- Email: [email protected]
Thiazolidinedione use is associated with better survival in hemodialysis patients with non-insulin dependent diabetes
original article
http://www.kidney-international.org & 2009 International Society of Nephrology
Thiazolidinedione use is associated with better survival in hemodialysis patients with non-insulin dependent diabetes Steven M. Brunelli1,2, Ravi Thadhani3, T. Alp Ikizler4 and Harold I. Feldman1,2 1
Renal, Electrolyte and Hypertension Division, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania, USA; 2Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania, USA; 3Renal Unit, Department of Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts, USA and 4Division of Nephrology, Vanderbilt University Medical Center, Nashville, Tennessee, USA
Cardiovascular mortality is especially high among dialysis patients with diabetes, as is morbidity due to protein energy wasting. Given that both of these factors may be decreased by thiazolidinedione treatment, we studied the effect of thiazolidinedione use on survival among chronic dialysis patients in a national cohort of 5290 incident dialysis patients with diabetes. Thiazolidinedione use was assessed according to prescription data, and the analyses were stratified based on insulin use due to observed interaction. In the primary analysis, thiazolidinedione treatment was associated with significantly lower all-cause mortality among insulin-free but not insulin-requiring subjects, with adjusted hazards ratios of 0.53 (0.31–0.89) and 0.82 (0.46–1.47) respectively. Sensitivity analyses found the findings to be robust with respect to confounding by indication, severity of the diabetes, potential reverse causality, and time varying exposure patterns. The mechanism of this decline in all-cause mortality will need to be examined after these studies are confirmed. Kidney International (2009) 75, 961–968; doi:10.1038/ki.2009.4; published online 4 February 2009 KEYWORDS: diabetes; dialysis; epidemiology; mortality; survival thiazolidinediones
There are over 340,000 patients on chronic hemodialysis (HD) in the United States; of these over 40% have diabetes (DM).1 The mortality rate among HD patients is exceedingly high (228.9 deaths/1000 patient-years), and higher yet among those with co-existing DM (251.4 deaths/1000 patient-years).1 Cardiovascular (CV) disease is the leading cause of death among HD patients.2 Protein energy wasting is a potent risk factor for death among HD patients,3 and may interact with CV disease to portend an even graver prognosis.4 HD patients with DM are known to be at a higher risk for CV disease and wasting than their non-diabetic counterparts.1,2,5,6 Thiazolidinediones (TZDs) are a class of oral diabetic medications that function by binding the peroxisome proliferator-activated receptor g, thereby increasing insulin sensitivity in peripheral tissues.7 Human data suggest that in addition to lowering serum glucose, TZDs favorably alter some CV risk factors such as increasing HDL8–17 and circulating adiponectin,7 decreasing triglycerides,12,14–17 visceral adiposity,18 circulating inflammatory mediators,19,20 and albuminuria,8 improving flow-mediated vasodilatation,21 and blunting carotid intima-media thickening.22,23 In addition, experiments in animal models suggest that TZDs may blunt muscle catabolism.24 Among human patients with DM and chronic kidney disease not on HD, treatment with TZDs is associated with a trend toward lower all-cause mortality.25 Given that susceptibility to CV disease and wasting is higher among patients on HD than among patients not yet on HD, it is plausible that TZD treatment might be particularly effective among HD patients. Thus, we conducted this retrospective study among Accelerated Mortality on Renal Replacement (ArMORR)26 cohort participants with diabetes to test that hypothesis that treatment with TZDs is associated with a decreased incidence in all-cause mortality.
Correspondence: Steven M. Brunelli, Renal Division, Brigham and Women’s Hospital, Harvard Medical School, 75 Francis Street, MRB-4, Boston, MA 02115, USA. E-mail: [email protected]
RESULTS Description of cohort
Received 23 September 2008; revised 4 November 2008; accepted 9 December 2008; published online 4 February 2009
Of 10,044 patients in the ArMORR cohort, 5290 were diabetic and survived at least until HD day 30, and were
Kidney International (2009) 75, 961–968
961
original article
SM Brunelli et al.: TZD and mortality in dialysis
included in the primary cohort. Demographic, comorbid disease and laboratory characteristics of the cohort according to baseline TZD and insulin exposure are given in Table 1. Among non-insulin requiring subjects, those receiving TZDs had significantly higher body mass indices, serum albumin levels, and facility standardized mortality ratios than those not receiving TZDs. Among insulin-requiring subjects, those exposed to TZDs were significantly older than those not. Within each insulin exposure stratum, TZD-exposed and unexposed subjects were otherwise similar. Pre-dialysis random glucose levels were similar among TZD-exposed and -unexposed patients (P ¼ 0.4); glycosylated hemoglobin levels were not available. Hemoglobin levels and equilibrated KT/V were similar among TZD exposure groups at baseline, and at all subsequent time points (data not shown). Serum
albumin was significantly higher among TZD-exposed (versus unexposed) subjects at days 90 (Po0.001) and 180 (P ¼ 0.05), but not at day 270 (P ¼ 0.2). Over the first month of HD, 9.6% patients were treated with a TZD, 39.8% with insulin, and 16.9% with a sulfonylurea (Table 2). The proportion of patients treated with zero, one, and two or more classes of anti-diabetic agents were 40.6, 50.2, and 9.2% respectively. b adrenergic antagonists were prescribed in 48.8%, and antagonists of the renin–angiotensin–aldosterone axis in 39.1%. Baseline survival analyses
In the primary analysis, exposures were assessed over HD days zero through 30, and survival measured following day 30 (corresponding to the start of at-risk time). A total of 719
Table 1 | Baseline characterization of subjects stratified by insulin and TZD exposure TZD INS (n=2832)
TZD+ INS (n=353)
TZD INS+ (n=1953)
TZD+ INS+ (n=152)
Age (years); mean (s.d.) Men; n (%) Non-white; n (%) Hypertension; n (%) Congestive heart failure; n (%) Arterial disease; n (%) Cirrhosis; n (%)
64.0 1499 1141 1203 409 480 83
(13.3) (52.9%) (40.3%) (42.5%) (14.4%) (17.0%) (2.9%)
64.4 187 147 156 47 51 11
(11.8) (53.0%) (41.6%) (44.2%) (13.3%) (14.5%) (3.1%)
61.1 1031 722 827 308 348 62
(13.2)* (52.8%) (37.0%) (42.4%) (15.8%) (17.8%) (3.2%)
63.9 72 54 75 21 26 4
(11.1)* (47.4%) (35.5%) (49.3%) (13.8%) (17.1%) (2.6%)
Body mass index (kg/m2); n (%) o20 20–25 25–30 30–35 435
233 836 826 464 467
** (8.2%) (29.6%) (29.2%) (16.4%) (16.5%)
11 91 94 74 83
** (3.1%) (25.8%) (26.6%) (21.0%) (23.5%)
142 489 549 387 380
(7.2%) (25.1%) (28.2%) (19.9%) (19.5%)
10 25 43 32 41
(6.6%) (16.6%) (28.5%) (21.2%) (27.2%)
Facility standardized mortality ratio; n (%) o0.75 0.75–1 1–1.25 41.25 Serum albumin (g/100 ml); mean (s.d.) Serum creatinine (mg/100 ml); mean (s.d.) Serum phosphate (mg/100 ml); mean (s.d.)
632 765 643 768 3.4 5.9 4.7
*** (22.5%) (27.2%) (22.9%) (27.4%) (0.5)** (2.4) (1.5)
72 112 92 75 3.6 5.8 4.5
*** (20.5%) (31.9%) (26.2%) (21.4%) (0.5)** (2.4) (1.5)
470 587 449 426 3.4 5.7 4.6
(24.3%) (30.4%) (23.2%) (22.1%) (0.5) (2.1) (1.5)
39 (26.4%) 54 (36.5%) 25 (16.9%) 30 (20.3%) 3.4 (0.5) 5.4(2.3) 4.6 (1.6)
INS, insulin; TZD, thiazolidinedione. *P=0.01; **Po0.001; ***P=0.03.
Table 2 | Characterization of therapies and baseline and selected follow-up time pointsa Agent TZD a glucosidase inhibitor Biguanide Insulin Meglitinide Sulfonylurea
Day 0 (n=5290) 505 14 19 2105 140 896
Day 90 (n=4598)
Day 180 (n=4180)
Day 270 (n=3804)
(9.6%) (0.3%) (0.4%) (39.8%) (2.7%) (16.9%)
527 11 12 2155 133 877
(11.5%) (0.2%) (0.3%) (47.9%) (2.9%) (19.1%)
557 12 16 2141 131 851
(13.3%) (0.3%) (0.4%) (51.2%) (3.1%) (20.1%)
551 11 15 2009 125 813
(14.5%) (0.3%) (0.4%) (52.8%) (3.3%) (21.4%)
Number of classes of anti-diabetic agents 0 2150 (40.6%) 1 2653 (50.2%) 42 487 (9.2%) b adrenergic antagonist 2581 (48.8%) RAAS antagonist 2069 (39.1%)
1436 2669 493 2599 2168
(31.2%) (58.1%) (10.7%) (56.5%) (47.2%)
1098 2520 562 2558 2210
(26.3%) (60.3%) (13.4%) (61.2%) (52.9%)
926 2305 573 2404 2151
(24.3%) (60.6%) (15.1%) (63.2%) (56.5%)
RAAS, renin–angiotensin–aldosterone. a Reported as n (%).
962
Kidney International (2009) 75, 961–968
original article
SM Brunelli et al.: TZD and mortality in dialysis
Table 3 | Mortality rates and unadjusted incidence rate ratios for all-cause mortality by diabetic treatment
P interaction = 0.02 IRR 0.51 (0.33 to 0.74)
IRR 1.03 (0.62 to 1.62)
Class of agent TZD Unexposed Exposed Insulin Unexposed Exposed
Mortality rate (95% CI) per 1000 patient-years
Incidence rate ratio (95% CI)
189.8 (176.0–204.7) 124.0 (93.7–164.1)
0.65 (0.48–0.87) P=0.002
194.5 (177.4–213.1) 166.3 (147.4–187.8)
0.86 (0.73–1.00) P=0.04
Mortality rate (per 1000 patient-years)
300 250 200 150 100 50 0 TZD- INS-
Sulfonylurea Unexposed Exposed
183.0 (168.9–198.3) 184.1 (154.4–219.5)
1.01 (0.82–1.22) P40.5
Meglitinide Unexposed Exposed
182.3 (169.2–196.3) 219.4 (143.1–336.5)
1.20 (0.74–1.85) P=0.4
TZD+ INS-
TZD- INS+
TZD+ INS+
Figure 1 | Unadjusted mortality rate according to TZD and insulin treatment status. Incidence rate ratios for TZD exposure stratified by insulin exposure are given in the figure. Considered alternately, the IRR (95% CI) for insulin exposure is 0.80 (0.68–0.94) among TZD-unexposed subjects, and 1.64 (0.88–2.99) among TZDexposed subjects (P interaction ¼ 0.02).
