Telemedicine for screening eye disease in the remote Peruvian Amazon: proof-of-concept

Abstract

Background

Improvements in technology could facilitate task-shifting and ocular disease screening in rural areas.

Methods

Visual acuity (VA) was tested using a Ministry of Health 3-m VA card. Anterior segment photographs were taken using a three-dimensional printed cellphone attachment and remotely graded.

Results

Of 326 photographed eyes, 1 was ungradable. Of 123 eyes with non-refractive visual impairment, cataract was identified in 35.8%, pterygium in 41.5%, corneal opacity in 5.7% and phthisis in 2.4%.

Conclusions

While the cause of visual impairment cannot be determined without a posterior segment examination, the smartphone attachment proved to be easy to use by non-specialist workers and identified anterior segment pathology in most cases.

Introduction

Technological advancements can redistribute limited human resources in eye-care programs.¹ This includes task-shifting eye disease screenings from ophthalmologists to non-specialists, allowing ophthalmologists to devote more time to surgery. Because the aim of a screening program is to detect instead of manage disease it is well suited to asynchronous (i.e. store-and-forward) telemedicine where images are captured and sent elsewhere for interpretation.² We present a proof-of-concept study in which non-technical health workers used a low-cost smartphone attachment to take anterior segment images, providing telemedicine screening of anterior segment disease in a remote region of the Peruvian Amazon.

Materials and Methods

A convenience sample of 16 villages was selected in Alto Amazonas, Peru. Within each village, all adults aged ≥50 y from 30 randomly selected households were invited to participate.

Visual acuity (VA) and photographs were collected for all participants. VA was tested using a 3-m VA card developed by the Peruvian Ministry of Health,³ with available refractive correction (i.e. the WHO's definition of presenting VA) and pinhole occlusion. Visual impairment was defined as VA worse than 20/60 but 20/400 or better and blindness as VA worse than 20/400; individuals were classified based on the better-seeing eye. Three anterior segment photographs were taken per eye using a 12-megapixel (ƒ/1.8) camera of a mobile phone (Xioami A2) with proper attachment (Corneal CellScope, University of California Berkeley, Berkeley, CA USA). The smartphone attachment consists of a three-dimensional (3D)-printed housing, +25-diopter lens and two light-emitting-diode lights; it improves the native smartphone camera by providing magnification and external illumination.⁴ Because there was no network coverage in the study area, the photographs were synced once the team returned to their base of operations (i.e. a store-and-forward approach). The photographs were presented to seven trained ophthalmic assistants in random order and graded for the presence or absence of corneal opacity, phthisis bulbi, pterygium, cataract and pseudophakia, with reliability assessed using an intra-class correlation coefficient (ICC). Cataract severity was on a 1–4 scale modified from lens opacities classification system II; scores ≥3 were considered visually significant. Grading and severity discrepancies were settled using the majority consensus for each eye.

Results and Discussion

Of 207 eligible individuals, VA and imaging were completed on 163 (78.7%) individuals from 127 households. The median age was 59 (IQR 53–67) y and 42.9% (N=70) were female. Anterior segment photographs were assessed by seven photo-graders and the majority consensus was used for analyses. Of 326 photographed eyes, 93.9% (N=306) were judged to be of good quality, 5.8% (N=19) were poor but gradable and 0.3% (N=1) were ungradable. According to the consensus grade of the 325 gradable images, 59 had cataract, 105 had pterygium, 29 had corneal opacity and 3 had phthisis. Inter-rater agreement between the seven photo-graders was variable (ICC for cataract: 0.42, 95% CI 0.36 to 0.49; ICC for pterygium: 0.67, 95% CI 0.60 to 0.74; ICC for corneal opacity: 0.37, 95% CI 0.29 to 0.50; ICC for phthisis: 0.76, 95% CI 0.73 to 0.79).

Based on presenting VA, 63 people had visual impairment (38.7%, 95% CI 31.2 to 46.1%) and 7 were blind (4.3%, 95% CI 1.2 to 7.4%). After pinhole occlusion, the number of visually impaired and blind individuals was 49 and 7, respectively. Of 123 eyes with visual impairment or blindness on pinhole acuity (i.e. thought not due to refractive error), anterior segment photography identified visually significant cataract in 35.8% (N=44), pterygium in 41.5% (N=51), corneal opacity in 5.7% (N=7) and phthisis in 2.4% (N=3) (Figure 1 panels A and B). Representative images are shown in Figure 1 (panels C, D and E). Of the 325 photographed eyes, 1.2% (N=4) were pseudophakic.

Figure 1.

