Vitamin D Receptor Activator Use and Cause-specific Death among dialysis Patients: a Nationwide Cohort Study using Coarsened Exact Matching

Vitamin D receptor activators (VDRA) may exert pleiotropic effects on cardiovascular disease, malignancy, and infections among dialysis patients, but recent studies have mainly focused on cardiovascular outcomes. Among 8,675 patients who started dialysis in 2007 and who survived until January 1, 2010, listed in the Renal Data Registry of the Japanese Society for Dialysis Therapy, 5,365 VDRA users were matched to 3,203 non-users based on clinically relevant variables at the end of 2009 using the coarsened exact matching procedure. Until December 31, 2011, a total of 1,128 deaths occurred, of which 468 (42%) were cardiovascular deaths, 229 (20%) were infection-related deaths, and 141 (12%) were malignancy-related deaths. Multivariable survival analyses accounting for intra-region correlation revealed that VDRA use was significantly associated with lower rates of infection- and malignancy-related deaths [subhazard ratio 0.62 (95% CI, 0.52–0.73) and 0.70 (95% CI, 0.50–0.97), respectively] but not with cardiovascular death [subhazard ratio 0.86 (95% CI, 0.72–1.04)]. Future randomized clinical trials with a sufficient sample size and an adequate follow-up period are warranted to test the clinical effectiveness of VDRA on infection and malignancy, rather than cardiovascular disease, among dialysis patients.

Scientific RepoRts | 7:41170 | DOI: 10.1038/srep41170 we used a nationwide cohort of Japanese dialysis patients to test our hypothesis that VDRA use is more strongly associated with infection-or malignancy-related death compared to cardiovascular death. Among 8,675 ESRD patients who started dialysis in 2007 and who survived until January 1 st , 2010, a total of 5,365 VDRA users as of December 2009 were matched to 3,203 non-users based on age, sex, diabetes, history of CVD, initial dialysis modality (HD or PD), and estimated GFR at dialysis initiation using the coarsened exact matching procedure ( Fig. 1) 14 . Given the low prevalence of VDRA use during the pre-dialysis period in Japan (i.e. < 9%) 15 , we assumed that VDRA use reflects the degree of exposure to VDRA after dialysis initiation and that clinical data at dialysis initiation can be used as pretreatment variables for causal inference. Among 5,365 VDRA users, 3,782 (71%) and 1,746 (34%) patients used intravenous VDRAs (i.e., calcitriol or maxacalcitol) and oral VDRAs (i.e., calcitriol, alfacalcidol, or falecalcitriol), respectively. At dialysis initiation, patients in the matched cohort were 65 ± 13 years old and had a mean GFR of 4.7 ± 2.3 mL/min/1.73 m 2 at dialysis initiation; 65% were male, 42% were diabetic, 20% had history of cardiovascular disease, and 4% received peritoneal dialysis (Table 1). Compared to non-VDRA users, VDRA users showed small differences in serum albumin concentrations (3.4 ± 0.6 mg/dL vs. 3.3 ± 0.6 mg/dL; standardized difference = 0.16) and central venous catheter use as vascular access (28% vs. 34%; standardized difference = 0.11). There were no meaningful differences (i.e., standardized difference < 0.1) in the other variables, including hemoglobin and serum concentrations of corrected calcium, phosphorus, and C-reactive protein.
During the follow-up between January  (Fig. 3). Additionally, VDRA use showed consistent associations with subcategories of cardiovascular and infection-related death; VDRA use was associated with lower mortality due to sepsis and pneumonia, but not with congestive heart failure, ischemic heart disease, cerebral hemorrhage, or sudden death. The associations of VDRA use with death due to pulmonary and gastrointestinal cancer did not reach statistical significance, likely due to the limited number of each outcome. Sensitivity analyses with further adjustment for therapies that may be on the causal pathway between VDRA use and mortality (i.e., the use of calcium carbonate, non-calcium phosphorus binders, and cinacalcet and prior invasive parathyroid interventions as of the end of 2009) slightly attenuated these associations in most outcomes but did not change their significance.

