Assessing adoption of human and AI-enabled diabetic retinopathy screening in primary healthcare settings: findings from a pragmatic trial

Abstract

Introduction

Diabetic retinopathy (DR) screening with defined referral pathways is essential for early detection and effective management of DR. This study assessed the adoption of three DR screening (DRS) models in primary healthcare settings, focusing on referral adherence rates and stakeholders’ perceptions of the interventions.

Methods

A cross-sectional study was conducted in the Mohali district of Punjab, India, between February 2023 and January 2024. This pragmatic study compared three DRS arms (n = 200 each): I) facility-based screening at health and wellness centres (HWCs) by non-ophthalmologists, II) community-based AI DRS screening at home, and III) standard care involving counselling and referral to district hospitals (DHs). Participants with referable DR or ungradable images were advised for ophthalmology opinion, and their follow-up status and reasons for any non-compliance were assessed after one month. The adoption (acceptability and scalability) of the DRS was assessed via in-depth interviews with stakeholders involved in providing diabetes and DR care in public health settings.

Results

Among the 600 participants screened, the average age was 58.22 years (SD ± 11.52). Most participants, 300 (59.57%), were aged 51-70 years, comprising 245 (40.77%) males and 355 (59.23%) females. The referral adherence rates were low, ranging from 13% to 17% across Arms I, II, and III. Barriers to follow-up included lack of awareness, financial limitations, health concerns, perceived good eye health, and transportation challenges. Qualitative findings reveal that DRS, implemented through HWCs and community-based models, is feasible and well-accepted by patients. Stakeholders largely supported the implementation of DRS within primary healthcare settings, though responses varied. Likewise, DRS through HWCs and community-based models is feasible and well-accepted among patients.

Conclusion

The non-adherence to referrals among study participants is mainly attributable to economic constraints and knowledge gaps. Enhanced point-of-care counselling targeting groups at higher risk of non-adherence for follow-ups, along with a streamlined referral process, can improve the uptake of referral recommendations.

Trial registration

Clinical Trial Registry of India (CTRI): 2022/10/046283

Supplementary Information

The online version contains supplementary material available at 10.1186/s13690-025-01757-3.

Text box 1. Contributions to literature
• Evidence on integrating AI-assisted screening for diabetic retinopathy in routine public health systems in LMICs remains limited.
• The findings identify critical implementation bottlenecks, including workforce capacity, referral uptake, and system readiness in real-world settings.
• The study adds to the literature by examining digital innovation adoption within primary care infrastructure in a lower-middle-income country context.

Introduction

Diabetes mellitus imposes a significant burden on healthcare systems, primarily due to its associated complications [1]. The increasing prevalence of diabetes is projected to increase the health and economic burden due to its associated complications [2, 3]. Diabetic retinopathy (DR) is the most prevalent microvascular complication of diabetes and a leading cause of blindness among working-age adults in many countries [4, 5]. An estimated 28 million people worldwide are affected by vision-threatening DR (VTDR) [6]. DR screening (DRS) enables early detection, enabling timely treatment to prevent or delay diabetes-related blindness [7, 8].

The All India Ophthalmological Society (AIOS) recommends DRS for all people with diabetes, including those with RBS ≥ 200 mg/dl, HbA1c >6.5%, or gestational diabetes at diagnosis [9]. Despite increasing evidence of the effectiveness of national screening programs globally, comprehensive DRS strategies have not been widely adopted in India [10–12]. Retinal photography is an effective screening tool for DRS, but reliance on trained graders often results in delayed care and higher costs [13]. As the global burden of diabetes continues to rise, artificial intelligence (AI) algorithms have emerged as a scalable approach for DRS [14].

