Article
Bisphosphonates and Mortality in Women with CKD and the Presence or Absence of Cardiovascular Disease Robert M. Perkins,*† H. Lester Kirchner,*‡ Kunihiro Matsushita,§ Ion D. Bucaloiu,† Evan Norfolk,† and James E. Hartle†
Abstract Background and objectives A modest protective association between bisphosphonate prescription and mortality among women with CKD but without clinically manifest cardiovascular disease has been shown. Whether a prior cardiovascular event (myocardial infarction, stroke, or heart failure) modifies this association is unknown. Design, setting, participants, & measurements A cohort of adult women with stages 3 and 4 CKD receiving primary care in a rural integrated health care system during the period 2004–2011 without history of advanced malignancy or organ transplantation (n=6756, median age=74 years, median follow-up=4.3 years) was retrospectively assembled. The primary analysis compared those patients prescribed bisphosphonates (both prevalent and incident use during follow-up) with those patients not prescribed. Additional approaches were taken to account for survival and indication biases. The primary outcome was time to death by Cox multivariable regression. Results In the primary analysis, compared with women not prescribed a bisphosphonate, the hazard ratio (95% confidence interval) for death among women prescribed a bisphosphonate was 0.90 (0.78 to 1.04) if there was no history of cardiovascular event but 1.22 (1.04 to 1.42) if there was history of cardiovascular event (P for interaction=0.004). In the additional approaches, associations between bisphosphonate prescription and mortality among those patients with a prior cardiovascular history varied: hazard ratios (95% confidence intervals) were 1.25 (1.01 to 1.57), 1.48 (1.16 to 1.88), and 0.94 (0.66 to 1.34). Interaction by prior cardiovascular event history varied across these three approaches (P=0.07, P=0.22, and P=0.05). Conclusion In this study of women with CKD, the association between bisphosphonate treatment and mortality risk was inconclusive across a series of analyses designed to account for various types of selection and indication bias. Clin J Am Soc Nephrol 9: 874–880, 2014. doi: 10.2215/CJN.07790713
Introduction Among patients with CKD, vascular calcification may amplify cardiovascular risk (1,2). Bisphosphonates might attenuate this pathologic mineral metabolism and in turn, modify cardiovascular risk. However, the use of these agents is controversial because of limited knowledge of their pharmacokinetics and dynamics among those patients with CKD (3–5). Risk–benefit analysis is further clouded by discrepancies between animal studies and outcome studies in humans. Although these agents have been shown to improve vascular calcification scores in animal models, some studies have suggested that they increase the risk of cardiac arrhythmias and stroke in humans, although no clear risk mechanism has been established (6–8). An additional complicating factor is the limited trial data on bisphosphonate use among those patients with CKD (3). Dosing guidelines discourage the use of these agents among those patients with more advanced (stage 4 or higher) CKD. These gaps limit the data that might be used to test hypotheses about potential cardiovascular benefit (or risk) among those patients 874
Copyright © 2014 by the American Society of Nephrology
*Geisinger Center for Health Research, Danville, Pennsylvania; † Nephrology Department and ‡ Division of Medicine, Geisinger Medical Center, Danville, Pennsylvania; and § Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland Correspondence: Dr. Robert M. Perkins, Geisinger Center for Health Research, 100 North Academy Drive, Danville, PA 17822. Email: rmperkins@ geisinger.edu
with CKD. Observational studies of community-based cohorts offer an opportunity to examine associations between treatment with these agents and key outcomes but present methodological challenges related primarily to survival and treatment indication biases (9). We have previously reported that, among adult patients with CKD and without clinically manifest cardiovascular disease, bisphosphonate therapy is associated with a lower risk of death (10). Of some concern in subgroup analyses was that this association was lost among those patients with coronary disease or cardiac dysrhythmia, suggesting the possibility that the presence of cardiovascular disease may modify any protective mortality benefit among those patients with CKD treated with a bisphosphonate. This prior study excluded any individual with a history of a major cardiovascular event, thus limiting the extent to which this issue might be explored. We wished to test the hypothesis that the mortality risk associated with bisphosphonate therapy among women patients with CKD differs between those patients with and without a prior major cardiovascular event. www.cjasn.org Vol 9 May, 2014
Clin J Am Soc Nephrol 9: 874–880, May, 2014
Materials and Methods This study was approved by the Geisinger Medical Center Institutional Review Board. The data source was EpicCare, Geisinger Medical Center’s electronic health record. Study Population Patients eligible for the study included women between the ages of 18–88 years as of the study index date who were enrolled for primary care at any Geisinger facility and had CKD stage 3 or 4 (defined as two outpatient eGFR values between 15 and ,60 ml/min per 1.73 m2 on at least two dates separated by at least 90 days but no more than 730 days). We limited the analysis to women because few men were treated with bisphosphonates. Cohort enrollment took place between January 1, 2004, and March 31, 2011; followup for outcomes occurred through September 30, 2011. The index date was the first qualifying CKD date (defined as the second of the two eGFR values) after January 1, 2004, or January 1, 2004, for those patients with prevalent CKD. Prevalent CKD was accounted for using left truncation; information on CKD status for the study population was available back to January 1, 2001 (11). GFR was estimated from serum creatinine using the Chronic Kidney Disease Epidemiology Collaboration estimating equation (12). Serum creatinine was measured at a single Geisinger laboratory using the isotope dilution/mass spectroscopy–traceable Roche enzymatic method throughout the study period (13). Patients were excluded for hospital-acquired AKI occurring within 30 days of the index date (defined by an increase of 50% or more from baseline serum creatinine during a hospitalization for any cause, with the baseline serum creatinine defined as the lowest recorded value between 90 days before the index admission and the hospital discharge date) (14), pregnancy within 6 months of the index date, history of ESRD or prior inpatient dialysis, history of solid organ transplantation, or metastatic cancer. Patients were censored for death or end of the study period (September 30, 2011), whichever occurred first. Exposure Bisphosphonate therapy was identified using electronic prescription orders. Any outpatient prescription for a medication (oral or parenteral) within this class on or after the index date was qualifying. The analysis treated bisphosphonates as a time-dependent exposure; a participant accumulated exposure time only after the first bisphosphonate medication order date. After a subject began therapy with a bisphosphonate, active treatment until end of follow-up was assumed. Outcomes The study outcome was all-cause death. Information on vital status for Geisinger primary care recipients is updated monthly by institutional query of the Social Security Administration’s death master file through the National Technical Information Service (15). Study Covariates Variables considered included demographic information, medications active at the time of the index date (b-blockers, aspirin, clopidogrel, HMG-CoA reductase inhibitors, and
Bisphosphonates and Mortality, Perkins et al.
