This was a study to evaluate the feasibility and value of a adding a novel strabismus screening smartphone application to conventional school vision screening, as well as calculating threshold value for a positive screening to maximize sensitivity and specificity.
The findings suggest that a school nurse with no special training in the measurement of strabismus can easily add the EyeTurn app to their iPhone (available for iphone 6 to 8 or an iPad pro owned by the school), use the app alongside conventional school screening, and improve the identification strabismic children ages 6–12 educated in elementary school. There was value in adding the app to the screening with 3 children identified who were missed by conventional screening and had a previously undiagnosed intermittent exotropia. They were subsequently referred by the school nurse to an eye doctor for management of the intermittent strabismus, which may include prism glasses, vision therapy, or monitoring depending on the symptomatology. The school occupational therapist also reported an interest in the results of the screening for the possible relationship between intermittent exotropia, motor developmental problems and academic performance.
Discussion of Functionalities, Settings, and Methods in Lieu of the Results: the app has successfully captured images through glasses, but in order to maximize image capture success for the purposes of the study (glasses introduce extra reflections and distortions), it was recommended to remove glasses. Leaving the glasses on for screening is not needed. The app will presumably detect the strabismus without the glasses and the recommendation for referral can be ignored by parents when the issue has already been identified. The nurse was instructed to use the app at 30 cm to 40 cm, which is ecologically valid for school work and learning which primarily involve these near distances. Distance does not need to be precisely controlled and there is no need to add any landmark / indicating marker of known size in the plane of corneal reflection (e.g. credit card stripe as is used in some pupillary distance apps) as the app auto-calibrates for distance.16 The app can also make measurements at distances up to approximately 1.5 m; however, resolution of any photographic system worsens with distance and so the results of this study do not apply if the app is used at distances greater than 40 cm where false positives and negatives are likely to be higher. It is worth noting that the nurse’s iPhone 7 used in this study has a 12MP camera, which when used at 40 cm, provides resolution of ≈ 1/14th of a mm (1.4Δ) using a Hirschberg ratio of 19Δ/mm). The resolution drops to ≈ 1/5th of a mm at 1.5 m. There may be concern that the Hirschberg ratio (HR) population variability is too great to use the computerized Hirschberg method without individual measurement. Measuring the HR with the iPhone has been done on a smartphone via a monocular measurement at various gaze positions of known eccentricity in each eye,16 and requires precise fixation but is otherwise not difficult to obtain; however, a prior study with the app found that a population norm HR was of essentially equivalent accuracy;16 therefore, this extra step which could jeopardize feasibility is not needed. The version of the EyePhone app used in this study was limited to measurement of strabismus in horizontal direction only, but our data here suggest it is probably not needed for screening in elementary school children. A prime example was the child with congenital 4th nerve palsy in our sample who also manifested an exodeviation, and was easily detected by the app. However, future studies will involve measurement of eye deviation in both horizontal and vertical direction.
Discussion of Image Capture and Measurement Success and Failures: The results of the study suggest that it is not difficult to obtain at least one successful image for over 90% of elementary school students tested, given 3 attempts. The most common issue reported by the nurse was sensitivity to the flash resulting in eye closure and the failure of the software to automatically and accurately locate the eyes and limbus; however, of the 87 images with fitting failures, only 13 were due to eye closing. Therefore while the flash may have been bothersome to some children it did not seem to have a substantial impact on successful measurement. Iris misfitting was a greater issue representing a little more than half of all measurement failures. The app did equally well at measuring strabismus in children with light irises as in dark irises, which is encouraging. Our sample was fairly balanced between dark and light irises, and while the school tested would not be considered racially diverse, the spectrum of possible iris colors were well represented and should generalize to other populations. The overall image capture failure rate of 23% can be compared to a prior study of the GoCheckKids app at about 12%.15 This seeming difference is somewhat surprising since both methods require flash carrying the risk of eye closure, and automatic processing in both apps requires iris border detection. It is possible that precise iris fitting is less critical for the GoCheckKids’s red reflex method.
