We recognize Dr. Cao’s comments regarding the study design, analysis, and interpretation of the data presented in our manuscript, and we will address his concerns, many of which are related to the inherent nature of the Nationwide Inpatient Sample (NIS) files, in a direct, stepwise fashion.

With regard to outcomes, the data available in the NIS prevent the use of Valve Academic Research Consortium definitions as the NIS is not an organized protocol, and procedural end points, such as paravalvular regurgitation, are simply unavailable. With regard to patient selection, many subgroups of patients who underwent surgical aortic valve replacement (SAVR) (emergent, conversion from transcatheter aortic valve implantation [TAVI], re-do sternotomy) were not definable in the NIS. However, many subgroups, such as a history of coronary bypass surgery (CABG), were definable. Cao points out that the percentage of previous CABG in the SAVR arm rose from 5.5% (overall cohort) to 19.1% (after propensity matching), which results from matching to the raw TAVI arm that had a 25.8% rate of previous CABG. Although Cao is correct in stating that our matched data, therefore, are not representative of the general population of patients who are considered for AVR procedures, it is not intended to be as TAVI patients in 2011 were of the highest risk category (TAVI, in the United States, was technically indicated for only transfemoral, inoperable patients during this time period).

With respect to our matching protocol, we agree with Cao that it would be ideal to match patients according to their EuroSCORE and Society of Thoracic Surgeons score. However, the NIS does not include all variables contained in the 2 aforementioned scoring systems (it should be noted that ethnicity, however, is in fact part of the Society of Thoracic Surgeons score, so we feel it should be included). Cao’s comments regarding 3:1 matching (vs 2:1 or 1:1) resulting in potentially higher bias is true; but 3:1 matching also results in decreased sampling variability and increased efficiency because more cases of SAVR could be used for matching.

We agree that our absolute cost numbers are, indeed, higher than other published data. This is likely because (1) the population being extreme risk (arguably higher risk than in the analysis quoted by Cao) and (2) our numbers actually representing charge. We considered it not necessary to introduce cost-to-charge ratios (to estimate reimbursement) as the intent of our analysis was the relative comparison between TAVI and SAVR. With respect to the large standard deviation observed, it is in fact at par with other studies, including that quoted by Cao.

We do agree that the higher major bleeding rate in the TAVI group is more pronounced in our analysis versus other studies. However, this is not entirely unexpected, as this initial phase of commercial TAVI cases involved centers at the beginning of their learning curve, whereas all these same centers were well versed at traditional SAVR. Moreover, there was no adjudication of adverse events, and, as stated in the manuscript, complications (all based on billing codes) were likely to be less severe than those in the randomized trials. Finally, given that we are comparing index hospitalization costs, neither the estimated costs in the TAVI or SAVR arm are comprehensive and do not account for potential differences in preoperative and postdischarge costs between the groups and is, indeed, a limitation of the study.