719 Deaths; 3925.3 patient-years
Kidney International (2009) 75, 961–968
Insulin –
1.5 Hazard ratio (95% Cl)
deaths occurred over 3925.3 patient-years of at-risk time; crude mortality rate (95% CI) was 183.2 (170.3–197.1) deaths/1000 patient-years. Median survival time was 335 days. On bivariable analysis, the crude mortality rate was significantly lower among TZD-exposed subjects than among non-TZD exposed subjects: IRR (95% CI) 0.65 (0.48–0.87) (Table 3). Survival was similar regardless of TZD agent used (P ¼ 0.45; Figure S1). The IRRs (95% CIs) for all-cause mortality were 0.86 (0.73–1.00), 1.01 (0.82–1.22) and 1.20 (0.74–1.85) according to exposure to insulin, sulfonylureas, and meglitinides, respectively (Table 3). The data were too scant to examine relationships on the basis of other classes of diabetic medications. Stratified analysis was conducted to examine the effect modification of the TZD – mortality association on the basis of insulin exposure (hypothesized a priori). Among subjects not receiving insulin, TZD exposure was associated with reduced all-cause mortality: IRR (95% CI) 0.51 (0.33–0.74); among subjects receiving insulin, there was no association between TZD exposure and all-cause mortality; IRR (95% CI) 1.03 (0.62–1.62; P for interaction ¼ 0.02; Figure 1). Based on this finding, all subsequent analyses were stratified on insulin exposure status. (Considered alternately, the IRR (95% CI) for insulin exposure was 0.80 (0.68–0.94) among TZD unexposed subjects, and 1.64 (0.88–2.99) among TZDexposed subjects (P interaction ¼ 0.02)). The association with improved survival (among patients not on insulin) was specific to TZDs, and was not seen with sulfonylureas or meglitinides (Figure S2). Cox proportional hazard models were fit to examine the adjusted HR for all-cause mortality according to TZD exposure status. TZD exposure was associated with adjusted HRs (95% CIs) for all-cause mortality of 0.53 (0.31–0.89; P ¼ 0.02) and 0.82 (0.46–1.47; P40.5) among subjects not receiving and receiving insulin, respectively (Figure 2). Propensity score analysis yielded nearly identical results:
Insulin +
1
0.5
0 Unadjusted
Multivariable
Propensity
Figure 2 | Hazard ratios for all-cause mortality according to TZD exposure as estimated by Cox proportional hazards models, stratified by insulin exposure status (insulinunexposed white, insulin-exposed gray). Multivariable models were adjusted for age, sex, race, body mass index (categorized r20, 20–25, 25–30, 30–35, 435), facility standardized mortality ratio (categorized r0.75, 0.75–1.0, 1.0–1.25, 41.25), hypertension, arterial disease, congestive heart failure, cirrhosis, number of classes of diabetic medications (categorized 0, 1, Z2), b adrenergic antagonist use, renin–angiotensin–aldosterone antagonist use, and serum albumin, creatinine, and phosphate. Propensity score models were adjusted for the same covariates: number of classes of diabetic medications was adjusted as a covariate in the survival, and the remaining covariates were fit into the propensity model.
HRs (95% CIs) 0.48 (0.29–0.80; P ¼ 0.005) and 0.79 (0.44–1.41; P ¼ 0.4) for patients not receiving and receiving insulin, respectively (Figure 2). The association with improved survival (among patients not on insulin) was specific to TZDs, and was not seen with sulfonylureas or meglitinides (Table S1). Baseline sensitivity analyses
Owing to the possibility that intensity of diabetic therapy may have influenced findings, sensitivity analyses were conducted in which consideration was limited to patients on only one class of oral diabetic medication (Table 4). 963
original article
SM Brunelli et al.: TZD and mortality in dialysis
Table 4 | Results of sensitivity analyses in which single oral agent TZD exposure was compared to other single oral agent therapy, stratified by insulin exposure Off insulin TZD (n=666)
Age-adjusted HR (95% CI) all-cause mortality; TZD versus other
88/495.8 16/182.7 177.5 (144.0–218.7) 87.6 (53.7–143.0) 0.49 (0.27–0.85) P=0.006 0.56 (0.33–0.96) P=0.03
Among patients not receiving insulin, TZD exposure (versus treatment with a single oral agent of another class) was associated with an IRR (95% CI) for all-cause mortality of 0.49 (0.27–0.85); among those receiving insulin the IRR (95% CI) was 0.80 (0.39–1.56). Diminution in the number of outcomes prevented complete multivariable adjustment. Age adjusted HRs (95% CIs) for all-cause mortality according to TZD exposure were 0.56 (0.33–0.96) and 0.82 (0.43–1.55) among patients not receiving and receiving insulin, respectively. Results were similar when the comparator group was further restricted to patients receiving single oral agent therapy with a sulfonylurea (data not shown). To explore whether TZDs influenced mortality through CV mechanisms, non-CV mechanisms or both, competing risks models were fit. Among patients not on insulin, TZDs significantly reduced non-CV mortality, but not CV mortality: HRs (95% CIs) 0.35 (0.15–0.77) and 0.67 (0.36–1.24), respectively (P for difference between CV and nonCV ¼ 0.002). Among subjects not on insulin, TZD exposure did not affect either non-CV or CV mortality: HRs (95% CIs) 0.73 (0.32–1.67) and 0.88 (0.45–1.74), respectively (P for difference between CV and non-CV mortality ¼ 0.7). There was no significant difference in the risk of CV mortality according to TZD agent used (P rosiglitazone versus pioglitazone ¼ 0.6). Because of the concern that we might be observing a reverse-causal relationship (that is, that pre-terminal patients were less likely to be exposed to TZDs), sensitivity analyses were conducted in which exposure and outcome assessment was lagged by 1, 2, and 3 months (Figure 3). Among patients not receiving insulin, the multivariable adjusted HRs (95% CIs) were 0.52 (0.30–0.90), 0.56 (0.31–1.00) and 0.63 (0.34–1.18) in the 1, 2, and 3 month-lagged models, respectively. Estimates were nearly identical when propensity score adjustment was used: HRs (95% CIs) 0.47 (0.28–0.80), 0.50 (0.28–0.88), 0.56 (0.31–1.04), respectively. Among patients receiving insulin, the corresponding adjusted HRs (95% CIs) were 0.98 (0.54–1.81), 1.14 (0.60–2.18), and 1.20 (0.61–2.36). Again, estimates were nearly identical when propensity score adjustment was used: HRs (95% CIs) 0.95 (0.52–1.73), 1.12 (0.59–2.14), 1.19 (0.61–2.34), respectively. Reassuringly, the point estimate for TZD exposure was similar between baseline and lagged models, though statistical significance was lost in the 3-month lagged model, possibly due to diminution of the number of events. 964
TZD+ (n=230)
TZD (n=189)
TZD+ (n=115)
28/139.1 14/87.5 201.3 (139.0–291.6) 160.1 (94.8–270.3) 0.80 (0.39–1.56) P=0.5 0.82 (0.43–1.55) P40.5
Insulin – Insulin +
2 Adjusted HR (95% Cl)
Deaths/patient-years at risk Mortality rate (95% CI); per 1000 patient-years IRR (95%CI) all-cause mortality; TZD versus other
On insulin
1.5 1 0.5 0 1-month lag
2-month lag
3-month lag
Figure 3 | Multivariable adjusted hazard ratios for all-cause mortality according to TZD exposure in lagged analyses. In all models, exposure was defined over days zero through 30, and outcome assessment was lagged to varying degrees (x axis). Separate models were fit according to insulin exposure status (insulin-unexposed white, insulin-exposed gray). Models were adjusted for age, sex, race, body mass index (categorized r20, 20–25, 25–30, 30–35, 435), facility standardized mortality ratio (categorized r0.75, 0.75–1.0, 1.0–1.25, 41.25), hypertension, arterial disease, congestive heart failure, cirrhosis, number of classes of diabetic medications (categorized 0, 1, Z2), b adrenergic antagonist use, renin–angiotensin–aldosterone antagonist use, and serum albumin, creatinine, and phosphate using Cox proportional hazards models.
Time-updated survival analyses
Finally, because exposure (and covariate) status may vary over time, we fit time-updated models in which exposures and covariates were updated at monthly intervals. Using this approach, the adjusted odds ratio (95% CI) for all-cause mortality according to TZD exposure was 0.64 (0.43–0.96) among patients not on insulin, and 0.90 (0.59–1.38) among patients on insulin (Table 5). Analogous models in which exposure and outcome assessment lagged by 1 month yielded nearly identical estimates. DISCUSSION
Survival is very poor among patients on HD in general, particularly among those with co-existing DM.1 CV disease and protein energy wasting are potently associated with allcause HD-related mortality,2,3 which may be of particular importance among diabetic patients.5,6 In addition to their glycemic effects, TZDs have a number of pleiotropic actions on CV risk factors and muscle wasting, suggesting that TZD treatment might improve survival among diabetic HD Kidney International (2009) 75, 961–968
original article
SM Brunelli et al.: TZD and mortality in dialysis
Table 5 | Adjusted odds ratios (95% CI) for all-cause mortality according to TZD exposure status estimated by time-updated logistic regression modelsa
Off insulin On insulin
Un-lagged
1 Month lagged
0.64 (0.43–0.96) P=0.03 0.90 (0.59–1.38) P40.5
0.63 (0.42–0.95) P=0.03 0.88 (0.57–1.36) P40.5
a Adjusted for age, race, sex, body mass index (categorized p20, 20–25, 25–30, 30–35, 435), facility standardized mortality ratio (categorized p0.75, 0.75–1.0, 1.0–1.25, 41.25), hypertension, congestive heart failure, arterial disease, cirrhosis, number of classes of diabetic medication (categorized 0, 1, X2), b adrenergic antagonist use, renin–angiotensin–aldosterone antagonist use, and serum albumin, creatinine and phosphate using time-updated logistic regression models as described in the text.
