Records of all episodes between 1998 and 2017 were retrieved from the MBDS. These records had to include ICD-9 procedural codes 35.21 or 35.22 or ICD-10 codes 02RF07Z, 02RF08Z, 02RF0KZ, 02RF47Z, 02RF48Z, 02RF4KZ, X2RF032, X2RF432, 02RF0JZ, or 02RF4JZ.

We then excluded all patients who had undergone any other major cardiac procedure during the same admission (operations on other valves, thoracic great vessel repair, or coronary artery bypass grafting). Also not considered were patients younger than 18 years of age and those who had undergone any congenital defect repair. We additionally excluded patients who had undergone a transcatheter aortic valve implantation (TAVI), valvuloplasty, or aortic valve repair or those who had endocarditis. TAVI was excluded by eliminating all records with ICD-9 codes 35.05 and 35.06 or ICD-10 codes 02RF37Z, 02RF38Z, 02RF3JZ, 02RF3KZ, X2RF332, 02RF37H, 02RF38H, 02RF3JH, or 02RF3KH after 2013 and those who had received an aortic tissue valve (code 35.22) without extracorporeal circulation (code 39.61) before 2014 (in 2014, the specific coding for TAVI was included in the ICD-9). Given that it is not possible to differentiate between bioprostheses and aortic homografts in the ICD-9, we considered both to be bioprostheses for the purposes of this study. Nonetheless, the proportion of homografts coded as bioprostheses should be small because endocarditis has been excluded and homografts are uncommonly implanted in Spain for any other indication.13 The patient selection algorithm is shown in figure 1.

Figure 1.

Flow diagram showing patient and episode selection. CABG, coronary artery bypass grafting; rec., records; TAVI, transcatheter aortic valve implantation.

(0.24MB).

The first admission of a patient during the study period was considered the index event and the concatenated episodes of transfer between hospitals were considered a single event with an admission date equal to that of the first concatenated episode and a discharge date equal to that of the last.12

The full period of time (1998-2017) was divided into four 5-year intervals (1998-2002, 2003-2007, 2008-2012, and 2013-2017). Comorbidities, mortality, and type of aortic valve prosthesis were analyzed according to the time interval.

To estimate the number of SAVRs per million inhabitants per year, we used the Spanish population reported by the Spanish National Institute of Statistics.14 Hospitals performing SAVR were classified according to the distribution of the mean volume of interventions per year in each period (low-volume centers if their mean number of interventions/y in a period was in quartile 1, intermediate–low-volume centers if their mean number of interventions/y in a period was in quartile 2, and so on) (figure 1 of the supplementary data).

Patients were classified into 4 groups according to age (≤ 60, 60-70, 70-80, and> 80 years). In these groups, we analyzed the trends in the prevalence of various comorbidities (table 1). The age-modified Charlson comorbidity index was calculated,15,16 and the population was divided into 4 groups according to quartile. Predictors of mortality, variabilities in mortality, and trends in risk-adjusted mortality were investigated throughout the study period.

We investigated the changes in the proportion of and factors associated with the type of aortic prosthesis (mechanical or biological). Codes 35.21, 02RF07Z, 02RF08Z, 02RF0KZ, 02RF47Z, 02RF48Z, 02RF4KZ, X2RF032, and X2RF432 were used to identify biological prostheses and homografts and 35.22, 02RF0JZ, and 02RF4JZ to identify mechanical prostheses.

Categorical variables are presented as absolute and relative frequencies (%) and were compared with a chi-square test. The normality of the quantitative variables was analyzed with normality plots. They are expressed as mean (standard deviation) or median and interquartile range [IQR]. Quantitative variables were compared with ANOVA or via nonparametric median comparisons if the distribution was not normal. Further analyses were performed to check for linear trends (LTs). Relative risk reductions (RRRs) and odds ratios (ORs) were calculated to estimate the strengths of associations between baseline variables and mortality.

A joinpoint regression analysis was applied to study varying trends in mortality over time to identify the time point(s) at which the trend changed.17 We calculated the overall annual percentage change (APC) as well as the APC of the periods defined by joinpoint analysis. Joinpoint regression analyses were performed using Joinpoint Regression Program version 4.8.0.1, provided by the Surveillance, Epidemiology, and End Results Program (National Cancer Institute). Further information on joinpoint regression is available in figure 2 of the supplementary data.

Through multivariate stepwise binary logistic regression, factors associated with in-hospital mortality were investigated. The variables in the model were selected according to theoretical criteria or if they were significantly related (P <.05) to the dependent variable in a previous univariate analysis. Multivariate model performance was studied using the area under the curve and Hosmer-Lemeshow test.

To adjust mortality for patient comorbidity, we estimated predicted risks of death with different multivariate logistic regression models. The best model was selected considering the best Akaike's information criteria and area under the receiver operating characteristic curves. Using the probability of in-hospital mortality predicted with this model, we calculated the risk-adjusted mortality ratio (RAMR). RAMR represents the ratio between the observed mortality and the predicted risks estimated by the logistic model. More information on the construction of the RAMR can be found in the figure 3 and figure 4 of the supplementary data.

Variations in the proportions of biological and mechanical valves were studied, and factors associated with the type of implanted prosthesis were detected with stepwise binary logistic regression. The variables in the model were selected according to theoretical criteria or if they had been significantly related (P <.05) to the dependent variable in a previous univariate analysis.

All statistical analyses were performed with Stata version 15.0 (StataCorp, College Station, Texas, United States).

