Predictors of Poor Outcomes After Transcatheter Aortic Valve Replacement: Results from the PARTNER Trial

Abstract

Background

Transcatheter Aortic Valve Replacement (TAVR) has emerged as a less invasive option for valve replacement for high-risk patients with severe aortic stenosis. However, not all patients derive a mortality or quality of life (QoL) benefit from TAVR. As such, we sought to build and validate a prediction model to identify patients at high risk for a poor outcome after TAVR, using a novel definition of outcome that integrates QoL with mortality.

Methods and Results

We examined QoL and mortality outcomes among 2137 patients who underwent TAVR in the PARTNER randomized trial or the associated continued access registry. QoL was assessed using the Overall Summary Scale of the Kansas City Cardiomyopathy Questionnaire (KCCQ-OS, range 0-100; higher=better) at baseline and at 1, 6, and 12 months. A poor 6-month outcome—defined as death, KCCQ-OS <45, or a decrease in KCCQ-OS by ≥10 points compared with baseline—occurred in 704 patients (33%). A multivariable model was constructed using a split-sample design to identify a parsimonious set of covariates to identify patients at high-risk for poor outcome. The resulting model demonstrated adequate discrimination (c-index=0.66) and good calibration with the observed data and performed similarly in the separate validation cohort. Based on pre-procedure characteristics alone, the model identified 211 patients (10% of the population) with a ≥50% likelihood of a poor outcome after TAVR. These individuals were more likely to have low body weights, low mean aortic valve gradients, oxygen-dependent lung disease, and poor baseline functional and cognitive status. A second model that explored predictors of poor outcome at 1 year identified similar predictors as the original model and was able to identify 1102 patients (52%) with ≥50% likelihood and 178 (8%) with ≥70% likelihood of a poor 1-year outcome after TAVR

Conclusions

Using a large, multicenter cohort of patients undergoing TAVR, we have developed and validated predictive models that can identify patients at high-risk for a poor outcome after TAVR. These models may help guide treatment choices and offer patients realistic expectations of outcomes based on their presenting characteristics.

Among patients with severe aortic stenosis, transcatheter aortic valve replacement (TAVR) has emerged as a less invasive option for aortic valve replacement and offers substantial reductions in mortality and improvement in quality of life (QoL) compared with medical therapy^1-2 and similar long-term outcomes to surgical valve replacement.^3-4 With increasing experience, however, it has become evident that some patients do not improve functionally or live longer following TAVR. For example, in the Placement of AoRTic TraNscathetER Valve (PARTNER) Trial, approximately 1 in 4 patients who were treated with TAVR were dead at 1 year.^{1, 3} Furthermore, there were a number of patients who received TAVR who, while alive at 1 year, continued to have very poor QoL after TAVR, with significant heart failure symptoms and functional limitation.^{2, 4}

In light of these observations, there has been a recognition by practitioners,^5-6 regulators,⁷ and third party payers⁸ that TAVR should not be offered to patients in whom valve replacement would not be expected to positively impact either their survival or quality of life. To date, however, there has been little guidance as to how best to identify these patients. To address this need, several efforts have focused on predicting mortality after TAVR, with the goal of trying to identify patients at high risk for poor outcomes.^9-10 However, in this elderly population of patients, often with extensive comorbidity and impaired health status, it is likely that prolonged survival alone (without improved QoL) would not be viewed as a desirable outcome. Therefore, any definition of a successful outcome of TAVR (and conversely, of a poor outcome) must consider both survival and QoL. The goal of this study was to build and validate a prediction model to prospectively identify patients at high risk of a poor outcome (including death or very poor QoL) after TAVR.

