A Comparison of Statistical Analysis Between "Real" Patients Reported in Kaplan-Meier Curves and "Reconstructed" Patients Estimated Through the IPDfromKM Method: Analysis of Eight Trials Evaluating Catheter Ablation of Ventricular Tachycardia

Time-to-event endpoints are most widely used in oncology and, to a lesser extent, in cardiology. Typical statistical parameters employed in this context include overall survival, progression-free survival, and recurrence-free survival. The graphical presentation of the results is based on the Kaplan-Meier plot. When Kaplan-Meier curves are included in a meta-analysis, the typical methodological approach is a simplified one because the results of each trial (as well as those of the meta-analysis itself) are expressed through a 2x2 contingency; the methodological simplification is that the follow-up is left out from the analysis and, consequently, the Kaplan-Meier curves are omitted as well. The IPDfromKM method, developed in 2021, is an artificial intelligence algorithm designed to be used in these situations. According to this method, to keep the Kaplan-Meier curves in the meta-analysis, each curve is converted into a database of individual patients (which are denoted as "reconstructed" individual patients). In this way, for the purposes of the meta-analysis, the statistical methods are based on individual patients (like those of clinical trials) so that the Kaplan-Meier curves must not be excluded, and the effect of the follow-up can, therefore, be investigated. This technical report describes the IPDfromKM method in all of its operational details. To present the method, a meta-analysis investigating the effects of catheter ablation to prevent ventricular tachyarrhythmia (VT) has been taken as an example. The original meta-analysis, which included nine controlled trials, was published in February 2023 and adopted the simplified approach based on 2x2 contingency tables. We have reanalyzed these trials by using the IPDfromKM method. Overall, both the standard binary meta-analysis and the IPDfromKM method showed that ablation significantly reduces VT recurrence (hazard ratio, 0.820 for binary meta-analysis vs 0.728 for the IPDfromKM method). By contrast, while no heterogeneity was found by the binary method, the IPDfromKM found significant heterogeneity, which was confirmed by visual inspection of the Kaplan-Meier curves. This suggests that the results of the IPDfromKM method are more accurate because they include the effect of the follow-up on patients' outcomes. In conclusion, our reanalysis confirms the significant benefit determined by ablation, but a more pronounced degree of between-trial heterogeneity has been found. Finally, it should be stressed that, outside the field of meta-analysis, the IPDfromKM method is also applicable to carry out an indirect comparison between treatments that have never been compared in real clinical trials. In this case, reconstructed patients are analyzed by conducting a simulated comparative trial.