Data sources
We merged two different data sources to conduct this study: the DaVita Clinical Data Warehouse and the United States Renal Data System (USRDS). Permission to merge these files was obtained from the project officers of the National Institute of Diabetes and Digestive and Kidney Diseases. We obtained laboratory data from the DaVita Clinical Data Warehouse and demographic, comorbidity, dialysis facility, hospitalization, and outcomes data from the USRDS database.
Study population and design
We derived the study cohort from the source population of all DaVita in-center hemodialysis facilities from September 1, 2009, through December 31, 2010, reflecting the most contemporary 16-month period of data available for the linked DaVita-USRDS database.
Eligible facilities had at least 16 months of data during the study period to allow for a 4-month baseline period (September 1, 2009, through December 31, 2009) and a 12-month follow-up period (January 1, 2010, through December 31, 2010). Additionally, we required facilities to have had the same ownership for at least 1 year before the index date and to care for at least ten patients who met all patient-level eligibility criteria. The study cohort consisted of point prevalent hemodialysis patients derived on January 1, 2010, from the eligible facilities. Eligible patients: (1) were aged ≥ 18 years; (2) were alive on the index date; (3) were on hemodialysis for at least 1 year before the index date; (4) received care at their respective facilities during the entire 4-month baseline period; (5) were continuously enrolled in Medicare in 2009; and (6) had at least one value for each CKD-MBD-related biochemical parameter (PTH, calcium, and phosphate) during the baseline period (Fig. 1). Patients were followed from January 1, 2010 (index date), until the earliest of death, kidney transplant, change in provider or modality, Medicare disenrollment, loss to follow-up, or end of study.
Exposure, outcome, and other measurements
The proportion of patients at each facility with at least two of three CKD-MBD biochemical parameters out of target was our primary exposure of interest. We chose this definition based on prior patient-level work by Danese et al. [3] demonstrating that at least two CKD-MBD biochemical parameters out of target best maximized identification of patients at risk of adverse clinical outcomes and minimized identification of patients not at risk. Consistent with Danese et al. [3], we considered biochemical parameters to be out of target if they were above or below the following pre-defined target ranges during the 4-month baseline period: PTH, 150–600 pg/mL; calcium, 8.4–10.2 mg/dL; phosphate, 3.5–5.5 mg/dL. For each biochemical parameter, we considered the average value over the baseline period.
We assigned each dialysis facility a single CKD-MBD composite score, calculated as the proportion of patients at that facility with at least two of three CKD-MBD biochemical parameters out of target during the baseline period. Facilities were then categorized into five groups based on quintiles of the distribution of the facility-level CKD-MBD composite score. The exposure variable for patients at each facility was defined as the quintile group for that facility. As such, patients within each facility were assigned the same quintile score, regardless of their individual, patient-level CKD-MBD composite score. Only patients who contributed to the determination of facility eligibility were considered in the calculation of the facility-level CKD-MBD composite score. However, we allowed all eligible patients at a facility to contribute events and person-time for descriptive and modeling analyses.
In secondary analyses, we evaluated two alternative exposures. First, we redefined our primary exposure using an above-target composite score, defined as the proportion of patients at each facility with at least two of three CKD-MBD biochemical parameters above the pre-defined target ranges during the 4-month baseline period. Second, we ascertained our exposures using a PTH range of 150–300 pg/mL to define out-of- or above-target ranges for this laboratory measure; target ranges for calcium and phosphate did not change.
The primary outcome was the first occurrence of a cardiovascular hospitalization or death. Secondary outcomes included death and parathyroidectomy, separately. The time at risk for each outcome was independently calculated for the primary endpoint and for each secondary endpoint. For cardiovascular hospitalizations, we used an algorithm from the USRDS Annual Data Report that requires the presence of one of the following primary diagnosis codes for the hospitalization: 394–398.99, 401–405, 410–420, 421.9, 422.90, 422.99, 423–438, and 440–459. For parathryoidectomy, we used a previously published approach requiring International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) procedure code 06.8x in any position on the hospital discharge claim [5].
We assessed patient-level covariates including demographic factors, time on dialysis, cause of renal failure, access type, body mass index (BMI), length of hospital stay, and comorbid conditions as of the index date. We considered a comorbid condition as being present if at least one inpatient, home health, or skilled nursing facility claim, or at least two outpatient or physician/supplier claims separated by 90 days, were found with the corresponding ICD-9-CM diagnosis code during the 12 months before the index date [6]. Facility-level covariates included facility size and geography.
Statistical analyses
Patient characteristics and facility CKD-MBD composite score quintiles were examined using descriptive statistics for continuous (mean, standard deviation [SD]) and categorical (percentage [%]) variables.
In the primary analysis, we fitted Poisson regression models using generalized estimating equations to estimate the association between facility-level CKD-MBD composite score (categorized as quintiles) and the patient-level risk of the composite outcome (cardiovascular hospitalization or death) during the 1-year follow-up period. We used an independent correlation structure and robust standard error estimates. We provide crude and adjusted relative risk (RR) estimates and 95 % confidence intervals (CIs); models were adjusted for patient characteristics, comorbid conditions, hospital days, and facility characteristics. In secondary analyses, we fitted Poisson regression models using generalized estimating equations to estimate the association between facility-level CKD-MBD composite score and the risk of death and parathyroidectomy, separately. We evaluated this same series of relations using the above-target composite score for defining the exposure. Finally, we replicated this series of analyses (out of target and above target) using the Kidney Disease Outcomes Quality Initiative (KDOQI) definition for PTH (150–300 pg/mL). Quintile 1 (facilities with the lowest proportion of patients with out-of- or above-target values) was the reference category for all analyses.
All analyses were conducted using SAS software version 9. (Cary, NC).