Clinical Evaluations – Endpoints

Ori Ben-Yehuda, MD

Endpoint selection is of utmost importance and an essential part of trial design. Clearly, the endpoints should be clinically relevant and provide an opportunity to demonstrate the presence of a meaningful impact of the intervention. Cardiovascular study endpoints can be divided to “hard” endpoints such as mortality, myocardial infarction, and stroke; intermediate endpoints such as hospitalizations for heart failure; and “soft” endpoints which can include symptoms, quality of life measures (e.g., New York Heart Association [NYHA] class), and surrogate measures previously proven to correlate with clinically relevant endpoints (e.g., ejection fraction [EF]). Earlier transcatheter aortic valve replacement (TAVR) trials, such as the early PARTNER (1,2) or CoreValve (3) trials, compared TAVR to surgical or medical only treatments and included inoperable patients or those at high surgical risk. Understandably, the population in these trials had high rates of “hard” endpoints and statistically proven differences could be demonstrated with a reasonable sample size. Current and next phase TAVR trials are enrolling lower-risk patients, some asymptomatic or at an earlier stage of aortic valve disease. Since event rates may be lower and effect size smaller, a large sample size will be needed to prove benefit. The primary method to reduce sample size is to use “soft,” or surrogate, endpoints and/or to combine several endpoints to one composite endpoint with its inherent limitations (4-6). A somewhat different approach, combining endpoints in a hierarchical clinically intuitive way can be achieved by the Finkelstein and Schoenfeld method (7) or the Pocock Win Ratio (8).

In this section, we will review how innovative approaches to clinical trial design can inform future trials in aortic stenosis therapy.

Aortic stenosis (AS) is the most common type of valvular heart disease in the western world (9). Our current understanding of disease progression in AS relies on small datasets, some more than 50 years old. In the 1930s, the cardinal features of AS — angina pectoris, syncope, and heart failure (HF) — were recognized as indicating rapid deterioration after a long asymptomatic period (10-12). The seminal work of Ross and Braunwald, cited repeatedly to this day, matched different survival periods to the onset of angina, syncope, or HF, highlighting the importance of intervention once symptoms developed (13).  Earlier intervention in select patients is an area of emerging interest, tied to a more detailed understanding of the natural history of AS. Current guidelines (14-16) recommend aortic valve replacement in symptomatic patients with severe AS (Class I recommendation) or asymptomatic patients with severe AS and EF <50%. Asymptomatic patients may also be referred for intervention in the presence of a pathological stress test (Class IIa) or in patients undergoing other cardiac surgery (Class I for severe AS and Class IIa for moderate AS). Advances in medicine throughout the years and better management of risk factors, coupled with the introduction of balloon valvuloplasty in 1986 (17) and TAVR in 2002 (18) may alter the risk-benefit ratio and favor earlier intervention. While TAVR is a valid alternative to surgical aortic valve replacement (SAVR) in high- and intermediate-risk symptomatic patients (16), the question of whether the reduced risk of TAVR should affect the timing of AS intervention and expand the indication for intervention to asymptomatic and low-risk patients is being determined in ongoing trails detailed below.

Several trials are trying to assess the utility of TAVR in lower-risk populations than those studied thus far. Endpoints and statistical analysis vary between trials. EARLY TAVR (Evaluation of Transcatheter Aortic Valve Replacement Compared to Surveillance for Patients With Asymptomatic Severe Aortic Stenosis; NCT03042104) is enrolling 1,109 asymptomatic patients with severe AS to either TAVR or regular clinical surveillance. The nonhierarchical primary endpoint is a composite of all-cause death, stroke, or unplanned cardiovascular hospitalization. The secondary endpoint is death or disabling stroke. While asymptomatic, the study population is expected to be older, with baseline comorbidities, and a Society of Thoracic Surgeons (STS) score that is not necessarily low, and so event rates are expected to be high enough for a reasonable sample size trial. On the other hand, the PARTNER 3 trial (NCT02675114) is randomizing 1,328 patients with severe AS and low surgical risk (STS <4) to TAVR or SAVR. The primary endpoint is the composite of all-cause mortality, all stroke, and rehospitalization. Since event rates are not expected to be high in this population, the composite endpoint will be evaluated by a noninferiority analysis based on a relative noninferiority margin of 35%. A noninferiority design provides an opportunity to utilize “hard” endpoints such as mortality and stroke with a reasonable relative margin. Finally, the TAVR UNLOAD trial (NCT02661451) is enrolling 600 patients with moderate AS and HF randomized to either TAVR or optimal medical therapy. The primary outcome is a hierarchical composite of death, disabling stroke, hospitalizations and Kansas City Cardiomyopathy Questionnaire (KCCQ) change from baseline. Combining these endpoints increases event rates and allows enrollment of a smaller sample size with sufficient power to demonstrate an expected significant difference. The hierarchical ordering provides further clinical validity, as described below.

