FormalPara Key Points for Decision Makers

From a policy maker’s perspective, a diabetic retinopathy (DR) screening programme should be incorporated for the diabetic population in India. Among the different models of screening, annual tele-supported screening of the diabetic population with a minimum of 5 years’ duration of disease is the most efficient strategy.

The choice of strategy should be guided by the level of glycaemic control among patients with diabetes at the population level.

From a primary care providers’ perspective, our study underscores the importance of counselling to ensure adequate glycaemic level in patients with diabetes, failure of which would make the DR screening programme less efficient.

The study calls for more technological advancement in diagnostic accuracy of AI-based technologies, since an increase in sensitivity of such devices to at least 96.5% would make AI-supported screening a more cost-effective strategy compared with the other modes of screening.

1 Introduction

Demographic and epidemiological transitions have resulted in a significant burden of non-communicable diseases including ophthalmic disorders, as populations experience longer life expectancies and a shift toward chronic health conditions at the global level. The leading causes of blindness in the global context are cataract, glaucoma, uncorrected refractive errors, age-related macular degeneration, and diabetic retinopathy (DR) [1]. Although there has been a decline in prevalence of blindness attributed to cataract, glaucoma and uncorrected refractive errors, the case load for DR has intensified in the last three decades in India [2]. In India, DR has emerged as a significant public health issue. Once the 20th leading cause of blindness in the 1970s, it has surged to become the sixth leading cause since 2015 [3].

The Government of India launched the Ayushman Bharat programme in 2018 with one of the stated goals to ensure the provision of essential primary eye care services including annual screening for DR through use of non-mydriatic fundus cameras at the upgraded primary health centres (PHCs) [4]. However, there can be different implementation models for provision of systematic screening services, and advancing technology in the fields of ophthalmology, telemedicine and artificial intelligence (AI) present newer avenues for improvement in coverage, accessibility, and quality of care.

Cost-effectiveness analysis (CEA) is a significant tool to guide policymakers in formulating evidence-based decisions on the introduction and implementation of efficient health interventions or programmes [5]. The extant literature on the cost effectiveness of DR screening suggests the intervention to be cost effective in high-income countries (Table S14, see electronic supplementary material [ESM]). A few researchers from Africa and Asian countries such as Singapore, Thailand, and China have also contributed to economic evaluations related to DR (Table S14, see ESM). However, there is a dearth of robust literature that compares the alternative models for roll-out of population-based DR screening in the low- and middle-income country (LMIC) setting to inform the most efficient ways to scale-up DR screening strategies. Further, despite the proliferating AI-based devices with optimal diagnostic accuracy, there have been limited attempts at their economic assessment, with no study conducted in LMICs, which are home to 80% of the world diabetic population [6].

The current study was conducted with an objective to assess the cost effectiveness of different models of population-based screening strategies for DR at PHCs (face-to-face screening by optometrists versus teleophthalmology-supported screening versus AI-supported screening) compared with usual care. The study also aims to assess the impact of screening strategies in terms of reduction in incident cases of vision-threatening DR (VTDR) and blindness and gain in quality-adjusted life years (QALYs). Conventionally, a targeted approach based on individual risk factors has been recommended to optimize resource allocation, directing resources primarily toward patients considered to be at higher risk of developing DR [7, 8]. Therefore, our study also assessed population-based screening strategies that incorporate risk stratification according to the duration of diabetes and the status of glycaemic control.

2 Methods

A mathematical model, comprising a decision tree and Markov model, was developed in MS Excel to compare three non-mydriatic fundus camera-based annual screening strategies for DR (face-to-face screening, teleophthalmology-supported screening and AI-supported screening), and a usual-care scenario.

2.1 Details of Comparative Strategies

The potential comparative screening strategies were identified through targeted literature review (Table S14, see ESM), followed by expert consultations with policy makers and ophthalmologists to outline context-specific interventions in terms of mode of screening, and target populations. We considered three screening modalities—face-to-face screening, tele-ophthalmology supported screening, and AI-supported screening, to evaluate the costs and health benefits of each method, as well as to compare them with the usual-care scenario. The usual-care scenario was equivalent to ‘no population-based screening at PHCs’, which implied an opportunistic diagnosis for diabetic patients visiting an ophthalmologist.

