Recommendations for future research

The present analysis characterized the association between vitamin D exposure and fracture risk but it was not designed to establish causal inference. A robust clinical trial can confirm the validity of our observed associations but these trials would likely be limited by under-ascertainment of fracture events and questions regarding clinical equipoise. The impracticability of randomized controlled trials warrants the use of robust non-experimental studies to address this salient issue. The following text proposes research questions,

corresponding study designs, and data sources that could be employed in non-experimental studies to substantiate our results and address the current gaps in the nephrology literature. In the absence of clinical trial evidence, the grouped-treatment approach extends the natural experiment methodology commonly found in epidemiology by taking advantage of variations in practice policies to estimate the marginal effect of differences in treatment

selection.218 As in this dissertation, the grouped-treatment approach (where the main variable of interest is measured ecologically while outcomes are measured at the individual-level) is advantageous in epidemiological studies where individual-level factors like biochemical parameters that determine the decision to prescribe a medication are not readily available in the dataset of interest. An individual-level analysis using a data source with adequate laboratory and biochemical measures could be conducted to substantiate whether there is indeed no association between vitamin D exposure and fracture risk among hemodialysis patients. To this end, one could employ the approach used by Block and colleagues to investigate cardiovascular outcomes among hemodialysis patients. Data from DaVita Inc., the second largest dialysis provider in the United States, could be linked using unique patient identifiers to the Centers for Medicare & Medicaid Services (CMS) ESRD database by the United States Renal Data System (USRDS) through a data licensing agreement.219

Unlike using USRDS data alone, merged DaVita and USRDS data contain detailed laboratory values, IV medication use, home medication use, and vascular access information from DaVita while simultaneously providing relevant Medicare claims and hospitalization data from the USRDS for each patient.219 The DaVita dataset provides two distinct

advantages. Firstly, the data contain the important laboratory values of calcium, phosphorus, and PTH levels that are used to guide vitamin D administration. Secondly, merged DaVita and USRDS data would allow researchers to measure exposure to cinacalcet, an oral calcimimetic also used to treat SHPT. Our analysis was restricted to the years prior to the widespread use of cinacalcet to allow us the ability to ascertain the association between vitamin D and fracture risk without the effects of confounding from the administration of a drug also indicated to treat SHPT. A retrospective cohort study using the most recent years

of available data could be conducted to assess the effect of vitamin D exposure on fracture risk with patient-level vitamin D prescription treated as a time-dependent variable. Vitamin D exposure and covariates (e.g., baseline comorbidity, laboratory data, cinacalcet and phosphate binder use) could be measured during a 6-month baseline period. Adjusted time- dependent Cox proportional hazard regression models could be employed to assess 3-year fracture risk and eligible patients would be censored in the event of transfer out of DaVita, renal transplantation or loss to follow-up.

Furthermore, a natural extension of the recommended study described above could be an investigation exploring the comparative effectiveness of IV vitamin D formulations with respect to fracture outcomes. In this dissertation, we report the overwhelming use of paricalcitol between 1999 and 2008. However, trends have changed in recent years and contemporary data of hemodialysis patients contend that doxercalciferol and paricalcitol are now being administered at relatively equal frequency, with calcitriol use now virtually non- existent.183 It is unknown whether the pharmacological differences between these

formulations translate into differential effects on important clinical outcomes like fracture risk. Using the aforementioned, linked USRDS and DaVita data, the comparative

effectiveness of paricalcitol versus doxercalciferol with respect to fracture risk could be assessed with a retrospective cohort study. A cohort of incident hemodialysis patients could be selected under the new user design. Following a 3-month waiting period for claim

ascertainment post-dialysis initiation, vitamin D exposure (use and dosage) could be measured over a 6-month baseline period for patients treated exclusively with either

paricalcitol or doxercalciferol. Patients could be followed over 1 or 3 years and censored if they died, switched to another formulation, switched dialysis facilities, or underwent renal

transplantation. Laboratory values could be averaged over the 6-month baseline period and controlled for within Cox proportional hazard regression models along with baseline

comorbidity and clinical attribute data.

Cinacalcet, also used to treat SHPT among dialysis patients, is currently covered under Medicare Part D but will be included under the new bundle as of January 2014. Given that financial incentives may compel the substitution of IV vitamin D for the cheaper

cinacalcet for certain patient populations in 2014, future studies are needed to investigate the comparative efficacy and safety of vitamin D versus these various therapeutic options with respect to fracture outcomes. The secondary data needed to explore these issues will take years before becoming available to researchers and confirmatory studies will be required to ensure that the bone health of patients with ESRD is not compromised under the new payment system. For now, the independent effect of cinacalcet on fracture risk could be assessed using the methodology published by Frankenfield and colleagues to allow for comparison with studies of vitamin D exposure and to inform future analyses post the 2014 reimbursement policy changes.220

Using CMS Medicare Part D data linked to USRDS files, a retrospective cohort study could be conducted with a point-prevalent cohort of adult hemodialysis patients alive

between July 1, 2006 and December 31, 2006, the latter months of the calendar year of Medicare Part D’s initiation. Patients would then be followed from December 31, 2006 until death, renal transplantation or the last day of available data. Cinacalcet exposure could be defined as a dichotomous, time-dependent variable indicating the presence or absence of a cinacalcet prescription during the study period. Time-varying Cox proportional hazard regression models would assess the effect of cinacalcet prescription on fracture risk with

adjustment for baseline characteristics along with time-varying laboratory and IV vitamin D use.

Lastly, potentially inappropriate use of vitamin D therapy has been observed in an internationally representative sample of dialysis patients where investigators found that vitamin D was potentially overused in up to 46% of patients with low PTH (concentration <100 pg/mL) and potentially underused in up to 34% of patients with high PTH

(concentration >400 pg/mL).221 To understand the appropriateness of vitamin D

administration to dialysis patients and to elucidate the relationship between vitamin D use and fracture outcomes, the nephrology community must ensure that treatment decisions are based on current, reliable evidence.