METHODS

Study Population and Protocol

The study population was derived from patients with severe symptomatic AS who were enrolled in Cohort A or Cohort B of the PARTNER trial or in the associated non-randomized continued access PARTNER registry. Enrolled patients in both the trial and the registry had severe aortic stenosis (aortic valve area of <0.8 cm² with either a mean aortic valve gradient ≥40 mmHg or a peak aortic jet velocity ≥4.0 m/s); New York Heart Association (NYHA) class II or greater heart failure symptoms; and high surgical risk based on the Society for Thoracic Surgeons (STS) mortality risk score and other factors.^{1, 3} Patients with severe (4+) mitral or aortic regurgitation were excluded. Eligible patients were classified into 2 cohorts: Cohort A patients were high-risk but suitable for surgical aortic valve replacement³ whereas Cohort B patients were deemed ineligible for cardiac surgery due to coexisting medical or anatomic conditions associated with a predicted probability of perioperative death or permanent disability ≥50%.¹ In the randomized trial, Cohort A patients were randomized to surgical AVR or TAVR, and Cohort B patients were randomized to medical therapy or TAVR. For this study, we included only patients who received TAVR. After enrollment in the PARTNER trial was complete but before commercial approval of the Edwards-Sapien valve, a limited number of patients who would have been eligible for the PARTNER randomized trial were allowed to receive TAVR, via either the transfemoral or transapical route as clinically indicated, as part of a continued access registry. Inclusion and exclusion criteria, study protocols, and follow-up procedures were identical in the registry as in the randomized trial.

Patients were assessed for clinical factors and QoL/health status at baseline and at 1, 6, and 12 months after randomization. Baseline QoL/health status questionnaires were administered before randomization (or study registration for the continued access registries), and follow-up questionnaires were administered during in-person visits to the enrolling centers or by mail. The study was approved by the institutional review board at each participating site, and all patients provided written informed consent for baseline and follow-up assessments.

Health Status Data

Disease-specific health status (i.e., symptoms, functional status, and QoL) was assessed with the Kansas City Cardiomyopathy Questionnaire (KCCQ),¹¹ a 23-item self-administered questionnaire that assesses specific health domains pertaining to heart failure that yields an overall summary scale, which was the primary health status outcome for our study. Values for the KCCQ overall summary score (KCCQ-OS) range from 0 to 100 with higher scores indicating less symptom burden and better QoL. Linguistically and culturally validated translations of the KCCQ were provided to non-English speakers. Previous studies have suggested that KCCQ-OS scores correlate roughly with New York Heart Association Functional Class as follows: Class I ≈ KCCQ-OS 75-100; Class II ≈ 60-74; Class III ≈ 45-59; and Class IV ≈ 0-44.¹² Among outpatients with heart failure, small, moderate, and large clinical improvements/deteriorations, as rated by treating physicians, correspond to changes in the KCCQ-OS of approximately 5, 10, and 20 points, respectively.¹² The KCCQ has undergone extensive reliability and validity testing in various heart failure populations^{11, 13-14} as well as in patients with severe aortic stenosis.¹⁵

Functional status was assessed objectively by means of a 6-minute walk test (6MWT). If a patient could not perform the test, the value for the 6MWT distance was set to zero. Generic health status was assessed with the physical and mental summary scores of the Medical Outcomes Study Short-Form 12 (SF-12) Health Survey.¹⁶ Since the SF-12 physical summary score represents functional status, which was also assessed by means of the KCCQ and the 6-minute walk test, we did not include this as a potential predictor for poor outcome.

Definition of a Poor Outcome

While any single definition of a “poor outcome” after TAVR is a somewhat arbitrary construct, we have previously examined several potential definitions that combine mortality and QoL.¹⁷ For the purposes of this study, we used outcomes at 6 months after TAVR as the basis for our primary endpoint, as we believe that survival for at least 6 months with a reasonable QoL would be the minimum acceptable outcome for the procedure. As previously described, for our primary analysis, a poor outcome was defined as any of the following at 6 months after TAVR (Definition #1): 1) death; 2) KCCQ overall summary score <45; or 3) decrease of ≥10 points in the KCCQ-OS score from baseline to 6 months.^17-18

