UACR adjustment attenuates associations of selected routine blood markers with documented diabetic retinopathy more than eGFR adjustment: a cross-sectional study in type 2 diabetes

Background Studies of routine blood markers and diabetic retinopathy often adjust for estimated glomerular filtration rate (eGFR), although albuminuria captures a different aspect of kidney damage. We compared these associations under eGFR and urinary albumin-to-creatinine ratio (UACR) adjustment. Methods We analyzed 8,159 adults hospitalized with type 2 diabetes from January 2020 to March 2026. Primary exposures were serum albumin, total bilirubin, fibrinogen and the FIB-4 index. Four models were fitted in a fixed complete-case sample, with additional adjustment for eGFR, log UACR, or both. The outcome was documented diabetic retinopathy (DR), based on ophthalmological assessment and discharge diagnosis; DR = 0 included unassessed patients without a DR diagnosis. Prevalence ratios were estimated by marginal standardization. Results Among 3,978 complete-case patients, 958 (24.1%) had documented DR. Adjustment for log UACR attenuated the serum albumin association more than adjustment for eGFR (73.3% versus 34.3%; difference 38.9 percentage points, 95% CI 30.7 to 49.0) and moved the fibrinogen estimate close to one. Binary albuminuria produced substantially less attenuation than continuous log UACR. Multiple imputation supported the renal-adjustment comparison, although the albumin interval included the null after mutual renal adjustment. Serum albumin remained associated with documented DR after restriction to eGFR ≥90 mL/min/1.73 m 2 (prevalence ratio 0.762, 95% CI 0.649 to 0.873), whereas no clear evidence of an eGFR association was observed within this stratum (0.981, P = 0.823). Conclusions: In this single-center inpatient cohort, eGFR and UACR adjustment yielded different estimates for the studied blood-marker associations with documented DR. Attenuation denotes proportional reduction in the absolute log prevalence-ratio estimate, not a causal proportion explained. Dichotomizing UACR discarded information and reduced attenuation in these models.