Artificial intelligence Applications In Coagulation

by Donna Castellone, MS, MT (ASCP) SH • April 06, 2023

The interpretations below are provided by Donna Castellone, MS, MT (ASCP) SH for Aniara Diagnostica.


The first form of Artificial Intelligence (AI) was an automated checkers bot and was invented in England in 1951. Today it is used in everything from Alexa, to cars as well as in healthcare. The utilization of AI is increasing with constantly improving algorithms but what are the pros and cons of this technology? What applications can be adapted in coagulation?(1)

The benefits of AI is making lives easier. Humans can become distracted and make mistakes and perform in a limited time frame. AI algorithms can run 24/7 at the same standard it is programmed. It can also analyze large sets of data very fast, in minutes as opposed to manually taking weeks and can also pick up patterns and predictions for what might happen in the future.(1)

AI does have its limitations. It bases its decision on what has happened in the past and as a result it lacks creativity. Many people also feel it will reduce employment, but in many situations, they enhance what people are already doing allowing for workers to be more efficient in other tasks. AI is purely logical and leaves little room for nuance and emotion working with fixed rules for analysis. This limits ethics and morality into the algorithm, it is only as good as the parameters in which the creators set leading to potential bias.(1)

Utilization of AI in coagulation can be used to track multiple parameters throughout a treatment or diagnosis. This can identify patient-specific patterns that can assist in proper treatment or underlying causes. This will be important in such instances as in Point of Care (POC) testing when looking at the INR and with other technologies such as challenges posed by novel anticoagulant measurement and their impact on other coagulation tests.(2)



For warfarin to be clinically effective, most patients are recommended to remain within a narrow therapeutic range (International Normalized Ratio (INR) between 2.0 and 3.0). Warfarin also had dietary and drug interactions, and patients require monitoring which is a burden to many patients. A retrospective study of adults who had used a form of AI that included various machine learning (ML) to predict suboptimal anticoagulation control for warfarin in patients with atrial fibrillation (AF). This was defined as time in therapeutic range (TTR) <70% based on the International Normalized Ratio (INR). There were 35,479 patients eligible for inclusion. Various ML techniques were employed including support vector machines; random forests; stochastic gradient boosting; neural networks [NN]) and time-varying data (6-week intervals up to 30 weeks (long-short term memory [LSTM] NN)) were applied. The MI algorthims were useful in predicting patients who are at a risk of suboptimal control. Having prior INR measures helped in the predictive performance for identifying patients who may benefit from more frequent INR monitoring or switching to alternate therapies. Combining the INR with the information from other clinical characteristics can be used to assess risk and the patient having suboptimal INR control. Alcohol consumption has been linked to a 3-fold increase in poor INR control in this study. The largest driver of suboptimal INR was the use of anti-depressant medication resulting in a 30% more likely occurrence to have poor INR control. Using AI can look at complex interrelationships and be able to identify non-linear associations.(3)


A randomized parallel group study (n=28) evaluated the use of AI on mobile devices to measure medication adherence in stroke patients on anticoagulation. Direct oral anticoagulants do not require monitoring, therefore more emphasis is placed on patients to self-manage. DOAs have a short half-life making adherence even more important. There are also no routine coagulation testing that can demonstrate adherence making patients at a risk of stroke or bleeding.(4)

Patients were randomized to daily monitoring by AI (n=15) and compared to a control group (n=13) which had no monitoring. AI app data fell into 5 categories: (1) visual confirmation of ingestion, (2) self-reported dose via the AI app, (3) self-reported dose by clinic staff, (4) missed dose, and (5) dose taken in clinic. Adherence was measured by pill counts and plasma sampling. Staff also received automated text messages or emails if doses were missed or late or incorrect. Results showed that adherence based on the AI platform was 90.5%, with plasma drug concentrations being 100% in the AI group versus 50% in the control group. The outcome demonstrated the ability of using AI having the potential for patients to adhere to medication, in particular DOAs.(4)