patients. Using data from a national cohort of incident HD patients with DM, we demonstrated that among those not receiving insulin, TZD therapy was associated with a 47% reduction in all-cause mortality at 1 year. To our knowledge, this is the first report of the positive association between TZD therapy and survival among HD patients. Clearly, if the observed association proves causal, it would have great ramifications on survival among diabetic patients on HD. Plausibility of the observed association is suggested by results from a post hoc analysis of the PROactive trial, which demonstrated that among patients with chronic kidney disease, there was a trend towards improved survival among those randomized to TZD therapy (HR (95% CI) 0.75 (0.55–1.03)).25 However, that trial excluded patients on HD, in whom the benefits of TZD therapy might be expected to be more pronounced. The precise mechanism through which TZDs may reduce mortality among non-insulin-dependent diabetic HD patients is uncertain. CV disease (and attendant mortality) is highly prevalent among HD patients, and exceeds that which would be predicted on the basis of traditional Framingham risk factors.27 Some of this excess risk is thought to relate, in part, to the effects of chronic inflammation.28 Likewise, malnutrition and muscle wasting are common among HD patients,29 and are potent risk factors for HD-related mortality.3 The protein energy wasting syndrome is shown to occur as a result of membrane-induced complement activation,30 systemic release of inflammatory mediators,31 HD-associated nutrient loss,32,33 and poor appetite.33 Human studies demonstrate that diabetic HD patients have greater net loss of muscle protein stores,5 and greater loss of lean body mass over the first year on HD than their non-diabetic counterparts.6 Recent data suggest that the ill-effects of CV disease and protein energy wasting may be synergistic.4 TZDs bind PPAR g receptors, and the resultant receptor – ligand complexes influence transcription of a number of downstream targets.7 The net effect is improvement in glycemic control, favorable modification of some traditional (for example, HDL, triglycerides)8–17 and non-traditional (for example, fat distribution, circulating inflammatory mediators, albuminuria)8,18–20 CV risk factors, as well as decrease in insulin resistance-related muscle catabolism and wasting.24 Kidney International (2009) 75, 961–968
Our data suggest that TZDs mitigate mortality risk in non-insulin requiring HD patients through primarily nonCV mechanisms. Sensitivity analyses suggest a potent association with reduced non-CV mortality, and a more modest (and non-statistically significant) association with reduced CV mortality. Recent reports in the literature have suggested an association between TZDs in general, and rosiglitazone in particular (possibly on the basis of its adverse impacts on LDL,8–12 which are not seen with pioglitazone13–17,34) with increased rates of acute myocardial infarction,35–37 and decompensated congestive heart failure.35,37 One meta-analysis suggested that rosiglitzaone was associated with increased CV mortality;36 however, when data from this study were re-analyzed with inclusion of three large trials designed specifically to look at adverse CV outcomes, this association was no longer appreciated,37 consistent with other published data.35 It is reassuring that although our study did not detect a benefit of TZD use with respect to CV mortality, there was no suggestion of harm (that is, HR o1). Moreover, we did not appreciate a differential association between individual TZDs and allcause or CV mortality. It must be emphasized that the analyses of cause-specific mortality and differential effects of individual TZDs were exploratory in nature, and should be considered as hypothesis generating until further data become available. Moreover, we lacked data on non-fatal CV events, so no conclusions can be drawn regarding the association between TZD use and non-fatal CV events in dialysis patients. Two observations suggest that improvement in glycemic control is not the operative mechanism for the effects of TZD on mortality. One, the beneficial effects on survival were specific to TZDs, and were not seen with other classes of oral agents, which are expected to provide similar degrees of glycemic control.38 Two, the effects of TZDs on glycemic control are modest (lowering hemoglobin A1c by approximately 0.6% in diabetics on HD39), and would therefore not be expected to reduce 1-year mortality to the degree seen.40 Regardless of the mechanism, the observed association between TZD exposure and improved survival is important. These data suggest that the beneficial effects of TZDs are limited to patients not receiving insulin. The likely explanation is that exogenous insulin overwhelms insulin resistance, thereby obviating the benefits of TZDs. However, some proportion of insulin-treated patients were likely type I diabetics (data on type of DM were not available), which may have biased inference in this group. The subgroup of patients not receiving insulin should not have included any type I diabetics, thereby excluding the possibility that observations in this group (including the observed reduction in all-cause mortality related to TZD use) were biased by the type of diabetes. As with all observational study, there remains the possibility of residual confounding. Specifically, one must consider whether TZD-treated patients received these agents 965
original article
because they: (1) were less sick or (2) were receiving better care. With respect to question one, overall health may have differed across TZD groups either with respect to diabetic severity or non-diabetic illness. Assessment of diabetic severity in HD patients is difficult. In non-HD patients one commonly used marker is glycosylated hemoglobin level. However, among HD patients, the utility of glycosylated hemoglobin is limited due to shortened and variable half life of peripheral erythrocytes;41 moreover, these data were not available. Random, pre-dialysis glucose levels were similar among TZD-exposed and -unexposed patients, however, the utility of these values as markers of glycemic control is questionable. Another marker of diabetic severity is intensity of anti-diabetic therapy. Therefore, analyses were stratified on the basis of insulin therapy, and further adjusted for total number of diabetic medications. Additionally, sensitivity analyses were conducted in which analyses were restricted to subjects on single oral agent therapy, presuming that diabetic severity would be similar among TZD-exposed and unexposed members of this subgroup, all of whom received mono-therapy for their diabetes. Consistency of results of this analysis with those from the primary analysis is reassuring in this regard. With respect to non-diabetic illness, TZD-exposed and unexposed patients were similar on the basis of nearly all demographic and comorbid disease characteristics. The exceptions were that among non-insulin-receiving subjects, those TZD exposed had higher facility standardized mortality ratios (which would portend a worse prognosis), serum albumin levels and body mass indices (which would portend a better prognosis), and among insulin-requiring subjects those exposed to TZDs were older. Analyses were therefore adjusted on the basis of these variables, as well as many additional demographic, comorbid disease and laboratory variables. In addition, propensity score models were fit to reduce the influence of confounding by indication, and lagged sensitivity analyses were performed to account for the possibility that there was differential exposure to TZDs on the basis of pre-terminal illness. Consistency of results of these analyses with those from the primary analysis is again reassuring. Quality of health care was addressed in several ways. We adjusted the analyses on the basis of facility-specific mortality rates, as well as exposure to use of b adrenergic blockers and antagonists of the renin–angiotensin–aldosterone axis, which are associated with improved survival,42–44 and could have confounded results. In addition, makers of quality of dialytic care (hemoglobin level, equilibrated KT/V) were similar among TZD-exposed and -unexposed patients. As in all observational study, there was the possibility of exposure misclassification. There was no opportunity to assess patient adherence to medication, thus medication exposures were based on prescription data. This approach is consistent with intention-to-treat principles, and is conservative in that exposure misclassification would be expected to bias results toward the null. In addition, because patient 966
SM Brunelli et al.: TZD and mortality in dialysis
exposures are not static over time, we fit time-updated models, which reassuringly, yielded similar estimates to the primary analysis. Finally, this cohort consisted of diabetic patients incident to HD. Generalization to prevalent HD patients should be undertaken cautiously; generalization to non-diabetic HD patients is not advisable. In addition, assessment of survival in this study was limited to 1 year of follow up; further study is needed to clarify longer-term effects. We conclude that among incident hemodialysis patients with non-insulin requiring diabetes, TZD use was associated with significantly lower rate of all-cause mortality at 1 year of follow up. Further studies are necessary to confirm and generalize findings and to explore the causal nature of this relationship. METHODS Study population This protocol was approved by the University of Pennsylvania Institutional Review Board. We conducted a retrospective analysis of subjects in the ArMORR cohort. Details of the cohort have been previously published.26 Briefly, the ArMORR cohort consists of all patients incident to HD at any of over 1000 Fresenius Medical Care units in North America (FMC-NA) between June 2004 and August 2005 (n ¼ 10 044). We restricted inclusion to adult patients with DM, as defined by (1) attribution of kidney failure to DM (Centers for Medicare and Medicaid Services Form 2728), (2) an International Classification of Disease-9 code of DM upon HD initiation, or (3) active prescription for anti-diabetic medication during the first 30 HD days. To enable baseline characterization of TZD and covariate exposure, we further limited inclusion to subjects surviving to HD day 31 (the start of at-risk time). Study data All study data were collected prospectively. Demographic and comorbid disease characteristics were collected by study investigators at the time of HD initiation. In addition, data on each unit’s standardized mortality ratio was collected to enable adjustment on the basis of the ‘excellent-center effect.’ The standardized mortality ratio is the facility-specific mortality rate relative to all other FMCNA units, and describes facility-level mortality beyond what would be expected on the basis of age, race, gender, nutritional status, and HD adequacy. All laboratory tests were performed in a centralized laboratory (Spectra Laboratories, Rockleigh, NJ, USA). Medication data were collected at baseline, and subsequently updated at each dialysis session. No opportunity existed to assess medication adherence, thus patients were classified as exposed if they had an active prescription spanning that treatment. TZD exposure was considered to be an active prescription for either rosiglitazone (Avandia, Glaxo Smith Kline) or pioglitazone (Actos, Takeda Pharmaceutical Company Ltd.). For the purpose of analysis, medication exposure was considered monthly: a patient was considered to be exposed if he/she had an active prescription for that medication any time during the month. The primary outcome of interest was time to death from any cause. Subjects were followed prospectively until they died or were censored at the time of transfer of care away from FMC-NA, recovery of kidney function, renal transplantation, or change in dialytic modality. As of the time of these analyses, complete Kidney International (2009) 75, 961–968
original article
SM Brunelli et al.: TZD and mortality in dialysis
outcome data were available through 1 year, thus subjects were administratively censored at this time point. In the primary analyses, exposure was defined over the first 30 days following the onset of chronic HD. Exposure to TZDs was then related to survival beginning on 31st day after the onset of HD (corresponding to the start of at-risk time) and continuing until death or censoring. Statistical analysis Continuous data were analyzed graphically and by calculation of summary statistics. Between-group comparisons were conducted using the Student’s t-test. Categorical data were analyzed in terms of counts and proportions, and compared using the w2 Test. In bivariable analyses, survival was compared among groups by calculation of stratum specific (for example, TZD exposed, unexposed) mortality rates, and estimation of incidence rate ratios (IRRs). Pre-specified interaction of the TZD – mortality association on the basis of insulin exposure status was examined using stratified analysis and Mantel–Haenzel methods. Given the potent interaction identified, all subsequent analyses were stratified on the basis of insulin treatment. Adjusted hazard ratios (HRs) were estimated using Cox proportional hazards models, which included terms for pre-specified demographic, comorbid disease, laboratory, and medication covariates of interest. Separate models were fit for insulin-exposed and -unexposed patients. The proportional hazards assumption was tested graphically and by inclusion of two-way cross product terms with time. To account for potential confounding by indication, propensity score adjustment was used. Propensity models were specified as logistic regression models with TZD exposure as the response variable and covariates (aside from number of classes of diabetic medication) as predictor variables. Association between TZD and outcome was assessed using proportional hazards models with covariate terms for propensity score (as a continuous variable; linearity assumption tested graphically) and number of classes of diabetic medication. The effect of TZD use on cause-specific mortality (that is, CV, non-CV) was examined using competing risks models. These models were specified according to the method of Lunn.45 Cause of death was adjudicated by International Classification of Disease-9 code (list given in Supplementary Table S2, Supplementary Materials). A series of lagged analyses was performed in which TZD (and covariate) exposure over HD days zero through 30 was used to predict survival beginning on HD days 61, 91, and 121. Methods were otherwise analogous to those used in the primary analyses. Finally, we conducted analyses using time-updated exposure and covariate data. In these models, each patient’s experience was decomposed into individual patient-months. Time-updated logistic regression models were fit such that survival in month t was predicted on the basis of exposures in month t1. Analyses were stratified on insulin status by inclusion of a two-way cross product term with TZD exposure. Robust variance estimates were used to correct for non-independence of observations within patient.