The DRS programs involves multifaceted implementation strategies, and understanding their adoption across organizational settings is essential for evaluating current and potential impact [15, 16]. Previous studies have cited examples of limited space, poor recruitment, high costs, and inadequate leadership, time, and training as key barriers [17, 18]. The success of a DRS intervention also depends on patients treatment adherence for DR management [9]. Suboptimal adherence to follow-up reduces treatment effectiveness and adversely affects patient outcomes [20]. Barriers to attending eye examinations include transportation challenges, high costs, and prolonged waiting times at healthcare facilities [21]. However, some factors (individual and health-related) influencing adherence and non-adherence to referral services remain unknown [22].

Artificial intelligence (AI) in medical imaging is advancing rapidly, with systems now enabling non-specialists to perform opportunistic DRS and make referral recommendations [23]. Despite this potential, evidence on their clinical adoption within healthcare systems remains limited [23]. Introducing such complex interventions in primary care poses additional challenges, particularly when they do not align with local contexts or stakeholder priorities [24]. Tailoring strategies to stakeholder feedback and evaluating adoption in real-world settings is, therefore, critical [25]. In this context, we examined innovative screening approaches at health and wellness centres (HWCs) alongside an AI-based automatic grading algorithm for DR detection in community settings. This study focused on referral uptake and adherence to DRS within primary healthcare, guided by the RE-AIM framework, while Reach, Implementation, and Effectiveness outcomes will be reported separately.

Methods

This manuscript was prepared using a template for intervention description and replication (TIDieR) for qualitative studies [28]. (see Additional file 1)

Study design and settings

This pragmatic three-arm observational study was conducted from February 2023 to January 2024 in Block Boothgarh District, Mohali, Punjab, India [29].DRS screening was conducted among people with diabetes mellitus (PwDM) aged 30 years and above, across three equal groups of 200 participants each. Arm I implemented non-ophthalmologist-led DRS at the Health and Wellness Centre (HWC), Khijrabad, Block Boothgarh; Arm II employed AI-based DRS; and Arm III received standard care without community-based screening (participant homes) (Fig. 1). Non-ophthalmologist screeners were optometrists holding a bachelor’s degree in optometry with prior retinal imaging experience. This paper reports findings from a larger study, focusing on the adoption of DRS interventions and the factors influencing their implementation at both the health system and participant levels [16]. The study [29] used the RE-AIM framework (reach, effectiveness, adoption, implementation, maintenance) to evaluate DRS implementation and outcome variations. Qualitative and quantitative data from participants and research team observations were analyzed to assess performance across RE-AIM domains [16].

Fig. 1.

Study design illustrating adoption of human and AI-enabled diabetic retinopathy screening in primary healthcare settings. 

In this study, adoption was defined as (1) the number of referral recommendations and adherence rates for each DRS intervention, and (2) stakeholders perspectives, including PwDM, research staff, healthcare providers (HCP), and program officers, perceptions of intervention usability and acceptability.

The study was approved by the Institutional Ethics Committee of the Postgraduate Institute of Medical Education and Research (PGIMER), Chandigarh (PGI/IEC/2020/000741) and registered under the Clinical Trial Registry of India (CTRI): 2022/10/046283.

Participants and sampling

Quantitative participants (PwDM)

The study included individuals aged 30 years and above with diabetes. Exclusion criteria comprised those under 30 without diabetes, individuals who did not consent, and those with a history of intraocular surgery, retinal treatments, or active eye infections or injuries [29].

Sample size

Sample size estimation was based on a binary outcome (DR: yes/no), assuming equal group allocation, alpha = 0.05 (1.96), power = 0.80 (beta = 0.20) [30],, with expected DR prevalence: p1 = 0.16, q1 = 0.84 (human grading) [31], and p2 = 0.28, q2 = 0.75 (AI screening) [32]. Based on prior studies evaluating DRS interventions in similar settings, the dropout rate was estimated at 15%. Therefore, the final required sample size was 600 PwDM.

