Data completeness as an unmeasured confounder in dialysis facility performance comparison with 1-year follow-up .

Abstract

Using the United States Renal Data System ESRD database, we included patients who initiated hemodialysis July 1 - December 31 in each of the years 2006-2010, had Medicare as primary payer, were aged ≥66 years, and had no prior transplant. Patients were followed from dialysis initiation to the earliest of death, transplant, modality change, or 1 year. SMRs and SHRs were calculated for for-profit/non-profit and rural/urban facilities for ME-based and claims-based comorbidity, separately. Cox models were used for expected number of deaths and piecewise Poison models for expected number of hospitalizations. Comorbidity agreement was measured by κ-statistic. Testing of differences between ME-based and claims-based SMRs/SHRs was performed by bootstrap. In all, 73,950 incident hemodialysis patients were included. κ-values for comorbidity agreement were low, <0.5, except for diabetes (0.77). Percentages of claims-based comorbidity were similar for for-profit and non-profit facilities; ME-based comorbidity was lower for for-profit facilities. Differences between ME-based and claims-based SMRs/SHRs were statistically significant. Compared with ME-based SMRs/SHRs, claims-based ratios decreased 0.9/0.6% for for-profit and 1/0.7% for urban facilities and increased 3.4/2.8% for non-profit and 5.9/4.1% for rural facilities. Comorbidity data source may affect performance evaluation. The impact is larger for smaller groups. .