Utilization of a self-refracting system on the streets to improve vision in an urban unhoused population

Abstract

Background

To evaluate the effectiveness of the USee self-refracting eyewear kit in an urban unhoused population and identify predictors of instances in which the USee does not improve vision.

Methods

In this retrospective cross-sectional study, 110 patients (220 eyes) experiencing homelessness in Phoenix, Arizona, were evaluated before refractive correction (control) and after USee refraction. Visual acuity (VA) testing and USee refraction were performed in mobile outreach settings. Mixed-effects models were used to determine VA outcomes. USee failure was defined as USee-corrected visual acuity (USeeVA) worse than 20/40 or any worsening of USeeVA compared to uncorrected visual acuity (UCVA). Predictors of USee failure were evaluated using receiver-operating characteristic curves (AUC-ROC). Thresholds were determined using Youden’s Index and Liu’s method. The primary outcome was the change in binocular VA. Secondary outcomes included the rate of achieving 20/40 vision or better and determining what factors predicted USee device failure.

Results

Mean binocular VA improved from 20/64 to 20/31. The proportion of patients achieving 20/40 vision or better increased from 40% to 81%. Pinhole VA alone (AUC: 0.74), cylinder alone (AUC: 0.67), and their combination (AUC: 0.81) predicted USee failure. A pinhole VA worse than 20/60 had sensitivity 0.667, specificity 0.819 (positive likelihood ratio, LR⁺: 3.69, negative likelihood ratio, LR⁻: 0.41). A cylinder threshold less than − 0.5 diopters (D) had sensitivity 0.346, specificity 0.511 (LR⁺: 0.71, LR⁻: 1.28).

Conclusions

The USee device improves vision in low-resource, transient populations. Screening using pinhole VA and cylinder may help optimize patient selection for USee-based refractive correction.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12886-026-05077-0.

Vision impairment is a significant burden among unhoused individuals around the world. Sayal et al. report a prevalence of 25.6%, six times higher than that of housed individuals and ten times that of the global population [1]. Visual impairment can be broadly classified into refractive and non-refractive errors. Refractive errors include myopia, hyperopia, astigmatism, and presbyopia, which can often be corrected with glasses, while non-refractive visual pathology typically requires more advanced care.

For individuals experiencing homelessness, vision impairment negatively affects their daily lives, contributing to poor job and academic performance, reduced workforce participation, and social isolation. For unhoused individuals, there are additional risks of safety and theft associated with low visual acuity (VA) [1]. Only 22% of unhoused individuals visited an eye care provider (either an optometrist or ophthalmologist) in 2020 [1]. A study by Hennein et al. demonstrated that less than half of the patients referred to a non-profit organization to receive free glasses after being seen at a free ophthalmology clinic showed up for their appointment [2]. Systemic barriers to health care, such as the hesitancy to leave or secure their belongings and inadequate transportation, are encountered when referring unhoused people to brick-and-mortar locations for care. Social barriers to healthcare, such as the perception of prejudice and poor service of providers towards those experiencing homelessness, further worsen their ocular health inequality [3]. However, mobile health care and point-of-service health care delivery are methods that can lessen health care disparities in this population.

A novel manual-refracting device, the USee device (Global Vision 2020, Easton, Maryland, USA), may overcome many of these barriers. The USee vision kit includes the USee refraction device, snap-in lenses, glasses frames, and vision charts, containing everything required to distribute glasses immediately and inexpensively in low-resource environments, with an estimated cost of $5 per unit [4]. The ease of use and ability to immediately dispense glasses are especially important for unhoused populations due to the transient nature of their lives and the need for prompt vision correction. It has the potential to address these concerns, but currently, only a single peer-reviewed study has examined its utility, efficacy, and accuracy compared to standard methods of refraction [5].

Street Medicine Phoenix (SMP) is a medical outreach program at the University of Arizona College of Medicine – Phoenix that provides healthcare services for individuals experiencing homelessness. In this study, we assess the performance of the USee device in a community setting, discuss modifications made to the refraction process to improve use, and review limitations of the device.

Methods

This study consisted of a retrospective chart review of patients’ prior vision measurements. The study complied with the Health Insurance Portability and Accountability Act and followed the tenets of the Declaration of Helsinki, with institutional review board approval (STUDY00003510) obtained from the University of Arizona College of Medicine – Phoenix. Inclusion criteria were adults aged 18 years or older who underwent a vision examination and USee refraction with the SMP outreach team at the Lodestar Day Resource Shelter, Andre House, Grace Lutheran Church, or the surrounding streets of the Keys to Change Arizona Campus (Phoenix, Arizona, USA) between October 2023 and January 2025. Exclusion criteria included those who were not unhoused, under 18 years of age, did not complete both uncorrected visual acuity testing and USee refraction, and those whose data could not be linked to a unique Central Arizona Shelter Services ID (CASS-ID, an unhoused identification number). No exclusion criteria were applied based on refractive error magnitude, astigmatism, or known ocular pathology, as the study aimed to evaluate USee device performance in a real-world, unselected outreach population. Global Vision 2020 (GV2020) was not affiliated with our research.