Diabetic retinopathy screening (Quantitative data Collection)

Participant recruitment began with sensitization meetings involving local leaders and healthcare providers to ensure community awareness and support. The research team conducted line-listing and geographic mapping of households to identify eligible participants, followed by household visits to confirm diabetes status. In Arms II and III, a trained research optometrist performed DRS using a handheld fundus camera at participants homes. Arm III participants additionally received structured referral facilitation and follow-up. Two trained non-ophthalmologists (research optometrists) conducted non-mydriatic colour fundus imaging for screening. Two field-centered images (macula and disc) per eye were captured at a 45-degree field of vision (FOV). A Forus 3nethra classic (benchtop) [33, 34] was used to capture images at the HWC, and the Remidio NM FOP 10 (smartphone-based) integrated with the offline Medios AI algorithm, was used in the community [35].

Referral and follow-up

The screened participants were referred if they had (a) referable DR [36] or (b) ungradable images [37]. In Arms I and II, those with moderate non-proliferative DR and ungradable images were referred to the district hospital, while severe NPDR or proliferative DR cases were referred to tertiary care [29]. Arm III participants were advised to visit an ophthalmologist at the district hospital without a fundus examination. Telephone follow-up one-month post-screening assessed adherence to referral recommendations. The questionnaire, initially prepared in English, was administered in Punjabi or Hindi after obtaining telephone consent. Participants answered pre-set questions (yes, no, don’t know), and diagnoses of those who visited, reasons for non-attendance, and additional comments were recorded.

Qualitative data collection

Adoption was assessed across three stakeholder groups: (1) PwDM, (2) research optometrists, and (3) healthcare providers (HCPs) and program officers (POs). PwDM who underwent DRS were purposively sampled with support from accredited social health activists (ASHAs). Research optometrists involved in the interventions were interviewed. HCPs included Medical Officers from PHCs, Community Health Officers from HWCs, optometrists from CHCs and PHCs, ophthalmologists from DHs, and retina specialists from tertiary units. These HCPs, involved in diabetes and eye care across Mohali district public facilities, were purposively recruited with assistance from a Senior Medical Officer. POs from the state’s National Health Mission, responsible for implementing national health programs, were also purposively selected. (more details in Supplementary Table 1)

The in-depth interview guide was developed in English and translated into Hindi and Punjabi. Pre-specified open-ended questions and probes were organized logically and adapted to the interview flow. Pilot interviews with two PwDM, an ASHA, and a CHO prompted revisions, replacing ambiguous or complex terms with simpler language to improve participant understanding [38–40]. (Supplementary Tables 2,3)

Two research fellows (AC, MS) with qualitative experience conducted in-depth interviews. The research optometrist interviewed PwDM at their homes. HCPs and POs were interviewed face-to-face or via Zoom based on availability, while the research optometrist was interviewed at the PGIMER retina clinic. PwDM and ASHAs were interviewed in Hindi or Punjabi; research optometrists, HCPs, and POs mostly in English. Informed consent was obtained in person or verbally via Zoom. Interviews lasted 40–45 min, were audio-recorded, and continued until data saturation was reached [41].

Rigour and trustworthiness

The qualitative component was conducted with attention to recognised principles of trustworthiness, including credibility, transferability, dependability, and confirmability [41]. These standards align with recent qualitative research on the experiences of people with DR, which highlights the importance of transparency and methodological coherence in eye-care research [42, 43].

Credibility was supported through interviewer training, piloting of the guide, and triangulation across PwDM, optometrists, healthcare providers, and programme officers. Transferability was addressed through purposive sampling in both facility- and community-based contexts, providing sufficient detail on participant characteristics to facilitate comparison with similar settings. Dependability was supported by maintaining clear records of interview transcripts and coding, as well as by using structured interview guides tailored to each stakeholder group. Confirmability was strengthened through reflexive notes, peer review of emerging themes, and the use of anonymised quotations to illustrate findings.

Interviews continued until thematic saturation was reached. These steps enhanced the robustness of the qualitative enquiry and reduced potential concerns about selection bias.