875
glucocorticoids), comorbid conditions (diabetes, osteoporosis, chronic obstructive pulmonary disease [COPD], peripheral arterial disease, thyroid disease, tobacco use, history of cardiac dysrhythmia, and history of coronary artery disease), subspecialty rheumatology consultation during the 12 months before the index date, number of visits to a primary care physician during the 12 months before the index date, vital signs (body mass index and systolic and diastolic BPs), and laboratory results (serum creatinine, serum albumin, serum HDL and LDL, and proteinuria) within 6 months before the index date. Comorbid conditions were coded if present on a patient’s permanent medical problem list or identified at two or more outpatient encounters using International Classification of Diseases (ICD)-9 codes. Proteinuria was coded as positive, negative, or not assessed. A positive test was defined as any semiquantitative dipstick analysis result of 30 mg/dl or greater, a urinary protein-to-creatinine ratio of 0.2 mg protein/1 g creatinine or greater, or a 24-hour urine protein of at least 300 mg protein. Statistical Analyses Patients were stratified by baseline cardiovascular event status at the time of cohort entry. A cardiovascular event was coded if present on the permanent medical problem list, coded at least two times in conjunction with an outpatient clinic visit, or listed as a primary or secondary hospital discharge diagnosis before the index date. A cardiovascular event was defined as myocardial infarction (ICD-9 410.XX, 411.0, 411.1, or 411.89), congestive heart failure (ICD-9 414.8, 425, 425.4–425.9, 428.XX, or 429.4), or stroke (ICD-9 430, 431, 432.9, 433.0–433.21, 433.81, 433.91, 434.01, 434.10, 434.11, 434.91, 435.3, 435.9, 436, 437.8, 437.9, or 438.XX). For any continuously measured covariate with more than 10% missing baseline values (serum calcium, serum albumin, or serum HDL and LDL cholesterol), we used multiple imputation with five datasets to limit bias (16). For proteinuria, a dummy variable was created to indicate whether this test was ordered (17). A sensitivity analysis was performed in which, for the primary analysis alone, missing case values were analyzed using dummy variable assignment for the calcium, albumin, and cholesterol laboratory results. Baseline comparisons (by cardiovascular event status and bisphosphonate treatment status) were made using the Kruskal–Wallis nonparametric and Pearson’s chi-squared tests as appropriate. A Poisson regression model was used to estimate the mortality rate across groups, and the results were expressed as incident rate ratios. Time until death was analyzed using the Cox proportional hazard regression model and treating bisphosphonate therapy as time-dependent. Variables significant at P,0.10 in the univariate analysis or previously associated with death among patients with CKD, were included in the Cox model. An interaction variable (bisphosphonate therapy3prior cardiovascular event) was created and entered into the survival model to test whether the presence of a prior cardiovascular event modifies the association between bisphosphonate treatment and mortality risk. A P value ,0.05 for the coefficient of the interaction term would indicate a very low likelihood (i.e., chance) of observing that coefficient under the assumption of no interaction. The practical interpretation would be that a prior cardiovascular event modifies the effect of bisphosphonate treatment on the risk of death. The
876
Clinical Journal of the American Society of Nephrology
Cox proportionality assumption was tested with timedependent covariates created by interacting predictors and a function of survival time. These time-dependent covariates were nonsignificant. Sensitivity Analyses Given the observational design of this study, survival and indication bias were of practical concern in the primary analysis. Secondary analyses were, therefore, performed. Prevalent Versus Incident Bisphosphonate Treatment (Approach 2). Those patients treated with a bisphosphonate at study entry might be healthier than those patients who ceased treatment or died on therapy before study entry because of adverse effects or low adherence, potentially resulting in a bias to protective effects of bisphosphonate. Therefore, we repeated the analysis after excluding patients receiving bisphosphonate therapy at the time of the study index date (i.e., only including incident bisphosphonate users). Treatment Lag Effect (Approach 3). It was considered possible that any potential pathophysiologic bisphosphonate effect would increase mortality risk only after prolonged treatment; an additional analysis was, therefore, performed using the approach 2 cohort, in which deaths occurring within the 365-day period after treatment initiation among the treated cohort subset were attributed to the unexposed group. Treatment Indication and Timing (Approach 4). There were likely important differences in clinical characteristics between those patients prescribed and not prescribed bisphosphonates, potentially confounding the association between bisphosphonate treatment and mortality. A propensity score-adjusted analysis was, therefore, performed. For this sensitivity analysis, the characteristics and clinical profiles of the treated subset were characterized at the time of bisphosphonate initiation rather than the study index date used in the primary analysis across the incident treatment cohort (i.e., the approach 2 cohort, above). All potential confounding variables were considered for inclusion in the propensity score; final variable selection was based on a boosted regression tree model and comparison of the relative influence on bisphosphonate treatment (18). The inverse of the propensity score was then used to weight the observations (inverse probability weighting). Given the degree of variability typically observed with these weights, they were further standardized using the marginal probability of bisphosphonate treatment. The Cox proportional hazard regression model was then fit using bisphosphonate use as the variable of interest and weighted by the stabilized inverse probability weights (19). The interaction of interest (bisphosphonate treatment with cardiovascular disease) was tested like in the other approaches. All analyses were performed using SAS version 9.2 (SAS Corporation) and Stata version 12.0 (StataCorp.).
Results For the primary analysis, 6756 patients met cohort entry criteria; 2173 (32.2%) of the study population patients had a history of a prior major cardiovascular event. Nearly 40% of those patients with and without a prior cardiovascular event were treated with a bisphosphonate.
Among other differences (Table 1), those patients with a history of prior cardiovascular event were more likely to be older, have a smoking history, have diabetes, COPD, and vascular disease, be prescribed antiplatelet and HMGCoA reductase inhibitor therapy, have lower LDL and HDL cholesterol levels, and have proteinuria. They were also seen more frequently by their primary care provider during the year before study entry. Those women prescribed a bisphosphonate were more likely to be older and have seen a rheumatologist in the year before the study entry; they were more likely to have been prescribed an HMG-CoA reductase inhibitor and have COPD, but they were less likely to have diabetes or proteinuria. They were more likely to have a higher HDL cholesterol and a lower body mass index (Table 2). The study population was followed for a median (interquartile range) of 4.3 (2.2–6.6) years (29,555 person-years). There were 1592 deaths during the study period. Ageadjusted mortality rates were significantly higher among those patients with (relative to those without) a prior cardiovascular event (incident rate ratio, 1.95; 95% confidence interval [95% CI], 1.77 to 2.15; P,0.001). In the whole study population, no association between bisphosphonate treatment and age-adjusted mortality risk was observed (incident rate ratio, 0.98; 95% CI, 0.88 to 1.08; P=0.67). When stratifying by prior cardiovascular event status, the age-adjusted mortality rate associated with bisphosphonate treatment among those patients with a prior cardiovascular event was 15% higher relative to those patients not treated (incident rate ratio, 1.15; 95% CI, 0.99 to 1.34; P=0.06). Among those patients without a history of cardiovascular disease, no clear association between bisphosphonate use and ageadjusted mortality risk was observed (incidence rate ratio, 0.96; 95% CI, 0.83 to 1.10; P=0.57). Table 3 shows the unadjusted and fully adjusted Cox proportional mortality hazards associated with bisphosphonate treatment for the primary analysis (approach 1) and each of the three sensitivity analyses (approaches 2–4). In the primary adjusted analysis, bisphosphonate treatment was associated with a higher risk of death among those patients with but not without a history of major cardiovascular event (similar findings were observed when the analysis was limited to those patients with a history of congestive heart failure and also when missing values for serum calcium, albumin, and HDL and LDL cholesterol were handled using dummy variable assignment instead of multiple imputation; data not shown). The interaction was significant (P=0.004), suggesting that a prior major cardiovascular event modifies the impact of treatment with bisphosphonates in this population. The results differed under the conditions of the sensitivity analyses. For the analysis focusing on new initiators of bisphosphonates (approach 2), 1691 (25%) patients from the primary analysis cohort were excluded, because they were on treatment with a bisphosphonate at the time of the study index date, leaving 5065 subjects for the analysis. These subjects were followed for a median (interquartile range) of 3.5 (1.7–5.7) years. As shown in Table 3, the trend toward a beneficial association of bisphosphonate treatment and mortality risk among those patients without a prior history of a major cardiovascular event is weaker in this analysis, whereas the adjusted risk persists for those
Clin J Am Soc Nephrol 9: 874–880, May, 2014
Bisphosphonates and Mortality, Perkins et al.
877
Table 1. Baseline characteristics of women with CKD by cardiovascular event status
Prior Heart Disease Variable Age at index (yr), median (IQR) Bisphosphonate prescription, n (%) Prevalent Incident None Caucasian, n (%) Smoking status, n (%) Never Past Current Thyroid disorder, n (%) Osteoporosis, n (%) Any rheumatologist visits in the past year, n (%) Aspirin, n (%) Clopidogrel, n (%) b-Blockers, n (%) Statins, n (%) Diabetes, n (%) Cardiac dysrhythmia, n (%) Chronic obstructive pulmonary disease, n (%) Peripheral arterial disease, n (%) Atherosclerotic coronary artery disease, n (%) Proteinuria, n (%) Not ordered Negative Positive Body mass index (kg/m2), median (IQR) Systolic BP (mmHg), median (IQR) Diastolic BP (mmHg), median (IQR) Albumin (g/dl), median (IQR) LDL (mg/dl), median (IQR) HDL (mg/dl), median (IQR) eGFR (ml/min per 1.73 m2), median (IQR) Number of visits to PCP 12 mo before index date, median (IQR)
Yes (n=2173)
No (n=4583)
75.1 (68.9–79.2)
73.9 (67.4–78.4)
534 (24.6) 282 (13.0) 1357 (62.4) 2148 (98.8)
1157 (25.2) 652 (14.2) 2774 (60.5) 4525 (98.7)
1137 (57.3) 664 (33.5) 182 (9.2) 660 (30.4) 643 (29.6) 266 (12.2) 350 (16.1) 264 (12.2) 1068 (49.2) 943 (43.4) 1153 (53.1) 739 (34.0) 428 (19.7) 387 (17.8) 943 (43.4)
2703 (64.7) 1100 (26.3) 374 (9.0) 1309 (28.6) 1452 (31.7) 512 (11.2) 477 (10.4) 134 (2.9) 1644 (35.9) 1635 (35.7) 1767 (38.6) 511 (11.2) 417 (9.1) 351 (7.7) 798 (17.4)
925 (42.6) 797 (36.7) 451 (20.8) 31.1 (27.3–36.7) 128.0 (116–140) 70.0 (60–76) 4.1 (3.8–4.3) 89.0 (70–113) 49.0 (41–58) 53.7 (46.2–57.2) 5.0 (3–7)
2375 (51.8) 1513 (33.0) 695 (15.2) 31.1 (27.1–35.8) 130.0 (120–142) 70.0 (64–80) 4.1 (3.9–4.3) 100.0 (79–126) 52.0 (44–62) 54.2 (47.8–57.6) 4.0 (2–6)
P Value ,0.001 0.24
0.69 ,0.001
0.13 0.08 0.20 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001
0.02 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001
IQR, interquartile range; PCP, primary care physician.
patients with a prior history of cardiovascular event treated with bisphosphonates. The interaction by the presence of such a history was not statistically significant (P=0.07). In the second sensitivity analysis (approach 3), mortality risk attributable to treatment with a bisphosphonate was limited to the period beginning 1 year after the date of first bisphosphonate prescription. The approach 2 cohort was, therefore, modified by deaths occurring during this 1-year lag period; 184 deaths in bisphosphonate-treated subjects occurred during this lag period. Median follow-up was 2.5 years. Regardless of cardiovascular disease history, bisphosphonate treatment was associated with an increased risk of death in the fully adjusted analysis. The association was not statistically significant for those patients without cardiovascular disease, and the strength of the association was stronger for those patients with preexisting cardiovascular disease. The interaction term was nonsignificant (Table 3). In the final sensitivity analysis (approach 4), in which a propensity score for the likelihood of bisphosphonate treatment was used to adjust the Cox proportional hazard
model, the approach 2 cohort was again used as the initial study population. This cohort was further modified by characterization of treated patients at the time of bisphosphonate initiation as opposed to the study index date used in the primary analysis; 403 patients were excluded, resulting in a sample size of 4662 patients. For those patients without preexisting cardiovascular disease, bisphosphonate treatment was associated with a reduced mortality risk in this analysis (hazard ratio, 0.55; 95% CI, 0.37 to 0.82) but not in those patients with established cardiovascular disease. The interaction term was statistically significant, suggesting that cardiovascular disease modifies the risk association (Table 3).