It was assumed that the face-to-face screening/AI-supported screening at PHCs would be conducted by the optometrist, and supported by the medical officers on fixed days, which would be preceded by social mobilization for screening by community health workers to ensure a high level of participation (Appendix, p7; ESM). Face-to-face screenings would utilize a non-mydriatic fundus camera, while AI-supported screenings would employ the same camera, augmented with deep learning algorithms to assist optometrists in decision making. In contrast, the tele-ophthalmology-supported screening would involve transferring fundus images through a tele-platform, where they would be graded by an ophthalmologist at a central hub to provide a provisional diagnosis.

Furthermore, in terms of target populations, we assessed the three modalities of screening in ‘all of the diabetic population’, ‘the diabetic population with duration of disease of ≥5 years’, and ‘the diabetic population with duration of disease of ≥10 years’. Additionally, we conducted a subgroup analysis to assess the cost effectiveness of each strategy when the screening was limited to diabetic individuals with either controlled or uncontrolled glycaemic status.

2.2 Model Characteristics

The model began with a hypothetical cohort of 1000 patients with diabetes aged 45 years old considering the mean age of type 2 diabetes [9]. The cycle length of the model was assumed to be 1 year. Lifetime consequences were valued in terms of gain in number of life years (LY), QALYs, and the years of blindness averted per 1000 diabetic patients. We took an abridged societal perspective, that is, we considered direct medical (consultation, diagnostics, medicines, laser/surgical procedures, and bed charges) and non-medical costs (transport, boarding and lodging, and food-related out-of-pocket expenditure) in the evaluation and discounted the future costs and consequences at an annual rate of 3% in accordance with the Indian methodological guidelines [10]. We did not include indirect costs related to productivity losses in the analysis, in accordance with the Indian reference case for conducting economic evaluations [10]. The incremental cost-effectiveness ratio (ICER) was computed for the alternative screening strategies. Based on the Indian guidelines for conducting economic evaluations, the screening strategy was considered cost effective if its ICER in terms of incremental costs per QALY gained was below the gross domestic product (GDP) per capita of 2021–22 (₹171,498; US$2182) [10, 11]. In order to gauge the effectiveness of population-based screening at the regional level, and in states in different epidemiological transition levels [12], we relied on state-level population projections from the census and the state-specific prevalence rates of diabetes obtained from the ICMR-INDIAB-17 study [13, 14].

2.3 Model Overview

A decision tree was developed, assuming that a population cohort of 1000 patients would be screened using one of the systematic strategies or usual care (Fig. S1, see ESM). Based on the diagnostic accuracy of the strategies and care-seeking behaviour, the DR patients were either ‘diagnosed’ or ‘not diagnosed’, and ‘treated’ or ‘not treated’ in each cycle (Fig. S1). The screening coverage in the three arms was assumed to be 75% based on studies in Indian settings [15]. It was considered that the population suspected of moderate or severe non-proliferative DR (NPDR), proliferative DR (PDR), and any stage of DR with macular edema or with ungradable images would be referred to higher centres for confirmatory diagnosis, based on Indian standard treatment guidelines [16]. However, there would be a loss to follow-up of 44% to confirmatory diagnosis at higher centres (Table S3, see ESM) [17]. Among those with a confirmed diagnosis, a 27.5% loss to follow-up for initiation of treatment was considered in all the scenarios [18]. The usual-care scenario was equivalent to ‘no population-based screening at PHCs’, which implied an opportunistic diagnosis in 10% of diabetic patients visiting an ophthalmologist (Table S3).

This was followed by the development of a Markov model which simulated the natural history of the disease (Fig. S2, see ESM). We considered the classification of DR recommended by the International Council of Ophthalmology, which is the most commonly used classification in LMICs [15]. The cohort comprised a population with controlled (23.4%) and uncontrolled (76.6%) glycaemic status [19]. The progression rate to blindness varied according to the status of treatment. The ‘death’ state was considered an absorbing state for all the other Markov states.

2.4 Model Parameters

The annual incidence and progression of DR was calculated from the Sankara Nethralaya Diabetic Retinopathy Epidemiology and Molecular Genetic Study (SN-DREAMS), (Table S1, see ESM) [20]. The incidence was varied according to the duration of diabetes and glycaemic status using relative risk ratios from the United Kingdom Prospective Disease Study (UKPDS) [20, 21]. Age-specific mortality was calculated from the Sample Registration System (SRS) life tables [22]. Mortality multipliers of 2.34 and 1.89 were used for blindness and diabetes, respectively, based on hazard ratios from published literature [23, 24].