Measures and analysis

Qualitative and quantitative data collected after screening were used to determine the adoption of the intervention. We analyzed descriptive statistics for participant characteristics, presenting continuous variables as the mean (SDs) and categorical variables as absolute counts (n) with corresponding percentages. All the statistical data were analyzed using Stata SE 15.0 (StataCorp LLC) [44]. The interviews conducted in Hindi and Punjabi were transcribed and translated into English by two trained investigators (SD, HR), who are proficient in both languages. The research scholar (AC) and trained investigators (SD, HR) familiarized themselves with the data. An inductive thematic analysis approach was employed to identify issues relevant to the usability and acceptability of DRS adoption. Two reviewers (AC, SD) reviewed and coded all the texts, and themes were discussed with other authors (MD and AK). The experienced public health experts, MD and AK, merged codes with similar meanings to prevent redundancy in the coded data. The two groups of participants are denoted in italics with a unique ID number: P 1–15 (PwDM); Opt 1–5 (Ophthalmologists); ASHA 1–4 (ASHA workers); CHO 1–3 (Community Health Officer), Optom 1–5 (Optometrists); and MO 1 (Medical Officer). Quotes were edited for readability to retain their original meaning; ellipses (…) show removed text.

Results

Quantitative findings

Sociodemographic characteristics of the PwDM

The screened participants (n = 600) had a mean age of 58.22 ± 11.52 years. Most participants, 300 (59.57%), were between 51 and 70 years. The participants aged ≤ 40 accounted for 46 (7.65%), while 12.98% were over 70 years old. The gender distribution revealed that 245 (40.77%) male participants and 355 (59.23%) female participants were affected. Most participants, 363 (60.91%), had a monthly family income of less than INR 30,000. A significant portion, 388 (64.77%), were unemployed or engaged in home duties. Education levels varied, with 193 (32.49%) being illiterate and only 3.03% having graduated or above (see Supplementary Table 4 for arm-wise socio-demographic details).

Referral and Follow-Up

A total of 83 participants in arm I, 86 in arm II, and 200 in arm III were referred for further follow-up after DRS. The mean age of the referred participants was highest in Arm I (61.2), followed by Arm II [58] and Arm III (58.5). Among those advised for follow-up, 72/83 (86.7%), 71/86 (82.6%), and 173/200 (86.5%) participants did not visit Arms I, II, and III, respectively (Supplementary Table 5).

Adherence to referral recommendations

Only 11/83 (13.25%) of Arm I patients visited higher facilities for follow-up. Severe NPDR patients had limited engagement, whereas one PDR patient received laser and anti-vascular endothelial growth factor (VEGF) therapy. In Arms II, only 15/86 (17.4%) patients visited higher facilities; severe NPDR cases received laser and anti-VEGF, while ungradable images resulted in cataract surgery scheduled for follow-up, and some patients received no treatment. Arm III participants (27/200, 13.5%) showed varied engagement across the DR stages. One severe NPDR patient visited higher facilities and received laser therapy, while 2 PDR patients sought care, with one receiving laser therapy (Supplementary Table 6). Based on the chi-squared test results, there was no evidence of a difference in adherence between the arms (p = 0.6).

Reasons for non-adherence to referral recommendations

The factors were divided into technical, family and social responsibilities, work and occupation demands, health perceptions, competing priorities, access, and practical barriers. Technical issues, such as faulty phone numbers, were common across all arms (Supplementary Table 7). Non-adherent participants cited reliance on family, personal matters, and work commitments as reasons for their non-adherence. Factors such as unawareness, financial constraints, medical concerns, family loss, perceived eye health, relocation, and harvesting affect follow-up. The PwDM stated that transportation issues, predominantly in the comments section, significantly hindered access to the eye care facilities. The lack of local transport (“Local transport is an issue. There are some Chandigarh Transport Union (CTU) buses, but they are rare, so traveling to (name of the facility) is an issue”); owning a vehicle did not guarantee health facility visits (“I do not know how to drive a car and depend on my son. When he is available, then I can visit the hospital); high cost of transportation (“Travelling to the hospital by two-wheeler is costly; I am a daily wager and cannot afford to waste my day’s wages.”) was presented by the participants as an additional but primary reason for not adhering to referral instructions. Further analysis revealed that 76 (12.67%) participants owned a two-wheeler, four-wheeler, or both, yet did not visit any health facility.