Age, date of service, CASS-ID, uncorrected visual acuity (UCVA), pinhole visual acuity (PHVA), autorefractor (AR) measurements, lens code from the USee refractor, and USee-corrected visual acuity (USeeVA) were obtained. USeeVA refers to VA measured with correction (i.e., with the USee device and corresponding glasses created from USee refraction). All data was de-identified.

USee device

The USee device uses a trial frame (Fig. 1A) with a bar of step gradation lenses (Fig. 1A, label 5) positioned in front of each eye aperture (Fig. 1A, label 3). A knob is used to rotate the gradation lenses (Fig. 1A, label 5) until the patient feels their vision is clear while looking at a visual acuity chart [6].

Fig. 1.

A USee device (1: Focus dial, 2 and 4: Components of the device frame; 3: Eye aperture, 5: Gradation lenses); B Components of the USee kit

Each marking on the lens code strip corresponds to a particular lens of a specific sphere. Therefore, the USee device measures sphere only and does not measure cylinder or provide a spherical equivalent; the measurement range of the device is − 6.00 D to + 4.00 D. The power of each lens was verified with a lensometer. The sphere-to-color code conversion chart was only available to SV and SZ, who completed data analysis, to ensure that the remainder of the research team remained masked.

REDCap (Research electronic data capture)

Retrospective chart review was performed by analyzing data stored in REDCap (Vanderbilt University, Nashville, Tennessee, USA), the electronic medical record used by SMP. The REDCap form utilized in our study is provided as an example to aid other outreach programs with data collection (Online Resource 1).

Vision screening process

Before participating in data collection, students administering the study watched a training video provided by the manufacturer. However, slight modifications were made to our protocol to obtain more efficient and accurate results. To start, we identified an open space that was free of obstacles and illuminated with either sunlight or room lighting to perform vision testing. An Early Treatment Diabetic Retinopathy Study (ETDRS) vision chart was positioned at eye level, aligned with the seated patient. Depending on the available space, a chair was placed at either five or ten feet from the chart, and the corresponding vision chart was used.

First, the patient’s VA was obtained. An occluder was held over the left eye (OS), and VA for the right eye (OD) was tested. Patients were asked to read each line of the vision chart out loud. If they correctly identified more than half of the letters, they were asked to read the next line. When they were unable to identify more letters, their uncorrected VA was recorded. A pinhole was placed over the right eye, and the process was repeated to obtain PHVA. This process was then repeated for the left eye.

Patients were then instructed to wear the USee device like a pair of glasses. Researchers were not masked to the previously recorded VAs. Adjustments were made to position the aperture in the patient’s line of sight. The right eye was tested first, while the left eye was covered with an occluder. The USee gradation lens element was initially set at the highest plus power (White 7), corresponding to the lowest part of the lens element. As the researcher rotated the knob to decrease plus power, moving the lens element downward, the patient was asked to notify the researcher once they could read the first line of the vision chart (ETDRS 20/100).

Patients were then asked to read each line out loud. If they correctly identified more than half of the letters, they were asked to read the line below. If they were unable to, the knob would be rotated once to move the lens element down by one lens equivalent. If they still could not read the next line, the knob was rotated two additional times to confirm that each additional increment would not improve their vision. If their vision improved (able to read the line below) with subsequent adjustments of the USee refractor, the process would be repeated until the patient noted no further improvement in their vision despite additional increments or their VA reached 20/20. After the refractor was set to the lowest lens equivalent that produced their best USeeVA, the lens color code and VA for that eye were recorded. The occluder was then switched to the right eye, and the process was repeated for the left eye.

If patients could not read any letters on the vision chart, the following descriptors and associated logarithm of the Minimum Angle of Resolution (LogMAR) were recorded: count fingers (CF, LogMAR 1.9), hand motion (HM, LogMAR 2.3), light perception (LP, LogMAR 2.7), or no light perception (NLP, LogMAR 3.0) [7].

Measurements from the autorefractor (QuickSee, PlenOptika, Cambridge, MA) were obtained following USee refraction. Pupillary distance (PD) was measured with a PD ruler, and the PD on the AR was adjusted accordingly. Patients were instructed to hold the AR like a pair of binoculars and confirm that they could see a red light. When confirmed, they were asked to focus on a distant point. Measurements were acquired and recorded when indicated by the device. These values were not used to prescribe eyewear but only as a comparison with the USee device for sphere measurements.