(Satellite Dialysis Research Fellowship) to SMB. SMB had full access to all of the data and takes responsibility for the integrity and accuracy of the data analysis. SUPPLEMENTARY MATERIAL Table S1. Adjusted hazard ratios (95% CIs) for all-cause mortality according to baseline sulfonylurea use and meglitinide use, stratified by insulin exposure*. Table S2. International classification of disease-9 codes used to define cardiovascular mortality. Figure S1. Kaplan–Meier survivor function according to TZD agent used. Note that the y axis has been rescaled. Figure S2a. Unadjusted mortality rate according to sulfonylurea use, stratified on insulin treatment status. SU, sulfonylurea; INS, insulin Figure S2b. Unadjusted mortality rate according to meglitinide use, stratified on insulin treatment status. MG, meglitinide; INS, insulin Supplementary material is linked to the online version of the paper at http://www.nature.com/ki REFERENCES 1.
2. 3.
4.
5.
6.
7. 8.
9.
10.
11.
12.
13.
DISCLOSURE
14.
R. Thadhani receives grant support from Abbott and has received honoraria from Abbott and Genzyme. Neither company produces thiazolidinediones.
15.
ACKNOWLEDGMENTS
We thank Raymond Boston and Russell Localio for their advice regarding competing risks models. This study was supported in part by awards from the NIDDK (1K23DK079056), the American Heart Association (0775021N), and the National Kidney Foundation Kidney International (2009) 75, 961–968
16.
National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, U.S. Renal Data System, USRDS 2007 Annual Data Report: Atlas of Chronic Kidney Disease and End-Stage Renal Disease in the United States. http://www.usrds.org/ Accessed May 17, 2008, Bethesda, MD. Foley RN, Parfrey PS, Sarnak MJ. Clinical epidemiology of cardiovascular disease in chronic renal disease. Am J Kidney Dis 1998; 32(Suppl 3): S112–S119. Kalantar-Zadeh K, Kilpatrick RD, Kuwae N et al. Revisiting mortality predictability of serum albumin in the dialysis population: time dependency, longitudinal changes and population-attributable fraction. Nephrol Dial Transplant 2005; 20: 1880–1888. de Mutsert R, Grootendorst DC, Axelsson J et al. Excess mortality due to interaction between protein-energy wasting, inflammation and cardiovascular disease in chronic dialysis patients. Nephrol Dial Transplant 2008; 23: 2957–2964. Pupim LB, Flakoll PJ, Majchrzak KM et al. Increased muscle protein breakdown in chronic hemodialysis patients with type 2 diabetes mellitus. Kidney Int 2005; 68: 1857–1865. Pupim LB, Heimburger O, Qureshi AR et al. Accelerated lean body mass loss in incident chronic dialysis patients with diabetes mellitus. Kidney Int 2005; 68: 2368–2374. Yki-Jarvinen H. Thiazolidinediones. N Engl J Med 2004; 351: 1106–1118. Lebovitz HE, Dole JF, Patwardhan R et al. Rosiglitazone monotherapy is effective in patients with type 2 diabetes. J Clin Endocrinol Metab 2001; 86: 280–288. Fonseca V, Rosenstock J, Patwardhan R et al. Effect of metformin and rosiglitazone combination therapy in patients with type 2 diabetes mellitus: a randomized controlled trial. JAMA 2000; 283: 1695–1702. Gomez-Perez FJ, Fanghanel-Salmon G, Antonio Barbosa J et al. Efficacy and safety of rosiglitazone plus metformin in Mexicans with type 2 diabetes. Diabetes Metab Res Rev 2002; 18: 127–134. Vongthavaravat V, Wajchenberg BL, Waitman JN et al. An international study of the effects of rosiglitazone plus sulphonylurea in patients with type 2 diabetes. Curr Med Res Opin 2002; 18: 456–461. Raskin P, Rendell M, Riddle MC et al. A randomized trial of rosiglitazone therapy in patients with inadequately controlled insulin-treated type 2 diabetes. Diabetes Care 2001; 24: 1226–1232. Aronoff S, Rosenblatt S, Braithwaite S et al. Pioglitazone hydrochloride monotherapy improves glycemic control in the treatment of patients with type 2 diabetes: a 6-month randomized placebo-controlled doseresponse study. The Pioglitazone 001 Study Group. Diabetes Care 2000; 23: 1605–1611. Rosenblatt S, Miskin B, Glazer NB et al. The impact of pioglitazone on glycemic control and atherogenic dyslipidemia in patients with type 2 diabetes mellitus. Coron Artery Dis 2001; 12: 413–423. Einhorn D, Rendell M, Rosenzweig J et al. Pioglitazone hydrochloride in combination with metformin in the treatment of type 2 diabetes mellitus: a randomized, placebo-controlled study. The Pioglitazone 027 Study Group. Clin Ther 2000; 22: 1395–1409. Kipnes MS, Krosnick A, Rendell MS et al. Pioglitazone hydrochloride in combination with sulfonylurea therapy improves glycemic control in patients with type 2 diabetes mellitus: a randomized, placebo-controlled study. Am J Med 2001; 111: 10–17.
967
original article
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
968
Rosenstock J, Einhorn D, Hershon K et al. Efficacy and safety of pioglitazone in type 2 diabetes: a randomised, placebo-controlled study in patients receiving stable insulin therapy. Int J Clin Pract 2002; 56: 251–257. Miyazaki Y, Mahankali A, Matsuda M et al. Effect of pioglitazone on abdominal fat distribution and insulin sensitivity in type 2 diabetic patients. J Clin Endocrinol Metab 2002; 87: 2784–2791. Haffner SM, Greenberg AS, Weston WM et al. Effect of rosiglitazone treatment on nontraditional markers of cardiovascular disease in patients with type 2 diabetes mellitus. Circulation 2002; 106: 679–684. Marx N, Froehlich J, Siam L et al. Antidiabetic PPAR gamma-activator rosiglitazone reduces MMP-9 serum levels in type 2 diabetic patients with coronary artery disease. Arterioscler Thromb Vasc Biol 2003; 23: 283–288. Caballero AE, Saouaf R, Lim SC et al. The effects of troglitazone, an insulinsensitizing agent, on the endothelial function in early and late type 2 diabetes: a placebo-controlled randomized clinical trial. Metabolism 2003; 52: 173–180. Sidhu JS, Kaposzta Z, Markus HS et al. Effect of rosiglitazone on common carotid intima-media thickness progression in coronary artery disease patients without diabetes mellitus. Arterioscler Thromb Vasc Biol 2004; 24: 930–934. Mazzone T, Meyer PM, Feinstein SB et al. Effect of pioglitazone compared with glimepiride on carotid intima-media thickness in type 2 diabetes: a randomized trial. Jama 2006; 296: 2572–2581. Wang X, Hu Z, Hu J et al. Insulin resistance accelerates muscle protein degradation: Activation of the ubiquitin-proteasome pathway by defects in muscle cell signaling. Endocrinology 2006; 147: 4160–4168. Schneider CA, Ferrannini E, Defronzo R et al. Effect of pioglitazone on cardiovascular outcome in diabetes and chronic kidney disease. J Am Soc Nephrol 2008; 19: 182–187. Teng M, Wolf M, Ofsthun MN et al. Activated injectable vitamin D and hemodialysis survival: a historical cohort study. J Am Soc Nephrol 2005; 16: 1115–1125. Longenecker JC, Coresh J, Powe NR et al. Traditional cardiovascular disease risk factors in dialysis patients compared with the general population: the CHOICE Study. J Am Soc Nephrol 2002; 13: 1918–1927. Krane V, Drechsler C, Wanner C. Dyslipidemia, inflammation and dialysis outcomes: what we know now. Curr Opin Nephrol Hypertens 2006; 15: 566–570. Pifer TB, McCullough KP, Port FK et al. Mortality risk in hemodialysis patients and changes in nutritional indicators: DOPPS. Kidney Int 2002; 62: 2238–2245.
SM Brunelli et al.: TZD and mortality in dialysis
30.
31. 32. 33.
34.
35. 36.
37.
38.
39.
40.
41.
42.
43. 44.
45.
Gutierrez A, Alvestrand A, Wahren J et al. Effect of in vivo contact between blood and dialysis membranes on protein catabolism in humans. Kidney Int 1990; 38: 487–494. Caglar K, Peng Y, Pupim LB et al. Inflammatory signals associated with hemodialysis. Kidney Int 2002; 62: 1408–1416. Ikizler TA, Flakoll PJ, Parker RA et al. Amino acid and albumin losses during hemodialysis. Kidney Int 1994; 46: 830–837. Kalantar-Zadeh K, Block G, McAllister CJ et al. Appetite and inflammation, nutrition, anemia, and clinical outcome in hemodialysis patients. Am J Clin Nutr 2004; 80: 299–307. Khan MA, St Peter JV, Xue JL. A prospective, randomized comparison of the metabolic effects of pioglitazone or rosiglitazone in patients with type 2 diabetes who were previously treated with troglitazone. Diabetes Care 2002; 25: 708–711. Singh S, Loke YK, Furberg CD. Long-term risk of cardiovascular events with rosiglitazone: a meta-analysis. JAMA 2007; 298: 1189–1195. Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial infarction and death from cardiovascular causes. N Engl J Med 2007; 356: 2457–2471. Dahabreh IJ. Meta-analysis of rare events: an update and sensitivity analysis of cardiovascular events in randomized trials of rosiglitazone. Clin Trials 2008; 5: 116–120. Kahn SE, Haffner SM, Heise MA et al. Glycemic durability of rosiglitazone, metformin, or glyburide monotherapy. N Engl J Med 2006; 355: 2427–2443. Nakamura T, Ushiyama C, Osada S et al. Effect of pioglitazone on dyslipidemia in hemodialysis patients with type 2 diabetes. Ren Fail 2001; 23: 863–864. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet 1998; 352: 837–853. Ansari A, Thomas S, Goldsmith D. Assessing glycemic control in patients with diabetes and end-stage renal failure. Am J Kidney Dis 2003; 41: 523–531. Ritz E, Dikow R, Adamzcak M et al. Congestive heart failure due to systolic dysfunction: the Cinderella of cardiovascular management in dialysis patients. Semin Dial 2002; 15: 135–140. Efrati S, Zaidenstein R, Dishy V et al. ACE inhibitors and survival of hemodialysis patients. Am J Kidney Dis 2002; 40: 1023–1029. Takahashi A, Takase H, Toriyama T et al. Candesartan, an angiotensin II type-1 receptor blocker, reduces cardiovascular events in patients on chronic haemodialysis–a randomized study. Nephrol Dial Transplant 2006; 21: 2507–2512. Lunn M, McNeil D. Applying Cox regression to competing risks. Biometrics 1995; 51: 524–532.
