Diabetes mellitus is now among the most consequential non-communicable diseases worldwide. The International Diabetes Federation estimated that 10.5% of adults aged 20–79 years, or 536.6 million people, were living with diabetes in 2021, with the steepest relative growth expected in middle-income countries.¹ India carries a disproportionate share of this burden; the ICMR-INDIAB national survey has documented a high prevalence of diabetes across urban and rural areas of the country.² Because the microvascular complications of diabetes accumulate silently over years, the size of this population translates directly into a large, partly hidden reservoir of end-organ damage.

Diabetic retinopathy (DR) is the most specific of these complications and remains a leading cause of preventable vision loss in working-age adults. A meta-analysis of 59 population-based studies estimated a global DR prevalence of 22.27% among people with diabetes, corresponding to 103.12 million adults in 2020 and a projected 160.50 million by 2045.³ An earlier pooled analysis of individual participant data reported a higher age-standardised prevalence of 34.6% and showed that every DR end point increased with diabetes duration, glycated haemoglobin and blood pressure.⁴ In India, the nationally representative SMART-India study estimated a DR prevalence of 12.5% and a vision-threatening DR prevalence of 4.0% among people aged 40 years or older with diabetes,⁵ and diabetes-related vision impairment has been shown to be a substantial national problem.⁶ Hospital- and population-based studies from South India have consistently identified diabetes duration and glycaemic control as the dominant risk factors for DR.^7,8

Clinical guidelines recommend regular retinal examination for all people with type 2 diabetes mellitus (T2DM) from diagnosis,⁹ and annual screening for albuminuria to detect diabetic kidney disease.¹⁰ In practice, however, these two surveillance activities are not always delivered with equal intensity. Where retinal examination is constrained by limited access to ophthalmologists, the urinary albumin result is sometimes used informally as a proxy for microvascular status, on the assumption that a patient without albuminuria is unlikely to have retinopathy. This assumption derives largely from the close coupling of nephropathy and retinopathy seen in type 1 diabetes, and it is less secure in T2DM.

The consequences of such an assumption are amplified by patient behaviour. Retinopathy is asymptomatic until late, and a qualitative study nested in SMART-India found that most people with sight-threatening retinopathy who were referred from community screening did not attend eye hospitals, citing low perceived susceptibility, cost and difficulty of access.¹¹ Physicians who see these patients in general medical clinics are therefore often the only professionals in a position to recognise risk and insist on referral. Any reassurance conveyed by a normal urine albumin result, whether intended or not, may further reduce the likelihood that the patient ever has a fundus examination. Identifying simple clinical markers that flag retinal risk in normoalbuminuric patients is therefore of direct practical value in general medicine, not only in ophthalmology.

Several lines of evidence suggest that retinal and renal microvascular injury frequently diverge in T2DM. In the Italian RIACE cohort of 15,773 patients, 56.6% of those with reduced estimated glomerular filtration rate were normoalbuminuric, and this phenotype correlated less strongly with retinopathy than albuminuric disease did.¹² Indian and North American studies have described a 'retino-renal dissociation', in which retinopathy and nephropathy progress with considerable discordance in T2DM.^13,14 Most directly relevant, the multinational PRIORITY study found retinopathy in 17.3% of 1,758 people with T2DM and normoalbuminuria, and its presence predicted subsequent microalbuminuria and cardiovascular events.¹⁵ In a Taiwanese cohort of normoalbuminuric inpatients, retinopathy independently predicted all-cause and cardiovascular mortality.¹⁶ Retinopathy in the normoalbuminuric patient is therefore neither rare nor clinically trivial.

If albuminuria cannot be relied on to flag retinal risk, clinicians need other readily available markers. Three candidates are measurable at any primary-care or outpatient visit. Blood pressure is an established and modifiable determinant of DR; in the United Kingdom Prospective Diabetes Study, tight blood pressure control reduced two-step retinopathy progression and the need for photocoagulation.^17,18 Diabetes duration is the most consistent risk factor across populations.^4,7 Anaemia is less often considered: low haemoglobin was associated with a two-fold higher odds of retinopathy in a Finnish clinic population,¹⁹ lower haematocrit predicted high-risk proliferative disease in the Early Treatment Diabetic Retinopathy Study,²⁰ and anaemia has shown a joint effect with diabetic kidney disease on DR severity in Chinese patients with T2DM.²¹ Anaemia is highly prevalent in India and is inexpensive to detect and correct, which makes it a potentially important target.

Most published evidence on these predictors, however, comes from mixed populations in which albuminuria itself is a powerful covariate, and few studies have examined them specifically in normoalbuminuric T2DM in a South Indian hospital population. Whether anaemia, hypertension and diabetes duration identify retinopathy in patients whose kidneys show no albuminuric signal is the gap this study addresses. We hypothesised that, among patients with T2DM, normoalbuminuria and normal renal function, a substantial proportion would have DR, and that lower haemoglobin, systemic hypertension and longer diabetes duration would be independently associated with its presence.

OBJECTIVES

The study had two objectives:

(i) To estimate the prevalence of diabetic retinopathy in patients with type 2 diabetes mellitus who have normoalbuminuria.

(ii) To study the predictors of diabetic retinopathy (anaemia, duration of type 2 diabetes mellitus and hypertension) in patients with normoalbuminuria.