Discussion In this study of women with CKD, the association between bisphosphonate treatment and mortality risk was inconclusive across a series of analyses designed to account for various types of selection and indication bias. The pattern of these associations would suggest that the
878
Clinical Journal of the American Society of Nephrology
Table 2. Baseline characteristics of women with CKD by bisphosphonate treatment status
Bisphosphonate Use Variable
Age at index (yr), median (IQR) Prior heart disease, n (%) Caucasian, n (%) Smoking status, n (%) Never Past Current Thyroid disorder, n (%) Osteoporosis, n (%) Any rheumatologist visits in the past year, n (%) Aspirin, n (%) Clopidogrel, n (%) b-Blockers, n (%) Statins, n (%) Diabetes, n (%) Cardiac dysrhythmia, n (%) Chronic obstructive pulmonary disease, n (%) Peripheral arterial disease, n (%) Atherosclerotic coronary artery disease, n (%) Proteinuria, n (%) Not ordered Negative Positive Body mass index (kg/m2), median (IQR) Systolic BP (mmHg), median (IQR) Diastolic BP (mmHg), median (IQR) Albumin (g/dl), median (IQR) LDL (mg/dl), median (IQR) HDL (mg/dl), median (IQR) eGFR (ml/min per 1.73 m2), median (IQR) Number of visits to PCP 12 mo before index date, median (IQR)
P Valuea
Prevalent (n=1691)
Incident (n=934)
None (n=4131)
76.3 (72.2–80.0) 534 (31.6) 1680 (99.4)
75.0 (70.1–78.5) 282 (30.2) 928 (99.4)
73.1 (65.4–77.9) 1357 (32.8) 4065 (98.4)
965 (60.3) 519 (32.4) 116 (7.2) 508 (30.0) 1477 (87.3) 249 (14.7)
550 (63.9) 248 (28.8) 63 (7.3) 278 (29.8) 250 (26.8) 71 (7.6)
2325 (62.8) 997 (27.0) 377 (10.2) 1183 (28.6) 368 (8.9) 163 (4.0)
0.51 ,0.001 ,0.001
207 (12.2) 105 (6.2) 688 (40.7) 710 (42.0) 549 (32.5) 335 (19.8) 259 (15.3)
110 (11.8) 54 (5.8) 352 (37.7) 343 (36.7) 338 (36.2) 166 (17.8) 113 (12.1)
510 (12.4) 239 (5.8) 1672 (40.5) 1525 (36.9) 2033 (49.2) 749 (18.1) 473 (11.4)
0.89 0.81 0.25 ,0.001 ,0.001 0.27 ,0.001
191 (11.3) 453 (26.8)
82 (8.8) 238 (25.5)
465 (11.3) 1050 (25.4)
0.08 0.54
,0.001 0.24 0.003 ,0.001
,0.001 830 (49.1) 666 (39.4) 195 (11.5) 29.1 (25.4–33.3) 129.0 (119–140) 70.0 (62–78) 4.2 (3.9–4.4) 95.0 (76–120) 54.0 (46–64) 54.6 (48.9–57.6) 4.0 (3–6)
521 (55.8) 306 (32.8) 107 (11.5) 30.8 (26.8–35.5) 130.0 (120–142) 70.0 (64–80) 4.1 (3.9–4.3) 98.0 (80–124) 53.0 (45–63.5) 54.3 (47.9–58.1) 4 (2–6)
1949 (47.2) 1338 (32.4) 844 (20.4) 31.5 (28.0–37.3) 130.0 (120–142) 70.0 (64–78) 4.1 (3.8–4.3) 95.0 (75–122) 50.0 (42–59) 53.9 (46.5–57.4) 4 (2–6)
,0.001 ,0.001 ,0.001 ,0.001 0.09 ,0.001 ,0.001 ,0.001
a Kruskal–Wallis test for difference across groups. A significant value (,0.05) indicates that at least one group differs from another group among the three groups. IQR, interquartile range; PCP, primary care physician.
mortality risk association with bisphosphonate among women with CKD and a history of cardiovascular disease may differ from similar women without cardiovascular disease, which was evidenced by the statistically significant interaction term in two of four approaches. Both a protective effect of treatment (among women without clinically apparent cardiovascular disease) and an increased risk of treatment (among women with cardiovascular disease) were observed in some but not all of the analyses in this study; consistency across all analyses was not seen. No definitive conclusions regarding risk association can, therefore, be drawn. The varying results across the series of approaches provide cautionary evidence for the influence of selection and indication biases in assessing drug treatment effects in an observational setting. Across two of four analytic approaches in this study, we observed a statistically significant mortality risk among
bisphosphonate-treated subjects with a history of prior cardiovascular disease. The reasons for this finding cannot be elucidated given the nature of our study design. However, some reports have identified an increased risk of atrial fibrillation among bisphosphonate users, a finding not confirmed by the US Food and Drug Administration (6–8,20). Other studies have identified an association between low bone turnover—a bisphosphonate effect—and vascular calcification, an adverse outcome that might potentially be amplified by delayed drug clearance among those patients with CKD (21,22). However, a systematic review of clinical trials examining the impact of bisphosphonate treatment on coronary and aortic calcification and carotid arterial thickness found inconsistent associations among the nine reported trials included (23). Although our analysis does not confirm higher risk, the associations of treatment with risk among those patients with
Clin J Am Soc Nephrol 9: 874–880, May, 2014
Bisphosphonates and Mortality, Perkins et al.
879
Table 3. Cox proportional mortality hazard for women with CKD prescribed bisphosphonate therapy by cardiovascular event status
HR (95% CI)
P Value
HR (95% CI)
P Value
P Value for Interaction (Bisphosphonate and Prior CV Event)
0.97 (0.85 to 1.12) 0.90 (0.78 to 1.04)
0.71 0.15
1.17 (1.01 to 1.36) 1.22 (1.04 to 1.42)
0.04 0.01
0.07 0.004
1.01 (0.83 to 1.24) 0.95 (0.78 to 1.17)
0.89 0.65
1.15 (0.93 to 1.43) 1.25 (1.01 to 1.57)
0.21 0.05
0.40 0.07
1.27 (1.02 to 1.58) 1.21 (0.97 to 1.51)
0.03 0.09
1.33 (1.05 to 1.68) 1.48 (1.16 to 1.88)
0.02 0.001
0.78 0.22
0.93 (0.70 to 1.23) 0.55 (0.37 to 0.82)
0.60 0.003
0.85 (0.64 to 1.13) 0.94 (0.66 to 1.34)
0.26 0.73
0.65 0.05
No Prior CV Event Model Approach 1 Crude Adjusteda Approach 2 Crude Adjusteda Approach 3 Crude Adjustedb Approach 4 Crude Adjustedc
Prior CV Event
CV, cardiovascular; HR, hazard ratio; 95% CI, 95% confidence interval. a Adjusted for age; smoking status; prior CV event (myocardial infarction, hospitalization for congestive heart failure, or stroke); history of cardiac dysrhythmia, atherosclerotic coronary artery disease, diabetes, or chronic obstructive pulmonary disease; HMG-CoA reductase inhibitor prescription; body mass index; systolic BP; and baseline eGFR, serum albumin, urine protein excretion, and LDL and HDL cholesterol. b Adjusted for the above covariates (with the exception of LDL cholesterol), prescription for b-blocker or aspirin, and history of a dualenergy x-ray absorptiometry (DEXA) scan in the year before index date. c Propensity score-adjusted approach. The propensity score was derived from the following variables: age, body mass index, height, systolic and diastolic BP, glucocorticoid therapy at any time during the 12 months before the index date, HMG-CoA reductase inhibitor prescription, b-blocker prescription, clopidogrel prescription, aspirin use, history of smoking, hyper- or hypothyroidism, osteopenia, osteoporosis, diabetes, history of prior cardiovascular event, eGFR, LDL and HDL cholesterol, serum albumin, proteinuria, DEXA study in the prior 3 years, number of visits with primary care physician in the prior 12 months, referral to a rheumatologist in the prior 12 months, and number of visits with a rheumatologist in the prior 12 months.
