tetano
Editor, Senior Moderator
Semin Thromb Hemost
. 2026 Mar 23.
doi: 10.1055/a-2832-1944. Online ahead of print.
Artificial Intelligence Applications in COVID-19-Associated Coagulopathy: Lessons Learned
Gerard Gurumurthy[SUP] 1 2 [/SUP], Filip Kisiel[SUP] 3 [/SUP], Lianna Reynolds[SUP] 4 [/SUP], Jecko Thachil[SUP] 4 [/SUP]
Affiliations
Coronavirus disease 2019 (COVID-19)-associated coagulopathy (CAC) is a thromboinflammatory syndrome marked by endothelial injury, micro- and macrovascular thrombosis, and, in a minority, late consumptive bleeding. D-dimer elevation was found to be one of the most consistent laboratory abnormalities and an established marker of severity. In this review, we explore how artificial intelligence methods were applied during the pandemic to predict thrombotic and mortality outcomes and to clarify mechanisms. Across studies, a consistent lesson was that D-dimer is most informative even when embedded within multivariable and time-aware models. Mechanistic machine learning (ML) analyses converged on an IL-6-centered immunothrombotic network, which associated cytokine signaling with complement activation. They highlighted how integrated proteomic and coagulation phenotyping can identify potentially actionable pathways. ML also quantified residual thrombotic risk despite standard thromboprophylaxis, suggesting that a high-risk minority may be detectable for closer monitoring or trial enrolment. Most CAC models, however, remain retrospective with limited external validation. They are also vulnerable to D-dimer assay heterogeneity and may drift across variants and vaccination eras. Future progress will depend on prospectively validated models that can stratify treatable subgroups and guide risk-adapted anticoagulation and immunomodulatory strategies beyond acute COVID-19.
. 2026 Mar 23.
doi: 10.1055/a-2832-1944. Online ahead of print.
Artificial Intelligence Applications in COVID-19-Associated Coagulopathy: Lessons Learned
Gerard Gurumurthy[SUP] 1 2 [/SUP], Filip Kisiel[SUP] 3 [/SUP], Lianna Reynolds[SUP] 4 [/SUP], Jecko Thachil[SUP] 4 [/SUP]
Affiliations
- PMID: 41871611
- DOI: 10.1055/a-2832-1944
Coronavirus disease 2019 (COVID-19)-associated coagulopathy (CAC) is a thromboinflammatory syndrome marked by endothelial injury, micro- and macrovascular thrombosis, and, in a minority, late consumptive bleeding. D-dimer elevation was found to be one of the most consistent laboratory abnormalities and an established marker of severity. In this review, we explore how artificial intelligence methods were applied during the pandemic to predict thrombotic and mortality outcomes and to clarify mechanisms. Across studies, a consistent lesson was that D-dimer is most informative even when embedded within multivariable and time-aware models. Mechanistic machine learning (ML) analyses converged on an IL-6-centered immunothrombotic network, which associated cytokine signaling with complement activation. They highlighted how integrated proteomic and coagulation phenotyping can identify potentially actionable pathways. ML also quantified residual thrombotic risk despite standard thromboprophylaxis, suggesting that a high-risk minority may be detectable for closer monitoring or trial enrolment. Most CAC models, however, remain retrospective with limited external validation. They are also vulnerable to D-dimer assay heterogeneity and may drift across variants and vaccination eras. Future progress will depend on prospectively validated models that can stratify treatable subgroups and guide risk-adapted anticoagulation and immunomodulatory strategies beyond acute COVID-19.