clinically negligible changes in hemoglobin levels across 7.0 g/dL were associated with substantial differences in the probability of RBC transfusion among hospitalized patients yet did not translate to differences in mortality or 30-day readmission. The findings might motivate research on cognitive biases in transfusion behavior and caution against overreliance on hemoglobin thresholds for red cell transfusion.
