clinical investigation
http://www.kidney-international.org & 2014 International Society of Nephrology
see commentary on page 238
Early mortality in patients starting dialysis appears to go unregistered Robert N. Foley1,2, Shu-Cheng Chen1, Craig A. Solid1, David T. Gilbertson1 and Allan J. Collins1,2 1
United States Renal Data System, Minneapolis Medical Research Foundation, Minneapolis, Minnesota, USA and 2Department of Medicine, University of Minnesota, Minneapolis, Minnesota, USA
Clinical experience suggests a heightened risk associated with the transition to maintenance dialysis but few national studies have systematically examined early mortality trajectories. Here we calculated weekly mortality rates in the first year of treatment for 498,566 adults initiating maintenance dialysis in the United States (2005–2009). Mortality rates were initially unexpectedly low, peaked at 37.0 per 100 person-years in week 6, and declined steadily to 14.8 by week 51. In both early (weeks 7–12) and later (weeks 13–51) time frames, multivariate mortality associations included older age, female, Caucasian, non-Hispanic ethnicity, end-stage renal disease (ESRD) from hypertension and acute tubular necrosis, ischemic heart disease, estimated glomerular filtration rate of 15 ml/min per 1.73 m2 or more, shorter duration of nephrologist care, and hemodialysis, especially with a catheter. For early mortality risk, adjusted hazard ratios of 2 or more were seen with age over 65 (5.80 vs. under 40 years), hemodialysis with a catheter (2.73 vs. fistula), and age 40–64 (2.33). For later mortality risk, adjusted hazard ratios of 2 or more were seen with age over 65 (4.32 vs. under 40 years), hemodialysis with a catheter (2.10 vs. fistula), and age 40–64 (2.00). Thus, low initial mortality rates question the accuracy of data collected and are consistent with deaths occurring in the early weeks after starting dialysis not being registered with the United States Renal Data System. Kidney International (2014) 86, 392–398; doi:10.1038/ki.2014.15; published online 12 February 2014 KEYWORDS: ascertainment bias; dialysis; incident; mortality
Correspondence: Robert N. Foley, United States Renal Data System, Minneapolis Medical Research Foundation, 914 South 8th Street, Suite S4.100, Minneapolis, Minnesota 55404, USA. E-mail: [email protected] Received 15 March 2013; revised 4 December 2013; accepted 19 December 2013; published online 12 February 2014 392
The initial weeks of maintenance dialysis treatment are a time of clinical uncertainty. Although end-stage renal disease (ESRD) is associated with high mortality in and of itself, initiation of dialysis often adds features that could also put patients at risk. For example, initiation of in-center hemodialysis could heighten the risk of blood-borne infection, dialysis-related hypotension, hypokalemia, and loss of intrinsic renal function. Similarly, initiation of dialysis remains the most common time for introducing typical treatments for ESRD that may be dangerous if used inappropriately, with overaggressive use of erythropoiesisstimulating agents being a notable example.1,2 Although it is not known with certainty whether mortality risks associated with noninitiation of dialysis outweigh those associated with the dialysis procedure itself, at least one large study of patients initiating hemodialysis in a large chain of dialysis units in the United States suggested that this may be the case, as weekly mortality rates were maximal at the time of dialysis initiation.3 Dialysis initiation is a period of great flux, and clinical experience suggests that successful navigation through this period has long-term ramifications; however, the evolution of mortality risk early in the course of dialysis treatment remains poorly studied. This lack of information is surprising, as more information could help with clinical decision making, especially for patients making judgments about when, whether, and how to initiate renal replacement therapy (RRT). Accurate knowledge of survival expectations early in the course of dialysis treatment could be especially enlightening. Regarding mortality patterns at that time, large national dialysis patient registries have inherent attractions and potential caveats. For example, registries are typically representative, and large sample size allows for precise risk estimates and examination of findings within subgroups. On the other hand, registries may encompass potential biases, especially in the period surrounding the transition to RRT. In particular, there is concern that an early ascertainment bias may be present, with ESRD patients who die soon after dialysis initiation not being registered in national registries. In this scenario, the desire to understand outcomes ever earlier in RRT would have to be tempered against the ever-increasing likelihood of invalid findings. In addition, the fact that Medicare coverage in the Kidney International (2014) 86, 392–398
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
United States requires 90 days to have elapsed from initiation of hemodialysis may have created the impression that survival rates in the first 90 days cannot be studied when, in actuality, registration in the United States Renal Data System (USRDS) has been required for all new RRT patients for decades. As few large national studies have systematically examined early mortality trajectories in dialysis patients, we set out to address this issue among adult patients in the USRDS data.
