Table 4

Associations of time-dependent covariates with risk of further three or more–step progression from DCCT closeout or new PDR in EDIC and the impact of adjusting for each covariate on the DCCT intensive diabetes treatment effect

Time-dependent covariate(s)Further ≥3-step progressionPDR
Risk associated with the covariateEffect of DCCT intensive diabetes therapyRisk associated with the covariateEffect of DCCT intensive diabetes therapy
Hazard ratio (95% CI)P valueRisk reduction (%) (95% CI)P value% explained by covariateHazard ratio (95% CI)P valueRisk reduction (%) (95% CI)P value% explained by covariate
Baseline adjusted model*46 (36, 54)<0.000146 (29, 59)<0.0001
HbA1c (per 10% increase)0 (−28, 22)0.9821100%−1 (−51, 32)0.9555100%
 DCCT mean1.17 (1.07, 1.29)0.00061.22 (1.06, 1.42)0.0063
 Updated EDIC mean1.57 (1.46, 1.68)<0.00011.66 (1.49, 1.85)<0.0001
Hyperlipidemia vs. not§1.39 (1.16, 1.67)0.000342 (31, 52)<0.000137%1.39 (1.06, 1.82)0.01844 (24, 58)0.000124%
Sustained AER >30 mg/24 h vs. not1.79 (1.42, 2.25)<0.000144 (34, 53)<0.000110%2.54 (1.93, 3.34)<0.000139 (19, 53)0.000640%
Currently smoking vs. not1.34 (1.10, 1.61)0.002846 (36, 54)<0.00010%1.27 (0.94, 1.71)0.1246 (29, 59)<0.00011.6%
Mean arterial blood pressure1.02 (1.01, 1.03)<0.000147 (37, 55)<0.00010%1.04 (1.02, 1.05)<0.000146 (30, 59)<0.00010%
RAAS inhibitor use0.96 (0.76, 1.21)0.7046 (36, 54)<0.00010.1%1.34 (1.00, 1.81)0.05245 (28, 59)<0.00012.8%
  • * Basic Weibull proportional hazards models evaluated the associations of DCCT treatment group with risk of further three or more–step progression or new PDR in EDIC, respectively, after adjustment for diabetes duration, HbA1c at DCCT entry, and retinopathy level at DCCT closeout. Risk reduction associated with intensive diabetes therapy is calculated as (1 − hazard ratio of intensive vs. conventional diabetes therapy with or without adjustment for covariate) × 100%.

  • Separate Weibull models evaluated associations of each time-dependent covariate with risk of retinopathy progression or new PDR to generate each covariate hazard ratio. Hazard ratio for covariates is evaluated per 10% increase in DCCT or EDIC HbA1c (e.g., from 8 to 8.8%), unit change in other quantitative covariates, or status change in binary covariates. DCCT mean HbA1c (a fixed covariate) and EDIC mean HbA1c (a time-dependent covariate) modeled together.

  • Additional models assessed the interaction between the covariate and the DCCT/EDIC weighted mean HbA1c. None was significant at the 0.05 level.

  • Separate Weibull models then evaluated the risk reduction with DCCT intensive vs. conventional diabetes therapy on risk of retinopathy progression or new PDR in EDIC, adjusting for each time-dependent covariate one at a time in addition to the covariates adjusted in the basic model (and modeling DCCT HbA1c along with the EDIC HbA1c). The percent of the DCCT treatment effect explained by group differences in each covariate is computed as the percentage change in the DCCT treatment group χ2 test value from the basic treatment effect model to the model adjusted for the time-dependent covariate. Since each covariate is evaluated in a separate model, the proportions do not sum to 100%.

  • § Hyperlipidemia was defined by an LDL cholesterol level ≥130 mg/dL (3.4 mmol/L) or the use of lipid-lowering agents.