Introduction
As the LMIC with the largest population in the world, China has 8·69 million citizens with blindness. Population ageing and growth, as well as urbanisation, have shifted the spectrum of blindness-causing eye diseases to non-communicable conditions, and cataracts, glaucoma, age-related macular degeneration, diabetic retinopathy, and pathological myopia have become the leading blindness-causing eye diseases in China.
,
Encouragingly, most vision impairment is avoidable and treatable. Identifying asymptomatic patients and referring them for treatment in a timely fashion could improve prognosis and reduce the disease burden.
However, a routine screening programme is not available in China, and the implementation and maintenance of a large-scale, population-based programme is made difficult by time, transportation, financial, and workforce constraints.
Especially in LMICs, where ophthalmic services are insufficient due to fewer trained vision care experts, many patients cannot receive early diagnoses and effective treatment.
With image examination as the main assisted diagnosis method, telemedicine offers ophthalmology great advantages by helping to resolve unbalanced medical resource distribution and reducing the burden of travel on patients. Regular teleophthalmology-based screening for diabetic retinopathy using retinal fundus photography in place of hospital examination is dependable and cost-effective, potentially increasing screening accessibility, especially for rural and underserved communities.
,
The rapid development of artificial intelligence (AI) has brought new breakthroughs in detecting various ophthalmic diseases, showing a high sensitivity and specificity,
which substantially increases service coverage. There have been only a handful of economic studies on AI in ophthalmology, although the diagnosis of eye disease has been the leading field in which AI has been applied. Several studies have analysed the economic and clinical benefits of AI-based screening for diabetic retinopathy in the UK and Singapore.
,
However, the result of an economic assessment could vary substantially across disease profiles and health-care systems.
Thus, substantially more economic data are required across economic levels worldwide, especially in LMICs. Moreover, in recent studies, using AI to screen for multiple eye diseases showed good sensitivity and specificity,
,
and it could be more clinically beneficial and economically valuable for clinical application in the real world, especially for primary care and community settings where the AI model would most probably be used. However, sufficient evidence and cost-effectiveness data are needed to support the decisions of policy makers regarding costs and budget allocations. We aimed to fill these evidence gaps and build a comprehensive Markov model to investigate the cost-effectiveness of combined screening for leading eye diseases in China.
Evidence before this study
We searched PubMed, Ovid MEDLINE, Scopus, ProQuest, Google, Web of Science, and the Centre for Studies for publications in English from database inception to Nov 15, 2021, using “OR/AND” operators with the following keywords and the Medical Subject Headings terms: “eye diseases”, “ophthalmology”, “diabetic retinopathy”, “age-related macular degeneration”, “glaucoma”, “cataract”, “pathologic myopia”, “screening”, “health care”, “machine learning”, “artificial intelligence”, “deep learning”, “economic evaluation”, “cost-effectiveness analysis”, “cost-utility analysis”, “cost-benefit analysis”, “cost-minimization analysis”, “China”, and “Chinese”. We also searched the references listed in the identified papers. We found no cost-effectiveness analysis on the combined screening of age-related macular degeneration, glaucoma, diabetic retinopathy, cataracts, and pathological myopia screening in China. Previous studies have focused on cost-effectiveness analyses of screening for single blindness-causing eye diseases. Regular screening programmes for diabetic retinopathy have been implemented in several communities in low-income and middle-income countries (LMICs) and in high-income countries and have been shown to be reliable and cost-effective. However, the cost-effectiveness of screening for age-related macular degeneration or glaucoma remains undetermined. Cataract surgery has been shown to be cost-effective, but few studies have explored the cost-effectiveness of population-based screening for cataract or screening for pathological myopia. Studies on the economics of screening based on artificial intelligence (AI) in ophthalmology are even fewer and scarcer, with only a few publications on the economic evaluation of AI in screening for diabetic retinopathy in high-income countries. Additionally, those studies only considered costs and outcomes over a 1-year period rather than the whole life course. To our knowledge, there are no studies analysing the cost-effectiveness of AI-based screening of multiple eye diseases.