RESULTS
Of the study population, 49.7% of subjects were aged 65 years or older, 28.8% were black, 13.7% were Hispanic, and diabetes was the cause of ESRD in 45.4% (Table 1). Peritoneal dialysis was the initial mode of RRT for 6.0% of patients, and in 77.7% hemodialysis was initiated with a catheter. Table 1 also compares patients according to duration of predialysis nephrologist care and mode of dialysis. With adjusted logistic regression, nephrologist care for o6 months was associated
Table 1 | Study population characteristics at initiation of dialysis (n ¼ 498,566), compared by duration of predialysis nephrologist care and mode of dialysis therapy Predialysis nephrologist care, months Characteristics
All
Percent of population
X6
0–5
45.6
54.4
AOR, 0–5 (vs. X6)
Mode of dialysis HD
PD
94.0
6.0
AOR, PD (vs. HD)
Age, years o40 40–64 X65
7.9 42.4 49.7
7.2 43.1 49.7
8.9 43.0 48.1
1 (Referent) 0.88 (0.86–0.91) 0.78 (0.76–0.80)
7.6 41.8 50.6
13.0 51.6 35.4
1 (Referent) 0.70 (0.67–0.73) 0.38 (0.36–0.39)
Sex Men Women
56.1 43.9
55.5 44.5
56.3 43.7
1 (Referent) 0.94 (0.93–0.95)
56.2 43.8
54.6 45.4
1 (Referent) 1.10 (1.07–1.13)
Race White Black Other
65.7 28.8 5.5
65.8 28.4 5.9
64.0 30.5 5.5
1 (Referent) 1.20 (1.19–1.22) 1.20 (1.17–1.24)
65.3 29.3 5.4
72.1 21.2 6.7
1 (Referent) 0.54 (0.53–0.56) 0.96 (0.91–1.01)
Hispanic ethnicity No Yes
86.3 13.7
85.5 14.5
85.1 14.9
1 (Referent) 1.35 (1.32–1.37)
86.2 13.8
88.4 11.6
1 (Referent) 0.67 (0.65–0.70)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
45.4 28.4 6.5 2.0 3.0 14.7
49.8 26.9 7.6 3.0 1.0 11.7
41.8 29.6 5.6 1.2 4.6 17.2
1 (Referent) 1.34 (1.32–1.36) 0.93 (0.90–0.95) 0.69 (0.66–0.72) 4.57 (4.36–4.79) 1.67 (1.64–1.70)
45.6 28.6 6.1 1.8 3.2 14.8
42.7 24.6 13.1 6.1 0.3 13.3
1 (Referent) 1.10 (1.06–1.13) 1.80 (1.73–1.87) 2.38 (2.25–2.52) 0.13 (0.11–0.17) 0.97 (0.93–1.01)
Ischemic heart disease No Yes
78.3 21.7
77.2 22.8
80.8 19.2
1 (Referent) 0.75 (0.74–0.76)
78 22
84.4 15.6
1 (Referent) 0.7 (0.68–0.72)
GFR, ml/min per 1.73 m2 o15 X15
82.6 17.4
82.9 17.1
81.4 18.6
1 (Referent) 1.17 (1.15–1.19)
82.4 17.6
84.9 15.1
1 (Referent) 1.03 (1.00–1.07)
Nephrologist care, months 412 6–12 0–5
23.2 22.4 54.4
50.9 49.1 0
— — —
22.1 22.0 55.9
40.4 29.5 30.1
1 (Referent) 0.77 (0.75–0.79) 0.32 (0.31–0.33)
Dialysis and access HD, fistula HD, graft HD, catheter PD
12.7 3.5 77.7 6
15.9 4.5 71.7 7.9
1 (Referent) 1.64 (1.58–1.7) 4.58 (4.49–4.67) 1.29 (1.25–1.33)
13.6 3.8 82.7 0
0 0 100
5.8 2.3 88.6 3.4
0 0 0 100
— — — —
Abbreviations: AOR, adjusted odds ratio; ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Data are presented as column percent or odds ratio (95% confidence interval). Logistic regression was used to calculate odds ratios, with adjustment for all the variables shown in the first column. Po0.001 for all comparisons, except AOR of GFR 415 ml/min per 1.73 m2 and association with mode of dialysis (P ¼ 0.08). Missing data: dialysis and access, 1.1%; GFR, 1.9%.