CKD and cardiovascular disease would warrant additional investigation and does support a cautious approach to the use of these medications in this population. Animal models show clearly that bisphosphonates accumulate and show activity in calcified arterial vessels; whether this result, in turn, increases risk is not clear (24,25). In a previous study, we reported a modest protective association of bisphosphonates in a similar study population without a prior history of a major cardiovascular event but with a different study enrollment period, inclusion criteria, and follow-up period (10). In the analysis presented here, the derivation of the study population across the sensitivity analyses had a considerable impact on risk associations; in three of four approaches, the protective association was again observed among those patients without cardiovascular disease, although this association was statistically significant in only the analysis adjusting for treatment indication (approach 4). Observational studies designed to address drug effects— both adverse and intended—are complicated by multiple potential sources of bias. It is not possible to determine which of the approaches applied in this observational study is least biased. Although the risk estimates suggest potential relationships, conclusions regarding definitive risk or benefit await additional study. Accounting for cumulative exposure, graded dose effects, and intraclass drug differences represents additional challenges when examining adverse drug effects observationally. Despite a growing understanding of the atypical pathophysiology
of vascular calcification among those patients with CKD, no targeted therapies to reduce cardiovascular risk exist for this population of patients beyond those therapies available to the general population. Interventional trials may help untangle the multitude of factors that influence outcomes in this unique population, and the results of this observational study, in which both risk and benefit are seen depending on cardiovascular disease burden, may justify the effort and resources required to carry out such trials. This study has additional limitations beyond its observational design. Nonadherence or changes in treatment plan after the first bisphosphonate prescription are not accounted for in this analysis to include discontinuation of therapy, which may result in bias to the null. Similarly, bisphosphonate dose was not accounted for in this study, and potential intraclass differences among the individual agents may exist, which in turn, may differentially impact mortality risk. This study was not designed to assess this potentially important heterogeneity. Exposure misclassification because of treatment occurring outside the Geisinger system is an additional potential limitation, although it is likely minimal because of Geisinger referral patterns for the primary care population. Follow-up time was modest and may have been inadequate to detect longer-term effects. Also, because the definition of CKD was based on eGFR, additional examination is needed to determine whether our results can be generalizable to those patients with CKD based on albuminuria. Confounding by factors not included in the model is an additional potential limitation; two potentially influential variables not available
880
Clinical Journal of the American Society of Nephrology
for this analysis include a history of bone fracture and information about serum vitamin D levels. For example, if bisphosphonate treatment indication was based on the occurrence of pathologic bone fracture for a substantial proportion of study subjects, this confounding might mask a protective effect of these agents, given the strong mortality risk associated with such events. In conclusion, bisphosphonate treatment among women with CKD was not consistently significantly associated with mortality, regardless of the presence of a prior cardiovascular event. Risk estimates varied across sensitivity analyses, suggesting the need for careful attention to bias when designing observational drug studies. Interventional trials likely are needed to elucidate the true nature of any risk or benefit relationship with these drugs among those patients with CKD, and they may be warranted in light of these results. Acknowledgments This study was presented in modified form at the American Society of Nephrology Annual Meeting in San Diego, CA (November 3, 2012). Disclosures None. References 1. Blacher J, Guerin AP, Pannier B, Marchais SJ, London GM: Arterial calcifications, arterial stiffness, and cardiovascular risk in end-stage renal disease. Hypertension 38: 938–942, 2001 2. London GM: Cardiovascular calcifications in uremic patients: Clinical impact on cardiovascular function. J Am Soc Nephrol 14 [Suppl 4]: S305–S309, 2003 3. Ott SM: Bisphosphonate safety and efficacy in chronic kidney disease. Kidney Int 82: 833–835, 2012 4. Toussaint ND, Lau KK, Strauss BJ, Polkinghorne KR, Kerr PG: Effect of alendronate on vascular calcification in CKD stages 3 and 4: A pilot randomized controlled trial. Am J Kidney Dis 56: 57–68, 2010 5. Ariyoshi T, Eishi K, Sakamoto I, Matsukuma S, Odate T: Effect of etidronic acid on arterial calcification in dialysis patients. Clin Drug Investig 26: 215–222, 2006 6. Cummings SR, Schwartz AV, Black DM: Alendronate and atrial fibrillation. N Engl J Med 356: 1895–1896, 2007 7. Abrahamsen B, Eiken P, Brixen K: Atrial fibrillation in fracture patients treated with oral bisphosphonates. J Intern Med 265: 581–592, 2009 8. Black DM, Delmas PD, Eastell R, Reid IR, Boonen S, Cauley JA, Cosman F, Lakatos P, Leung PC, Man Z, Mautalen C, Mesenbrink P, Hu H, Caminis J, Tong K, Rosario-Jansen T, Krasnow J, Hue TF, Sellmeyer D, Eriksen EF, Cummings SR; HORIZON Pivotal Fracture Trial: Once-yearly zoledronic acid for treatment of postmenopausal osteoporosis. N Engl J Med 356: 1809–1822, 2007 9. Armstrong K: Methods in comparative effectiveness research. J Clin Oncol 30: 4208–4214, 2012 10. Hartle JE, Tang X, Kirchner HL, Bucaloiu ID, Sartorius JA, Pogrebnaya ZV, Akers GA, Carnero GE, Perkins RM: Bisphosphonate therapy, death, and cardiovascular events among female patients with CKD: A retrospective cohort study. Am J Kidney Dis 59: 636–644, 2012
11. Foreman AJ, Lai GP, Miller DP: Surviving left truncation using PROC PHREG. Presented at the Western Users of SAS Software Meeting, University City, CA, November 5, 2008 12. Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF 3rd, Feldman HI, Kusek JW, Eggers P, Van Lente F, Greene T, Coresh J; CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration): A new equation to estimate glomerular filtration rate. Ann Intern Med 150: 604–612, 2009 13. Myers GL, Miller WG, Coresh J, Fleming J, Greenberg N, Greene T, Hostetter T, Levey AS, Panteghini M, Welch M, Eckfeldt JH; National Kidney Disease Education Program Laboratory Working Group: Recommendations for improving serum creatinine measurement: A report from the laboratory working group of the National Kidney Disease Education Program. Clin Chem 52: 5– 18, 2006 14. Lafrance JP, Miller DR: Acute kidney injury associates with increased long-term mortality. J Am Soc Nephrol 21: 345–352, 2010 15. National Technical Information Service: Available at: http:// www.ntis.gov/. Accessed March 5, 2013 16. Rubin DB: Multiple imputation. In: Multiple Imputation for Nonresponse in Surveys, New York, John Wiley & Sons, 1987, pp 15–19 17. Cohen J, Cohen P: Missing data. In: Applied Multiple Regression; Correlation Analysis for the Behavioral Sciences, Hillsdale, NJ, Erlbaum, 1983, pp 303–320 18. McCaffrey DF, Ridgeway G, Morral AR: Propensity score estimation with boosted regression for evaluating causal effects in observational studies. Psychol Methods 9: 403–425, 2004 19. Cole SR, Herna´n MA: Adjusted survival curves with inverse probability weights. Comput Methods Programs Biomed 75: 45– 49, 2004 20. FDA: Update of Safety Review Follow-up to the October 1, 2007 Early Communication about the Ongoing Safety Review of Bisphosphonates, 2008. Available at: http://www.fda.gov/Drugs/ DrugSafety/PostmarketDrugSafetyInformationforPatientsandProviders/DrugSafetyInformationforHeathcareProfessionals/ucm136201.htm. Accessed March 1, 2013 21. London GM, Marty C, Marchais SJ, Guerin AP, Metivier F, de Vernejoul MC: Arterial calcifications and bone histomorphometry in end-stage renal disease. J Am Soc Nephrol 15: 1943–1951, 2004 22. Elmariah S, Delaney JA, O’Brien KD, Budoff MJ, Vogel-Claussen J, Fuster V, Kronmal RA, Halperin JL: Bisphsophonate use and prevalence of valvular and vascular calcification in women MESA (the Multi-Ethnic Study of Atherosclerosis). J Am Coll Cardiol 56: 1752–1759, 2010 23. Santos LL, Cavalcanti TB, Bandeira FA: Vascular effects of bisphosphonates-a systematic review. Clin Med Insights Endocrinol Diabetes 5: 47–54, 2012 24. Ylitalo R, Mo¨nkko¨nen J, Urtti A, Ylitalo P: Accumulation of bisphosphonates in the aorta and some other tissues of healthy and atherosclerotic rabbits. J Lab Clin Med 127: 200–206, 1996 25. Abedin M, Tintut Y, Demer LL: Vascular calcification: Mechanisms and clinical ramifications. Arterioscler Thromb Vasc Biol 24: 1161–1170, 2004 Received: July 23, 2013 Accepted: January 22, 2014 Published online ahead of print. Publication date available at www. cjasn.org.
