Digital Tools for the Self-Assessment of Visual Acuity: A Systematic Review

Many digital tools are available to self-test VA, though a clinical validation is often lacking [6]. This systematic review presents the 17 publications on 13 different tools for the self-assessment of VA currently available.

Our systematic review identified low mean differences of the digital tools when compared to reference standards for assessing distance VA, suggesting a low systematic bias. The mean differences ranged from – 0.08 to 0.10 logMAR. The digital tools only slightly underestimate the VA score of the patients and we consider these low values to be negligible in clinical practice. The 95% LoA vary between studies, ranging from ± 0.08 logMAR to ± 0.47 logMAR from the mean difference. Most of the 95% LoA are rather wide, suggesting considerable variability of the VA assessments of the digital tools. As stated before, there is always a certain variation in repeated measurements in the same person [1, 20]. A study on the variability of VA assessments in a large eye clinic reported a test–retest variability of ± 0.15 logMAR, when different charts and different examiners assessed the same patient’s VA within a short time interval [20]. The authors conclude that, in general, differences of less than 0.15 logMAR (i.e. 1.5 lines) are considered measurement variation and should therefore not be considered as indicative of an actual clinical change. Surprisingly, despite the different chart designs and examiners that are compared in this study, this reported variability does not substantially differ from the test–retest variability of same-chart measurements. The test–retest variability of Snellen charts is actually even wider, as depicted in Fig. 4. This figure also illustrated that the 95% LoA of most digital tools exceed the 95% LoA reflecting test–retest variability of traditional VA charts. On the basis of these findings, the digital tools appear less precise than traditional VA charts. Obviously, as with any medical technological device, quality differences of the different tools affect performance. Importantly, these technologies are developing continuously and an improvement in accuracy is attained with every design iteration.

The accuracy of the near vision tests seems better than the distance VA assessments, expressed by the smaller 95% LoA. The near vision cards and the assessments using tablets are very comparable in nature, which might explain the good agreement. Although near vision testing has a remarkably smaller role in clinical practice, these tools are very easy to use. It has been proven effective to have patients do a self-test by handing over a tablet at an emergency unit [34]. Especially in an emergency setting, convenience and time-effectiveness outweigh accuracy.

To the best of our knowledge, this is the first systematic review that focusses on clinically validated digital tools for self-assessing VA. We conducted a comprehensive literature search and reviewed all reference lists of included studies using PRISMA guidelines. Two reviewers independently performed the literature screening, data extraction and risk of bias assessment. The bias analysis identified concerns regarding blinding of the subject and researcher for the outcome of the tests in nine studies. We do not consider this a major problem since VA tests report a quantifiable outcome which can be recorded without interpretation. In addition, the two compared tests were different, so learning effects are not expected. One study, Nik Azis et al. [26], reported a convenience sampling method for recruitment, by excluding children who were deemed uncooperative. This may have negatively affected generalizability of the outcomes.

When comparing VA assessments two factors are important to consider. First the type of chart used significantly affects the obtained VA outcome [4]. Consequently, this affects the observed differences between the digital tools and the reference charts. There is always variation when two different VA charts are compared, and one should be careful when using charts interchangeably [5]. This also explains the difference in agreement of the Peek Acuity tool, as studied by Bastawrous et al., when different reference charts are used (Snellen vs. ETDRS, see Fig. 4) [28].