Discussion
In this nationwide cohort of Japanese dialysis patients, we demonstrated the difference in the strength of the relationship between VDRA use and three major causes of death; VDRA use was associated with lower infection-and malignancy-related mortality but not with cardiovascular mortality. Vitamin D stimulates innate immunity and enhances antimicrobial activity by upregulating the synthesis of the antimicrobial peptide cathelicidin 16 , which may explain the association between VDRA and infection-related death observed in this study. Interestingly, the association with lower mortality due to pneumonia is consistent with previous studies demonstrating a lower incidence of hospitalization due to acute respiratory infection among hemodialysis patients treated with vs. without oral VDRAs 17,18 . Another study also found that both intravenous and oral VDRAs showed an association with lower infection-related mortality 19 . Our study had relatively greater study power due to a larger sample size and thus enabled the examination of subcategory outcomes (i.e., sepsis and pneumonia) independently. The lower risk associated with VDRA use was consistently observed for both causes of death, suggesting a robust relationship between VDRA use and infection-related death.
Given that vitamin D also has antiproliferative effects, activates apoptotic pathways, inhibits angiogenesis, and promotes cell differentiation, nutritional vitamin D supplementation or VDRA use may prevent the development and progression of various cancers 20 . Although, there are scarce data on the benefit of VDRA use on cancer mortality among hemodialysis patients, partly due to its relatively low incidence of malignancy-related death compared to cardiovascular death, pre-transplant vitamin D deficiency is associated with a risk of post-transplant malignancy among kidney transplant recipients, who have decreased kidney function and high incidence of cancer 21 . Moreover, kidney transplant recipients with VDRAs, compared to those without VDRAs, showed a lower incidence of cancer 22 , which is consistent with the results of the present study. However, VDRA may be less effective in reducing cardiovascular risk for patients receiving renin-angiotensin system (RAS) inhibitors. VDRAs have been suggested to have a protective effect against cardiovascular disease because vitamin D suppresses RAS activity by downregulating renin expression in the kidney via its interaction with the vitamin D receptor 23 . Indeed, a small prospective observational study, which enrolled hemodialysis patients in 1992 when RAS inhibitors were not frequently used in this population, showed that patients on oral VDRA treatment exhibited lowered cardiovascular mortality compared to non-oral VDRA users 24 . However, recent clinical trials failed to demonstrate the cardio-protective effect of VDRA among pre-dialysis CKD patients with high prevalence (i.e., 70-80%) of RAS inhibitor use 10,11 . The prevalence of RAS inhibitor use is also increasing in the hemodialysis population from ~30% in 2002 to ~40% in 2008 in Japan 25,26 . These secular trends may explain the discrepancy among studies and the reason why the present study showed a trend, albeit not statistically significant due to the limited number of events, toward lower cardiovascular mortality among dialysis patients on VDRAs.
Although other mineral and bone disorder-related therapies, such as cinacalcet, invasive parathyroid interventions (i.e., parathyroidectomy and percutaneous ethanol injection therapy), and non-calcium containing phosphorus binders may partly explain the association between VDRA use and mortality [27][28][29][30] , the between-group differences in their prevalence as of the end of 2009 were not meaningful, and the adjustment for these potential intermediates slightly attenuated but did not change the significance in the associations of VDRA use with all-cause and cause-specific mortality. Importantly, the prevalence of cinacalcet use was low (i.e., 4% overall) because cinacalcet became available in clinical practice from January 2008 in Japan. Thus, this study did not enable an evaluation of the impact of cinacalcet on the association between VDRA use and mortality as suggested in the ADVANCE study 31 .
