High blood sugar doubles risk of preterm eye disease, dose-response study finds

Extremely premature infants whose blood sugar climbs too high in the first weeks of life face more than twice the risk of developing retinopathy of prematurity, one of the world's leading causes of preventable childhood blindness, according to a new multicenter study from northwest China. The research, published in Frontiers in Pediatrics, followed 308 infants born before 32 weeks of gestation across four neonatal units in the Ningxia Hui Autonomous Region and found that the danger did not merely rise with high glucose levels — it climbed steeply and predictably with every increment of hyperglycemia severity. Retinopathy of prematurity, or ROP, is a neurovascular disease of the developing retina in which abnormal blood vessels proliferate in the eyes of premature infants.

It can progress to retinal detachment and severe visual impairment, and despite advances in neonatal intensive care, its global incidence is increasing as more extremely preterm infants survive. Current treatment relies on laser photocoagulation or injections of anti-VEGF agents into the eye — interventions that are invasive, costly, and carry risks including systemic side effects and disease reactivation. That is why researchers are increasingly focused on identifying modifiable risk factors that could be addressed before the disease ever takes hold.

Neonatal hyperglycemia is a common metabolic disturbance in preterm infants, arising from immature insulin secretion and immature glucose regulation. Previous observational studies had suggested a link between high blood sugar and ROP, but the findings were inconsistent, and many were limited by small sample sizes and inadequate control for critical confounders such as gestational age and birth weight. Few studies had rigorously examined whether ROP risk rises in a dose-dependent fashion with the severity of hyperglycemia.

The new study was designed to fill precisely those gaps. Blood glucose was measured at least three times daily for the first seven days of hospitalization, and hyperglycemia was defined as a venous glucose concentration above 7 mmol/L (126 mg/dL) confirmed on at least two independent measurements. ROP was diagnosed by certified ophthalmologists using wide-field digital fundus imaging, staged according to international classification criteria.

By the end of follow-up, 89 infants — 28.9 percent of the cohort — had developed ROP. The crude contrast was stark: 39.6 percent of the 182 hyperglycemic infants developed the disease compared with 13.5 percent of the 126 normoglycemic infants. But the hyperglycemic infants were also smaller, less mature, and sicker in other ways, with lower birth weights and gestational ages, longer ventilation times, and higher rates of brain hemorrhage and chronic lung disease.

The central question was whether high blood sugar mattered independently of those disadvantages. To answer it, the investigators built a series of multivariable Cox proportional-hazards models, adjusting step by step for perinatal characteristics including sex, gestational age, birth weight, maternal age and body mass index, mode of delivery, multiple birth, antenatal corticosteroids and five-minute Apgar score; then for maternal diabetes, hypertension and thyroid disease; and finally for neonatal complications including intraventricular hemorrhage, patent ductus arteriosus, bronchopulmonary dysplasia, late-onset sepsis and necrotizing enterocolitis. The hazard ratio barely moved across the sequence — from 2.58 down to 2.54 — indicating that hyperglycemia carried an effect that was not simply a proxy for immaturity or illness.

Additional sensitivity models further adjusted for respiratory support variables and glucose infusion rate, and a Firth penalized regression corrected for sparse-data bias, all yielding consistent estimates around 2.4. The most striking finding was the dose-response gradient. When infants were stratified by peak blood glucose, those with peaks between 8.33 and 10 mmol/L (150 to 180 mg/dL) had roughly 2.7 times the ROP risk of normoglycemic infants, while those exceeding 10 mmol/L (180 mg/dL) had roughly 3.6 times the risk.

Across model specifications, each incremental rise in glucose tier corresponded to an approximately 50 to 80 percent increase in average ROP risk, and the trend test was highly significant with no meaningful deviation from linearity. Infants in the mildest hyperglycemia band, just above 7 mmol/L, showed no significant excess risk, which points to a candidate vigilance threshold of 8.33 mmol/L. The biology behind such a gradient is plausible, the authors note.

Persistently elevated glucose can promote the accumulation of advanced glycation end-products and induce oxidative stress, which in turn upregulates vascular endothelial growth factor and drives the aberrant retinal neovascularization that defines ROP. Consistent with a broader vascular effect, hyperglycemia was also independently associated with intraventricular hemorrhage in this cohort, with an adjusted odds ratio of 2.68 that survived correction for multiple testing. Associations with bronchopulmonary dysplasia and patent ductus arteriosus were attenuated after adjustment, and no significant links emerged with late-onset sepsis or necrotizing enterocolitis.

The study's limitations are substantial and the authors are candid about them. As a retrospective analysis, it cannot establish causation, and intermittent hyperglycemic episodes may have been missed by the fixed three-times-daily monitoring schedule. The requirement for at least four weeks of hospitalization introduced possible survivor bias, since the sickest infants who died or were transferred early are absent from the cohort.

Insulin therapy was not systematically documented, continuous glucose monitoring data were unavailable, and with 89 ROP events spread across models containing up to 24 covariates, some degree of overfitting cannot be excluded. Severe ROP, defined as stage 3 or higher, occurred in only 13 infants, and its exploratory analysis showed no statistically significant association with hyperglycemia after penalized adjustment — a result the authors attribute to limited statistical power. The cohort also came exclusively from one Chinese region, so the thresholds await external validation.

Even so, the findings carry practical weight for neonatal units. The authors suggest that infants whose peak glucose exceeds 8.33 mmol/L be prioritized for more frequent ROP screening and closer glycemic surveillance, and they argue for a shift from passive monitoring of hyperglycemia toward proactive, protocol-driven glucose management. They stop short, deliberately, of recommending a specific insulin initiation threshold, noting that insulin in preterm infants carries a well-recognized risk of iatrogenic hypoglycemia and that the clinical benefit of any vigilance threshold requires prospective validation.

Future studies integrating continuous glucose monitoring, they conclude, are essential to refine glycemic targets and to determine whether tighter metabolic control can genuinely protect the developing retinas of the smallest infants.

Subject of Research: Neonatal hyperglycemia as an independent and dose-dependent risk factor for retinopathy of prematurity in extremely preterm infants

Article Title: Neonatal hyperglycemia as an independent risk factor for retinopathy of prematurity: a multicenter retrospective cohort study with dose-response analysis

Article References: Wang, Y., Wang, X., Yin, H., Xu, F., Xu, W., & He, J. (2026). Neonatal hyperglycemia as an independent risk factor for retinopathy of prematurity: a multicenter retrospective cohort study with dose-response analysis. Frontiers in Pediatrics, 14, Article 1975596. https://doi.org/10.3389/fped.2026.1975596

Image Credits: AI Generated

DOI: 10.3389/fped.2026.1975596

Keywords: retinopathy of prematurity, neonatal hyperglycemia, preterm infants, Cox regression, dose-response, retinal neovascularization, intraventricular hemorrhage, glycemic control, neonatology, childhood blindness, retrospective cohort, Neonatal