Study design and setting This was a hospital-based cross-sectional study with prospective data collection, conducted in the Department of General Medicine, , Raichur institute of medical sciences (Rims), Raichur, Karnataka, India, between June and August 2014. Fundus examination was carried out with the support of the Department of Ophthalmology of the same institution. The study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement for cross-sectional studies. Participants Adults with T2DM attending the General Medicine services were screened for eligibility. T2DM was defined by a fasting plasma glucose ≥126 mg/dL or a 2-hour post-prandial plasma glucose >200 mg/dL, consistent with American Diabetes Association criteria,22 or by current treatment with dietary modification alone or with oral hypoglycaemic agents or insulin. Participants were required to have normoalbuminuria, defined as a urinary albumin excretion rate <20 μg/min or a urinary albumin-to-creatinine ratio <30 mg/g in two of three samples collected over 2–3 consecutive months, and normal renal function on routine renal function testing. Patients with a serum creatinine above 2 mg/dL were excluded; estimated glomerular filtration rate was not calculated. Exclusion criteria were age <18 or >80 years; dyslipidaemia; hepatic failure; renal failure; current smoking; thyroid dysfunction; acute systemic infection; strenuous exercise before urine collection; current treatment for anaemia; pregnancy; malignancy; and refractive errors that precluded reliable fundus assessment. These exclusions were chosen to remove conditions that independently alter urinary albumin excretion, haemoglobin concentration or the retinal examination. Participants were recruited by non-probability (convenience) sampling: patients attending the department during the study period who fulfilled the eligibility criteria were enrolled until the target of 100 participants was reached. The number of patients screened and excluded was not recorded. A total of 100 eligible participants were enrolled and all were included in the analysis (Figure 1). Sample size No formal sample-size calculation was performed; the sample size of 100 was fixed in advance as the number of eligible patients who could be recruited during the three-month study period. With 100 participants, the observed prevalence of DR was estimated with a 95% confidence interval half-width of approximately ±9.5 percentage points (Wilson interval 30.9–49.8% around 40.0%; half-width = (49.8 − 30.9)/2). Data collection and measurements Each participant underwent a structured clinical history, general and systemic examination and laboratory investigations using a standard proforma. The history captured age, sex and the duration of T2DM in completed years from the date of diagnosis. Laboratory investigations included haemoglobin, red cell indices, packed cell volume, peripheral blood smear, urine albumin, renal function tests, liver function tests and a lipid profile; the latter tests were used to apply the exclusion criteria. All laboratory tests were performed in the central clinical laboratory of Government Rajaji Hospital using its standard routine methods. Blood pressure was measured during the clinical examination at the study visit and recorded in mmHg. Hypertension status was assigned from the recorded value; antihypertensive treatment status was not recorded. All participants underwent ophthalmoscopic examination of the fundus in the Department of Ophthalmology to screen for DR. DR was diagnosed by the presence of characteristic lesions, such as microaneurysms, retinal haemorrhages, hard exudates, cotton-wool spots, venous beading, intraretinal microvascular abnormalities or new vessels, in either eye. DR was recorded as present or absent; severity grades were not recorded in the study dataset. Where severity grading is reported in future work, the International Clinical Diabetic Retinopathy Disease Severity Scale is recommended.23 Variables and operational definitions The outcome was the presence of DR in either eye. The three pre-specified exposures were haemoglobin concentration, systemic hypertension and duration of diabetes. Haemoglobin was analysed as a continuous variable (g/dL) and in four categories (<6, 6–7.9, 8–11.9 and ≥12 g/dL). Anaemia was defined as haemoglobin <12 g/dL; because no male participant had a haemoglobin value between 12.0 and 12.9 g/dL, this threshold classified participants identically to sex-specific cut-offs of 12 g/dL for women and 13 g/dL for men. Following the study protocol, a haemoglobin concentration <8 g/dL was used as the primary anaemia exposure in multivariable models. Hypertension was defined as a systolic blood pressure ≥140 mmHg or a diastolic blood pressure ≥90 mmHg and staged according to the Seventh Report of the Joint National Committee (JNC 7) categories,24 as stage 1 (140–159/90–99 mmHg) and stage 2 (≥160/≥100 mmHg). Diabetes duration was analysed as a continuous variable (years), in three categories (<5, 5–15 and >15 years) and as a binary exposure (≥5 years versus <5 years). Age and sex were treated as potential confounders. Bias Selection bias was limited by applying uniform eligibility criteria to all screened patients, and information bias by using a common proforma for all participants. Because the study was conducted at a single tertiary referral centre, the sample may over-represent patients with longer or more complicated diabetes; this is considered in the Discussion. Statistical analysis Continuous variables were assessed for normality with the Shapiro–Wilk test; because all were non-normally distributed, they are summarised as median (interquartile range, IQR) alongside mean (standard deviation, SD), and compared between participants with and without DR using the Mann–Whitney U test. Categorical variables are presented as counts and percentages with denominators. The prevalence of DR is reported with a Wilson score 95% confidence interval (CI). Associations between each exposure and DR were examined with Pearson's chi-square test, and crude odds ratios (ORs) with 95% CIs were calculated. Linear trends in DR prevalence across ordered categories of haemoglobin, diabetes duration and the number of predictors present were tested with the Mantel–Haenszel linear-by-linear association test. Two multivariable logistic regression models were fitted, with DR as the dependent variable and all covariates entered simultaneously rather than by stepwise selection. Model A, the primary model, included the three dichotomised exposures (haemoglobin <8 g/dL, hypertension, duration ≥5 years) together with age (continuous) and male sex. Model B included haemoglobin (per g/dL), systolic blood pressure (per mmHg) and diabetes duration (per year) as continuous variables, with age and sex. With 40 outcome events, each model had eight events per candidate variable. Multicollinearity was assessed with variance inflation factors, model fit with the likelihood-ratio test and McFadden pseudo-R², calibration with the Hosmer–Lemeshow test and discrimination with the area under the receiver operating characteristic curve (AUC), for which 95% CIs were obtained from 2,000 bootstrap resamples. Because sparse cells and near-separation can inflate maximum-likelihood estimates in small samples, Model A was refitted using Firth's penalised likelihood as a sensitivity analysis.25 Because the predictors were expected to co-occur, their inter-relationships were described with chi-square tests and Spearman correlation, and the associations of anaemia and hypertension with DR were examined within strata of diabetes duration (<5 versus ≥5 years) using Fisher's exact test and the Mantel–Haenszel pooled OR (with a 0.5 continuity correction for zero cells), with homogeneity across strata assessed by the Breslow–Day test. A further sensitivity analysis replaced haemoglobin <8 g/dL with anaemia defined as haemoglobin <12 g/dL in Model A. As an exploratory, post hoc analysis, the number of predictors present (0–3) was cross-tabulated against DR status, and the sensitivity, specificity and predictive values of having at least one predictor were calculated. There were no missing data for any study variable. All tests were two-tailed, with p <0.05 considered statistically significant. Analyses were performed in Python 3 using SciPy (version 1.18.1), statsmodels (version 0.15.0) and scikit-learn (version 1.9.1). Ethical considerations The study protocol was approved by the Institutional Review Board/Independent Ethics Committee of , Raichur institute of medical sciences (Rims), Raichur, Karnataka, India. Written informed consent was obtained from every participant, and confidentiality of participant information was maintained throughout. The study was conducted in accordance with the Declaration of Helsinki.

