We read with interest the study by Deharo et al. published in the European Heart Journal earlier this year (1). The authors use data from the national administrative PMSI database, which relies on hospital data regarding diagnosis and treatment during an index hospitalisation. The main objective of such a database is to facilitate the interaction between hospitals and insurance authorities on the one side, inform administrators on particular trends in healthcare management, and enable policy making (PMID: 28168290).
The main advantages of administrative data are well known: firstly, the creation of real-world longitudinal data, based on the fact that a unique identification number identifies each patient and any subsequent hospitalisations are connected to the initial one; secondly, data collection is comprehensive, as all hospitalisations are assigned to specific primary and secondary diagnoses and treatments (2). However, discrepancies in volumes, classification of comorbidities, lack of adjudication of adverse events and, subsequently, inadequate risk adjustment are common when comparing administrative data to scientific data (3). Most importantly, misclassifications of surgical procedures can occur, affecting group stratification and increasing the risk of misleading comparisons (4). It is essential to recognise that the PMSI database is not constructed according to the rigorous criteria required for clinical research (PMID: 28168290). Additionally, other studies have reported significant coding errors that have impacted research findings (PMID: 29673880), as there is no data audit process to ensure that data are consistent, accurate, and complete (PMID: 30898564).
Contrary to the administrative value of this data, there are significant drawbacks to its scientific validity. The methodological shortcomings can be summarised as follows:
First, The exact aetiology of mitral regurgitation is missing. There is no information on the percentage of primary vs secondary disease. Table 1 implies that even after propensity matching, higher rates of mitral valve endocarditis were observed in patients undergoing surgery.
Second, the mean duration of the follow-up was 1 year, according to the graphical abstract. Based on this fact, the conclusion on long-term follow-up does not reflect the data presented. However, long-term administrative data collection would have been possible for the authors. Data analysis is based on administrative data collected between 2012 and 2022. This means that all patients have already accomplished at least the landmark of a 1-year follow-up. According to Figure 2, the percentage of patients reaching 3 years is less than 10%. This means that 90% of the patients were either operated on between 2019 and 2022 or excluded during propensity matching. This fact questions the main advantage of administrative data in creating a real-world environment.
Moreover, there are significant differences between the clinical events reported before and after 3 months of follow-up. This refers to the primary endpoint (all-cause death) and the combined endpoint (all-cause mortality, hospitalisation for heart failure or atrial fibrillation, stroke). Supplementary tables 5 and 6 verify previous studies showing a higher risk of surgery during the first 3 months after the procedure, which is mainly related to the perioperative risk. However, the risk of clinical events rises for TEER at longer follow-up. Since the mean follow-up is short (only 1 year), the two curves will probably cross if earlier patients are included or the authors wait for a representative number of patients to reach a 3-5 years follow-up. This is a major prerequisite to draw meaningful conclusions.
Finally, there is a discrepancy between the discussion and the supplementary data related to the disease classification. Although the unavailability of data on the mechanism of mitral regurgitation is recognised as one of the study's limitations, the interaction between primary and secondary MR is presented in supplementary table 8. According to this table, there are differences in cardiovascular mortality in short-term follow-up favouring TEER only in secondary MR. This benefit is not demonstrated for primary MR. This fact strengthens the assumption that the model included mainly patients with secondary MR due to propensity matching. If this is true, then the conclusions mainly apply to secondary MR patients and cannot be generalised to the whole population of mitral regurgitation. To support their argument, the authors are expected to provide results for both pathologies or revise the conclusions accordingly.
Our mission: To reduce the burden of cardiovascular disease.