After refraction was completed for both eyes, the lenses corresponding to the patient’s recorded lens color codes were retrieved, placed into empty frames, and given to the patient as a pair of glasses. USeeVA was measured monocularly and binocularly. Patients were given a microfiber bag to store their new glasses, and their visit was concluded. All measurements were obtained on the same visit.

Group classification

Recorded eyes were separated into two categories: eyes for which the USee worked (USeeSuccesses) and eyes for which the USee failed (USeeFails). USeeFails were defined in two ways: USeeVA worse than 20/40, or any worsening of USeeVA compared to UCVA (i.e., UCVA of 20/30 to USeeVA of 20/63). The 20/40 threshold was selected because it represents the minimum VA required for legal driving in Arizona, is consistent with the manufacturer’s reported benchmark, and has been used as the primary success criterion in prior USee studies [5, 19]. USeeSuccesses were defined as eyes that did not meet these criteria. Eyes with less than a two-line difference on the ETDRS chart (0.2 LogMar of difference) were categorized as noise or clinically insignificant and excluded from our analysis [9].

Statistical analysis

Data was analyzed with Google Python Colab. Because right and left eye data points are correlated with each other, mixed-effects models were used to account for any inter-eye correlation [10]. ParticipantID and left/right eye were random intercepts, and the variable of interest was a fixed effect. To assess whether researchers’ proficiency with the device influenced our results, we used mixed-effects models to evaluate the relationship between the date of the measurement and both the number of USeeFails and the amount of patient improvement in LogMar with USee glasses. We also used mixed-effects models to determine whether USee refraction significantly improved VA, and if there was a difference between VA obtained from the USee device prescription versus the final lens prescription.

Bayesian logistic mixed-effects regression was used to identify univariate odds ratios (ORs) associated with USeeFails. A multivariable mixed-effects regression was then performed to assess the combined influence of all candidate predictors. Model specification was reviewed for appropriateness prior to fitting the regression. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC-ROC). Predictions were assessed at a probability threshold of 0.5 using classification metrics including accuracy, precision, recall, and F1 score. ROC curves were constructed and compared visually across models. Thresholds for significant predictors were determined using both the Youden and Liu methods.

Missing data were assessed for consistency with a Missing at Random (MAR) mechanism across key variables, including age, sex, UCVA, PHVA, and AR measurements. Patients missing data for the primary outcome were excluded from the relevant sub-analyses.

Results

220 eyes of 110 patients who underwent vision testing (62 males, 45 females) with a mean (± standard deviation) age of 50.7 (± 13.71) years were included in the study. Several sub-analyses may have data missing from some eyes or patients. Baseline and outcome characteristics are summarized in Table 1.

Table 1.

Baseline and outcome characteristics of study population (n = 110 patients)

Measure	OD mean ± SD (n)	OS mean ± SD (n)
Demographics
Age, mean ± SD	52.4 ± 12.8 years (range 23–78, n = 101)
Sex	Male (62, 57.9%), Female (45, 42.1%) (n = 107)
Visual acuity (LogMAR)
UCVA	0.68 ± 0.56 (n = 101)	0.69 ± 0.56 (n = 102)
PHVA	0.44 ± 0.45 (n = 87)	0.38 ± 0.32 (n = 87)
USeeVA	0.29 ± 0.33 (n = 99)	0.31 ± 0.36 (n = 102)
Autorefractor (Diopters)
Sphere	−0.96 ± 2.99 (range − 10.50 to 7.75, n = 91)	−0.85 ± 2.96 (range − 10.75 to 6.75, n = 95)
Cylinder	−0.89 ± 0.86 (range − 4.75 to 0.00, n = 89)	−0.94 ± 0.92 (range − 6.00 to 0.00, n = 94)
Spherical equivalent	−1.35 ± 2.85 (range − 8.25 to 7.25, n = 80)	−1.30 ± 2.74 (range − 8.50 to 5.25, n = 82)

Abbreviations: OD = Right eye; OS = Left eye; UCVA = Uncorrected visual acuity; PHVA = Pinhole visual acuity; USeeVA = USee-corrected visual acuity. Cylinder reported in minus-cylinder notation. LogMAR reference: 0.0 = 20/20, 0.301 = 20/40, 1.176 = 20/300. Sample sizes vary due to missing data (see Table 3 for missing data summary)

Mean VA OU before refraction was 20/64, and mean VA OU after refraction was 20/31, an improvement of 3 lines. Of 200 eyes with complete paired LogMAR data (99 OD, 101 OS; 102 patients), 146 eyes (73.0%) improved by one or more Snellen lines (LogMAR improvement ≥ 0.1), including 72 of 99 right eyes (72.7%) and 74 of 101 left eyes (73.3%). It is important to note that UCVA OU was not systematically collected; the paired analysis was therefore performed per eye with the patient as the random effect. From our cohort of 110 patients, the proportion of subjects correctable to 20/40 OU was 81.1%, which is similar to GV2020’s estimate of 80%. Among the 98 patients with complete paired data, USee refraction significantly improved VA outcomes across various vision thresholds, with the proportion of patients achieving 20/32 or better increased from 30.6% to 70.4%, and 20/25 or better from 12.2% to 54.1% (Table 2).

