Refining the Video Head Impulse Test Diagnostic Accuracy: A Case-Control Study

The bedside head impulse test was formulated in 1988 as a measure of the lateral semicircular canal function [Halmagyi and Curthoys, 1988]. For this test, pathological response is based on the detection of re-fixation saccades (RSs) that compensate for the low gain of the failing vestibulo-ocular reflex (VOR). In 2009, the video head impulse test (vHIT) was introduced [MacDougall et al., 2009]. This technology, which is based on the capturing of eye movements at a frequency of around 250 Hz, enables the quantification of the VOR gain and the recording of RS both during the head movement and following it (“covert” and “overt” RS, respectively). The RS parameters of latency, frequency, and velocity are provided by the commercially available vHIT systems [Halmagyi et al., 2017]. However, current clinical practice considers the VOR gain, for which norms and pathological values have been published, as the primary measure for semicircular canal function, while the role of the registered RS is still under evaluation [MacDougall et al., 2009; Matiño-Soler et al., 2015; McGarvie et al., 2015; Korsager et al., 2017; Kim and Kim, 2018].

It was found that the RS velocity increases with decreasing gain [Anson et al., 2016; Janky et al., 2018]. Also, in a group of dizzy patients with normal VOR gains, the frequency of the RS was reported to increase with age [Anson et al., 2016]. Several previous studies have examined the possible role of RS parameters in addition to the VOR gain towards the diagnosis of canal dysfunction. High-velocity RSs were demonstrated together with improved gain values among patients recovering from vestibular neuritis, supporting the possible diagnostic value of RS in addition to the gain criterion [Yang et al., 2018]. Overt but not covert RSs were frequently recorded in asymptomatic older patients [Rambold, 2016; Janky et al., 2018] differentiated from those found in unilateral semicircular canal dysfunction by their lower frequency and slower peak velocity. The combination of gain values <0.78 and RS frequency >82% was recently suggested to improve diagnostic accuracy over the low gain parameter alone [Janky et al., 2018]. Others even assert that the interpretation of vHIT results should first rely on the occurrence of RS and only secondarily on the gain values [Korsager et al., 2016]. Another study proposed that the presence of RS, albeit with normal gain values, indicates the existence of peripheral vestibulopathy and localizes the side of the lesion [Perez-Fernandez and Eza-Nuñez, 2015]. For the recently introduced protocol of suppressive head impulse (SHIMP), the amplitude of anti-compensatory saccades was suggested as an indicator of residual semicircular canal function [Shen et al., 2016].

When mono-ocular eye movement registration is employed, the horizontal vHIT gain values were found to vary according to the side of the eye against which the recording camera is placed. Higher gains and longer saccade latencies were found for rightward impulses when the right eye movements only were captured [Matiño-Soler et al., 2015; Rambold, 2016; Janky et al., 2018; Kim and Kim, 2018; Rey-Martinez et al., 2018].

The purpose of the study was to appraise the added benefit of RS parameters alongside the VOR gain values towards the improvement of the vHIT diagnostic accuracy. Specifically, the horizontal vHIT (hvHIT) results of patients with leftward VOR gain values <0.8, currently defined as pathological [MacDougall et al., 2009; Matiño-Soler et al., 2015; McGarvie et al., 2015; Curthoys et al., 2016; Korsager et al., 2017; Kim and Kim, 2018; Kim et al., 2018], which was registered by right eye position of the testing system camera, were retrospectively evaluated for the presence of RS. The contribution of low VOR gain values with and without the addition of RS findings to the patients’ final diagnosis of left vestibular dysfunction (VD) was analysed. The focus of our study was on the diagnosis of left horizontal canal dysfunction as the mono-ocular right eye placement of the recording camera might introduce potential bias in lower leftward VOR gain values [Matiño-Soler et al., 2015; Janky et al., 2018; Kim and Kim, 2018; Rambold, 2018; Rey-Martinez et al., 2018].

When only mono-ocular registration of the eye movements is carried out by the vHIT system, head thrusts away from the side on which the camera is located yield lower gains than that calculated for head thrusts towards the ipsilateral side of the camera [Matiño-Soler et al., 2015; Rambold, 2016; Janky et al., 2018; Kim and Kim, 2018; Rey-Martinez et al., 2018]. This might result in false-positive interpretation of the leftward vHIT gain results while using the Otometrics – Natus ICS Impulse device (Taastrup, Denmark) in which the camera captures the right eye movements alone. The present study aimed to critically evaluate the borderline leftward gain results combined with RS characteristics in the diagnosis of left horizontal semicircular canal dysfunction.