From a conceptual standpoint, this definition means that if a patient had a very poor QoL prior to TAVR, his or her QoL would have to improve to a minimum threshold (~NYHA Class III symptoms or better) in order to be considered an acceptable outcome. On the other hand, if a patient had a satisfactory QoL prior to TAVR, then survival for at least 6 months with a QoL that had not deteriorated substantially from baseline would be considered an acceptable result. This combined definition integrates the two potential benefits of TAVR, reduced mortality and improved QoL, and recognizes that patients that have good QoL at baseline may not improve symptomatically after TAVR but could still derive a mortality benefit (which would represents a clinically meaningful benefit of the procedure). In addition, we constructed an alternative, expanded definition of poor outcome (Definition #2) that included any of the following at 1 year after TAVR: 1) death; 2) KCCQ overall summary score <60; or 3) decrease of ≥10 points in the KCCQ-OS score from baseline to 1 year. This definition allows for a longer time frame of analysis and may be beneficial to patients and providers in conjunction with, or as an alternative to, Definition #1 depending on the patient’s goals of care.

Statistical Analysis

Demographics, cardiac and non-cardiac comorbidities, and echocardiographic variables were compared between patients who had a poor outcome and patients who had an acceptable outcome after TAVR using t-tests for continuous variables and chi square tests for categorical variables. We then randomly split the cohort of TAVR patients into 2 groups: 2/3 for model derivation and 1/3 for validation. In the derivation cohort, we developed a multivariable logistic regression model to predict a poor 6-month outcome (using Definition #1 as described above). Variables for the prediction model were selected from 25 candidate variables (Table 1). Baseline data had a high rate of completion, with an average of 0.55 missing data items per patient, which were imputed with a single imputation dataset using IVEware (Institute for Social Research, University of Michigan, Ann Arbor, MI).

Table 1. Baseline characteristics of patients with acceptable vs. poor outcomes according to Definition #1.

	All Patients n=2137	Acceptable Outcome n=1433	Poor Outcome n=704	P- Value
Age (y)	84.4 (7.2)	84.3 (7.2)	84.6 (7.0)	0.357
Male (%)	52.8	51.3	56.0	0.042
Coronary artery disease (%)	78.2	78.1	78.6	0.807
Cerebrovascular disease (%)	26.6	26.5	26.8	0.872
Carotid disease (%)	27.0	27.6	25.7	0.364
Peripheral vascular disease (%)	42.2	42.0	42.6	0.791
Diabetes mellitus (%)	36.6	38.4	33.1	0.017
Major arrhythmia* (%)	50.9	47.5	57.8	<0.001
Creatinine (mg/dL)	1.31 (0.48)	1.29 (0.46)	1.36 (0.51)	0.002
Hemoglobin (g/dL)	11.7 (1.5)	11.8 (1.5)	11.6 (1.5)	0.033
Mean arterial pressure (mmHg)	87.1 (13.1)	87.2 (12.8)	87.0 (13.4)	0.656
Body mass index (kg/m2)	26.9 (6.3)	27.1 (6.1)	26.6 (6.8)	0.093
Oxygen-dependent lung disease (%)	10.9	9.3	14.3	<0.001
Pulmonary hypertension (%)	40.0	39.2	41.5	0.316
Mitral regurgitation (>1+) (%)	25.3	24.8	26.3	0.473
Aortic regurgitation (>1+) (%)	10.5	10.8	9.9	0.536
Mean aortic gradient (mmHg)	43.8 (14.3)	45.1 (14.3)	41.2 (14.0)	<0.001
Ejection fraction (%)	52.0 (13.1)	52.4 (13.0)	51.2 (13.4)	0.053
Stroke volume (mL/beat)	64.5 (21.1)	65.1 (20.0)	63.2 (23.1)	0.053
Mini-Mental Status Exam score	27.4 (3.0)	27.5 (2.9)	27.1 (3.0)	0.008
6-Min Walk Test (% able to perform)	65.0	70.1	54.8	<0.001
6-Min Walk Test distance (m)	110.8 (118.1)	123.9 (122.2)	84.0 (104.3)	<0.001
KCCQ Overall Summary score	42.0 (21.7)	43.8 (20.8)	38.4 (23.0)	<0.001
SF-12 Mental Summary score	47.6 (11.2)	48.3 (11.0)	46.2 (11.5)	<0.001
STS mortality risk score	11.5 (4.4)	11.4 (4.5)	11.9(4.1)	0.005

