ABSTRACT
Objective
To determine whether repetition‐based saccadic fatigability measured with three‐dimensional video‐oculography (VOG) distinguishes myasthenia gravis (MG) from unilateral ocular motor cranial nerve palsy (CNP) in patients with diplopia.
Methods
In this prospective cross‐sectional diagnostic study conducted from August 2022 to July 2024, consecutive patients with MG or unilateral third, fourth, or sixth CNP, and healthy controls underwent repetitive horizontal and vertical saccade testing (±°15 targets; 75 cycles at 0.25 Hz). Prespecified index metrics were percentage change in saccadic range and time to target acquisition from the second saccade to the mean of the last five saccades. Diagnostic discrimination was assessed using receiver operating characteristic (ROC) curves.
Results
The cohort comprised 231 participants: 93 with MG, 54 with CNP and 84 healthy controls. MG showed greater repetition‐related range decrement and longer time to target acquisition than healthy controls and axis‐matched CNP. In the horizontal comparison (MG vs. sixth CNP), AUCs were 0.800 for percentage range change and 0.787 for percentage time‐to‐target change. In the vertical comparison (MG vs. third/fourth CNP), percentage range change showed the best discrimination (AUC 0.868, 95% CI 0.783 to 0.934); a threshold of −5.95% yielded 79.6% sensitivity and 87.5% specificity. CNP showed early baseline slowing, whereas MG showed progressive hypometria with relatively preserved peak velocity.
Conclusions
Repetition‐based VOG captures a dynamic fatigability signature in MG that differs from the more static deficit of unilateral ocular motor CNP. Quantitative saccadic fatigability may provide a useful objective adjunct to the neuro‐ophthalmic evaluation of diplopia.
Keywords: cranial nerve diseases, diplopia, eye movements, myasthenia gravis, saccades, video‐oculography
1. Introduction
Myasthenia gravis (MG) and unilateral ocular motor cranial nerve palsy (CNP) are among the most common causes of acquired binocular diplopia encountered in neuro‐ophthalmic practice [1, 2, 3, 4]. MG produces fluctuating weakness of the extraocular muscles and levator palpebrae, whereas third, fourth, and sixth nerve palsies more typically produce a direction‐specific ocular misalignment arising from a relatively fixed paresis [1, 3, 4]. In daily practice, however, the distinction is not always straightforward. Ocular MG may present without generalised features; serological results may be negative, and bedside examination may capture only a single time point within a fluctuating course [5]. Conversely, partial CNP may resemble incomplete ophthalmoplegia from MG.
Because the diagnostic pathways often differ, early separation of these entities has practical consequences. Suspected MG prompts confirmatory serology or electrophysiology and may alter treatment decisions, whereas suspected CNP may warrant etiological evaluation and selective neuroimaging when a microvascular cause cannot be confidently established on clinical grounds. Static alignment and duction testing remain central to the examination, but they do not quantify the defining physiological feature of MG: fatigability during repeated activation.
Video‐oculography (VOG) offers a way to measure ocular motor behaviour across repeated trials with greater precision than bedside fatigue manoeuvres [6, 7, 8]. Prior work suggests that repetitive saccade paradigms can detect range decrement and delayed target acquisition in MG, but the available studies have been small and have not directly addressed the clinically relevant differential diagnosis of unilateral ocular motor CNP [9, 10]. Crucially, no prospective study has evaluated these metrics against CNP as the comparator—the principal alternative diagnosis that may prompt selective neuroimaging or structural evaluation, depending on clinical context, rather than serological or electrophysiological testing. We therefore evaluated whether repetition‐based saccadic fatigability measured using three‐dimensional VOG distinguishes MG from unilateral third, fourth and sixth nerve palsies. We hypothesised that MG would show progressive deterioration across repeated saccades, whereas CNP would show baseline deficits with comparatively little additional decline.
2. Methods
2.1. Study Design and Participants
This prospective cross‐sectional diagnostic study was conducted at Jeonbuk National University Hospital from August 2022 through July 2024. The protocol was approved by the institutional review board of Jeonbuk National University Hospital (IRB No 2022‐04‐044‐001), and all participants provided written informed consent. The study followed the Declaration of Helsinki and is reported in accordance with STARD 2015. The completed STARD 2015 checklist is provided as Table S5.
We enrolled three groups: patients with MG, patients with unilateral ocular motor CNP, and healthy controls. Patients with MG were eligible if they had a neurologist‐confirmed diagnosis supported by at least one standard diagnostic test. Patients with CNP were eligible if they had unilateral third, fourth, or sixth nerve involvement with clinically evident extraocular movement limitation; the CNP cohort was not restricted to idiopathic or magnetic resonance imaging (MRI)‐negative palsies. Neuroimaging and clinical work‐up were performed when clinically indicated to identify the underlying aetiology and to exclude non‐CNP mimics. Healthy controls were recruited from the community and hospital staff and were excluded if they had neurological or ophthalmic disorders affecting eye movements. Across all groups, participants were excluded if they had other neurological or ophthalmic diseases known to alter ocular motor function. Sample size was pragmatic and based on the number of eligible participants recruited during the study period.
2.2. Reference Standard Diagnosis and Masking
Myasthenia gravis (MG) was diagnosed by a senior neurologist on the basis of clinical examination and at least one supportive result from routine clinical care: decrement on repetitive nerve stimulation, positive acetylcholine receptor antibody, positive muscle‐specific kinase antibody, or a positive neostigmine response [11]. CNP was diagnosed by clinical examination, with neuroimaging and clinical work‐up used when clinically indicated to identify the underlying aetiology and exclude non‐CNP mimics; cases were classified by the involved nerve and by recorded aetiology [1]. The CNP group was therefore an aetiologically heterogeneous clinical comparator rather than a cohort restricted to idiopathic or MRI‐negative palsies.
Clinicians responsible for the reference‐standard diagnosis were blinded to VOG fatigability metrics and did not use them for diagnostic classification. VOG analysts and statisticians were blinded to the diagnostic group through coded deidentified datasets. The intended clinical role of the index test was as an adjunct to routine evaluation in patients with diplopia when both MG and unilateral CNP were plausible.
Diagnostic assignment was based on the reference‐standard clinical diagnosis established by the treating neuro‐ophthalmologists using routine clinical assessment and ancillary investigations, including serological testing, electrophysiological studies, pharmacological or therapeutic response, and neuroimaging when clinically indicated. The clinicians responsible for diagnostic classification were blinded to the VOG‐derived fatigability metrics, and VOG results were not used for diagnostic assignment. Available follow‐up records were reviewed to identify any subsequent diagnostic cross‐over between ocular MG and CNP groups. No such cross‐over reassignment was identified; therefore, the original diagnostic grouping was retained for the primary analysis.
2.3. Video‐Oculography Protocol
Saccadic eye movements were recorded with binocular three‐dimensional VOG (SLMED, Seoul, Korea) at 120 Hz in a darkened room while participants wore VOG goggles. A red target was projected on a screen. Five‐point calibration comprised a central fixation target and four eccentric targets located at 15° left, 15° right, 15° up and 15° down, matching the eccentricity of the saccade targets used in the repetitive paradigm. No predefined minimum ocular‐excursion or single‐saccade amplitude threshold was used as an inclusion criterion; however, participants had to yield reliable calibration/gaze tracking and sufficient analysable data for inclusion in the final analysis. Participants were instructed to abstain from caffeine and alcohol for 24 h before testing and, when clinically safe, to withhold acetylcholinesterase inhibitors for 24 h and corticosteroids for 72 h before recording [12, 13]. In participants with clinically significant ptosis, the upper eyelid was secured with hypoallergenic tape before VOG recording to maintain pupil exposure and support reliable vertical pupil tracking.
