PB0542 - Recommendation model to automatically assign immunosuppressants to low-dose immune tolerance induction individually in severe hemophilia A patients with inhibitors via causal inference

Monday, June 24, 2024

13:45 – 14:45 ICT

Room: Exhibition Hall

Presenting Author(s)

Zhengping Li, Medical Doctor in progress

M.D. in progress
zhengp_li@163.com
Beijing, Beijing, China (People's Republic)

Co-Author(s)

YX

Yanghao Xiao

Dr
State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China
Beijing, Beijing, China (People's Republic)
YT

Yongqiang Tang

associate professor
State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China
Beijing, Beijing, China (People's Republic)
Runhui Wu, PhD, MD

Professor
Department of Hematology, Beijing Children’s Hospital, Capital Medical University, National Center for Children’s Health, Beijing, 100045, China
Beijing, Beijing, China (People's Republic)

Background: Adding immunosuppressants (IS) in the low-dose immune tolerance induction (LD-ITI) individually was significant but lacked relative studies.

Aims: To fill the gap, we aim to develop a recommendation model that can help doctors to assign IS treatment to LD-ITI individually.

Methods: This was a single-center, retrospective study on Chinese severe hemophilia A (SHA) children with high-titer inhibitors undergoing LD-ITI with or without IS therapy for ≥24 months, with assessable ITI outcomes. Utilizing causal inference methods, a model that could recommend individually patients to add IS during LD-ITI treatment based on the clinical baseline information was developed.

Results: Total 195 SHA patients with high-titer inhibitors were included, with 145 (74.4%) patients achieving success. As table 1 shown, shorter interval-time (from inhibitor-diagnosis to ITI-start) (11.9 vs. 20.1 months), younger age at ITI-start (4.0 vs. 6.1 years), lower immediate pre-ITI inhibitor-titer (33.8 vs. 81.5 BU/mL), historical peak inhibitor-titer (60.5 vs. 110.0 BU/mL) and peak inhibitor-titer during ITI (55.7 vs. 146.4 BU/mL) were found in success group.
Among 195 patients receiving factual treatment recommended by doctors, 123 (63.1%, 123/195) patients received consistent recommendation generated by the model, including 76.4% (94/123) achieving success. Among 72 (36.9%, 72/195) patients gained inconsistent recommendations, 29.2% (21/72) patients failed to eradicate inhibitors.
Compared to all 195 patients, mean older age at ITI-start (8.8 vs. 4.5 years) and longer interval-time (25.5 vs. 14.0 months) was found in the 21 patients with inconsistent recommendation and failure outcome (Table 2).

Conclusion(s): Through the casual inference and machine learning. we developed a recommendation model that can help doctors to assign IS treatment to LD-ITI individually. This model displayed relatively good consistence with judgements from doctors. Additionally, the model implied the older age at ITI-start and longer interval-time may be potential predictors of adding IS in LD-ITI treatment.