STS, Society of Thoracic Surgeon

^*Defined as a history of atrial fibrillation or flutter, supraventricular tachycardia, ventricular arrhythmias, or high-degree AV block Values in parentheses represent standard deviations

Harrell’s backward selection strategy was used to select a parsimonious set of variables for the final model.¹⁹ The contribution of each covariate in the multivariable model was ranked by F-value, and variables with the smallest contribution to the model were sequentially eliminated. This iterative process continued until further variable elimination led to a greater than 5% loss in model prediction, as compared with the initial model. The remaining covariates comprised the final parsimonious model and explained over 95% of the variance of the full model. This selection strategy supports inclusion of only variables that provide incremental prognostic value, avoids over-fitting, and maximizes the potential clinical usefulness of the model.¹⁹ Non-linear spline terms were considered for all continuous variables. Model discrimination was assessed with the c-index, and model calibration was assessed by plotting deciles of predicted risk against the observed event rate and comparing the regression line to theline of unity. These analyses were performed in both the derivation and validation cohorts and were repeated using Definition #2 of a poor outcome.

As sensitivity analyses, to further explore the associations of the covariates with the outcome, we applied the model to each of the component endpoints to explore its ability to predict: 1) death or 2) poor QoL or QoL decline (among survivors). Second, to ensure that our results were not heavily influenced by the consequences of early complications (which might be considered outliers), we conducted a sensitivity analysis excluding patients who had a major peri-procedural complication (i.e., major stroke, bleeding complication, vascular complication, or surgical aortic valve replacement within 7 days of the TAVR procedure). Third, we applied the model separately to patients according to access site (transfemoral or transapical) to ensure that model performance did not vary markedly by site of valve delivery. Finally, we constructed an alternative model including the KCCQ instead of the 6MWT, for ease of implementation. All statistical analyses were performed with the use of SAS software, version 9.2 (SAS Institute, Inc., Cary, NC).

RESULTS

Patient Population

Of the 1057 patients with severe aortic stenosis who were enrolled in the PARTNER randomized trial, 527 patients underwent TAVR (Figure 1; 348 in Cohort A, 179 in Cohort B), and an additional 2068 patients underwent TAVR in the non-randomized continued access registry via the transfemoral (n=1503) or transapical route (n=1092). Of these, 400 died within 6 months of TAVR. Among the 2195 who survived 6 months, KCCQ data were available for 1737 (79%). Thus, our analytic population included 2137 patients who underwent TAVR and were either dead or were assessed with the KCCQ at 6 months after their procedure. Most baseline characteristics were similar for patients with vs. without complete KCCQ data. However, patients with missing KCCQ data had lower baseline 6 minute walk distances (85 vs. 119 m, p<0.001), lower body mass indices (26.3 vs. 27.2, p=0.008), and lower hemoglobin levels (11.6 vs. 11.8, p=0.026).

Figure 1. Patient flow.

The mean age of the analytic population was 84 years, and 53% were male (Table 1). The mean aortic valve gradient was 44 mmHg, and 93% were classified as NYHA Class III-IV. At 6-month follow-up, among the 2137 study patients, 400 (19%) had died, 260 (12%) had a very poor QoL (i.e., KCCQ-OS <45), and an additional 44 (2%) had worsened QoL (i.e., decrease in KCCQ-OS of ≥10 points) (Figure 1). Thus, a total of 704 patients (33%) had a poor 6-month outcome. The baseline characteristics of patients with an acceptable vs. poor outcome after TAVR are summarized in Table 1.