Table 2.

Vision threshold achievement before and after USee refraction (n = 98 patients with complete paired data)

Vision threshold	Before USee refraction, n (%)	After USee refraction, n (%)	Absolute increase (pp)
20/20	4 (4.1%)	27 (27.6%)	+ 23.5 pp
20/25	12 (12.2%)	53 (54.1%)	+ 41.8 pp
20/32	30 (30.6%)	69 (70.4%)	+ 39.8 pp
20/40	41 (41.8%)	80 (81.6%)	+ 39.8 pp

Proportions reflect patients with at least one eye meeting or exceeding the vision threshold. Absolute increase expressed in percentage points (pp). All improvements in achieving each threshold from before to after USee refraction were statistically significant

There was no statistically significant association between time and failure rate (β = -0.001, SE = 0.001, p = 0.073), though the coefficient suggests a slight downward trend in failure rates over time. Improvement in LogMAR per day did not systematically increase or decrease over the study period (β = 0.000, SE = 0.000, p = 0.411). There was a statistically significant improvement in vision from UCVA to USeeVA after wearing USee glasses (β = -0.391, SE = 0.032, p < 0.001, 95% CI [-0.455, -0.328]). There was no difference between VA obtained with the USee refractor device and VA obtained while wearing glasses created with the USee prescription (β = -0.003, SE = 0.023, p = 0.881, 95% CI [-0.041, 0.048]). USeeVA was significantly better than PHVA across paired eyes (β = −0.065, SE = 0.024, p = 0.007, 95% CI [− 0.112, − 0.018]; n = 171 eyes, 90 patients), indicating that USee refraction provides meaningful correction beyond the improvement achievable with pinhole alone.

Out of 220 eyes and 110 patients, we identified 61 eyes as USeeFails, 105 eyes as USeeSuccesses, and 54 eyes as noise. Among the 98 patients with complete paired data, 41 (41.8%) already met the binocular 20/40 threshold at baseline. This result is per patient and the per eye proportion of individual eyes meeting 20/40 at baseline was lower than 40%. From our per-eye analysis framework, 60 patients had the capacity to improve (defined as at least one eye worse than 20/40 at baseline), with 41 (68.3%) achieving 20/40 or better with USee refraction, and 19 (31.7%) did not. To characterize the 19 non-improvers, we compared them to the 41 who succeeded. Failures had significantly worse baseline PHVA (median logMAR 1.176, IQR 0.699–1.479) than successes (median 0.498, IQR 0.301–1.205; Mann-Whitney U, p = 0.002). Among the 18 non-improvers with available PHVA data, 4 had PHVA of 20/40 or better, likely due to uncorrected astigmatism which the USee cannot correct. Fourteen had PHVA worse than 20/40, suggesting a non-refractive component such as ocular pathology. Failures were also more likely to have AR sphere exceeding the USee range of −6.00 to +4.00 D: 9 of 18 (50.0%) vs. 4 of 55 (7.3%; Fisher's exact, p = 0.0002), respectively. Cylinder trended higher in USeeFailures (median 1.38 D) than successes (median 1.00 D) but did not reach significance (p = 0.070).

Missing data were assessed for all key variables. The extent of missing data within age, sex, and UCVA was less than 10%. PHVA was missing in 20 patients (18.2%), and AR measurements were missing in 14 to 21 patients (12.7–19.1%) depending on the eye and parameter. We compared age, sex, and baseline UCVA between patients with and without PHVA data. Patients with PHVA data did not differ significantly in age from those without (52.2 ± 11.9 years vs. 52.4 ± 13.0 years). However, patients with PHVA data had worse mean baseline LogMAR than those without (0.736 vs. 0.320), attributable to two identifiable factors: a tendency to omit pinhole testing when baseline UCVA was already adequate and operational constraints at a single high-volume event (September 7, 2024, accounting for 7 of 20 missing cases). While this pattern is consistent with MAR, where missingness is explained by observed variables rather than unobserved values, we acknowledge that complete verification of MAR is not possible. Given that PHVA was used as a predictor rather than the primary outcome and that patients missing pinhole data were excluded only from the relevant sub-analyses, we do not believe this introduces meaningful bias into the primary results.

