Stop Blindness in Coastal Bangladesh: Testing the Effectiveness of Community-Based, Artificial Intelligence-Assisted Eye Disease Screening in Coastal Bangladesh
试验速览
- 阶段
- 不适用
- 状态
- 招募中
- 发起方
- 入组人数
- 20,000
- 试验地点
- 1
- 主要终点
- Number of individuals screened
研究概览
简要总结
Eye disease affects 2.2 billion people globally, which in turn adversely affects schooling, economic productivity, and participation in social life. The primary conditions contributing to visual impairment and blindness include cataracts, age-related macular degeneration (AMD), glaucoma, diabetic retinopathy (DR), refractive error, and presbyopia. Early detection of eye disease can provide substantial benefits in prompting treatment to reduce progression and mitigate disability.
Compared with other regions, South Asia has the most cases of visual impairment due to cataracts and uncorrected refractive error. The combination of poverty, poor living and working environments, and limited health care access have long endangered eye health in Bangladesh. Coastal Bangladesh is particularly impacted by eye disease due to economic deprivation and limited healthcare access. The coastal population mostly works in fishing and agriculture, have prolonged sunlight exposure, and inadequate occupational eye protection. This low-lying region, with 35 million people, is especially vulnerable to climate disasters and global warming. High rates of chronic disease, especially diabetes mellitus Type 2 and hypertension, coupled with limited screening and treatment, shape the area's health profile, with the increasing prevalence of eye diseases such as DR, glaucoma, and visual impairment.
To address the issues of poor health, accessibility, and affordability of eye care, Artificial Intelligence (AI) applications, such as Artificial Intelligence (AI)-assisted fundus imaging, can be applied in eye screening. Medical AI applications have the potential to improve the quality and efficiency of healthcare, reduce healthcare costs, optimize treatment plans, and bolster the development of primary healthcare. They can identify presumptive DR, hypertensive retinopathy (HR), AMD, and glaucoma by analyzing the retina and optic disc of fundus images with moderate accuracy and high efficiency, thus helping address the lack of local eye care professionals.
Data Yakka developed a human-AI collaboration that delivers affordable and transformative community-based eye screening to underserved communities in the coastal Bangladesh region of Char Fasson. The "Amar Chokh Amar Alo" (My Eyes, My Light) initiative creates and implements comprehensive eye screening that combines AI-assisted eye screening and grassroots partnerships with trusted non-health non-governmental organizations (NGOs). It has three objectives: 1) Enhancing accessibility and affordability of eye screening; 2) Supporting high quality and efficient treatment of those problems detected via screening, 3) Collecting fundus images to refine or train AI algorithms in the future. This project was designed to evaluate the feasibility, performance, equity, and cost of this model of eye screening and its implications for global eye disease.
The implementation of participant recruitment, data collection, screening, and follow-up was separated into twelve steps. This standardized framework ensured the integration of screening with data collection and follow-up eye care services. Based on risk stratification by diabetes, hypertension, age 50+ years, and/or optometrist recommendation, fundus imaging was offered selectively to higher-risk patients.
详细描述
Setting Location and Participants Selection of a target region within Bangladesh considered four factors that would facilitate successful screening: 1) Relatively high population density and adequate transportation network, 2) the presence of a Cyclone Preparedness Program (CPP) volunteer network, 3) An eye care hospital willing to serve referred clients, and 4) Support and assistance from local and regional government. Based on these criteria, the investigators selected and implemented the screening project in Char Fasson of coastal Bangladesh.
Char Fasson is a sub-district (Upazila) of the Bhola District and has a population of 518,792. While the investigators initially considered another coastal sub-district of comparable size, Char Fasson was chosen because of data collection challenges in the original choice. Char Fasson showed greater levels of poverty and more limited healthcare access while governmental coordination was exemplary. Within Char Fasson, the investigators made special efforts to serve outlying, smaller islands within the sub-district.
Partnership Team Successful delivery of the program relied on effective collaboration across a range of diverse partners, including major operational organizations, clinical affiliates, commercial suppliers, and technical advisors. The major partners responsible for day-to-day work included the organizer, Data Yakka (Palo Alto, California, USA) which is a healthcare technology company specializing in AI-assisted medical screening platforms and data management systems, the regional government (Char Fasson Upazila), the Bangladesh Disaster Preparedness Centre (BDPC, Dhaka), and the national government's Cyclone Preparedness Program (CPP, Dhaka). Among them, BDPC provided personnel for program execution and supervision while CPP mobilized its volunteers for paid work in clinical site staffing and program outreach to remote areas. The total number of CPP community volunteers in coastal Bangladesh is around 76,000, with 3,300 in Char Fasson.
The main clinical partner was the Dr. K. Zaman Bangladesh National Society for the Blind Eye Hospital (BNSB, Mymensingh, Bangladesh), which provided on-site ophthalmology and optometry staffing as well as remote fundus image review and clinical consultation as needed. The Ad-din Medical College Hospital (Dhaka) provided additional resources. Clients requiring cataract surgery were referred to regional facilities vetted by BNSB or to services provided by the Ad-Din Hospital.
Commercial supplier partners included VisionSpring (New York, New York, USA), a not-for-profit organization providing low-cost reading glasses that the program distributed free to those in need, Topcon Corporation (Tokyo, Japan) for its NW-500 robotic fundus imaging camera, and Aurolab (Madurai, India) for its HAWK I T2 slit lamp examination device. Thirona Retina (Nijmegen, Netherlands) supplied its RetCAD AI software algorithms for the detection of suspected DR, AMD, and glaucoma. The software evaluates image quality, generates heatmaps with deep learning, and provides disease likelihood scores. Previous research demonstrated that RetCAD showed high performance: 96% sensitivity and 94% specificity on the Messidor-2 dataset for DR, 95% sensitivity and 97% specificity on the private1 dataset for AMD, and 95% sensitivity and 86% specificity on the REFUGE dataset for glaucoma. Technical advisors, including Stanford University (Palo Alto, California, USA), Royal Victorian Eye and Ear Hospital (Melbourne, Australia), and Glaucoma Australia (Sydney, Australia), shared their expertise, guided technical program components, and provided external clinical oversight and case-by-case review for complex or unclear diagnosis.
研究设计
- 研究类型
- Observational
- 观察模型
- Ecologic Or Community
- 时间视角
- Prospective
入排标准
- 年龄范围
- 35 Years 至 —（Adult, Older Adult）
- 性别
- All
- 接受健康志愿者
- 是
入选标准
- •Age 35+ years
- •Residing in the Sub-District of Char Fasson in the Bhola District
排除标准
- •Known eye disease diagnosis
结局指标
主要结局
Number of individuals screened
时间窗: 12 months
The feasibility goal was to demonstrate that high-quality free services could be provided to the population through the eye screening model's work-flow, computer platform, and use of AI-assisted interpretation of fundus images.
次要结局
- Eye Disease Prevalence(12 months)