Participant characteristics

All 100 enrolled participants had complete data and were included in the analysis . Their mean age was 51.9 years (SD 12.7; range 30–72), and 52 (52.0%) were men. The median duration of diabetes was 4.0 years (IQR 3.8–13.0; range 2–23). Thirty-nine participants (39.0%) were anaemic (haemoglobin <12 g/dL), of whom 31 had a haemoglobin concentration <8 g/dL, and 30 (30.0%) were hypertensive. The characteristics of participants with and without DR are shown in Table 1.

Compared with participants without DR, those with DR had a lower median haemoglobin (7.5 g/dL [IQR 6.8–12.4] versus 14.9 g/dL [IQR 14.0–15.0]; p < 0.001), a higher median systolic blood pressure (142.0 mmHg versus 120.0 mmHg; p < 0.001) and diastolic blood pressure (94.0 mmHg versus 82.0 mmHg; p = 0.002), and a substantially longer median duration of diabetes (13.5 years [IQR 9.0–18.2] versus 4.0 years [IQR 3.0–4.0]; p < 0.001). Age did not differ significantly between the groups (median 55.5 versus 49.0 years; p = 0.20). Men made up 62.5% (25/40) of participants with DR and 45.0% (27/60) of those without (p = 0.086).

Prevalence of diabetic retinopathy

DR was detected in 40 of the 100 normoalbuminuric participants, a prevalence of 40.0% (95% CI 30.9–49.8). Prevalence was 48.1% (25/52) in men and 31.2% (15/48) in women. Across age groups it was 46.2% (6/13) at 18–34 years, 23.5% (8/34) at 35–49 years, 54.5% (18/33) at 50–64 years and 40.0% (8/20) at 65 years or older, without a consistent age gradient (Table 2).

Distribution of DR by predictor

DR prevalence varied markedly with each of the three pre-specified predictors (Table 2; Figure 1). Among participants with haemoglobin ≥12 g/dL, 16.4% (10/61) had DR, compared with 75.0% (6/8) at 8–11.9 g/dL, 85.7% (18/21) at 6–7.9 g/dL and 60.0% (6/10) below 6 g/dL (test for trend p < 0.001). Overall, 76.9% (30/39) of anaemic participants had DR, and 77.4% (24/31) of those with haemoglobin <8 g/dL.

Among normotensive participants, 24.3% (17/70) had DR, compared with 77.8% (14/18) of those with stage 1 hypertension and 75.0% (9/12) of those with stage 2 hypertension. DR prevalence rose steeply with diabetes duration, from 10.5% (6/57) at <5 years to 75.0% (18/24) at 5–15 years and 84.2% (16/19) at >15 years (test for trend p < 0.001). Of the 43 participants with diabetes of 5 years or longer, 34 (79.1%) had DR.

Crude associations

On bivariate analysis (Table 3), haemoglobin <8 g/dL (crude OR 11.36; 95% CI 4.13–31.20), anaemia defined as haemoglobin <12 g/dL (17.00; 6.21–46.54), hypertension (10.24; 3.74–28.04) and diabetes duration ≥5 years (32.11; 10.47–98.46) were each strongly associated with DR (all p < 0.001). Male sex (2.04; 0.90–4.61; p = 0.086) and age ≥50 years (2.27; 0.99–5.18; p = 0.050) showed weaker associations whose confidence intervals included the null.

Multivariable analysis

In the primary multivariable model (Model A; Table 4), adjusted for age and sex, haemoglobin <8 g/dL (adjusted OR 9.81; 95% CI 2.00–48.00; p = 0.005), hypertension (5.35; 1.17–24.46; p = 0.031) and diabetes duration ≥5 years (35.19; 8.19–151.18; p < 0.001) remained independently associated with DR, whereas age (1.02 per year; 0.97–1.07; p = 0.48) and male sex (0.82; 0.22–3.04; p = 0.76) were not (Figure 2). The model was highly significant (likelihood-ratio χ² 73.41, 5 degrees of freedom; p < 0.001), explained a substantial share of outcome variation (McFadden R² 0.545) and showed no relevant multicollinearity (maximum variance inflation factor 1.35).

Firth penalised regression produced smaller but directionally identical estimates: haemoglobin <8 g/dL, OR 7.54 (95% CI 1.77–32.05; p = 0.006); hypertension, 4.44 (1.08–18.23; p = 0.038); and duration ≥5 years, 23.54 (6.36–87.08; p < 0.001). When predictors were modelled continuously (Model B), each 1 g/dL increase in haemoglobin was associated with 34% lower odds of DR (adjusted OR 0.66; 95% CI 0.52–0.84; p < 0.001), each 1 mmHg increase in systolic blood pressure with 4% higher odds (1.04; 1.01–1.08; p = 0.017) and each additional year of diabetes with 43% higher odds (1.43; 1.23–1.67; p < 0.001).

Discrimination and calibration

Model A discriminated well between participants with and without DR (AUC 0.925; bootstrap 95% CI 0.853–0.984), and Model B slightly better (AUC 0.953; 0.896–0.994) (Figure 3). Among single markers, diabetes duration had the highest AUC (0.894; optimal cut-off 7 years, sensitivity 82.5%, specificity 88.3%), followed by haemoglobin (0.790; cut-off ≤11.8 g/dL, sensitivity 75.0%, specificity 85.0%) and systolic blood pressure (0.736; cut-off 142 mmHg, sensitivity 57.5%, specificity 88.3%). Calibration was imperfect: the Hosmer–Lemeshow test indicated lack of fit for Model A (χ² 22.19, 8 degrees of freedom; p = 0.005) and, to a lesser degree, for Model B (χ² 16.84; p = 0.032).