Kidney International (2014) 86, 392–398
393
Mortality rate, per 100 patient-years
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
50 40 Care < 6 Months Hemodialysis All Care . 6 Months Peritoneal dialysis
30 20 10 0 0
20 Weeks
40
Figure 1 | Weekly mortality rates during weeks 1–51, inclusive, following initiation of dialysis therapy. Care refers to duration of nephrologist care before dialysis initiation.
with younger age, male sex, black race, ESRD from hypertension and acute tubular necrosis, lack of ischemic heart disease, estimated glomerular filtration rate X15 ml/min per 1.73 m2, and hemodialysis with a catheter. Associations of initial peritoneal dialysis treatment, as opposed to hemodialysis, included younger age, female sex, white race, nonHispanic ethnicity, ESRD from glomerulonephritis and cystic disease, lack of ischemic heart disease, and longer duration of nephrologist care. During the 1-year study (mean follow-up 0.86 years), 2.2% of patients were censored because of change in RRT, whereas 23.4% of patients died from cardiac causes (35.8%), withdrawal of dialysis (9.6%), and other causes (54.6%). Figure 1 shows weekly mortality rate intervals following hemodialysis initiation. Mortality rates in the dialysis population were 5.9 per 100 person-years in week 1, peaked at 37.0 per 100 person-years in week 6, and declined steadily to intermediate levels by week 51 (14.8 per 100 person-years). Figure 1 also shows weekly mortality rates in subgroups defined by duration of nephrologist care and initial mode of dialysis. At all time points, rates were highest with nephrologist care o6 months followed by hemodialysis, nephrologist care X6 months, and peritoneal dialysis. Mortality rates peaked after 4 weeks in all subgroups and gradually declined in all subgroups except in peritoneal dialysis patients, for whom rates remained stable. Table 2 shows hazard ratios for mortality in two time frames, 7–12 weeks (inclusive) and 13–51 weeks (inclusive). In each of these time frames, multivariate analysis showed that mortality was associated with older age, female sex, white race, non-Hispanic ethnicity, ESRD from hypertension and acute tubular necrosis, ischemic heart disease, estimated glomerular filtration rate X15 ml/min per 1.73 m2, shorter duration of nephrologist care, and hemodialysis, especially with a catheter. For early mortality risk, adjusted hazard ratios (AHRs) X2 were seen with age 465 years (AHR 5.80 vs.o40 years), hemodialysis with a catheter (AHR 2.73 vs. hemodialysis with a fistula), and age 40–64 years (AHR 2.33); for later mortality, AHRs X2 were seen with age 465 years 394
