Surrogate and clinical endpoints in interventional cardiology: are statistics the brakes?

Statistical analysis

All sample-size estimates were calculated with nQuery/nTerim version 2.0 (Statistical Solutions Ltd., Cork, Ireland). Test hypotheses were divided in terms of endpoints (clinical and surrogate) and test hypotheses (noninferiority and test for difference). For simplicity reasons we will refer to the ‘test for difference’ as a ‘superiority’ design.

Reimbursement impact

Due to the plethora of reimbursement systems worldwide, a grossly simplified approach was chosen to determine the level of impact for selected study endpoints in the field of interventional cardiology. The National Institute of Health and Care Excellence (NICE) in the UK publishes technology appraisals in a number of evaluation pathways for major indications. Within the framework of coronary artery disease (CAD), NICE appraised a total of five technologies for DESs [NICE, 2010a], BMSs [NICE, 2011], bio-absorbable scaffolds [NICE, 2014a], drug-coated balloons [NICE, 2010b] and CAD-relevant comedication [NICE, 2003]. These appraisals were investigated for the frequency of study endpoints that constituted the basis for their relative reimbursement impact in this review.

Surrogate endpoint study designs

Late lumen loss

LLL, PS and MLD are based on edge-detection algorithms applied in coronary quantitative angiography (QCA), which requires a calibration based on the outer diameter of a contrast-dye-filled guiding-catheter tip (6F = 2.0 mm). Defined on this reference calibre, all diameters obtained from the shadow images of an angiogram are the foundation to compute LLL, PS and MLD. The major methodological advantage is that a blinding relative to the treatment group can be elegantly performed when the device type cannot be recognized in the angiogram. True blinding (investigator versus QCA operator) is therefore apparent for device studies, such as drug-coated balloon (DCB) versus POBA or BMS versus DES. In trials comparing DCB and DES, a systematic QCA error may concur given that a stent per definition has a sharper radiopaque edge, whereas a nonimplant angiogram may have a higher level of haziness, thereby leading to smaller MLDs. Based on the pioneering work by Mauri and coworkers, the mean LLL correlates well with binary-restenosis [Mauri et al. 2005b] rates but mean LLL is not a significant determinant of TLR [Mauri et al. 2005a]. Without questioning the scientific value of LLL in interventional trials that led to the paramount level of evidence we have at our disposal today, it must be mentioned that LLL has limited value from a reimbursement viewpoint. In addition, LLL may not be applicable when bio-absorbable vascular scaffolds (BVS) are studied due to their different time-dependent mechanical or biological behaviour in terms of recoil, vessel remodelling and neointimal changes [Bangalore and Mauri, 2009]. In these devices, after the degradation of the stent or scaffold postimplantation, late recoil may play a pivotal point that cannot be quantified with the standard angiographic LLL follow ups of 6–9 months. The major finding of our sample-size calculations revealed the ‘hopelessness’ to run head-to-head comparisons in terms of LLL, especially when the treatment effects are on a low level, in the 0.1–0.2 mm range.

Minimal lumen diameter

The advantages and disadvantages of LLL measurements also apply to the MLD. However, it is our opinion that the MLD seems to be hemodynamically indifferent because it does not consider the lesion length that impacts on the translesion pressure drop. This has been established in the human domain by Brosh and coworkers using FFR [Brosh et al. 2005], and also in experimental animal models [Young et al. 1975]; the latter demonstrated that there is a lesion-length-dependent pressure gradient across artificially produced stenoses. We should also not ignore the Navier–Stokes equation [Sherman, 1990] that provides the theoretical basis for pressure gradients across narrowed fluid boundaries defined by the diameter and length of the stenosis.

Fractional-flow reserve endpoints

By intuition, not every pressure drop across a lesion manifests itself as hemodynamically or clinically important. However, due to the averaging algorithm during pullback of the FFR wire, measurement variations are very small and in the ± 0.12 mmHg range [Pijls et al. 2000].

Even though a difference in FFR as an endpoint seems intuitive, it ought to be pointed out that a mere pressure drop independent of the lesion location may not have the merit to determine clinical efficacy. Also, the fact that FFR cannot be interpreted in a linear fashion, that is, a ΔFFR from 0.95 to 0.90 mmHg, is certainly less severe than a difference between 0.80 to 0.75 mmHg. Only when the lesion locations have the same impact on the coronary circulation, for example, with a lesion inclusion criterion such as the proximal left-anterior descending artery, may it be worthwhile to further explore FFR endpoints. Needless to say, the absolute FFR values at follow up should be above the established cut-off value of 0.80 [De Bruyne et al. 2012]. A recently published randomized trial comparing DCB and DES treatment in de novo lesions by Shin and coworkers revealed FFR values of 0.85 ± 0.08 in DCB versus 0.85 ± 0.05 in the stent group, which provides additional support that an FFR-based endpoint can be feasible even when the balloon dilatation is compared with a stent implantation, whereas the latter would certainly provide more postprocedural lumen gain [Shin et al. 2015].