There are two studies showing conflicting results on the VDRA-mortality association. Naves-Diaz et al. showed that in a propensity-matched cohort of incident hemodialysis patients, oral VDRA use was associated with a similar reduction (i.e., 40-50%) in the risk of mortality across causes of death, including cardiovascular disease, infection, and malignancy 5 . The reason for the difference in the results between their study and the present study is unclear, but may be due to the difference in the definition of VDRA use and patient characteristics (i.e., race, history of cardiovascular disease, and the use of CV catheter, RAS inhibitor, and vitamin D supplementation). In contrast, Tentori et al. showed no association between VDRA use and all-cause mortality using the instrumental variable method, an alternative approach for estimating the causal associations using observational data. It uses an exogenous variable that is associated with exposure but not with outcome, and this method is considered less biased because it can eliminate both residual and unmeasured confounding under ideal conditions. However, it is challenging to identify appropriate variables that satisfy the assumptions of instruments. Furthermore, the instrumental variable method relies on only a portion of the total variance and leads to less precise and more conservative estimates than regression models, as indicated by wide standard errors, especially in cases with weak instrument relevance 12 . Treatment preferences of groups (e.g., clinical facilities) have been proposed as instruments, but several guidelines on VDRA might have made the preference less relevant to this exposure 32,33 . Also, this approach is valid only when unmeasured confounding effects exist within groups but not between groups, which is likely not the case with the real world scenario 13 . These limitations might have biased the results toward the null in the previous study of VDRA among hemodialysis patients using the instrumental variable method 9 . We also acknowledge several imitations in this study. First, the observational nature of this study does not allow us to confirm causality between VDRA use and cause-specific death, and there may be residual and unmeasured confounders. Specifically, neither serum 25-hydroxyvitamin D nor 1,25-dihydroxyvitamin D was available in this study, which was based on nation-wide surveys consisting of practical questionnaires, because their measurements have not been reimbursed and not been routinely performed under the bundled payment system in Japan. Second, we used the baseline Cox models and competing-risks regression models with VDRA use as of December 2009 as the exposure variable but did not take into account the doses or their changes overtime. We also do not have exact data on VDRA usage at dialysis initiation, although the prevalence of VDRA use during the pre-dialysis period was reasonably low in Japan (i.e. < 9%) 15 . Third, there may be survivor bias because our cohort consisted of dialysis patients who initiated dialysis in 2007 and survived until January 1 st , 2010, which might have resulted in bias toward the null, particularly for cardiovascular mortality, by excluding fragile patients. Finally, our result may not be extrapolated to patients with pre-dialysis CKD or those receiving dialysis treatments in countries with different proportions of causes of death or a different prescription pattern of RAS inhibitors. The proportion of patients on peritoneal dialysis was also limited.
In conclusion, VDRA use was associated with infection-and malignancy-related mortality, but not with cardiovascular mortality, among dialysis patients. Future randomized clinical trials with a sufficient sample size and an adequate follow-up period are needed to test the clinical effectiveness of VDRA on infection and malignancy, rather than cardiovascular disease, in this population.

Methods
The Japanese Society for Dialysis Therapy conducts annual surveys of dialysis facilities throughout Japan in December. The surveys address epidemiological background, treatment conditions and the outcomes of treatment with dialysis. Dataset JRDR-13104 was used with the permission of the Committee of the Renal Data Registry of the Japanese Society for Dialysis Therapy (JRDR). The study protocol was approved by the Medicine Ethics Committee of the Japanese Society for Dialysis Therapy as exempt from informed consent. The study was performed in accordance with the Declaration of Helsinki.

Study Cohort Restriction.
To examine the association of exposure to VDRA with all-cause and cause-specific mortality, we identified 24,490 ESRD patients who started dialysis treatment during 2007 and who survived until January 1 st , 2010 (Fig. 4). Information on dialysis patients and data on VDRA use were available in the surveys at 2007 and 2009, respectively. Thus, data on demographics, comorbidities, and laboratory variables at dialysis initiation were extracted from the 2007 survey. Hemoglobin and serum concentrations of albumin, creatinine, calcium, phosphorus, and C-reactive protein were measured using standard laboratory techniques at each center. VDRA users and non-users were defined based on data on VDRA use extracted from the 2009 survey. VDRA available in Japan between 2007 and 2009 included oral calcitriol, oral alfacalcidol, oral falecalcitriol, intravenous calcitriol, and intravenous maxacalcitol. Out of 24,490 patients, we excluded 4,100 with missing data on VDRA use and 11,295 with missing information on diabetes, history of cardiovascular disease, or estimated glomerular filtration rate (GFR). We also excluded 413 and 7 patients with history of cancer and history of human immunodeficiency virus infection and acquired immune deficiency syndrome (HIV/AIDS), respectively.
