Routine CBC–based machine learning for individualized correction of lipemia-induced hemoglobin bias

This study tested whether routine complete blood count values could estimate the hemoglobin overestimation caused by lipemia, using plasma-replacement hemoglobin as the reference, in 283 clinical specimens. A random forest model using routine CBC parameters predicted the relative bias well (out-of-fold R² of 0.848) and produced corrected hemoglobin values closely matching the reference (R² of 0.975). The authors note that accuracy was lower in the highest-bias cases, and that the model is not ready to replace established analytical correction methods without independent prospective validation.