In the CARE-HF study AV delay was optimized using Doppler echocardiography of trans-mitral flow using the iterative method [20]. The AV delay which was found to provide maximum separation of the E and A waves was programmed. This method for AV delay optimization is often recommended in guidelines [21]. Estimation of stroke volume using echocardiography, is recommended as the method of choice for performing VV optimization in guidelines for echocardiographic optimization [21].

Outside the heart, the most readily observable effects of changes in efficacy of CRT are changes in pressure and flow. Of these pressure is more straightforward to measure non-invasively with automatic equipment.

We have previously demonstrated that changing AV delay has a larger effect on haemodynamics than changing VV delay and results in a curvilinear acute blood pressure response, which fits closely to a parabola [22]. A superior signal to noise ratio is seen when this is performed at higher heart rates [14]. Of the components of blood pressure, systolic blood pressure appears to be the most informative and reproducible when used for AV delay optimization [18]. The algorithm requires multiple alternations between a tested and reference delay and comparisons are made immediately before and after the change [23]. This method is amenable to performing large numbers of replicates, which narrows the confidence interval of the estimated optimum [24]. We have found that it is possible to closely predict the exercise optimum from resting measurements by pacing at different heart rates [25]. Very recently we showed how this could be done with an appropriately transformed signal from a pulse oximeter [26].

In BRAVO (British randomised controlled trial of AV and VV optimization) we adopted a standardised protocol based on our previous reports, which we describe in the methods below. In principle, it could be carried out with invasive pressure or with non-invasive pressure or pulse oximetry recorded beat-to-beat with any appropriate sensor.

The BRAVO study was set up with the aim of determining whether optimization using non-invasive haemodynamics is a possible clinical alternative to the recommended echocardiographic optimization. If the clinical outcomes are equivalent, this method would offer important logistical advantages over echocardiography as it can be operated by a single operator, could be automated, causes minimal discomfort to the patient, and the technology could be developed to become integral to new CRT devices. BRAVO is designed as a non-inferiority study with the primary endpoint of objective exercise capacity measured using peak oxygen uptake (peak VO₂). Secondary endpoints are symptoms, ventricular volumes, and NT-proBNP.

Study inclusion and exclusion criteria are shown in Table 1. Patients must have a previous diagnosis of symptomatic chronic heart failure and have been implanted with a biventricular pacemaker or biventricular defibrillator device. The device must have been implanted at least six months prior to enrolment, because cardiac reverse remodelling is known to occur after implantation and one of the secondary outcome measures is left ventricular volumes. We wanted to minimize the potential for any differential effect on remodelling of the two optimization schemes to be interfered with by effects arising from recent implantation of the device.

Since the primary endpoint is peak VO₂ measured using a treadmill cardiopulmonary exercise test, ability and willingness to walk on a treadmill is an inclusion criterion.

The endpoints of BRAVO are designed to assess physiological markers which are consistently reported to improve following CRT implantation. The rationale for this is that if these markers are less favourable with haemodynamic optimization than with standard echocardiographic optimization, then there would be good reason to reject the use of this haemodynamic optimization since it is biologically implausible that longer term outcomes would be satisfactory when these physiological markers were worse.

The primary endpoint is objective exercise capacity defined as peak VO₂ on cardiopulmonary exercise testing [27].

Peak VO₂ measures a physical capacity that integrates many elements and has strong prognostic value in heart failure [28] Reproducibility of repeated measurements of peak VO₂ at separate hospital visits is well defined [29] and measurement variability is sufficiently low to allow it to be a suitable measure for assessing the impact of optimization of CRT device settings on exercise capacity without requiring an excessively large study population.

Secondary outcome measures are left ventricular reverse remodelling, quality of life score and N-terminal pro B-type Natriuretic Peptide (NT-pro BNP). Left ventricular end diastolic volume (LVEDV) is measured by echocardiography and has been reported to decrease after implantation of CRT devices [30].

Quality of life is assessed using standardised scores for heart failure: SF-36v2 [31] and Minnesota living with heart failure score [32].

The third secondary endpoint is NT-pro BNP. Neuroendocrine activation may be both a marker of and a mechanism for progressive deterioration in cardiac function [10, 33].

This study is conducted in compliance with Good Clinical Practice (GCP) guidelines, consistent with the most recent version of the Declaration of Helsinki. The study has been approved by the South West London Research Ethics Committee (3) and site specific assessments were performed for each participating hospital.