Each session consisted of 75 consecutive cycles of horizontal saccades and 75 consecutive cycles of vertical saccades at 0.25 Hz, with targets placed at ±15° from centre. The 75‐cycle duration was selected a priori based on previous repetitive‐saccade fatigability paradigms and feasibility considerations, with the aim of providing sustained activation while maintaining tolerability and recording quality; it was not optimised from the present diagnostic dataset. One cycle was defined as a full 30° back‐and‐forth movement between the two peripheral targets without stopping at the central position; therefore, 75 cycles comprised 150 directional saccades per axis. The order of horizontal and vertical tasks was randomly assigned and counterbalanced, with a 3‐min rest interval between tasks. To preserve clinical interpretability, analyses were axis matched to the expected deficit. Horizontal metrics were compared between MG and sixth nerve palsy, whereas vertical metrics were compared between MG and third or fourth nerve palsy. For CNP, analyses used the clinically affected eye. For MG, the more affected eye was analysed, defined a priori as the eye with the greater decrement in saccadic range across repetitions [7]. In healthy controls, the right eye was analysed. To explore intra‐series performance changes, we additionally generated descriptive trial‐by‐trial trajectories of saccadic range and peak velocity across the full 75‐cycle sequence.
The repeated‐saccade task consisted of visually guided saccades to a fixed 15° target displacement. Random‐amplitude saccades were not included in this protocol. Therefore, nominal target amplitude was controlled by design. However, because the actually executed saccadic range could be reduced in patients with hypometria or fatigability, we performed post hoc amplitude‐adjusted analyses to evaluate whether group differences in peak velocity persisted after accounting for measured saccadic range. Peak velocity was used as the dependent variable, and measured saccadic range and diagnostic group were included as independent variables. Separate models were fitted for horizontal and vertical saccades and for the second‐saccade and last‐five‐mean measurements. These analyses were considered sensitivity analyses rather than formal main sequence modelling, because the protocol used a single fixed target displacement rather than multiple target amplitudes.
2.4. Kinematic Parameters and Fatigability Metrics
Data were processed offline using custom MATLAB scripts (MathWorks). Saccade onset was defined as eye velocity exceeding 30°/s and offset as velocity falling below 30°/s [14, 15, 16]. Prespecified kinematic parameters were saccadic range, peak velocity, latency, and time to target acquisition. Time to target acquisition was defined as the interval from saccade onset to the first point at which gaze remained within ±1° of the target for at least 100 ms, including corrective saccades required before stabilisation [17, 18, 19].
Trials were excluded using prespecified criteria: blink after target onset, loss of fixation, incorrect saccade direction, or saccadic gain outside 0.4 to 1.5 [20, 21]. To quantify fatigability, the second saccade was used as baseline to minimise initial acclimatisation effects, and the mean of the last five saccades was used as the fatigued state [9, 10]. Percentage change was calculated as (mean of last 5 − second)/second × 100. The primary index metrics were percentage change in saccadic range and time to target acquisition.
2.5. Statistical Analysis
Continuous variables are summarised as mean (SD) or median (IQR), depending on distribution. Parametric comparisons used Welch t‐tests or Welch analysis of variance, with Holm‐Bonferroni adjustment for multiple pairwise comparisons. Non‐parametric variables were compared using the Mann–Whitney U test. Categorical variables were compared using the χ 2 test or Fisher exact test. Because peak saccadic velocity depends on saccadic amplitude, we performed post hoc amplitude‐adjusted linear regression analyses with peak velocity as the dependent variable and measured saccadic range and diagnostic group as independent variables. Models were fitted separately for horizontal and vertical saccades and for the second saccade and the mean of the last five saccades. These exploratory analyses are reported in the Table S3. To examine intra‐series behaviour descriptively, saccadic range and peak velocity were also plotted across trial order over the full 75‐cycle sequence and summarised as group mean with SD at each cycle; this visualisation was exploratory and was not used for threshold selection or formal subject‐level fluctuation analysis (Figure S3).
Diagnostic discrimination was assessed using receiver operating characteristic curve analyses with MG defined as the positive class. AUCs with 95% CIs were estimated using the DeLong method. Because prespecified operating thresholds were not available, candidate cut points were selected using the Youden index. Internal validation used 1000 stratified bootstrap replicates with threshold optimisation in each replicate. No imputation was performed. Participants without analysable VOG recordings were excluded from the final analysis; this participant‐level exclusion reflected calibration/recording quality or inability to complete the protocol, not a predefined minimum saccadic‐amplitude threshold. Sensitivity analyses repeated the receiver operating characteristic analyses in high‐certainty MG, defined by seropositivity and/or decrement on repetitive nerve stimulation. All tests were two‐sided, and p < 0.05 was considered statistically significant. Analyses were conducted using SPSS Statistics V.23.0 (IBM) and Python V.3.11.3 (scikit‐learn).
3. Results
3.1. Participants and Clinical Characteristics
Of 382 individuals assessed for eligibility, 265 were enrolled after providing informed consent; 117 were excluded before enrolment because they did not meet the inclusion criteria, had another neurological or ophthalmic disorder, or declined to participate and 34 were excluded after enrolment because of calibration failure, inability to complete the protocol, or insufficient analysable VOG data (Figure S1). The final analytic cohort comprised 231 participants: 93 with MG, 54 with ocular motor CNP and 84 healthy controls. Among participants with MG, 61 (66%) had ocular MG and 32 (34%) had generalised MG. Within the CNP group, 22 participants (41%) had sixth nerve palsy, 18 (33%) had third nerve palsy and 14 (26%) had fourth nerve palsy. The CNP group was not restricted to idiopathic or MRI‐negative cases: presumed vascular palsies were most common (39/54, 72%), followed by inflammatory, demyelinating or immune‐mediated causes, including multiple sclerosis, Tolosa‐Hunt syndrome or Miller‐Fisher syndrome (8/54, 15%), aneurysm (4/54, 7%) and neoplasm (3/54, 6%); no traumatic or undetermined cases were included. Clinical characteristics are summarised in Table 1.
TABLE 1.
Baseline clinical characteristics of study participants.
