Risk stratification tools used in patients with severe aortic stenosis have been mostly derived from surgical series. Although specific predictors of early mortality with transcatheter aortic valve replacement (TAVR) have been identified, the prognostic impact of their combination is unexplored. We sought to develop a simple score, using preprocedural variables, for prediction of 30-day mortality after TAVR. A total of 1,878 patients from a national multicenter registry who underwent TAVR were randomly assigned in a 2:1 manner to development and validation data sets. Baseline characteristics of the 1,256 patients in the development data set were considered as candidate univariate predictors of 30-day mortality. A bootstrap multivariate logistic regression process was used to select correlates of 30-day mortality that were subsequently weighted and integrated into a scoring system. Seven variables were weighted proportionally to their respective odds ratios for 30-day mortality (glomerular filtration rate <45 ml/min [6 points], critical preoperative state [5 points], New York Heart Association class IV [4 points], pulmonary hypertension [4 points], diabetes mellitus [4 points], previous balloon aortic valvuloplasty [3 points], and left ventricular ejection fraction <40% [3 points]). The model showed good discrimination in both the development and validation data sets (C statistics 0.73 and 0.71, respectively). Compared with the logistic European System for Cardiac Operative Risk Evaluation in the validation data set, the model showed better discrimination (C statistic 0.71 vs 0.66), goodness of fit (Hosmer-Lemeshow p value 0.81 vs 0.00), and global accuracy (Brier score 0.054 vs 0.073). In conclusion, the risk of 30-day mortality after TAVR may be estimated by combining 7 baseline clinical variables into a simple risk scoring system.
A logistic European System for Cardiac Operative Risk Evaluation (EuroSCORE) version I ≥20% has been suggested as an indication for transcatheter aortic valve replacement (TAVR). The EuroSCORE and the recently updated EuroSCORE II were developed and validated in the setting of standard cardiac surgery, and thereby their algorithms miss important TAVR-specific variables and predictors of operative mortality. Used as a tool for risk stratification, the EuroSCORE is known to markedly overestimate the risk of death at 30 days after TAVR, which is expected at ∼8% using Valvular Academic Research Consortium definitions. Identifying patients with severe aortic stenosis (AS) in whom the risk of operative mortality with TAVR is very high (e.g., >15%), or even prohibitive, might have important implications in terms of patient selection, avoidance of futile procedures, and cost-effectiveness. However, there is still a lack of simple preprocedural, reproducible, bedside assessment tools specifically developed and validated in the TAVR scenario. The aim of the present study was to develop and validate a simple risk scoring system that could be readily applied at the bedside to evaluate the individual risk of patients with AS dying within 30 days from a TAVR procedure.
Methods
The risk score was built from the TAVR cohort of the Observational Study Of Appropriateness, Efficacy And Effectiveness of AVR-TAVR Procedures For the Treatment Of Severe Symptomatic Aortic Stenosis (OBSERVANT) study, whose details have been reported elsewhere. Briefly, OBSERVANT is a national, observational, prospective, cohort study that enrolled patients with AS (n = 7,618) undergoing TAVR (n = 1,911) or surgical aortic valve replacement (n = 5,707) at 95 Italian centers (34 hemodynamic centers and 61 cardiac surgery centers) from December 2010 to June 2012, run by the Italian National Health Institution in cooperation with the Italian Ministry of Health, the National Agency for Regional Health Services, Italian regions, and Italian scientific societies and federations representing Italian professionals involved in the topic of AS management (see Supplementary Appendix ).
Administrative in-hospital mortality data was obtained through a record linkage with the first admission recorded in the National Hospital Discharge records. Thirty-day clinical follow-up was obtained by clinical visits and telephone contacts and integrated through a record linkage with the Tax Register Information System.
The eligible study cohort was randomized in a 2:1 manner to create a development and a validation data set, respectively. Baseline parameters in the development data set were initially examined by univariate logistic regression analysis for their relation to 30-day mortality. All variables in Tables 1 and 2 were available for selection in this model. Age, body mass index, glomerular filtration rate (GFR), albumin level, hemoglobin level, left ventricular ejection fraction (LVEF), frailty, and mean aortic pressure gradient were dichotomized and treated as binary variables, based on established cutoffs and previous studies. Frailty was assessed with the Geriatric Status Scale, with values of 2 or 3 indicating moderate to severe frailty. Severe left ventricular systolic dysfunction was defined as LVEF <40%. Based on World Health Organization criteria, moderate or severe anemia was defined as a baseline hemoglobin level of <11.0 g/dl for both men and women. GFR was calculated by the Modification of Diet in Renal Disease formula, and renal impairment was defined based on the median of GFR (45 ml/min). Pulmonary arterial hypertension was defined as systolic pulmonary artery pressure >60 mm Hg. Critical perioperative state was defined as ventricular tachycardia, ventricular fibrillation, or aborted sudden death, preoperative cardiac massage, preoperative ventilation, preoperative inotropes or intra-aortic balloon pump, or preoperative acute renal failure (anuria or oliguria <10 ml/hour).
