PhD Scientific Days 2022

Translational Medicine II. (Poster discussion will take place in the Aula during the Coffee Break)

Examination and prediction of laboratory parameters caused by electrical shocks on animal model

Text of the abstract

Keywords: defibrillation threshold, efficacy prediction, laboratory tests
Introduction Time plays a crucial role in the medical management of resuscitation cases - necessitating fast, organized, well-coordinated teamwork. Physicians' work can be facilitated by state-of-the-art data analysis and forecasting methods that are able to predict both additional diagnostic and therapeutic directions based on patient parameters in real time.
Aims The aim of the research is to analyse the database of defibrillations (DF) in an animal model and to predict the efficacy of DF. The relationships between subjects' (n=15) laboratory parameters and the defibrillation threshold (DFT) were analysed using classical data analysis techniques and machine learning algorithms.
Methods In the experimental setting ventricular fibrillation was induced at by 50 Hz DC, and then DFT was determined by a step-down protocol. Blood samples were taken before and after defibrillations, and levels of PaCO2, PaO2, pH, Hct, Na+, K+, Cl-, Creatinine, Urea, HCO3-, ALT, AST, CK, LDH were measured. Using algorithms of machine learning and artificial intelligence, it has become possible to manage, interpret, and visualize large amounts of data recorded through the experiments. Multiple machine learning models were trained and their accuracy in predicting DFT from laboratory parameter input data were compared.
Results The scatterplot matrix showed no linear association between observed parameters and DFT levels. Statistically significant correlations (p<0.0001) were identified between Na+, Urea, Creatinine, ALT, and DFT (0,373, 0,393, 0,346, 0,348). The Random Forest Classifier algorithm achieved the best prediction performance among the trained models, with an accuracy of 83.3%. The most important laboratory parameters in the prediction method were Urea, PaO2, CREJ2, Hct, CK, LDH, and Na+. A prediction application was developed by integrating the trained model into a web platform and building a user-friendly interface. According to our results, the DFT value can be predicted by entering the laboratory parameters into the developed website.

Conclusion According to our results, the DFT value can be predicted by entering the laboratory parameters into the self-developed predictive website.