We thank Drs. Lanzillo and Moccia for the comment on our article, [1] but we disagree. Indeed, the results obtained for relapses was above a 1.0 confidence inerterval. If this was not the case, the results would have been insignificant.
A Cox model, is appropriate for dealing with censored times-to-event, which was not pertinent in our short-term study. Our principal analyses were performed at 1-year post-treatment initiation and no censoring occurred as only patients with a treatment initiation at least 1 year before the data extraction were included. Drs. Lanzillo and Moccia evidently misinterpreted the statistical model. In face, a direct relationship exists between the proportion of an event at a given time (as we proposed with logistic regression) and the hazard function up to this time (as fitted by a Cox model). Using logistic regression or the Cox model is not a matter of bias, it depended on the kind of indicators we proposed. We preferred to present the results in proportions of event vs instantaneous hazards and corresponding ratios because it is generally more useful and easily understood by clinicians.
There is also a misunderstanding concerning the propensity-score method as Drs. Lanzillo and Moccia refer to an alternative method based on the matching of propensity scores which was not the case in our study. Further explanation can be found in our recent paper on the topic. [2]
We agree with the conclusion of Drs. Lanzillo and Moccia: our study has several limitations and further validation studies must be performed. The main limitations of our study were missing values, sample size, and possible confounding factors. Therefore, even if our conclusion was close to the one proposed, the reasons were obviously different.
1. Barbin L, Rousseau C, Jousset N, et al. Comparative efficacy of fingolimod vs natalizumab: A French multicenter observational study. Neurology 2016;86:771-778.
2. Borgne FL, Giraudeau B, Querard AH, Giral M, Foucher Y. Comparisons of the performance of different statistical tests for time-to-event analysis with confounding factors: practical illustrations in kidney transplantation. Stat Med 2016;35:1103-1116.
For disclosures, please contact the editorial office at [email protected].
We thank Drs. Lanzillo and Moccia for the comment on our article, [1] but we disagree. Indeed, the results obtained for relapses was above a 1.0 confidence inerterval. If this was not the case, the results would have been insignificant.
A Cox model, is appropriate for dealing with censored times-to-event, which was not pertinent in our short-term study. Our principal analyses were performed at 1-year post-treatment initiation and no censoring occurred as only patients with a treatment initiation at least 1 year before the data extraction were included. Drs. Lanzillo and Moccia evidently misinterpreted the statistical model. In face, a direct relationship exists between the proportion of an event at a given time (as we proposed with logistic regression) and the hazard function up to this time (as fitted by a Cox model). Using logistic regression or the Cox model is not a matter of bias, it depended on the kind of indicators we proposed. We preferred to present the results in proportions of event vs instantaneous hazards and corresponding ratios because it is generally more useful and easily understood by clinicians.
There is also a misunderstanding concerning the propensity-score method as Drs. Lanzillo and Moccia refer to an alternative method based on the matching of propensity scores which was not the case in our study. Further explanation can be found in our recent paper on the topic. [2]
We agree with the conclusion of Drs. Lanzillo and Moccia: our study has several limitations and further validation studies must be performed. The main limitations of our study were missing values, sample size, and possible confounding factors. Therefore, even if our conclusion was close to the one proposed, the reasons were obviously different.
1. Barbin L, Rousseau C, Jousset N, et al. Comparative efficacy of fingolimod vs natalizumab: A French multicenter observational study. Neurology 2016;86:771-778.
2. Borgne FL, Giraudeau B, Querard AH, Giral M, Foucher Y. Comparisons of the performance of different statistical tests for time-to-event analysis with confounding factors: practical illustrations in kidney transplantation. Stat Med 2016;35:1103-1116.
For disclosures, please contact the editorial office at [email protected].