Table 3.

Missing data summary of key study variables (N = number of patients)

Variable	N present	N missing	% missing
Age	101	9	8.2%
Sex	107	3	2.7%
Uncorrected VA OD/OS	101	9	8.2%
Pinhole VA OD/OS	90	20	18.2%
AR sphere OD/OS	91–96	14–19	12.7–17.3%
AR cylinder OD/OS	89–95	15–21	13.6–19.1%
USeeVA OD/OS (LogMAR)	99–102	8–11	7.3–10.0%

Abbreviations: VA = Visual acuity; OD = Right eye; OS = Left eye; AR = Autorefractor; USeeVA = USee-corrected visual acuity; LogMAR = Logarithm of the minimum angle of resolution

We identified several variables of interest as potential predictors of USee failure: baseline refractive error (measured by UCVA), PHVA, the difference between UCVA and PHVA, cylinder, sphere, and age. Spherical equivalent was not analyzed separately, as its components (sphere and cylinder) were already included.

Univariate ORs were significant for UCVA, PHVA, and cylinder. UCVA lost significance from unadjusted to adjusted odds ratios (Table 4). The adjusted ORs for the multivariable model exhibited wide confidence intervals (Table 4), attributable to the limited sample size available for complete-case multivariable analysis (75 patients, 126 eyes) and the amplification of uncertainty inherent to logistic regression on a logarithmic scale. The AUC-ROC analysis and threshold-based metrics in Table 5 therefore serve as the primary tools for clinical interpretation. To evaluate whether PHVA and cylinder predicted USee failure, we performed logistic regression modeling. Three models were tested: (1) PHVA alone, (2) cylinder alone, and (3) both predictors combined. The combined model demonstrated the highest discriminative ability (AUC = 0.81), compared to PHVA-only (AUC = 0.74) and cylinder-only (AUC = 0.67) (Fig. 2). At a probability threshold of 0.5, the combined model achieved an overall accuracy of 80%, with a precision of 84% and an F1 score of 0.67 for failure cases.

Table 4.

Unadjusted and adjusted odds ratios of USee failures

Predictor	Unadjusted OR (95% CI)	Patients	Eyes	Adjusted OR (95% CI)
Uncorrected LogMAR	12.69 (2.57, 84.10)	91	164	2.20 (0.16, 40.61)
PinHole LogMAR	418.22 (21.16, 8283.33)	83	148	491.76 (6.88, 41856.40)
Pin Diff	1.41 (0.10, 26.36)	82	146	9.54 (0.27, 328.98)
MR sphere	0.81 (0.56, 1.16)	91	160	0.80 (0.51, 1.31)
AR sphere	0.85 (0.61, 1.19)	85	148	1.12 (0.70, 1.73)
Cylinder	0.25 (0.08, 0.68)	84	146	0.19 (0.06, 0.58)
Age	1.02 (0.93, 1.13)	92	166	0.99 (0.91, 1.08)

Adjusted Odds Ratios for 75 patients and 126 eyes

Abbreviations: OR = Odds ratio; CI = Confidence interval; LogMAR = Logarithm of the Minimum Angle of Resolution; Pin Diff = The difference between uncorrected visual acuity and pinhole visual acuity; MR Sphere = Manual refractor sphere; AR Sphere = Autorefractor sphere

Table 5.

Diagnostic performance of pinhole visual acuity and cylinder thresholds for predicting failure of USee-based vision correction, using a pre-test probability of 36.7%

Variable	Method used	Threshold	Sensitivity	Specificity	LR⁺	LR⁻	Post-test probability (Positive)	Post-test probability (Negative)
PHVA	Youden/Liu	0.4771 (20/60 on Snellen Chart)	0.667	0.819	3.69	0.41	68.10%	19.40%
Cylinder	Liu	-0.5	0.346	0.511	0.71	1.28	29.00%	43.00%
Cylinder	Youden	Not applicable	—	—	—	—	—	—

Abbreviations: LR = Positive likelihood ratio; LR = Negative likelihood ratio; PHVA = Pinhole visual acuity

Fig. 2.

The receiver operating characteristic curves of the best parameters for discriminating between USeeFails and USeeSuccesses. Abbreviations: PHVA = Pinhole visual acuity; AUC = Area under the curve; ROC = Receiver operating characteristic

We developed thresholds to identify values at which a patient’s eye is more likely to be classified as a USeeFail and reported on its sensitivity, specificity, likelihood ratios, and post-test probabilities (Table 5). Excluding noise, the pretest probability of USeeFailures in our dataset is 36.7%.