The sensitivity and specificity of each DR screening strategy in the diabetic population was based on published studies (Table S2, see ESM) [25,26,27]. It was considered that 22% of the retinal images would be ungradable, based on a national-level study [28]. The individuals with ungradable images would be referred to higher centres for DR evaluation. Furthermore, it was assumed that diagnosis at higher centres would result in diagnosis of the true health state.

2.5 Cost of Screening and Treatment

The cost of screening and treatment included the health system costs and the direct non-medical costs incurred by the patients. We assumed that the screening services would be integrated with the services currently being provided by the existing PHCs, based on expert consultations with policy makers and ophthalmologists. We calculated screening costs for the scenarios using the data from studies included as part of the National Health System Cost Database of India (Appendix, p7–9; ESM) [29]. The mean cost per scan in the case of AI-supported screening was acquired through three rounds of consultation with three major suppliers of this technology in India from a payer perspective.

The cost of screening was calculated at different rates of screening, by altering the variable costs, considering the mean population served by a PHC and the prevalence of diabetes in India (Table S6, Fig. S3, see ESM) [30]. The direct non-medical costs, inclusive of cost of transportation and food, which would be incurred by the population for screening were adapted from a South Indian study [31].

The considered treatments included counselling to maintain glycaemic control in NPDR patients, and three sittings of pan-retinal photocoagulation (PRP) in PDR patients. Additionally, it was considered that 45% and 15% of PDR patients would require supplemental PRP and vitrectomy, respectively [32]. Eighty percent of patients with NPDR with macular edema were assumed to be treated with anti-vascular endothelial growth factor (anti-VEGF) injections, while 20% would receive focal laser treatment, to simulate the current practice in India [33]. Patients who had PDR with macular edema were presumed to be treated with PRP and anti-VEGF injections in 90% of cases while 10% of such patients with high-risk PDR would be treated with vitrectomy. Following the initial treatment, 6% and 10% of patients with PDR with macular edema and NPDR with macular edema would require additional vitrectomy, respectively.

Treatment costs for the procedures were obtained from standard sources (Table S5, see ESM) [29, 34, 35]. The annual follow-up care for patients with PDR and macular edema included the cost of outpatient consultation and diagnostic services. The non-medical costs for treatment and follow-up, inclusive of out-of-pocket expenditure on transportation, food and accommodation, were estimated through interviews with 300 patients with DR who visited the ophthalmic outpatient department in an Indian tertiary care hospital. All costs are reported in US dollars, using the 2022 conversion rate (1 USD = 78.60 INR), as specified by the World Bank [36].

2.6 Health State Utility Values

A facility-based cross-sectional study was conducted to measure the quality of life of DR patients. The study involved recruitment of 300 patients with DR who visited the ophthalmic outpatient department in a tertiary care hospital in northern India. Face-to-face interviews were conducted with the participants of the study using the EuroQoL-5-dimensions-5-levels (EQ-5D-5L) tool. The mean utility scores along with standard errors were calculated for patients in different health states (NPDR, PDR, any stage with macular edema) using the EQ-5D-5L Indian value set (Table S4, see ESM) [37].

2.7 Sensitivity Analysis

To test the impact of uncertainty in the parameter values, we undertook multivariate and univariate probabilistic sensitivity analysis (PSA), using relevant distributions (Tables S1–S5, see ESM). Wherever the standard error was not available from the published literature, the epidemiological, diagnostic accuracy parameters and the compliance rates were varied by 10%. Due to perceived significant regional variations in costs, we allowed these parameters to vary within a large floating range of 50%. Subsequently, 999 Monte Carlo simulations were run, from which a median ICER along with 2.5th and 97.5th percentile values were analysed.

2.8 Dominance Analysis

Extended dominance analysis was conducted, comparing each arm against the next best alternative to evaluate the comparative cost effectiveness among different scenarios [38]. The comparative cost effectiveness was assessed in terms of incremental cost per QALY gained. The net monetary benefits of the interventions were calculated by analysing the difference in monetized QALYs (1 QALY equivalent to US$2182) and costs, attributed to the screening strategies versus the usual-care scenario.

