(P-BC-6) A Web Application for Probabilistic Assessment of Donors’ Hemoglobin Level Accounting for Measurement Error

Background/Case Studies: On-site donor eligibility is often determined using fingerstick hemoglobin (Hb) measurement, which is subject to measurement error. Many blood collectors perform a second measurement in donors who fail on the initial test and accept donations if the second measure exceeds a sex-specific threshold. Understanding the probability that a donor’s true Hb exceeds the threshold while accounting for measurement variability could inform how blood donation centers use fingerstick Hb to determine donors’ eligibility.

Study

Design/Methods: We created a Bayesian hierarchical model to estimate donors’ underlying Hb distribution based on one or two imperfect fingerstick Hb measures and created a web application to estimate the probability that a donors’ true Hb exceeds a threshold based on one or two Hb measurements. We used Hb measurements of 20,000 donor visits to a large US blood provider between 2017 to 2022, including 900 visits with two Hb measurements recorded. We used paired measurements performed on the same day to isolate measurement variability from intra-individual variation. Using importance sampling, our model estimates the posterior distribution of a donor’s true Hb. We used stan (version 2.36) to develop the Bayesian model and R Shiny (version 4.4) to create the web application.

Results/Findings: The application allows users to enter the sex of the donor, the sex-specific Hb threshold, the on-site fingerstick Hb measurement, and, if available, a second Hb measurement. It then estimates the likelihood of the underlying true Hb values using importance sampling. The likelihood and the posterior distribution of the donor’s true distribution are presented visually as density plots, and the probability that the true Hb exceeds the threshold is displayed. Warning messages are displayed when the estimates are potentially unreliable, e.g., when a Hb measure is an outlier. Figure illustrates three scenarios in which the threshold Hb level is 13 g/dL and the first Hb measurement is 12.8 g/dL. In the first scenario (A), the app estimates a 47.3% probability that the true Hb exceeds the threshold. When a second measurement of 12.4 g/dL is added, the estimated probability is updated to 17.4% (scenario B), and a second measurement of 13.2 g/dL updated the probability to 59.1% (scenario C).

Conclusions: The web application provides a convenient interface for understanding the role of measurement error in on-site donor Hb estimation and estimating the likelihood that donors’ Hb exceeds an eligibility threshold. The app could be used to inform decisions on collecting a second fingerstick Hb and deferring donors.