Abstract: Background: Platelet transfusion efficacy affects clinical outcomes and clinical decision making, and there is a lack of specific efficacy prediction models for platelet transfusion in children with acute lymphoblastic leukemia (ALL). Objective: To develop and validate a predictive model for platelet corrected count increment (CCI) in pediatric ALL after platelet transfusion. Design: A retrospective cohort study. Methods: Hospitalized ALL children who had been transfused with platelets in the Department of Hematology and Oncology at Affiliated Children's Hospital of Chongqing Medical University from January 2022 to March 2023 for initial diagnosis and treatment were included, and were divided into the modeling group (before December 31, 2022 ) and the validation group (after December 31, 2022). The history data that might affect the efficacy of platelet transfusion were collected including gender, age, splenomegaly or hyper-splenism, hemorrhagic grade, degree of anemia, platelet storage days, medications, morphology, risk grouping, and treatment stage. Univariable COX regression and Lasso regression analysis were performed, followed by multivariable COX regression to determine the final predictors. The column-line graph prediction model was established built on the results. To detect the predictive efficacy, the model was then evaluated for its differentiation and calibration. Main outcome measures: Predictive efficacy of the CCI line graph model for platelet transfusion efficacy. Results: A total of 133 cases with 247 platelet transfusions were included in this study during the study period. There were a total of 198 transfusions in the modeling group and 49 transfusions in the validation group, and the baseline differences in the variables between the two groups were not statistically significant. Six factors, namely, gender, age, ALL risk grouping, pre-transfusion leukocyte count, splenomegaly or hypersplenism, and infection status, were finally incorporated to construct a column-line graph prediction model. When CCI was 11.4 (P25), 18.5 (P50), and 28.2 (P75), the area under curve (AUC) values were respectively 0.783, 0.695, and 0.654 in the modeling group, and 0.765, 0.714, and 0.580 in the validation group. The calibration curves showed that the model-predicted incidence rates and the actual incidence rates were basically the same in the modeling and validation group when CCI was 11.4, 18.5, and 28.2. Conclusion: A predictive model for platelets CCI after platelet transfusion in children with ALL was established. Good predictive efficacy for platelet transfusion efficacy in children with ALL was confirmed by a column-line graph predictive model using influencing factors of gender, age, pre-transfusion leukemia count, infection, ALL risk grouping, and splenomegaly or hypersplenism.

Key words: Pediatric acute lymphoblastic leukemia, Platelet transfusion, CCI, Predictive model