Model Development

After backward stepwise elimination, the final predictive model consisted of 10 covariates, which are summarized in Table 2 (model parameters in Supplemental Table 1). According to F-values, the distance walked on the 6MWT had the strongest association with poor outcome after TAVR, with each additional 10 meters walked associated with a 3% lower risk of poor outcome (adjusted OR 0.97, 95% CI 0.96-0.98). Higher mean aortic valve gradients were also strongly associated with a lower risk of poor outcome; each 10 mmHg increase in mean gradient was associated with a 18% lower odds of a poor outcome (adjusted OR 0.82, 95% CI 0.75-0.89). Other baseline factors associated with a poor outcome after TAVR included oxygen-dependent chronic lung disease, renal dysfunction, decreased cognition, and cardiac arrhythmias (defined as a history of atrial fibrillation or flutter, supraventricular tachycardia, ventricular arrhythmia, or high-degree AV block). The c-index of the model was 0.66, indicating adequate discriminatory capacity.

Table 2. Association of pre-procedure factors with poor outcomes after TAVR.

	Poor Outcome at 6 Months		Poor Outcome at 1 Year
	(Definition #1)		(Definition #2)
Predictor	OR (95% CI)	P-Value	OR (95% CI)	P-Value
Male sex	1.23 (0.96-1.57)	0.097	1.22 (0.97-1.53)	0.097
Diabetes mellitus	0.82 (0.63-1.06)	0.130
Major arrhythmia	1.29 (1.02-1.63)	0.036	1.13 (0.91-1.40)	0.280
Serum creatinine (per 1 mg/dL)	1.32 (1.03-1.70)	0.028	1.41 (1.11-1.79)	0.005
Mean arterial pressure (per 1 mmHg)	1.01 (1.00-1.02)	0.209
Body mass index (per 1 kg/m2)	0.98 (0.96-1.00)	0.104	1.00 (0.98-1.02)	0.791
Oxygen-dependent lung disease	1.77 (1.23-2.54)	0.002	1.80 (1.25-2.61)	0.002
Mean aortic valve gradient (per 10 mmHg)	0.82 (0.75-0.89)	<0.001	0.84 (0.77-0.90)	<0.001
Mini-Mental Status Exam (per 1 point)	0.96 (0.92-1.00)	0.036	0.94 (0.90-0.97)	0.001
6-Minute Walk Test distance (per 10 m)	0.97 (0.96-0.98)	<0.001	0.97 (0.96-0.98)	<0.001

The observed versus predicted risk of poor outcome after TAVR within risk deciles is shown in Figure 2a. In general, the model demonstrated good calibration with the observed outcomes with an intercept of −0.01 (p-value for difference from 0=0.806), a slope of 1.03 (p-value for difference from 1=0.793), and an R² of 93%.

Figure 2. Calibration plots for prediction of poor outcome at 6 months after TAVR (Definition #1) in the derivation cohort (a) and validation cohort (b).

(a) Intercept of −0.01 (SE 0.04; p-value [for difference from 0]=0.806), a slope of 1.03 (SE 0.10; p-value [for difference from 1]=0.793), and R² of 93%. (b) Intercept of 0.07 (SE 0.06; p-value [for difference from 0]=0.277), a slope of 0.75 (SE 0.18; p-value [for difference from 1]=0.203), and R² of 69%.

Model Validation

The baseline characteristics of the derivation vs. validation cohort are presented in Supplemental Table 2. Patients were similar in terms of baseline demographic and clinical characteristics, except that patients in the validation cohort had somewhat worse QoL at baseline (KCCQ-OS: 40 vs. 43, p=0.007). The model performed well in the validation cohort, with adequate discrimination (c-index=0.64) and reasonable calibration (Figure 2b), with an intercept of 0.07 (p-value for difference from 0=0.277) and a slope of 0.75 (p-value for difference from 1=0.203).

