Artificial intelligence prediction of therapy response in newly diagnosed adult severe primary immune thrombocytopenia.

Background: Newly diagnosed severe primary immune thrombocytopenia (ITP) with a platelet count of ≤10 × 10 Objectives: Using early kinetics of complete blood counts (CBC) to predict time interval to confirmed response in newly diagnosed severe primary ITP in adults. Methods: We used a training cohort of 85 adults with newly diagnosed severe primary ITP to develop an artificial intelligence model predicting confirmed response. The model used CBC values over the initial 5 days of therapy to classify subjects into 3 strata: high-confidence fast response, lower-confidence fast response, and slow/no response. We validated the model in an independent cohort of 59 subjects. Results: Concordance between the model-computed risk stratification and time to confirmed response is 0.78 (95% CI, 0.73, 0.82) and 0.78 (95% CI, 0.72, 0.83) in the training and validation cohorts, respectively. SHapley Additive exPlanations analyses indicated that the most informative covariates are age and kinetics of platelet count, mean red blood cell (RBC) volume, immature platelet fraction, standard deviation of RBC distribution width, mean RBC hemoglobin concentration, and immature reticulocyte percentage during days 3-5 of therapy. Conclusion: Our study should be treated as hypothesis-generating. Early dynamic patterns of CBC values might be used by artificial intelligence to predict the likelihood and speed of reaching confirmed response in adults with newly diagnosed severe primary ITP and might assist decision on timely therapy adjustment. Prospective testing and independent validation in non-Chinese cohorts and cohorts treated with different paradigms are needed.