The MBDS database contains 147 921 records of patients who underwent SAVR between 1998 and 2017. Of these, 95 225 underwent SAVR with or without concomitant coronary artery bypass grafting (CABG). Finally, 77 668 (77.4%) underwent isolated SAVR (figure 1).

The SAVR volume and the number of procedures per million inhabitants increased linearly over time (figure 2). SAVR procedures were reported by 40 hospitals in 1998 vs 50 in 2017. The median number of interventions per center and year increased progressively—from 59.5 in 1998 to 90 in 2017 (P <.001)—but at a lower rate than the number of procedures per million inhabitants (figure 2B). The median [IQR] numbers of procedures per year and institution were 61.8 [53.2-88.2], 71.6 [56.8-103], 90 [66.8-112.4], and 84 [71.6-117.4] in the first, second, third, and fourth periods, respectively (PLT <.001) (figure 1 of the supplementary data).

Figure 2.

Number of procedures. A: number of surgical aortic valve replacements (SAVRs) in Spain. B: number of procedures per million inhabitants (circles) and median number of procedures/hospital in Spain (triangles) (P value for trend analysis <.001).

(0.22MB).

Age increased by 3 years (68.9 vs 71.3, PLT <.001) from the first to the last period. In addition, 31 782 of the patients were women (43.1%). A worsening of patients’ risk profile was observed, with an increased prevalence of almost all comorbidities (table 1). The mean (standard deviation) age-modified Charlson index worsened from 2.7 (1.4) to 3.6 (1.7) (P <.001).

In-hospital mortality more than halved from 7.2% (169 of 2342) in 1998 to 3.3% in 2017 (155 of 4629) (RRR, 69.7%; PLT <.001) (figure 3A). This decrease occurred in all age groups (figure 3B): <60 years, 4.9% vs 1.4% (PLT <.001); 60-70 years, 6.2% vs 3.3% (PLT <.001); 70-80 years, 9.5% vs 3.6% (PLT <.001); and> 80 years, 11.8% vs 4.3% (PLT <.001). More information on mortality rates is shown in table 1 of the supplementary data.

Figure 3.

In-hospital mortality after isolated surgical aortic valve replacements in Spain. A: reduction in global mortality (P value for trend analysis <.001). B: trend in the risk-adjusted mortality ratio (RAMR) over time (P value for trend analysis <.001).

(0.18MB).

The study period and hospital volume, among others, were associated with mortality (table 2). A multivariate logistic model including factors associated with mortality showed fairly good performance (area under the curve, 0.76; 95% confidence interval [95%CI], 0.75-0.76; P=.23 [Hosmer-Lemeshow]).

The overall average APC in mortality was −3.6 (95%CI, −4.5 to −2.7). A relevant change in the APC was detected in 2006. Between 1998 and 2006, the APC was 0.4 (95%CI, −1.3 to 2.1). From 2006 to 2017, it was −6.3 (95%CI, −7.5 to −5.1) (figure 2 of the supplementary data).

We detected a linear decrease in the RAMR between 1998 (1.3) and 2017 (0.7) (PLT <.001), which is consistent with the observed worsening risk profile of patients and the reduced mortality (figure 3B). Further information regarding the performance of the model used to estimate the RAMR can be found in figure 3 of the supplementary data.

In total, 32 860 patients (44.7%) received a bioprosthesis. Male patients received mechanical prostheses more frequently than women (58.8% vs 50.7%; P <.001).

From 1998 to 2002, 20.7% of implanted prostheses were tissue valves. This proportion increased linearly: from 2003 to 2007 (37.9%), from 2008 to 2012 (49.7%), and from 2013 to 2017 (59.6%) (PLT <.001). As shown in table 1 and figure 4, there was a decrease in the ratio of mechanical to biological prostheses. As expected, the use of mechanical prostheses was greater among patients <65 years of age (86.7%, 17 757 of 20 488) than among those ≥ 65 (43.2%, 22 877 of 53 006) (P <.001). There was a progressive increase in the use of tissue valves in the 2 groups throughout the series (figure 4A of the supplementary data) (P <.001). We also detected an inverse relationship between the volume of hospital activity and the use of bioprostheses (figure 4B of the supplementary data) (PLTs <.001).

Figure 4.

Changes in the type of aortic prosthesis. The proportions of bioprostheses are shown (P value for trend analysis <.001).

(0.15MB).

Factors related to the implantation of tissue valves are shown in table 2 of the supplementary data.

Clinical interpretations of administrative data, no matter how carefully performed, should always be taken with caution. Coding errors in clinical information and the lack of availability of ICD-9/ICD-10 codes to cover the entire variety of procedures and diagnoses prevent us from adequately defining variables to adjust the baseline risk of patients. For the same reason, it was not possible to estimate risk of frailty scores for SAVR (such as EuroSCORE,29 Katz's index, or Fried's score), which could have helped to better define the baseline risk. Given the information available in the MBDS, the native valve lesion (stenosis or regurgitation) and its etiology are unknown. Stratified analyses according to these variables would have undoubtedly added valuable information to the study.

We do not have enough information to explain the differences in the choice between bio- and mechanical prostheses with respect to other countries and according to center volume. The reported SAVR volume in Spain might be underestimated because records from hospitals that do not report to the MBDS have not been collected. According to the Spanish Department of Health and previous reports, some records have been missed during 2016 and 2017 due to the conversion from the ICD-9 to ICD-10.30 Therefore, the volume of procedures during these 2 years might, once again, be underestimated.