Kidney International (2009) 75, 961–968
http://www.kidney-international.org & 2009 International Society of Nephrology
Thiazolidinedione use is associated with better survival in hemodialysis patients with non-insulin dependent diabetes Steven M. Brunelli1,2, Ravi Thadhani3, T. Alp Ikizler4 and Harold I. Feldman1,2 1
Renal, Electrolyte and Hypertension Division, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania, USA; 2Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania, USA; 3Renal Unit, Department of Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts, USA and 4Division of Nephrology, Vanderbilt University Medical Center, Nashville, Tennessee, USA
Cardiovascular mortality is especially high among dialysis patients with diabetes, as is morbidity due to protein energy wasting. Given that both of these factors may be decreased by thiazolidinedione treatment, we studied the effect of thiazolidinedione use on survival among chronic dialysis patients in a national cohort of 5290 incident dialysis patients with diabetes. Thiazolidinedione use was assessed according to prescription data, and the analyses were stratified based on insulin use due to observed interaction. In the primary analysis, thiazolidinedione treatment was associated with significantly lower all-cause mortality among insulin-free but not insulin-requiring subjects, with adjusted hazards ratios of 0.53 (0.31–0.89) and 0.82 (0.46–1.47) respectively. Sensitivity analyses found the findings to be robust with respect to confounding by indication, severity of the diabetes, potential reverse causality, and time varying exposure patterns. The mechanism of this decline in all-cause mortality will need to be examined after these studies are confirmed. Kidney International (2009) 75, 961–968; doi:10.1038/ki.2009.4; published online 4 February 2009 KEYWORDS: diabetes; dialysis; epidemiology; mortality; survival thiazolidinediones
There are over 340,000 patients on chronic hemodialysis (HD) in the United States; of these over 40% have diabetes (DM).1 The mortality rate among HD patients is exceedingly high (228.9 deaths/1000 patient-years), and higher yet among those with co-existing DM (251.4 deaths/1000 patient-years).1 Cardiovascular (CV) disease is the leading cause of death among HD patients.2 Protein energy wasting is a potent risk factor for death among HD patients,3 and may interact with CV disease to portend an even graver prognosis.4 HD patients with DM are known to be at a higher risk for CV disease and wasting than their non-diabetic counterparts.1,2,5,6 Thiazolidinediones (TZDs) are a class of oral diabetic medications that function by binding the peroxisome proliferator-activated receptor g, thereby increasing insulin sensitivity in peripheral tissues.7 Human data suggest that in addition to lowering serum glucose, TZDs favorably alter some CV risk factors such as increasing HDL8–17 and circulating adiponectin,7 decreasing triglycerides,12,14–17 visceral adiposity,18 circulating inflammatory mediators,19,20 and albuminuria,8 improving flow-mediated vasodilatation,21 and blunting carotid intima-media thickening.22,23 In addition, experiments in animal models suggest that TZDs may blunt muscle catabolism.24 Among human patients with DM and chronic kidney disease not on HD, treatment with TZDs is associated with a trend toward lower all-cause mortality.25 Given that susceptibility to CV disease and wasting is higher among patients on HD than among patients not yet on HD, it is plausible that TZD treatment might be particularly effective among HD patients. Thus, we conducted this retrospective study among Accelerated Mortality on Renal Replacement (ArMORR)26 cohort participants with diabetes to test that hypothesis that treatment with TZDs is associated with a decreased incidence in all-cause mortality.
Correspondence: Steven M. Brunelli, Renal Division, Brigham and Women’s Hospital, Harvard Medical School, 75 Francis Street, MRB-4, Boston, MA 02115, USA. E-mail: [email protected]
RESULTS Description of cohort
Received 23 September 2008; revised 4 November 2008; accepted 9 December 2008; published online 4 February 2009
Of 10,044 patients in the ArMORR cohort, 5290 were diabetic and survived at least until HD day 30, and were
Kidney International (2009) 75, 961–968
961
original article
SM Brunelli et al.: TZD and mortality in dialysis
included in the primary cohort. Demographic, comorbid disease and laboratory characteristics of the cohort according to baseline TZD and insulin exposure are given in Table 1. Among non-insulin requiring subjects, those receiving TZDs had significantly higher body mass indices, serum albumin levels, and facility standardized mortality ratios than those not receiving TZDs. Among insulin-requiring subjects, those exposed to TZDs were significantly older than those not. Within each insulin exposure stratum, TZD-exposed and unexposed subjects were otherwise similar. Pre-dialysis random glucose levels were similar among TZD-exposed and -unexposed patients (P ¼ 0.4); glycosylated hemoglobin levels were not available. Hemoglobin levels and equilibrated KT/V were similar among TZD exposure groups at baseline, and at all subsequent time points (data not shown). Serum
albumin was significantly higher among TZD-exposed (versus unexposed) subjects at days 90 (Po0.001) and 180 (P ¼ 0.05), but not at day 270 (P ¼ 0.2). Over the first month of HD, 9.6% patients were treated with a TZD, 39.8% with insulin, and 16.9% with a sulfonylurea (Table 2). The proportion of patients treated with zero, one, and two or more classes of anti-diabetic agents were 40.6, 50.2, and 9.2% respectively. b adrenergic antagonists were prescribed in 48.8%, and antagonists of the renin–angiotensin–aldosterone axis in 39.1%. Baseline survival analyses
In the primary analysis, exposures were assessed over HD days zero through 30, and survival measured following day 30 (corresponding to the start of at-risk time). A total of 719
Table 1 | Baseline characterization of subjects stratified by insulin and TZD exposure TZD INS (n=2832)
TZD+ INS (n=353)
TZD INS+ (n=1953)
TZD+ INS+ (n=152)
Age (years); mean (s.d.) Men; n (%) Non-white; n (%) Hypertension; n (%) Congestive heart failure; n (%) Arterial disease; n (%) Cirrhosis; n (%)
64.0 1499 1141 1203 409 480 83
(13.3) (52.9%) (40.3%) (42.5%) (14.4%) (17.0%) (2.9%)
64.4 187 147 156 47 51 11
(11.8) (53.0%) (41.6%) (44.2%) (13.3%) (14.5%) (3.1%)
61.1 1031 722 827 308 348 62
(13.2)* (52.8%) (37.0%) (42.4%) (15.8%) (17.8%) (3.2%)
63.9 72 54 75 21 26 4
(11.1)* (47.4%) (35.5%) (49.3%) (13.8%) (17.1%) (2.6%)
Body mass index (kg/m2); n (%) o20 20–25 25–30 30–35 435
233 836 826 464 467
** (8.2%) (29.6%) (29.2%) (16.4%) (16.5%)
11 91 94 74 83
** (3.1%) (25.8%) (26.6%) (21.0%) (23.5%)
142 489 549 387 380
(7.2%) (25.1%) (28.2%) (19.9%) (19.5%)
10 25 43 32 41
(6.6%) (16.6%) (28.5%) (21.2%) (27.2%)
Facility standardized mortality ratio; n (%) o0.75 0.75–1 1–1.25 41.25 Serum albumin (g/100 ml); mean (s.d.) Serum creatinine (mg/100 ml); mean (s.d.) Serum phosphate (mg/100 ml); mean (s.d.)
632 765 643 768 3.4 5.9 4.7
*** (22.5%) (27.2%) (22.9%) (27.4%) (0.5)** (2.4) (1.5)
72 112 92 75 3.6 5.8 4.5
*** (20.5%) (31.9%) (26.2%) (21.4%) (0.5)** (2.4) (1.5)
470 587 449 426 3.4 5.7 4.6
(24.3%) (30.4%) (23.2%) (22.1%) (0.5) (2.1) (1.5)
39 (26.4%) 54 (36.5%) 25 (16.9%) 30 (20.3%) 3.4 (0.5) 5.4(2.3) 4.6 (1.6)
INS, insulin; TZD, thiazolidinedione. *P=0.01; **Po0.001; ***P=0.03.
Table 2 | Characterization of therapies and baseline and selected follow-up time pointsa Agent TZD a glucosidase inhibitor Biguanide Insulin Meglitinide Sulfonylurea
Day 0 (n=5290) 505 14 19 2105 140 896
Day 90 (n=4598)
Day 180 (n=4180)
Day 270 (n=3804)
(9.6%) (0.3%) (0.4%) (39.8%) (2.7%) (16.9%)
527 11 12 2155 133 877
(11.5%) (0.2%) (0.3%) (47.9%) (2.9%) (19.1%)
557 12 16 2141 131 851
(13.3%) (0.3%) (0.4%) (51.2%) (3.1%) (20.1%)
551 11 15 2009 125 813
(14.5%) (0.3%) (0.4%) (52.8%) (3.3%) (21.4%)
Number of classes of anti-diabetic agents 0 2150 (40.6%) 1 2653 (50.2%) 42 487 (9.2%) b adrenergic antagonist 2581 (48.8%) RAAS antagonist 2069 (39.1%)
1436 2669 493 2599 2168
(31.2%) (58.1%) (10.7%) (56.5%) (47.2%)
1098 2520 562 2558 2210
(26.3%) (60.3%) (13.4%) (61.2%) (52.9%)
926 2305 573 2404 2151
(24.3%) (60.6%) (15.1%) (63.2%) (56.5%)
RAAS, renin–angiotensin–aldosterone. a Reported as n (%).
962
Kidney International (2009) 75, 961–968
original article
SM Brunelli et al.: TZD and mortality in dialysis
Table 3 | Mortality rates and unadjusted incidence rate ratios for all-cause mortality by diabetic treatment
P interaction = 0.02 IRR 0.51 (0.33 to 0.74)
IRR 1.03 (0.62 to 1.62)
Class of agent TZD Unexposed Exposed Insulin Unexposed Exposed
Mortality rate (95% CI) per 1000 patient-years
Incidence rate ratio (95% CI)
189.8 (176.0–204.7) 124.0 (93.7–164.1)
0.65 (0.48–0.87) P=0.002
194.5 (177.4–213.1) 166.3 (147.4–187.8)
0.86 (0.73–1.00) P=0.04
Mortality rate (per 1000 patient-years)
300 250 200 150 100 50 0 TZD- INS-
Sulfonylurea Unexposed Exposed
183.0 (168.9–198.3) 184.1 (154.4–219.5)
1.01 (0.82–1.22) P40.5
Meglitinide Unexposed Exposed
182.3 (169.2–196.3) 219.4 (143.1–336.5)
1.20 (0.74–1.85) P=0.4
TZD+ INS-
TZD- INS+
TZD+ INS+
Figure 1 | Unadjusted mortality rate according to TZD and insulin treatment status. Incidence rate ratios for TZD exposure stratified by insulin exposure are given in the figure. Considered alternately, the IRR (95% CI) for insulin exposure is 0.80 (0.68–0.94) among TZD-unexposed subjects, and 1.64 (0.88–2.99) among TZDexposed subjects (P interaction ¼ 0.02).