3 Results

We found that all the annual population-based screening strategies were cost effective, when compared with the usual-care scenario (Table S8, see ESM). Table 1 presents discounted costs and outcomes, as well as ICERs based on extended dominance analysis. After excluding the strategies that were strongly dominated (higher costs with lesser QALYs compared with other strategies), and extendedly dominated (Table 1), tele-supported screening in the population with diabetes duration of ≥5 years was the most cost-effective strategy with maximum QALY gains within the willingness-to-pay threshold (Table 1). It resulted in a highest net monetary gain of ₹38,202 (US$486) per diabetic patient when compared with the usual-care scenario (Table S9, see ESM). The tele-supported screening in all the diabetic population yielded maximum QALYs (12.091) with a median ICER of ₹213,721 (US$2719), which was above the national GDP per capita (Table 1).

Table 1 Discounted life years, QALY, cost per person and ICERs for different annual DR screening strategies and usual care
Full size table

Our analysis also concluded that the implementation of annual DR screening at PHCs, supported by tele-ophthalmology, has the potential to avert 0.105–0.145 years of blindness per diabetic patient, depending on the specific target population groups for screening (Table S13, see ESM).

We found that the systematic tele-supported screening strategy in the population with diabetes duration of ≥5 years yielded a gain in QALYs with additional costs in comparison with the previous best strategy (Fig. S4, see ESM). The probability of annual DR screening being a cost-effective strategy was 97% for tele-supported screening in the population with diabetes duration of ≥5 years in comparison with the previous best strategy (AI-supported screening in the population with diabetes duration of ≥10 years) (Fig. S5, see ESM). On the other hand, tele-supported screening in the all-diabetes population had a 26% probability of being cost effective below the cost-effectiveness threshold value when compared with tele-supported screening limited to the population with diabetes duration of ≥5 years (Fig. S6, see ESM).

The sub-group analyses according to glycaemic control indicated that tele-supported screening for all of the diabetic population would be the most efficient strategy if screening is limited to the population with uncontrolled glycaemic status, with maximum QALYs (11.498) and an ICER value below the willingness-to-pay threshold (Table 2).

Table 2 Discounted QALYs, cost per person and incremental cost-effectiveness ratios in DR screening limited to the diabetic population with uncontrolled glycaemic status
Full size table

In univariate sensitivity analysis, we found that with glycaemic control of 5–78% among the population with diabetes, tele-supported screening in the population with a minimum disease duration of 5 years remains the preferred strategy (Fig. 1, Table S10, see ESM). This highlighted the robustness of the results in favour of tele-supported screening in the diabetic population with duration of disease of ≥5 years. However, at very low levels of glycaemic control at the population level (<5%), this annual tele-supported screening would have to be expanded to all the diabetic population to ensure efficiency. Contrarily, if a high degree of glycaemic control is achieved among 79–92% of diabetics, then AI-supported DR screening of patients with diabetes who have a minimum disease duration of 10 years is most efficient. These findings signify the importance of linking diabetes care with the DR screening programme.

Fig. 1
figure 1

Effect of change in proportion of population with controlled diabetes on cost effectiveness of dominant strategies. AI artificial intelligence, ICER incremental cost-effectiveness ratio

Full size image

Counselling the population with uncontrolled glycaemic level who have developed NPDR is expected to achieve glycaemic control status. We observed that if the counselling (as a treatment for the diabetic population with NPDR) results in a change of <5% from uncontrolled to controlled glycaemic levels at a population level, the screening strategy of tele-supported screening in the population with diabetes duration of ≥5 years ceases to be cost effective (Fig. 2). On the other hand, as the effect of counselling increases, an inverse relationship was observed between the ICER values and the impact of counselling. In one scenario, if counselling is effectively integrated in the programme to improve the proportion of the population with adequate glycaemic control to 70%, tele-supported screening in the whole diabetes population would become the most efficient strategy, with maximum QALYs (12.15) among all strategies and ICER below the GDP per-capita threshold.

Fig. 2
figure 2

Effect of counselling in early stages of diabetic retinopathy in terms of change in glycaemic control on ICER values of dominant strategies. AI artificial intelligence, ICER incremental cost-effectiveness ratio, WTP willingness-to-pay threshold

Full size image

Finally, we also found that if the sensitivity of AI-supported screening is improved to a minimum of 96.5%, the strategy yields highest QALYs, thus maximizing the health gains compared with other strategies (Fig. S7, see ESM).

