Figure 2:

Point estimates (95% confidence intervals) of hazard ratios from Model B. Hazard ratios in continuous covariates, including long-term glycemia levels and the three estimated glomerular filtration rate (eGFR) constructs, are for hazard differences between the average values in the 4^th versus 1^st quartiles in the analytic sample. Model B adjusted for time-varying diabetic retinopathy duration– (–diagnosis within past 5 years, –diagnosis over 5 years ago), glaucoma and age-related macular degeneration duration, age, gender, self-reported non-Hispanic White (yes/no), carrying at least one APOE ε4 allele, degree beyond high school at Adult Changes in Thought (ACT) study entry (yes/no), type of ACT recruitment cohort, and time-varying smoking status (never vs. current/former), microalbuminuria (worst urine albumin-to-creatinine ratio [ACR] status in past 5 years defined as normal/reference with ACR<30 mcg/mg, microalbuminuria with ACR 30–300 mcg/mg, or macroalbuminuria with ACR >300 mcg/mg), long-term glycemia (average glucose over previous 5 years), and renal function based on eGFR measures (3 eGFR constructs: average eGFR, eGFR standard error, eGFR trajectory).

^asee “Long-term glycemia” in “Study Covariates” section of the Methods for details on how the covariate was defined.

^bSee “Renal function, based on eGFR constructs” in “Study Covariates” section of the Methods for details on how the covariates were defined.