Objective Decision making for intervention in symptomatic aortic stenosis should balance the risks of surgery and of transcatheter aortic valve implantation (TAVI). We identified the factors associated with early mortality after TAVI and aimed to develop and validate a simple risk score.
Methods A population of 3833 consecutive patients was randomly split into two cohorts comprising 2552 and 1281 patients, used respectively to develop and validate a scoring system predicting 30-day or in-hospital mortality.
Results TAVI was performed using the Edwards Sapien prosthesis in 2551 (66.8%) patients and the Medtronic Corevalve in 1270 (33.2%). Approach was transfemoral in 2801 (73.4%) patients, transapical in 678 (17.8%), subclavian in 219 (5.7%) and other in 117 (3.1%). Early mortality was 10.0% (382 patients). A multivariate logistic model identified the following predictive factors of early mortality: age ≥90 years, body mass index <30 Kg/m2, New York Heart Association class IV, pulmonary hypertension, critical haemodynamic state, ≥2 pulmonary oedemas during the last year, respiratory insufficiency, dialysis and transapical or other (transaortic and transcarotid) approaches. A 21-point predictive score was derived. C-index was 0.67 for the score in the development cohort and 0.59 in the validation cohort. There was a good concordance between predicted and observed 30-day mortality rates in the development and validation cohorts.
Conclusions Early mortality after TAVI is mainly related to age, the severity of symptoms, comorbidities and transapical approach. A simple score can be used to predict early mortality after TAVI. The moderate discrimination is however a limitation for the accurate identification of high-risk patients.
Statistics from Altmetric.com
Transcatheter aortic valve implantation (TAVI) is now a well-established technique in patients who are at high risk for surgical aortic valve replacement (AVR).1 ,2 Although surgical risk scores, such as the Euroscore or the Society of Thoracic Surgeons (STS) score, have limitations in the estimation of operative mortality, they are integrated into the decision-making process in order to choose the most appropriate intervention. Immediate results of TAVI can now be accurately assessed from large series.3 However, only a few of them specifically analysed the factors associated with early mortality and were based on relatively small series.4 ,5 In addition, there has not been any attempt to develop a scoring system to predict early mortality following TAVI.
We used the data from the French Aortic National CoreValve and Edwards (FRANCE 2) registry to analyse the predictive factors of early mortality after TAVI. We also aimed to develop and validate a simple scoring system aiming at predicting early mortality after TAVI.
Between 1 January 2010 and 31 December 2011, 3933 consecutive patients undergoing TAVI in the 33 French centres and in the Monaco centre were included in the FRANCE 2 registry. The criteria for the authorisation to perform TAVI, patient inclusion and the organisation of the FRANCE 2 registry were previously described.6 Patients were selected for TAVI if they had severe, symptomatic aortic stenosis and if surgery was contraindicated or judged to be high risk by a multidisciplinary team.
The registry was approved by the Institutional Review Board of the French Ministry of Health. All patients gave written informed consent. Device manufacturers funded the registry but did not have any role in data collection or analysis or in the preparation of the manuscript.
For the purpose of the elaboration and validation of the score predicting early mortality, we excluded 35 patients for whom procedural data were missing and 65 patients in whom TAVI was performed because of degeneration of aortic bioprosthesis (‘valve-in-valve’ procedures). Therefore, the study population comprised 3833 patients.
The two devices used were the balloon-expandable Sapien or Sapien XT prosthesis (Edwards Lifesciences) or the self-expandable CoreValve (Medtronic). Both devices were used in 30 centres, and only the Edwards Sapien was implanted in four centres. The balloon-expandable prostheses were used in 23 and 26 mm sizes with transfemoral approach using the Retroflex 3 or Novaflex delivery catheters, respectively, or the transapical approach using the Ascendra catheter. The 29 mm prosthesis was used only with the transapical approach. The balloon-expandable prosthesis was used in 26, 29 and 31 mm sizes with transfemoral or subclavian approaches.