While patients are recruited from many different hospitals, all study visits are performed at one of two testing centres. The two testing centres are Aberdeen Royal Infirmary and Imperial College London. The advantage of this approach is that the equipment for making clinical measurements is kept constant during the study. It also facilitates training of staff performing the study in order to keep study costs to a minimum.

Patients in sinus rhythm have the following AV delays tested unless the rhythm becomes intrinsic at a shorter AV delay: 40, 80, 120, 160, 200, 240, 280, and 320 ms. Some devices do not permit 240 and 280 ms: in which case 250 and 275 ms are used instead.

Tested VV delays are LV first by 80 ms, 60 ms, 40 ms, 20 ms, 0 ms, and RV first by 20 ms, 40 ms, 60 ms and 80 ms. In devices that do not allow programming of these values, the nearest permitted value is used.

Beat-to-beat blood pressure is measured non-invasively using the Finometer device (Finapres Medical Systems, Netherlands). This digital photoplethysmograph uses an inflatable finger cuff and applies a volume-clamp technique [34] to yield a continuous arterial pressure waveform. This technique has been extensively validated for measuring changes in blood pressure [35‐38].

The data is read into a custom data acquisition system. In brief, this consists of an analogue-to-digital card (DAQCard-6062E, National Instruments, Austin, TX, USA) and workstation running custom-written software in the Labview instrument control language (v7.0, National Instruments, Austin, TX, USA). Data is analysed off-line using custom-written software [39‐41] based largely on Matlab (Natick, MA, USA), which has been developed, and validated at the lead centre [41‐43].

When using a physiological variable in order to guide the optimization of CRT settings it is important that enough steps are taken to minimize the effects of biologic variability. Without these steps noise is likely to overwhelm the signal and the results obtained are unlikely to be reliable [44, 45].

In this study we use a protocol designed to minimise the effect of noise, this has been described in detail previously but in brief consists of the following steps [14, 22, 23, 25]. Each tested delay is compared to a reference delay (for AV delay 120 ms, for VV delay 0 ms) and the relative change in acute blood systolic blood pressure is calculated as the difference of the mean of 8 beats immediately before the transition to the tested delay compared to the mean of the 8 beats immediately after the transition. For each tested delay multiple transitions between the tested and reference delays (a minimum of six transitions) are made and the mean relative change in acute systolic blood pressure is calculated (Figure 2).

The optimal AV or VV delay is identified as the peak of a fitted curve [22] unless the longest tested delay results in the greatest increase in pressure, in which case this is taken as the optimal value since fitting a curve in this case will result in a non-physiological AV delay being reported as optimal.

In order to maximize signal-to-noise ratio, optimization is performed during atrial pacing with an elevated heart rate (40 bpm above the resting heart rate if paced or 45 bpm if sensed). This identified optimum is programmed as the paced AV delay. The sensed AV delay for haemodynamic optima is programmed to be 60 ms shorter based on data previously acquired with this protocol.

AV optimization is performed by assessing the pattern of pulsed wave Doppler recorded through the mitral valve using the iterative method (Figure 3) [11]. Doppler data is acquired using a GE Vivid-I machine (GE Healthcare). Delays which produce intrinsic ventricular conduction are excluded from analysis. The AV delay with the best filling pattern, defined using the published protocol from CARE-HF, [11] is programmed as the sensed AV delay. The paced AV delay is programmed as 30 ms longer than the optimal sensed AV delay as per recent guidelines [21].

VV optimization is performed after AV delay optimization. The AV delay is set to the value determined as optimal during the AV delay optimization during VV optimization. The VV delay yielding the highest mean LVOT VTI, determined from 5 beats, is selected as the optimal VV delay.

Additional echocardiographic measurements are made using a Philips ie33 machine (Philips, Netherlands) using an S5-1 probe with the patient semi recumbent in the left lateral position. For each parameter the mean of three measurements are calculated, the following parameters are measured: left ventricular dimensions, volumes and ejection fraction from the four and 2 chamber views, quantification of mitral regurgitation and maximum velocity through the aortic valve. Tissue Doppler imaging (TDI) is used to measure longitudinal function at the mitral valve annulus level on the lateral, septal, and inferior walls. Where image quality is suitable a 3D dataset is obtained using an X3-1 probe and used for off-line calculation of volumes and ejection fraction by a physiologist blinded to the study arm.

