Introduction

Fractional flow reserve (FFR) is a well validated index of ischemia which is measured during maximal hyperemia usually achieved during peripheral or intracoronary adenosine administration. FFR adoption has increased by approximately 300% in recent years, and now stands at 30% relative to PCI volume in the United States [1] with improving uptake across Europe according to a recent survey [2]. Despite this, global uptake has continued room for growth given the clearly demonstrated benefits to patient outcomes [3–5]. In an effort to facilitate interpretation of physiology data, efforts have focused on improving the user interface in order to optimize workflow [6]. A particular challenge exists in trying to discard spurious data which can be caused by artifacts from a variety of sources (such as aortic or pressure wire noise, signal distortion, or arrhythmia) [7]. Steady State FFR refers to FFR measured when the Pd/Pa curve displays relatively stable values over a period of analysis during hyperemia which may not always be as low as the minimal FFR value observed during the complete physiology study. These values can be selected by the operator prospectively as per the VERIFY 2 Trial [8] or an automated algorithm can be applied which looks for sequential high quality measurements and then calculates the so called “smart” minimum FFR (smFFR) which was recently described Johnson et al. [9]. The smFFR algorithm has been validated against Core Laboratory FFR and found to produce equivalent results using an automated software [9].

In this study we aimed to compare both operator defined steady state and minimal FFR versus core lab derived smFFR as well as automatically detected FFR as displayed by the Philips Volcano s5® console (Volcano FFR). We quantified reclassification of vessels among operator-defined steady state and minimum FFR, console-reported FFR, and smFFR .

Methods

All vessels which comprised the VERIFY 2 dataset were eligible for analysis. As all patients had already consented to anonymized analysis of their data no further consent was obtained beyond that which was acquired for the main study. The methodology of this study has been described elsewhere [8] but in brief, patients with angiographically intermediate coronary stenoses (30–80% diameter stenosis) in which FFR measurement was clinically indicated were eligible to be included. All physiology data was acquired by or under the supervision of BH or SW to minimize inter-observer variation. This was a single center study performed in Golden Jubilee National Hospital, Glasgow, U.K. Standard exclusion criteria for pressure wire studies applied and included the following: severe calcific coronary disease, severe tortuosity rendering pressure wire studies difficult or impossible, recent myocardial infarction within the previous 72 h, ongoing unstable chest pain, known intolerance of adenosine or severe asthma. A Prestige® or Verrata® Wire (Philips Volcano) was inserted into the guide catheter, calibrated and passed to the distal third of the coronary artery beyond the lesion of interest. Once in position and following the administration of intra-coronary isosorbide dinitrate (200 mcg) patients had adenosine administered via large bore cannulae at a rate of 140 mcg/kg/min for a minimum of 2 min. The operating cardiologist interpreted results during the index procedure and both the minimal FFR value (excluding artifact) as well as the steady state FFR value were noted on a standardized case report form. Results as reported by the Philips Volcano software were manually extracted from the study files. Following completion of enrollment, the study data was anonymized and electronically transferred for analysis and calculation of smFFR using an automated algorithm at the Weatherhead PET Center For Preventing and Reversing Atherosclerosis, Houston, Texas. Details and results from the smFFR algorithm have been previously published [9] but, in brief, pressure signals from the aorta and coronary artery were analysed to identify beats of sufficient quality. Segments of the tracing that did not meet quality criteria (for example, loss of pressure wire signal or aortic pressure damping) were excluded. The Pd/Pa trend line was computed and averaged for valid portions of the tracings, and its lowest value defines the smart minimum FFR. Figure 1 visually summarizes the transformation of raw data into a Pd/Pa trendline.

Fig. 1
figure 1

SmFFR example raw tracing with analysis (reproduced with permission [7]). Left phasic aortic pressure (Pa) (red) and distal coronary pressure (Pd) (blue) tracings. An intravenous infusion of adenosine at 140 μg/kg/min began after approximately ten cardiac cycles and continued for the duration of the recording. Right analysis of the phasic data shows the average Pa (red) and Pd (blue) pressures for each cardiac cycle, as well as derived heart rate (green) and Pd/Pa (black). Vertical dashed gray lines denote the hyperemic period. Horizontal bars mark the smart minimum fractional flow reserve (FFR) and its associated pressures and heart rate. Note the stability of Pd/Pa during the hyperemic period [coefficient of dispersion (COD) 1.5%] compared with the much wider variations in aortic pressure (COD 7.6%), coronary pressure (COD 7.5%), and heart rate (COD 7.2%)

Full size image

Statistical analysis was performed (BH) using version 23 SPSS, IBM Corp, Armonk, NY and reviewed by an independent biostatistician (JM) not involved in other parts of the study. Values were correlated using Spearman’s coefficients and the mean values were compared using paired t tests. The level of agreement between tests was examined using Bland Altman plots using Graphpad Prism ver. 6.

Results

Data from 256 of the 257 vessels was analyzable using the smFFR algorithm. Summary demographics are shown in Table 1. Means and standard deviations of FFR measured by all methods were similar and are shown in Table 2.

Table 1 Clinical characteristics and vessel data of study population (n = 197)
Full size table
Table 2 Descriptive statistics for FFR data
Full size table

Spearman’s correlation coefficients are shown in Table 3. Although the FFR values derived by each modality are numerically similar there is a statistically significant difference between all pairs (Table 4) except minimal and Volcano derived FFR (p = 0.752).

