Approval for this study was granted by the University of California, San Francisco (UCSF) Institutional Review Board. Written informed consent was provided by all adult participants. Minors expressed their verbal assent, and their parent(s) provided formal informed consent. Data were deidentified, coded anonymously, encrypted, and stored on a server in a secure location.

The participants were 35 patients from a cohort of 64 consecutive eligible patients referred with a diagnosis of intermittent exotropia to a single pediatric ophthalmologist (JCH) over 5 years. Data from 19 patients were reported previously, but without segmentation into distance vs near viewing. Furthermore, 16 new patients were added. Of 64 patients, 29 were excluded because exotropia occurred <1% of the time (18), the video file was corrupt (4), or the data were collected in an older file format incompatible with our analysis software (7).

The 35 patients ranged in age from 3 to 79 years, with a median age of 10 years. All but 3 were younger than 40 years, and therefore not presbyopic. The corrective lenses used with the Tobii eye tracker are single vision, not bifocal. Consequently, the 3 presbyopic patients in this study may have performed less near work than customary during their recording session. Eligibility criteria were as follows: (1) best-corrected visual acuity of 20/20 in each eye, (2) no ocular disease except strabismus, (3) an intermittent exotropia ≥10 prism-diopters measured by prism and alternate cover test, (4) ability to fuse at near for at least a few seconds to allow eye tracker calibration, (5) stereopsis of 80 arc-sec or better by Stereo Optical Randot circles (Stereo Optical Company, Inc.), and (6) willingness and ability to wear the eye-tracking glasses for at least a few hours. Subjects with a spherical equivalent refractive error of ≤± 2.00 diopters (D) were tested without correction. If the spherical equivalent refractive error was ≥± 2.00 D, subjects wore their own contact lenses or auxiliary spherical equivalent corrective lenses were placed into the eye-tracking glasses. Corrective lenses were available from −5.00 sphere to +3.00 sphere in steps of 0.50 D.

Measurements of eye position in ambulatory patients were obtained with Tobii Pro Glasses 3 ( www.tobiipro.com ). The instrument consists of a pair of eyeglasses containing 8 infrared illuminators and 2 cameras embedded in each plano lens. A single camera located between the eyes on the bridge of the glasses captures the scene over 95° horizontally by 63° vertically at 25 Hz. The glasses connect via a cable to a recording unit. The device is calibrated at the beginning of the recording by having the subject fixate on a bull’s-eye target. The center of gaze for each eye relative to the image captured by the scene camera is computed at 50 Hz from the locations of the pupil center and the 8 illuminator reflections on the cornea. Data are saved to an SD card in the recording unit for later analysis. The recording unit is powered by an internal battery connected to an external rechargeable battery pack that provides up to 12 hours of operation. Both are secured inside a small knapsack that can be worn comfortably by patients, even when seated in a chair. When outdoors in bright sunlight, slip-on infrared-blocking lenses are worn to avoid washing out the corneal illuminator reflections.

When an eligible patient was seen during a clinic visit, ambulatory eye tracking was proposed to assess the occurrence rate of intermittent exotropia. If the patient chose to participate, the glasses were placed on the subject, and their function was explained. Participants observed how the glasses tracked their gaze position and understood that a video and audio recording would be made of everything they looked at and heard during the monitoring session. Subjects consented explicitly to the preservation and review of the audiovisual recording by the investigators. They left the clinic wearing the eye-tracking glasses and usually wore them for the rest of the day while engaged in their regular repertoire of activities. Participants removed the eye-tracking glasses when engaged in private activities, such as using a bathroom. The device was shipped prepaid back to our laboratory the next morning by a package delivery company. After a few days, the data showing the percentage of time that exotropia was present were emailed to subjects, followed by a phone call to discuss the results.

Data on the SD card from the recording session were analyzed using custom software available at Open Science Framework via this link: https://osf.io/fesdp . A previous publication provides details about how the moment-to-moment position of each eye was calculated and how the data were filtered to compensate for artifacts and brief interruptions such as blinks. Each subject’s ocular alignment during the ambulatory recording was compiled in a histogram (0.2° bin size) that displayed the time spent at each angle of vergence. For orthotropic subjects fixating at infinity, the vergence angle was 0°. Previous testing in normal subjects viewing a distant target has shown that the eye-tracking glasses generate offset errors of up to ±1° and instrument noise causes peaks to widen slightly.