| MG (n = 93) | CNP (n = 54) | HCs (n = 84) | ||||||
|---|---|---|---|---|---|---|---|---|
| Ocular MG (n = 61) | Generalised MG (n = 32) | Total (n = 93) | 6th CNP (n = 22) | 3rd CNP (n = 18) | 4th CNP (n = 14) | Total (n = 54) | HCs (n = 84) | |
| Age, mean (SD), y | 55.8 (14.1) | 52.4 (17.3) | 54.6 (15.2) | 59.8 (15.3) | 59.1 (14.1) | 54.4 (6.7) | 58.2 (13.1) | 46.9 (12.4) |
| Sex | ||||||||
| Female, No. (%) | 25 (41) | 20 (62) | 45 (48) | 8 (36) | 8 (44) | 4 (29) | 20 (37) | 46 (54) |
| Male, No. (%) | 36 (59) | 12 (38) | 48 (52) | 14 (63) | 10 (56) | 10 (71) | 34 (63) | 40 (46) |
| Clinical presentation | ||||||||
| Ptosis, No. (%) | 49 (80) | 21 (66) | 70 (75) | 0 (0) | 11 (61) | 0 (0) | 11 (20) | — |
| Diplopia, No. (%) | 54 (89) | 19 (59) | 73 (79) | 22 (100) | 13 (72) | 14 (100) | 49 (91) | — |
| Bulbar symptoms, No. (%) | 0 (0) | 21 (66) | 21 (23) | 0 (0) | 1 (6) | 0 (0) | 1 (2) | — |
| Limb weakness, No. (%) | 0 (0) | 21 (66) | 21 (23) | 0 (0) | 0 (0) | 0 (0) | 0 (0) | — |
| Disease duration, mean (SD), mo | 55.1 (80.1) | 75.2 (76.7) | 62.0 (79.1) | 9.1 (21.3) | 15.7 (23.9) | 3.6 (6.5) | 9.8 (19.8) | |
| Antibody test | ||||||||
| AChR antibody positive, No. (%) | 44 (72) | 25 (78) | 69 (74) | — | — | — | — | — |
| MuSK antibody positive, No. (%) | — | 1 (3) | 1 (1) | — | — | — | — | — |
| AChR antibody titre, mean (SD), nM | 3.1 (4.3) | 6.8 (6.2) | 4.4 (5.3) | — | — | — | — | — |
| RNS test positive, No. (%) | 19 (31) | 23 (72) | 42 (45) | — | — | — | — | — |
| Neostigmine test positive, No. (%) | 21/36 (58) | 4/11 (36) | 25/47 (53) | — | — | — | — | — |
| Chest CT | ||||||||
| Thymoma, No. (%) | 11 (18) | 5 (16) | 16 (17) | — | — | — | — | — |
| Thymic hyperplasia, No. (%) | 6 (10) | 9 (28) | 15 (16) | — | — | — | — | — |
| Not tested, No. (%) | 5 (8) | 3 (9) | 8 (9) | — | — | — | — | — |
| Laterality of CNP | ||||||||
| Left/Right/Bilateral | — | — | — | 12 (55)/8 (36)/2 (9) | 10 (55)/7 (39)/1 (6) | 7 (50)/7 (50)/0 (0) | 29 (54)/22 (41)/3 (5) | — |
| Aetiology | ||||||||
| Vascular (microischemic) | — | — | — | 16 (73) | 10 (55) | 13 (93) | 39 (72) | — |
| Head trauma | — | — | — | 0 (0) | 0 (0) | 0 (0) | 0 (0) | — |
| Neoplasm | — | — | — | 2 (9) | 1 (6) | 0 (0) | 3 (6) | — |
| Aneurysm | — | — | — | 0 (0) | 4 (22) | 0 (0) | 4 (7) | — |
| Other (e.g., MS, THS, MFS) | — | — | — | 4 (18) | 3 (17) | 1 (7) | 8 (15) | — |
| Undetermined | — | — | — | 0 (0) | 0 (0) | 0 (0) | 0 (0) | — |
Note: Data are presented as mean (SD) or No. (%). — indicates not applicable. Percentages for bulbar symptoms and limb weakness in MG are calculated against the total MG cohort (n = 93).
Abbreviations: AChR, acetylcholine receptor; CNP, cranial nerve palsy; CT, computed tomography; HCs, healthy controls; MFS, Miller‐Fisher syndrome; MG, myasthenia gravis; MS, multiple sclerosis; MuSK, muscle‐specific kinase; RNS, repetitive nerve stimulation; THS, Tolosa‐Hunt syndrome.
3.2. Baseline Saccadic Performance and Repetition‐Related Change
Analyses were axis matched to the expected motility deficit. Selected baseline and fatigability metrics are shown in Table 2, and the full kinematic dataset is provided in Figure S2; effect sizes for all pairwise comparisons are provided in Tables S1 and S2. On the second saccade, sixth nerve palsy showed reduced horizontal saccadic range and peak velocity compared with MG and healthy controls. Similarly, third nerve palsy showed marked reductions in vertical range and peak velocity at baseline, while fourth nerve palsy showed smaller but directionally similar reductions. These findings were consistent with a relatively fixed paretic deficit present from the outset of testing.
TABLE 2.
Baseline and repetition‐related saccadic parameters by diagnostic group.
| Panel A. Horizontal saccadic parameters | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| MG (n = 93) | 6th CNP (n = 22) a | HCs (n = 84) | Adjusted p value c | ||||||
| Ocular MG (n = 61) | Generalised MG (n = 32) | p | Total (n = 93) | MG vs HCs | 6th CNP vs HCs | MG vs 6th CNP | |||
| Second saccade | |||||||||
| Range, mean (SD), ° | 28.3 (2.9) | 28.8 (1.8) | 0.305 | 28.5 (2.6) | 21.8 (6.9) | 29.6 (2.3) | 0.003 | < 0.001 | < 0.001 |
| Peak velocity, mean (SD), °/s | 463.7 (70.0) | 471.9 (46.5) | 0.504 | 466.5 (62.7) | 347.2 (121.9) | 478.4 (59.4) | 0.202 | < 0.001 | < 0.001 |
| Latency, mean (SD), ms | 241.7 (43.8) | 230.2 (36.0) | 0.178 | 237.7 (41.4) | 244.7 (65.3) | 205.0 (28.7) | < 0.001 | 0.021 | 0.635 |
| Time to target acquisition, mean (SD), ms | 109.5 (42.4) | 113.9 (31.7) | 0.58 | 111.0 (39) | 152.9 (109.1) | 103.2 (21.2) | 0.179 | 0.136 | 0.179 |
| Mean of last 5 saccades | |||||||||
| Range, mean (SD), ° | 23.5 (4.3) | 26.1 (3.4) | 0.002 | 24.4 (4.2) | 21.5 (6.4) | 29.6 (3.3) | < 0.001 | < 0.001 | 0.052 |
| Peak velocity, mean (SD), °/s | 451.6 (83.3) | 452.3 (64.6) | 0.967 | 451.9 (77) | 333.7 (107.8) | 461.7 (67.3) | 0.367 | < 0.001 | < 0.001 |
| Latency, mean (SD), ms | 238.1 (43.6) | 234.0 (45.7) | 0.679 | 236.7 (44.1) | 230.8 (49.2) | 205.4 (42.6) | < 0.001 | 0.07 | 0.612 |
| Time to target acquisition, mean (SD), ms | 196.6 (88.0) | 205.9 (82.5) | 0.618 | 199.8 (85.8) | 160.3 (111.5) | 114 (28.6) | < 0.001 | 0.134 | 0.134 |
| Difference between second saccade and mean of last 5 saccades a | |||||||||
| Range, mean (SD), ° | −4.8 (3.8) | −2.7 (3.1) | 0.005 | −4.1 (3.7) | −0.3 (2.1) | −0.6 (3.3) | < 0.001 | 0.643 | < 0.001 |
| Peak velocity, mean (SD), °/s | −12.0 (87.6) | −19.6 (53.2) | 0.609 | −14.6 (77.3) | −13.5 (66) | −16.6 (37.5) | 1 | 1 | 1 |
| Latency, mean (SD), ms | −3.6 (40.8) | 3.9 (34.0) | 0.353 | −1.0 (38.6) | −13.9 (65.5) | 0.5 (36.3) | 0.995 | 0.995 | 0.995 |
| Time to target acquisition, mean (SD), ms | 87.1 (86.2) | 92.0 (88.1) | 0.798 | 88.8 (86.4) | 7.4 (59.4) | 10.8 (20.7) | < 0.001 | 0.793 | < 0.001 |
| Panel B. Vertical saccadic parameters | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MG (n = 93) | 3rd CNP (n = 18) b | 4th CNP (n = 14) b | HCs (n = 84) | Adjusted p value c | ||||||||
| Ocular MG (n = 61) | Generalised MG (n = 32) | p | Total (n = 93) | MG vs HCs | 3rd CNP vs HCs | 4th CNP vs HCs | MG vs 3rd CNP | MG vs 4th CNP | ||||
| Second saccade | ||||||||||||
| Range, mean (SD), ° | 28.4 (2.8) | 29.3 (2.5) | 0.126 | 28.7 (2.7) | 20.1 (6.1) | 27.4 (1.5) | 29.4 (2.4) | 0.064 | < 0.001 | < 0.001 | < 0.001 | 0.024 |