Added value of this study
Our study showed that a population screening programme for multiple blindness-causing eye diseases was likely to be highly cost-effective in both rural and urban China. All three screening strategies (non-telemedicine screening, non-AI telemedicine screening, and AI telemedicine screening) met the criterion of a highly cost-effective health intervention. In addition, our study showed that annual AI screening in both rural and urban areas in China was the optimal screening strategy.
Implications of all the available evidence
Our results suggest that routine screening for multiple blindness-causing eye diseases could be highly cost-effective in China, providing robust economic evidence for informed policy making regarding its large-scale promotion. Moreover, our findings indicate that novel screening strategies, such as AI-based screening, could achieve greater cost-effectiveness in population screening. These novel screening approaches could play a unique role in high-quality eye-care delivery and improve the equity and accessibility of eye health resources. More importantly, they can serve as a feasible example for other countries, especially LMICs with similar settings or epidemiological features.
Discussion
The results of our study suggest that combined screening of multiple blindness-causing diseases is likely to be highly cost-effective and fulfil WHO’s criteria for population screening in China. Among all investigated screening intervals and strategies, annual AI screening was the most cost-effective strategy under the current willingness-to-pay threshold in China.
Moreover, there is increasing evidence showing the potential to deliver more cost-effective and accessible screening for diabetic retinopathy with teleophthalmology compared with non-telemedicine methods.
Screening for age-related macular degeneration has been shown to be cost-effective and reduce the number of patients with blindness,
but in another study in Japan, a different conclusion was reached as the ICER exceeded the willingness-to-pay threshold.
A recent study found that combined screening for age-related macular degeneration and diabetic retinopathy is very cost-effective in China.
Population-based screening for glaucoma was not considered to be cost-effective in high-income countries,
but an economic evaluation by Tang and colleagues
showed that screening for glaucoma is likely to be cost-effective in China due to the high disease incidence and low screening costs. These findings emphasise the importance of taking into consideration the various economic levels, demographic statuses, epidemiology, and intervention costs while implementing screening programmes across countries and diseases.
Additionally, with an ageing global population and associated vision loss burden, urgent action is needed to deliver comprehensive services and address the full range of eye conditions on a large scale. The Chinese Government has made efforts to continuously improve health management systems, introducing specific screening programmes, such as the China DR Screening and Prevention Project and One Million Poor Cataract Patients Restoring Vision. However, like most countries, China does not have combined screening for multiple blindness-causing eye diseases, and the diagnosis of related diseases mainly depends on opportunistic diagnosis. However, there are some problems with opportunistic diagnoses, such as low visit rates, low diagnosis rates, and long waiting times for secondary health-care appointments. Considering that comprehensive screening programmes for multiple eye diseases might become the major trend in the future, we believe that a related economic assessment has a crucial role for all stakeholders in health care. However, few economic studies have explored combined screening for multiple eye diseases. The results of our study help to fill this evidence gap by reviewing the complete literature on epidemiology and screening programmes, indicating that population-level multiple screening for eye diseases is likely to be cost-effective in China.
,
,
,
A study in Singapore showed that fully automated screening for diabetic retinopathy had a cost saving of 14·3{a5ceed037b574a4d8c6b44a0a7290437cee40655417128da3b56d864fe64414f} compared with the human grading strategy from the health system perspective.
It is essential to evaluate the difference in these costs among countries because they might not be replicated in LMICs, where labour costs are lower. A Scottish study showed that automated screening for diabetic retinopathy led to £201 600 savings per year.
In another UK study, two semi-AI models achieved sufficient specificity to be cost-effective compared with non-AI grading.
However, in those studies, costs and consequences were considered over only 1 year and not over a lifetime. Most chronic eye diseases are lifelong diseases; thus, analysis using a Markov model has more advantages and offers clearer evidence on the costs and health rewards of AI applications and interventions over their lifetime.
In this study, we built a thorough and dynamic Markov model that considers transitions across disease stages and lifetime costs instead of a static decision-analytic model.
