Author + information
- aDivision of Cardiology, Lahey Hospital & Medical Center, Burlington, Massachusetts
- bHarvard Clinical Research Institute, Boston, Massachusetts
- cDivision of Cardiac Surgery, University of British Columbia, Vancouver, British Columbia
- dJohns Hopkins Bloomberg School of Public Health, Baltimore, Maryland
- ↵∗Reprint requests and correspondence:
Dr. Mathew R. Reynolds, Harvard Clinical Research Institute, 930 Commonwealth Avenue, Boston, Massachusetts 02215.
“Truth is one, paths are many.”
—Mahatma Gandhi (1)
Despite the great success of transcatheter aortic valve replacement (TAVR) in transforming the treatment of aortic stenosis, close to 25% of TAVR patients die within 1 year of their procedure (2) and others survive the intervention but remain with poor overall health status (3). Risk prediction models can help clinicians and patients understand the potential likelihood of such undesirable outcomes, information that may be valuable in choosing and planning optimal treatment pathways.
To address this need, a number of TAVR-specific risk prediction models have recently been developed using data from major clinical trial programs and multicenter registries (3–8). In this issue of the Journal, Arnold et al. (9) make an important addition to the burgeoning literature on this topic, by externally validating risk models previously developed using data from the PARTNER (Placement of Aortic Transcatheter Valves) I trial and its continued access registry, using newer data from the CoreValve US Extreme and High Risk trials and their associated continued access registries.
With the recent publication of several new TAVR risk models (7–9), the time seems ripe to ask: How do these models compare and contrast with each other? What are we learning from them?
Most of the previously published studies in this area have focused on short-term (in-hospital or 30-day) mortality (4,5,7,8); a few have additionally or alternatively examined 1-year mortality (6–8). All but 1 of the mortality risk models were developed using data from multicenter registries (4–7), whereas 1 was drawn from the CoreValve US trial program (8). Because registries include less narrowly selected patients than trials, registry-based risk models might be more representative of “typical” or real-world practice and might identify important predictive factors not observed in trials. It is unsurprising, then, that no 2 models have produced identical results. Nonetheless, some common themes are apparent.
The registry-based models of early mortality have consistently found that measures of disease or symptom severity as well as patient acuity and operative urgency are associated with increased risk. Thus, New York Heart Association functional class IV symptoms (4,5,8), “critical preoperative state” (4), and “acuity category” (8) stand out in several of these models, highlighting the well understood fact that patients with symptomatic, severe aortic stenosis fare much better when their valves are replaced electively, rather than waiting for debilitating symptoms or a moment of crisis.
As to be expected, all of the published models have shown that the risks of both early and 1-year mortality are increased in the presence of severe comorbidities and advanced age (>85 or 90 years). Both chronic kidney disease and chronic pulmonary disease—particularly when accompanied by the need for home oxygen—appear to increase risk markedly.
Recently, there has been growing interest in understanding, defining, and measuring the concept of frailty in the older age groups commonly considered for TAVR and exploring the relationship between those measures and subsequent outcomes. As frailty is a multidimensional concept, various ways have been suggested to measure it, some of which overlap considerably with traditional measures of functional status, disability, nutrition, and cognition. By whatever name, these factors do correlate strongly with outcomes after TAVR. For example, in a recent report from the CoreValve study database, among the factors most strongly associated with 30-day mortality were a lack of independence in activities of daily living—which seems to be encapsulated nicely by residency in an assisted living environment—and a serum albumin of <3.3 g/dl (8). Further, a history of recent falls was independently associated with 1-year mortality risk. The FRANCE 2 (French Aortic National CoreValve and Edwards) registry also reported an association between mortality and low body mass index (5).
