AI telemedicine screening in ophthalmology: health economic considerations

Abstract

Digital technology is revolutionising health care and this transformation is evident by the increase in telemedicine, the internet of medical things, and artificial intelligence (AI) health diagnostics.1Ting DSW Carin L Dzau V Wong TY Digital technology and COVID-19.Nat Med. 2020; 26: 459-461Crossref PubMed Scopus (710) Google Scholar Within ophthalmology, deep-learning models have shown excellent diagnostic performance in the screening of eye diseases, such as diabetic retinopathy,2Ting DSW Cheung CY Lim G et al.Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes.JAMA. 2017; 318: 2211-2223Crossref PubMed Scopus (999) Google Scholar, 3Gulshan V Peng L Coram M et al.Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs.JAMA. 2016; 316: 2402-2410Crossref PubMed Scopus (3460) Google Scholar glaucoma,4Hemelings R Elen B Barbosa-Breda J Blaschko MB De Boever P Stalmans I Deep learning on fundus images detects glaucoma beyond the optic disc.Sci Rep. 2021; 1120313Crossref PubMed Scopus (15) Google Scholar and cataract.5Tham Y Goh JHL Anees A et al.Detecting visually significant cataract using retinal photograph-based deep learning.Nat Aging. 2022; 2: 264-271Crossref Scopus (1) Google Scholar Nonetheless, the widespread clinical implementation of these AI technologies is still hampered by barriers such as ethical and accountability concerns, administrative difficulties, and the scarcity of health economic analyses to support policy-level implementation. With the advent of digital health technologies, there are questions about whether these interventions are truly more efficient than face-to-face screening. Health economic outcomes of these new technologies can be assessed by cost-effectiveness, cost-utility analysis, cost-minimisation analysis, and cost–benefit analysis.6Kwee A Teo ZL Ting DSW Digital health in medicine: important considerations in evaluating health economic analysis.Lancet Reg Health West Pac. 2022; 23100476Google Scholar In addition, important considerations specific to the use of digital health technologies in screening include deployment mode (ie, tier of human graders and whether AI technology is fully autonomous or semi-autonomous), human and technology grading fees, and digital infrastructure considerations.6Kwee A Teo ZL Ting DSW Digital health in medicine: important considerations in evaluating health economic analysis.Lancet Reg Health West Pac. 2022; 23100476Google Scholar In The Lancet Global Health, Hanruo Liu and colleagues7Liu H Li R Zhang Y et al.Economic evaluation of combined population-based screening for multiple blindness-causing eye diseases in China: a cost-effectiveness analysis.Lancet Glob Health. 2023; (published online Jan 23.)https://doi.org/10.1016/S2214-109X(22)00554-XGoogle Scholar evaluate the cost-effectiveness and cost-utility of screening methods for the simultaneous screening of five major eye diseases in China. Three screening methods—non-telemedicine (face-to-face) screening, telemedicine screening without AI, and AI telemedicine screening—were compared with one another and were further analysed in rural and urban settings. Non-telemedicine screening consisted of face-to-face examinations by ophthalmologists. Non-AI telemedicine screening consisted of retinal and anterior segment photography, autorefraction, intraocular pressure, and medical history collected by primary care staff and analysed by graders. In AI telemedicine screening, retinal fundus photographs were collected by primary care staff and further analysed by AI technology. The five ocular diseases chosen for screening were age-related macular degeneration, glaucoma, diabetic retinopathy, cataract, and pathological myopia, which are the leading causes of visual impairment in China. A decision-analytic Markov model of adults aged 50 years or older for a total of 30 1-year cycles was developed for each setting. Incremental cost-utility ratios (ICURs) using quality-adjusted life-years, and incremental cost-effectiveness ratios (ICERs) in terms of cost per year of blindness avoided, were calculated for the three screening methods in rural and urban settings. Both direct and indirect costs were included, where direct costs consisted of costs incurred for screening, hospital definitive ophthalmic reviews, interventions, food, transportation, and accommodation by patients and accompanying family members. Indirect costs included the monetary value of work time loss of the patient and one accompanying family member. Liu and colleagues7Liu H Li R Zhang Y et al.Economic evaluation of combined population-based screening for multiple blindness-causing eye diseases in China: a cost-effectiveness analysis.Lancet Glob Health. 