Strut coverage

Strut coverage, by definition, can only provide an estimate on the temporal healing characteristics. It is, by definition, not an efficacy marker for vessel patency. Furthermore, it appears important that the formation of a functional endothelium can neither be expected nor truly documented with optical coherence tomography (OCT). In the current interventional cardiology community, there is an unmet need to reduce the duration of DAPT. Nevertheless, given that a completely covered stent has a lower risk for ST, an estimate on how fast stent struts can be covered seems desirable. A novel healing-response endpoint that includes the ratios of uncovered and malapposed stent struts was proposed by the US National Institute of Health and elaborated by Räber and coworkers, which seems to include more healing properties than the mere number of uncovered struts [Räber et al. 2015]. Unfortunately, there is no defined cut-off value for stent-strut coverage that translates to a no-risk situation for ST. There are literature reports, however, that state when 90–95% of stent struts are covered, the risk of ST without DAPT is sufficiently reduced [Tahara et al. 2011]. Nonetheless, it is debatable if the percentage of covered stent struts is an adequate measure to assess the risk of late ST and its continued need for DAPT. In the case of late ST, neoatheroma formation may be the culprit without underlying mechanical factors such as strut coverage.

Clinical endpoints

One side aspect of this review was to investigate other attractive endpoints to keep patient numbers low in early proof-of-concept trials. Clinical endpoints are most likely not the first approach when conducting early studies. However, it is undisputable that from a clinical standpoint and in terms of patient benefits, MACE, TVF or TLR rates are the most meaningful measures and the ‘gold standard’ for documenting efficacy. In this context, the importance to standardize clinical endpoints in coronary stent trials remains unchanged [Cutlip et al. 2007]. Recently, a large noninferiority margin of 4.5% with event rates of 7% was used to demonstrate noninferiority between resorbable scaffolds and DES by Ellis and coworkers [Ellis et al. 2015].

TLR, as the preferred efficacy endpoint was suggested by Silber and Herdeg, inasmuch as it establishes a direct cause-and-effect relationship between the lesion treatment and its failure [Silber and Herdeg, 2008].

Follow-up period

This review is an attempt to provide a first glance at study population sizes for preferred endpoints in patients with CAD. When biometric planning for a trial begins, comparable target populations and their expected outcomes for a specific treatment are elucidated first. If the number of patients is not manageable, study-budget-related questions gain importance. The provided estimates in this review may serve as a first guidance to eliminate possible trial designs that cannot be conducted with a reasonable number of patients. Often neglected, confounding factors may be introduced during the course of the follow up. By definition, once the study patients are discharged to the point of the follow up, there is a black box of confounding factors which may not be fully described. The patients’ reluctance to adhere to their DAPT may increase the risk of ST [Iakovou et al. 2005] or the level of exercise can have an impact on the recurrence of MI. Without the intention to delve into the postinterventional psychological aspects, there are also established relationships between compliance with comedication and psychological support for the patient [Warner et al. 2013].

Reimbursement

To have a clinical or angiographic benefit of one treatment group is the first step towards demonstrating clinical efficacy. However, which incremental benefit is worth additional spending in a particular healthcare system? We will have difficulties explaining a LLL improvement of 0.1 mm to a health-insurance provider, unless these angiographic results translate into clinical improvements. To our knowledge, there is no accepted methodology available that gauges the reimbursement value for various endpoints as a function of the number of patients. A first step on this terrain is attempted with the endpoints illustrated in Figure 6 on the basis of the sample-size estimate in Figures 1 –5. We are aware that this is a first crude attempt to relate endpoints, their sample-size estimates and their frequency of use in NICE appraisals to formulate a relative reimbursement value in %.

OPEN IN VIEWER

Figure 6.

Number of patients versus relative reimbursement value and clinical value (size of the bubbles) for various endpoints.

Given that TLR directly correlates with the treatment success or efficacy, as previously suggested [Räber et al. 2015], the reimbursement value of 70% seems adequate, whereas MACE remains the gold standard. In addition, TLR rates were the most useful endpoints of the main Markov transition-model inputs in a cost-efficacy study conducted by NICE [NICE, 2010b] and Bonaventura and coworkers, comparing DES and DCB angioplasty [Bonaventura et al. 2012]. The reimbursement value for the rate of ST as a primary endpoint is rather limited (20%) inasmuch as it is not an efficacy endpoint but a safety endpoint with a low incidence rate of 0.5–2.0% at 1 year. Despite this semiquantitative and certainly somewhat subjective attempt to portray these complex relationships of reimbursement value, clinical importance and study-population sizes, it seems obvious that angiographic endpoints are not the favourites in this context. Cardiac death is also of high interest from a reimbursement point of view. Nevertheless, there is the ethical dilemma due to fact that there are no additional treatment costs when this endpoint is reached.

In this review, we refrained from quality-of-life endpoints, even though they have been used in various patient subsets, such as patients undergoing aortic-valve replacements [Tully et al. 2015]. In multimorbid patients, by nature, multiple factors contribute to the overall health status so that causal relationships between the primary treatment and the overall health status are difficult to determine.

Limitations

The estimated number of patients for a given test hypothesis are based on available literature references. They do not replace a properly conducted sample-size calculation for a particular design. Furthermore, most endpoints, such as those related to safety and efficacy, are not interchangeable for a given target population and the objective of a trial. To gauge reimbursement values, each endpoint was based on NICE appraisals only and should be expanded to other cost-benefit analyses.