The choice of the prosthesis, approach and type of anaesthesia was left at the discretion of the teams. The transfemoral approach was favoured as the first approach. Alternate approaches were transapical with the balloon-expandable prosthesis, subclavian with the self-expandable prosthesis and, less frequently, transaortic or transcarotid.
The performance of the procedures has been previously detailed.6 All patients received aspirin (75–160 mg daily) and clopidogrel (300 mg loading dose and 75 mg daily) before the procedure and for 1 month after the procedure, and then a single antiplatelet drug regimen.
Mortality was adjudicated by an independent clinical events committee. The endpoint used for the elaboration and validation of the scoring system was all-cause mortality 30 days after the procedure or longer if the patient was not discharged, according to the recommendations of the Valvular Academic Research Consortium (VARC) classification system.7
Data were prospectively collected using a standardised electronic case report form and sent to a central database (Axonal). Quality control was performed by checking data against source documents for 10% of patients in randomly selected centres.
Continuous variables were expressed as mean± SD. Categorical variables were expressed as percentages and were grouped if needed according to risk progression.
The population of 3833 patients was split into two randomly selected cohorts: a development cohort of 2552 patients used for model building and a validation cohort of 1281 patients.8 ,9 Model coefficients were not re-estimated and applied to the validation cohort. Univariate analysis was performed with a logistic model including the 56 variables listed in table 1.
Variables with p<0.20 in univariate analysis were entered in a multivariate adjusted logistic model with a backward selection procedure and a significance level of p=0.05. Continuous variables were kept continuous at this stage. Two-way interactions were studied between significant variables in multivariate analysis and selected if significant at the level of p=0.01.
We also tested a multilevel hierarchical logistic model including the covariates of the final model and the volume centre as a random effect to analyse the potential heterogeneity across centres. Volume centre was classified into <25, 25–50 and >50 procedures per year.
Then, continuous variables of the final model were categorised according to clinically relevant cut-off points. A score-based prediction rule for early mortality was developed from the multivariate model of the development cohort. Regression coefficients were multiplied by 10/3 and rounded to the nearest integer. The score was the sum of the points corresponding to each variable of the multivariate model.
The discrimination obtained with the final multivariate model, the multilevel hierarchical model and the score was assessed using the c-index and its 95% CI. Overall calibration was tested with the Hosmer–Lemeshow goodness-of-fit test. Calibration performance according to the score value was assessed by computing predicted mortality rates and observed mortality rates with 95% CI and visually represented using calibration plots.8 Analyses were performed with SAS statistical software (SAS Institute, Cary, North Carolina, USA, V.9.3) and R software (R foundation for statistical computing, Vienna, Austria, V.2.12.0).
Patient characteristics are described in table 1 for the development and validation cohorts. Age was ≥90 in 435 patients (11.4%) and ≥80 in 2776 (72.4%). Mean logistic Euroscore was 21.5±13.8 in the development cohort and 21.9±14.1 in the validation cohort.
The type of prosthesis and the approach are detailed in table 1. Other approaches were transaortic in 103 patients and transcarotid in 14. Prosthesis implantation was successful in 3698 patients (97.2%).
Follow-up was complete in 3831 patients (99.9%). There were 382 early deaths (10.0%), including 253 (9.9%) in the development cohort and 129 (10.1%) in the validation cohort.
Death occurred on the day of the procedure in 84 cases (22.0%), between the second day and the end of the first week in 110 cases (28.8%), during the second week in 59 (15.4%), the third week in 49 (12.8%), between day 21 and day 30 in 52 (13.6%) and after day 30 in 28 (7.4%).
When dividing the 2 years of the study into eight trimesters, early mortality rates varied between 8.6 and 11.5% and there was no difference between the eight trimesters (p=0.85). There was no difference in early mortality rates according either to the centre (p=0.27) or the annual volume centre: 10.0% in the 5 centres performing <25 procedures, 10.9% in the 10 centres performing 25–50 procedures and 9.9% in the 19 centres performing >50 procedures (p=0.89).