Quality of life (QoL) is scored using the Minnesota Living with Heart Failure Questionnaire (MLHFQ) (http://​www.​mlhfq.​org/​) [32] and SF-36 [31] (http://​www.​sf-36.​org/​tools/​sf36.​shtml). The MLHFQ questionnaire is a validated condition-specific measure with good sensitivity and discrimination. It will be complemented with the SF-36, a validated generic measure of QoL, which enables comparison of findings with other outcome studies.

Blood samples are taken for analysis of N-terminal pro-B-type Natriuretic Peptide (NT-pro BNP).

In a subset of patients reproducibility of cardiopulmonary exercise testing is investigated by conducting a second exercise test, within a two week period, prior to programming the AV delay. Sub-studies are also performed to assess the reproducibility of the two optimization protocols. In a further subgroup the impact of AV delay on longer term blood pressure is assessed by 24 hour ambulatory blood pressure measurement. This is carried out, following optimization, at the initial visit (time of randomization) and again at 6 month follow- up.

Distributions will be described by the mean and standard deviation, where they meet criteria for normality. Skewed data will be appropriately transformed prior to analysis. Analysis of the cross-over data will be performed using linear mixed effects models [47]. Sample size calculations were carried out using nQuery Advisor and are given in detail in the section below.

All sample size calculations were based on paired t-tests comparing outcomes for echocardiographic optimization versus haemodynamic optimization.

The reproducibility of peak VO₂ in heart failure patients between two separate visits using a smoothed modified Bruce protocol on a standard treadmill is 2.4 ml/kg/min, expressed as standard deviation of difference between two independent measurements [48]. Implantation of a CRT pacemaker results in a 0.5-2.5 (mean 1.5) ml/kg/min increase in peak VO₂[8, 9, 49‐53]. Simply implanting a CRT device yields a haemodynamic improvement of 5–6 mmHg of systolic blood pressure [11, 12]. Optimizing the settings after implantation appears in preliminary work to add a further 3-6 mmHg increase in pressure, [1] i.e. between 50-100% of the effect of CRT.

The study sample size was chosen to have 90% power to detect a margin of equivalence of 0.75 ml/kg/min at the 5% significance level. This is the smallest difference between the treatment mean and the reference mean that still results in the conclusion of non-inferiority and corresponds to half the expected clinically meaningful difference as advocated by Jones et al. [54] using a one sided t-test of equivalence in means. On this basis 338 patients (N1 = 177/N2 = 177) are required. Mean mortality and dropout in similar populations over a 12-month period has been observed to be between 10% and 15%. In order to accommodate possible losses to follow-up as high as this, the target enrolment is 400 patients.

The standard deviation of the difference of LV end-diastolic volume measurements has previously been reported as 23.8 ml [55]. Following CRT reductions in LV end-diastolic volumes are variable but of the order of 25 ml [56] On this basis 338 patients provides >99% power to detect a margin of equivalence of 12.5 ml (50% of clinically significant difference), and permits a difference of 7.4 ml to be detected with 90% power at a significance level of 5%.

For SF-36, the standard deviation of the difference of scores has previously been reported to be 26 [17]. Previous studies of CRT have reported improvements of ~17. With 338 patients we have 92% power to detect a margin of equivalence of 8.5 at a significance level of 5%. For the Minnesota questionnaire, mild improvements in QoL are represented by a change of 12 with a typical standard deviation of 15 [48]. With 338 patients we would be able to easily detect a mild improvement of this magnitude (>99% power) and have 90% power to detect an equivalence limit of 4 points in the Minnesota score.

For NT-proBNP, CARE-HF reported median reductions of 656 pg/ml in NT-pro BNP (adjusted for change in control group) [10]. The within subject coefficient of variation (σ_w/mean) in log NT-proBNP levels in people with chronic heart failure over 12 months has been reported to be ~10%, i.e. around 190 pg/ml [57]. Assuming a log normal distribution of NT-proBNP, with 338 patients we will be able to easily detect an equivalence limit of half the clinically reported difference with >99% power at 5% significance level and have 90% power to detect an equivalence limit of 0.062 log units (i.e. a fall in baseline NT-pro BNP from 1900 pg/ml to 1647 pg/ml) at a 5% significance level.

All adverse events will be documented. Serious adverse events are defined as those which are not anticipated or not known to be related to the condition being studied or the intervention being used that would result in any of the following outcomes: death, life-threatening condition, unexpected/unplanned in-patient hospitalisation, persistent or significant disability or incapacity, is a congenital anomaly or birth defect.

The first patient was randomised in December 2010. Recruitment is expected to be complete by December 2013. The study will report in 2015 after follow up of all participants is complete.