Qualitative findings

Participant characteristics

Of the 18 PwDM who were approached, 15 agreed to participate. The HCPs included one retina specialist, four ophthalmologists (out of 5 approached), five optometrists, one MO, 3 CHOs (out of 5 approached), and 4 ASHAs (out of 6 approached). Most PwDM participants (6 males, four females) were over 50 years of age, with a mean age of 60.18 ± 11.13 years. The HCPs included nine women and ten men. The average age of the HCPs was 38.3 ± 4.2 years, and the mean years of experience were 8.86 ± 3.9 years. The PO group comprised three males and one female, with a mean age of 46.8 ± 4.6 years. There were two male optometrists and one female optometrist (Supplementary Table 8).

Stakeholders’ perspective on the usability and acceptability of DRS interventions

Health and wellness centre -Based screening

• PwDM Perspective.

Participants expressed a favourable perception of HWC-based screening, appreciating its proximity and value to the broader community. Two PwDM reported encouraging peers to get screened at HWCs. While supportive overall, some participants voiced concern over diagnostic accuracy, stating a preference for screening conducted by ophthalmologists due to their experience. (Table 1)

Table 1.

Representative quotes on usability and acceptability of DRS adoption from the PwDM

Stakeholder	Screening facility	Representative quotes
PwDM (n = 15)	HWC based screening	(P): Such facilities (fundus screening) should be near our house within 5–10 km. It should be in our health centre as PGI is far from my village. P3 (P): It is an excellent initiative. Let me tell you, people in the village are already accepting it. I also recommend that my friends get their health checked like this. P5 (P): I have a problem with my eyes, and I have to travel far, which wastes my whole day; if something like this is established near my dispensary, this will be helpful. P8 (C): The screening will help only if the doctor has checked it personally, not by others (doctor nu dekhna paina). P2
	Community-based screening	(P): Yes, obviously, “bhoole nu raah paata” (showed me the direction). The team showed my eyes “retina” (parda dikhaya) on the screen, which gave me confidence that my eyes were fine. P5 (P): You came home; I am thankful because I have a knee problem, can barely walk, and low vision makes my life more terrible. If the screening had been done earlier, my vision would have been saved. P1 (P): Yes, this would be beneficial for everyone. One hundred eight villages fall under our block, which would benefit whole block screening on alternate days. P3 (C): “I don’t trust a machine to give complete results; it can sometimes go wrong, while the doctors here have so many years of experience.” P8

P: pros (in favor), C: cons (not in favor)

Abbreviations: HWC: Health and wellness centre, Optom: Optometrist, P: Patient

• Research Optometrist Perspective.

Optometrists preferred HWC-based screening due to its better infrastructure, including reliable power, a dark room, and patient waiting areas. They noted barriers, including low awareness and accessibility issues, among elderly individuals or those with mobility restrictions. Optometrists appreciated the quick AI-supported results in HWC settings, especially for patients with limited access to specialists. However, they emphasized the need for improved patient engagement strategies, as rural households may initially resist unfamiliar providers. Table 2 provides representative quotes from the optometrists.

Table 2.