What is most unique about the work of Arnold et al. (9) is a definition of poor outcome that looks beyond simple survival. In both the original PARTNER-based report and the current study, a novel definition of poor outcome was used, defined as death, very poor quality of life (QOL) (a value of <45 at 6 months or <60 at 12 months on the Kansas City Cardiomyopathy Questionnaire overall summary score), or a moderate worsening of QOL (a ≥10-point decrease in the Kansas City Cardiomyopathy Questionnaire summary score from baseline). This patient-centric definition of a poor outcome explicitly recognizes that, for many patients considering TAVR, the goal of the intervention is not merely to survive, but to feel and function better. It is noteworthy that at 1 year following TAVR, 50.8% of the patients in the CoreValve program experienced a poor outcome (death 30.2%, poor QOL 19.6%, QOL decline 1.0%) by this definition.
The previously developed “poor outcome” risk models described in the Arnold et al. (3) original paper included several of the same variables identified in 1 or more independent mortality risk models (oxygen-dependent lung disease, diabetes, male sex, low body mass index) along with standard measures of functional status (6-min walk test distance) and cognition (as assessed by the Mini-Mental State Examination). Additionally, the earlier work identified lower aortic valve gradients to be associated with higher risk of a poor outcome, suggesting that patients referred for TAVR with lower valve gradients may not be most symptomatic or disabled from their aortic stenosis but rather from other health conditions.
In the current study, the 6- and 12-month risk models developed using PARTNER trial data also performed well in the CoreValve study population, with similar moderate discrimination (c-indexes: 0.64 to 0.67) and excellent calibration, as seen in the earlier study, marking the first time that a TAVR risk model has been externally validated in an independent data set. Empirical research has shown that risk prediction models infrequently undergo external validation, and usually perform less well when this step is taken (10). The investigators are to be congratulated for this effort.
Additionally, Arnold et al. (9) carefully evaluated the incremental value of the various frailty measures available in the more richly developed CoreValve database to their risk model. They confirmed the previously reported associations between measures of frailty and outcome following TAVR, but found that adding these measures to their previous variables improved the calibration and discrimination of their models by very small amounts.
An important area of agreement between this and another recent CoreValve-based study (8) is that measures of frailty can be incorporated into TAVR risk models without the need for laborious formal assessments, such as grip strength or 5-m gait speed. In both studies, easily measured items such as impairment in activities of daily living or markers of catabolism or poor nutrition (unintentional weight loss, low albumin) provided as much or more prognostic information.
The “poor outcome” risk models, similar to others, have limitations. Despite including simplified “clinical” as well as “full” models, these risk assessments cannot quite be implemented rapidly at the bedside. Their discrimination is only moderate, in part because QOL outcomes are hard to predict and, possibly, in part because this model has only been studied in 2 somewhat similar clinical trial programs within a relatively narrow range of patients.
At this point, there does not appear to be a single TAVR risk model that perfectly serves all of the needs of all people who might wish to use one—but that is not a realistic goal. The information generated from independent groups developing different models is sometimes confirmatory and sometimes complementary. There will always be an interest in mortality risk models for high-risk interventions such as TAVR, and some are now available from large and representative patient samples. Arnold et al. have shown that with some additional work, we can make some reasonably accurate predictions about survival with improved QOL after TAVR. Work in this area is incomplete, but seems to be on the right track.
↵∗ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
Dr. Reynolds has served as a consultant to Edwards Lifesciences and Medtronic. Dr. Hong has reported that he has no relationships relevant to the contents of this paper to disclose. Deepak L. Bhatt, MD, MPH, served as Guest Editor-in-Chief, for this article. Michael Mack, MD, served as Guest Editor for this article.
- American College of Cardiology Foundation
- ↵Goodreads. Mahatma Gandi quotes. Available at: http://www.goodreads.com/quotes/29887-truth-is-one-paths-are-many. Accessed August 17, 2016.
- Arnold S.V.,
- Reynolds M.R.,
- Lei Y.,
- et al.
- Iung B.,
- Laouenan C.,
- Himbert D.,
- et al.
- Hermiller J.B. Jr..,
- Yakubov S.J.,
- Reardon M.J.,
- et al.
- Arnold S.V.,
- Afilalo J.,
- Spertus J.A.,
- et al.