| Peak velocity, mean (SD), °/s | 442.4 (82.9) | 465.9 (59.8) | 0.12 | 450.5 (76.2) | 292.4 (100.0) | 397.0 (72.3) | 469.8 (67.7) | 0.079 | < 0.001 | 0.008 | < 0.001 | 0.04 |
| Latency, mean (SD), ms | 249.1 (48.1) | 234.7 (43.6) | 0.149 | 244.1 (46.9) | 255.7 (45.9) | 216.1 (55.2) | 210.3 (45.0) | < 0.001 | 0.003 | 0.715 | 0.681 | 0.27 |
| Time to target acquisition, mean (SD), ms | 112.8 (34.8) | 117.5 (26.5) | 0.466 | 114.4 (32.2) | 166.1 (64.8) | 163.0 (63.6) | 114.0 (29.4) | 0.922 | 0.018 | 0.04 | 0.018 | 0.04 |
| Mean of last 5 saccades | ||||||||||||
| Range, mean (SD), ° | 23.9 (3.9) | 25 (3.5) | 0.16 | 24.3 (3.8) | 20.3 (6.6) | 27.2 (1.8) | 28.8 (3.4) | < 0.001 | < 0.001 | 0.026 | 0.026 | < 0.001 |
| Peak velocity, mean (SD), °/s | 415.5 (76.8) | 448.7 (63.2) | 0.029 | 426.9 (73.8) | 279.0 (123.8) | 381.6 (70.6) | 461.1 (69.0) | 0.004 | < 0.001 | 0.003 | < 0.001 | 0.039 |
| Latency, mean (SD), ms | 255.8 (51.9) | 234.1 (50.9) | 0.057 | 248.4 (52.3) | 238.5 (65.9) | 200.9 (43) | 210.1 (44.1) | < 0.001 | 0.289 | 0.946 | 0.946 | 0.005 |
| Time to target acquisition, mean (SD), ms | 203.0 (71.3) | 195.4 (60.9) | 0.593 | 200.3 (67.7) | 184.8 (62.5) | 173.2 (61.9) | 121.5 (30.1) | < 0.001 | 0.002 | 0.025 | 0.349 | 0.297 |
| Difference between second saccade and mean of last 5 saccades a | ||||||||||||
| Range, mean (SD), ° | −4.5 (3.5) | −4.26 (3.1) | 0.744 | −4.4 (3.4) | 0.1 (1.7) | −0.2 (0.9) | −0.6 (3.1) | < 0.001 | 0.363 | 0.363 | < 0.001 | < 0.001 |
| Peak velocity, mean (SD), °/s | −26.9 (70.0) | −17.11 (67.6) | 0.517 | −23.5 (69) | −13.4 (57.4) | −15.4 (34.1) | −8.6 (50.3) | 0.516 | 1 | 1 | 1 | 1 |
| Latency, mean (SD), ms | 6.78 (51.6) | −0.49 (45.0) | 0.483 | 4.3 (49.3) | −17.2 (72.1) | −15.1 (31.2) | −0.1 (40.2) | 0.721 | 0.721 | 0.509 | 0.721 | 0.294 |
| Time to target acquisition, mean (SD), ms | 90.1 (69.7) | 74.7 (58.0) | 0.26 | 84.8 (66) | 18.7 (35.8) | 10.2 (26.7) | 8.7 (28.0) | < 0.001 | 0.551 | 0.846 | < 0.001 | < 0.001 |
Abbreviations: CNP, cranial nerve palsy; HCs, healthy controls; MG, myasthenia gravis; SD, standard deviation.
Values for sixth cranial nerve palsy are from the clinically affected eye or, in bilateral cases, the more affected eye.
Values for third and fourth cranial nerve palsy are from the clinically affected eye or, in bilateral cases, the more affected eye.
p values were derived from Welch t‐tests and adjusted for multiple comparisons using the Holm‐Bonferroni method.
Post hoc amplitude‐adjusted regression analyses showed the same qualitative pattern. After adjustment for measured saccadic range, peak velocity did not differ significantly between MG and healthy controls at either assessed time point, whereas peak velocity remained significantly lower in sixth nerve palsy for horizontal saccades and in third and fourth nerve palsies for vertical saccades. MG also showed significantly higher amplitude‐adjusted peak velocity than the axis‐matched CNP groups (Table S3).
Across repetitions, the dominant phenotype in MG was progressive fatigability. Compared with healthy controls and axis‐matched CNP, MG showed larger decrements in saccadic range from the second saccade to the mean of the last five saccades (horizontal, −4.1° in MG vs. −0.3° in sixth nerve palsy and −0.6° in controls; vertical, −4.4° in MG vs. +0.1° in third nerve palsy, −0.2° in fourth nerve palsy and −0.6° in controls). MG also showed larger increases in time to target acquisition (horizontal, +88.8 ms in MG vs. +7.4 ms in sixth nerve palsy and +10.8 ms in controls; vertical, +84.8 ms in MG vs. +18.7 ms in third nerve palsy, +10.2 ms in fourth nerve palsy and +8.7 ms in controls). Peak velocity changed less across repetitions than range, particularly in MG, suggesting increasing hypometria with greater reliance on corrective saccades rather than marked slowing of the initial saccade. Exploratory trial‐by‐trial visualisation across the full 75‐cycle sequence supported this interpretation: healthy controls showed stable range and peak velocity, MG showed gradual range decrement with variation in timing and degree across the sequence, and axis‐matched CNP groups showed lower initial range and/or peak velocity with comparatively flat trajectories; a consistent rebound‐type decline–recovery–decline pattern was not evident at the group level (Figure S4).
3.3. Diagnostic Performance
Receiver operating characteristic analyses evaluated repetition‐based fatigability metrics derived from horizontal and vertical saccades (Figure 1 and Table 3). In the axis‐matched horizontal comparison, percentage change in saccadic range and percentage change in time to target acquisition showed similar performance for distinguishing MG from sixth nerve palsy (AUC 0.800, 95% CI 0.700 to 0.885 and AUC 0.787, 95% CI 0.705 to 0.867, respectively). In the axis‐matched vertical comparison, percentage change in vertical saccadic range had the highest discrimination for MG versus third or fourth nerve palsy (AUC 0.868, 95% CI 0.783 to 0.934). A threshold of −5.95% yielded sensitivity of 79.6% and specificity of 87.5%, with a positive likelihood ratio of 6.28 and a negative likelihood ratio of 0.23. Percentage change in time to target acquisition also performed well (AUC 0.836, 95% CI 0.760 to 0.906). Distributions of the principal fatigability metrics are shown in Figure 2, and predictive value curves across disease prevalence are shown in Figure S3.
FIGURE 1.
Receiver operating characteristic curves for fatigability metrics derived from repetitive saccade testing. Panels show percent change in saccadic range (ΔRange; A and C) and percent change in time to target acquisition (ΔTime‐to‐target; B and D) for horizontal (A and B) and vertical (C and D) saccades. Δ metrics were calculated as percent change from the second saccade to the mean of the last 5 saccades: (mean of last 5 − second)/second × 100. Curves compare myasthenia gravis (MG) vs. healthy controls (HC) and myasthenia gravis (MG) vs. axis‐matched cranial nerve palsy (CNP) (horizontal: Sixth CNP; vertical: Third or fourth CNP). Solid curves indicate the primary estimates; faint curves show ROC curves from 1000 bootstrap resamples; the dashed diagonal indicates no discrimination. Area under the curve (AUC) values are shown in each panel. Abbreviations: AUC, area under the curve; CNP, cranial nerve palsy; HC, healthy control; MG, myasthenia gravis.
TABLE 3.