Representative quotes from the research optometrists (n = 3) on the implemented DRS interventions

Stakeholder	Screening facility	Representative quotes
Research optometrist (n = 3)	HWC based screening	(P): Keeping the participant’s/patient’s comfort in mind, the facility has tables and chairs, making assessment more straightforward and accessible. All the devices are connected to continuous power. Optom 1 (P): Everything is set up, and the screening takes little time. Proper dark room conditions can be met for non-mydriatic fundus imaging, and adequate distance can be maintained for visual acuity measurement. Optom 2 (C): Patients avoid visiting the health facility due to lack of awareness, mobility issues due to old age, and lack of transportation, so I feel we will miss the elderly population for screening. Optom 2
	Community-based screening	(P): Patients who don’t have access to the facility benefit the most. It is now more accessible with offline AI software to deliver the results within a few minutes and refer for early treatment. Optom 1 (C): Maintaining proper distance to record VA at every household is challenging. Finding suitable (dark) room conditions for non-mydriatic fundus imaging at home is hard. Optom 3 (C): During home visits, the entire family needs to be explained and convinced for the checkup as people in rural areas prefer to avoid outsiders being allowed in their houses. Optom 1

P: pros (in favor), C: cons (not in favor)

Abbreviations: HWC: Health and wellness centre, Optom: Optometrist, P: Patient

• HCP and Program Officer Perspective.

Medical officers and CHOs highlighted the importance of integrating DRS into routine services at HWCs. CHOs advocated linking HWC-based screening with Ayushman Bharat activities and telemedicine services for specialist review. Optometrists and ophthalmologists emphasized the importance of training HWC-based staff in fundus photography and grading to minimize unnecessary referrals. Program officers expressed support for integrating AI into HWC workflows, while recognizing the need for a structured rollout plan to ensure smooth implementation. The representative quotes from the HCP and PO are reported in Table 3.

Table 3.

Representative quotes on usability and acceptability of DRS adoption from the HCP and PO

Stakeholder	Representative quotes
Ophthalmologist (n = 5) Includes one retina specialist	(P): Conducting diabetic retinopathy screening at a primary level will save an ophthalmologist precious time and reduce unnecessary referrals. Opt S3 (P): We must go somewhere to screen the patients. We must train an optometrist or technical person to capture good-quality images and use artificial intelligence software to obtain results. Opt S3 (P): Every technology and new thing will face problems, but we should still advance in that field. We must strengthen our support systems, like the medical or trained staff. Opt S4 (C): We may miss the cases through AI; obviously, it is a machine, and I prefer hospital settings for screening. Opt 1
Medical officer (n = 1)	(P): Whether facility or community-based, the screening program will be helpful for those who have never been screened due to a lack of resources. MO1
CHO (n = 3)	(P): If the patient is not getting a facility nearby or locally, then the patient does suffer. Fundus screening can be included in HWC, but we need training as we have been trained for other diseases. CHO1 (P): AB is a wonderful program because people feel connected with the healthcare services at their doorstep. This screening at HWC will be an excellent initiative as people will not have to go to other distant hospitals. CHO2 (P): We have desktops, and some of our fellow CHOs also have laptops. Your screening camera can be attached to it, and we can also connect to the telemedicine hub for further review by an expert and decide on further referral. CHO2
Health system optometrist (n = 5)	(P): First thing is equipment, and second is training. The basic advantage of thought and vision is giving all fundus cameras to peripheral centers to see the fundus, capture it, and refer the patient. Optom4 (P): I reviewed and referred further because there was no screening equipment. The screening idea is acceptable at the facility level, but we will need training and resources to implement it. Optom7 (C): Retina is not easy to understand and takes time. Anyone can capture the image, but to grade the image, an eye care professional is always better. Optom 6
ASHA (n = 4)	(P): I think the screening facility near rural areas will save patients time and money as they do not have to travel to (name of the hospital) more than 20 km from our village (pura din marna painda), and a full day is required to visit the hospital. ASHA1 (P): Yes, it would be beneficial. If we gather ten people with diabetes, at least three will agree to screen if it is near their home. ASHA2 (P): More eye screenings should be organized in villages because our village is difficult to reach from the city. ASHA3
Program officer (n = 4)	(P): Let’s always make the best of technology. We can use artificial intelligence to select a patient requiring referral to a higher centre at risk of developing retinopathy. And I think if we train our staff to do such interventions, that will be good.” PO3 (C): “It’s not like you have made an operational guideline the next day; it will be launched and implemented. You require step-by-step implementation, which requires much time and effort; it is a machine, and we must see its interaction with our healthcare workers.” PO1

P: pros (in favor), C: cons (not in favor)

Abbreviations: ASHA: Accredited social health activist, CHO: Community health officer, HCP: Healthcare provider, HWC: Health and wellness centre, Opt: Ophthalmologist, Optom: Optometrist, PO: Program officer

Community-Based screening

• PwDM Perspective.