(AHR 4.32), hemodialysis with a catheter (AHR 2.10), and age 40–64 years (AHR 2.00). Table 3 shows adjusted hazards for death in subgroups, defined by duration of predialysis nephrologist care. Although risk association patterns were similar to patterns in the overall population, risk gradients were generally steeper in the subgroup with shorter duration of care, with the exception of initial mode of dialysis and access in which risk gradients were similar. Table 4, summarizing mortality risk in subgroups defined by mode of dialysis, shows that hazard ratios were generally similar for hemodialysis and peritoneal dialysis patients; risk estimates differed noticeably, however, for age X65 years (vs. o40 years), especially for later mortality (AHR 4.24 for hemodialysis vs. 9.36 for peritoneal dialysis patients), and for acute tubular necrosis as cause of ESRD, especially for early mortality (AHRs (vs. diabetes)) 1.60 for hemodialysis and 7.33 for peritoneal dialysis). DISCUSSION
We found that mortality rates in US dialysis patients were approximately six times higher in week 6 than in week 1 after initiation, and declined by approximately twofold between week 6 and week 52. Early mortality rates were especially high in older subgroups, and in patients in whom acute tubular necrosis with incomplete recovery was the cause of ESRD, in patients whose duration of predialysis nephrologist care was short, and in hemodialysis patients, especially those with catheters for vascular access. In addition, the unexpectedly low mortality rates in the initial week of dialysis suggest a possible initial ascertainment bias, with high-risk patients not being registered. If this initial bias applies, mortality rates reported in this study for the first few weeks of RRT will be underestimates. Examining outcomes in short intervals requires large sample sizes for precise risk estimates. Few studies have reported mortality risk evolution in weekly increments following initiation of maintenance hemodialysis. In the Dialysis Outcomes and Practice Patterns Study, Bradbury and colleagues examined mortality risk in US patients who were started on hemodialysis between 1996 and 2004, and found that mortality rates were 25% higher in the first 3 months of treatment than in months 4–12, with greater disparities for cardiovascular than for non-cardiovascular mortality.4 Chan et al.3 examined mortality rates in 303,289 patients who initiated dialysis between 1997 and 2009 in 1733 facilities affiliated with a large US dialysis organization. They found that mortality estimates were highest in the first 2 weeks of treatment and declined successively thereafter. Unlike in our study, in which mortality estimates rose by sixfold between baseline and the 6-week peak, mortality risk was close to maximum in the first week. The observation that the relative mortality of patients on peritoneal dialysis vs. those on hemodialysis widens with the passage of time was also noted in the study by Chan et al.3 Although the cause of this disparity in time-dependent Kidney International (2014) 86, 392–398
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
Table 2 | Hazard ratios of baseline characteristics for death between 7–12 and 13–52 weeks (inclusive) after starting dialysis therapy Death, 7–12 weeks, n ¼ 481,377; 4.6% dead Characteristics
Death, 13–51 weeks, n ¼ 459,877; 16.9% dead
Unadjusted