3.1 Implications of Implementation of an Annual DR Screening Programme at the National and Sub-National Level

It was noted that there would be a reduction in the annual incidence of VTDR and blindness by 17.3% (252,068 cases), and 38.5% (130,900 cases), respectively, through tele-supported screening in a population with diabetes duration ≥5 years (Fig. S8, Table S11, see ESM). The states experiencing high and medium levels of epidemiological transition, such as Goa, Sikkim, Himachal Pradesh, Kerala, Tamil Nadu, Delhi, Puducherry, and Punjab are expected to derive greater benefits from a population-based DR screening programme compared with states in a low epidemiological transition level (Fig. S9, Table S12, see ESM).

3.2 Model Validation

We estimated the prevalence of VTDR and blindness in the usual-care scenario at 4.12% (4.16 million) and 1.01% (1.01 million), respectively, from the model. This was in line with the SMART-India population-based cross-sectional study, which determined the prevalence of VTDR as 4% (3.4–4.8%) [39]. Moreover, our model-estimated prevalence of blindness also corresponds with the findings from the national RAAB survey 2015–2019 [40]. We estimated annual incidence of VTDR and blindness at 1.4% and 0.3% in the absence of population-based screening, which was similar to that found by Raman et al. [20].

4 Discussion

In our model-based economic evaluation, we found that the systematic screening strategies led to greater improvement in health outcomes compared with the usual-care scenario and are cost effective. We found that DR screening is likely to result in a three times higher gain in QALYs than in life years in diabetic patients. The gain in life expectancy is primarily attributed to a reduction in the progression of DR to blindness, which is associated with a higher risk of mortality. In contrast, the improvements in QALYs stem from the preservation of vision, which has been linked to reduced disability and dependence, and the psychosocial benefits related to enhanced mental and emotional well-being. Prioritizing interventions for improvement in QALYs are more meaningful than those focused solely on extension of life years. Therefore, these findings emphasize the importance of early detection interventions, positioning DR screening as a critical strategy in improving overall patient outcomes.

Our analysis concluded that the implementation of annual DR screening at PHCs, supported by tele-ophthalmology in the population with diabetes duration of ≥5 years was the most cost-effective strategy, in comparison with other screening modalities and usual care (Table S13, see ESM). Our findings are consistent with the existing literature which strongly supports the implementation of teleophthalmology-supported DR screening at the primary-care level [30, 41,42,43,44,45]. Our findings also reinforce the importance of a strong primary care system and diabetes care alongside screening for DR.

Furthermore, screening was most cost effective in the population with a duration of diabetes of ≥5 years. This conclusion on the most efficient target population sub-group for screening is in concurrence with the biological plausibility of the disease, although we could not compare our results in economic terms due to limited published literature [46]. Nonetheless, the implementation of such a strategy would warrant strengthening of information systems to keep a track of the diabetic population and their duration of disease. Furthermore, the current rate of underdiagnosis and delay in diagnosis of diabetes (57%) in India adds to the operational challenges in the implementation of a restricted screening strategy according to the duration of disease [47, 48].

The findings of the study suggest the adoption of novel technology-supported screening approaches at the primary care level to improve technical efficiency. The Indian public healthcare system is embracing digital technologies to improve the access to healthcare service delivery [49, 50]. Moreover, the governments in states like Kerala have initiated implementation of the tele-model of eye care [51]. This evolving digital healthcare ecosystem presents a valuable opportunity to leverage technology for eye care at the primary care level. However, in the current scenario, the limited number of ophthalmologists in India (1:100,000 population), with skewed distribution towards urban areas and in the southern part of the nation, may make it challenging to implement a tele-supported screening in some parts of the country [52]. Therefore, while tele-supported screening is the optimal mode of screening from an efficiency lens, the states with deficient human resources may consider AI-based non-mydriatic fundus cameras to empower the primary care teams at peripheral facilities to conduct DR screening. This highlights that in a country as diverse as India, a ‘one strategy fits all’ approach may not be appropriate from an economic perspective. Therefore, states should prioritize the most efficient screening models, depending on epidemiological transition level and economic status.