Community-based screening was valued for its convenience, especially for elderly individuals or those with low vision. Participants appreciated the ease of access but expressed mixed feelings about diagnostic reliability when specialists were not involved. Some voiced greater trust in ophthalmologists and concern over possible errors in diagnosis. (Table 1)

• Research Optometrist Perspective.

While acknowledging the outreach potential of community-based screening, optometrists highlighted challenges such as suboptimal lighting, difficulty maintaining the visual acuity testing distance, and patient hesitancy toward unfamiliar visitors. They noted that engaging family members was often necessary to build trust. AI-supported results were also valued in these settings, provided that adequate imaging conditions were met. (Table 2)

• HCP and Program Officer Perspective.

ASHA workers strongly advocated for community-level DRS to reduce the travel burden and associated costs. They suggested integrating DRS with other outreach programs, such as maternal-child health or immunization camps. Program officers emphasized that community screening is crucial in reaching underserved populations and should be strengthened in conjunction with HWC services. (Table 3)

Discussion

This study evaluated follow-up rates and perspectives from PwDM, HCPs, POs, and the screening optometrist, alongside the effectiveness of DRS interventions in supporting PwDM. Findings indicate that DRS at HWCs and community-based models are feasible and well-accepted. As one of the first studies to track patient journeys from screening to referral and treatment, it provides valuable insights into the continuity of DR care. Stakeholders expressed diverse views on the screening initiatives. PwDM and HCPs strongly favored HWC- and community-based screenings, especially for individuals with limited mobility (255, 42.5%) or vision impairments. National guidelines support CHOs in conducting DRS and referring patients as needed [45]. Point-of-care screening with portable cameras [46] and AI algorithms has demonstrated good diagnostic accuracy in identifying VTDR [45]. AI-based screening was generally well-received, with some participants reporting increased confidence in clinical decisions. The PO noted implementation delays despite the national rollout, citing operational challenges. HCPs, POs, and research optometrists emphasized the need for infrastructure, equipment, and trained staff at primary care to support early detection, timely referral, and treatment [47]. Primary care offers more efficient access than tertiary care, with regular visits and shorter waiting times [48], though effective counselling remains essential to raise DR awareness [45]. ASHAs recommended rural screening centres and integrating DRS with national programs such as maternal-child health and immunization services.

Timely treatment of VTDR and regular follow-ups can prevent severe vision loss in up to 90% of cases [49, 50]. However, only a few patients with VTDR attend their recommended ophthalmology visits on time, with over 80% in our study failing to follow through with referrals. Contributing factors included personal and family issues, work commitments, financial constraints, other health problems, relocation, and perceived good eye health. Many participants remained unclear about the next steps after receiving results and referrals, while limited awareness and fear of tests further hindered timely care-seeking [51, 52].

In our study, 193 (33%) participants had no formal education, likely limiting their understanding of DR risks. This highlights the need for targeted counseling and improved health literacy, especially among low-income groups [53]. Phone follow-ups identified transportation as a common barrier, with vehicle ownership (12.7%) not guaranteeing visits. Additionally, 65% faced work or home responsibilities limiting travel to distant facilities (Supplementary Table 2). These findings are consistent with previous research showing that transportation difficulties, treatment costs, and lost income impede healthcare access [54–56]. Providing immediate DRS results can improve engagement and support timely referrals through counselling within a month [49]. In our study, although participants received immediate results and counseling, referral adherence remained low, ranging from 13% to 17% across study arms (Supplementary Table 3), showing little difference from standard care. These findings highlight patients at risk of non-adherence and the need for targeted counselling. Beyond individual-level interventions, broader strategies are required to address social, cultural, and economic factors that contribute to non-adherence [22]. Additionally, unclear care pathways continue to challenge the development of effective DR policies and screening programs. Despite increasing attention to DR, many policy gaps persist in LMICs, with limited engagement from policymakers and weak advocacy for improved DR care [57].