Adjusted
Unadjusted
Adjusted
Age, years o40 40–64 X65
1 (Referent) 2.36 (2.13–2.6) 6.74 (6.12–7.43)
1 (Referent) 2.33 (2.11–2.58) 5.80 (5.24–6.41)
1 (Referent) 2.07 (1.98–2.17) 5.06 (4.84–5.29)
1 (Referent) 2.00 (1.91–2.09) 4.32 (4.12–4.52)
Sex Men Women
1 (Referent) 1.05 (1.02–1.08)
1 (Referent) 1.06 (1.04–1.09)
1 (Referent) 1.05 (1.03–1.06)
1 (Referent) 1.05 (1.04–1.07)
Race White Black Other
1 (Referent) 0.64 (0.62–0.66) 0.50 (0.46–0.54)
1 (Referent) 0.74 (0.72–0.77) 0.58 (0.54–0.63)
1 (Referent) 0.69 (0.68–0.70) 0.55 (0.53–0.57)
1 (Referent) 0.77 (0.76–0.78) 0.61 (0.59–0.63)
Hispanic ethnicity No Yes
1 (Referent) 0.62 (0.59–0.65)
1 (Referent) 0.68 (0.64–0.71)
1 (Referent) 0.65 (0.63–0.66)
1 (Referent) 0.67 (0.65–0.69)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
1 (Referent) 1.36 (1.32–1.40) 0.56 (0.52–0.61) 0.29 (0.24–0.35) 2.52 (2.38–2.67) 1.83 (1.76–1.89)
1 (Referent) 1.17 (1.13–1.21) 0.79 (0.73–0.86) 0.53 (0.44–0.64) 1.61 (1.52–1.71) 1.71 (1.65–1.78)
1 (Referent) 1.16 (1.14–1.18) 0.52 (0.50–0.54) 0.26 (0.24–0.29) 1.75 (1.69–1.81) 1.38 (1.35–1.4)
1 (Referent) 1.04 (1.02–1.05) 0.70 (0.67–0.72) 0.40 (0.37–0.44) 1.21 (1.16–1.25) 1.35 (1.32–1.37)
Ischemic heart disease No Yes
1 (Referent) 1.49 (1.45–1.54)
1 (Referent) 1.16 (1.12–1.19)
1 (Referent) 1.56 (1.54–1.59)
1 (Referent) 1.22 (1.2–1.24)
GFR, ml/min per 1.73 m2 o15 X15
1 (Referent) 1.81 (1.76–1.87)
1 (Referent) 1.53 (1.49–1.58)
1 (Referent) 1.76 (1.73–1.79)
1 (Referent) 1.51 (1.48–1.53)
Nephrology care, months 412 6–12 0–5
1 (Referent) 1.34 (1.28–1.41) 2.05 (1.97–2.13)
1 (Referent) 1.26 (1.2–1.32) 1.66 (1.6–1.73)
1 (Referent) 1.17 (1.14–1.2) 1.50 (1.47–1.53)
1 (Referent) 1.12 (1.1–1.15) 1.33 (1.3–1.35)
Dialysis and access HD, fistula HD, graft HD, catheter PD
1 (Referent) 1.36 (1.22–1.52) 2.83 (2.68–2.99) 0.64 (0.57–0.72)
1 (Referent) 1.50 (1.34–1.68) 2.73 (2.56–2.91) 0.86 (0.76–0.97)a
1 (Referent) 1.41 (1.34–1.48) 2.13 (2.07–2.18) 0.86 (0.82–0.9)
1 (Referent) 1.45 (1.38–1.52) 2.10 (2.04–2.16) 1.07 (1.02–1.12)b
Abbreviations: ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Cox regression was used to calculate hazard ratios (95% confidence intervals). Adjustment variables were age, sex, race, Hispanic ethnicity, cause of ESRD, ischemic heart disease, GFR, predialysis nephrology care, dialysis, and access. Po0.001, unless otherwise stated. a 0.01pPo0.05 b 0.001pPo0.01
mortality cannot be determined from observational studies, it is tempting to speculate that it may relate to the usual practice in which hemodialysis is almost always used when acute declines in kidney function lead to life-threatening complications, or when the burden of comorbid illnesses is large. The interplay between starting maintenance dialysis, Medicare, and the USRDS is complex. Regardless of future intentions to use Medicare as payer, a Centers for Medicare & Medicaid Services (CMS) Medical Evidence Report (form CMS-2728) must be filed within 45 days of the first dialysis Kidney International (2014) 86, 392–398