| Variable | Overall (n = 1,878) | Development (n = 1,256) | Validation (n = 622) | p Value |
|---|---|---|---|---|
| Age (yrs) | 81.9 ± 6.2 | 81.9 ± 5.9 | 81.7 ± 6.6 | 0.291 |
| Age >85 yrs | 520 (27.7) | 342 (27.2) | 178 (28.6) | 0.527 |
| Female gender | 1,087 (57.9) | 724 (57.6) | 363 (58.4) | 0.767 |
| Body mass index (kg/m 2 ) | 25.9 ± 4.8 | 25.9 ± 4.7 | 26.1 ± 4.9 | 0.511 |
| Body mass index <25 kg/m 2 | 842 (46.2) | 572 (47.0) | 270 (44.6) | 0.331 |
| Diabetes mellitus | 506 (26.9) | 337 (26.8) | 169 (27.2) | 0.867 |
| Smoker | 208 (11.1) | 151 (12.0) | 57 (9.2) | 0.061 |
| Creatinine (mg/dl) | 1.3 ± 0.9 | 1.3 ± 0.9 | 1.3 ± 0.9 | 0.870 |
| Glomerular filtration rate (ml/min) | 48.6 ± 23.7 | 48.2 ± 23.6 | 49.5 ± 23.9 | 0.247 |
| Glomerular filtration rate <45 ml/min | 916 (48.8) | 621 (49.4) | 295 (47.4) | 0.313 |
| Dialysis | 44 (2.3) | 29 (2.3) | 15 (2.4) | 0.890 |
| Albumin (mg/dl) | 3.5 ± 0.8 | 3.5 ± 0.8 | 3.6 ± 0.8 | 0.059 |
| Albumin <4 mg/dl | 659 (35.1) | 443 (35.3) | 216 (37.4) | 0.644 |
| Hemoglobin (mg/dl) | 11.6 ± 1.6 | 11.5 ± 1.6 | 11.6 ± 1.6 | 0.387 |
| Moderate anemia ∗ | 906 (48.9) | 605 (49.0) | 301 (48.6) | 0.883 |
| Active infective endocarditis | 1 (0.1) | 1 (0.1) | 0 (0.0) | 0.669 |
| Previous acute myocardial infarction | 333 (17.7) | 229 (18.2) | 104 (16.7) | 0.428 |
| Unstable angina pectoris | 53 (2.8) | 40 (3.2) | 13 (2.1) | 0.174 |
| Chronic obstructive pulmonary disease | 525 (27.9) | 363 (28.9) | 162 (26.0) | 0.194 |
| Oxygen dependency | 117 (6.2) | 80 (6.4) | 37 (5.9) | 0.713 |
| Pulmonary hypertension | 340 (18.1) | 226 (18.0) | 114 (18.3) | 0.862 |
| Neurologic dysfunction | 128 (6.8) | 84 (6.7) | 44 (7.1) | 0.785 |
| Chronic liver disease | 58 (3.1) | 38 (3.0) | 20 (3.2) | 0.827 |
| Active neoplastic disease | 71 (3.8) | 48 (3.8) | 23 (3.7) | 0.882 |
| Peripheral arteriopathy | 500 (26.6) | 334 (26.6) | 166 (26.7) | 0.995 |
| Previous cardiac surgery | 320 (17.0) | 209 (16.6) | 111 (17.8) | 0.503 |
| Porcelain aorta | 157 (8.4) | 104 (8.3) | 53 (8.5) | 0.866 |
| Difficult thoracic approach | 51 (2.7) | 32 (2.5) | 19 (3.1) | 0.525 |
| Frailty score † | ||||
| 1 | 377 (20.1) | 250 (19.9) | 127 (20.4) | 0.794 |
| 2 | 404 (21.5) | 270 (21.5) | 134 (21.5) | 0.982 |
| 3 | 55 (2.9) | 36 (2.9) | 19 (3.1) | 0.820 |
| 2 or 3 | 459 (24.4) | 306 (24.4) | 153 (24.6) | 0.911 |
| Previous percutaneous coronary intervention | 500 (26.6) | 328 (26.1) | 172 (27.7) | 0.453 |
| Previous balloon aortic valvuloplasty | 269 (14.3) | 189 (15.0) | 80 (12.9) | 0.205 |
| Critical preoperative state ‡ | 88 (4.7) | 57 (4.5) | 31 (4.9) | 0.653 |
| New York Heart Association class | ||||
| III | 696 (37.1) | 631 (50.2) | 338 (54.3) | 0.094 |
| IV | 255 (13.6) | 176 (14.0) | 79 (12.7) | 0.435 |
| III or IV | 1,224 (65.2) | 807 (64.3) | 417 (67.0) | 0.232 |
∗ Defined as baseline hemoglobin level <11.0 g/dl for both men and women (World Health Organization definition ).