2023; (published online Jan 23.)https://doi.org/10.1016/S2214-109X(22)00554-XGoogle Scholar showed that any form of screening provided significant ICURs and ICERs over no screening, and this was more prominent in rural settings. In rural settings, when compared with no screening, the ICUR for non-telemedicine screening was US$2494 and for non-AI telemedicine screening was $2326. Compared with no screening, AI telemedicine screening was noted to be dominating, although no value was stated. The ICER for non-telemedicine screening was $12 487 and for non-AI telemedicine screening was $11 766; AI telemedicine screening again dominated no screening. In urban settings, when compared with no screening, the ICUR for non-telemedicine screening was $624, for non-AI telemedicine screening was $581, and for AI telemedicine screening was $244. The ICER for non-telemedicine screening was $7251, for non-AI telemedicine screening was $6920, and for AI telemedicine screening was $2567. Cost-effectiveness acceptability curves showed that AI telemedicine screening was the dominant strategy in 90% of rural areas and 67% of urban areas, based on willingness to pay thresholds in each area. The authors further evaluated screening intervals from 1 to 5 years and found that annual AI screening was the optimal screening interval for both rural and urban settings. The study has several strengths. First, the cost-effectiveness analyses and utility analyses were compared in both rural and urban settings, which often have differences in costs, willingness to pay thresholds, as well as prevalence of disease, which are significant factors in the analyses. Second, in contrast to single-disease telemedicine screening, the piggybacking of multiple eye diseases during telemedicine screening is more like existing ophthalmological clinical examination where incidental findings might be picked up during evaluation for a specific condition. The majority of existing AI screening studies are diagnostic accuracy studies aimed at evaluating the ability of the AI model to pick up a single disease. The evaluation of multidisease screening models and health economic analyses on these models is a step towards much needed clinical translation. Although the demonstration of cost utility and effectiveness of telemedicine screening (with or without AI) in the five ocular conditions aids in the justification for implementation, there are still key challenges that need to be addressed before widespread clinical adoption is possible. First, the conclusions of health economic analyses for the same intervention can vary between populations due to varying disease prevalence, screening costs, and treatment strategies. For example, the screening of pathological myopia in this study might be deemed cost effective given the high myopia prevalence in people from east Asia, but might not be so in White populations for whom myopia prevalence is substantially lower.8Rudnicka AR Kapetanakis VV Wathern AK et al.Global variations and time trends in the prevalence of childhood myopia, a systematic review and quantitative meta-analysis: implications for aetiology and early prevention.Br J Ophthalmol. 2016; 100: 882-890Crossref PubMed Scopus (276) Google Scholar This limits the generalisability of the study findings beyond the studied Chinese population. Second, poor image quality and false positives during telemedicine screening might lead to unnecessary tertiary referrals, resulting in a substantial rise in health-care costs. Referral thresholds must thus be carefully designed for the specific disease and the target population. Third, despite cost-effective telemedicine screening and high diagnostic accuracy of AI screening models, low compliance to subsequent referrals for interventions will not achieve the intended health economic benefits of screening. In this study, the authors noted that compliance to referrals could be as low as 57% in urban settings and 19% in rural settings. Last, for a new telemedicine screening intervention to be effective, defining the intended use environment is crucial and should be customised to each country and tailored to each city or town when possible. Within a large country such as China, there is a wide range of resources across different states and a current absence of a population-based national screening programme for eye diseases.9Li R Yang Z Zhang Y et al.Cost-effectiveness and cost-utility of traditional and telemedicine combined population-based age-related macular degeneration and diabetic retinopathy screening in rural and urban China.Lancet Reg Health West Pac. 2022; 23100435Google Scholar In urban areas, ophthalmology screening performed in tertiary centres is opportunistic and scarce, as shown by data suggesting that more than half of people with known diabetes have never had an eye examination.10Xiao B Mercer GD Jin L et al.Outreach screening to address demographic and economic barriers to diabetic retinopathy care in rural China.PLoS One. 2022; 17e0266380Crossref Scopus (0) Google Scholar The scarcity of screening is even more apparent in