Stratified and sensitivity analyses

The three predictors clustered within the same individuals. Hypertension was present in 19 of the 31 participants (61.3%) with haemoglobin <8 g/dL, compared with 11 of 69 (15.9%) with higher haemoglobin (p < 0.001), and haemoglobin was inversely correlated with both diabetes duration (Spearman ρ −0.38; p < 0.001) and systolic blood pressure (ρ −0.31; p = 0.002). Because diabetes duration was the strongest correlate of DR, the associations of anaemia and hypertension were examined within duration strata. Among participants with diabetes for less than 5 years, DR was present in 30.0% (3/10) of those with haemoglobin <8 g/dL and 6.4% (3/47) of those without (Fisher p = 0.060); among those with diabetes for 5 years or more, the corresponding figures were 100% (21/21) and 59.1% (13/22) (p = 0.001). The duration-stratified Mantel–Haenszel OR for haemoglobin <8 g/dL was 12.07 (95% CI 2.97–49.03), with no evidence of heterogeneity between strata (Breslow–Day p = 0.29). For hypertension, DR prevalence was 33.3% (3/9) versus 6.2% (3/48) in the shorter-duration stratum (p = 0.044) and 95.2% (20/21) versus 63.6% (14/22) in the longer-duration stratum (p = 0.021), giving a Mantel–Haenszel OR of 7.56 (95% CI 2.10–27.26; Breslow–Day p = 0.91).

When anaemia was defined at the conventional threshold of haemoglobin <12 g/dL in place of <8 g/dL in Model A, anaemia (adjusted OR 36.21; 95% CI 3.97–330.41) and diabetes duration ≥5 years (80.38; 9.55–676.55) remained strongly associated with DR, but the association with hypertension was attenuated and no longer statistically significant (2.34; 0.41–13.40; p = 0.339). The very wide intervals in this model reflect the small number of participants without DR in the higher-risk cells.

Cumulative predictor burden (exploratory)

DR prevalence increased with the number of predictors present: 7.3% (3/41) with none, 30.0% (9/30) with one, 92.3% (12/13) with two and 100% (16/16) with all three (test for trend p < 0.001; Figure 2D). All 21 participants who had both haemoglobin <8 g/dL and diabetes duration ≥5 years had DR. The presence of at least one predictor identified DR with a sensitivity of 92.5% (37/40; 95% CI 80.1–97.4) and a specificity of 63.3% (38/60; 95% CI 50.7–74.4); the negative predictive value was 92.7% (38/41; 95% CI 80.6–97.5) and the positive predictive value 62.7% (37/59; 95% CI 50.0–73.9).

Table 1. Characteristics of participants with type 2 diabetes and normoalbuminuria, overall and by diabetic retinopathy status (N = 100)

Percentages are column percentages. ᵃMann–Whitney U test. ᵇPearson chi-square test. Abbreviations: BP, blood pressure; DR, diabetic retinopathy; Hb, haemoglobin; IQR, interquartile range; SD, standard deviation.

Table 2. Prevalence of diabetic retinopathy by demographic characteristics and predictor categories (N = 100)

Prevalence is row percentage with Wilson score 95% CI. ᵃPearson chi-square test (blood pressure: hypertension versus normal). ᵇMantel–Haenszel test for linear trend. ᶜPredictors: haemoglobin <8 g/dL, hypertension and diabetes duration ≥5 years (exploratory, post hoc). Stage 1 hypertension: 140–159/90–99 mmHg; stage 2: ≥160/≥100 mmHg. Abbreviations: CI, confidence interval; DR, diabetic retinopathy.

Table 3. Crude (unadjusted) associations of candidate predictors with diabetic retinopathy (N = 100)

p values from Pearson chi-square test (1 degree of freedom). Reference category for each exposure is its absence (female sex for male sex). Abbreviations: CI, confidence interval; DR, diabetic retinopathy; Hb, haemoglobin; OR, odds ratio.

Table 4. Multivariable logistic regression models for diabetic retinopathy (N = 100; 40 events)

All covariates entered simultaneously. Model A: dichotomised predictors with age and sex. Firth: penalised-likelihood logistic regression (Wald 95% CI) as sensitivity analysis for small-sample bias. Model B: predictors as continuous variables. AUC 95% CI from 2,000 bootstrap resamples. Maximum variance inflation factor 1.35 (Model A) and 1.30 (Model B). ᵃModel A with anaemia defined as Hb <12 g/dL in place of Hb <8 g/dL, also adjusted for age and sex. Reference categories: Hb ≥8 g/dL (≥12 g/dL in the sensitivity model), normotension, duration <5 years, female sex. Abbreviations: aOR, adjusted odds ratio; AUC, area under the receiver operating characteristic curve; BP, blood pressure; CI, confidence interval; HL, Hosmer–Lemeshow test.

Figure 1. Prevalence of diabetic retinopathy by predictor category. Bars show the percentage of participants with diabetic retinopathy in each category of (A) haemoglobin, (B) blood pressure, (C) duration of diabetes and (D) number of predictors present; error bars are Wilson 95% confidence intervals and labels show n/N. The reference category in each panel is shown in blue. *Predictors: haemoglobin <8 g/dL, hypertension and diabetes duration ≥5 years (exploratory analysis). Stage 1 hypertension, 140–159/90–99 mmHg; stage 2, ≥160/≥100 mmHg.

Figure 2. Crude and adjusted odds ratios for diabetic retinopathy. Forest plot of crude odds ratios (circles) and adjusted odds ratios from multivariable Model A (squares) with 95% confidence intervals, on a logarithmic scale. Model A included haemoglobin <8 g/dL, hypertension, duration of diabetes ≥5 years, age (per year) and male sex. The dashed line marks an odds ratio of 1. Age was modelled continuously and has no crude dichotomous estimate shown.

Figure 3. Receiver operating characteristic curves for identification of diabetic retinopathy. Curves for multivariable Model B (continuous haemoglobin, systolic blood pressure, duration, age and sex), Model A (dichotomised predictors, age and sex), duration of diabetes alone and haemoglobin alone (lower values indicating higher risk). The diagonal dashed line indicates no discrimination. AUC, area under the curve. Discrimination is apparent (in-sample) and is likely to be optimistic.

