Figure 1.
Schematic representation of the joint probability model structure.
Figure 2.
Predictors of cardiac syncope in the derivation populations.
Figure 3.
Receiver-operator curves (ROC) of the predicted probabilities against the actual outcome, for each dataset.
Model results are shown using all available variables (closed triangles, solid line), for a parsimonious model using only variables common to all datasets (age, gender, structural heart disease, number of spells, and prodromal symptoms; open circles, short dashed line), and for a model using data resampled to create standard distributions of number of spells and age (categorical; open inverted triangles).
Figure 4.
Predicted probability of cardiac syncope (median, inter-quartile range [box], 10-90th percentile [whiskers] and outliers [crosses]) using three sampling methods.
The full model includes all patients with a final diagnosis of cardiac syncope (C) and non-cardiac syncope (NC) using the full model; the parsimonious model uses only the 5 variables common to all datasets; and the resampled model uses resampled data with standardized distributions of age and number of spells.