A Preliminary Study on the Usage of a Data-Driven Probabilistic Approach to Predict Valve Performance Under Different Physiological Conditions

Predicting potential complications after aortic valve replacement (AVR) is a crucial task that would help pre-planning procedures. The goal of this work is to generate data-driven models based on logistic regression, where the probability of developing transvalvular pressure gradient (DP) that exceeds 20 mmHg under different physiological conditions can be estimated without running extensive experimental or computational methods. The hemodynamic assessment of a 26 mm SAPIEN 3 transcatheter aortic valve and a 25 mm Magna Ease surgical aortic valve was performed under pulsatile conditions of a large range of systolic blood pressures (SBP; 100-180 mmHg), diastolic blood pressures (DBP; 40-100 mmHg), and heart rates of 60, 90 and 120 bpm. Logistic regression modeling was used to generate a predictive model for the probability of having a DP > 20 mmHg for both valves under different conditions. Experiments on different pressure conditions were conducted to compare the probabilities of the generated model and those obtained experimentally. To test the accuracy of the predictive model, the receiver operation characteristics curves were generated, and the areas under the curve (AUC) were calculated. The probabilistic predictive model of DP > 20 mmHg was generated with parameters specific to each valve. The AUC obtained for the SAPIEN 3 DP model was 0.9465 and that for Magna Ease was 0.9054 indicating a high model accuracy. Agreement between the DP probabilities obtained between experiments and predictive model was found. This model is a first step towards developing a larger statistical and data-driven model that can inform on certain valves reliability during AVR pre-procedural planning.