We have read with interest the study by McInerney et al.1 published in Revista Española deCardiología and offer some comments that we believe may be of interest to readers.

We note that none of the authors of the article are cardiovascular surgeons. Surgeons could have contributed their experience to the interpretation of the analysis of administrative databases which, as is well known, are subject to highly significant biases in the analysis of clinical indicators.2,3 For example, the article reported an incidence of postoperative atrial fibrillation of 2.8%. This value is 5 to 10 times lower than the value known until now4 and is the lowest ever published. However, it seems that the authors overlooked this anomaly, which can only be understood by reference to coding errors and would have been noticed by authors familiar with postoperative complications.

Likewise, we believe that errors were made in patient selection. The authors claim to have excluded patients undergoing mitral or tricuspid surgery from the surgical group. However, after assessing the selection codes in the appendix, they did not exclude other associated procedures, such as thoracic aortic surgery, septal defects, or any type of mitral or tricuspid repairs. We performed our own analysis of the minimum data set (MDS) using stricter exclusion criteria and found that both the number of patients in the surgical arm (n=3446) and their mortality (4.7%) decreased by more than 30%.

Furthermore, the authors decided to exclude all events related to transcatheter aortic valve implantation (TAVI) and percutaneous coronary intervention (PCI) during the same hospital admission, as it was impossible to determine if these were planned or rescue procedures. We performed an MDS query with similar codes for TAVI and detected 187 TAVI and PCI events during the same hospital admission (representing almost 20% of the percutaneous arm referred to in the article), with mortality of 8.6% (n=16). We suggest that the elimination of these patients may involve serious bias and that, in any case, such bias is greater than if they had been included while taking into account that the indication for PCI may be difficult to determine with certainty.

The selection of patients undergoing TAVI with PCI in the previous 6 months is also questionable for the following reasons: a) it is difficult to univocally identify different events concerning the same patient in the MDS; b) the implementation of ICD-105 has led to some hospitals having very deficient MDS data; c) it is arbitrary to exclude patients treated with PCI in the same or a subsequent care episode; and d) it is impossible to determine whether the previous TAVI and PCI procedures were performed for the same clinical syndrome or for a different one.

Regarding the propensity score analysis, it should be emphasized that the characteristics of the matched surgical cohort differed from those of the original cohort (eg, older age, more women, chronic kidney disease, chronic obstructive pulmonary disease, heart failure, etc). It is highly likely that, between 2016 and 2019, any patient with this profile would have been a candidate for TAVI, because the surgical risk would have been high or prohibitive. If the local multidisciplinary teams had decided to opt for surgery, it would have probably been due to the technical impossibility of using a percutaneous approach. Thus, we wonder what useful conclusion for daily clinical practice can be drawn from such a comparison.

In the matched-groups comparison, 10 of the 15 adjustment variables had a standardized mean difference at least 0.1, indicating that the matching was suboptimal. The fact that the estimated propensity score distribution (see figure 2B1) was similar does not imply that the measured baseline covariates were balanced between the 2 groups, nor does it imply that the propensity score model was correctly specified. The area under the curve of the propensity score model is also of little consequence, as it is well known that the area under the curve does not give any indication of whether an important confounding variable has been omitted from the propensity score estimation model.6

In summary, we believe that the findings of this study should be interpreted with extreme caution. We understand that the analysis of databases that include a surgical group may give rise to selection biases and flaws in the interpretation of the results. The participation of surgeons in the analysis of these data may help to ameliorate such problems. Furthermore, we believe that the comparison of clinical indicators should be based on clinical registries. In this regard, the Spanish Society of Cardiovascular and Endovascular Surgery has set up the Spanish Registry of Cardiac Surgery, which can analyze more than 1200 process and outcome indicators at the patient level. We encourage other scientific societies to implement this initiative so that future comparative studies can be performed without the need to resort to administrative databases.