Diagnostic performance of repetition‐related saccadic fatigability metrics for distinguishing Myasthenia Gravis from cranial nerve palsy and healthy controls.
| AUC (95% CI) a | Threshold (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) | LR+ (95% CI) | LR‐ (95% CI) | |
|---|---|---|---|---|---|---|---|---|
| MG vs healthy controls (horizontal) | ||||||||
| ΔRange | 0.836*** (0.762–0.897) | −7.310 (−8.250–3.670) | 0.742 (0.645–0.828) | 0.988 (0.926–1.000) | 0.986 (0.928–1.000) | 0.764 (0.708–0.832) | 33.097 (9.332–65.323) | 0.269 (0.176–0.359) |
| ΔPeak velocity | 0.523 (0.432–0.606) | −6.240 (−15.380–5.520) | 0.441 (0.161–0.570) | 0.765 (0.642–0.963) | 0.691 (0.600–0.871) | 0.542 (0.492–0.596) | 1.947 (1.306–5.516) | 0.737 (0.591–0.898) |
| ΔLatency | 0.506 (0.419–0.586) | 0.000 (−15.060–24.400) | 0.570 (0.183–0.978) | 0.568 (0.086–0.926) | 0.591 (0.542–0.765) | 0.537 (0.480–0.825) | 1.261 (1.032–2.832) | 0.751 (0.185–0.944) |
| ΔTime to target acquisition | 0.789*** (0.716–0.855) | 49.250 (18.150–53.220) | 0.624 (0.505–0.785) | 0.975 (0.815–1.000) | 0.967 (0.821–1.000) | 0.692 (0.637–0.773) | 23.516 (3.910–54.000) | 0.387 (0.256–0.496) |
| MG vs sixth CNP (horizontal) | ||||||||
| ΔRange | 0.800*** (0.700–0.885) | −9.090 (−12.910–1.650) | 0.688 (0.538–0.817) | 0.864 (0.682–1.000) | 0.959 (0.910–1.000) | 0.400 (0.320–0.514) | 5.362 (2.366–15.849) | 0.355 (0.223–0.503) |
| ΔPeak velocity | 0.559 (0.426–0.696) | −7.650 (−11.806–9.270) | 0.645 (0.172–0.849) | 0.636 (0.364–1.000) | 0.889 (0.838–1.000) | 0.278 (0.210–0.432) | 1.774 (1.204–5.914) | 0.615 (0.310–0.890) |
| ΔLatency | 0.617 (0.457–0.762) | −5.610 (−20.070–1.480) | 0.688 (0.441–0.935) | 0.682 (0.273–0.909) | 0.895 (0.837–0.959) | 0.333 (0.243–0.556) | 2.011 (1.212–5.441) | 0.473 (0.189–0.738) |
| ΔTime to target acquisition | 0.787*** (0.705–0.867) | 40.270 (8.700–70.590) | 0.667 (0.484–0.849) | 0.909 (0.682–1.000) | 0.966 (0.915–1.000) | 0.392 (0.306–0.552) | 5.677 (2.481–15.613) | 0.367 (0.192–0.538) |
| MG vs healthy controls (vertical) | ||||||||
| ΔRange | 0.879*** (0.818–0.929) | −5.210 (−8.690–3.200) | 0.806 (0.699–0.903) | 0.914 (0.815–0.975) | 0.912 (0.839–0.974) | 0.805 (0.731–0.882) | 9.083 (4.525–32.661) | 0.212 (0.117–0.321) |
| ΔPeak velocity | 0.580 (0.498–0.670) | −5.940 (−7.110–3.950) | 0.495 (0.376–0.613) | 0.815 (0.679–0.901) | 0.746 (0.658–0.847) | 0.580 (0.532–0.638) | 2.555 (1.677–4.839) | 0.630 (0.494–0.766) |
| ΔLatency | 0.523 (0.438–0.605) | −2.960 (−25.330–36.560) | 0.677 (0.086–0.968) | 0.481 (0.111–0.988) | 0.597 (0.549–0.900) | 0.556 (0.479–0.733) | 1.288 (1.059–6.260) | 0.697 (0.317–0.947) |
| ΔTime to target acquisition | 0.868*** (0.810–0.927) | 26.980 (16.285–47.860) | 0.785 (0.656–0.892) | 0.914 (0.802–0.988) | 0.916 (0.830–0.986) | 0.786 (0.708–0.878) | 9.332 (4.246–52.585) | 0.238 (0.121–0.359) |
| MG vs third or fourth CNP (vertical) | ||||||||
| ΔRange | 0.868*** (0.783–0.934) | −5.950 (−11.950–1.606) | 0.796 (0.677–0.925) | 0.875 (0.688–0.969) | 0.949 (0.892–0.987) | 0.596 (0.451–0.728) | 6.280 (2.839–25.462) | 0.225 (0.098–0.359) |
| ΔPeak velocity | 0.525 (0.410–0.644) | −13.810 (−24.730–11.270) | 0.849 (0.140–0.968) | 0.375 (0.156–0.969) | 0.798 (0.760–0.952) | 0.423 (0.272–0.667) | 1.359 (1.089–4.989) | 0.469 (0.172–0.921) |
| ΔLatency | 0.621* (0.500–0.731) | −2.960 (−10.680–2.520) | 0.634 (0.452–0.817) | 0.719 (0.469–0.844) | 0.862 (0.794–0.924) | 0.400 (0.316–0.500) | 2.151 (1.327–4.129) | 0.516 (0.344–0.746) |
| ΔTime to target acquisition | 0.836*** (0.760–0.906) | 34.260 (13.080–48.410) | 0.763 (0.634–0.914) | 0.875 (0.688–0.969) | 0.946 (0.890–0.987) | 0.556 (0.456–0.757) | 5.884 (2.767–23.742) | 0.275 (0.111–0.411) |
Note: Δ indicates percent change from the second saccade to the mean of the last 5 saccades: [(mean of last 5 – second)] × 100. Sensitivity analyses restricted to high‐certainty MG (n = 78) showed similar discrimination for key vertical fatigability metrics. Cross‐tabulations (2 × 2 tables) for prespecified ΔRange thresholds are provided in Table S3.
Abbreviations: AUC, area under the receiver operating characteristic curve; CI, confidence interval; CNP, cranial nerve palsy; LR, likelihood ratio; MG, myasthenia gravis; NPV, negative predictive value; PPV, positive predictive value.
DeLong test for AUC > 0.5.
p < 0.05.
p < 0.001.
FIGURE 2.
Distribution of key fatigability metrics across HC, MG, and CNP subgroups. Panels show ΔRange (A and C) and ΔTime‐to‐target (B and D) for horizontal (A and B) and vertical (C and D) saccades. Δ metrics were calculated as percent change from the second saccade to the mean of the last 5 saccades: (mean of last 5 − second)/second × 100. Plots show HC, MG (total, ocular MG, and generalised MG), and axis‐matched CNP subgroups (horizontal: Sixth CNP; vertical: Third and fourth CNP). Points represent individual participants; shaded distributions depict density; boxes indicate median and interquartile range with whiskers. The vertical reference line at 0 indicates no change across repetitions; negative ΔRange values indicate decrements across repetitions, and positive ΔTime‐to‐target values indicate prolongation across repetitions. Analyses used the clinically affected eye for CNP, the more affected eye for MG, and the right eye for HCs; sample sizes are shown in each panel and may vary by metric because of incomplete analysable data. Abbreviations: CNP, cranial nerve palsy; HC, healthy control; MG, myasthenia gravis.
When the MG group was restricted to high‐certainty cases supported by seropositivity and/or decrement on repetitive nerve stimulation (n = 78), discrimination of the key vertical fatigability metrics remained similar to the primary analysis. Cross‐tabulations for the prespecified percentage range‐change thresholds are provided in Table S4.
4. Discussion
In this prospective diagnostic study, repetitive saccade testing with three‐dimensional VOG identified a dynamic ocular motor phenotype in MG that differed from the more static deficit of unilateral ocular motor CNP. Patients with MG showed progressive decrement in saccadic range and prolongation of time to target acquisition across repeated trials, whereas patients with CNP showed axis‐appropriate deficits that were already present on the earliest analysable saccade and changed little thereafter. For the clinically relevant comparison of MG versus third or fourth nerve palsy, percentage change in vertical saccadic range provided the strongest discrimination.