Secondly, the precision of the assessment depends strongly on the achieved VA of the tested subject. When examining the Bland–Altman plots of individual studies, we noticed that the accuracy of the tests improves for patients with better VA scores. This was also demonstrated in the subgroup analyses in Fig. 4. For the study by Zhang et al., the variability of the difference was remarkably smaller for subjects with an achieved VA < 1.0 logMAR than for the ≥ 1.0 logMAR subgroup (± 0.12 vs. ± 0.22). From our own study data, we also learned that the measurement accuracy of one specific tool (Easee) differed for various VA ranges and study populations. For example, the additional subanalysis of the data of Wisse et al. revealed an evidently smaller 95% LoA for the healthy individuals with higher VA scores (e.g. 95% LoA of 0.00; 0.58 for visual acuities ≤ 0.5 logMAR, versus 95% LoA of – 0.75; – 0.25 for visual acuities > 0.5 logMAR) [16]. More importantly, the measurements of this specific digital tool were substantially more accurate when assessing corrected VA (in patients with uveitis), compared to assessing uncorrected VA (in healthy individuals and patients with keratoconus, with refractive errors) [10, 11, 16]. The digital self-assessments provide less accurate results for poorer VA ranges, regardless of underlying ocular comorbidities. Thus, wide 95% LoA do not necessarily imply inadequate testing conditions, or a low technical quality of the studied digital tools, but can be largely attributed to the poor vision of the tested population. A lower accuracy in poorer ranges is a common feature of VA assessments. A recurrent clinical reason is that the person administering the conventional test might terminate the assessment too early. The Snellen chart in particular shows a poorer precision in lower VA ranges and the differences between Snellen and ETDRS increase in this range [21]. This can be explained by the chart design. When testing Snellen acuity, the tester uses a line assignment method. The poor vision lines usually contain only 1 or 2 letters. Thus, missing a letter on these lines can make a huge difference in this range [5]. In clinical practice, both time and convenience are essential, and therefore Snellen charts remain popular. We suggest that the accuracy of the digital VA self-assessments in poorer VA ranges could be improved by extending the initial assessment of individuals with poor scores for retesting (i.e. performing another assessment with different optotypes for internal validation and adjustment of the initial VA score).

Two studies show a remarkable narrow 95% LoA: Ansell et al. [29] and Zhang et al. [18] in the subgroup with VA better than 0.1 Snellen. These papers show a better agreement than well-executed test–retest studies of VA wall charts, evaluated in controlled, experimental settings.

Future research should focus on the performance of the digital tools in unsupervised conditions. Some of the included studies reported specifically that tests were performed at fixed distances, for example with the head of the participant leaning against an ophthalmologic chin piece and the smartphone or chart fixed on a desk (Brucker et al. [14]). We expect the accuracy to be different in real-world, less-controlled conditions. We encourage stratifying for different VA ranges when evaluating agreement between VA charts. We strongly recommend using logMAR charts as reference charts as they are the most accurate and consistent. For follow-up purposes, good repeatability (test–retest reliability) of the tools is important. This creates excellent opportunities for follow-up and signaling worsening of vision, as obtaining baseline measurements will allow future vision comparisons. Only two of the included validation studies reported on repeatability [13, 28].

Digital testing with mobile technology has many important advantages that outweigh accuracy. Traditional in-hospital VA testing requires patients to physically attend a clinic and consumes substantial hospital resources. The former might be particularly difficult for patients from rural areas, the elderly and immobilized patients. With the increasing digitization and availability of mobile devices, the digital tools have the potential to identify the most important cause of visual impairment worldwide: uncorrected refractive errors. Visual impairment has a negative impact on school performance, employability and quality of life in general. Strikingly, 53% of the visually impaired people have poor vision that can be prevented or remedied by glasses or contacts [35]. For screening of visual impairment, one is mostly interested in identifying a low VA range, rather than determining an exact value. When looking at the accuracy of the Peek Acuity tool (Bastawrous et al. [28]) in the forest plot, the 95% LoA are wider than the previously discussed test–retest studies of traditional charts. Notwithstanding, the tool has been successfully implemented in school screening programmes for identifying visual impairment. This has been investigated among various study populations around the globe [36‐38]. Sensitivity, specificity and predictive value varied among the studies, but included some promising results. A screening study in the USA illustrated how sensitivity differed per age category and found the highest sensitivity for detecting decreased vision in 3–5-year-olds (93–100%) [38]. In a study in Kenya, the sensitivity (77%) was reported to be similar to the conventional screening method: standard E-card assessments by school teachers [37]. Importantly, when visual impairment was detected by the smartphone application, personalized SMS reminders to attend the hospital were sent to the children’s parents or guardians, increasing adherence to hospital referral. Specificity of the digital screening was a bit lower compared to the conventional method (91% vs. 97%). The authors of the original article suggested testing strategies to reduce the false positive rate, including retesting the children who tested positive at first and a local triage service to review all children who initially screened positive.

Our systematic review indicates that the low rate of bias makes digital tools for self-assessment of VA a promising avenue for delivering eye care remotely. The precision of most tools seems lower than traditional VA charts, though these differences diminish when assessing individuals with better VA. These self-assessments of visual function have great potential for screening purposes, particularly to increase access to eye care, which is acutely relevant in the current COVID-19 pandemic. The landscape of digital medicine has been rapidly changing, especially over the last few years: we expect the accuracy of the current tools to improve with every iteration and new tools to be introduced in the coming years.