Nowadays, the first-line therapy for paroxysmal atrial fibrillation (PAF) is pulmonary vein isolation through catheter ablation. However, the success rate of this procedure is still not as high as desirable. Thus, preoperative prediction of early AF recurrence after ablation is a challenge to select optimal candidates for the intervention. To this end, some promising predictors based on the P-wave in short ECG signals have been proposed in the last years. However, evolution of the P-wave along the time has still not been analyzed. Hence, the present work studies how time variability of two features of the P-wave predicts midterm cryoablation failure. For 45 PAF patients, a standard 12-lead ECG signal was obtained for 5 minutes before ablation. An automatic algorithm was then used to delineate all P-waves in lead II, and duration and amplitude were computed. The resulting time series were characterized by their mean, standard deviation and coefficient of variation (CV). Correlating these measures with ablation outcome, the CV for both parameters obtained the best discrimination between patients. In fact, compared with the mean value, the CV for both features obtained accuracies 10% greater, thus achieving values of 70%. These outcomes entail that time variability of the P-wave can reveal new information about the proarrhythmic condition of the patients, thus improving predictions of ablation failure.