A thrombin-driven neural net diagnoses the antiphospholipid syndrome without the need for interruption of anticoagulation

ABSTRACT

Thrombosis is an important manifestation of the antiphospholipid syndrome (APS). The thrombin generation (TG) test is a global hemostasis assay, and increased TG is associated with thrombosis. APS is currently diagnosed based on clinical and laboratory criteria, the latter defined as anti-cardiolipin, anti-β2-glycoprotein I antibodies, or lupus anticoagulant (LA). APS testing is often performed after a thrombotic episode and subsequent administration of anticoagulation, which might hamper the interpretation of clotting assays used for LA testing. We set out to develop an artificial neural network (NN) that can diagnose APS in patients who underwent vitamin K antagonist (VKA) treatment, based on TG test results. Five NNs were trained to diagnose APS in 48 VKA-treated patients with APS and 64 VKA-treated controls, using TG and thrombin dynamics parameters as inputs. The 2 best-performing NNs were selected (accuracy, 96%; sensitivity, 96%-98%; and specificity, 95%-97%) and further validated in an independent cohort of VKA-anticoagulated patients with APS (n = 33) and controls (n = 62). Independent clinical validation favored 1 of the 2 selected NNs, with a sensitivity of 88% and a specificity of 94% for the diagnosis of APS. In conclusion, the combined use of TG and NN methodology allowed for us to develop an NN that diagnoses APS with an accuracy of 92% in individuals with VKA anticoagulation (n = 95). After further clinical validation, the NN could serve as a screening and diagnostic tool for patients with thrombosis, especially because there is no need to interrupt anticoagulant therapy.

Tailoring the effect of antithrombin-targeting therapy in haemophilia A using in silico thrombin generation

Factor (F) VIII deficiency causes bleeding in haemophilia A patients because of the reduced formation of procoagulant enzyme thrombin, which is needed to make the blood clot. We measured the dynamics of coagulation in haemophilia A patients by measuring thrombin generation (TG). Additionally, we quantified the procoagulant process of prothrombin conversion and anticoagulant process of thrombin inhibitor complex formation. In haemophilia A, prothrombin conversion is severely reduced, causing TG to be low. Nevertheless, the thrombin inactivation capacity of these patients is comparable to that in healthy subjects, leading to a severe imbalance between procoagulant and anticoagulant processes and a subsequent increased bleeding risk. A novel therapy in haemophilia A is the targeting of anticoagulant pathway, e.g. thrombin inhibitor antithrombin (AT), to restore the haemostatic balance. We simulated the effect of AT reduction on TG in silico. Lowering AT levels restored TG dose-dependently and an AT reduction of 90-95% led to almost normal TG in most patients . However, the variation in response to AT reduction was large between patients, indicating that this approach should be tailored to each individual patients. Ideally, TG and thrombin dynamics simulation could in the future contribute to the management of patients undergoing AT targeting therapy.

Deciphered coagulation profile to diagnose the antiphospholipid syndrome using artificial intelligence

The antiphospholipid syndrome (APS) is diagnosed by the presence of lupus anticoagulant and/or antibodies against cardiolipin or β₂-glycoprotein-1 and the occurrence of thrombosis or pregnancy morbidity. The assessment of overall coagulation is known to differ in APS patients compared to normal subjects. The accelerated production of key factor thrombin causes a prothrombotic state in APS patients, and the reduced efficacy of the activated protein C pathway promotes this effect. Even though significant differences exist in the coagulation profile between normal controls and APS patients, it is not possible to rely on a single test result to diagnose APS. A neural network is a computing system inspired by the human brain that can be trained to distinguish between healthy subjects and patients based on subject specific data. In a first cohort of patients, we developed a neural networking that diagnoses APS. We clinically validated this neural network in a separate cohort consisting of APS patients, normal controls, controls visiting the hospital for other indications and two diseased control groups (thrombosis patients and auto-immune disease patients). The positive predictive value ranged from 62% in the hospital controls to 91% in normal controls and the negative predictive value of the neural network 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. In conclusion, we developed a neural network that accurately diagnoses APS in the validation cohort. After further clinical validation in newly diagnosed patients, this neural network could possibly be clinically implemented to diagnose APS based on thrombin generation data.

Towards standardization of thrombin generation assays: Inventory of thrombin generation methods based on results of an International Society of Thrombosis and Haemostasis Scientific Standardization Committee survey

BACKGROUND

Thrombin generation (TG) is a better determinant of the overall function of the hemostatic system than the routinely used clotting time-based assays. Nowadays, TG is widely used in hemostasis research teams, for both clinical and basic research. However, there is significant variability between laboratories regarding preanalytics, reagents, TG protocol, analysis interpretation, and normalization.

OBJECTIVES

To document the extent of variation in the methodology of TG, we aim to collect all the methods that are being used to measure TG in a survey.

METHODS

We organized a questionnaire through the Standardization committee for Lupus Anticoagulant/Antiphospholipid Antibodies of the International Society of Thrombosis and Haemostasis Scientific Standardization committee. The questionnaire consisted of 51 questions regarding the different aspects of TG: type of users, methods, sample type, analysis, and interpretation of results, normalization, and quality control.

RESULTS

Of the 240 surveys that were started, 108 were completed (45%). However, not all questions were in scope for all laboratories. One-half of the laboratories were research laboratories and half diagnostic laboratories. The most used TG assay was the calibrated automated thrombinography-based assay (56%). There was a divergence regarding several aspects of the TG assay: type of needle for blood collection, blood tubes, centrifugation, sample storage and thawing, reagents, sample dilution, calibration, reference ranges, data normalization, and so on.

CONCLUSIONS

There is an important variation in the methods used for measuring TG. A standardized protocol and data normalization should lead to a better reproducibility and for comparing data from different laboratories.

Reference values for thrombin dynamics in platelet rich plasma

Thrombin generation (TG) is a better determinant of the overall function of the hemostatic system than routinely used clotting time-based assays and can be studied more in detail by thrombin dynamics analysis. Platelet poor plasma is often used to measure TG, however, measuring the contribution of the platelets is also important as patients with a low platelet count or with dysfunctional platelets have an increased risk of developing bleeding. In this study, platelet rich plasma (PRP) was collected from 117 healthy individuals. PRP was measured undiluted and diluted to a varying platelet concentration of 10*10⁹/L to 400*10⁹/L. Prothrombin conversion and thrombin inactivation were calculated from the data obtained by the TG parameters and coagulation factor levels (antithrombin, α₂Macroglobulin (α₂M) and fibrinogen). Reference ranges of TG and thrombin dynamics in PRP of 117 healthy individuals were established. Peak, velocity index and the maximum rate of prothrombin conversion increased linearly with platelet count, but endogenous thrombin potential reached a maximum at 150*10⁹/L as seen in a subset population (n = 20). More extensive analysis revealed that a platelet count below 50*10⁹/L did not affect TG parameters (except for the ETP). Correlation analysis indicated that the platelet count mainly affected the rate of prothrombin conversion. Inhibition of thrombin by antithrombin and α₂M increased with increasing TG, but the ratio of inhibition by antithrombin or α₂M remained the same independently of the total thrombin formed. In conclusion, TG and thrombin dynamics were assessed in PRP of healthy donors to provide reference values for future TG studies in PRP. Increasing the platelet count mainly affected the rate of prothrombin conversion and TG, rather than the total amount of thrombin formed.