This study comprised 46 patients diagnosed with BV at the Division of Balance Disorders at Maastricht University Hospital, based on the diagnostic criteria for BV from the Bárány Society . Since VOR gain obtained by VHIT was used as an outcome parameter in this study, this criterium was removed from the inclusion criteria. Patients diagnosed with BV solely based on VHIT outcomes were, therefore, not part of this study population. Inclusion criteria comprised (1) reduced caloric response (sum of bithermal maximum peak slow phase eye velocities of < 6°/s on each side), (2) and/or reduced horizontal angular VOR gain < 0.1 on rotatory chair and a phase lead > 68°. Exclusion criteria comprised being unable to stop vestibular suppressants for 1 week (cinnarizine and all psychiatric medication), and the inability to undergo one of the vestibular examinations.
Experimental setup 
One trained examiner (FL) performed all VHIT’s. A fixed distance of two metres from the back of the chair to the point of fixation was ensured . Patients were seated on a static chair, to prevent upper body movement during head impulses. The room was well lit, to ensure a small pupil in every patient. Patients fixated on a green (532-nm) 1-mw laser dot projected on a large full visual field black (or white) painted wall. This facilitated a wider range for measuring the eye movements. At the same time, it minimised the change of artefacts due to light reflections onto the pupil. The fixating point was adjusted to the eye level of every patient. Each test started with calibration of the system. The examiner assessed the quality of the calibration and determined whether the process needed to be repeated. The examiner stood behind the patient, holding the head firmly during head impulses. Patients were instructed to relax their neck, keep their eyes wide open and fixate on the target in front of them. The examiner continuously repeated these instructions to facilitate optimal awareness of the patient. The head impulses comprised fast horizontal rotational head movements (> 120°/s) with a low amplitude, unpredictable in timing and direction. Only outward impulses were given .
The camera of the Interacoustics and Otometrics systems is head fixed and is integrated in a pair of goggles. Therefore, before start of testing, goggle movement was minimised by tightly fastening the strap of the goggles around the patients’ head. The camera was always set on the right eye and focused on the pupil while the patient looked at the point of fixation with eyes wide open. In case the eyelids were in front of the pupil, the examiner adjusted the rim of the goggles so they would hold the eyelids back. After calibration, the patient was instructed to not touch (the strap of) the goggles, their face and/or their hair. The camera of the Synapsys system is space fixed, and therefore, no goggles were used. The camera that measured eye and head movements was placed in front of the patient. Eye movements from both eyes were measured (Fig. 1).
Three different VHIT systems were used in this study: EyeSeeCam (Interacoustics VOG; Munich, Germany), ICS Impulse (GN Otometrics; Taastrup, Denmark), and Ulmer (Synapsys, Marseille, France). Each patient sequentially underwent the horizontal VHIT with the different VHIT systems. The Synapsys system was not used in 17 patients, and the Interacoustics system was not used in one patient, due to the unavailability of the systems at the time of testing. The order of testing of the different VHIT systems was randomised by draw.
VOR gain calculation by the different VHIT systems
VOR gain, as calculated by the systems, was used as main outcome parameter. The three systems calculate VOR gain differently. Interacoustics uses instantaneous gain; it divides eye- and head velocity at a certain point in time (small window around 60 ms) after onset of the head movement . Otometrics calculates VOR gain as the ratio of the area under the eye velocity and head velocity curve (from 60 ms before peak head acceleration to the last value of 0°/s as the head returns to rest). If needed, the eye movement is desaccaded by the system before the VOR gain is calculated . The Synapsys system calculates the VOR gain over the period from 40 ms before to 80 ms after peak head acceleration for each impulse. In case of a covert saccade, the 80-ms window is reduced, and stops at time of onset of the covert saccade . However, the method of gain calculation used by the Synapsys system was unknown to this research group, despite multiple efforts to obtain more information from the manufacturer.
Covert saccades might influence VOR gain (calculation). Therefore, covert saccades in this study population were analysed separately to assess whether they differed between tests (as an adaptation effect) in this BV population when repeatedly tested. The frequency of occurrence of covert saccades, and the latency of the first covert saccade of a trace were analysed.
To extract saccades, head and eye velocity (Interacoustics and Otometrics) and position (Synapsys) traces were exported and processed using Wolfram Mathematica 11.3 (Wolfram Research, Champaign, IL, USA). Only traces that were accepted by the systems were exported.