Bisphosphonates and Mortality in Women with CKD and the Presence or Absence of Cardiovascular Disease Robert M. Perkins,*† H. Lester Kirchner,*‡ Kunihiro Matsushita,§ Ion D. Bucaloiu,† Evan Norfolk,† and James E. Hartle†
Abstract Background and objectives A modest protective association between bisphosphonate prescription and mortality among women with CKD but without clinically manifest cardiovascular disease has been shown. Whether a prior cardiovascular event (myocardial infarction, stroke, or heart failure) modifies this association is unknown. Design, setting, participants, & measurements A cohort of adult women with stages 3 and 4 CKD receiving primary care in a rural integrated health care system during the period 2004–2011 without history of advanced malignancy or organ transplantation (n=6756, median age=74 years, median follow-up=4.3 years) was retrospectively assembled. The primary analysis compared those patients prescribed bisphosphonates (both prevalent and incident use during follow-up) with those patients not prescribed. Additional approaches were taken to account for survival and indication biases. The primary outcome was time to death by Cox multivariable regression. Results In the primary analysis, compared with women not prescribed a bisphosphonate, the hazard ratio (95% confidence interval) for death among women prescribed a bisphosphonate was 0.90 (0.78 to 1.04) if there was no history of cardiovascular event but 1.22 (1.04 to 1.42) if there was history of cardiovascular event (P for interaction=0.004). In the additional approaches, associations between bisphosphonate prescription and mortality among those patients with a prior cardiovascular history varied: hazard ratios (95% confidence intervals) were 1.25 (1.01 to 1.57), 1.48 (1.16 to 1.88), and 0.94 (0.66 to 1.34). Interaction by prior cardiovascular event history varied across these three approaches (P=0.07, P=0.22, and P=0.05). Conclusion In this study of women with CKD, the association between bisphosphonate treatment and mortality risk was inconclusive across a series of analyses designed to account for various types of selection and indication bias. Clin J Am Soc Nephrol 9: 874–880, 2014. doi: 10.2215/CJN.07790713
Introduction Among patients with CKD, vascular calcification may amplify cardiovascular risk (1,2). Bisphosphonates might attenuate this pathologic mineral metabolism and in turn, modify cardiovascular risk. However, the use of these agents is controversial because of limited knowledge of their pharmacokinetics and dynamics among those patients with CKD (3–5). Risk–benefit analysis is further clouded by discrepancies between animal studies and outcome studies in humans. Although these agents have been shown to improve vascular calcification scores in animal models, some studies have suggested that they increase the risk of cardiac arrhythmias and stroke in humans, although no clear risk mechanism has been established (6–8). An additional complicating factor is the limited trial data on bisphosphonate use among those patients with CKD (3). Dosing guidelines discourage the use of these agents among those patients with more advanced (stage 4 or higher) CKD. These gaps limit the data that might be used to test hypotheses about potential cardiovascular benefit (or risk) among those patients 874
Copyright © 2014 by the American Society of Nephrology
*Geisinger Center for Health Research, Danville, Pennsylvania; † Nephrology Department and ‡ Division of Medicine, Geisinger Medical Center, Danville, Pennsylvania; and § Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland Correspondence: Dr. Robert M. Perkins, Geisinger Center for Health Research, 100 North Academy Drive, Danville, PA 17822. Email: rmperkins@ geisinger.edu
with CKD. Observational studies of community-based cohorts offer an opportunity to examine associations between treatment with these agents and key outcomes but present methodological challenges related primarily to survival and treatment indication biases (9). We have previously reported that, among adult patients with CKD and without clinically manifest cardiovascular disease, bisphosphonate therapy is associated with a lower risk of death (10). Of some concern in subgroup analyses was that this association was lost among those patients with coronary disease or cardiac dysrhythmia, suggesting the possibility that the presence of cardiovascular disease may modify any protective mortality benefit among those patients with CKD treated with a bisphosphonate. This prior study excluded any individual with a history of a major cardiovascular event, thus limiting the extent to which this issue might be explored. We wished to test the hypothesis that the mortality risk associated with bisphosphonate therapy among women patients with CKD differs between those patients with and without a prior major cardiovascular event. www.cjasn.org Vol 9 May, 2014
Clin J Am Soc Nephrol 9: 874–880, May, 2014
Materials and Methods This study was approved by the Geisinger Medical Center Institutional Review Board. The data source was EpicCare, Geisinger Medical Center’s electronic health record. Study Population Patients eligible for the study included women between the ages of 18–88 years as of the study index date who were enrolled for primary care at any Geisinger facility and had CKD stage 3 or 4 (defined as two outpatient eGFR values between 15 and ,60 ml/min per 1.73 m2 on at least two dates separated by at least 90 days but no more than 730 days). We limited the analysis to women because few men were treated with bisphosphonates. Cohort enrollment took place between January 1, 2004, and March 31, 2011; followup for outcomes occurred through September 30, 2011. The index date was the first qualifying CKD date (defined as the second of the two eGFR values) after January 1, 2004, or January 1, 2004, for those patients with prevalent CKD. Prevalent CKD was accounted for using left truncation; information on CKD status for the study population was available back to January 1, 2001 (11). GFR was estimated from serum creatinine using the Chronic Kidney Disease Epidemiology Collaboration estimating equation (12). Serum creatinine was measured at a single Geisinger laboratory using the isotope dilution/mass spectroscopy–traceable Roche enzymatic method throughout the study period (13). Patients were excluded for hospital-acquired AKI occurring within 30 days of the index date (defined by an increase of 50% or more from baseline serum creatinine during a hospitalization for any cause, with the baseline serum creatinine defined as the lowest recorded value between 90 days before the index admission and the hospital discharge date) (14), pregnancy within 6 months of the index date, history of ESRD or prior inpatient dialysis, history of solid organ transplantation, or metastatic cancer. Patients were censored for death or end of the study period (September 30, 2011), whichever occurred first. Exposure Bisphosphonate therapy was identified using electronic prescription orders. Any outpatient prescription for a medication (oral or parenteral) within this class on or after the index date was qualifying. The analysis treated bisphosphonates as a time-dependent exposure; a participant accumulated exposure time only after the first bisphosphonate medication order date. After a subject began therapy with a bisphosphonate, active treatment until end of follow-up was assumed. Outcomes The study outcome was all-cause death. Information on vital status for Geisinger primary care recipients is updated monthly by institutional query of the Social Security Administration’s death master file through the National Technical Information Service (15). Study Covariates Variables considered included demographic information, medications active at the time of the index date (b-blockers, aspirin, clopidogrel, HMG-CoA reductase inhibitors, and
Bisphosphonates and Mortality, Perkins et al.