† Defined according to the Geriatric Status Scale.
‡ Defined as ventricular tachycardia, ventricular fibrillation, or aborted sudden death, preoperative cardiac massage, preoperative ventilation, preoperative inotropes or intra-aortic balloon pump, or preoperative acute renal failure (anuria or oliguria <10 ml/hour).
| Variable | Overall (n = 1,878) | Development (n = 1,256) | Validation (n = 622) | p Value |
|---|---|---|---|---|
| No. of coronary arteries narrowed | 575 (30.6) | 391 (31.1) | 184 (29.6) | 0.493 |
| 1 | 336 (17.9) | 229 (18.2) | 107 (17.2) | 0.584 |
| 2 | 109 (5.8) | 75 (5.9) | 34 (5.5) | 0.660 |
| 3 | 130 (6.9) | 87 (6.9) | 43 (6.9) | 0.991 |
| Mitral valve regurgitation | ||||
| Mild | 1,000 (53.2) | 662 (52.7) | 338 (54.3) | 0.504 |
| Moderate | 490 (26.1) | 329 (26.2) | 161 (25.9) | 0.886 |
| Severe | 51 (2.7) | 32 (2.5) | 19 (3.1) | 0.525 |
| Moderate or severe | 541 (28.8) | 361 (28.7) | 180 (28.9) | 0.929 |
| LVEF (%) | 51.9 ± 12.4 | 52.1 ± 12.5 | 51.5 ± 12.1 | 0.363 |
| LVEF <40% | 311 (16.6) | 205 (16.3) | 106 (17.0) | 0.698 |
| Aortic valve pattern | ||||
| Valve area (cm 2 ) | 0.6 ± 0.3 | 0.6 ± 0.3 | 0.7 ± 0.3 | 0.192 |
| Peak gradient (mm Hg) | 80.3 ± 22.7 | 80.7 ± 23.1 | 80.1 ± 22.6 | 0.565 |
| Mean gradient (mm Hg) | 49.3 ± 15.0 | 49.5 ± 15.1 | 49.2 ± 15.0 | 0.698 |
| Mean gradient <40 mm Hg | 397 (21.1) | 266 (21.2) | 131 (21.1) | 0.948 |
Multivariate logistic regression analysis of significant predictors selected by univariate analysis was performed to identify predictors of 30-day mortality and to estimate odds ratios and 95% confidence intervals. To avoid overfitting data and select the best subset of risk factors, a bootstrap method was used, in which a stepwise logistic regression analysis (entry and exit criteria p = 0.05 and p = 0.10, respectively) was repeated for each of 1,000 bootstrap samples, the list of variables selected being stored. For variables selected in at least 90% of the samples, the odds ratio of 30-day mortality was calculated in a multivariate logistic regression model. Further processing with clinically relevant predictors of TAVR 30-day mortality selected in at least 50% of the samples was attempted to refine the adequacy of the model. Each of the variables in the final model was assigned an integer weighted score of 2 proportional to the multivariate odds ratio for 30-day mortality. The final OBSERVANT model was evaluated in terms of discrimination and calibration in the development and validation data sets. In addition, the performance of the logistic EuroSCORE was compared with that of the OBSERVANT model in the validation data set.
Discrimination is the probability that the score will assign higher values of risk to patients who were dead at 30 days compared with those who were not. It was measured with the C statistic, which ranges from 0.50 (no discrimination) to 1.0 (perfect discrimination). In addition, discrimination was assessed in the validation data set using the Somers’ D xy rank correlation between predicted probabilities and observed responses (when D xy = 0, the model is making random prediction, when D xy = 1, the predictions are perfectly discriminating). Calibration evaluates the degree of correspondence between the estimated probabilities produced by a model and the actual observation. It was measured by the Hosmer-Lemeshow test (with p = NS indicating goodness of fit) and by generating calibration plots that visually compare the prediction with the observed probability to evaluate bias or nonconstancy of variance across different levels of predicted risk, defined by tertiles. Global accuracy was tested calculating the Brier score (quadratic difference between predicted probability and observed outcome for each patient), an overall performance measure that consists of only positive values ranging from 0 (perfect prediction) to 1 (worst possible prediction), with lower scores indicating greater accuracy.