Among the remaining 8,675 dialysis patients, 5,365 VDRA users were matched to 3,203 non-users based on age, sex, diabetes, history of CVD, initial dialysis modality (HD or PD), and estimated GFR at dialysis initiation using the coarsened exact matching procedure 14 . First, we coarsened age and estimated GFR at dialysis initiation into categories using cut-points of 50, 60, 70, and 80 years and Sturge's rule, respectively. We then sorted all patients by each stratum of the coarsened variables, as well as by sex, diabetes, central venous catheter as vascular access, and dialysis modality (PD or HD). Within each stratum that included at least one patient in each group, non-VDRA users were given a weight of 1, and VDRA users were given a weight that equalized the ratio of sum of weights in each group of the stratum to the ratio of total matched patients on each group. Coarsened exact matching has several advantages over other matching approaches, including propensity-score matching; requiring fewer ad-hoc post-estimation assumptions about how to define a match; automatic balancing of treatment and control populations; and superior computational properties for large datasets. It is also particularly suitable for applications where most independent variables can be categorized appropriately. There were no meaningful differences in variables in Table 1 except for a small difference in estimated GFR among included vs. excluded patients (5.0 ± 2.4 vs. 5.7 ± 4.6 mL/min/1.73 m 2 , Supplemental Table 1).
Statistics. Data are presented as the mean ± SD or percentage, as appropriate. Differences between groups were compared by standardized differences due to the large sample size of this study 34,35 . Standardized differences of 0.8, 0.5, and 0.2 in absolute values are considered large, medium, and small differences, and ≥ 0.1 was defined as meaningful imbalance 34,35 . The primary outcome was time to infection-related death. The secondary outcomes included time to all-cause, cardiovascular, and malignancy-related death. Information on all-cause, cardiovascular, infection-related, and malignancy-related death were extracted from the surveys at 2010 and 2011. Cardiovascular death was defined as death due to heart failure, pulmonary edema, ischemic heart disease, arrhythmia, endocarditis, valvular disease, other cardiac disease, subarachnoid hemorrhage, cerebral hemorrhage, cerebral infarction, other cerebrovascular disease, or sudden death. Infection-related death was defined as death due to sepsis, pneumonia, gastrointestinal infection, cholecystitis/cholangitis, peritonitis, tuberculosis, HIV/AIDS, hepatitis virus infection, influenza virus infection, or other infection. The association of VDRA use and cause-specific mortality was examined by competing-risks regression based on Fine and Gray's proportional subhazards model 36 with and without adjustment for central venous catheter use, body weight, mean arterial blood pressure, hemoglobin, and serum concentrations of albumin, calcium, phosphorus, and C-reactive protein, using the other causes of death as competing risks. With regard to the sensitivity analyses, we further adjusted for other mineral and bone disorder-related therapies, including calcium carbonate, non-calcium containing phosphorus binders (i.e., sevelamer hydrochloride and lanthanum carbonate), cinacalcet, and history of invasive parathyroid interventions (i.e., parathyroidectomy and percutaneous ethanol injection therapy) as of the end of 2009. Clustered sandwich estimator was used to allow intragroup correlation within regions (i.e., prefectures). The multiple imputation method with 5 data sets was used to address missing values using multivariate normal regression based on all baseline data and an indicator of death. We performed Cox regression analyses to determine the association of VDRA use and all-cause mortality. P-values < 0.05 were considered significant.