AFFILIATIONS
1 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
2 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
3 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
4 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
5 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
6 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
7 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
8 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
9 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
10 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
11 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
12 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
13 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
14 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
15 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
16 Institute of Hematology and Blood Diseases Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Tianjin, China
ABSTRACT
Background: Response to eltrombopag is hard to predict in adults with primary immune thrombocytopenia (ITP). We aimed to develop and validate a machine learning (ML) model for predicting whether adult ITP patients receiving eltrombopag treatment would achieve an adequate response within 2 weeks of therapy start.
Methods: We conducted a retrospective analysis of 490 adults with ITP who received eltrombopag, a thrombopoietin receptor agonist, for the first time. 330 (67.3%) subjects had persistent or chronic primary ITP. 481 (98.2%) and 356 (72.7%) subjects had received ≥ 1 and ≥ 3 drugs targeting ITP at least 1 week before starting eltrombopag treatment. 412 subjects who initiated eltrombopag during January 2016 – December 2023 were designated the training cohort whilst the other 78 initiating eltrombopag during January 2024 – June 2025, the validation cohort. Adequate response to eltrombopag was defined as reaching a platelet count ≥ 100 × 10?/L within 2 weeks after starting eltrombopag therapy. 4 predictors, including anti-glycoprotein (GP) IIb/IIIa antibodies, the number of drugs given in previous treatment courses, the number of drugs given concurrently with eltrombopag (within 7 days of initiating eltrombopag), and maximum eltrombopag dosage during the first 7 days, were selected from 13 candidate variables for further analyses based on LASSO regression on the training cohort. 6 models were subsequently developed with the 4 selected predictor variables using logistic regression (LR), k-nearest neighbors, support vector machines, random forests, and extreme gradient boosting. Cross-validation and bootstrapping experiments within the training cohort indicated that the LR model had the best performance in predicting adequate response to eltrombopag. We then tested the LR model on the validation cohort.
Results: Adequate response within 2 weeks of initiating eltrombopag was highly correlated with sustained response off treatment (defined as a platelet count ≥ 30×10?/L maintained for at least 6 months after stoppage of medication targeting ITP) in our study cohort, with a concordance score of 0.70 (standard error = 0.06, 95-percent confidence interval [CI]: 0.58, 0.82). Therefore, ability to predict adequate response to eltrombopag is of considerable clinical value. Using the LR model to predict adequate response to eltrombopag, the area under the receiver operating characteristic (ROC) curve was 0.82 (95-percent CI, 0.72, 0.92) in the validation cohort. Presence of anti-GPIIb/IIIa antibodies (odds ratio [OR] = 3.79 [95-percent CI, 2.22, 6.46]), fewer prior drugs given (0.19 per drug [0.11, 0.31]), fewer drugs given concurrently with eltrombopag (0.54 per drug [0.43, 0.68]), and a lower maximum eltrombopag dose within 7 days (0.12 per 25mg/d [0.05, 0.31]) were associated with a higher likelihood of adequate response to eltrombopag. A nomogram and an online calculator (https://eltrombopagpredict.shinyapps.io/predict/) are made publicly accessible to facilitate verification in independent datasets.
Conclusions: We have developed a machine learning model and a nomogram for predicting adequate response to eltrombopag in adults with ITP. We have made our tools publicly accessible for other researchers to validate our findings.
DISCLOSURES
Xinmiao Qu: No relevant conflicts of interest to declare.