Discussion

Refracting homeless individuals using the USee device resulted in substantial improvement in vision, increasing the proportion of individuals achieving a VA of 20/40 or better from 40% to 80%, meeting the benchmark supported by the manufacturer. Of the 60 patients with capacity to improve, approximately two-thirds (41/60, 68.3%) achieved 20/40 or better with USee refraction, while approximately one-third (19/60, 31.7%) did not. Unlike prior USee studies, our cohort was unscreened for non-refractive pathology; basic screening (e.g., excluding cataract patients, common in this population) would likely raise this conservative estimate. While 20/40 is not an optimal cutoff for VA, it is a reasonable goal in populations with very limited access to any vision care. Furthermore, in most states in the United States, such as in Arizona, the minimum mandated VA to legally drive is 20/40 or better in at least one eye [8, 11]. The U.S. Preventive Services Task Force has noted that VA worse than 20/40 is associated with reduced ability to perform both basic and instrumental activities of daily living, work, and drive safely,⁸ and the Salisbury Eye Evaluation demonstrated significant functional decline across activities of daily living in individuals with binocular VA worse than 20/40 [12]. We acknowledge that 20/40 is not sufficient for all visual tasks, and patients achieving this threshold may still experience difficulty with near-vision activities, such as reading. However, by enabling a significant portion of the study population to achieve legal driving VA, the USee device may improve access to functional vision thresholds.

The refractive profile of our population was broadly comparable to US national estimates, with the exception of hyperopia, which appeared less prevalent in our sample [13–16]. Critically, 13 of 97 patients (13.4%) with available AR data had at least one eye with sphere outside the USee range (− 6.00 to + 4.00 D); these patients had significantly worse USeeVA outcomes than in-range patients (mean LogMAR 0.677 vs. 0.224; Mann-Whitney U, p < 0.0001), underscoring that high myopia exceeding − 6.00 D is a key device limitation.

Internal checks: USee refractor visual acuity compared to lenses visual acuity and time

Throughout our study, we noted minor discrepancies between the VA obtained while patients were using the USee device and while wearing glasses with USee-derived prescription lenses. However, as there was no statistically significant difference between the measurements, these differences could be attributed to random error and the subjective nature of testing VA. This was a valuable internal check to ensure that the device’s prescription matched with the provided lenses.

We also found that the number of failures per day did not statistically decrease over time, nor was there a correlation between the amount of improvement in LogMAR and time. This served as an additional internal check, ensuring that familiarity with the device did not affect our results.

Characteristics of patients for whom the USee failed to improve vision

Poor PHVA emerged as the strongest predictor of USee failure, with the combined PHVA and cylinder model demonstrating the best overall discriminative ability. Although neither threshold alone achieves the likelihood ratios typically required for a definitive screening test, PHVA offers clinically actionable guidance: when PHVA does not improve beyond 20/60 and/or cylinder exceeds − 0.5 D, the USee device is less likely to provide meaningful benefit. PHVA has long been used as a screening tool to determine whether a patient would benefit from refractive correction alone, and its emergence as the strongest predictor in this study is consistent with that role.

The USee device provides only spherical correction, making it suboptimal for patients with significant astigmatism. Cylinder was a significant predictor of failure, though less discriminative than PHVA, consistent with the known limitation of the device in handling astigmatic correction [16].

Changes made to the training program

We implemented several changes to GV2020’s USee training to improve the efficiency and accuracy of visual outcomes. For instance, GV2020 recommended that patients turn the knob on the device to refract themselves. In our experience, patients would skip through the lens markings very quickly or stop the dial in between lens markings. Having the healthcare worker control the device allowed for more precise adjustments to determine if a higher lens power would improve vision. We also found that checking three successive increments in lens power was more accurate at evaluating vision improvement than checking only one or two higher power lens increments. For example, patients reading at the 20/50 line may not be able to distinguish a difference between Green 1 and Green 2 but would likely perceive an improvement at Green 3, 4, or 5, highlighting the importance of checking incremental changes in lens power systematically. GV2020 also recommended that patients physically hold a glasses bag over the non-tested eye aperture as an occluder during vision measurements. We designed a clip-on occluder that further minimized patient interactions with the device (Online Resource 2).