Discussion of the Value of Adding Automated Hirschberg to Amblyopia Screening: Combination of the automated Bruckner and Hirschberg methods may be possible in an app as is done for multiple dedicated photoscreeners (see introduction), although this is likely to increase image processing time potentially creating a feasibility issue in school screening. Adding strabismus detection with photographic Hirschberg may not be of value if the only issue of concern is amblyopia detection, which has thus far been the principal target of AAPOS and AAP. However, stakeholder interest (educators, OTs, parents) in the negative impact of intermittent exotropia and convergence insufficiency on academic performance is growing and may drive new policies around targeted screening for these conditions if a feasible method exists.21 The school OT in this study expressed great interest in the potential of screening for intermittent exotropia and convergence insufficiency.
Discussion of the 1 Case of Strabismus Missed: The app was successful in detecting four cases of intermittent exotropia and one case of constant esotropia. When the optimum threshold was determined (ROC analysis) and reset to 3.0△, the EyeTurn app would have narrowly missed a case of 14Δ accommodative esotropia (as determined by in-person assessment with 20/30 letter as fixation). This was the only case of strabismus that the app would have missed in this cross-sectional sample. The smaller amount of esotropia with the app compared to ground-truth may be because using the camera lens as the fixation target did not introduce much accommodative demand. Similar occurrences were reported in a study about long-term follow-up in accommodative esotropia.22 Providing a 20/30 letter target on the phone near the camera lens may solve this problem. Conventional school screening only identified one case of strabismus (exotropia) with decreased stereoacuity, which was also identified by the app. This suggests that traditional school screening may miss many cases of intermittent strabismus where visual acuity and stereoacuity are good.23 It is worth noting that some children with esotropia detected by the app without glasses were previous diagnosed, therefore may have been appropriately “missed” by the conventional screening if they were wearing their glasses.
The sensitivity and specificity of (76.5% and 83.0%) measured in this study can be compared to a study with the Welch Allyn Spot Vision Screener’s ability to detect strabismus (50% and 96%) where a trained non-medical staff or volunteer took the images with the ground truth coming from those children whom ultimately followed up with an eye doctor.24 In another study of the Spot screener in detecting strabismus, this time in a pediatric ophthalmology office, sensitivity of (77% and 93%) were reported,25 but it should be noted that intermittent cases were not considered as positive in that study, which likely inflates the values. A prior study of the GoCheckKids app was similar to this study only in that it was also deployed on a phone, but used the automated Bruckner method, and was conducted in a different age group (ages 6 months-6 years) finding the sensitivity of (74.6%) and specificity of (67.2%) for comparison.26 The Plusoptix photoscreener is also used for detecting strabismus in children (2–14 years) with the sensitivity (40.7%) and specificity (98.3%) respectively.27
Discussion of the app for Dissociated Measurements: Sensitivity to bright light is a characteristic often associated with intermittent exotropia due to a “dazzling” of the retina and disruption of fusion.28, 29 This kind of disruption after dazzling can last more than 0.8 sec, which is long enough for the app (about 0.2 sec) to capture the eye misalignment.30 Therefore the flash may be helpful in dissociating the eyes of children with intermittent exotropia and convergence insufficiency, revealing the conditions as our data suggests. Of the 8 false positives, 2 of those were found to have convergence insufficiency by in-person testing which may be a detriment to near work and learning.31 Our findings suggest screening for convergence insufficiency may be possible with this app, and could be valuable given its negative impacts on near work activities important for learning which are treatable with vision therapy.32
Limitations: The true false negative rate cannot be known in this study, because the school could not, for logistical reasons, allow in-person measurements on all the children who were screened with the app. Therefore, there may be a couple children that were missed by both the app and the traditional screening. In addition, this was not a random sample so it is unknown if it was representative of the entire school. It is also possible there were cases missed by both the app and the conventional screening; however, the prevalence in our sample of 6/133 (4.5%) is within the reported prevalence in similar populations (3–8%).1, 2 In terms of eye detection errors, which can be due to eye closing or other sources of interference, the app prompts the user to retake the image and does not save the image (although it can be overridden and saved, which occurred in at least 5 instances, Table 2). Given that we had 133 students with 377 images and the protocol was to attempt 3 images per child (399 images), there were potentially 22 images unaccounted for which may have been eye detection failure with eye closure. Given that the nurse’s report of eye closing being an issue was contrary to the data available, the image capture failure rate could have been a little worse, 27.3% (109/399) rather than the 23.1% reported. Regardless, use of the app was feasible in the setting tested without undue burden reported by the nurse.