In this hospital-based study of adults with T2DM, normoalbuminuria and normal renal function, two in five participants had diabetic retinopathy. Anaemia and diabetes duration of five years or more were robustly associated with DR after adjustment for age and sex, across penalised estimation and alternative definitions; systemic hypertension was also independently associated in the primary model, though less robustly. Together, the three markers discriminated well between those with and without retinopathy. The absence of albuminuria therefore offered no reassurance about the retina in this population. Prevalence of retinopathy in normoalbuminuric T2DM The prevalence of 40.0% is higher than the 17.3% reported among normoalbuminuric participants in the PRIORITY study15 and well above the 12.5% national estimate for all people with diabetes aged 40 years or older in SMART-India.5 It is also above the 17.5% reported in the population-based CURES Eye Study from Chennai, in which participants were not selected on albuminuria status.26 The difference is best explained by setting rather than biology. Our participants were drawn from the medical services of a tertiary referral hospital, a population enriched for longer, more symptomatic and less well-controlled diabetes; 43% had diabetes for five years or longer, and nearly two-fifths were anaemic. Population-based and screening studies, by contrast, include many people with recently detected or undiagnosed diabetes in whom retinopathy has had little time to develop. The confidence interval around our estimate is also wide (30.9–49.8%). Our figure should therefore be read as the burden in a hospital-attending normoalbuminuric population, not as a community prevalence. Even the lower bound, however, implies that close to one in three such patients has retinopathy, which is ample reason not to use a normal albumin result to defer retinal examination. The finding fits a growing body of evidence that retinal and renal microvascular disease frequently dissociate in T2DM. In RIACE, normoalbuminuric renal impairment was the dominant phenotype of diabetic kidney disease and was less strongly associated with retinopathy than albuminuric disease,12 and Indian and North American data describe substantial discordance in the severity of retinopathy and nephropathy.13,14 Unlike type 1 diabetes, in which retinopathy and albuminuria tend to evolve in parallel, T2DM is a heterogeneous condition in which hypertension, ageing, atherosclerosis and non-diabetic renal disease all shape the kidney, so that albuminuria is an imperfect proxy for diabetic microangiopathy elsewhere. Retinopathy in the normoalbuminuric patient also carries prognostic weight beyond vision: it predicted incident microalbuminuria and cardiovascular events in PRIORITY15 and mortality in a Taiwanese cohort.16 A normoalbuminuric patient with retinopathy is therefore not a low-risk patient. Anaemia Low haemoglobin showed one of the clearest relationships with retinopathy in this study, with roughly ten-fold higher adjusted odds below 8 g/dL and a graded inverse association per g/dL on continuous modelling. The direction of this association is concordant with a consistent international literature. Qiao and colleagues found that patients with haemoglobin <12 g/dL had twice the odds of any retinopathy and five times the odds of severe retinopathy, after adjustment for creatinine and proteinuria.19 In the Early Treatment Diabetic Retinopathy Study, lower haematocrit was an independent baseline risk factor for progression to high-risk proliferative retinopathy.20 More recent cross-sectional studies have reported adjusted ORs between 1.44 and 3.71 for anaemia in T2DM,21,27 an inverse haemoglobin–DR relationship in large NHANES and Chinese clinical datasets,28 and a synergistic effect of anaemia and long diabetes duration.29 Similar associations have been described in the Middle East30 and, prospectively, in urban India, where anaemia predicted incident DR over 15 years.31 The magnitude of our estimate is considerably larger than these published effects. Three features of the study plausibly contribute. First, anaemia here was frequent and severe, with 31 participants below 8 g/dL, whereas most published studies compare mildly anaemic with non-anaemic patients. Second, the sample was small, so the estimates are imprecise; the wide confidence interval (2.00–48.00) and the attenuation to 7.54 under Firth regression indicate that the true effect is probably nearer the lower end of the interval. Third, and most importantly, glycated haemoglobin was not measured, so poor glycaemic control, which is associated with both lower haemoglobin and retinopathy, may have confounded the association. In addition, anaemia in diabetes is partly a consequence of early, non-albuminuric renal disease through impaired erythropoietin production, and normal routine renal function tests do not exclude reduced glomerular filtration. The anaemia–DR association in our study may therefore partly reflect unmeasured kidney dysfunction, which would be consistent with the joint effect of anaemia and diabetic kidney disease described by Wang and colleagues.21 A biological basis for a direct effect is nonetheless plausible. Reduced oxygen-carrying capacity aggravates retinal hypoxia, the principal stimulus for vascular endothelial growth factor (VEGF) release, and VEGF concentrations in ocular fluid are closely linked to active proliferative retinopathy.32 Small clinical series have described regression of hard exudates and macular oedema after correction of severe anaemia with erythropoietin.33,34 These observations do not establish that treating anaemia prevents retinopathy, but they support routine measurement of haemoglobin in people with T2DM and suggest that correcting anaemia is a reasonable, low-risk adjunct that warrants formal evaluation. Hypertension Hypertension was associated with a roughly five-fold increase in the adjusted odds of DR, and each 1 mmHg rise in systolic pressure with a 4% increase. This is concordant with the pooled analysis by Yau and colleagues, in which all DR end points increased with blood pressure,4 and with the UKPDS, in which tight blood pressure control reduced microaneurysm counts, hard exudates and two-step progression of retinopathy.17,18 Takao and colleagues further showed that systolic pressure over the preceding five years influences the development of retinopathy.35 The literature is not uniform, however. In the CURES Eye Study, hypertension did not emerge as a major risk factor in cross-sectional analysis,8 and the Chennai CURES analysis of microvascular complications identified age, glycaemia, duration and triglycerides, rather than blood pressure, as the risk factors common to all three complications.26 Differences in hypertension definitions, in the proportion of treated hypertensive participants and in adjustment for glycaemia probably account for much of this heterogeneity. In our data, DR prevalence was similar in stage 1 and stage 2 hypertension, so we could not demonstrate a dose–response across hypertension stages, possibly because of small numbers. Blood pressure was also recorded at a single visit, and treatment status was not analysed, which may have led to misclassification. Most importantly, the independent contribution of hypertension was the least robust of the three associations: it persisted within duration strata and under Firth