Early mortality was 8.0% (142/1774) with the transfemoral approach using the Edwards Sapien prosthesis, 9.3% (95/1019) with the transfemoral approach using the Medtronic CoreValve prosthesis, 8.8% (19/217) with the subclavian approach using the Medtronic CoreValve prosthesis, 15.1% (102/677) with the transapical approach using the Edwards Sapien prosthesis and 12.8% (15/117) with other approaches. There was no interaction between the type of prosthesis and the approach in relation to early mortality (p=0.30).
Predictive factors of early mortality
Predictive analysis of early mortality was performed in the 2552 patients from the development cohort. Univariate analysis is detailed in table 2.
Multivariate analysis identified nine independent predictive factors of early mortality (see online supplementary table S2). No interaction was significant. The c-index obtained with the final logistic model in the development cohort without coding of continuous variables was 0.68 (95% CI 0.65 to 0.72). The c-index obtained with the multilevel hierarchical model was 0.69 (95% CI 0.65 to 0.72), showing no effect of the heterogeneity across centres on the prediction. There was no significant difference between predicted and observed mortality (p=0.60). The c-index obtained with the logistic Euroscore was 0.59 (95% CI 0.55 to 0.64).
The scoring system was derived from the final multivariate model in the development cohort. The final model is detailed in online supplementary table S3.The score was calculated by adding each component and ranged theoretically from 0 to 21 (table 3). In the development cohort, the score ranged between 0 and 14.
The relationship between the score value and predicted early mortality is shown in figure 1. The OR associated with a one-point increase of the score was 1.33 (95% CI 1.25 to 1.40). The c-index was 0.67 (95% CI 0.64 to 0.71) for the score in the development cohort. The p value of the Hosmer–Lemeshow test was 0.40. Predicted and observed early mortality rates according to score values are detailed for the development cohort in figure 2 and in the online supplementary table S4.
The predictive model and the score were tested in the validation cohort.
The c-index was 0.59 (95% CI 0.54 to 0.64) for the final multivariate model and 0.59 (95% CI 0.54 to 0.64) for the score in the validation cohort. The corresponding p values of the Hosmer–Lemeshow test were 0.50 with the final multivariate model and 0.50 with the score. Predicted and observed early mortality rates according to score values are detailed for the validation cohort in figure 3 and in the online supplementary table S5. The good calibration was also shown by the fact that the line fitting the data was close from the diagonal line for predicted mortality up to 20%.
In this large nationwide registry comprising all TAVI procedures performed with different devices and approaches, eight patient-related and one procedure-related variables were identified as predictive factors of early mortality. It was possible to derive a simple multivariate score to estimate early mortality after TAVI. However, its moderate discrimination limits its ability to reliably identify patients at high risk of early death after TAVI.
The present study population reflects contemporary TAVI practices in high-risk patients. A strength of this registry is to include all consecutive procedures performed with the two most commonly used prosthetic devices and all approaches.10 We excluded TAVI procedures performed for bioprosthesis degeneration, which is an off-label indication, and is the subject of specific studies.11
Early mortality after TAVI
Early all-cause mortality is a standardised endpoint proposed by the VARC and is consistent with guidelines for reporting mortality after cardiac valve interventions.7 ,12 Mortality at 30 days or longer if the patient was not discharged is also the usual endpoint for assessing early mortality in surgical databases and with risk scores.13 ,14
The 10.0% early mortality rate is consistent with other multicentre registries, ranging from 5.4 to 12.4%.4 ,5 ,15–18 A meta-analysis on more than 16 000 procedures reported a 30-day mortality rate of 8.1%.3
The absence of difference in mortality rates according to centres, volume activity and between trimesters can be related to the fact that 30 of the 34 centres had previous experience with TAVI.6
Predictive factors of early mortality
Older age was associated with a modest but significant increase in early mortality.