719 Deaths; 3925.3 patient-years
Kidney International (2009) 75, 961–968
Insulin –
1.5 Hazard ratio (95% Cl)
deaths occurred over 3925.3 patient-years of at-risk time; crude mortality rate (95% CI) was 183.2 (170.3–197.1) deaths/1000 patient-years. Median survival time was 335 days. On bivariable analysis, the crude mortality rate was significantly lower among TZD-exposed subjects than among non-TZD exposed subjects: IRR (95% CI) 0.65 (0.48–0.87) (Table 3). Survival was similar regardless of TZD agent used (P ¼ 0.45; Figure S1). The IRRs (95% CIs) for all-cause mortality were 0.86 (0.73–1.00), 1.01 (0.82–1.22) and 1.20 (0.74–1.85) according to exposure to insulin, sulfonylureas, and meglitinides, respectively (Table 3). The data were too scant to examine relationships on the basis of other classes of diabetic medications. Stratified analysis was conducted to examine the effect modification of the TZD – mortality association on the basis of insulin exposure (hypothesized a priori). Among subjects not receiving insulin, TZD exposure was associated with reduced all-cause mortality: IRR (95% CI) 0.51 (0.33–0.74); among subjects receiving insulin, there was no association between TZD exposure and all-cause mortality; IRR (95% CI) 1.03 (0.62–1.62; P for interaction ¼ 0.02; Figure 1). Based on this finding, all subsequent analyses were stratified on insulin exposure status. (Considered alternately, the IRR (95% CI) for insulin exposure was 0.80 (0.68–0.94) among TZD unexposed subjects, and 1.64 (0.88–2.99) among TZDexposed subjects (P interaction ¼ 0.02)). The association with improved survival (among patients not on insulin) was specific to TZDs, and was not seen with sulfonylureas or meglitinides (Figure S2). Cox proportional hazard models were fit to examine the adjusted HR for all-cause mortality according to TZD exposure status. TZD exposure was associated with adjusted HRs (95% CIs) for all-cause mortality of 0.53 (0.31–0.89; P ¼ 0.02) and 0.82 (0.46–1.47; P40.5) among subjects not receiving and receiving insulin, respectively (Figure 2). Propensity score analysis yielded nearly identical results:
Insulin +
1
0.5
0 Unadjusted
Multivariable
Propensity
Figure 2 | Hazard ratios for all-cause mortality according to TZD exposure as estimated by Cox proportional hazards models, stratified by insulin exposure status (insulinunexposed white, insulin-exposed gray). Multivariable models were adjusted for age, sex, race, body mass index (categorized r20, 20–25, 25–30, 30–35, 435), facility standardized mortality ratio (categorized r0.75, 0.75–1.0, 1.0–1.25, 41.25), hypertension, arterial disease, congestive heart failure, cirrhosis, number of classes of diabetic medications (categorized 0, 1, Z2), b adrenergic antagonist use, renin–angiotensin–aldosterone antagonist use, and serum albumin, creatinine, and phosphate. Propensity score models were adjusted for the same covariates: number of classes of diabetic medications was adjusted as a covariate in the survival, and the remaining covariates were fit into the propensity model.
HRs (95% CIs) 0.48 (0.29–0.80; P ¼ 0.005) and 0.79 (0.44–1.41; P ¼ 0.4) for patients not receiving and receiving insulin, respectively (Figure 2). The association with improved survival (among patients not on insulin) was specific to TZDs, and was not seen with sulfonylureas or meglitinides (Table S1). Baseline sensitivity analyses
Owing to the possibility that intensity of diabetic therapy may have influenced findings, sensitivity analyses were conducted in which consideration was limited to patients on only one class of oral diabetic medication (Table 4). 963
original article
SM Brunelli et al.: TZD and mortality in dialysis
Table 4 | Results of sensitivity analyses in which single oral agent TZD exposure was compared to other single oral agent therapy, stratified by insulin exposure Off insulin TZD (n=666)
Age-adjusted HR (95% CI) all-cause mortality; TZD versus other
88/495.8 16/182.7 177.5 (144.0–218.7) 87.6 (53.7–143.0) 0.49 (0.27–0.85) P=0.006 0.56 (0.33–0.96) P=0.03
Among patients not receiving insulin, TZD exposure (versus treatment with a single oral agent of another class) was associated with an IRR (95% CI) for all-cause mortality of 0.49 (0.27–0.85); among those receiving insulin the IRR (95% CI) was 0.80 (0.39–1.56). Diminution in the number of outcomes prevented complete multivariable adjustment. Age adjusted HRs (95% CIs) for all-cause mortality according to TZD exposure were 0.56 (0.33–0.96) and 0.82 (0.43–1.55) among patients not receiving and receiving insulin, respectively. Results were similar when the comparator group was further restricted to patients receiving single oral agent therapy with a sulfonylurea (data not shown). To explore whether TZDs influenced mortality through CV mechanisms, non-CV mechanisms or both, competing risks models were fit. Among patients not on insulin, TZDs significantly reduced non-CV mortality, but not CV mortality: HRs (95% CIs) 0.35 (0.15–0.77) and 0.67 (0.36–1.24), respectively (P for difference between CV and nonCV ¼ 0.002). Among subjects not on insulin, TZD exposure did not affect either non-CV or CV mortality: HRs (95% CIs) 0.73 (0.32–1.67) and 0.88 (0.45–1.74), respectively (P for difference between CV and non-CV mortality ¼ 0.7). There was no significant difference in the risk of CV mortality according to TZD agent used (P rosiglitazone versus pioglitazone ¼ 0.6). Because of the concern that we might be observing a reverse-causal relationship (that is, that pre-terminal patients were less likely to be exposed to TZDs), sensitivity analyses were conducted in which exposure and outcome assessment was lagged by 1, 2, and 3 months (Figure 3). Among patients not receiving insulin, the multivariable adjusted HRs (95% CIs) were 0.52 (0.30–0.90), 0.56 (0.31–1.00) and 0.63 (0.34–1.18) in the 1, 2, and 3 month-lagged models, respectively. Estimates were nearly identical when propensity score adjustment was used: HRs (95% CIs) 0.47 (0.28–0.80), 0.50 (0.28–0.88), 0.56 (0.31–1.04), respectively. Among patients receiving insulin, the corresponding adjusted HRs (95% CIs) were 0.98 (0.54–1.81), 1.14 (0.60–2.18), and 1.20 (0.61–2.36). Again, estimates were nearly identical when propensity score adjustment was used: HRs (95% CIs) 0.95 (0.52–1.73), 1.12 (0.59–2.14), 1.19 (0.61–2.34), respectively. Reassuringly, the point estimate for TZD exposure was similar between baseline and lagged models, though statistical significance was lost in the 3-month lagged model, possibly due to diminution of the number of events. 964
TZD+ (n=230)
TZD (n=189)
TZD+ (n=115)
28/139.1 14/87.5 201.3 (139.0–291.6) 160.1 (94.8–270.3) 0.80 (0.39–1.56) P=0.5 0.82 (0.43–1.55) P40.5
Insulin – Insulin +
2 Adjusted HR (95% Cl)
Deaths/patient-years at risk Mortality rate (95% CI); per 1000 patient-years IRR (95%CI) all-cause mortality; TZD versus other
On insulin
1.5 1 0.5 0 1-month lag
2-month lag
3-month lag
Figure 3 | Multivariable adjusted hazard ratios for all-cause mortality according to TZD exposure in lagged analyses. In all models, exposure was defined over days zero through 30, and outcome assessment was lagged to varying degrees (x axis). Separate models were fit according to insulin exposure status (insulin-unexposed white, insulin-exposed gray). Models were adjusted for age, sex, race, body mass index (categorized r20, 20–25, 25–30, 30–35, 435), facility standardized mortality ratio (categorized r0.75, 0.75–1.0, 1.0–1.25, 41.25), hypertension, arterial disease, congestive heart failure, cirrhosis, number of classes of diabetic medications (categorized 0, 1, Z2), b adrenergic antagonist use, renin–angiotensin–aldosterone antagonist use, and serum albumin, creatinine, and phosphate using Cox proportional hazards models.
Time-updated survival analyses
Finally, because exposure (and covariate) status may vary over time, we fit time-updated models in which exposures and covariates were updated at monthly intervals. Using this approach, the adjusted odds ratio (95% CI) for all-cause mortality according to TZD exposure was 0.64 (0.43–0.96) among patients not on insulin, and 0.90 (0.59–1.38) among patients on insulin (Table 5). Analogous models in which exposure and outcome assessment lagged by 1 month yielded nearly identical estimates. DISCUSSION
Survival is very poor among patients on HD in general, particularly among those with co-existing DM.1 CV disease and protein energy wasting are potently associated with allcause HD-related mortality,2,3 which may be of particular importance among diabetic patients.5,6 In addition to their glycemic effects, TZDs have a number of pleiotropic actions on CV risk factors and muscle wasting, suggesting that TZD treatment might improve survival among diabetic HD Kidney International (2009) 75, 961–968
original article
SM Brunelli et al.: TZD and mortality in dialysis
Table 5 | Adjusted odds ratios (95% CI) for all-cause mortality according to TZD exposure status estimated by time-updated logistic regression modelsa
Off insulin On insulin
Un-lagged
1 Month lagged
0.64 (0.43–0.96) P=0.03 0.90 (0.59–1.38) P40.5
0.63 (0.42–0.95) P=0.03 0.88 (0.57–1.36) P40.5
a Adjusted for age, race, sex, body mass index (categorized p20, 20–25, 25–30, 30–35, 435), facility standardized mortality ratio (categorized p0.75, 0.75–1.0, 1.0–1.25, 41.25), hypertension, congestive heart failure, arterial disease, cirrhosis, number of classes of diabetic medication (categorized 0, 1, X2), b adrenergic antagonist use, renin–angiotensin–aldosterone antagonist use, and serum albumin, creatinine and phosphate using time-updated logistic regression models as described in the text.