estimation, but was attenuated to non-significance when anaemia was defined at the broader threshold of haemoglobin <12 g/dL. Hypertension should therefore be regarded as a probable but less certain independent predictor in this population. Clustering of predictors A notable feature of the data is that the three predictors did not occur independently. Nearly two-thirds of participants with haemoglobin <8 g/dL were hypertensive, compared with about one in six of those with higher haemoglobin, and lower haemoglobin accompanied both longer diabetes duration and higher systolic pressure. This clustering is biologically coherent. Longer exposure to hyperglycaemia damages the renal tubulointerstitium and impairs erythropoietin production, producing anaemia before albuminuria appears, and the same process contributes to sodium retention and hypertension. Normoalbuminuric renal impairment, which is now the most common phenotype of diabetic kidney disease in T2DM,12 may therefore be the shared substrate that links all three markers to retinopathy, and its absence from the dataset is a key limitation. From an analytical perspective, clustering means that the separate adjusted ORs should not be over-interpreted as independent causal effects; the combination of markers, rather than any single one, is the more robust and clinically useful signal. The duration-stratified analyses support this reading: within each duration stratum, anaemia and hypertension were each associated with a several-fold higher prevalence of DR, indicating that their associations are not simply a reflection of longer diabetes. Duration of diabetes Diabetes duration was the strongest single correlate of retinopathy, with an AUC of 0.894 on its own and a steep rise in prevalence after five years. This is the most reproducible finding in DR epidemiology, reported in global pooled data4 and in South Indian cohorts,7,26 in young-onset T2DM, where DR odds roughly doubled per five years of duration,36 and in Southeast Asian screening populations.37 The very large adjusted OR in our study (35.19) must nevertheless be interpreted with caution. Duration was self-reported from the date of diagnosis, and because T2DM is frequently diagnosed years after onset, true exposure is underestimated in a way that may differ between patients. Retinopathy is already present in a proportion of people at the time T2DM is diagnosed,8 which is consistent with this underestimation. Sparse cells also inflated the estimate; the Firth estimate of 23.54 is likely to be closer to the true association. Combined predictor burden and clinical implications The exploratory analysis of cumulative burden is perhaps the most clinically useful result. Prevalence rose from 7.3% with none of the three predictors to 100% with all three, and every participant with both severe anaemia and five or more years of diabetes had DR. In a setting where retinal screening capacity is limited, these three inexpensive measurements could help to prioritise referral, and the 92.7% negative predictive value of having no predictor is reassuring for triage. They should not, however, replace universal screening: three of the 41 participants with no predictor still had retinopathy, the cut-offs were derived and tested in the same sample, and the Hosmer–Lemeshow tests indicated that predicted probabilities were imperfectly calibrated. The models should therefore be regarded as explanatory rather than as validated prediction tools. Current guidelines already recommend retinal examination for every person with T2DM at diagnosis and periodically thereafter;9 our results reinforce that this recommendation applies regardless of albuminuria status, and suggest that anaemic, hypertensive patients with longer-standing diabetes should be examined promptly. Because retinopathy has been linked to renal outcomes even in normoalbuminuric patients,15,38 a finding of DR in such a patient should also prompt closer surveillance of kidney function and cardiovascular risk. Strengths The study focused specifically on normoalbuminuric T2DM with normal renal function, a group in which retinal risk is easily underestimated. Eligibility required normoalbuminuria on at least two of three samples, which reduces misclassification arising from the day-to-day variability of albumin excretion. Several conditions that independently affect albuminuria, haemoglobin or retinal assessment were excluded. There were no missing data. The analysis reported crude and adjusted estimates with confidence intervals, modelled predictors both as categories and continuously, and examined robustness with Firth penalised regression, discrimination and calibration. Limitations This study has important limitations. First, the cross-sectional design does not allow causal or temporal inference; anaemia, in particular, could be a consequence rather than a cause of the microvascular process. Second, the sample was small and drawn from a single tertiary centre over three months, which produced wide confidence intervals and limits generalisability to primary-care or community populations. Third, glycated haemoglobin, lipid levels (other than for exclusion), estimated glomerular filtration rate and antihypertensive treatment were not included in the analysis, so residual confounding, especially by glycaemic control and subclinical renal impairment, is likely. Fourth, DR was diagnosed by clinical ophthalmoscopy and recorded only as present or absent; severity grades and macular oedema were not captured, pupil dilation and masking of the examiner to haemoglobin and blood pressure results were not documented, and ophthalmoscopy is less sensitive than standardised retinal photography for mild lesions. The number of patients screened for eligibility was also not recorded, so the participation fraction cannot be reported. Fifth, blood pressure was based on clinic measurements at a single visit and duration of diabetes on self-report. Sixth, the exclusion of patients with dyslipidaemia, smokers and those already treated for anaemia, while reducing confounding, produced a selected population. Finally, the predictor cut-offs and the cumulative-burden analysis were derived and assessed in the same data, and are therefore likely to be optimistic; they require external validation. Clinical implications Normoalbuminuria should not be used to defer or de-prioritise retinal examination in people with T2DM. Haemoglobin, blood pressure and diabetes duration are available at almost every outpatient visit and can help clinicians identify normoalbuminuric patients who need prompt referral for fundus examination. Routine haemoglobin estimation and the investigation and correction of anaemia deserve a more explicit place in comprehensive diabetes care, and the detection of retinopathy in a normoalbuminuric patient should prompt closer renal and cardiovascular surveillance. Future research Larger, multicentre prospective studies in Indian populations are needed to confirm these associations with adjustment for glycated haemoglobin, estimated glomerular filtration rate, lipids and antihypertensive therapy, to grade retinopathy by standard photographic protocols, and to establish temporal sequence. A simple risk score combining haemoglobin, blood pressure and duration should be developed and externally validated before use for triage, and pragmatic trials should test whether correcting anaemia slows the onset or progression of retinopathy.