Table 3 Correlation coefficients between FFR Measurements
Full size table
Table 4 Paired t tests for all FFR modalities
Full size table

The operator-defined steady state FFR correlated and agreed well with the smart minimum FFR (Fig. 2): r = 0.984 (p < 0.001), average bias 0.008 (SD 0.014), 95% limits of agreement −0.02 to 0.036 (Fig. 3, panel a). The Volcano FFR also correlated well and agreed well with the smart minimum FFR: r = 0.997 (p < 0.001), average bias 0.004 (SD 0.006), 95% limits of agreement −0.016 to 0.008 (Fig. 3, panel b). The Volcano FFR correlated well and agreed well with the operator-reported steady state FFR: r = 0.985 (p < 0.001), average bias 0.012 (SD 0.014), 95% limits of agreement 0.039 to −0.015 (Fig. 3, panel c). The minimal FFR value reported by the operator correlated well and agreed well with the Volcano FFR with an r value of 0.997, average bias of 0.0001 (SD 0.007), 95% limits of agreement −0.0134–0.0137 (Fig. 3, panel d).

Fig. 2
figure 2

Scatterplot of steady state FFR versus smart minimum FFR

Full size image
Fig. 3
figure 3

a Bland Altman of smart minimum FFR versus steady state FFR. 95% limits of agreement indicated by solid lines. Average bias indicated by dashed line. b Bland Altman of smart minimum FFR versus volcano derived FFR. 95% limits of agreement indicated by solid lines. Average bias indicated by dashed line. c Bland Altman of steady state FFR versus volcano derived FFR. 95% limits of agreement indicated by solid lines. Average bias indicated by dashed line. d Bland Altman of volcano derived FFR versus minimal FFR value. 95% limits of agreement indicated by solid lines. Average bias indicated by dashed line. Note that there is the same number of points on all graphs but some have graphs have more points with identical averages and differences and these get overlaid on one another

Full size image

The number of lesions considered to be functionally significant (FFR ≤ 0.80) by each parameter were as follows: 108/256 (42.2%) using the operator-defined steady state FFR, 123 (48%) using the Volcano FFR, 121 (47%) using the minimal FFR values and 115 (45%) using the smFFR algorithm. When a cutoff of ≤0.75 is applied for ischemic FFR the numbers of vessels categorized as ischemic are 57 (22.3%) for Steady State FFR, 71 (27.7%) for Volcano FFR, 71 (27.7%) for minimum FFR and 65 (25.4%) for smFFR.

In 244/256 lesions [95%, CI (92, 98)%], the operator reported that steady-state FFR, smFFR, and Volcano FFR agreed with each other to within 0.04, which is within the previously reported test/retest limits of agreement of FFR reported by an experienced core lab [9].

Discussion

We have shown that high quality physiology data obtained by experienced interventional cardiologists is similar to that obtained by using the “smart” minimum algorithm in a core laboratory setting with a reclassification in only 7/256 [2.7%, CI (1.1, 5.6)%] vessels under study between the operator defined steady state FFR and smFFR (Fig. 4). No cases were excluded in the primary analysis of VERIFY 2 [8] in order for the dataset to represent a reaI world population and similarly we only excluded 1/257 of the VERIFY 2 cases in this analysis due to technical difficulty in applying the algorithm. It is also clearly evident that the Philips Volcano algorithm reports very similar data to the operator determined minimal FFR value. Our prior study has demonstrated that 40% of patients may not reach a “steady state” during IV adenosine infusion [9]. These scenarios almost certainly occur due to insufficient adenosine delivery as a result of the long and dynamic pathway between peripheral IV infusion and central coronary artery action, especially considering effects of variable venous return, altered pulmonary transit due to tachypnoea, and red blood cell deactivation of adenosine. As a result, the lowest FFR value derived from valid data reflects the peak action of adenosine vasodilation and hence maximal hyperemia. However, the lowest FFR value without quality checking, as currently reported by several commercial platforms including the Boston Scientific Comet Wire™/Polaris™ system (Boston Medical, Marlborough, USA) and Opsens Medical system (Québec, Canada), does not reflect valid physiology. It is therefore vital that improvements with pressure wire technology are paralleled by improvements in software design which incorporates data filtering through modalities such as the smFFR algorithm. With improved confidence in reliability, less experienced operators may be more comfortable in making decisions based on physiological data especially when FFR values are close to ischemic cut-offs.

Fig. 4
figure 4

Pd/Pa during rest and hyperemia during a coronary physiology study. smFFR is indicated by the black line (0.782) and operator defined steady state FFR is indicated by the dotted red line (0.82). In this case there was a significant discordance between measurements which would have led to a reclassification as ischemic if the smFFR measurement was adopted

Full size image

Limitations of this study

This is a single center study from a unit that performs high volume coronary physiology as part of local and international studies. This may mean that operator experience may not be generalizable to other centers and there is therefore potential that the difference in smFFR and operator-defined FFR values may be underestimated in this study. Furthermore, our prior work measured FFR twice [10] which provided an independent method to judge values based on test/retest similarity unlike the single measurements obtained in the current study.

Conclusions

Smart minimum FFR results are similar to operator-defined FFR values when they are specifically sought and recorded in experienced coronary physiology centers. Using the automated minimal values on pressure wire consoles results in reclassification of up to 6% of cases as ischemic when a cut-off value of ≤0.80 is utilized and up to 5% change in reclassification as ischemic when a ≤0.75 cut-off is adopted. Operators must ensure careful observation for artifact during recordings.

Future directives

Future use of algorithms to improve catheterization laboratory workflow will be important in determining the usefulness of the “smart” minimum algorithm in everyday practice. Furthermore, software/hardware that provides automated recognition of drift during a case as well as exclusion of measurements with pressure damping or other artifacts would all serve to improve the quality of physiology data. Clearly there are a large number of patients whose FFR data falls within the FFR grey-zone whose results are open to interpretation depending on the institutional/operator ischemic cut-off and whether minimal or steady state values are applied. Future prospective studies may provide interesting data on the prevalence of ischemia in this group to aid treatment strategies (NCT02425969).