Comparison with other studies evaluating the USee device

Several other portable refraction devices have been developed to address uncorrected refractive error in underserved populations. The EyeNetra (EyeNetra Inc., Cambridge, MA), a smartphone-based subjective refraction system, has been shown to produce small but clinically meaningful differences from subjective refraction, with a tendency toward myopic overcorrection when used without professional assistance. The QuickSee (PlenOptika, Cambridge, MA), a handheld wavefront autorefractor, has demonstrated strong agreement with subjective refraction across low-resource settings in multiple countries. Salmeron-Campillo et al. evaluated a web-based smartphone app measuring spherical and cylindrical over-refraction, the residual refractive error in patients already wearing prescribed correction, reporting agreement within ± 0.50 D of clinical refraction in 85% of cases among healthy young university students with known baseline prescriptions [17]. This approach is not directly applicable to populations such as ours, in which patients present with no existing glasses, no prior prescription, and no established eye care. Luo et al. demonstrated that a smartphone app measuring spherical equivalent via far-point distance improved VA by a mean of five lines in myopic patients in Ethiopia, with app-prescribed correction matching optometrist performance in 90% of eyes. This is a design more analogous to the USee in that both generate a de novo sphere-only prescription in a low-resource setting [18]. However, Luo et al. excluded hyperopic refractive error, required an optometrist for administration, and did not report the source, cost, or inclusion of glasses dispensing as part of the intervention. By contrast, the USee is a fully integrated point-of-service kit: the refracting device, lens inventory, and dispensed glasses are provided in a single encounter, with no external supply chain, optometrist, or follow-up visit required. This model is particularly critical for transient unhoused populations, for whom any multi-step dispensing pathway represents a major barrier to receiving correction.

The only available research on the outcomes of USee refraction is from studies by Annadanam et al. and Miller et al. [5, 19] Annadanam et al. provided the first study reporting on the accuracy and ease of use of the USee refractor among adults in optometry clinics. They excluded patients with coexisting ocular pathology as determined by an optometrist and those with high refractive error (plus or minus 6.00 D). Miller et al. also used similar exclusion criteria in addition to excluding those with astigmatism greater than + 2.00 D. Both studies excluded patients who may be encountered in the low-resourced populations that the device is meant to service.

We found it important to design a protocol for using the USee device in a patient population without access to the technology or expertise to exclude patients. Ideally, for appropriate use in low-resource settings, the USee device would be used on patients for whom we have limited to no knowledge of their ocular parameters. Our study offers insight into how well the USee performs in these real-world conditions.

Table 6 compares several results between our study and that of Annadanam et al., demonstrating the differences between the two populations. While we defined USeeFail as VA worse than 20/40, Annadanam et al. used a VA cutoff of worse than 20/30 without accounting for noise. For this comparison, we redefined our definition of USeeFail.

Table 6.

Comparison of results between studies evaluating efficacy of the USee

	Vetsa	Annadanam
Age (mean ± SD)	51.78 ± 12.57 years	53.1 ± 18.6 years
Number of eyes	220	60
Mean sphere of MR (± SD, range)	-1.14 (± 2.42, -6.00 to 4.00)	− 1.22 D (± 2.42, − 6.00 to 3.00)
Improvement in ETDRS eye chart	3 lines	4 lines
Age odds ratio**
< 40	Reference	Reference
40–65	0.586 (95% CI [0.0982–3.49], p = 0.556)	1.88 (95% CI [0.32 -11.0], p = 0.487)
66+	5.09 (95% CI [0.256–101], p = 0.286)	6.00 (95% CI [1.05–34.3], p = 0.044)
Cylindrical power odds ratio**
0-0.25	Reference	Reference
0.5-1.0	8.96 (95% CI [1.53–52.5], p = 0.015)	6.38 (95% CI [0.74–55.1], p = 0.092)
1.25+	52.3 (95% CI [5.63–485], p = 0.001)	21.3 (95% CI [1.91–236], p = 0.013)
% within 20/30*	54.9%	83.3%

*20/32 in our dataset was considered sufficiently close to 20/30 to qualify as a success, consistent with the criteria used by Annadanam et al

**The odds ratios represent the odds that a subgroup will have a visual acuity worse than 20/30 per eye relative to the reference group (patients aged < 40 years; cylindrical power 0-0.25 D)

Abbreviations: SD = Standard deviation; MR = Manual refraction; ETDRS = Early treatment diabetic retinopathy study

The age distribution of our patients is similar to that of Annadanam et al., but older than that of Miller et al. (25.2 ± 4.5 years). Given this difference, our patient population likely had a higher prevalence of non-refractive ocular pathology than in the Miller cohort. Additionally, we did not apply exclusion criteria based on sphere or astigmatism in determining eligibility criteria for USee use. Annadanam et al.’s cohort had access to treatment for non-refractive pathology and benefited from an optometrist-led screening process prior to USee use. These factors may explain the lower percentage of patients in our study achieving a USeeVA with their acquired glasses better than 20/30 [5].

Unlike Annadanam and Miller et al., we did not apply exclusion criteria based on age, sphere, astigmatism, or ocular pathology risk, meaning our cohort likely had a higher prevalence of age-related non-refractive pathology. One would expect age to have correlated with rates of USee fails, but this was not supported by our results. We find it reassuring that the OR for cylinder is significant in both papers for predicting USee failure because it is a known limitation of the device.

