A multicenter analysis revisiting the PLASMIC score: performance limitations and the impact of age, MCV, and LDH

1. PLASMIC score performance declined with age; MCV lacked predictive value, while LDH was a strong predictor. 2. Replacing MCV with LDH modestly improved accuracy, highlighting structural limits of equally weighted PLASMIC variables. The PLASMIC score is widely used to guide early identification of immune thrombotic thrombocytopenic purpura (iTTP), but recent studies suggest decreased performance in different populations and low predictive yield of specific components. Refining rapid diagnosis of iTTP may improve outcomes. We conducted a multicenter retrospective analysis of patients who underwent ADAMTS13 testing. We validated the PLASMIC score, evaluated predictive performance of its individual components using multivariable logistic regression, and assessed age-stratified performance in two age groups (<60 and ≥60). We generated modified score variants by excluding low-yield components and/or incorporating lactate dehydrogenase (LDH). We included 482 patients (iTTP n=80, controls n=402). PLASMIC retained good performance overall (AUC 0.854), but sensitivity declined with older age (83% vs 62%). Mean corpuscular volume (MCV) was the only PLASMIC component that lacked independent predictive value (OR 1.1, p=0.68), while LDH was a strong independent predictor of iTTP diagnosis (OR 3.2, p<0.001). Excluding MCV (“PLASIC” variant) improved AUC (0.881) and sensitivity (91%), and adding LDH (“PLASIC+L”) yielded highest AUC (0.883). All score variants showed reduced performance in older adults. In the largest multicenter validation to date, the PLASMIC score remained a reliable tool, but demonstrated structural weaknesses: diminished performance in older adults, inclusion of a low-yield variable (MCV), and exclusion of a high-performing variable (LDH). The modest gains seen with PLASIC+L variant underscore the limitations of traditional scoring systems that assign equal weight to unequally informative components. Future iTTP models should incorporate high-yield variables and account for their individual predictive strength to enhance diagnostic precision.