While device versus device trials have been infrequent, they may highlight important differences in secondary endpoints such as paravalvular leak, rate of permanent pacemaker implantation, and vascular complications.

Patients with AS suffer from exercise intolerance (usually due to HF or chest pain), overall poor quality of life (QOL), and frequent hospitalizations. Mortality is also increased in this population. However, running clinical trials testing new devices aimed at showing a reduction in mortality rates has become increasingly difficult due to requirements for a large patient population and a long duration of follow-up. Justifiably, a search for more prevalent, yet clinically meaningful, surrogate or “soft” endpoints has produced numerous candidates. The advantages of using surrogate endpoints include not only a higher number of events with every patient contributing during a shorter follow up, but also capturing clinically relevant endpoints affecting the everyday life of patients such as QOL or functional capacity measures. The clear caveat of using “soft” endpoints is that association is not necessarily causal (19). Moreover, some endpoints are difficult to combine into one composite endpoint.

Possible surrogate endpoints for TAVR trials include: 1) imaging endpoints, such as echocardiography (valvular function, EF) and magnetic resonance imaging (e.g., left ventricular [LV] hypertrophy reduction, infarct size); 2) functional capacity endpoints (e.g., 6-minute walking test [6MWT], exercise treadmill test, cardiopulmonary exercise test); 3) quality of life measures with such examples as generic questionnaires such as SF-36/SF-12 (20), EQ-5 (21), and disease specific ones such as Seattle angina questionnaire (22), Minnesota Living with Heart Failure Questionnaire (MLWHFQ) (23), and KCCQ (24,25); and 4) HF worsening assessed by rates of hospitalizations/hospitalization equivalents, and biomarkers (e.g., N-terminal pro-B-type natriuretic peptide [NT-proBNP], STS).

Valvular intervention trials regularly utilize imaging to assess device performance. Long-term echocardiographic follow-up provides important information when comparing trial arms. Nevertheless, to reach a clinically meaningful and valid conclusion, a trial must contain a clinically meaningful endpoint and cannot rely solely on imaging. Imaging endpoints may also be combined into a single composite endpoint. EARLY TAVR, for example, uses a combination of imaging measurements as a secondary endpoint termed “Integrated measure of LV health,” which includes LV global longitudinal strain, LV mass index, and left atrial volume index.

Rigorous assessment of exercise tolerance as an endpoint is useful because it not only assesses the impact on a primary limitation faced by the study population but also correlates with mortality and rates of hospitalizations. Moreover, exercise measures such as cardiopulmonary exercise testing (CPET)/peak oxygen consumption (peak VO₂) seem to outperform subjective measures (e.g., NYHA class, KCCQ) in predicting future “hard” outcomes. These outcomes have been validated in HF trials and should be suitable for trials assessing interventions for aortic valve disease. They are frequently accompanied by HF symptoms.

New York Heart Association Classification

The NYHA classification, first presented in 1928 (26) and extensively incorporated in HF guidelines (27), is often used for patient classification and risk stratification. While clearly having prognostic implications, this traditional four group division is subjective and limited in assessing more subtle changes in health status that may still be clinically meaningful.

Exercise Treadmill Test (ETT)

Physical fitness assessed by an ETT has long been shown to be a predictor of mortality (28). ETT was further incorporated into the guidelines with recommendations for AVR in asymptomatic patients with a positive test. Furthermore, ETT may also be used for risk stratification and prognostication in asymptomatic patients with AS (29). The use of ETT as a comparison tool between trial arms or to compare patients before and after intervention has been limited and impacted by comorbidities such as respiratory and orthopedic disease.

6-Minute Walking Test

The 6MWT in use today has been repeatedly modified since its inception as a 15-minute-run field test (30). A shorter 12-minute walk test was first used for the assessment of chronic bronchitis (31) and later shortened to 6 minutes to assess respiratory disease (32) and heart failure (33). The test is also frequently used in the evaluation and management of pulmonary hypertension (34). The primary measure is distance walked in a hallway during an uninterrupted 6-minute period. The test is most useful in those with limited exercise tolerance and in those limited by symptoms since patients are allowed brea ks (timer continues), a significant advantage over regular ETT assessment in the population enrolled in TAVR trials. The 6MWT has been shown to provide added prognostic information to the Euroscore in patients undergoing aortic valve replacement (35). In the TAVR population, the test seems sensitive enough to demonstrate improved function following TAVR and adds to risk stratification (36-39). A sub-analysis of the PARTNER trial showed that 6MWT improved following TAVR and that baseline 6MWT did not predict procedural outcomes but did predict long-term mortality (40). 6MWT is useful in medically managed patients with aortic stenosis as well (41). With proven prognostic implications, 6MWT seems a plausible, safe, and clinically meaningful surrogate marker in the population enrolled in TAVR trials.

