HIT Event Early Detection System – 25-103

Distinguishing true heparin‑induced thrombocytopenia (HIT) from the expected postoperative platelet decline after cardiac surgery is challenging, producing delayed or missed diagnoses that increase thrombosis, morbidity and mortality [1]. The clinical problem is amplified by the large volume of cardiac surgery worldwide and the limited discrimination of existing HIT scoring systems in the post‑cardiac‑surgery setting.

Researchers at McMaster University have developed a time‑series anomaly detection system trained on postoperative platelet trajectories to flag patterns consistent with HIT. The method learns “normal” postoperative platelet behavior from non‑HIT data and raises per‑day HIT risk scores when observed platelet sequences produce anomalously large reconstruction or residual errors; thresholding is tuned to maximize F1 on held‑out splits from the TRUST registry.

Applications

• Real‑time clinical decision support for HIT surveillance in cardiac surgery wards and ICUs.
• Retrospective cohort surveillance and quality‑improvement analyses of postoperative thrombocytopenia.

Advantages

• Early automated detection tuned to post‑cardiac‑surgery platelet dynamics (reduces diagnostic delay).
• Higher detection performance (example: LSTM‑AE F1 ≈ 83% vs ARIMA F1 ≈ 66%).
• Uses minimal, EHR‑friendly inputs (daily platelet counts), easing integration and deployment.

References:

1. Pishko, A. M., & Cuker, A. (2017). Heparin-Induced Thrombocytopenia in Cardiac Surgery Patients. Seminars in thrombosis and hemostasis, 43(7), 691–698. https://doi.org/10.1055/s-0037-1602664

Image obtained from: https://www.istockphoto.com/photo/blood-clot-gm1329520940-413225209?searchscope=image%2Cfilm