Frequency of eye disease diagnosed from anterior segment photography and representative anterior-segment images taken using the CellScope. (A) Eye-level frequency of each type of eye disease stratified by pinhole visual acuity status, with 95% CIs accounting for non-independence of eyes from the same person. (B) Venn diagram demonstrating the overlap of diagnoses in the total study population. (C) Corneal opacities and nasal pterygium; (D) phthisis bulbi; (E) cataract; and (F) 3D printed attachment and smartphone used to capture anterior segment images.

The prevalence of blindness and visual impairment was high in this area of the Peruvian Amazon and anterior segment pathology was identified in a majority of cases. Although the cause of visual impairment cannot be definitively established without a posterior segment examination, non-specialist health workers were capable of using the smartphone attachment to estimate the proportion of visually impaired adults with potentially treatable eye conditions (e.g. cataract, pterygium). We did not assess the reproducibility of CellScope images taken in different settings or by different individuals, although a previous study found excellent reproducibility of CellScope imaging for corneal opacity.⁵ Future studies could improve upon this effort by combining anterior segment imaging with posterior segment non-mydriatic imaging. Such a tool would have the potential to expand access to care in rural areas if integrated into existing health systems. This would be especially useful in an area like Alto Amazonas where eye care is accessible through state-sponsored programs (i.e. Seguro Integral de Salud) but the only ophthalmologist in the region is located in the province capital, which can be several days of travel from some of the more remote villages.

Conclusions

Smartphone diagnostic tools may expand identification of visually significant cataract and other anterior-segment diseases to remote areas with no access to eye care. The attachment described in this article was easily adopted by non-specialist health workers and took high-quality images. With appropriate validation the technology could be scaled up for use in population-based prevalence surveys (i.e. Rapid Assessment of Avoidable Blindness surveys) or alternatively as a referral tool for individuals with vision impairment, especially given the financial and logistical barriers for people in remote areas.⁶ We anticipate making the plans for the Corneal CellScope available to non-commercial organizations. Although image interpretation for the present study was conducted by humans as proof-of-concept, artificial intelligence algorithms could present unique opportunities for automated point-of-screening image grading and classification without the need for internet connectivity or reading centers.

Authors’ contributions

JMN, JDK, AGL, HAM and MM conceived the study; JMN, MM, SLT and JDK designed the protocol; JMN, MM, SLT and JDK carried out the fieldwork; JMN and JDK analyzed and interpreted the data; JMN and JDK drafted the manuscript; all authors critically revised the manuscript for intellectual content. All authors read and approved the final manuscript. JMN and JDK are guarantors of the paper.

Acknowledgements

We would like to acknowledge all the fieldworkers who did the truly hard work that made the study possible: Jim Bill Oliveira Garay, Técnico en Enfermería; Segundo Rosbel Soria Saavedra, Técnico en Enfermería; Rider Isaias Pizango Taminchi, Técnico en Enfermería; Hernan Dario Tapayuri Curitima, Técnico en Enfermería; Geiner Armas Damacen, Técnico en Enfermería; Kike Guevara Alarcón, Técnico Sistema de Información Geográfica; Lester Lenin Vela Tello, Técnico en Enfermería; Abdias Valles Davila, Técnico en Enfermería; Luis Edgar Payaba Pacaya, Técnico en Producción Agropecuario; Lizandro Guerra Rios, Técnico en Enfermería; Francisco Javier Noriega Morey, Técnico en Enfermería; Jorge Chasnamote Macahuachi, Técnico en Enfermería; and Gene Lucio Panaifo Garces, Técnico en Enfermería.

We also thank Edgardo Nepo Linares, MD as representative of PAHO Peru whose help laid the foundations for this project; Salvith Karen Melendez Ruiz, RN and Rosario Avellaneda, RN for sharing their experience and help in coordinating this project; and Nilda Trejo Maguiña, RN and Héctor Shimabuku Ysa, MD from the technical team of the Ocular Health and Prevention of Blindness department of the Peruvian Ministry of Health.

Funding

This work was supported by the National Eye Institute and the Fogarty International Center of the National Institutes of Health (NIH) under Award Number D43TW009343 as well as the University of California Global Health Institute (UCGHI) in the form of a Fogarty grant to John Nesemann. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or UCGHI. Support was also provided by That Man May See and Research to Prevent Blindness.

The funders had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; preparation, review or approval of the manuscript; and decision to submit the manuscript for publication.

Competing interests

None to report.

Ethical approval

Ethical approval was obtained from the University of California, San Francisco, Universidad Peruana Cayetano Heredia and the Pan-American Health Organization.

Data availability

The data underlying this article will be shared upon reasonable request to the corresponding author.