Sensitivity Analyses

When the same covariates were applied to the individual components of the composite endpoint (i.e., mortality and poor QoL), there was general concordance of the 2 models such that the direction of association of each covariate with the 2 endpoint components was similar (although the magnitude of the association varied; Supplemental Table 3). When patients with major peri-procedural complications (n=194) were excluded from the analytic population, the model c-index was unchanged at 0.66 and the calibration remained good (Supplemental Figure 1). When the transfemoral and transapical cohorts were analyzed separately (baseline characteristics of 2 groups, the model demonstrated similar discrimination within each access site group (c-index 0.65 vs. 0.66), although the calibration was worse within the transapical group, with somewhat higher rates of poor outcomes than predicted (Supplemental Figure 2). Finally, when we substituted the KCCQ-OS for the 6MWT distance (for ease of implementation), the discrimination decreased slightly (c-index 0.64; Supplemental Table 4).

Outcomes According to Predicted Risk

In order to better understand the ability of the model to inform clinical decisions, we stratified patients into 3 groups according to their predicted risk of a poor outcome: low risk (<25%; n=612), intermediate risk (25% to <50%; n=1314), and high risk (≥50%; n=211). The baseline characteristics of the 3 groups are shown in Table 3. Compared with patients at low risk of a poor outcome, high-risk patients had less diabetes, lower body mass indices, worse kidney function, more frequent oxygen-dependent lung disease, lower mean aortic valve gradients, worse cognitive function, worse functional status, and worse QoL at baseline. In the 6 months after TAVR, 31% of high-risk patients died, and an additional 24% had very poor QoL or a decline in QoL. In contrast, among intermediate risk patients, 17% had died, and an additional 20% had a very poor QoL or a decline in QoL, while among low-risk patients, only 8% had died and 10% had a very poor QoL or a decline in QoL (p<0.001; Figure 3a).

Table 3. Patient characteristics according to predicted risk of poor outcome (Definition #1).

	Predicted Risk*
	Low n=612	Intermediate n=1314	High n=211	P-Value
Age (y)	83.4 (8.2)	85.0 (6.6)	84.0 (7.0)	<0.001
Male (%)	43.0	53.3	78.2	<0.001
Coronary artery disease (%)	77.6	78.4	79.1	0.878
Cerebrovascular disease (%)	25.3	27.1	27.5	0.685
Carotid disease (%)	30.1	25.8	25.1	0.119
Peripheral vascular disease (%)	42.5	42.8	37.9	0.410
Diabetes mellitus (%)	45.1	34.2	27.5	<0.001
Major arrhythmia (%)	29.1	57.3	74.4	<0.001
Creatinine (mg/dL)	1.17 (0.41)	1.32 (0.46)	1.60 (0.60)	<0.001
Hemoglobin (g/dL)	11.9 (1.5)	11.7 (1.5)	11.5 (1.6)	0.018
Mean arterial pressure (mmHg)	86.6 (12.2)	87.4 (13.2)	86.9 (13.7)	0.420
Body mass index (kg/m2)	28.0 (7.4)	26.6 (5.9)	25.6 (5.2)	<0.001
Oxygen-dependent lung disease (%)	2.5	9.5	44.5	<0.001
Pulmonary hypertension (%)	35.8	41.6	42.2	0.043
Mitral regurgitation (>1+) (%)	20.9	26.6	30.3	0.006
Aortic regurgitation (>1+) (%)	9.2	11.5	8.5	0.181
Mean aortic gradient (mmHg) (%)	53.0 (15.2)	41.3 (12.0)	32.7 (9.7)	<0.001
Ejection fraction (%)	55.5 (11.2)	51.2 (13.3)	46.6 (15.0)	<0.001
Stroke volume (mL/beat)	68.9 (24.0)	62.9 (19.4)	61.1 (20.1)	<0.001
Mini-Mental Status Exam score	28.1 (2.1)	27.3 (2.9)	25.3 (4.3)	<0.001
6-Min Walk Test (% able to perform)	89.2	61.2	19.0	<0.001
6-Min Walk Test distance (m)	208.9 (128.9)	80.7 (88.6)	13.2 (33.1)	<0.001
KCCQ Overall Summary score	50.2 (21.6)	40.3 (20.8)	29.4 (18.5)	<0.001
SF-12 Mental Summary score	49.6 (10.9)	47.3 (11.2)	43.9 (11.1)	<0.001
STS mortality risk score	10.7 (4.3)	11.8 (4.3)	12.6 (4.6)	<0.001
Outcomes at 6 Months
Dead (%)	8.8	20.3	37.4	<0.001
Poor Outcome (%)	17.8	36.5	55.0	<0.001

