The analysis of data from a retrospective cohort study used records from the European Clinical Database (EuCliD), a registry of medical data from HD patients at EU-FME dialysis centres, methodological details of which have been previously described
. Briefly, EuCliD contains detailed patient-level information on medical and drug history for all HD patients registered with the EU-FME, including records of key biochemical measurements and medications, and data on patient diagnoses, hospitalisations, and death codes based on the World Health Organization’s International Classification of Diseases, 10th Revision (ICD-10) coding scheme. The present analysis comprised EuCliD data for randomly selected patients who underwent HD between January 1, 2005 and December 31, 2006 at participating EU-FME dialysis facilities in 10 countries: Czech Republic, France, Hungary, Italy, Poland, Portugal, Slovak Republic, Slovenia, Spain, and Turkey. Selection criteria and characteristics of this source population have been reported elsewhere
. Patients attending centres where the majority of data on key dialysis parameters were missing were excluded, as were patients from the UK because information on medication use was not available for UK centres
. Patient data were anonymised, and informed consent was obtained from all patients by EU-FME
The cohort study did not include any human or animal experimentation for which ethical approval was required. All ethical and regulatory obligations concerning the use of patient data were met at each participating European Fresenius Medical Care study site. Because our analysis used existing data, there were no risks posed to human life. All patient information had been anonymised by Fresenius Medical Care prior to transfer of data for analysis. Permission was received from Fresenius Medical Care to use the database for this analysis. Fresenius Medical Care data were provided through a data license agreement which covered observational research use.
To be eligible for the present analysis, patients were required to have at least one iPTH value recorded, and at least 1 month of follow-up after a 3-month baseline period during which SHPT parameters (iPTH, total calcium, and phosphate) were assessed. The baseline period was defined as the 3-month period starting on the date of the first recorded iPTH measurement. The SHPT parameters were defined as the mean value of the above laboratory measures taken during the 3-month baseline period. Since these parameters fluctuate over time
, use of mean values during the 3-month period rather than at a single assessment helped to minimise potential exposure misclassification. Time at risk was defined as the time after the 3-month baseline period until death, successful renal transplantation, loss to follow-up, or the end of follow-up (December 31, 2006).
Healthcare resource utilisation outcomes for this study included SHPT-related hospitalisations and medications. Hospitalisations potentially related to SHPT were defined as those associated with CVD, fracture, and parathyroidectomy (PTX). EuCliD records for the selected patients were used to calculate both the rate per 100 patient-years and the 12-month cumulative incidence of relevant hospitalisations, identified on the basis of ICD-10 codes as listed in Additional file
1: Table S1. CVD-, fracture-, and PTX-related hospitalisations were assessed separately and also combined as all–SHPT-related hospitalisations.
SHPT-related medications identified in the database included phosphate binders, oral vitamin D sterols, and Mimpara® (cinacalcet). Drugs for treating diabetes and CVD were also assessed separately. Included drugs are listed in Additional file
2: Table S2. Medication outcomes were number of days of treatment per patient-year and percentage of patients prescribed these medications.
Economic outcomes were the cost per month of CVD-, fracture-, and PTX-related hospitalisations, as well as the monthly cost of medications associated with treatment of SHPT, diabetes, and CVD. Diabetes and CVD medications were included because both comorbidities are associated with elevated SHPT parameters
[13, 22] and with morbidity and mortality in ESRD patients on HD
[16, 23]. The cost of each episode of hospitalisation related to CVD, fracture, or PTX was calculated by mapping the relevant ICD-10 code to country-specific diagnosis-related groups (DRGs) and their associated costs. French DRGs were used for costing for Turkey, where national DRG costs were not yet available. French DRGs were chosen over other nations’ for this purpose because they offered the combination of being available online, well documented, and weighted by different diagnoses. Costs of each episode of medication use were computed by mapping the medication and prescribed dose to an associated cost using national price lists for each country. Cost analyses included costs for both SHPT-related and non–SHPT-related (i.e., diabetes- and CVD-related) drug use. Hospitalisation and medication price lists were obtained for 2006 wherever available, with missing prices retrieved for other years and adjusted for inflation to 2006 Euros using the healthcare component of the Consumer Price Index.
Outcomes were stratified by baseline iPTH, total calcium, and phosphate. Values for each of these SHPT parameters were divided into clinically relevant categories, using the K/DOQI target range for each parameter for CKD Stage 5 as the reference category: 150–300 pg/mL for iPTH, 2.10–2.37 mmol/L for total calcium, and 1.13–1.78 mmol/L for phosphate
. Target ranges from the 2003 K/DOQI guidelines
 were used rather than those from the 2009 KDIGO guidelines
 because the former were followed during the study period (2006).
All countries were included in the analyses of healthcare resource utilisation. The cost analyses comprised patients only from the following five countries with ≥ 1000 patients enrolled in EU-FME/EuCliD: Hungary, Italy, Portugal, Spain, and Turkey.
Hospitalisation rates were calculated by summing the number of relevant hospitalisation episodes across all patients and dividing by the person-time at risk for all patients. The 95% confidence interval was calculated assuming a Poisson distribution.
One-year cumulative incidence of hospitalisation (all-cause, as well as related to CVD, fractures, or PTX) was obtained with the Kaplan–Meier method and 95% confidence intervals using the arcsine-square root transformation. Data for patients not hospitalised were censored at the last known follow-up date or the end of follow-up (December 31, 2006).
The number of days of treatment per patient-year with phosphate binders, oral vitamin D sterols, and cinacalcet was calculated by summing the number of days when these medications were received during the time at risk and dividing by the person-time at risk.
The total cost per month was defined for each patient as the sum of costs for hospitalisations (CVD-, fracture-, and PTX-related) and medications (for SHPT, diabetes, and CVD) divided by the number of months at risk.
Univariate and multivariate analyses exploring the association of baseline SHPT parameters with cost per month were performed using generalised linear models (GLM) with a gamma probability distribution and a log link function. To account for patients with zero costs in the analyses, €1 was added to the total cost per month.
The univariate analysis for each SHPT parameter was adjusted for the duration of follow-up. The multivariate analysis included all SHPT parameters and duration of follow-up, and was also adjusted for the following potentially confounding factors: age, sex, CKD aetiology, history of CVD, diabetes and cancer, dialysis vintage, C-reactive protein, albumin, haemoglobin, ferritin, total cholesterol, and blood leukocyte count.
Results are presented for observed data, with no imputation of missing values. All statistical analyses were performed using SAS® (version 9.1, Cary, NC).
To assess heterogeneity in duration of follow-up times between patients, a sensitivity analysis was conducted to compare results for patients with fixed periods of follow-up, defined as starting after the baseline period. The multivariate GLM in the main analysis was repeated for patients with a fixed follow-up time of 3 months, and again for patients with a fixed follow-up time of 6 months. These two sensitivity analyses included only patients with at least 3 and 6 months of follow-up, respectively, and in the analyses their follow-up time was truncated to these fixed periods.
In addition to the main analysis based on a one-part model, a two-part multivariate model was fitted in order to assess if the choice of the €1 additive factor in the cost per month affected the results. A two-part model is a mixture model in which the first part models the probability of a patient having a cost and the second part estimates the cost given a patient has a cost. The overall cost is the product of the first-part estimated probability and the second-part predicted conditional cost
Lastly, to account for the presence of censored cost data, the weighted linear regression model proposed by Lin
 was fitted. The study period was partitioned into several time intervals in order to include complete interval cost information where available. This approach aims to recreate the pseudo-population that would have been observed had no censoring occurred.