Indeed, unless more effort is made in terms of equity, disparities between and within countries will probably widen and lead to a feedback loop, as socially advantaged people are more able to use innovative medical care services. As a result, we believe that the implementation of a broad-based screening programme requires stronger leadership and a substantial financial commitment from the government, especially for rural and resource-limited areas.
AI-based teleophthalmology has the potential to drive not only delivery enhancements but also notable cost savings and improvements to the health system. However, the implementation of AI in China might pose concerns and challenges. First, one of the crucial factors in successful implementation is the involvement and collaboration of various stakeholders. Although eye doctors are the core of eye health services, the most engaged group comprises optometrists, general practitioners, and nurses. Therefore, training and management of these practitioners are essential for the success of teleophthalmology services. Second, digital literacy and previous experience working with computers could affect the acceptance of technology-based solutions and thus the satisfaction of users or patients in older populations, especially in rural areas. Another key factor in the wide adoption of AI is its economic viability. Currently, in China, the most common purchase plans for using AI services include one-time purchases, capital leasing plans, and non-capital leasing plans with a minimum number of yearly examinations for research. From the provider perspective, a deficit could still exist, notwithstanding the purchase plan, and a fixed charge per patient and reimbursement would be needed for a break-even marginal revenue. Further benefits will inevitably come from piggybacking multidisease eye screening onto outreach screening programmes for non-communicable diseases (eg, diabetes and hypertension), which will enhance the value of data use for the health-care system and might be the most cost-effective strategy. Finally, policy makers and health sector specifications must consider processes for the continuous updating of AI policies, guidelines, regulations, and health systems to adapt flexibly and efficiently to AI innovations and manage monitoring and risk management practices. Despite the great interest in AI-based teleophthalmology, we advocate following Confucian doctrine—that is, to learn extensively, inquire thoroughly, ponder prudently, discriminate clearly, and practise devotedly—to ensure that affordable, high-quality, equitable, people-centred eye care can be delivered.
Our study has several limitations. First, our analysis was not based on prospective data, and we used some parameters from non-Chinese studies or expert advice. In reality, individuals who frequent primary care services are likely to be patients with chronic systemic disease, such as hypertension and diabetes, which might result in a cohort with greater prevalence of diabetes retinopathy or cataract, and this might alter the economic analysis. However, the model was shown to be robust and insensitive to a wide fluctuation of parameters by extensive sensitivity analyses. Second, because the data incorporated were all Chinese based, it is not clear how relevant our economic results are to other countries with different health settings and medical costs. Third, only five leading common blindness-causing diseases were included in this study. Moreover, given the complexity of the model, we considered one of the most common situations in clinical practice, that patients might have cataracts combined with one of the other four diseases, but we assumed that these four diseases would not all present simultaneously in one person. Indeed, other eye diseases could also be screened for under the proposed setting, and such a screening could even be useful for identifying individuals at risk of cardiovascular disease or cognitive impairment. However, these diseases were not considered in this analysis. Screening for more conditions might not show a linear relationship with cost-effectiveness, so the average number of conditions picked up during a single screening and which diseases should be screened for in order to be most cost-effective could be evaluated in future studies. In addition, the effect on cost-effectiveness of sex, age group, cardiovascular risk factors, and economic level and medical resources in different regions should be considered. Fourth, the costs of AI-based telehealth care have not yet been standardised or determined clearly. Furthermore, the increasing amounts of patient data might cause differences in maintenance costs, and an extra work burden might also be generated by false positives and poor-quality images, which are potential issues in the long-term use of AI.
In conclusion, our study shows that combined screening for multiple blindness-causing diseases is likely to be cost-effective in China. We believe that AI coupled with telemedicine and mobile health would produce social and economic benefits in LMICs by expanding the coverage of medical resources and improving the quality, sustainability, and equity of eye health services.
HLiu, NW, and JTa conceived and designed the study. RL, KZ, YZ, MY, JTi, HLi, SF, XC, YL, and DM acquired or interpreted the data. HLiu drafted the manuscript. NW and JTa critically revised the manuscript for important intellectual content. RL and KZ did the statistical analysis. HLiu obtained funding. HLiu, NW, and JTa supervised the study and accessed and verified the data. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.