Fig. 1.

Schematic representation of predictor, coefficient, and output matrices in logistic regression analysis. In the example presented in this paper, the matrix of the predictors X contains information about the scale factors used to generate a population of 1000 AP models (only 100 cases are shown here for the sake of simplicity) by perturbing the baseline value of 19 parameters (defined in Table 1). The same proarrhythmic protocol is simulated with all the models in the population, and the presence of EADs is assessed in each case, thereby obtaining the binary elements of the matrix of the outputs Y. The result of logistic regression analysis is the matrix of the coefficients B, used to estimate the effect of perturbations in model parameters on the probability of arrhythmogenesis. Here both Y and B are arrays because only one output (i.e., EAD development) is evaluated in this example. Values in X (obtained from the scale factors via log-transformation and conversion into z-scores, see text) and B (regression coefficients, shown also in Fig. 4) are dimensionless quantities, while elements of Y are Boolean variables.