A research team from Chulalongkorn University including faculty from medicine and engineering have combined forces and invented a tool to assess the chance of developing a stroke. The tool evaluates stroke risk caused by heart disease. It is called "AICute". This will be a tool to support care given to patients in community hospital and small hospitals in remote areas where there may be a lack of tools and cardiologists. This is a web application with AI.(5)

AICute is a web application with artificial intelligence (AI) to analyze and evaluate patients for the risk of stroke from heart disease. It enables doctors in the community and small hospitals to make decisions and send patients to heart examinations more rapidly, making the treatment of stroke more effective. To access AICute clinicians log in through the web application and fill in 2 sets of data including the patient symptoms and their partial history together with 30-32 brain scans. The data will be processed and summarized in a report of the likelihood of stroke caused by heart disease and if the patient should be referred to a cardiologist. The accuracy is between 92-94% according to the database of 40,000 high resolution X-ray images.(5)


Antiphospholipid syndrome is characterized by either the presence of lupus anticoagulant or antibodies against cardiolipin or β2-glycoprotein-1. It is very difficult to diagnosis and requires several tests that should be repeated over time to determine the presence of APS. There must be at least one clinical criteria such as the occurrence of thrombosis or pregnancy morbidity. The laboratory criteria are defined as anti-cardiolipin (aCL) antibodies, anti-β2-glycoprotein I (aβ2-GPI) antibodies, or lupus anticoagulant (LA) in plasma on two or more occasions. Further complicating this diagnosis are that there are several assays for this testing which have different sensitivity and levels of detection.(6)

A neural network which is a computing system inspired by the human brain to help distinguish between healthy subjects and patients that fit into a subject specific data in this case APS. This was developed based on results from the thrombin generation (TG) test that is known to differ in APS patients versus healthy subjects. This is based on the accelerated rate of prothrombin conversion which indicated a prothrombotic state in APS patients.(6)

A comparison of normal controls, APS patients and two diseased control groups, and those with autoimmune antibodies were compared. The results were varied and ranged from a positive predictive value of 62% in hospital controls to 91% in normal controls while the negative predictive value ranged from 86% in the thrombosis control group to 95% in the hospital controls. The sensitivity of the neural network was higher than 90% in all control groups. This neural network could accurately diagnosis APS in the validation cohort.(6)


Artificial intelligence uses a series of algorithms mimicking human intelligence. Applications of AI can be used for diagnostic procedures including forecasting events, prediction of risk for disease, monitoring anticoagulants. AI can aid in foreseeing adverse events minimizing bleeding and or clotting events. There are many applications in coagulation that would lead to better patient outcomes and enhance diagnosis and monitoring of anticoagulation.


  1. The Pros And Cons Of Artificial Intelligence. - Powering a Personal Wealth Movement, Dec 1, 2022,
  3. Jason Gordon a, Max Norman a, Michael Hurst a c, Thomas Mason a b, Carissa Dickerson a, Belinda Sandler c, KevinG. Pollock c, Usman Farooqui c, Lara Groves a, Carmen Tsang a, David Clifton d, Ameet Bakhai e, Nathan R. Hill c
  4. Using machine learning to predict anticoagulation control in atrial fibrillation: A UK Clinical Practice Research Datalink study Get rights and content
  5. Daniel L. Labovitz, Laura Shafner, Morayma Reyes Gil, Deepti Virmani and Adam Hanina, Using Artificial Intelligence to Reduce the Risk of Nonadherence in Patients on Anticoagulation Therapy, 6 Apr 2017, Stroke. 2017;48:1416–1419
  6. An assessment tool for ischemic stroke risk to reduce disability and death
  7. March 10, 2023, Chula’s AICute innovation., LABline,
  8. Romy M W de Laat-Kremers, Denis Wahl, Stéphane Zuily, Marisa Ninivaggi, Walid Chayouâ, Véronique Regnault, Jacek Musial, Philip G de Groot, Katrien M J Devreese, Bas de Laat, Deciphered coagulation profile to diagnose the antiphospholipid syndrome using artificial intelligence, Thromb Res 2021 Jul;203:142-151.