To segment the recordings into epochs of near viewing vs far viewing, the scene videos were downsampled from 25 to 5 Hz. The subject’s gaze position was represented in the video as a colored spot subtending 8°. During periods of exotropia, the mean position of the eyes was displayed, rather than the position of each eye. This was done to mask knowledge of whether the subject was fusing or was exotropic from the investigator analyzing the data.

The scene video with the superimposed gaze position cursor was segmented using DaVinci Resolve ( www.blackmagicdesign.com ). In a previous study, we observed that vergence profiles compiled in normal ambulatory subjects during eye tracking usually have a bimodal distribution. More time is spent viewing either at far or near, with less time at intermediate distances. The goal of segmenting the video was to eliminate equivocal intermediate distances and thus isolate epochs of pure near viewing and pure far viewing. This segmentation was done by means of a 2-stage process. First, the video was viewed, and a timestamp was placed whenever the subject’s eye position cursor suggested a transition from viewing at far to near or vice versa. Operationally, one is more apt to assign fixation behavior to near than to far. As a result, the first review parsed the video into (1) epochs of pure far viewing and (2) epochs of mostly near viewing, containing some misclassified intermediate viewing. To eliminate contamination by misclassified intermediate segments, the mixed epochs were reviewed again. On the second review, timestamps were placed to mark segments when near viewing was certain. This second step distilled epochs of pure near viewing.

The task of extracting relatively pure epochs of near vs far viewing from the video data required investigator judgment (Supplementary Figure S1). The process varied in difficulty from subject to subject. Some patients engaged in a single activity for prolonged periods, allowing one to advance rapidly through portions of their video. Others multitasked, requiring the placement of many timestamps. During video segmentation, an average of 52.6 timestamps/h was placed during the first-stage review and an additional 31.7 timestamps/h in the second-stage review. Before editing, the videos had a mean duration of 8.16 ± 2.93 hours, containing a mean of 6.62 ± 2.70 hours of filtered binocular eye-tracking data. The time difference was due to patient rest breaks and occasional loss of eye tracking. The editing process extracted a mean of 3.28 ± 1.91 hours of far viewing and 2.11 ± 1.59 hours of near viewing. A mean of 19% of the video data could not be categorized into near or far and was therefore excluded from postsegmentation occurrence rate analysis.

Examples of far activities included driving, walking, playing outside, attending a meeting, watching television, or shopping. Examples of near activities included using a smartphone, working on a computer, writing, or eating. To qualify as a near activity during the second pass review of videos, visual targets were required to be within arm’s reach. In reviewing the videos, a strong clue was provided by optic flow. When subjects approached a near object, it loomed in the video camera. Skill at segmenting the video data improved with practice. The first 6 videos were analyzed independently by 2 investigators as a quality control measure. The resulting segmentations were so similar that subsequent videos were processed by a single person.

In this study, head movements were not monitored, and therefore “gaze” refers colloquially to eye position in the orbit. For each eye, a positive value denotes right horizontal gaze in degrees. The subject’s vergence angle was calculated by subtracting the right eye position from the left eye position. Negative values signify divergence, and positive values represent convergence. In normal subjects, there were 2 peaks, one at 0° representing far viewing and another at some positive value representing near viewing ( Figure 1 , A). In subjects with exotropia only during far viewing, there were 3 peaks ( Figure 1 , B). In more decompensated subjects, there were 4 peaks: 2 corresponding to far and near orthotropia and 2 corresponding to far and near exotropia ( Figure 1 , C and D). If a subject converged to look at a near target but not enough to achieve fusion (ie, orthotropia), exotropia could occur with the eyes registering a converged value ( Figure 1 , D).

Schematic diagram of 4 typical states recorded by wearable eye tracker. A. Normal subject showing a peak for far viewing (orange) near 0° and a peak for near viewing (purple) at some arbitrary converged angle. B. Exotropic subject, losing fusion only during far viewing. C. Exotropic subject, losing fusion during far and near viewing. During exotropia at near, absence of convergence effort accounts for the superimposed exotropia peaks for near and far viewing. D. Exotropic subject, also losing fusion during far and near viewing, but making full convergence effort during near viewing. Consequently, the degrees between the far and near orthotropia peaks equal the degrees separating the far and near exotropia peaks. In this study, a range of convergence effort was observed during exotropia while near viewing, yielding XT _near peaks that varied widely in width and location.