A strength of this study is its focus on referral non-adherence at the primary care level, the first point of integrated care [58]. It identifies gaps in follow-up and provides context-specific barriers relevant for program design. Limitations include potential selection bias: while all attending patients were enrolled, some PwDM in the Boothgarh block may have missed DRS and are likely non-adherent. The 70% telephone follow-up response rate also introduces bias, with missing or incorrect numbers limiting contact. Additionally, data on waiting times, accessibility, and staff attitudes were only captured qualitatively, and differences between adherent and non-adherent groups could not be systematically explored. Further research is needed to understand better why ophthalmology visits are missed after DRS.

A key strength of this study is its focus on referral non-adherence from primary to higher levels of care, highlighting gaps in follow-up and context-specific barriers relevant for program design. It also reports referral rates after DRS at the primary healthcare level, the first point of contact for integrated care [58]. The study has limitations, including potential selection bias. While all attending patients were enrolled, some diabetic patients in Boothgarh may have missed DRS and likely exhibit non-adherence. Telephone follow-up reached 70% (220/316), with 30% (96/316) affected by technical issues (Supplementary Table 5). We attempted to minimize bias through ASHA outreach and documentation of non-participation reasons. The study was conducted within a specific North Indian healthcare system; the study could not fully capture factors affecting adherence, such as waiting times, accessibility, and staff attitudes, beyond qualitative data. Additionally, we could not analyze differences between adherent and non-adherent groups, highlighting the need for further research on missed ophthalmology visits after DRS.

Policy implications and way forward

This study highlights the need to strengthen DRS at the primary care level through improved infrastructure, trained staff, and integrated counselling. Community-based and AI-supported screening can improve access for vulnerable groups while reducing unnecessary referrals. To improve adherence, targeted interventions addressing financial, educational, and logistical barriers are essential. Clear referral pathways, enhanced patient awareness, and the integration of DRS into broader national health programs are recommended to ensure continuity for DR care.

Abbreviations

ASHA

Accredited Social Health Activist

AI

Artificial Intelligence

CHO

Community Health Officer

DR

Diabetic Retinopathy

DH

District Hospital

DRS

Diabetic Retinopathy Screening

FOV

Field of View

HWC

Health and Wellness Centre

HCP

Health Care Provider

LMIC

Low and Middle-Income Countries

NPDR

Non-Proliferative Diabetic Retinopathy

PDR

Proliferative Diabetic Retinopathy

PO

Program Officer

PwDM

People with Diabetes Mellitus

VEGF

Vascular Endothelial Growth Factor

WHO

World Health Organization

VTDR

Vision-Threatening Diabetic Retinopathy

Author contributions

AC, MD, and LV contributed to the study’s conception and design. GK and N assisted with data collection, transcription, and translation. AC prepared the initial draft, while MD, LV, AK, NS, VG, BT, and SB reviewed it and provided significant revisions.

Funding

This research was funded by the Indian Council of Medical Research (ICMR) under approval number 5/4/6/13/OPH/2020-NCD-II. The funders have no role in our research’s study design, conduct, reporting, or dissemination plans.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to\u0000 ethical considerations and confidentiality agreements related to state health systems data\u0000 but are available from the corresponding author at a reasonable request.

Declarations

Ethical approval and consent to participate

The Institutional Ethics Committee (IEC) of PGIMER, Chandigarh, approved the study. (PGI/IEC/2020/000741). All participants gave informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.