The study team analyzed teleophthalmology records to determine how frequently early-stage NPDR improved, remained stable, or worsened over a follow-up window of 6–36 months. Patients initially classified with very mild NPDR most often regressed: 47% regressed to no DR (95% CI 0.45–0.50). By contrast, 27% of that group showed any progression. For patients who began with mild NPDR, outcomes were more mixed — 36% experienced any regression (to very mild NPDR or to no DR), while 34% showed progression. The cohort therefore exhibited greater reversal than progression for the very mild NPDR subgroup, and a roughly balanced pattern among those starting at mild NPDR. Investigators reported these proportions with multinomial 95% confidence intervals and tested differences in outcome distributions with chi-square tests (p < .0001 for both baseline-severity groups).
The study team used multivariable logistic regression to examine independent associations between background factors and DR regression (versus all other outcomes). Data indicated that higher HbA1c values were associated with lower odds of regression in both baseline-severity groups; the source provided odds ratios per percentage point increase in HbA1c (reported in the article). Insulin therapy at baseline was also associated with reduced odds of regression compared with oral agents only. Shorter diabetes duration increased odds of regression, but this association was limited to the very mild NPDR subgroup in adjusted models. Age showed a small positive association with regression for the very mild NPDR group in the multivariable model. The authors explicitly noted that they did not interpret insulin use as a causal factor in worsening outcomes, but rather as a marker that may reflect longer disease severity or other clinical differences recorded in the medical record.
This retrospective cohort used de-identified medical records from the Indian Health Service (IHS) teleophthalmology program. Baseline imaging dates were limited to calendar year 2015 and follow-up exams occurred between 2016 and 2019. Eligible patients were aged 20 years or older with type 2 diabetes who had a baseline diagnosis of either very mild NPDR (ETDRS level 20) or mild NPDR (ETDRS level 35) in one or both eyes. The program’s graders are certified optometrists using standardized protocols tied to ETDRS clinical levels, with ophthalmologist oversight.
More than 80% of examinations used nonmydriatic ultra-widefield imaging (UWFI) with stereo pairs centered on the macula; the remainder used nonmydriatic fundus photography (NMFP) with multiple stereo fields to achieve approximately 90°–135° retinal coverage. The outcome for very mild NPDR patients was categorized by worst-eye status as regression to no DR, no change, progression to mild NPDR, or progression to more advanced DR. For mild NPDR patients, outcomes were regression to no DR, regression to very mild NPDR, no change, or progression. The source reported that 6.9% of follow-up exams could not be graded and were therefore excluded from outcome calculations.
Investigators presented unadjusted proportions and means for background measures and then modeled DR regression (binary outcome) with covariates including age, sex, diabetes therapy, diabetes duration, and HbA1c. Significance was set at p < .01 to limit Type I error. Model outputs included adjusted odds ratios and predicted probabilities at selected HbA1c levels while holding age at 60, diabetes duration at 10 years, and sex as female. The reported model-based probabilities illustrate a consistent decline in the likelihood of regression as HbA1c increased, and lower predicted probabilities for patients on insulin compared with those on oral agents only at the same HbA1c.
For example, the source reported that at an HbA1c of 8% the predicted probability of regression was higher for very mild NPDR patients treated with oral agents (reported value in the article) than for those using insulin; predicted probabilities were lower across the board for the mild NPDR group compared with very mild NPDR at the same HbA1c.
The authors highlighted several limitations inherent to the dataset and study design. This was a retrospective analysis of routine teleophthalmology records, so follow-up was determined by available clinical exams rather than a prespecified research schedule; only 53.7% of the initial very mild NPDR cohort and 51.5% of the initial mild NPDR cohort had a graded follow-up exam within 6–36 months. The source reported substitution of follow-up or cohort-mean values when baseline HbA1c was missing, a potential source of measurement bias. A subset of follow-up images (6.9%) could not be graded and were excluded from outcome calculations. The follow-up window was limited to a maximum of 36 months, so longer-term patterns of regression or delayed progression were not captured in this analysis. Finally, the cohort was drawn from a primary care–based IHS teleophthalmology program; the authors reported these program characteristics and grading processes explicitly, and they did not claim broader generalizability beyond similar screening populations.
Beyond these documented limitations, the source left unanswered clinical questions relevant to practice. The study did not report specific behavioral or medication-change interventions that accompanied regression, and it did not establish causal pathways linking glycemic control or therapy type with anatomical DR changes. Future prospective work with standardized intervals, complete glycemic data, and prespecified behavioral measures would be required to determine whether interventions that lower HbA1c directly increase the probability of DR regression in early-stage NPDR.