The relationship between low body mass index (BMI) and increased early mortality is an original finding with regards to TAVI. This shows that early mortality is increased not only in patients with BMI<18.5, but also in those with normal weight, as compared with overweight patients with a BMI≥30. Consistent findings have been reported after AVR for aortic stenosis, with the lowest early postoperative mortality rates in patients with a BMI around 30.19 A BMI<24 was a predictive factor of 6-month mortality after AVR in octogenarians.20 Weight loss is also a component included in the frailty scoring systems.
The two comorbidities identified as predictive factors of 30-mortality after TAVI were respiratory insufficiency and dialysis, which is consistent with valve surgery.21
Four factors were directly related to the severity of hemodynamic consequences of aortic stenosis, that is, recent acute pulmonary oedema (APE), New York Heart Association (NYHA) class IV, pulmonary hypertension and critical state, which comprises conditions related to poor hemodynamic status according to the definition of the Euroscore.13 Pulmonary hypertension and the need for hemodynamic support were also associated with early mortality in the Canadian registry of TAVI.4 In the present series, haemodynamic indices were strong predictive factors of early mortality after TAVI. This points out the importance of a timely intervention to reduce the procedural risk and may suggest a role for balloon aortic valvuloplasty as a bridge to TAVI in patients with poor hemodynamic condition. Impaired LVEF was associated with early mortality in univariate but not in multivariate analysis. This illustrates a strong confounding effect between EF and NYHA class or pulmonary hypertension. The transapical approach and other approaches (transaortic and transcarotid) were associated with a higher early mortality. On the other hand, there were no differences between the transfemoral and subclavian approaches. Increased mid-term mortality with transapical approach has also been shown in another registry.17 The interpretation of the relationship between approach and mortality should be cautious. Thoracotomy may be harmful in high-risk patients. However, since the transfemoral approach was favoured in the FRANCE 2 registry, transapical and other approaches were likely to be used in sicker patients. In addition, the modalities of pain control influence postprocedural mortality.22 Even a multivariate analysis cannot control for all potential known and unknown confounding factors in this setting. There were no differences between the two prosthetic devices, as also shown in other recent analyses.3
The risk score combining nine variables enables early mortality to be estimated for any given patient. Although score values theoretically range between 0 and 21, no patient had a risk score >14, showing that patients combining all predictive factors of early mortality were denied the procedure according to clinical judgement of the heart team.
The score achieved, however, only a moderate discrimination, as attested by the value of the c-index of 0.67 in the development sample and 0.59 in the validation sample. Although predictive analyses of early and mid-term mortality after TAVI have been reported, they did not comprise a specific validation sample allowing for an unbiased estimation of their predictive performance.8 Surgical risk scores achieved a better discrimination, with c-indexes around 0.80.13 ,14 The German aortic valve risk score is specific to aortic stenosis and achieves a c-index of 0.81 but comprises 95% of AVR and only 573 TAVI procedures.23 Unpublished analyses from the SOURCE (for Edwards SAPIEN Aortic Bioprosthesis European Outcome) registry mentioned comparable values of c-index obtained with multivariate predictive models of early mortality (0.61 for transapical and 0.70 with transfemoral approach).24 In a population of patients aged ≥70 undergoing cardiac surgery, the c-index was 0.65 with the Euroscore and 0.67 with the STS score, attesting for limited discrimination performance of scores in specific populations of high-risk patients.25 Whatever TAVI or surgery is considered, very high-risk patients represent a small and particularly heterogeneous group, in which it is difficult to collect all variables and to estimate the contribution of each variable influencing early mortality. That is the reason why recent recommendations emphasise the importance of clinical judgment.1 ,2 Finally, a number of postprocedural complications may increase early mortality after TAVI.7 This may contribute to limit the predictive performance of a model based on preprocedural patient characteristics.