rural regions where two-thirds of patients with diabetes have never had an eye examination.10Xiao B Mercer GD Jin L et al.Outreach screening to address demographic and economic barriers to diabetic retinopathy care in rural China.PLoS One. 2022; 17e0266380Crossref Scopus (0) Google Scholar In rural areas, health care is delivered via village health workers and town health centres run by general practitioners, and willingness to pay thresholds and willingness to travel to tertiary care centres are vastly different from urban areas. Screening for ophthalmological diseases requiring tertiary centre care is often dependent on ad-hoc programmes led by non-governmental organisations, such as the diabetic retinopathy screening and cataract surgery programme by Lifeline Express.11Wong IYH Ni MY Wong IOL Fong N Leung GM Saving sight in China and beyond: the Lifeline Express model.BMJ Glob Health. 2018; 3e000766Crossref Scopus (10) Google Scholar Eye screening in a large country such as China thus has substantial challenges and will require adjustments for different environments on the basis of resource settings and population profile. By contrast, the well established national diabetic retinopathy screening programme provides screening for all UK citizens with diabetes and is covered by the National Health Service. The implementation of this national programme has contributed to diabetic eye disease no longer being the leading cause of certifiable blindness in the UK.12Scanlon PH The English National screening programme for diabetic retinopathy 2003–2016.Acta Diabetol. 2017; 54: 515-525Crossref PubMed Scopus (219) Google Scholar In the USA, eye screening is primarily provided by eye care providers, many of whom practise outside of major health systems, limiting communication with primary care providers and resulting in a different barrier to diabetic eye screening.13Liu Y Swearingen R Diabetic eye screening: knowledge and perspectives from providers and patients.Curr Diab Rep. 2017; 17: 94Crossref PubMed Scopus (20) Google Scholar This differs from Singapore, a country with a small population and geographical area, where a nationwide Singapore National Diabetic Retinopathy Screening Programme in primary care settings is recorded on the national electronic health records systems and referrals to tertiary centres are relatively more accessible.3Gulshan V Peng L Coram M et al.Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs.JAMA. 2016; 316: 2402-2410Crossref PubMed Scopus (3460) Google Scholar, 14Nguyen HV Tan GS Tapp RJ et al.Cost-effectiveness of a national telemedicine diabetic retinopathy screening program in Singapore.Ophthalmology. 2016; 123: 2571-2580Summary Full Text Full Text PDF PubMed Scopus (103) Google Scholar Hence, tailoring screening programmes to the intended use environment is essential. With the intention to reap the benefits that digital technology has to offer, telemedicine screening must be carefully designed and ultimately tailored to the targeted population. In the evaluation of new digital technologies, we recommend the following approach (appendix). First, identify the disease prevalence and define the intended use environment (such as in this study, resource rich vs resource constrained settings). Second, select the relevant health economic model, of which multiple models might be relevant. Third, evaluate the diagnostic performance of the new technology and decide on deployment modes and screening thresholds. Last, evaluate costs. Concomitant efforts to increase post-screening referral and treatment compliance via health education and the reduction of financial barriers are essential. Despite the challenges, Liu and colleagues'7Liu H Li R Zhang Y et al.Economic evaluation of combined population-based screening for multiple blindness-causing eye diseases in China: a cost-effectiveness analysis.Lancet Glob Health. 2023; (published online Jan 23.)https://doi.org/10.1016/S2214-109X(22)00554-XGoogle Scholar findings suggest that digital technology and AI can increase the cost-effectiveness and utility of simultaneous screening of multiple ocular disease in both rural and urban settings in China. This affirms the value of digital technology and AI in our quest to achieve a more efficient and equitable health-care ecosystem. DSWT holds a patent on a deep learning system for detection of retinal diseases, and is a cofounder of EyRIS Singapore for which he was a scientific adviser until more than 3 years ago. ZLT declares no competing interests. Download .pdf (1.51 MB) Help with pdf files Supplementary appendix Economic evaluation of combined population-based screening for multiple blindness-causing eye diseases in China: a cost-effectiveness analysisCombined screening of multiple eye diseases is cost-effective in both rural and urban China. AI coupled with teleophthalmology presents an opportunity to promote equity in eye health. Full-Text PDF Open Access