KCCQ-OS, Kansas City Cardiomyopathy Questionnaire Overall Summary; STS, Society of Thoracic Surgeons

^*Low Risk: <25%, Intermediate Risk: 25 to <50%, High Risk: ≥50%

Figure 3. Outcomes of patients by their predicted risk of poor outcome at baseline at 6 months (Definition #1; Panel a) and 1 year (Definition #2; Panel b).

Definition #2 for Poor Outcome

For Definition #2, the analytic population consisted of 2130 patients who had available QOL data or had died by 1 year follow-up. Of these, 1073 (50%) had a poor 1-year outcome based on either death (n=558), poor QoL (n=503) or QoL decline (n=12). After backward stepwise elimination, the final predictive model consisted of 8 covariates, which are summarized in Table 2 (model parameters in Supplemental Table 1). In general, the predictors were similar to those for Definition #1—both in the variables that were included and their magnitude of association, except that the alternative model did not include mean arterial pressure or diabetes mellitus.

The c-index of the model was again 0.66, but the calibration of the model was better than the 6-month model (Supplemental Figure 3a) with an intercept of −0.01 (p-value for difference from 0=0.875), a slope of 1.01 (p-value for difference from 1=0.871), and an R² of 97%. The model also performed well in the validation cohort, with a c-index of 0.62 and good calibration (Supplemental Figure 3b). We then divided the patients into 4 groups based on their predicted risk of poor outcome at 1 year: low risk (<25%; n=65), intermediate risk (25% to <50%; n=963), high risk (50% to <70%; n=924), and very high risk (≥70%, n=178). At 1 year after TAVR, 50% of very high-risk patients had died, and an additional 23% had poor QoL or a decline in QoL. In contrast, among low-risk patients, only 17% had died and 12% had a poor QoL or a decline in QoL (p<0.001; Figure 3b).

DISCUSSION

Among patients with severe aortic stenosis, TAVR is highly effective at relieving the hemodynamic obstruction and can lead to excellent outcomes in many patients. However, there are some patients who do not achieve either a survival or functional benefit from the intervention. In this study, we have identified a set of covariates and an associated prediction model that can prospectively identify patients at high-risk for poor outcomes after TAVR. Within the PARTNER trial, we found that one-third of patients had a poor outcome at 6 months using Definition #1 (conservative definition) and one-half of patients had a poor outcome at 1 year using Definition #2 (expanded definition). The most important predictors of poor outcomes, using either definition, were poor functional capacity (as assessed with the distance walked on the 6-minute walk test) and lower mean aortic valve gradients. Other important predictors were oxygen-dependent lung disease, renal dysfunction, and poorer baseline cognitive function. Of note, the STS mortality risk score was not a predictor in either model. Taken together in a validated statistical model, these factors allowed us to predict each patient’s probability of a poor outcome after TAVR. Such information could help inform clinical decision making for patients and their physicians when considering TAVR and could also help patients and their families set realistic expectations when they choose to undergo TAVR.

While our definitions for poor outcome after TAVR are unique among prior studies, we believe that the optimal definition for a poor outcome after TAVR should reflect a failure to achieve the goals of the intervention and therefore must include both a mortality and a QoL component. Combining these two endpoints into a single definition can be challenging, however.¹⁷ Nonetheless, in this elderly population of patients with multiple comorbidities, inclusion of QoL as a component of the determination of a poor (or acceptable) outcome is critical, since improved QoL can be the primary treatment goal of many patients considering TAVR. We believe that the definitions of poor outcome that we considered reflect the treatment goals of patients considering TAVR, since patients who have minimal symptoms choose TAVR primarily to achieve an expected improvement in survival while patients with substantial symptom burden most likely undergo TAVR for an expected improvement in symptoms.