Clinical, demographic, and procedural characteristics were compared between the development and validation data sets using Student’s unpaired t tests for continuous variables (expressed as mean ± SD) and chi-square tests or Fisher’s exact tests for categorical variables (expressed as count and percentage). For all analyses, a 2-sided p <0.05 was considered statistically significant. All data were processed using the Statistical Package for Social Sciences, version 20 (IBM SPSS, Chicago, Illinois).
Results
Of the 1,911 patients who underwent TAVR in the OBSERVANT registry, 33 (1.7%) were excluded because of lack of follow-up data at 30 days. A total of 1,256 patients (67%) were assigned to the development data set. Baseline demographic, clinical, angiographic, and echocardiographic characteristics of the study population are listed in Tables 1 and 2 . There were no significant differences in baseline characteristics between the development and validation data sets. At 30-day follow-up, 77 patients (6.1%) and 37 patients (5.9%) had died in the development and validation data sets, respectively. Statistically significant univariate predictors of 30-day mortality in the development data set are listed in Table 3 . The multivariate model of predictors of 30-day mortality was obtained using data for all 1,165 patients with no missing covariate values and included 70 of 77 patients (90.9%) who died at 30 days. Five predictors (diabetes, GFR <45 ml/min, pulmonary hypertension, critical state, and New York Heart Association class 4) were selected in at least 90% of the bootstrap samples by multivariate analysis as independent predictors of 30-day mortality ( Table 3 ). Because LVEF <40% and history of previous balloon aortic valvuloplasty (being selected in >60% of bootstrap samples) were previously shown to be significant predictors of mortality at 30 days after TAVR, they were forced into the final model and found to reduce the −2 log likelihood statistic from 482,435 to 479,718 (p <0.001), the chi-square from 5,432 (Hosmer-Lemeshow p value = 0.363) to 3,995 (Hosmer-Lemeshow p value = 0.677), and to increase the C statistic from 0.71 to 0.73, indicating that the presence of these variables in the final risk model improved its adequacy, reason for which they were retained.
| Variable | Patients (%) | 30-Day Mortality (%) | Odds Ratio | 95% Confidence Interval | p Value | Integer Score Assigned |
|---|---|---|---|---|---|---|
| Univariate predictors | ||||||
| GFR <45 ml/min | 49.4 | 8.9 | 3.12 | 1.81–5.38 | <0.001 | — |
| Albumin <4 mg/dl | 35.3 | 9.5 | 2.80 | 1.64–4.81 | <0.001 | — |
| New York Heart Association class IV | 14.0 | 11.9 | 2.45 | 1.46–4.21 | 0.001 | — |
| Pulmonary hypertension | 18.0 | 11.1 | 2.37 | 1.43–3.93 | 0.001 | — |
| Moderate or severe mitral regurgitation | 28.7 | 10.0 | 2.31 | 1.45–3.68 | <0.001 | — |
| Critical preoperative state | 4.5 | 17.5 | 2.29 | 1.17–4.46 | 0.016 | — |
| Frailty score 2 or 3 | 24.4 | 9.8 | 2.09 | 1.30–3.37 | 0.003 | — |
| Previous balloon aortic valvuloplasty | 15.0 | 10.6 | 2.09 | 1.23–3.57 | 0.007 | — |
| LVEF <40% | 16.3 | 10.2 | 2.01 | 1.19–3.40 | 0.009 | — |
| Moderate anemia | 49.0 | 7.8 | 1.81 | 1.12–2.93 | 0.016 | — |
| Diabetes mellitus | 26.8 | 8.6 | 1.71 | 1.06–2.76 | 0.028 | — |
| Multivariate predictors | ||||||
| GFR <45 ml/min | — | — | 3.11 | 1.76–5.51 | <0.001 | 6 |
| Critical preoperative state | — | — | 2.26 | 1.01–5.05 | 0.045 | 5 |
| New York Heart Association class IV | — | — | 2.02 | 1.12–3.67 | 0.020 | 4 |
| Pulmonary hypertension | — | — | 1.91 | 1.11–3.30 | 0.020 | 4 |
| Diabetes mellitus | — | — | 1.90 | 1.13–3.18 | 0.015 | 4 |
| Previous balloon aortic valvuloplasty ∗ | — | — | 1.44 | 0.78–2.67 | 0.24 | 3 |
| LVEF <40% ∗ | — | — | 1.41 | 0.78–2.55 | 0.25 | 3 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
