One of the modern laboratory markers that allows to adequately assess the state of intravascular coagulation activation during pregnancy is fibrin monomer. The lack of a selective approach of ordering this test for pregnant women causes an increase in laboratory costs and the need to optimize the ordering process. There is a need of new model development which could allow to identify patients which will benefit the most from direct fibrin-monomer (FM) measurement.

THE AIM OF THE STUDY

To develop a predictive model of exceeding the FM concentration the upper limit of the reference interval in pregnant women based on the results of routine clotting tests, D-dimer concentration and gestational age using machine learning.

MATERIALS AND METHODS

To develop and validate a predictive model we used laboratory tests results of 897 pregnant women divided into two groups: group I (n=512) was used to develop a predictive model (training set) and group II (n=385) was used to validate the predictive model (test set). The models were developed using the symbolic regression method with further global optimization by the particle swarm method. Optimization of model parameters and cross-validation were performed using the R software environment.

RESULTS

A predictive model of a FM concentration (as a marker of the plasma hemostasis active state at the study time) increase has been developed. An algorithm for identifying pregnant women with a high probability of increased FM concentration detection has been proposed.

CONCLUSIONS

Measurement of TT and D-dimer concentration with inclusion of the obtained results in the proposed algorithm can be useful for coagulation activation detection during pregnancy and for ensuring evidence-based selectivity during pregnant women stratification for direct FM measurement.