The hvHIT recordings of 40 dizzy patients with left-sided horizontal canal VOR gains <0.8 were retrospectively re-evaluated for the presence of RS. The study groups included 20 patients with a final diagnosis of left horizontal semicircular canal VD secondary to vestibular neuritis, vestibular schwannoma, and unilateral intra-tympanic gentamicin-induced vestibulopathy and 20 patients for whom vestibular disease was ruled out (NVD). The diagnosis of left vestibular disease was based on the cumulative data of the patient’s history, bedside otoneurological examination findings, videonystagmography caloric test results, vHIT findings, and imaging results. The NVD subjects had non-specific complaint of dizziness, history that was not typical for vestibular disease, normal bedside otoneurological examination as well as videonystagmography and imaging study results. Also, no persistent postural perceptual dizziness or vestibular migraine cases were included in the NVD group. The contribution of the VOR gain values, RS latency, frequency, and velocity parameters towards the prediction of left VD was assessed.

All hvHIT examinations were carried out by the same experienced right-handed examiner (M.K.) employing the Otometrics – Natus ICS Impulse device (Taastrup, Denmark) in which the camera captures the right eye movements alone. For the vHIT system used, the VOR gain is calculated as the whole area under the curve of the velocity over time recordings of the eye movement divided by that of the head. The vHIT system settings ignore microsaccades, which are characterized by eye displacement smaller than 0.25° with velocity <50°/sec. Such microsaccades might occur physiologically and are not considered to be pathological [Martinez-Conde et al., 2009]. The vHIT test was carried out as has been previously described for the Otometrics – Natus ICS Impulse system [Curthoys et al., 2016]. For each participant, 20 repetitions of the head impulse were randomly carried out on the left and right sides. The outcome measures for the hvHIT were the horizontal canal VOR gain; percentage of gain asymmetry ([rightward gain − leftward gain/rightward gain + leftward gain] × 100), peak head velocity (deg/s), RS velocity (deg/s), RS latency (ms), and RS frequency ([number of corrective saccades/number of eligible head thrusts] × 100).

42 healthy subjects with no vestibular complaints and normal otoneurological examination had vHIT testing of the horizontal semicircular canals employing the Otometrics – Natus ICS Impulse device (Taastrup, Denmark). All the examinations were carried out by the same experienced right-handed examiner (M.K.). The gain values obtained had a normal distribution averaging 0.88 ± 0.08 (mean ± standard deviation) for the left horizontal semicircular canal and 1 ± 0.14 (mean ± standard deviation) for the right semicircular canal. Taking into consideration normal range within mean ±2 standard deviations, the calculated normal ranges for the hvHIT gains are 0.72–1.04 and 0.72–1.28 for the left and right side, respectively.

The reference left horizontal semicircular canal vHIT gain value for this study was 0.8, while the cut-off pathological gain value in our laboratory is 0.72, reflecting a decrease of 10%. With an alpha error of 0.05 and statistical power of 80%, the sample size estimation was 16 subjects in each group.

Leftward hvHIT gain values, percentage of gain asymmetry, head velocity, RS velocity, latency, and frequency were compared between the VD and NVD groups by the Student’s unpaired two-tailed test or the non-parametric Mann-Whitney test. The data points which have been used were the means or medians of the 20 vHIT repetitions for the Student’s unpaired two-tailed test and the non-parametric Mann-Whitney test accordingly. Within-subjects rightward and leftward head velocities and VOR gains were compared by the Student’s paired two-tailed test or the non-parametric Wilcoxon matched-pairs signed-ranks test. Datasets were tested for normal distribution by the Shapiro-Wilk normality test, and the appropriate statistical tests were accordingly used. Proportions of the cut-off gain values, the presence of RS, detection of RS with specific velocity and frequency characteristics, and their combinations were compared between the groups employing Fisher’s exact test. Sensitivity, specificity, positive and negative predictive values towards the prediction of VD were calculated for the various outcome measures and their combinations. p values <0.05 were considered statistically significant. Statistical analysis was performed using the GraphPad InStat version 3.06 software (San Diego, CA, USA).

The VD and NVD groups each included 11 men and 9 women. The mean subjects’ age of the VD group participants was 56.45 ± 16.64 years (mean ± standard deviation) and did not significantly differ from that of the NVD group which was 50.75 ± 14.58 years (p = 0.26; unpaired t test). The final diagnosis of the 20 patients included in the VD group was vestibular neuritis in 17, vestibular schwannoma in 2, and intra-tympanic gentamicin-induced vestibulopathy in 1 patient. The diagnosis of vestibular neuritis was reached only if all the following symptoms and signs were present: acute onset of prolonged severe rotatory vertigo, the presence of spontaneous nystagmus and postural imbalance, and documentation of unilateral reduced caloric response (caloric lateralization >25%) on the caloric study of videonystagmography.