These findings are clinically relevant because MG and ocular motor CNP often occupy the same early differential diagnosis in patients presenting with diplopia. Ptosis, partial ophthalmoplegia, and variable symptom reporting may blur the distinction at the bedside, particularly when the examination is brief or performed at a single time point. A quantitative test that captures fatigue during repeated activation therefore addresses a feature of MG that is not directly measured by routine alignment and duction testing. The pattern observed in MG—a progressive loss of range with comparatively modest change in peak velocity—is consistent with impaired neuromuscular transmission rather than a primary deficit in central saccade generation [14, 22, 23, 24]. Patients with MG appear able to initiate a rapid eye movement, but the movement becomes increasingly hypometric across repetitions, leading to additional corrective saccades and longer time to target acquisition [9, 25, 26]. By contrast, unilateral third, fourth, and sixth nerve palsies showed a different signature. Range and peak velocity were already reduced on the second saccade in the direction of the paretic muscle, but these abnormalities remained relatively stable during the task. This pattern is compatible with a structural or ischaemic paresis that constrains ocular rotation from the outset [15, 27, 28, 29, 30, 31]. Peak saccadic velocity should be interpreted cautiously because of its dependence on saccadic amplitude. Although the present repeated‐saccade protocol used a fixed 15° target displacement and therefore controlled nominal target amplitude by design, the actually executed saccadic range could vary across participants because of hypometria or fatigability. We therefore performed post hoc analyses adjusting peak velocity for measured saccadic range. These analyses suggested that apparent velocity reduction in MG was largely explained by reduced saccadic range, whereas CNP retained lower range‐adjusted peak velocity in relevant axis‐matched comparisons. Nevertheless, because the protocol did not include random‐amplitude or multi‐amplitude saccades, this analysis should be regarded as a range‐adjusted sensitivity analysis rather than formal main sequence modelling.
A particular strength of the study is the use of axis‐matched comparisons, aligning horizontal fatigability metrics with sixth nerve palsy and vertical metrics with third or fourth nerve palsy. This approach avoids confounding from large baseline deficits in a non‐corresponding axis and makes the results easier to interpret clinically. The vertical comparison was especially informative, perhaps because vertical range decrement better captures the mismatch between fluctuating MG and the more direction‐specific restrictions of third and fourth nerve palsies. Whether this reflects comparator composition, vertical duction biomechanics or greater susceptibility of vertical extraocular muscles to transmission failure warrants further study [32].
The study also extends previous work by testing a clinically plausible comparator group rather than only healthy controls [9, 10, 33]. In the evaluation of diplopia, the central question is not whether MG differs from normal ocular motility but whether it can be distinguished from other causes of ophthalmoplegia that prompt different diagnostic pathways. From that perspective, an AUC of 0.868 for vertical range change, with 79.6% sensitivity and 87.5% specificity at the optimal threshold, is encouraging for an adjunctive test performed during a single visit. These results should nevertheless be interpreted as complementary to, not a replacement for, standard evaluation. Serology, electrophysiology, and neuroimaging remain essential. Where binocular VOG is already available, a short non‐invasive paradigm may provide objective support for MG suspicion in patients with equivocal bedside findings or unrevealing imaging, but it should not be construed as a universally accessible triage test.
Implementation considerations remain important before broader adoption. Although VOG is non‐invasive and can provide immediate quantitative results once equipment and trained personnel are available, binocular VOG remains concentrated in tertiary or specialist neuro‐ophthalmology and neurotology centres in many healthcare systems. Therefore, the current role of this paradigm should be viewed as a specialist‐centre adjunct for objectively documenting fatigability in selected patients with equivocal bedside findings, not as a replacement for serology, electrophysiology, or clinically indicated neuroimaging. Reliable calibration and artefact handling are essential, and some enrolled participants in this study did not yield analysable recordings. Medication exposure may also influence measured fatigability; although medication timing was standardised when clinically safe, future pragmatic studies should evaluate performance under usual treatment conditions. Additional work is needed to establish age‐stratified normative distributions, intersession repeatability, platform comparability, operator training requirements and external validity across clinical settings.
This study has several limitations. It was conducted at a single centre, and the CNP subgroups were modest in size. The CNP cohort was aetiologically heterogeneous rather than restricted to idiopathic or MRI‐negative palsies; this improves real‐world relevance but may increase comparator heterogeneity. Because VOG was performed at a single time point and follow‐up duration was not uniform, delayed diagnostic evolution and residual misclassification, particularly between ocular MG and CNP in clinically ambiguous early presentations, cannot be fully excluded. Generalisability to other causes of diplopia, such as thyroid eye disease, internuclear ophthalmoplegia, skew deviation, chronic progressive external ophthalmoplegia or restrictive strabismus, remains uncertain. Exclusion of participants without analysable VOG recordings may have introduced selection bias, as patients with severe paresis, large deviations, poor calibration, ptosis or incomplete task performance may have been underrepresented. The analysed eye in MG was selected according to greater range decrement, which may have favoured discrimination. VOG calibration and pupil tracking can be less reliable in severe paresis, large deviations or ptosis; unlike magnetic search coils, VOG cannot fully resolve uncertainty about fixation during calibration. Although within‐session relative metrics may reduce the impact of absolute calibration error, measurement error cannot be eliminated. Fatigability was primarily quantified using prespecified early‐to‐late changes. Exploratory intra‐series visualisation suggested progressive range decrement in MG, but formal subject‐level time‐series or fluctuation analyses were not performed; a consistent rebound‐type decline–recovery–decline pattern was not evident at the group level. Finally, post hoc range‐adjusted regression does not replace formal main sequence analysis, so peak velocity should not be interpreted as a standalone marker of intrinsic saccadic slowing. The 75‐cycle task length was a pragmatic a priori choice, the alternating paradigm may permit partial agonist recovery, and the 120‐Hz sampling rate may underestimate true peak velocity, although this applied equally across groups [34, 35].
Despite these limitations, the present findings support quantitative fatigability as a potentially useful biomarker in ocular MG. Beyond diagnosis, repeated VOG assessment may also have value as an outcome measure because ocular symptoms in MG fluctuate and may respond to treatment. Future studies should assess longitudinal responsiveness and determine whether combining saccadic fatigability with other ocular motor measurements, such as fixation stability, smooth pursuit, or ptosis metrics, further improves classification across a broader diplopia differential diagnosis.
In summary, repetition‐based VOG identified a dynamic fatigability profile in MG that differed from the more static ocular motor deficits seen in unilateral ocular motor CNP. Quantitative saccadic fatigability may provide a practical objective adjunct to the neuro‐ophthalmic evaluation of diplopia.
Author Contributions
Sun‐Young Oh: conceptualization, funding acquisition, project administration, supervision, writing – review and editing, methodology. Min‐Woo Lee: visualization, writing – review and editing, resources, formal analysis. Juhee Chae: data curation, writing – original draft, investigation, methodology, validation, visualization.
Funding
This work was supported by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (RS‐2025‐02223226); a National Research Foundation of Korea (NRF) grant funded by the Korean government (Ministry of Science and ICT) (RS‐2025‐00553480); and this research was also supported by the Biomedical Research Institute Fund, Jeonbuk National University Hospital. The funders had no role in study design, data collection, analysis, interpretation, or manuscript preparation.
Ethics Statement
This study involves human participants and was approved by the Institutional Review Board of Jeonbuk National University Hospital (2022‐04‐044‐001).
Consent
Participants gave informed consent to participate in the study before taking part.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Figure S1: Participant flow diagram.
Flow diagram of participant screening, enrolment, exclusions, and inclusion in the final analysis. Among 382 individuals assessed for eligibility, 265 were enrolled after providing informed consent; 117 were excluded before enrolment because they did not meet inclusion criteria, had other neurologic or ophthalmologic disease, or declined participation. After enrolment, 34 participants were excluded because of calibration failure, inability to complete the protocol, or insufficient analysable data. The final analytic cohort included 231 participants: 93 with myasthenia gravis, 54 with cranial nerve palsy, and 84 healthy controls.
Figure S2: Representative horizontal and vertical saccade traces.