Cardiopulmonary Exercise Test

CPET is an important tool in assessing functional capacity and carries prognostic implications (42). The test can be performed on a treadmill or a bicycle and several different protocols have been established. In HF, the cornerstone study by Mancini et. al. established the validity of peak VO₂ as a strong prognostic factor (43) and set the foundation for defining the cutoff value of 14ml/kg/min in heart transplant candidates (44). Other informative parameters from CPET (albeit less substantiated than peak VO₂) include exercise capacity, the ventilatory anaerobic threshold and ventilatory expired gas parameters (VE/Vco₂ slope). CPET has been assessed in patients with aortic stenosis, mainly for risk stratification and management guidance as compared to ETT and stress echocardiography (45). A baseline CPET that shows no hemodynamic compromise suggests conservative treatment is associated with a good prognosis (46).

In conclusion, impaired exercise capacity is a cardinal manifestation of HF and/or aortic valve disease and improvement in exercise capacity is a meaningful endpoint for patients. Exercise capacity (i.e., peak VO₂ and 6MWT distance) closely relates to outcomes across the spectrum of cardiovascular disease, particularly HF. Measurements of exercise capacity are safe and increasingly simple to perform, and each has advantages and limitations (Table 1). Done properly with core lab oversight, these results are highly reproducible and permit ascertainment of clinically meaningful endpoints. Despite providing less prognostic information, 6MWT is frequently preferred over CPET due to ease of administration and the lack of expertise or equipment required.

Table 1. Functional Capacity Measures

Although difficult to measure, health is not just the absence of disease but also the presence of physical, mental, and social well-being (57). Quantifying HF has been largely based on the presence of symptoms (NYHA classification) and the consequences of the disease/symptoms (e.g., mortality and HF hospitalizations). Aside from reducing mortality (especially sudden death), treatment goals in AS are to improve quality of life by reducing symptoms. HF treatments individually proven to impact mortality/disease progression have been demonstrated to affect QOL, such as medications (LCZ696 [58] and valsartan [59] but neutral results in a meta-analysis of beta-blockers [60]) , cardiac resynchronization therapy (61), and LV assist device use (49). Several questionnaires were constructed to better quantify disease burden, such as the MLHFQ introduced in 1987 and repeatedly validated since (62,63), the KCCQ (24), and the less known Chronic HF Questionnaire (64). These and other assessment tools have been compared with conflicting results (65-67).

The well-validated and widely used KCCQ was formulated to “create a valid, sensitive, disease-specific health status measure for patients with congestive heart failure (CHF)” and claimed to better assess clinical improvement and subcategories of QOL compared to other questionnaires (24). The score has several domains including physical limitation, symptoms (frequency, severity and change), self-efficacy, social, and QOL and has also been published in a shortened format (25). The KCCQ has been shown to be more sensitive to clinical change (assessed by blinded cardiologists) than other measures, including NYHA classification and 6MWT (68). An analysis of 1,358 patients from EPHESUS (Eplerenone’s Neurohormonal Efficacy and Survival Study) demonstrated the value of serial KCCQ assessments in predicting mortality and the combined endpoint of cardiovascular death and hospitalization, after multivariable analysis (hazard ratio [HR]: 1.09; 95% confidence interval [CI]: 1.00 to 1.18; and HR: 1.11; 95% CI: 1.05 to 1.17 for each 5-point decrease in KCCQ, respectively) (69). In an analysis of the PARADIGM-HF trial (58), LCZ696 treatment in HF patients demonstrated higher KCCQ scores in surviving patients (70). Although no specific tool exists for quality of life assessment in patients with AS, current generic health assessment tools such as KCCQ and MLHFQ have been used successfully. A sub-analysis from the PARTNER trial demonstrated KCCQ was reliable, sensitive enough to capture a difference post-TAVR (uncaptured by NYHA), and prognostic (71). Current-era TAVR trials (e.g., TAVR UNLOAD, EARLY TAVR, and PARTNER 3) all include KCCQ as a secondary endpoint or part of the composite primary endpoint.

While QOL provides a methodological assessment than can be easily compared between trials, it is limited in that it is somewhat subjective and easily biased by missing data.

QOL improvement is an important treatment goal and should be clearly quantified to allow comparisons and support fine-tuning therapy. Several validated measurement tools are available, with the widespread use of the KCCQ in clinical trials, providing it with some advantage over other questionnaires.