Diabetic retinopathy was present in 40% of hospital-attending adults with type 2 diabetes mellitus who had normoalbuminuria and normal renal function. Haemoglobin below 8 g/dL and diabetes duration of five years or more were robustly and independently associated with retinopathy; systemic hypertension was also associated, although its independent effect was less certain because it clustered closely with anaemia. The risk rose steeply as these predictors accumulated. Absence of albuminuria should not be a reason to defer retinal screening; instead, anaemia, high blood pressure and longer diabetes duration should be used to identify normoalbuminuric patients at the highest risk who need early fundus examination.

1. Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB, et al. IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract. 2022;183:109119. doi:10.1016/j.diabres.2021.109119 PMID: 34879977 2. Anjana RM, Unnikrishnan R, Deepa M, Pradeepa R, Tandon N, Das AK, et al. Metabolic non-communicable disease health report of India: the ICMR-INDIAB national cross-sectional study (ICMR-INDIAB-17). Lancet Diabetes Endocrinol. 2023;11(7):474-489. doi:10.1016/S2213-8587(23)00119-5 PMID: 37301218 3. Teo ZL, Tham YC, Yu M, Chee ML, Rim TH, Cheung N, et al. Global Prevalence of Diabetic Retinopathy and Projection of Burden through 2045: Systematic Review and Meta-analysis. Ophthalmology. 2021;128(11):1580-1591. doi:10.1016/j.ophtha.2021.04.027 PMID: 33940045 4. Yau JW, Rogers SL, Kawasaki R, Lamoureux EL, Kowalski JW, Bek T, et al. Global prevalence and major risk factors of diabetic retinopathy. Diabetes Care. 2012;35(3):556-64. doi:10.2337/dc11-1909 PMID: 22301125 5. Raman R, Vasconcelos JC, Rajalakshmi R, Prevost AT, Ramasamy K, Mohan V, et al. Prevalence of diabetic retinopathy in India stratified by known and undiagnosed diabetes, urban-rural locations, and socioeconomic indices: results from the SMART India population-based cross-sectional screening study. Lancet Glob Health. 2022;10(12):e1764-e1773. doi:10.1016/S2214-109X(22)00411-9 PMID: 36327997 6. Gurudas S, Vasconcelos JC, Prevost AT, Raman R, Rajalakshmi R, Ramasamy K, et al. National prevalence of vision impairment and blindness and associated risk factors in adults aged 40 years and older with known or undiagnosed diabetes: results from the SMART-India cross-sectional study. Lancet Glob Health. 2024;12(5):e838-e847. doi:10.1016/S2214-109X(24)00035-4 PMID: 38430915 7. Pradeepa R, Anitha B, Mohan V, Ganesan A, Rema M. Risk factors for diabetic retinopathy in a South Indian Type 2 diabetic population--the Chennai Urban Rural Epidemiology Study (CURES) Eye Study 4. Diabet Med. 2008;25(5):536-42. doi:10.1111/j.1464-5491.2008.02423.x PMID: 18346159 8. Rema M, Pradeepa R. Diabetic retinopathy: an Indian perspective. Indian J Med Res. 2007;125(3):297-310. PMID: 17496357 9. American Diabetes Association Professional Practice Committee. 12. Retinopathy, Neuropathy, and Foot Care: Standards of Care in Diabetes-2025. Diabetes Care. 2025;48(1 Suppl 1):S252-S265. doi:10.2337/dc25-S012 PMID: 39651973 10. American Diabetes Association Professional Practice Committee. 11. Chronic Kidney Disease and Risk Management: Standards of Care in Diabetes-2025. Diabetes Care. 2025;48(1 Suppl 1):S239-S251. doi:10.2337/dc25-S011 PMID: 39651975 11. Kumar S, Mohanraj R, Raman R, Kumar G, Luvies S, Machhi SS, et al. 'I don`t need an eye check-up'. A qualitative study using a behavioural model to understand treatment-seeking behaviour of patients with sight threatening diabetic retinopathy (STDR) in India. PLoS One. 2023;18(6):e0270562. doi:10.1371/journal.pone.0270562 PMID: 37319187 12. Penno G, Solini A, Bonora E, Fondelli C, Orsi E, Zerbini G, et al. Clinical significance of nonalbuminuric renal impairment in type 2 diabetes. J Hypertens. 2011;29(9):1802-9. doi:10.1097/HJH.0b013e3283495cd6 PMID: 21738053 13. Kumar R, Singh TP, Jaryal A, Thakur S. Retino-Renal Dissociation in Type 2 Diabetes Mellitus: An Observational Cross Sectional Study at a Tertiary Care Center of Sub-Himalayan Region. J Assoc Physicians India. 2022;70(5):11-12. PMID: 35598130 14. Fitzgerald A, Das R, Moezzi CJ, Salazar SR, Mankad R, Qualls CR, et al. Discordance of diabetic retinopathy severity in a cohort of diabetic nephropathy patients: a cross-sectional case-control study in a new Mexican population of type 2 diabetes. Front Endocrinol (Lausanne). 2025;16:1638415. doi:10.3389/fendo.2025.1638415 PMID: 40822949 15. Rotbain Curovic V, Tofte N, Lindhardt M, Adamova K, Bakker SJL, Beige J, et al. Presence of retinopathy and incident kidney and cardiovascular events in type 2 diabetes with normoalbuminuria - A post-hoc analysis of the PRIORITY randomized clinical trial. J Diabetes Complications. 2023;37(4):108433. doi:10.1016/j.jdiacomp.2023.108433 PMID: 36841085 16. Li YH, Sheu WH, Lee IT. Effects of retinopathy and chronic kidney disease on long-term mortality in type 2 diabetic inpatients with normal urinary albumin or protein: a retrospective cohort study. BMJ Open. 2018;8(7):e021655. doi:10.1136/bmjopen-2018-021655 PMID: 30049696 17. Matthews DR, Stratton IM, Aldington SJ, Holman RR, Kohner EM; UK Prospective Diabetes Study Group. Risks of progression of retinopathy and vision loss related to tight blood pressure control in type 2 diabetes mellitus: UKPDS 69. Arch Ophthalmol. 