875
glucocorticoids), comorbid conditions (diabetes, osteoporosis, chronic obstructive pulmonary disease [COPD], peripheral arterial disease, thyroid disease, tobacco use, history of cardiac dysrhythmia, and history of coronary artery disease), subspecialty rheumatology consultation during the 12 months before the index date, number of visits to a primary care physician during the 12 months before the index date, vital signs (body mass index and systolic and diastolic BPs), and laboratory results (serum creatinine, serum albumin, serum HDL and LDL, and proteinuria) within 6 months before the index date. Comorbid conditions were coded if present on a patient’s permanent medical problem list or identified at two or more outpatient encounters using International Classification of Diseases (ICD)-9 codes. Proteinuria was coded as positive, negative, or not assessed. A positive test was defined as any semiquantitative dipstick analysis result of 30 mg/dl or greater, a urinary protein-to-creatinine ratio of 0.2 mg protein/1 g creatinine or greater, or a 24-hour urine protein of at least 300 mg protein. Statistical Analyses Patients were stratified by baseline cardiovascular event status at the time of cohort entry. A cardiovascular event was coded if present on the permanent medical problem list, coded at least two times in conjunction with an outpatient clinic visit, or listed as a primary or secondary hospital discharge diagnosis before the index date. A cardiovascular event was defined as myocardial infarction (ICD-9 410.XX, 411.0, 411.1, or 411.89), congestive heart failure (ICD-9 414.8, 425, 425.4–425.9, 428.XX, or 429.4), or stroke (ICD-9 430, 431, 432.9, 433.0–433.21, 433.81, 433.91, 434.01, 434.10, 434.11, 434.91, 435.3, 435.9, 436, 437.8, 437.9, or 438.XX). For any continuously measured covariate with more than 10% missing baseline values (serum calcium, serum albumin, or serum HDL and LDL cholesterol), we used multiple imputation with five datasets to limit bias (16). For proteinuria, a dummy variable was created to indicate whether this test was ordered (17). A sensitivity analysis was performed in which, for the primary analysis alone, missing case values were analyzed using dummy variable assignment for the calcium, albumin, and cholesterol laboratory results. Baseline comparisons (by cardiovascular event status and bisphosphonate treatment status) were made using the Kruskal–Wallis nonparametric and Pearson’s chi-squared tests as appropriate. A Poisson regression model was used to estimate the mortality rate across groups, and the results were expressed as incident rate ratios. Time until death was analyzed using the Cox proportional hazard regression model and treating bisphosphonate therapy as time-dependent. Variables significant at P,0.10 in the univariate analysis or previously associated with death among patients with CKD, were included in the Cox model. An interaction variable (bisphosphonate therapy3prior cardiovascular event) was created and entered into the survival model to test whether the presence of a prior cardiovascular event modifies the association between bisphosphonate treatment and mortality risk. A P value ,0.05 for the coefficient of the interaction term would indicate a very low likelihood (i.e., chance) of observing that coefficient under the assumption of no interaction. The practical interpretation would be that a prior cardiovascular event modifies the effect of bisphosphonate treatment on the risk of death. The
876
Clinical Journal of the American Society of Nephrology
Cox proportionality assumption was tested with timedependent covariates created by interacting predictors and a function of survival time. These time-dependent covariates were nonsignificant. Sensitivity Analyses Given the observational design of this study, survival and indication bias were of practical concern in the primary analysis. Secondary analyses were, therefore, performed. Prevalent Versus Incident Bisphosphonate Treatment (Approach 2). Those patients treated with a bisphosphonate at study entry might be healthier than those patients who ceased treatment or died on therapy before study entry because of adverse effects or low adherence, potentially resulting in a bias to protective effects of bisphosphonate. Therefore, we repeated the analysis after excluding patients receiving bisphosphonate therapy at the time of the study index date (i.e., only including incident bisphosphonate users). Treatment Lag Effect (Approach 3). It was considered possible that any potential pathophysiologic bisphosphonate effect would increase mortality risk only after prolonged treatment; an additional analysis was, therefore, performed using the approach 2 cohort, in which deaths occurring within the 365-day period after treatment initiation among the treated cohort subset were attributed to the unexposed group. Treatment Indication and Timing (Approach 4). There were likely important differences in clinical characteristics between those patients prescribed and not prescribed bisphosphonates, potentially confounding the association between bisphosphonate treatment and mortality. A propensity score-adjusted analysis was, therefore, performed. For this sensitivity analysis, the characteristics and clinical profiles of the treated subset were characterized at the time of bisphosphonate initiation rather than the study index date used in the primary analysis across the incident treatment cohort (i.e., the approach 2 cohort, above). All potential confounding variables were considered for inclusion in the propensity score; final variable selection was based on a boosted regression tree model and comparison of the relative influence on bisphosphonate treatment (18). The inverse of the propensity score was then used to weight the observations (inverse probability weighting). Given the degree of variability typically observed with these weights, they were further standardized using the marginal probability of bisphosphonate treatment. The Cox proportional hazard regression model was then fit using bisphosphonate use as the variable of interest and weighted by the stabilized inverse probability weights (19). The interaction of interest (bisphosphonate treatment with cardiovascular disease) was tested like in the other approaches. All analyses were performed using SAS version 9.2 (SAS Corporation) and Stata version 12.0 (StataCorp.).
Results For the primary analysis, 6756 patients met cohort entry criteria; 2173 (32.2%) of the study population patients had a history of a prior major cardiovascular event. Nearly 40% of those patients with and without a prior cardiovascular event were treated with a bisphosphonate.
Among other differences (Table 1), those patients with a history of prior cardiovascular event were more likely to be older, have a smoking history, have diabetes, COPD, and vascular disease, be prescribed antiplatelet and HMGCoA reductase inhibitor therapy, have lower LDL and HDL cholesterol levels, and have proteinuria. They were also seen more frequently by their primary care provider during the year before study entry. Those women prescribed a bisphosphonate were more likely to be older and have seen a rheumatologist in the year before the study entry; they were more likely to have been prescribed an HMG-CoA reductase inhibitor and have COPD, but they were less likely to have diabetes or proteinuria. They were more likely to have a higher HDL cholesterol and a lower body mass index (Table 2). The study population was followed for a median (interquartile range) of 4.3 (2.2–6.6) years (29,555 person-years). There were 1592 deaths during the study period. Ageadjusted mortality rates were significantly higher among those patients with (relative to those without) a prior cardiovascular event (incident rate ratio, 1.95; 95% confidence interval [95% CI], 1.77 to 2.15; P,0.001). In the whole study population, no association between bisphosphonate treatment and age-adjusted mortality risk was observed (incident rate ratio, 0.98; 95% CI, 0.88 to 1.08; P=0.67). When stratifying by prior cardiovascular event status, the age-adjusted mortality rate associated with bisphosphonate treatment among those patients with a prior cardiovascular event was 15% higher relative to those patients not treated (incident rate ratio, 1.15; 95% CI, 0.99 to 1.34; P=0.06). Among those patients without a history of cardiovascular disease, no clear association between bisphosphonate use and ageadjusted mortality risk was observed (incidence rate ratio, 0.96; 95% CI, 0.83 to 1.10; P=0.57). Table 3 shows the unadjusted and fully adjusted Cox proportional mortality hazards associated with bisphosphonate treatment for the primary analysis (approach 1) and each of the three sensitivity analyses (approaches 2–4). In the primary adjusted analysis, bisphosphonate treatment was associated with a higher risk of death among those patients with but not without a history of major cardiovascular event (similar findings were observed when the analysis was limited to those patients with a history of congestive heart failure and also when missing values for serum calcium, albumin, and HDL and LDL cholesterol were handled using dummy variable assignment instead of multiple imputation; data not shown). The interaction was significant (P=0.004), suggesting that a prior major cardiovascular event modifies the impact of treatment with bisphosphonates in this population. The results differed under the conditions of the sensitivity analyses. For the analysis focusing on new initiators of bisphosphonates (approach 2), 1691 (25%) patients from the primary analysis cohort were excluded, because they were on treatment with a bisphosphonate at the time of the study index date, leaving 5065 subjects for the analysis. These subjects were followed for a median (interquartile range) of 3.5 (1.7–5.7) years. As shown in Table 3, the trend toward a beneficial association of bisphosphonate treatment and mortality risk among those patients without a prior history of a major cardiovascular event is weaker in this analysis, whereas the adjusted risk persists for those
Clin J Am Soc Nephrol 9: 874–880, May, 2014
Bisphosphonates and Mortality, Perkins et al.
877
Table 1. Baseline characteristics of women with CKD by cardiovascular event status
Prior Heart Disease Variable Age at index (yr), median (IQR) Bisphosphonate prescription, n (%) Prevalent Incident None Caucasian, n (%) Smoking status, n (%) Never Past Current Thyroid disorder, n (%) Osteoporosis, n (%) Any rheumatologist visits in the past year, n (%) Aspirin, n (%) Clopidogrel, n (%) b-Blockers, n (%) Statins, n (%) Diabetes, n (%) Cardiac dysrhythmia, n (%) Chronic obstructive pulmonary disease, n (%) Peripheral arterial disease, n (%) Atherosclerotic coronary artery disease, n (%) Proteinuria, n (%) Not ordered Negative Positive Body mass index (kg/m2), median (IQR) Systolic BP (mmHg), median (IQR) Diastolic BP (mmHg), median (IQR) Albumin (g/dl), median (IQR) LDL (mg/dl), median (IQR) HDL (mg/dl), median (IQR) eGFR (ml/min per 1.73 m2), median (IQR) Number of visits to PCP 12 mo before index date, median (IQR)
Yes (n=2173)
No (n=4583)
75.1 (68.9–79.2)
73.9 (67.4–78.4)
534 (24.6) 282 (13.0) 1357 (62.4) 2148 (98.8)
1157 (25.2) 652 (14.2) 2774 (60.5) 4525 (98.7)
1137 (57.3) 664 (33.5) 182 (9.2) 660 (30.4) 643 (29.6) 266 (12.2) 350 (16.1) 264 (12.2) 1068 (49.2) 943 (43.4) 1153 (53.1) 739 (34.0) 428 (19.7) 387 (17.8) 943 (43.4)
2703 (64.7) 1100 (26.3) 374 (9.0) 1309 (28.6) 1452 (31.7) 512 (11.2) 477 (10.4) 134 (2.9) 1644 (35.9) 1635 (35.7) 1767 (38.6) 511 (11.2) 417 (9.1) 351 (7.7) 798 (17.4)
925 (42.6) 797 (36.7) 451 (20.8) 31.1 (27.3–36.7) 128.0 (116–140) 70.0 (60–76) 4.1 (3.8–4.3) 89.0 (70–113) 49.0 (41–58) 53.7 (46.2–57.2) 5.0 (3–7)
2375 (51.8) 1513 (33.0) 695 (15.2) 31.1 (27.1–35.8) 130.0 (120–142) 70.0 (64–80) 4.1 (3.9–4.3) 100.0 (79–126) 52.0 (44–62) 54.2 (47.8–57.6) 4.0 (2–6)
P Value ,0.001 0.24
0.69 ,0.001
0.13 0.08 0.20 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001
0.02 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001 ,0.001
IQR, interquartile range; PCP, primary care physician.
patients with a prior history of cardiovascular event treated with bisphosphonates. The interaction by the presence of such a history was not statistically significant (P=0.07). In the second sensitivity analysis (approach 3), mortality risk attributable to treatment with a bisphosphonate was limited to the period beginning 1 year after the date of first bisphosphonate prescription. The approach 2 cohort was, therefore, modified by deaths occurring during this 1-year lag period; 184 deaths in bisphosphonate-treated subjects occurred during this lag period. Median follow-up was 2.5 years. Regardless of cardiovascular disease history, bisphosphonate treatment was associated with an increased risk of death in the fully adjusted analysis. The association was not statistically significant for those patients without cardiovascular disease, and the strength of the association was stronger for those patients with preexisting cardiovascular disease. The interaction term was nonsignificant (Table 3). In the final sensitivity analysis (approach 4), in which a propensity score for the likelihood of bisphosphonate treatment was used to adjust the Cox proportional hazard
model, the approach 2 cohort was again used as the initial study population. This cohort was further modified by characterization of treated patients at the time of bisphosphonate initiation as opposed to the study index date used in the primary analysis; 403 patients were excluded, resulting in a sample size of 4662 patients. For those patients without preexisting cardiovascular disease, bisphosphonate treatment was associated with a reduced mortality risk in this analysis (hazard ratio, 0.55; 95% CI, 0.37 to 0.82) but not in those patients with established cardiovascular disease. The interaction term was statistically significant, suggesting that cardiovascular disease modifies the risk association (Table 3).