There have been a number of previous efforts to determine the predictors of poor outcome after TAVR. To date, all of these analyses have been restricted to identifying predictors of mortality or specific complications, such as stroke or kidney dysfunction.^{9, 20-22} Despite differences in the study populations and settings, some factors have been consistently found to be prognostically important among patients undergoing TAVR. In a single-center German study, older age, lower BMI, NYHA class IV symptoms, depressed left ventricular function, and higher estimated surgical risk score were independently predictive of 1-year mortality.²⁰ In a large, multicenter French registry, higher estimated surgical risk score, NYHA class III-IV symptoms, transapical approach, and post-implant aortic regurgitation were associated with reduced survival.⁹ In a multicenter Canadian study, chronic lung disease, chronic kidney disease, atrial fibrillation, and a subjective assessment of frailty were all associated with long-term mortality after TAVR.²² Finally, in a single-center Canadian study, male sex, chronic lung disease, and the 6-minute walk test distance were associated with 1-year mortality after TAVR.²³

While our model also included many of these same factors (body mass index, lung disease, kidney disease, arrhythmias, functional and cognitive decline [components of frailty]), our study offers several important advantages compared with these previous studies. First, our model was based on pre-procedural factors alone and therefore can be used prospectively to inform clinical decision making. Second, ours is one of the few studies that performed model validation, which provides greater assurance of reliability. Finally, ours is the first such study to incorporate patient-reported outcomes as an explicit component of the endpoint. By integrating QoL into our definition of a poor outcome, we believe that our model may be better aligned with the goals of many patients who are considering TAVR. Although further research is needed to understand how best to present the predicted risk of poor outcome to patients and physicians, this information could be invaluable to patients who are considering whether or not to undergo TAVR. With these additional insights, these models (or future iterations) could eventually be applied in routine clinical care in order to improve decisional quality, reduce anxiety associated with the treatment decision, and provide patients and their families realistic expectations of recovery.

Our study has several important limitations. First, our study population was derived from the PARTNER trial, which included patients with a mean STS mortality risk score of ~12%—much higher than most previous studies. Consequently, it is uncertain whether our model applies to moderate risk patients. Nonetheless, since many of the factors that were included in our final model had been previously identified as associated with poor outcomes even in moderate risk TAVR patients, there is reason to believe such extrapolation may be reasonable until future studies evaluating the performance of our model in lower risk patients are performed. Second, the discriminative ability of our model was only modest. While this level of discrimination is similar to other prediction models in cardiology,^24-25 there may be additional patient factors that are associated with poor outcomes after TAVR but were not available for inclusion in our models. For example, when the appropriate dataset is available, it will be particularly interesting to determine the incremental prognostic information gained from the addition of other markers of frailty, such as grip strength and gait speed, as such measures have been predictive of poor outcomes with other therapies.^26-27 Since TAVR is evolving rapidly, novel predictors of poor outcomes will certainly continue to be discovered. When such factors are identified in the future, their incremental predictive value can be tested with our model, thereby leading to further refinement and improvement in the model over time.

In conclusion, using a large, multicenter cohort of patients with severe, symptomatic aortic stenosis who underwent TAVR, we have created and validated 2 related prediction models for poor outcome after TAVR based on readily available pre-procedure patient characteristics. Using these models, it is possible to identify subsets of patients who are relatively unlikely to derive meaningful survival or quality of life benefit from TAVR. In the future, providing this information prospectively to patients could offer valuable guidance as patients and their families decide whether or not to undergo TAVR. Further studies are necessary to understand the incremental value of additional risk factors to the models’ discrimination and to understand how the model performs in lower risk TAVR populations.