Figure 1 demonstrates left horizontal vHIT recordings of normal findings of a patient from the NVD and a patient from the VD group. For the NVD group, the average rightward hvHIT gain was 0.93 ± 0.1 (95% CI: 0.88–0.98) and the leftward gain was 0.75 ± 0.02 (95% CI: 0.74–0.77). The difference was of statistical significance (p < 0.0001; Wilcoxon matched-pairs signed-ranks test).

The rightward and leftward head velocity values did not significantly differ, reaching 210 ± 31.3 (95% CI: 195.7–224.9) and 213.9 ± 33.8 (95% CI: 198.1–229.7) deg/s, respectively (paired t-test). No significant difference was found between the leftward head velocities of the VD and NVD groups registered as 207.3 ± 33.2 (95% CI: 191.8–222.8) and 213.9 ± 33.8 (95% CI: 198.1–229.7), respectively (unpaired t test).

The average left hvHIT gain of the VD group was significantly lower as compared to the NVD group (0.445 ± 0.23 vs. 0.757 ± 0.024; p < 0.0001, unpaired t test; shown in Fig. 2), and the gain asymmetry average, where higher gains were recorded for the rightward stimuli, was significantly higher (39.2 ± 24.11 vs. 18.12 ± 9.05%; p < 0.002, unpaired t test; shown in Fig. 3).

The calculated normal range in our laboratory for the right hvHIT gain is 0.72–1.28 and 0.72–1.04 for the left hvHIT gain. Gain values ≥0.72 were found in all NVD patients and in 4 (20%) patients of the VD group (p < 0.0001; Fisher’s exact test).

Significantly higher left-sided RS velocity and frequency were found in the VD group (221.35 ± 89.18 vs. 131 ± 42.93 deg/s; p < 0.0002 and 83.2 ± 33.2 vs. 22.5 ± 26.0%; p < 0.0001, Mann-Whitney test, respectively; shown in Fig. 4, 5). No differences between the groups were found in the RS latency (196.71 ± 52.13 vs. 250.82 ± 114.78 msec for the VD and NVD groups, respectively; not significant by the unpaired t test, Welch corrected for different standard deviations). RSs were detected in 19 (95%) of the VD patients and 12 (60%) of the NVD group (p < 0.02; Fisher’s exact test). Among the 19 VD patients in whom RSs were detected, 17 (89%) had RS frequency >80%, while in all the 12 NVD patients having RS, the frequency value was <80% (p < 0.0001; Fisher’s exact test).

RS velocity was higher than 150 deg/s in 17 of the 19 VD patients (89%) and in only 3 of the 12 NVD patients (25%) having RS (p < 0.0004; Fisher’s exact test). 17 of the VD group out of 19 (89%) and 1 of the NVD group out of 12 (8%) had RS with both frequency >80% and velocity >150 deg/sec (p < 0.0001; Fisher’s exact test).

15 of the VD group participants (75%) had gain values <0.72 and RS frequency >80%, while in none of the NVD group, such association was recorded (p < 0.0001; Fisher’s exact test). 15 of the VD group (75%) and 3 NVD group participants (15%) presented with gain values <0.72 and saccade velocity >150 deg/s (p < 0.0004; Fisher’s exact test).

Participants’ data and the leftward vHIT results are detailed in Table 1. The combination of gain values <0.72, saccade frequency >80%, and saccade velocity >150 deg/sec was found in 15 (75%) of the VD group and none of the NVD group (p < 0.0001; Fisher’s exact test). The sensitivity, specificity, positive predictive value, and negative predictive value of the following outcome measures towards the prediction of VD are detailed in Table 2.

vHIT systems employing mono-ocular eye movements recording have an inherited right-left imbalance regarding the horizontal semicircular canal gain values, depending on the location of the camera. A study including 212 healthy subjects of all ages showed a mean 9.1% gain value difference in favour of the right side in all age groups while using right eye recording system [Matiño-Soler et al., 2015]. Another recent study of commercially available vHIT systems reported on 5% rightward gain bias when the right eye movements only were captured [Strupp et al., 2018]. While using scleral search coils for accurate binocular eye movements recording, it was found that the horizontal VOR gain for the adducting eye exceeded the gain of the abducting eye by an average of 15.3% [Weber et al., 2008]. Several explanations have been offered for this right-left imbalance, a longer neural pathway (trisynaptic) for the adducting medial rectus compared to the shorter (disynaptic) pathway of the abducting lateral rectus when the system camera captures the right eye movements alone [Weber et al., 2008; Matiño-Soler et al., 2015]. Alternatively, the higher gain might stem from the relative decrease in head velocity while a right-handed examiner uses his weaker left hand for the conduction of the rightward head thrust. Others have attributed the higher rightward gain to asymmetry in the maximal active force of the medial rectus when compared to the lateral rectus muscle which is about 26% greater for the adducting muscle [Collins et al., 1981; Kim and Kim, 2018].

