Background: Confounding-by-indication is a bias in nonexperimental studies that occurs when outcomes are compared for treated and untreated patients and the treatment or medication dose is related to predictors of the outcome. Two recent publications reported that greater epoetin alfa (EPO) doses were associated with increased mortality rates. We assessed whether confounding-by-indication might account for these results.
Study design: We used a retrospective cohort study design.
Setting & participants: Hemodialysis patients were randomly selected from a large dialysis organization from July 2000 to June 2002 and were required to have completed a 9-month baseline period.
Predictor: EPO dose assessed during months 7 to 9 of the baseline period and monthly throughout the follow-up period. Hemoglobin (Hb) was assessed as average value during months 4 to 6 of the baseline period and monthly throughout the follow-up period. All other covariates were assessed during months 1 to 6 of the baseline period.
Outcome: All-cause mortality during the 1 year of follow-up. Baseline Cox models were fitted with log EPO and Hb with and without adjustment for baseline patient characteristics. Time-dependent models were fitted with time-varying log EPO and Hb and, separately, lagged log EPO and Hb, with adjustment for baseline patient characteristics.
Results: 22,955 patients met our inclusion criteria. In the unadjusted model, we observed increased mortality risk with increasing EPO dose (hazard ratio [HR], 1.31 per log unit increase; 95% confidence interval [CI], 1.26 to 1.36). Adjustment for baseline patient characteristics resulted in an appreciably decreased HR (HR, 1.21; 95% CI, 1.15 to 1.28). In the lagged time-dependent analyses, estimates ranged from HR of 0.93 (95% CI, 0.92 to 0.95) to HR of 1.01 (95% CI, 0.99 to 1.03) for the 1- and 2-month lagged models, respectively.
Limitations: This analysis was limited to prevalent hemodialysis patients, and inhospital EPO dosing information was unavailable.
Conclusions: The observed mortality risk estimates associated with EPO dose in nonexperimental studies in dialysis patients may be highly sensitive to the analytic method used. This highlights the complexity of evaluating the association between EPO dose, Hb level, and mortality in these studies.