With regard to calibration, there was a good overall concordance between the numbers of predicted and observed deaths. The analysis of score calibration according to the score value shows a good concordance between predicted and observed deaths for predicted early mortality rates <20%, which corresponds to >90% of the patients. The concordance between predicted and observed mortality in very high-risk patients is difficult to ascertain given the small number of patients. This is consistent with the limitations of risk scores in surgical AVR.26 Even if the recent Euroscore II achieves a good discrimination, its calibration properties are weaker in patients with a predicted mortality >30%.27
The lack of standardised definition for certain comorbidities may introduce some heterogeneity, particularly for respiratory insufficiency. However, only one series used predefined spirometric criteria to analyse the impact of pulmonary disease after TAVI.28
The assessment of the predictive performance of the score is limited by the lack of external validation. Nevertheless, the validation of the score in a randomly selected cohort limits the risk of overestimation with the predictive value of the score.
The FRANCE 2 registry did not collect indices assessing functional or cognitive capacity, particularly indices of frailty. Taking such indices into account may improve the discrimination.
Since this score was developed from a cohort of patients at high risk for surgery, it cannot be extrapolated to patients at intermediate risk, who are currently considered as candidates for surgical AVR but are the subject of specific trials.
The analysis based on a contemporary registry allows for a better knowledge of patients who are at high risk of early mortality after TAVI. Four of the eight patient-related characteristics are related to the consequences of aortic stenosis, that is, NYHA class IV, previous APE, pulmonary hypertension and critical state, thereby highlighting the importance of timely indication of TAVI. The scoring system combining predictive factors allows an easy estimation of early mortality after TAVI performed with current devices and approaches. This score may contribute to decision making by a multidisciplinary team in high-risk patients with aortic stenosis. Despite a good concordance between predicted and observed mortality, the moderate discrimination illustrates the difficulties in achieving a reliable prediction of early outcome after TAVI in an individual patient. Improvement in patient selection will also require predictive analyses of mid-term outcome.
What is already known on this subject
Immediate results of transcatheter aortic valve implantation (TAVI) can now be assessed from large numbers of patients. The assessment of early mortality is an important component of decision making. Only a limited number of relatively small series specifically analysed the factors associated with early mortality. No scoring system has been developed to predict early mortality following TAVI.
What this study adds
This study identified nine predictive factors of early mortality after TAVI. It also describes a simple scoring system that can contribute to individualise risk assessment, although its moderate discrimination performance limits its ability to accurately identify patients at high risk of early death after TAVI.
How might this impact on clinical practice
Four of the eight patient-related characteristics are related to the consequences of aortic stenosis, that is, New York Heart Association class IV, previous acute pulmonary oedema, pulmonary hypertension and critical state. Their impact on early mortality after TAVI highlights the importance of timely indication. Despite limited predictive performance, the proposed score can be useful in decision making in high-risk patients with severe, symptomatic aortic stenosis.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
Collaborators FRANCE 2 Investigators: see supplementary table S1.
Contributors BI participated in the conception of the FRANCE 2 registry, designed the score study, participated in data analysis and drafted the manuscript. CL performed the statistical analysis. DH and AV revised the manuscript content critically. Other authors contributed to the design of the FRANCE 2 registry and revised the manuscript. All authors approved the final version of the submitted manuscript.
Funding The FRANCE 2 registry was funded by Edwards Lifesciences and Medtronic
Competing interests BI has received consultant fees from Abbott, Boehringer Ingelheim, Bayer, Valtech, and speaker's fees from Edwards Lifesciences. DH has received proctoring fees from Edwards Lifesciences and Medtronic. HE received consultant fees from Edwards Lifesciences. PL received consultant fees from Medtronic. ET received consultant fees from Medtronic. ML received grants from Medtronic. AV is member of Advisory Board for Medtronic, Abbott, Valtech, and Boehringer Ingelheim, and has received speaker's fees from Edwards Lifesciences and Siemens. The other authors report no conflicts.
Ethics approval Institutional Review Board of the French Ministry of Health.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.