Synapsys measures both eyes during VHIT, but in this study, it was chosen to only use traces from the right eye, to better facilitate comparison with Interacoustics and Otometrics, which only register data from the right eye. In case of missing values from the right eye, data from the left eye were used. Because of the lower resolution of the Synapsys camera (100 Hz), the original eye and head position data were resampled to 250 Hz using linear interpolation. By differentiating these eye and head position traces, the velocity traces were calculated for eye and head movements recorded with the Synapsys system. Eye and head velocity traces from Interacoustics and Otometrics were directly extracted from the system itself. Eye and head position data for these two systems were calculated using numerical integration. Head and eye acceleration data were calculated for all three systems by differentiating the eye and head velocity signal.
To establish artefact-free traces for analysis, traces were removed when (1) peak head velocity was < 120°/s, or (2) the head velocity trace contained a bounce at the end of the impulse of > 50% of peak head velocity, or (3) head velocity never crossed zero after peak head velocity (within the recorded time frame), or (4) the head velocity trace contained missing values, or (5) the shape of the head velocity trace implied an inadequate head impulse, assessed by visual inspection and consensus between three authors (RB, DS, TD), or (6) when the mean head velocity of the interval of 80 ms prior and 120 ms after a peak head velocity was not in the range of ± 3 SD of the set of mean head velocities calculated in the same interval in all traces of one patient [4, 15, 16].
A custom-made algorithm was developed in Mathematica, and applied to extract saccades from the eye traces. To increase accuracy, every saccade was verified by visual inspection in the eye and head velocity and position traces. Two authors needed to achieve consensus (TD, DS) before a saccade was approved. Head impulse onset was specified as head velocity exceeding 10°/s, head impulse offset was defined as head velocity crossing 0°/s. Onset of a saccade was marked as the point where eye velocity crossed 0°/s or eye acceleration reached 2000°/s2. Saccades were included when (1) they occurred after peak head velocity, and (2) had a magnitude of more than 60°/s, and (3) peak velocity of the saccade was recorded, and (4) occurred at least in two traces around the same location within the same trial and patient. A saccade was classified as covert when onset occurred before head velocity crossed zero, and as overt when onset occurred after head velocity crossed zero.
Saccade analysis: defining frequency and latency
In this study, the first covert saccades of the first seven artefact-free traces were used for analysis . The frequency and latency of the covert saccades were extracted from the original eye velocities in the Interacoustics and Otometrics system, and from the calculated eye velocities in the Synapsys system. The frequency of occurrence of a covert saccade was first registered as a binary outcome (Yes/No) for every trace separately. From these data, a ratio per patient was calculated (in percentage). Latency (in milliseconds) was registered as the onset of the covert saccade, and was normalised to the start of the head impulse .
Data were analysed using SPSS Statistics 24 for Windows and R (v.3.5.2.). The α-value was set on p < 0.05. In case of multiple comparisons, the Bonferroni correction was applied. When no interaction was found between leftwards and rightwards head impulses, the direction of the impulse was removed from the statistical model and both sides were analysed together.
Statistical analysis of VOR gain and agreement of VHIT systems regarding BV diagnosis
A repeated-measures ANOVA was used to compare mean VOR gain between the three systems. A VOR gain of < 0.6 was classified as “bilateral vestibulopathy”, a VOR gain of ≥ 0.6 was classified as “no bilateral vestibulopathy” . In case the VHIT systems showed a discrepancy in classifying BV, it was classified as “no agreement”.
Statistical analysis of VOR gain and repetitive testing (the order effect)
To evaluate the order effect, a repeated-measures ANOVA was used to compare mean VOR gain between the first and the last executed VHIT trial (regardless of the VHIT system).
Statistical analysis of peak head velocity
Peak head velocities (extracted from the raw traces of the VHIT systems) of all traces of all patients were combined per VHIT system. Median peak head velocities were compared between VHIT systems using a Mann–Whitney U test. In patients with “no agreement” between systems, peak head velocities were analysed separately within the BV patient. Median peak head velocities of those particular trials were compared between VHIT systems using a Mann–Whitney U test.
Statistical analysis of saccades
The frequency of occurrence of covert saccades was compared between the first and the last executed VHIT trial (regardless of the VHIT system) using a generalised linear mixed-effects model. Additionally, the latency of the first covert saccade was compared between the first and the last executed VHIT trial (regardless of the VHIT system) with a paired T test. Patients with missing values (no saccades) were not included in this last analysis.