It has been demonstrated that maintaining HbA1c ≤7% reduces development and progression of DR, with the positive effects persisting for up to 10–20 years [53]. Our study also highlights that changes in glycaemic control at the population level would necessitate alterations in the screening strategies due to its effect on incidence and progression of DR. Consequently, counselling for lifestyle modifications for adequate glycaemic control emerges as an important parameter of this analysis, emphasizing the need to prioritize prevention strategies within the programmes. The National Programme for Prevention and Control of Non-Communicable Diseases (NP-NCD) and DR screening under the Ayushman Bharat programme are likely to have a synergistic effect on treatment adherence among the diabetic population and contribute to an increase in the proportion of patients with diabetes who have adequate glycaemic control [54]. In such a scenario, a risk-stratified approach based on glycaemic control levels may be applied to achieve technical efficiency.

The study calls for more technological advancement in diagnostic accuracy of AI-based technologies. An increase in sensitivity of such devices to at least 96.5% would lead to generation of higher value than other modes of screening. Further, while most of the literature considers the assessment of AI-based non-mydriatic fundus cameras in non-primary care settings, and with pupil dilation to demonstrate its efficacy, it would be necessary to carefully validate the diagnostic accuracy of the devices in more realistic primary care settings without pupil dilation.

4.1 Strengths and Limitations

Our study has several merits. It is the first study that compares diverse models of DR screening in an LMIC setting in economic terms. Our analysis addressed the limitation of an earlier Indian study [31], which compared tele-based screening with a ‘no screening’ scenario, by considering a rational coverage rate of screening and compliance in accordance with the published literature, rather than assuming it to be 100%, and by inclusion of treatment costs in a usual-care/no-screening scenario. Furthermore, we also accounted for the potential occurrence of non-gradable images obtained through non-mydriatic fundus cameras. This consideration is significant as it influences the associated costs, since individuals with non-gradable images would need to be referred to higher-level facilities for additional diagnostic testing, irrespective of their disease status.

Thirdly, in the sensitivity analysis, we adjusted the cost of screening according to differential coverage, taking into account the economies of scale. Fourthly, recognizing that quality of life is influenced by socio-cultural context, we conducted primary data collection and derived utility values using the India value set [37]. We addressed the non-generalizable data limitations of another previous analysis through the use of India-specific literature on epidemiological parameters, and the estimation of screening costs utilizing the National Health Systems Cost Database [29, 55]. Since the cost of AI for DR is not standardized, we undertook stakeholder consultations to acquire market quotations for the government as a purchaser from the major suppliers of technology in India, to assess the costs from a payer’s perspective. The economic evaluation followed the Consolidated Health Economic Evaluation Standards (Table S15, see ESM) [56].

Our study results should be seen in the light of a few limitations due to a paucity of literature. Firstly, due to lack of India-specific data, the annual risk of incidence and progression of DR in patients with controlled glycaemic status relative to that in patients with uncontrolled glycaemic status was acquired from the UKPDS clinical trial [21]. While epidemiological studies are needed to estimate the rates of DR progression based on individuals' glycaemic levels, we believe this is unlikely to significantly affect our results. The close alignment between the model-derived incidence of VTDR and blindness, as compared to epidemiological surveys conducted in India, suggests that the input parameters related to DR progression at the population level are valid. Nonetheless, we varied these parameters by 25% during the sensitivity analysis and found that such changes did not alter the direction of the results or the overall conclusions.

Secondly, the incidence of macular edema was presumed to be the same irrespective of the duration of disease, however literature supports an insignificant change in the incidence of macular edema in relation to diabetes duration in Indian settings [20]. Thirdly, the coverage indicators such as screening rate and loss-to-follow-up rate in diagnosis and treatment were presumed to be the same for the face-to-face, tele-supported, and AI-supported screening arms and the diagnostic accuracy of tests by the ophthalmologists was assumed to be perfect in all the scenarios. This allowed for a uniform comparison of the screening strategies; however, we suggest on further research to evaluate the acceptability of different methods of screening within the populations. The additional benefits of an annual screening strategy, such as early detection and intervention for glaucoma and cataract, were overlooked for balancing model complexity. We did not consider the cost-saving impact of averting blindness in terms of productivity gains. However, inclusion of these benefits is likely to further raise the value of the intervention and make the conclusions stronger.

5 Conclusion

Our study presents important implications for decision makers regarding the most efficient models of DR screening at primary health centres. It provides strong evidence for an integrated approach and functional linkage between eye care and diabetes care within the context of primary healthcare, in view of the synergistic benefits in terms of control in glycaemic level as well as disease progression to improve quality of life. DR screening should be considered as an opportunity to emphasize the importance of adequate lifestyle modification and treatment compliance in the diabetic population, for alleviating the burden of blindness in India.