patients. Using data from a national cohort of incident HD patients with DM, we demonstrated that among those not receiving insulin, TZD therapy was associated with a 47% reduction in all-cause mortality at 1 year. To our knowledge, this is the first report of the positive association between TZD therapy and survival among HD patients. Clearly, if the observed association proves causal, it would have great ramifications on survival among diabetic patients on HD. Plausibility of the observed association is suggested by results from a post hoc analysis of the PROactive trial, which demonstrated that among patients with chronic kidney disease, there was a trend towards improved survival among those randomized to TZD therapy (HR (95% CI) 0.75 (0.55–1.03)).25 However, that trial excluded patients on HD, in whom the benefits of TZD therapy might be expected to be more pronounced. The precise mechanism through which TZDs may reduce mortality among non-insulin-dependent diabetic HD patients is uncertain. CV disease (and attendant mortality) is highly prevalent among HD patients, and exceeds that which would be predicted on the basis of traditional Framingham risk factors.27 Some of this excess risk is thought to relate, in part, to the effects of chronic inflammation.28 Likewise, malnutrition and muscle wasting are common among HD patients,29 and are potent risk factors for HD-related mortality.3 The protein energy wasting syndrome is shown to occur as a result of membrane-induced complement activation,30 systemic release of inflammatory mediators,31 HD-associated nutrient loss,32,33 and poor appetite.33 Human studies demonstrate that diabetic HD patients have greater net loss of muscle protein stores,5 and greater loss of lean body mass over the first year on HD than their non-diabetic counterparts.6 Recent data suggest that the ill-effects of CV disease and protein energy wasting may be synergistic.4 TZDs bind PPAR g receptors, and the resultant receptor – ligand complexes influence transcription of a number of downstream targets.7 The net effect is improvement in glycemic control, favorable modification of some traditional (for example, HDL, triglycerides)8–17 and non-traditional (for example, fat distribution, circulating inflammatory mediators, albuminuria)8,18–20 CV risk factors, as well as decrease in insulin resistance-related muscle catabolism and wasting.24 Kidney International (2009) 75, 961–968
Our data suggest that TZDs mitigate mortality risk in non-insulin requiring HD patients through primarily nonCV mechanisms. Sensitivity analyses suggest a potent association with reduced non-CV mortality, and a more modest (and non-statistically significant) association with reduced CV mortality. Recent reports in the literature have suggested an association between TZDs in general, and rosiglitazone in particular (possibly on the basis of its adverse impacts on LDL,8–12 which are not seen with pioglitazone13–17,34) with increased rates of acute myocardial infarction,35–37 and decompensated congestive heart failure.35,37 One meta-analysis suggested that rosiglitzaone was associated with increased CV mortality;36 however, when data from this study were re-analyzed with inclusion of three large trials designed specifically to look at adverse CV outcomes, this association was no longer appreciated,37 consistent with other published data.35 It is reassuring that although our study did not detect a benefit of TZD use with respect to CV mortality, there was no suggestion of harm (that is, HR o1). Moreover, we did not appreciate a differential association between individual TZDs and allcause or CV mortality. It must be emphasized that the analyses of cause-specific mortality and differential effects of individual TZDs were exploratory in nature, and should be considered as hypothesis generating until further data become available. Moreover, we lacked data on non-fatal CV events, so no conclusions can be drawn regarding the association between TZD use and non-fatal CV events in dialysis patients. Two observations suggest that improvement in glycemic control is not the operative mechanism for the effects of TZD on mortality. One, the beneficial effects on survival were specific to TZDs, and were not seen with other classes of oral agents, which are expected to provide similar degrees of glycemic control.38 Two, the effects of TZDs on glycemic control are modest (lowering hemoglobin A1c by approximately 0.6% in diabetics on HD39), and would therefore not be expected to reduce 1-year mortality to the degree seen.40 Regardless of the mechanism, the observed association between TZD exposure and improved survival is important. These data suggest that the beneficial effects of TZDs are limited to patients not receiving insulin. The likely explanation is that exogenous insulin overwhelms insulin resistance, thereby obviating the benefits of TZDs. However, some proportion of insulin-treated patients were likely type I diabetics (data on type of DM were not available), which may have biased inference in this group. The subgroup of patients not receiving insulin should not have included any type I diabetics, thereby excluding the possibility that observations in this group (including the observed reduction in all-cause mortality related to TZD use) were biased by the type of diabetes. As with all observational study, there remains the possibility of residual confounding. Specifically, one must consider whether TZD-treated patients received these agents 965
original article
because they: (1) were less sick or (2) were receiving better care. With respect to question one, overall health may have differed across TZD groups either with respect to diabetic severity or non-diabetic illness. Assessment of diabetic severity in HD patients is difficult. In non-HD patients one commonly used marker is glycosylated hemoglobin level. However, among HD patients, the utility of glycosylated hemoglobin is limited due to shortened and variable half life of peripheral erythrocytes;41 moreover, these data were not available. Random, pre-dialysis glucose levels were similar among TZD-exposed and -unexposed patients, however, the utility of these values as markers of glycemic control is questionable. Another marker of diabetic severity is intensity of anti-diabetic therapy. Therefore, analyses were stratified on the basis of insulin therapy, and further adjusted for total number of diabetic medications. Additionally, sensitivity analyses were conducted in which analyses were restricted to subjects on single oral agent therapy, presuming that diabetic severity would be similar among TZD-exposed and unexposed members of this subgroup, all of whom received mono-therapy for their diabetes. Consistency of results of this analysis with those from the primary analysis is reassuring in this regard. With respect to non-diabetic illness, TZD-exposed and unexposed patients were similar on the basis of nearly all demographic and comorbid disease characteristics. The exceptions were that among non-insulin-receiving subjects, those TZD exposed had higher facility standardized mortality ratios (which would portend a worse prognosis), serum albumin levels and body mass indices (which would portend a better prognosis), and among insulin-requiring subjects those exposed to TZDs were older. Analyses were therefore adjusted on the basis of these variables, as well as many additional demographic, comorbid disease and laboratory variables. In addition, propensity score models were fit to reduce the influence of confounding by indication, and lagged sensitivity analyses were performed to account for the possibility that there was differential exposure to TZDs on the basis of pre-terminal illness. Consistency of results of these analyses with those from the primary analysis is again reassuring. Quality of health care was addressed in several ways. We adjusted the analyses on the basis of facility-specific mortality rates, as well as exposure to use of b adrenergic blockers and antagonists of the renin–angiotensin–aldosterone axis, which are associated with improved survival,42–44 and could have confounded results. In addition, makers of quality of dialytic care (hemoglobin level, equilibrated KT/V) were similar among TZD-exposed and -unexposed patients. As in all observational study, there was the possibility of exposure misclassification. There was no opportunity to assess patient adherence to medication, thus medication exposures were based on prescription data. This approach is consistent with intention-to-treat principles, and is conservative in that exposure misclassification would be expected to bias results toward the null. In addition, because patient 966
SM Brunelli et al.: TZD and mortality in dialysis
exposures are not static over time, we fit time-updated models, which reassuringly, yielded similar estimates to the primary analysis. Finally, this cohort consisted of diabetic patients incident to HD. Generalization to prevalent HD patients should be undertaken cautiously; generalization to non-diabetic HD patients is not advisable. In addition, assessment of survival in this study was limited to 1 year of follow up; further study is needed to clarify longer-term effects. We conclude that among incident hemodialysis patients with non-insulin requiring diabetes, TZD use was associated with significantly lower rate of all-cause mortality at 1 year of follow up. Further studies are necessary to confirm and generalize findings and to explore the causal nature of this relationship. METHODS Study population This protocol was approved by the University of Pennsylvania Institutional Review Board. We conducted a retrospective analysis of subjects in the ArMORR cohort. Details of the cohort have been previously published.26 Briefly, the ArMORR cohort consists of all patients incident to HD at any of over 1000 Fresenius Medical Care units in North America (FMC-NA) between June 2004 and August 2005 (n ¼ 10 044). We restricted inclusion to adult patients with DM, as defined by (1) attribution of kidney failure to DM (Centers for Medicare and Medicaid Services Form 2728), (2) an International Classification of Disease-9 code of DM upon HD initiation, or (3) active prescription for anti-diabetic medication during the first 30 HD days. To enable baseline characterization of TZD and covariate exposure, we further limited inclusion to subjects surviving to HD day 31 (the start of at-risk time). Study data All study data were collected prospectively. Demographic and comorbid disease characteristics were collected by study investigators at the time of HD initiation. In addition, data on each unit’s standardized mortality ratio was collected to enable adjustment on the basis of the ‘excellent-center effect.’ The standardized mortality ratio is the facility-specific mortality rate relative to all other FMCNA units, and describes facility-level mortality beyond what would be expected on the basis of age, race, gender, nutritional status, and HD adequacy. All laboratory tests were performed in a centralized laboratory (Spectra Laboratories, Rockleigh, NJ, USA). Medication data were collected at baseline, and subsequently updated at each dialysis session. No opportunity existed to assess medication adherence, thus patients were classified as exposed if they had an active prescription spanning that treatment. TZD exposure was considered to be an active prescription for either rosiglitazone (Avandia, Glaxo Smith Kline) or pioglitazone (Actos, Takeda Pharmaceutical Company Ltd.). For the purpose of analysis, medication exposure was considered monthly: a patient was considered to be exposed if he/she had an active prescription for that medication any time during the month. The primary outcome of interest was time to death from any cause. Subjects were followed prospectively until they died or were censored at the time of transfer of care away from FMC-NA, recovery of kidney function, renal transplantation, or change in dialytic modality. As of the time of these analyses, complete Kidney International (2009) 75, 961–968
original article
SM Brunelli et al.: TZD and mortality in dialysis
outcome data were available through 1 year, thus subjects were administratively censored at this time point. In the primary analyses, exposure was defined over the first 30 days following the onset of chronic HD. Exposure to TZDs was then related to survival beginning on 31st day after the onset of HD (corresponding to the start of at-risk time) and continuing until death or censoring. Statistical analysis Continuous data were analyzed graphically and by calculation of summary statistics. Between-group comparisons were conducted using the Student’s t-test. Categorical data were analyzed in terms of counts and proportions, and compared using the w2 Test. In bivariable analyses, survival was compared among groups by calculation of stratum specific (for example, TZD exposed, unexposed) mortality rates, and estimation of incidence rate ratios (IRRs). Pre-specified interaction of the TZD – mortality association on the basis of insulin exposure status was examined using stratified analysis and Mantel–Haenzel methods. Given the potent interaction identified, all subsequent analyses were stratified on the basis of insulin treatment. Adjusted hazard ratios (HRs) were estimated using Cox proportional hazards models, which included terms for pre-specified demographic, comorbid disease, laboratory, and medication covariates of interest. Separate models were fit for insulin-exposed and -unexposed patients. The proportional hazards assumption was tested graphically and by inclusion of two-way cross product terms with time. To account for potential confounding by indication, propensity score adjustment was used. Propensity models were specified as logistic regression models with TZD exposure as the response variable and covariates (aside from number of classes of diabetic medication) as predictor variables. Association between TZD and outcome was assessed using proportional hazards models with covariate terms for propensity score (as a continuous variable; linearity assumption tested graphically) and number of classes of diabetic medication. The effect of TZD use on cause-specific mortality (that is, CV, non-CV) was examined using competing risks models. These models were specified according to the method of Lunn.45 Cause of death was adjudicated by International Classification of Disease-9 code (list given in Supplementary Table S2, Supplementary Materials). A series of lagged analyses was performed in which TZD (and covariate) exposure over HD days zero through 30 was used to predict survival beginning on HD days 61, 91, and 121. Methods were otherwise analogous to those used in the primary analyses. Finally, we conducted analyses using time-updated exposure and covariate data. In these models, each patient’s experience was decomposed into individual patient-months. Time-updated logistic regression models were fit such that survival in month t was predicted on the basis of exposures in month t1. Analyses were stratified on insulin status by inclusion of a two-way cross product term with TZD exposure. Robust variance estimates were used to correct for non-independence of observations within patient.