2004;122(11):1631-40. doi:10.1001/archopht.122.11.1631 PMID: 15534123 18. UK Prospective Diabetes Study Group. Tight blood pressure control and risk of macrovascular and microvascular complications in type 2 diabetes: UKPDS 38. BMJ. 1998;317(7160):703-13. PMID: 9732337 19. Qiao Q, Keinänen-Kiukaanniemi S, Läärä E. The relationship between hemoglobin levels and diabetic retinopathy. J Clin Epidemiol. 1997;50(2):153-8. doi:10.1016/s0895-4356(96)00335-6 PMID: 9120508 20. Davis MD, Fisher MR, Gangnon RE, Barton F, Aiello LM, Chew EY, et al. Risk factors for high-risk proliferative diabetic retinopathy and severe visual loss: Early Treatment Diabetic Retinopathy Study Report #18. Invest Ophthalmol Vis Sci. 1998;39(2):233-52. PMID: 9477980 21. Wang J, Xin X, Luo W, Wang R, Wang X, Si S, et al. Anemia and Diabetic Kidney Disease Had Joint Effect on Diabetic Retinopathy Among Patients With Type 2 Diabetes. Invest Ophthalmol Vis Sci. 2020;61(14):25. doi:10.1167/iovs.61.14.25 PMID: 33351059 22. American Diabetes Association Professional Practice Committee. 2. Diagnosis and Classification of Diabetes: Standards of Care in Diabetes-2025. Diabetes Care. 2025;48(1 Suppl 1):S27-S49. doi:10.2337/dc25-S002 PMID: 39651986 23. Wilkinson CP, Ferris FL, Klein RE, Lee PP, Agardh CD, Davis M, et al. Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology. 2003;110(9):1677-82. doi:10.1016/S0161-6420(03)00475-5 PMID: 13129861 24. Lenfant C, Chobanian AV, Jones DW, Roccella EJ; Joint National Committee on the Prevention, Detection, Evaluation, and Treatment of High Blood Pressure. Seventh report of the Joint National Committee on the Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC 7): resetting the hypertension sails. Hypertension. 2003;41(6):1178-9. doi:10.1161/01.HYP.0000075790.33892.AE PMID: 12756222 25. Heinze G, Schemper M. A solution to the problem of separation in logistic regression. Stat Med. 2002;21(16):2409-19. doi:10.1002/sim.1047 PMID: 12210625 26. Pradeepa R, Anjana RM, Unnikrishnan R, Ganesan A, Mohan V, Rema M. Risk factors for microvascular complications of diabetes among South Indian subjects with type 2 diabetes--the Chennai Urban Rural Epidemiology Study (CURES) Eye Study-5. Diabetes Technol Ther. 2010;12(10):755-61. doi:10.1089/dia.2010.0069 PMID: 20818974 27. Chung JO, Park SY, Chung DJ, Chung MY. Relationship between anemia, serum bilirubin concentrations, and diabetic retinopathy in individuals with type 2 diabetes. Medicine (Baltimore). 2019;98(43):e17693. doi:10.1097/MD.0000000000017693 PMID: 31651899 28. Mao K, Liu R, Jiang P, Cai W, Lin Y, Chen F, et al. Lower hemoglobin levels may increase the risk of diabetic retinopathy in patients with type 2 diabetes mellitus. BMC Endocr Disord. 2025;26(1):20. doi:10.1186/s12902-025-02135-4 PMID: 41444948 29. Liu Y, Meng Y, Duan R, Liu B, Ma Y, Tsai CL, et al. Risk Factors for Diabetic Retinopathy Change With Diabetes Duration: Synergistic Effect of Long Duration and Anemia. J Diabetes Res. 2026;2026(1):e8611325. doi:10.1155/jdr/8611325 PMID: 42087387 30. AlFalasi SM, Abdouli KA, Aldashti NA. Association of Anemia and Diabetic Retinopathy Among Patients With Type 2 Diabetes Mellitus: Retrospective Cross-Sectional Study. Cureus. 2024;16(8):e67995. doi:10.7759/cureus.67995 PMID: 39347127 31. Raghu K, Surya R J, Rani PK, Sharma T, Raman R. Incidence and Progression of Diabetic Retinopathy in Urban India: Sankara Nethralaya Diabetic Retinopathy Epidemiology and Molecular Genetics Study, 15yr Follow up. Ophthalmic Epidemiol. 2024;32(4):428-436. doi:10.1080/09286586.2024.2419015 PMID: 39531593 32. Aiello LP, Avery RL, Arrigg PG, Keyt BA, Jampel HD, Shah ST, et al. Vascular endothelial growth factor in ocular fluid of patients with diabetic retinopathy and other retinal disorders. N Engl J Med. 1994;331(22):1480-7. doi:10.1056/NEJM199412013312203 PMID: 7526212 33. Berman DH, Friedman EA. Partial absorption of hard exudates in patients with diabetic end-stage renal disease and severe anemia after treatment with erythropoietin. Retina. 1994;14(1):1-5. doi:10.1097/00006982-199401000-00001 PMID: 8016453 34. Friedman EA, L'Esperance FA, Brown CD, Berman DH. Treating azotemia-induced anemia with erythropoietin improves diabetic eye disease. Kidney Int Suppl. 2003;(87):S57-63. doi:10.1046/j.1523-1755.64.s87.9.x PMID: 14531775 35. Takao T, Matsuyama Y, Suka M, Yanagisawa H, Kikuchi M, Kawazu S. Time-to-effect relationships between systolic blood pressure and the risks of nephropathy and retinopathy in patients with type 2 diabetes. J Diabetes Complications. 2014;28(5):674-8. doi:10.1016/j.jdiacomp.2014.05.008 PMID: 24996979 36. Rajalakshmi R, Amutha A, Ranjani H, Ali MK, Unnikrishnan R, Anjana RM, et al. Prevalence and risk factors for diabetic retinopathy in Asian Indians with young onset type 1 and type 2 diabetes. J Diabetes Complications. 2014;28(3):291-7. doi:10.1016/j.jdiacomp.2013.12.008 PMID: 24512748 37. Raja SA, Chong VH, Rahman NA, Shakir LMP, Knights J. Prevalence and Associated Factors of Diabetic Retinopathy among Type 2 Diabetes Mellitus Patients in Brunei Darussalam: A Cross-sectional Study. Korean J Ophthalmol. 2022;36(1):26-35. doi:10.3341/kjo.2021.0040 PMID: 34743489 38. Moriya T, Hayashi A, Matsubara M, Suzuki A, Ouchi M. Glucose control, diabetic retinopathy, and hemodialysis induction in subjects with normo-microalbuminuric type 2 diabetic patients with normal renal function followed for 15 years. J Diabetes Complications. 2021;36(1):108080. doi:10.1016/j.jdiacomp.2021.108080 PMID: 34740544.