Discussion In this study of women with CKD, the association between bisphosphonate treatment and mortality risk was inconclusive across a series of analyses designed to account for various types of selection and indication bias. The pattern of these associations would suggest that the
878
Clinical Journal of the American Society of Nephrology
Table 2. Baseline characteristics of women with CKD by bisphosphonate treatment status
Bisphosphonate Use Variable
Age at index (yr), median (IQR) Prior heart disease, n (%) Caucasian, n (%) Smoking status, n (%) Never Past Current Thyroid disorder, n (%) Osteoporosis, n (%) Any rheumatologist visits in the past year, n (%) Aspirin, n (%) Clopidogrel, n (%) b-Blockers, n (%) Statins, n (%) Diabetes, n (%) Cardiac dysrhythmia, n (%) Chronic obstructive pulmonary disease, n (%) Peripheral arterial disease, n (%) Atherosclerotic coronary artery disease, n (%) Proteinuria, n (%) Not ordered Negative Positive Body mass index (kg/m2), median (IQR) Systolic BP (mmHg), median (IQR) Diastolic BP (mmHg), median (IQR) Albumin (g/dl), median (IQR) LDL (mg/dl), median (IQR) HDL (mg/dl), median (IQR) eGFR (ml/min per 1.73 m2), median (IQR) Number of visits to PCP 12 mo before index date, median (IQR)
P Valuea
Prevalent (n=1691)
Incident (n=934)
None (n=4131)
76.3 (72.2–80.0) 534 (31.6) 1680 (99.4)
75.0 (70.1–78.5) 282 (30.2) 928 (99.4)
73.1 (65.4–77.9) 1357 (32.8) 4065 (98.4)
965 (60.3) 519 (32.4) 116 (7.2) 508 (30.0) 1477 (87.3) 249 (14.7)
550 (63.9) 248 (28.8) 63 (7.3) 278 (29.8) 250 (26.8) 71 (7.6)
2325 (62.8) 997 (27.0) 377 (10.2) 1183 (28.6) 368 (8.9) 163 (4.0)
0.51 ,0.001 ,0.001
207 (12.2) 105 (6.2) 688 (40.7) 710 (42.0) 549 (32.5) 335 (19.8) 259 (15.3)
110 (11.8) 54 (5.8) 352 (37.7) 343 (36.7) 338 (36.2) 166 (17.8) 113 (12.1)
510 (12.4) 239 (5.8) 1672 (40.5) 1525 (36.9) 2033 (49.2) 749 (18.1) 473 (11.4)
0.89 0.81 0.25 ,0.001 ,0.001 0.27 ,0.001
191 (11.3) 453 (26.8)
82 (8.8) 238 (25.5)
465 (11.3) 1050 (25.4)
0.08 0.54
,0.001 0.24 0.003 ,0.001
,0.001 830 (49.1) 666 (39.4) 195 (11.5) 29.1 (25.4–33.3) 129.0 (119–140) 70.0 (62–78) 4.2 (3.9–4.4) 95.0 (76–120) 54.0 (46–64) 54.6 (48.9–57.6) 4.0 (3–6)
521 (55.8) 306 (32.8) 107 (11.5) 30.8 (26.8–35.5) 130.0 (120–142) 70.0 (64–80) 4.1 (3.9–4.3) 98.0 (80–124) 53.0 (45–63.5) 54.3 (47.9–58.1) 4 (2–6)
1949 (47.2) 1338 (32.4) 844 (20.4) 31.5 (28.0–37.3) 130.0 (120–142) 70.0 (64–78) 4.1 (3.8–4.3) 95.0 (75–122) 50.0 (42–59) 53.9 (46.5–57.4) 4 (2–6)
,0.001 ,0.001 ,0.001 ,0.001 0.09 ,0.001 ,0.001 ,0.001
a Kruskal–Wallis test for difference across groups. A significant value (,0.05) indicates that at least one group differs from another group among the three groups. IQR, interquartile range; PCP, primary care physician.
mortality risk association with bisphosphonate among women with CKD and a history of cardiovascular disease may differ from similar women without cardiovascular disease, which was evidenced by the statistically significant interaction term in two of four approaches. Both a protective effect of treatment (among women without clinically apparent cardiovascular disease) and an increased risk of treatment (among women with cardiovascular disease) were observed in some but not all of the analyses in this study; consistency across all analyses was not seen. No definitive conclusions regarding risk association can, therefore, be drawn. The varying results across the series of approaches provide cautionary evidence for the influence of selection and indication biases in assessing drug treatment effects in an observational setting. Across two of four analytic approaches in this study, we observed a statistically significant mortality risk among
bisphosphonate-treated subjects with a history of prior cardiovascular disease. The reasons for this finding cannot be elucidated given the nature of our study design. However, some reports have identified an increased risk of atrial fibrillation among bisphosphonate users, a finding not confirmed by the US Food and Drug Administration (6–8,20). Other studies have identified an association between low bone turnover—a bisphosphonate effect—and vascular calcification, an adverse outcome that might potentially be amplified by delayed drug clearance among those patients with CKD (21,22). However, a systematic review of clinical trials examining the impact of bisphosphonate treatment on coronary and aortic calcification and carotid arterial thickness found inconsistent associations among the nine reported trials included (23). Although our analysis does not confirm higher risk, the associations of treatment with risk among those patients with
Clin J Am Soc Nephrol 9: 874–880, May, 2014
Bisphosphonates and Mortality, Perkins et al.
879
Table 3. Cox proportional mortality hazard for women with CKD prescribed bisphosphonate therapy by cardiovascular event status
HR (95% CI)
P Value
HR (95% CI)
P Value
P Value for Interaction (Bisphosphonate and Prior CV Event)
0.97 (0.85 to 1.12) 0.90 (0.78 to 1.04)
0.71 0.15
1.17 (1.01 to 1.36) 1.22 (1.04 to 1.42)
0.04 0.01
0.07 0.004
1.01 (0.83 to 1.24) 0.95 (0.78 to 1.17)
0.89 0.65
1.15 (0.93 to 1.43) 1.25 (1.01 to 1.57)
0.21 0.05
0.40 0.07
1.27 (1.02 to 1.58) 1.21 (0.97 to 1.51)
0.03 0.09
1.33 (1.05 to 1.68) 1.48 (1.16 to 1.88)
0.02 0.001
0.78 0.22
0.93 (0.70 to 1.23) 0.55 (0.37 to 0.82)
0.60 0.003
0.85 (0.64 to 1.13) 0.94 (0.66 to 1.34)
0.26 0.73
0.65 0.05
No Prior CV Event Model Approach 1 Crude Adjusteda Approach 2 Crude Adjusteda Approach 3 Crude Adjustedb Approach 4 Crude Adjustedc
Prior CV Event
CV, cardiovascular; HR, hazard ratio; 95% CI, 95% confidence interval. a Adjusted for age; smoking status; prior CV event (myocardial infarction, hospitalization for congestive heart failure, or stroke); history of cardiac dysrhythmia, atherosclerotic coronary artery disease, diabetes, or chronic obstructive pulmonary disease; HMG-CoA reductase inhibitor prescription; body mass index; systolic BP; and baseline eGFR, serum albumin, urine protein excretion, and LDL and HDL cholesterol. b Adjusted for the above covariates (with the exception of LDL cholesterol), prescription for b-blocker or aspirin, and history of a dualenergy x-ray absorptiometry (DEXA) scan in the year before index date. c Propensity score-adjusted approach. The propensity score was derived from the following variables: age, body mass index, height, systolic and diastolic BP, glucocorticoid therapy at any time during the 12 months before the index date, HMG-CoA reductase inhibitor prescription, b-blocker prescription, clopidogrel prescription, aspirin use, history of smoking, hyper- or hypothyroidism, osteopenia, osteoporosis, diabetes, history of prior cardiovascular event, eGFR, LDL and HDL cholesterol, serum albumin, proteinuria, DEXA study in the prior 3 years, number of visits with primary care physician in the prior 12 months, referral to a rheumatologist in the prior 12 months, and number of visits with a rheumatologist in the prior 12 months.