Representative examples of saccadic eye movements from (A) a healthy control participant, (B) a patient with myasthenia gravis (horizontal saccades), (C) a patient with sixth cranial nerve palsy (horizontal saccades), (D) a patient with third cranial nerve palsy (vertical saccades), and (E) a patient with fourth cranial nerve palsy (vertical saccades). For each panel, traces are shown for an early trial (Initial) and a late trial (Last) during the repetitive saccade protocol. Grey traces indicate target position, red traces indicate eye position, and black traces indicate eye velocity. In myasthenia gravis, saccadic range decreases across repeated trials (fatigability) with relatively preserved peak velocity, whereas cranial nerve palsy shows a reduced range and slower peak velocity from the outset with minimal change across trials (static deficit). In cranial nerve palsy panels, traces correspond to the affected eye. Scale bar indicates 1 s.
Figure S3: Positive and negative predictive values across disease prevalence.
Positive predictive value (PPV; blue) and negative predictive value (NPV; orange) plotted as a function of assumed myasthenia gravis (MG) prevalence for diagnostic thresholds derived from the primary analysis. Panels show (A) percent change in saccadic range (ΔRange) and (B) percent change in time to target (ΔTime‐to‐Target) for the axis‐matched horizontal comparison (MG vs. sixth cranial nerve palsy; top row) and the axis‐matched vertical comparison (MG vs. third or fourth cranial nerve palsy; bottom row). Δ metrics are calculated as percent change from the second saccade to the mean of the last 5 saccades: [(mean of last 5 − second)/second] × 100. The red dashed vertical line indicates the MG prevalence in the study sample for each comparison (0.809 for horizontal; 0.744 for vertical), and the grey vertical line indicates a prevalence of 0.50.
Figure S4: Representative trial‐by‐trial intra‐series trajectories of saccadic parameters during repetitive saccade testing.
Exploratory intra‐series trajectories of saccadic parameters across the 75‐cycle repeated saccadic task: (A) saccadic range, (B) peak velocity, (C) latency, and (D) time‐to‐target. Lines represent trial‐by‐trial group‐level trajectories for healthy controls, myasthenia gravis, horizontal sixth cranial nerve palsy, vertical third cranial nerve palsy, and vertical fourth cranial nerve palsy. Healthy controls showed relatively stable range and time‐to‐target across the sequence, whereas myasthenia gravis showed a progressive reduction in saccadic range and an increase in time‐to‐target during repeated saccades. Cranial nerve palsy groups showed lower baseline range and peak velocity, with comparatively limited repetition‐related decline.
Table S1: Effect Sizes for Horizontal Saccade Parameters.
Table S2: Effect Sizes for Vertical Saccade Parameters.
Table S3: Post hoc amplitude‐adjusted linear regression analysis of peak velocity (PV).
Table S4: Cross‐Tabulation of Index Test Classification by Reference Standard for Prespecified ΔRange Thresholds.
Table S5: STARD 2015 checklist for diagnostic accuracy studies.
Acknowledgements
The authors thank the participants and the clinical staff who supported enrolment and eye‐movement testing.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
References
- 1. Richards B. W., F. R. Jones, Jr. , and Younge B. R., “Causes and Prognosis in 4,278 Cases of Paralysis of the Oculomotor, Trochlear, and Abducens Cranial Nerves,” American Journal of Ophthalmology 113, no. 5 (1992): 489–496. [DOI] [PubMed] [Google Scholar]
- 2. Kim H.‐J., Kim H.‐J., Choi J.‐Y., Yang H. K., Hwang J.‐M., and Kim J.‐S., “Diplopia: Characteristics and Etiologic Distribution in a Referral‐Based University Hospital,” Journal of Neurology 270, no. 2 (2023): 1067–1075. [DOI] [PubMed] [Google Scholar]
- 3. Gilhus N. E., “Myasthenia Gravis,” New England Journal of Medicine 375, no. 26 (2016): 2570–2581. [DOI] [PubMed] [Google Scholar]
- 4. Kaminski H. J., Sikorski P., Coronel S. I., and Kusner L. L., “Myasthenia Gravis: The Future Is Here,” Journal of Clinical Investigation 134, no. 12 (2024): e179742. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Rousseff R. T., “Diagnosis of Myasthenia Gravis,” Journal of Clinical Medicine 10, no. 8 (2021): 1736. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Nguyen M. N., Van Der Walt A., Fielding J., Clough M., and White O. B., “Tracking Eye Movements for Diagnosis in Myasthenia Gravis: A Comprehensive Review,” Journal of Neuro‐Ophthalmology 42, no. 4 (2022): 428–441. [DOI] [PubMed] [Google Scholar]
- 7. de Verdal M., Renard D., and Castelnovo G., “Ocular Myasthenia Gravis: A Diagnosis With Video Oculography,” Journal of Neuro‐Ophthalmology 45, no. 1 (2025): e18–e19. [DOI] [PubMed] [Google Scholar]
- 8. Wong S. H., Bancroft M. J., Tailor V. K., et al., “Ocular Myasthenia Gravis Saccades as a Measure of Extraocular Muscle Function,” Frontiers in Ophthalmology 2 (2022): 938088. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Chae J., Nguyen T. T., and Oh S.‐Y., “Quantification of Saccadic Fatigability and Diagnostic Efficacy for Myasthenia Gravis,” Journal of Neurology 271, no. 8 (2024): 1–11. [DOI] [PubMed] [Google Scholar]
- 10. Nguyen T. T., Kang J.‐J., Chae J.‐H., et al., “Oculomotor Fatigability With Decrements of Saccade and Smooth Pursuit for Diagnosis of Myasthenia Gravis,” Journal of Neurology 270, no. 5 (2023): 2743–2755. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Gilhus N. E., Tzartos S., Evoli A., Palace J., Burns T. M., and Verschuuren J. J. G. M., “Myasthenia Gravis,” Nature Reviews Disease Primers 5, no. 1 (2019): 30. [DOI] [PubMed] [Google Scholar]
- 12. De Lucena D. D. and Rangel É. B., “Glucocorticoids Use in Kidney Transplant Setting,” Expert Opinion on Drug Metabolism & Toxicology 14, no. 10 (2018): 1023–1041. [DOI] [PubMed] [Google Scholar]
- 13. Agarwal S., Gowda K., Mandal U., et al., “Analysis of Pyridostigmine Bromide in Human Plasma and Its Application in Bioequivalence Studies,” Journal of Liquid Chromatography & Related Technologies 30, no. 17 (2007): 2605–2615. [Google Scholar]
- 14. Barton J. and Sharpe J. A., “'Saccadic Jitter'is a Quantitative Ocular Sign in Myasthenia Gravis,” Investigative Ophthalmology & Visual Science 36, no. 8 (1995): 1566–1572. [PubMed] [Google Scholar]
- 15. Wong A. M., McReelis K., and Sharpe J. A., “Saccade Dynamics in Peripheral vs Central Sixth Nerve Palsies,” Neurology 66, no. 9 (2006): 1390–1398. [DOI] [PubMed] [Google Scholar]
- 16. Leigh R. J. and Zee D. S., The Neurology of Eye Movements (Oxford University Press, 2015). [Google Scholar]
- 17. Larrazabal A. J., Cena C. G., and Martínez C. E., “Video‐Oculography Eye Tracking Towards Clinical Applications: A Review,” Computers in Biology and Medicine 108 (2019): 57–66. [DOI] [PubMed] [Google Scholar]
- 18. Tzagarakis C., West S., and Pellizzer G., “Neural Encoding of the Reliability of Directional Information During the Preparation of Targeted Movements,” Frontiers in Neuroscience 15 (2021): 679408. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Lazarov A., Ben‐Zion Z., Shamai D., Pine D. S., and Bar‐Haim Y., “Free Viewing of Sad and Happy Faces in Depression: A Potential Target for Attention Bias Modification,” Journal of Affective Disorders 238 (2018): 94–100. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Ma W., Li M., Wu J., et al., “Multiple Step Saccades in Simply Reactive Saccades Could Serve as a Complementary Biomarker for the Early Diagnosis of Parkinson's Disease,” Frontiers in Aging Neuroscience 14 (2022): 912967. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Rahmouni S. and Madelain L., “Inter‐Individual Variability and Consistency of Saccade Adaptation in Oblique Saccades: Amplitude Increase and Decrease in the Horizontal or Vertical Saccade Component,” Vision Research 160 (2019): 82–98. [DOI] [PubMed] [Google Scholar]