(Satellite Dialysis Research Fellowship) to SMB. SMB had full access to all of the data and takes responsibility for the integrity and accuracy of the data analysis. SUPPLEMENTARY MATERIAL Table S1. Adjusted hazard ratios (95% CIs) for all-cause mortality according to baseline sulfonylurea use and meglitinide use, stratified by insulin exposure*. Table S2. International classification of disease-9 codes used to define cardiovascular mortality. Figure S1. Kaplan–Meier survivor function according to TZD agent used. Note that the y axis has been rescaled. Figure S2a. Unadjusted mortality rate according to sulfonylurea use, stratified on insulin treatment status. SU, sulfonylurea; INS, insulin Figure S2b. Unadjusted mortality rate according to meglitinide use, stratified on insulin treatment status. MG, meglitinide; INS, insulin Supplementary material is linked to the online version of the paper at http://www.nature.com/ki REFERENCES 1.
2. 3.
4.
5.
6.
7. 8.
9.
10.
11.
12.
13.
DISCLOSURE
14.
R. Thadhani receives grant support from Abbott and has received honoraria from Abbott and Genzyme. Neither company produces thiazolidinediones.
15.
ACKNOWLEDGMENTS
We thank Raymond Boston and Russell Localio for their advice regarding competing risks models. This study was supported in part by awards from the NIDDK (1K23DK079056), the American Heart Association (0775021N), and the National Kidney Foundation Kidney International (2009) 75, 961–968
16.
National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, U.S. Renal Data System, USRDS 2007 Annual Data Report: Atlas of Chronic Kidney Disease and End-Stage Renal Disease in the United States. http://www.usrds.org/ Accessed May 17, 2008, Bethesda, MD. Foley RN, Parfrey PS, Sarnak MJ. Clinical epidemiology of cardiovascular disease in chronic renal disease. Am J Kidney Dis 1998; 32(Suppl 3): S112–S119. Kalantar-Zadeh K, Kilpatrick RD, Kuwae N et al. Revisiting mortality predictability of serum albumin in the dialysis population: time dependency, longitudinal changes and population-attributable fraction. Nephrol Dial Transplant 2005; 20: 1880–1888. de Mutsert R, Grootendorst DC, Axelsson J et al. Excess mortality due to interaction between protein-energy wasting, inflammation and cardiovascular disease in chronic dialysis patients. Nephrol Dial Transplant 2008; 23: 2957–2964. Pupim LB, Flakoll PJ, Majchrzak KM et al. Increased muscle protein breakdown in chronic hemodialysis patients with type 2 diabetes mellitus. Kidney Int 2005; 68: 1857–1865. Pupim LB, Heimburger O, Qureshi AR et al. Accelerated lean body mass loss in incident chronic dialysis patients with diabetes mellitus. Kidney Int 2005; 68: 2368–2374. Yki-Jarvinen H. Thiazolidinediones. N Engl J Med 2004; 351: 1106–1118. Lebovitz HE, Dole JF, Patwardhan R et al. Rosiglitazone monotherapy is effective in patients with type 2 diabetes. J Clin Endocrinol Metab 2001; 86: 280–288. Fonseca V, Rosenstock J, Patwardhan R et al. Effect of metformin and rosiglitazone combination therapy in patients with type 2 diabetes mellitus: a randomized controlled trial. JAMA 2000; 283: 1695–1702. Gomez-Perez FJ, Fanghanel-Salmon G, Antonio Barbosa J et al. Efficacy and safety of rosiglitazone plus metformin in Mexicans with type 2 diabetes. Diabetes Metab Res Rev 2002; 18: 127–134. Vongthavaravat V, Wajchenberg BL, Waitman JN et al. An international study of the effects of rosiglitazone plus sulphonylurea in patients with type 2 diabetes. Curr Med Res Opin 2002; 18: 456–461. Raskin P, Rendell M, Riddle MC et al. A randomized trial of rosiglitazone therapy in patients with inadequately controlled insulin-treated type 2 diabetes. Diabetes Care 2001; 24: 1226–1232. Aronoff S, Rosenblatt S, Braithwaite S et al. Pioglitazone hydrochloride monotherapy improves glycemic control in the treatment of patients with type 2 diabetes: a 6-month randomized placebo-controlled doseresponse study. The Pioglitazone 001 Study Group. Diabetes Care 2000; 23: 1605–1611. Rosenblatt S, Miskin B, Glazer NB et al. The impact of pioglitazone on glycemic control and atherogenic dyslipidemia in patients with type 2 diabetes mellitus. Coron Artery Dis 2001; 12: 413–423. Einhorn D, Rendell M, Rosenzweig J et al. Pioglitazone hydrochloride in combination with metformin in the treatment of type 2 diabetes mellitus: a randomized, placebo-controlled study. The Pioglitazone 027 Study Group. Clin Ther 2000; 22: 1395–1409. Kipnes MS, Krosnick A, Rendell MS et al. Pioglitazone hydrochloride in combination with sulfonylurea therapy improves glycemic control in patients with type 2 diabetes mellitus: a randomized, placebo-controlled study. Am J Med 2001; 111: 10–17.
967
original article
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
968
Rosenstock J, Einhorn D, Hershon K et al. Efficacy and safety of pioglitazone in type 2 diabetes: a randomised, placebo-controlled study in patients receiving stable insulin therapy. Int J Clin Pract 2002; 56: 251–257. Miyazaki Y, Mahankali A, Matsuda M et al. Effect of pioglitazone on abdominal fat distribution and insulin sensitivity in type 2 diabetic patients. J Clin Endocrinol Metab 2002; 87: 2784–2791. Haffner SM, Greenberg AS, Weston WM et al. Effect of rosiglitazone treatment on nontraditional markers of cardiovascular disease in patients with type 2 diabetes mellitus. Circulation 2002; 106: 679–684. Marx N, Froehlich J, Siam L et al. Antidiabetic PPAR gamma-activator rosiglitazone reduces MMP-9 serum levels in type 2 diabetic patients with coronary artery disease. Arterioscler Thromb Vasc Biol 2003; 23: 283–288. Caballero AE, Saouaf R, Lim SC et al. The effects of troglitazone, an insulinsensitizing agent, on the endothelial function in early and late type 2 diabetes: a placebo-controlled randomized clinical trial. Metabolism 2003; 52: 173–180. Sidhu JS, Kaposzta Z, Markus HS et al. Effect of rosiglitazone on common carotid intima-media thickness progression in coronary artery disease patients without diabetes mellitus. Arterioscler Thromb Vasc Biol 2004; 24: 930–934. Mazzone T, Meyer PM, Feinstein SB et al. Effect of pioglitazone compared with glimepiride on carotid intima-media thickness in type 2 diabetes: a randomized trial. Jama 2006; 296: 2572–2581. Wang X, Hu Z, Hu J et al. Insulin resistance accelerates muscle protein degradation: Activation of the ubiquitin-proteasome pathway by defects in muscle cell signaling. Endocrinology 2006; 147: 4160–4168. Schneider CA, Ferrannini E, Defronzo R et al. Effect of pioglitazone on cardiovascular outcome in diabetes and chronic kidney disease. J Am Soc Nephrol 2008; 19: 182–187. Teng M, Wolf M, Ofsthun MN et al. Activated injectable vitamin D and hemodialysis survival: a historical cohort study. J Am Soc Nephrol 2005; 16: 1115–1125. Longenecker JC, Coresh J, Powe NR et al. Traditional cardiovascular disease risk factors in dialysis patients compared with the general population: the CHOICE Study. J Am Soc Nephrol 2002; 13: 1918–1927. Krane V, Drechsler C, Wanner C. Dyslipidemia, inflammation and dialysis outcomes: what we know now. Curr Opin Nephrol Hypertens 2006; 15: 566–570. Pifer TB, McCullough KP, Port FK et al. Mortality risk in hemodialysis patients and changes in nutritional indicators: DOPPS. Kidney Int 2002; 62: 2238–2245.
SM Brunelli et al.: TZD and mortality in dialysis
30.
31. 32. 33.
34.
35. 36.
37.
38.
39.
40.
41.
42.
43. 44.
45.
Gutierrez A, Alvestrand A, Wahren J et al. Effect of in vivo contact between blood and dialysis membranes on protein catabolism in humans. Kidney Int 1990; 38: 487–494. Caglar K, Peng Y, Pupim LB et al. Inflammatory signals associated with hemodialysis. Kidney Int 2002; 62: 1408–1416. Ikizler TA, Flakoll PJ, Parker RA et al. Amino acid and albumin losses during hemodialysis. Kidney Int 1994; 46: 830–837. Kalantar-Zadeh K, Block G, McAllister CJ et al. Appetite and inflammation, nutrition, anemia, and clinical outcome in hemodialysis patients. Am J Clin Nutr 2004; 80: 299–307. Khan MA, St Peter JV, Xue JL. A prospective, randomized comparison of the metabolic effects of pioglitazone or rosiglitazone in patients with type 2 diabetes who were previously treated with troglitazone. Diabetes Care 2002; 25: 708–711. Singh S, Loke YK, Furberg CD. Long-term risk of cardiovascular events with rosiglitazone: a meta-analysis. JAMA 2007; 298: 1189–1195. Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial infarction and death from cardiovascular causes. N Engl J Med 2007; 356: 2457–2471. Dahabreh IJ. Meta-analysis of rare events: an update and sensitivity analysis of cardiovascular events in randomized trials of rosiglitazone. Clin Trials 2008; 5: 116–120. Kahn SE, Haffner SM, Heise MA et al. Glycemic durability of rosiglitazone, metformin, or glyburide monotherapy. N Engl J Med 2006; 355: 2427–2443. Nakamura T, Ushiyama C, Osada S et al. Effect of pioglitazone on dyslipidemia in hemodialysis patients with type 2 diabetes. Ren Fail 2001; 23: 863–864. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet 1998; 352: 837–853. Ansari A, Thomas S, Goldsmith D. Assessing glycemic control in patients with diabetes and end-stage renal failure. Am J Kidney Dis 2003; 41: 523–531. Ritz E, Dikow R, Adamzcak M et al. Congestive heart failure due to systolic dysfunction: the Cinderella of cardiovascular management in dialysis patients. Semin Dial 2002; 15: 135–140. Efrati S, Zaidenstein R, Dishy V et al. ACE inhibitors and survival of hemodialysis patients. Am J Kidney Dis 2002; 40: 1023–1029. Takahashi A, Takase H, Toriyama T et al. Candesartan, an angiotensin II type-1 receptor blocker, reduces cardiovascular events in patients on chronic haemodialysis–a randomized study. Nephrol Dial Transplant 2006; 21: 2507–2512. Lunn M, McNeil D. Applying Cox regression to competing risks. Biometrics 1995; 51: 524–532.
Kidney International (2009) 75, 961–968