CKD and cardiovascular disease would warrant additional investigation and does support a cautious approach to the use of these medications in this population. Animal models show clearly that bisphosphonates accumulate and show activity in calcified arterial vessels; whether this result, in turn, increases risk is not clear (24,25). In a previous study, we reported a modest protective association of bisphosphonates in a similar study population without a prior history of a major cardiovascular event but with a different study enrollment period, inclusion criteria, and follow-up period (10). In the analysis presented here, the derivation of the study population across the sensitivity analyses had a considerable impact on risk associations; in three of four approaches, the protective association was again observed among those patients without cardiovascular disease, although this association was statistically significant in only the analysis adjusting for treatment indication (approach 4). Observational studies designed to address drug effects— both adverse and intended—are complicated by multiple potential sources of bias. It is not possible to determine which of the approaches applied in this observational study is least biased. Although the risk estimates suggest potential relationships, conclusions regarding definitive risk or benefit await additional study. Accounting for cumulative exposure, graded dose effects, and intraclass drug differences represents additional challenges when examining adverse drug effects observationally. Despite a growing understanding of the atypical pathophysiology
of vascular calcification among those patients with CKD, no targeted therapies to reduce cardiovascular risk exist for this population of patients beyond those therapies available to the general population. Interventional trials may help untangle the multitude of factors that influence outcomes in this unique population, and the results of this observational study, in which both risk and benefit are seen depending on cardiovascular disease burden, may justify the effort and resources required to carry out such trials. This study has additional limitations beyond its observational design. Nonadherence or changes in treatment plan after the first bisphosphonate prescription are not accounted for in this analysis to include discontinuation of therapy, which may result in bias to the null. Similarly, bisphosphonate dose was not accounted for in this study, and potential intraclass differences among the individual agents may exist, which in turn, may differentially impact mortality risk. This study was not designed to assess this potentially important heterogeneity. Exposure misclassification because of treatment occurring outside the Geisinger system is an additional potential limitation, although it is likely minimal because of Geisinger referral patterns for the primary care population. Follow-up time was modest and may have been inadequate to detect longer-term effects. Also, because the definition of CKD was based on eGFR, additional examination is needed to determine whether our results can be generalizable to those patients with CKD based on albuminuria. Confounding by factors not included in the model is an additional potential limitation; two potentially influential variables not available
880
Clinical Journal of the American Society of Nephrology
for this analysis include a history of bone fracture and information about serum vitamin D levels. For example, if bisphosphonate treatment indication was based on the occurrence of pathologic bone fracture for a substantial proportion of study subjects, this confounding might mask a protective effect of these agents, given the strong mortality risk associated with such events. In conclusion, bisphosphonate treatment among women with CKD was not consistently significantly associated with mortality, regardless of the presence of a prior cardiovascular event. Risk estimates varied across sensitivity analyses, suggesting the need for careful attention to bias when designing observational drug studies. Interventional trials likely are needed to elucidate the true nature of any risk or benefit relationship with these drugs among those patients with CKD, and they may be warranted in light of these results. Acknowledgments This study was presented in modified form at the American Society of Nephrology Annual Meeting in San Diego, CA (November 3, 2012). Disclosures None. References 1. Blacher J, Guerin AP, Pannier B, Marchais SJ, London GM: Arterial calcifications, arterial stiffness, and cardiovascular risk in end-stage renal disease. Hypertension 38: 938–942, 2001 2. London GM: Cardiovascular calcifications in uremic patients: Clinical impact on cardiovascular function. J Am Soc Nephrol 14 [Suppl 4]: S305–S309, 2003 3. Ott SM: Bisphosphonate safety and efficacy in chronic kidney disease. Kidney Int 82: 833–835, 2012 4. Toussaint ND, Lau KK, Strauss BJ, Polkinghorne KR, Kerr PG: Effect of alendronate on vascular calcification in CKD stages 3 and 4: A pilot randomized controlled trial. Am J Kidney Dis 56: 57–68, 2010 5. Ariyoshi T, Eishi K, Sakamoto I, Matsukuma S, Odate T: Effect of etidronic acid on arterial calcification in dialysis patients. Clin Drug Investig 26: 215–222, 2006 6. Cummings SR, Schwartz AV, Black DM: Alendronate and atrial fibrillation. N Engl J Med 356: 1895–1896, 2007 7. Abrahamsen B, Eiken P, Brixen K: Atrial fibrillation in fracture patients treated with oral bisphosphonates. J Intern Med 265: 581–592, 2009 8. Black DM, Delmas PD, Eastell R, Reid IR, Boonen S, Cauley JA, Cosman F, Lakatos P, Leung PC, Man Z, Mautalen C, Mesenbrink P, Hu H, Caminis J, Tong K, Rosario-Jansen T, Krasnow J, Hue TF, Sellmeyer D, Eriksen EF, Cummings SR; HORIZON Pivotal Fracture Trial: Once-yearly zoledronic acid for treatment of postmenopausal osteoporosis. N Engl J Med 356: 1809–1822, 2007 9. Armstrong K: Methods in comparative effectiveness research. J Clin Oncol 30: 4208–4214, 2012 10. Hartle JE, Tang X, Kirchner HL, Bucaloiu ID, Sartorius JA, Pogrebnaya ZV, Akers GA, Carnero GE, Perkins RM: Bisphosphonate therapy, death, and cardiovascular events among female patients with CKD: A retrospective cohort study. Am J Kidney Dis 59: 636–644, 2012
11. Foreman AJ, Lai GP, Miller DP: Surviving left truncation using PROC PHREG. Presented at the Western Users of SAS Software Meeting, University City, CA, November 5, 2008 12. Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF 3rd, Feldman HI, Kusek JW, Eggers P, Van Lente F, Greene T, Coresh J; CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration): A new equation to estimate glomerular filtration rate. Ann Intern Med 150: 604–612, 2009 13. Myers GL, Miller WG, Coresh J, Fleming J, Greenberg N, Greene T, Hostetter T, Levey AS, Panteghini M, Welch M, Eckfeldt JH; National Kidney Disease Education Program Laboratory Working Group: Recommendations for improving serum creatinine measurement: A report from the laboratory working group of the National Kidney Disease Education Program. Clin Chem 52: 5– 18, 2006 14. Lafrance JP, Miller DR: Acute kidney injury associates with increased long-term mortality. J Am Soc Nephrol 21: 345–352, 2010 15. National Technical Information Service: Available at: http:// www.ntis.gov/. Accessed March 5, 2013 16. Rubin DB: Multiple imputation. In: Multiple Imputation for Nonresponse in Surveys, New York, John Wiley & Sons, 1987, pp 15–19 17. Cohen J, Cohen P: Missing data. In: Applied Multiple Regression; Correlation Analysis for the Behavioral Sciences, Hillsdale, NJ, Erlbaum, 1983, pp 303–320 18. McCaffrey DF, Ridgeway G, Morral AR: Propensity score estimation with boosted regression for evaluating causal effects in observational studies. Psychol Methods 9: 403–425, 2004 19. Cole SR, Herna´n MA: Adjusted survival curves with inverse probability weights. Comput Methods Programs Biomed 75: 45– 49, 2004 20. FDA: Update of Safety Review Follow-up to the October 1, 2007 Early Communication about the Ongoing Safety Review of Bisphosphonates, 2008. Available at: http://www.fda.gov/Drugs/ DrugSafety/PostmarketDrugSafetyInformationforPatientsandProviders/DrugSafetyInformationforHeathcareProfessionals/ucm136201.htm. Accessed March 1, 2013 21. London GM, Marty C, Marchais SJ, Guerin AP, Metivier F, de Vernejoul MC: Arterial calcifications and bone histomorphometry in end-stage renal disease. J Am Soc Nephrol 15: 1943–1951, 2004 22. Elmariah S, Delaney JA, O’Brien KD, Budoff MJ, Vogel-Claussen J, Fuster V, Kronmal RA, Halperin JL: Bisphsophonate use and prevalence of valvular and vascular calcification in women MESA (the Multi-Ethnic Study of Atherosclerosis). J Am Coll Cardiol 56: 1752–1759, 2010 23. Santos LL, Cavalcanti TB, Bandeira FA: Vascular effects of bisphosphonates-a systematic review. Clin Med Insights Endocrinol Diabetes 5: 47–54, 2012 24. Ylitalo R, Mo¨nkko¨nen J, Urtti A, Ylitalo P: Accumulation of bisphosphonates in the aorta and some other tissues of healthy and atherosclerotic rabbits. J Lab Clin Med 127: 200–206, 1996 25. Abedin M, Tintut Y, Demer LL: Vascular calcification: Mechanisms and clinical ramifications. Arterioscler Thromb Vasc Biol 24: 1161–1170, 2004 Received: July 23, 2013 Accepted: January 22, 2014 Published online ahead of print. Publication date available at www. cjasn.org.