Although we have found in the NVD group a significant gain difference in favour of the rightward head impulse which was conducted by a right-handed examiner, the head velocities on both sides were comparable. Thus, for our study, the higher right-sided gain is not explained by possible examiner-related factors.

The reported bias in favour of lower left hvHIT gain when right eye movements alone are recorded, which is also supported by our results in the NVD group, might introduce false-positive results of left-sided vestibulopathy, especially when the VOR gain is used as the sole diagnostic criterion. In the current study, we evaluated the added benefit of RS in reaching a final diagnosis of left horizontal semicircular canal hypofunction using the right mono-ocular ICS impulse vHIT system.

The recommended published VOR gain cut-off value discriminating normal and pathological horizontal semicircular canal function is 0.8 [MacDougall et al., 2009; Matiño-Soler et al., 2015; McGarvie et al., 2015; Curthoys et al., 2016; Korsager et al., 2017; Kim and Kim, 2018; Kim et al., 2018]. Based on our laboratory norms, we have looked at a cut-off value of 0.72. Significantly lower average left-sided VOR gain and higher gain asymmetry was found for the VD group, and all the NVD group participants had left hvHIT gain >0.72. Still, the sensitivity of gain <0.72 alone towards the diagnosis of VD was only 80%.

The existence of RS has been previously reported in healthy subjects, with increased frequency and velocity with higher age and lower gain values [Matiño-Soler et al., 2015; Li et al., 2020]. In our study, the mere presence of RS had 100% sensitivity but only 40% specificity towards the diagnosis of VD. Recent research has suggested RS frequency of 80% as a parameter for their consistency [Perez-Fernandez and Eza-Nuñez, 2015]. Adding the restriction parameter of RS frequency >80% increased the hvHIT specificity to 100% with 89% sensitivity.

Various RS velocity criteria have been proposed to distinguish normal versus pathological vestibular function. These include critical RS velocity of 50 [Perez-Fernandez and Eza-Nuñez, 2015], 100 [Li et al., 2020], 133 [Schubert et al., 2006], and 135 deg/s [Janky et al., 2018]. Although RS average velocity was significantly higher in the VD group and significantly higher proportion of VD patients demonstrated RS with velocity >150 deg/s, this RS velocity criterion alone or adding it to the RS frequency >80% parameter did not improve the vHIT prediction values.

Based on our results, we recommend careful re-consideration of the hvHIT test results when left-sided VOR gain values <0.8 are merely recorded. Gain <0.72 is highly specific for the diagnosis of VD. However, for gain values in the range of 0.72–0.79, the presence of RS with frequency >80% would largely improve the hvHIT diagnostic accuracy (shown in Fig. 6).

Our study limitations include its retrospective nature and the relatively small size of the cohort included. Although vestibular disease was ruled out in our NVD subjects, the ideal control group should have comprised healthy subjects with no complaints of dizziness, which we found difficult to recruit. Most (85%) of the VD group have suffered from vestibular neuritis. Still, the inclusion of other diagnoses might have introduced some disparity in our results.

The study protocol was approved, and an exempt was granted from informed consent procedure by the Committee for Human Experiments, Meir Medical Center, Kfar Saba, Israel (confirmation number 009/19/COM1). Personal data were confidentially treated.

The authors have no conflicts of interest to declare.

No funding was received.

Nabil Faranesh designed the experimental set-up, collected the data, and participated in its analysis and interpretation. He wrote the manuscript draft and approved its final version. Khaldon Abo-Saleh participated in the collection of the data, contributed to its analysis, and approved the manuscript. Margalith Kaminer performed the video head impulse tests, participated in the data analysis, and approved the final version of the manuscript. Avi Shupak conceived the study and the experimental design and participated in data analysis and interpretation. He critically revised the manuscript and approved the version submitted for publication.

All data generated or analysed during this study are included in this article. Further enquiries can be directed to the corresponding author.