- 22. Kaminski H. J., Maas E., Spiegel P., and Ruff R. L., “Why Are Eye Muscles Frequently Involved in Myasthenia Gravis?,” Neurology 40, no. 11 (1990): 1663–1669. [DOI] [PubMed] [Google Scholar]
- 23. Bahill A., Brockenbrough A., and Troost B., “Variability and Development of a Normative Data Base for Saccadic Eye Movements,” Investigative Ophthalmology & Visual Science 21, no. 1 Pt.1 (1981): 116–125. [PubMed] [Google Scholar]
- 24. Bahill A. T., Clark M. R., and Stark L., “Glissades—Eye Movements Generated by Mismatched Components of the Saccadic Motoneuronal Control Signal,” Mathematical Biosciences 26, no. 3–4 (1975): 303–318. [Google Scholar]
- 25. Serra A., Liao K., and Leigh R. J., “Conjugacy of Horizontal Saccades: Application of Binocular Phase Planes,” Progress in Brain Research 171 (2008): 571–574. [DOI] [PubMed] [Google Scholar]
- 26. Serra A., Thurtell M. J., and Leigh R. J., “Changes in Dynamic and Kinematic Properties of Saccades in Ocular Myasthenia Following Intravenous Immunoglobulin Treatment,” Annals of the New York Academy of Sciences 1164, no. 1 (2009): 468–471. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. McReelis K., “Saccades in Unilateral Central and Peripheral Sixth Nerve Palsy,” Investigative Ophthalmology & Visual Science 45 (2004): 2513. [Google Scholar]
- 28. Sharpe J. A., McReelis K., and Wong A. M., “Recovery of Peripheral Versus Central Nerves Identified by Saccadic Velocity After Abducens Neuropathy,” Annals of the New York Academy of Sciences 1039, no. 1 (2005): 417–429. [DOI] [PubMed] [Google Scholar]
- 29. Metz H. S., Scott A. B., O'Meara D., and Stewart H. L., “Ocular Saccades in Lateral Rectus Palsy,” Archives of Ophthalmology 84, no. 4 (1970): 453–460. [DOI] [PubMed] [Google Scholar]
- 30. Sharpe J. A., Wong A. M., and Fouladvand M., “Ocular Motor Nerve Palsies: Implications for Diagnosis and Mechanisms of Repair,” Progress in Brain Research 171 (2008): 59–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Kommerell G., Olivier D., and Theopold H., “Adaptive Programming of Phasic and Tonic Components in Saccadic Eye Movements. Investigations of Patients With Abducens Palsy,” Investigative Ophthalmology & Visual Science 15, no. 8 (1976): 657–660. [PubMed] [Google Scholar]
- 32. Cleary M., Williams G. J., and Metcalfe R. A., “The Pattern of Extra‐Ocular Muscle Involvement in Ocular Myasthenia,” Strabismus 16, no. 1 (2008): 11–18. [DOI] [PubMed] [Google Scholar]
- 33. Kim Y. L., Nguyen T. T., Hwang S.‐B., Kang J.‐J., Chae J., and Oh S.‐Y., “Quantifying Oculomotor Fatigability for Early Detection of Myasthenia Gravis,” Journal of Clinical Neuromuscular Disease 26, no. 3 (2025): 122–132. [DOI] [PubMed] [Google Scholar]
- 34. Mack D. J., Belfanti S., and Schwarz U., “The Effect of Sampling Rate and Lowpass Filters on Saccades–a Modeling Approach,” Behavior Research Methods 49, no. 6 (2017): 2146–2162. [DOI] [PubMed] [Google Scholar]
- 35. Nyström M., Hooge I. T., Hessels R. S., et al., “The Fundamentals of Eye Tracking Part 3: How to Choose an Eye Tracker,” Behavior Research Methods 57, no. 2 (2025): 67. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1: Participant flow diagram.
Flow diagram of participant screening, enrolment, exclusions, and inclusion in the final analysis. Among 382 individuals assessed for eligibility, 265 were enrolled after providing informed consent; 117 were excluded before enrolment because they did not meet inclusion criteria, had other neurologic or ophthalmologic disease, or declined participation. After enrolment, 34 participants were excluded because of calibration failure, inability to complete the protocol, or insufficient analysable data. The final analytic cohort included 231 participants: 93 with myasthenia gravis, 54 with cranial nerve palsy, and 84 healthy controls.
Figure S2: Representative horizontal and vertical saccade traces.
Representative examples of saccadic eye movements from (A) a healthy control participant, (B) a patient with myasthenia gravis (horizontal saccades), (C) a patient with sixth cranial nerve palsy (horizontal saccades), (D) a patient with third cranial nerve palsy (vertical saccades), and (E) a patient with fourth cranial nerve palsy (vertical saccades). For each panel, traces are shown for an early trial (Initial) and a late trial (Last) during the repetitive saccade protocol. Grey traces indicate target position, red traces indicate eye position, and black traces indicate eye velocity. In myasthenia gravis, saccadic range decreases across repeated trials (fatigability) with relatively preserved peak velocity, whereas cranial nerve palsy shows a reduced range and slower peak velocity from the outset with minimal change across trials (static deficit). In cranial nerve palsy panels, traces correspond to the affected eye. Scale bar indicates 1 s.
Figure S3: Positive and negative predictive values across disease prevalence.
Positive predictive value (PPV; blue) and negative predictive value (NPV; orange) plotted as a function of assumed myasthenia gravis (MG) prevalence for diagnostic thresholds derived from the primary analysis. Panels show (A) percent change in saccadic range (ΔRange) and (B) percent change in time to target (ΔTime‐to‐Target) for the axis‐matched horizontal comparison (MG vs. sixth cranial nerve palsy; top row) and the axis‐matched vertical comparison (MG vs. third or fourth cranial nerve palsy; bottom row). Δ metrics are calculated as percent change from the second saccade to the mean of the last 5 saccades: [(mean of last 5 − second)/second] × 100. The red dashed vertical line indicates the MG prevalence in the study sample for each comparison (0.809 for horizontal; 0.744 for vertical), and the grey vertical line indicates a prevalence of 0.50.
Figure S4: Representative trial‐by‐trial intra‐series trajectories of saccadic parameters during repetitive saccade testing.
Exploratory intra‐series trajectories of saccadic parameters across the 75‐cycle repeated saccadic task: (A) saccadic range, (B) peak velocity, (C) latency, and (D) time‐to‐target. Lines represent trial‐by‐trial group‐level trajectories for healthy controls, myasthenia gravis, horizontal sixth cranial nerve palsy, vertical third cranial nerve palsy, and vertical fourth cranial nerve palsy. Healthy controls showed relatively stable range and time‐to‐target across the sequence, whereas myasthenia gravis showed a progressive reduction in saccadic range and an increase in time‐to‐target during repeated saccades. Cranial nerve palsy groups showed lower baseline range and peak velocity, with comparatively limited repetition‐related decline.
Table S1: Effect Sizes for Horizontal Saccade Parameters.
Table S2: Effect Sizes for Vertical Saccade Parameters.
Table S3: Post hoc amplitude‐adjusted linear regression analysis of peak velocity (PV).
Table S4: Cross‐Tabulation of Index Test Classification by Reference Standard for Prespecified ΔRange Thresholds.
Table S5: STARD 2015 checklist for diagnostic accuracy studies.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.