While ORs are effective for quantifying the relative effect of variables, such as cylinder, on the likelihood of USee failure, they lack intuitive clinical interpretability. Threshold-based metrics like Youden Index or Liu’s method are better at setting a clinical threshold for decision-making and provide more actionable insights.

Our sample size of 220 eyes is larger than that of Annadanam et al. (60 eyes) and Miller et al. (48 eyes) and therefore has more statistical power. We account for inter-eye correlations, while these two studies opted to choose only one eye per patient.

Limitations of our research

Despite the utility of the USee device, various factors may impact its applicability. In particular, optimizing our setting was an important aspect of obtaining accurate measurements. We performed vision screenings indoors and outdoors, and particularly in an outdoor setting, there were additional factors to consider while setting up our screenings. When possible, we avoided using laminated vision charts to minimize glare and tried to find shaded locations to avoid checking vision in bright sunlight, which could cause unwanted pupillary miosis.

When assessing VA, we assumed that all patients who were counting fingers were tested at a standard distance of 10 feet, where the VA chart is typically positioned. However, there were several instances when patients were unable to achieve CF at that distance, prompting the operator to move closer to the patient. We did not account for variations in CF distance, which may have affected the accuracy of their true UCVA and the extent of their improvement. Notably, eight patients in our study had a UCVA of CF.

We also lacked methods to evaluate non-refractive causes of vision loss, such as cataracts, glaucoma, or retinal disorders, beyond utilizing the PHVA as a screen. Patients often had no history of recent eye care. Other than failure of improvement in VA with USee refraction and their past medical history, we had no other means to encourage referrals to higher levels of care. Additionally, the USee device has a measurement range of − 6.00 to + 4.00 D; when this limit was reached during refraction, we dispensed the maximum available correction if the patient reported any subjective improvement, acknowledging that optimal correction was not achievable in these cases. We identified 13 of 97 patients (13.4%) with available AR data as having at least one eye with sphere outside this range, all due to high myopia; these patients had significantly worse USeeVA outcomes, and this measurement ceiling should be considered a meaningful device limitation when planning deployment in unselected outreach populations.

We encountered several patients who were only able to read the largest letters on the VA chart (ETDRS 20/300) despite using the highest-strength lenses available in the kit. Even though these patients were not refracting better than 20/300, they reported enough improvement with the glasses, which were ultimately provided to them. Although this practice does not necessarily align with our research goals, we felt it was in the best interest of the patients we were serving. In a highly transient population with chronically low access to healthcare, the slight improvement they felt with the USee glasses could provide some marginal benefit, even if they were not adequately refracted. Additionally, as the USee device provides spherical correction only, patients with clinically significant astigmatism are unlikely to achieve adequate correction; based on the refractive error distribution in our population, a substantial proportion of patients fall into this category, which should be considered when determining candidacy for USee-based refraction. Finally, prospective validation of the predictive thresholds in an independent cohort is needed before clinical implementation.

Despite the limitations of the USee device, we have witnessed how it can play a role in bringing visual improvement to patients. The benefits of the USee kit include its ease of use, portability, and point-of-service eyewear distribution. Compared with other methods of refraction and eyewear distribution, the USee device is straightforward for healthcare professionals to learn. We have heard many testimonials of people finding employment, regaining their ability to drive, and having greater ease in performing activities of daily living with their new eyewear. Our study independently evaluates the utilization of the USee device in appropriate situations and provides a protocol for other organizations attending to the needs of low-resourced patient populations.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

VA

Visual acuity

SMP

Street Medicine Phoenix

GV2020

Global Vision 2020

CASS-ID

Central Arizona Shelter Services ID

UCVA

Uncorrected visual acuity

PHVA

Pinhole visual acuity

USeeVA

USee-corrected visual acuity

REDCap

Research Electronic Data Capture

ETDRS

Early Treatment Diabetes Retinopathy Study

OD

Right eye

OS

Left eye

OU

Both eyes

LogMAR

Logarithm of the Minimum Angle of Resolution

CF

Count fingers

HM

Hand motion

LP

Light perception

NLP

No light perception

SE

Standard error

D

Diopters

AUC

Area under the curve

ROC

Receiver operating characteristic curve

Author contributions

SV and SZ contributed to study design, data collection, and statistical analysis. RZ, CE, and RU assisted with data acquisition. RF provided oversight, mentorship, and critical revision. All authors read and approved the final manuscript.

Funding

No external funding was received for this study.

Data availability

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study complied with the Health Insurance Portability and Accountability Act and followed the tenets of the Declaration of Helsinki. This study was approved by the University of Arizona Institutional Review Board (STUDY00003510). Informed consent from all participants was waived due to the retrospective nature of the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.