Any hospitalization, time to first hospitalization, or accumulating hospitalizations during follow-up are all clinically meaningful and acceptable endpoints. In the PARADIGM-HF trial, LCZ696 treatment significantly reduced the risk of hospitalization (58).  A sub-analysis of the PARADIGM-HF trial further demonstrated that combining HF hospitalization equivalents (outpatient intensification of HF therapy and emergency department visits) with HF hospitalizations and cardiovascular death into one composite endpoint contributed a significant number of events and shortened the time to achieve these event rates (72). Understandably, all current-era TAVR trials incorporate hospitalization (HF or valve related) or rehospitalization in the composite primary endpoint, increasing expected event rates by including a clinically important endpoint.

Current guidelines recommend measuring B-type natriuretic peptide (BNP) and NT-proBNP for HF diagnosis and for establishing prognosis (27). Several systematic reviews reported one or both peptides can independently predict mortality and/or HF hospitalization (73-78). In the PARADIGM-HF trial, NT-proBNP levels decreased with LCZ696 treatment and endpoint event rates were significantly higher in patients with NT-proBNP >1000 pg/ml (79). Consequently, clear definitions to correctly diagnosis HF worsening and provide a foundation for proper comparisons across trials have been published (80). Neurohormone levels change following TAVR (e.g., from 642 ± 634 pg/ml to 340 ± 253 pg/ml at 30 days in a small prospective study (37) could be used to suggest reduced wall stress and the effect of therapy. Moreover, several studies demonstrated the utility of measuring BNP or NT-proBNP for risk stratification and prognostic implications, including mortality (81,82).

The common use of composite endpoints, intended to increase statistical power and avoid type II error and to account for several effects of a studied intervention, has several limitations. First, pooling together several endpoints that are not considered equal and differ in their clinical importance is problematic. Second, traditional time-to-event statistics fail to capture events occurring after the first endpoint event occurred. Furthermore, death can cause a competing effect on the risk of nonfatal endpoints. Defining each component of the composite as an independent secondary endpoint has been recommended to address these concerns (6).

Several statistical methods aim to deal with the inherent limitations of using a composite endpoint: The Anderson-Gill, an extension of the Cox model, counts repeated events yet considers no dependence between events, which is not always true (83). The competing risk methods, an extension of survival analysis methods, consider the time lost for a potential event and its hazard risk, yet do not account for multiple events (84,85). The weighted composite endpoints, an extension of the regular time-to-event method, assigns a reduced (yet somewhat arbitrary) weight to a patient’s contribution to the cohort following a non-lethal event (86). The Finkelstein and Schoenfeld (FS) method compares each patient to every other patient, while accounting for both time to event and multiple events (7). The Win ratio, a further development of the FS method, also compares between patients, and not only generates a p value as a measure of significance, but also provides a measure of the relative effectiveness of the therapies (8).

The FS methodology (7) creates a hierarchy for all endpoints that can include clinically relevant hard and soft endpoints, for instance, mortality and quality of life. Each patient in Arm A is compared with each patient in Arm B. For each patient comparison and endpoint in the hierarchy, a win, loss, or tie is declared. If a win (and corresponding loss for the other patient) is declared, there is no need to continue to the next, lower tier, in the hierarchy. If a tie is declared, the comparison moves on to a lower tier. Then, the number of wins is counted in each group and statistically compared (e.g., by the Mann-Whitney test). This method provides the opportunity to increase power (and thus reduce the required sample size) by combining several endpoints yet ordering them in a predetermined, sensible hierarchy of importance to be clinically meaningful (unlike a “simple” composite endpoint). The FS methodology has been used repeatedly before, including in TAVR UNLOAD.

One limitation of the FS method is that the result is hard to interpret clinically, as it does not provide a measure of the magnitude of the difference between the study arms. The Win Ratio (also referred to as the Pocock Win Ratio) provides a ratio of the wins versus losses in the two arms and therefore an estimate of the effect size (8).

Proper trial design incorporating several endpoints in a clinically meaningful hierarchy provides the opportunity to assess the effect of new therapy in the smallest sample size acceptable. The Valve Academic Research Consortium-2 or VARC-2 has published recommended trial design principles and endpoint definitions (87). The VARC-2 consensus document is meant to provide standardization of endpoint definitions for TAVR studies that will allow easy comparisons between different studies. The consortium acknowledged the need for composite endpoints for achievable sample sizes. The document emphasized the need for composite endpoints to include components that have a similar impact and are expected to trend in the same direction: for example, device success combines mortality and echocardiographic parameters. Clinical efficacy is defined as a composite of mortality, stroke, hospitalization, NYHA class, and echocardiographic parameters. The widespread implementation of these recommendations will help uniform comparisons between treatments and across various trials, and ultimately improve patient care. The use of validated surrogate endpoints, and hierarchical composite endpoints (both “hard” and “soft” endpoints), provide further tools to decrease sample size yet increase clinical validity by capturing several relevant aspects of patient care and treatment success.