depicts the selection process for patients included in the analysis. The analysis for the death only outcome included 35,593 patients, of whom 19,034 (53%) were diabetic. During the follow-up period, 8,238 (23%) of the patients were censored and 7,179 (20%) died. The analysis for the composite outcome included 32,534 patients (after 3,059 patients were excluded for having had a stroke, MI, or CHF during the study baseline period), of whom 17,387 (53%) were diabetic. 6,933 patients (21%) were censored; 8,512 (26%) had a composite event including 4,868 deaths (57%), and 3,644 (43%) cardiovascular events. Of the 3,644 cardiovascular hospitalization events, 2,117 (24.7%) were for CHF, 726 (8.5%) were for stroke, and 801 (9.5%) were for an MI. The event rates for diabetics and nondiabetics were similar for both mortality and composite outcomes (data not shown).
Compared with nondiabetics, diabetics were younger, more likely to be male and black, had a more severe comorbidity burden (higher Charlson score) with almost three times as many cardiovascular comorbidities, and had an increased likelihood of being hospitalized and for longer periods of time (). Although diabetic patients were more likely to receive predialysis epoetin therapy, they received similar amounts of epoetin during the baseline period and achieved similar hematocrit levels at the end of baseline when compared to nondiabetic patients.
Characteristics of study population (N = 35,593).
shows the adjusted survival curves under three hypothetical epoetin dose levels: 15,000 U/wk, 30,000 U/wk, and 45,000 U/wk throughout the entire follow-up period and for each outcome separately ( shows death only and shows the composite outcome). The survival decreased with increasing doses. For the lowest epoetin doses of 15,000 U/wk, the 9-month risk of death was 20% (28% for composite endpoint) among diabetics and 24% (31%) among nondiabetics.
Figure 2 (A) Survival outcome probabilities for three selected epoetin dosage regimens: Low dosage (15,000 U/wk), medium dosage (30,000 U/wk), and high dosage (45,000 U/wk), based on the primary analysis which imputes epoetin for patients with a hospital stay (more ...)
The difference in mortality risk between 30,000 units/week and 15,000 units/week was 9% (95% CI 7%, 11%) among diabetics and 5% (95% CI 2%, 7%) among nondiabetics (p-value = 0.04 for heterogeneity between diabetics and nondiabetics). The difference in mortality risk between 45,000 units/week and 15,000 units/week was 13% (95% 10%, 16%) among diabetics compared to 5% (95% CI 2%, 9%) among nondiabetics (p-value = 0.002).
The difference in the risk of a composite endpoint between 30,000 units/week and 15,000 units/week was 6% (95% CI 3%, 8%) among diabetics and 5% (95% CI 2%, 8%) among nondiabetics (p-value = 0.57 for heterogeneity between diabetics and nondiabetics). The difference in mortality risk between 45,000 units/week and 15,000 units/week was 9% (95% 5%, 12%) among diabetics compared to 5% (95% CI 1%, 8%) among nondiabetics (p-value = 0.19).
presents the estimated average hazard ratios (HRs) for both mortality and the composite outcome. The estimates were consistent with the findings from the survival curves. Among both diabetics and nondiabetics, lower epoetin dose levels (less than 20,000 U/wk) were associated with lower risk for both mortality and composite outcomes. Compared with a dose of 20,000 – 30,000 U/wk, the hazard ratio of mortality was 1.32 (95% CI 1.11, 1.58) and the hazard ratio of a composite endpoint of death and cardiovascular events was 1.26 (95% CI 1.07, 1.50) among diabetic patients exposed to the highest ESA doses (>40,000 U/wk). The corresponding hazard ratios in non-diabetic patients were 1.06 (95% CI 0.88, 1.28) for mortality and 1.10 (95% CI 0.92, 1.32) for composite end point.
Cumulative average epoetin dose and hazard ratios (HR) based on inverse probability weighting.
In secondary analyses the estimates were similar when we made different assumptions about the amount of epoetin used during hospitalizations (). Our estimates did not materially change when we used: other summaries of epoetin use (i.e., total cumulative dosage from start of follow-up, recent and past cumulative average dosage) in the logistic model; cubic splines with knots at different locations; IP weights that were not truncated; IP weights were estimated under a gamma-distribution for the log of epoetin dosage; alternative categorizations of hematocrit values; expanded billable service (or claims) gap definition from 30 to 60 days; censoring that did not include change of provider; and censoring criteria that included reduced dialysis sessions.