session.5 Similarly, a Death Notification (form CMS-2746) must be filed when patients die, regardless of payer designation.6 The Medical Evidence Report performs the dual role of tracking the size and composition of the national RRT program and potential enrollment in Medicare for coverage of medical care. New Medicare coverage for incenter patients is available after 3 months of in-center hemodialysis. In this regard, it is plausible that some of the unexpectedly low mortality rates very early in the course of dialysis treatment may reflect the lack of a financial incentive to complete these forms for patients who lack Medicare 395
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
Table 3 | Adjusted hazard ratios of baseline characteristics for death between 7–12 and 13–52 weeks (inclusive), according to duration of predialysis nephrology care Death, 7–12 weeks
Death, 13–52 weeks
Care X6 months, n ¼ 221,698; 3.3% dead
Care 0–5 months, n ¼ 259,679; 5.7% dead
CareX6 months, n ¼ 214,598; 14.3% dead
Care 0–5 months, n ¼ 245,279; 19.1% dead
Age, years o40 40–64 X65
1 (Referent) 2.03 (1.67–2.46) 5.20 (4.31–6.27)
1 (Referent) 2.59 (2.31–2.92) 7.32 (6.53–8.21)
1 (Referent) 2.08 (1.91–2.26) 4.60 (4.23–5.00)
1 (Referent) 2.11 (2.00–2.23) 5.08 (4.82–5.35)
Sex Men Women
1 (Referent) 1.02 (0.97–1.07)a
1 (Referent) 1.10 (1.06–1.13)
1 (Referent) 1.03 (1–1.05)b
1 (Referent) 1.09 (1.07–1.11)
Race White Black Other
1 (Referent) 0.72 (0.68–0.76) 0.57 (0.50–0.66)
1 (Referent) 0.63 (0.60–0.65) 0.50 (0.46–0.55)
1 (Referent) 0.71 (0.69–0.73) 0.62 (0.58–0.66)
1 (Referent) 0.71 (0.70–0.73) 0.55 (0.52–0.57)
Hispanic ethnicity No Yes
1 (Referent) 0.70 (0.64–0.76)
1 (Referent) 0.58 (0.55–0.61)
1 (Referent) 0.69 (0.66–0.71)
1 (Referent) 0.61 (0.59–0.63)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
1 (Referent) 1.18 (1.12–1.25) 0.77 (0.68–0.87) 0.51 (0.40–0.67) 2.02 (1.74–2.34) 1.63 (1.52–1.75)
1 (Referent) 1.28 (1.23–1.33) 0.59 (0.53–0.65) 0.34 (0.26–0.45) 2.02 (1.89–2.16) 1.74 (1.67–1.82)
1 (Referent) 1.06 (1.03–1.09) 0.71 (0.67–0.75) 0.43 (0.38–0.49) 1.48 (1.36–1.61) 1.30 (1.26–1.35)
1 (Referent) 1.09 (1.07–1.12) 0.52 (0.49–0.55) 0.26 (0.22–0.30) 1.49 (1.43–1.55) 1.35 (1.31–1.38)
Ischemic heart disease No Yes
1 (Referent) 1.21 (1.15–1.27)
1 (Referent) 1.52 (1.47–1.58)
1 (Referent) 1.26 (1.23–1.29)
1 (Referent) 1.57 (1.54–1.6)
GFR, ml/min per 1.73 m2 o15 X15
1 (Referent) 1.62 (1.53–1.7)
1 (Referent) 1.71 (1.64–1.77)
1 (Referent) 1.51 (1.47–1.55)
1 (Referent) 1.72 (1.69–1.76)
Nephrology care, months 412 6–12 0–5
1 (Referent) 1.26 (1.20–1.32) —
— — —
1 (Referent) 1.12 (1.10–1.15) —
— — —
Dialysis and access HD, fistula HD, graft HD, catheter PD
1 (Referent) 1.35 (1.16–1.56) 2.49 (2.30–2.69) 0.81 (0.69–0.94)c
1 (Referent) 1.25 (1.06–1.47)c 2.23 (2.05–2.43) 0.57 (0.47–0.68)
1 (Referent) 1.43 (1.35–1.53) 2.10 (2.03–2.17) 1.09 (1.03–1.16)c
1 (Referent) 1.28 (1.19–1.39) 1.75 (1.68–1.83) 0.73 (0.67–0.79)
Characteristics
Abbreviations: ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Cox regression was used to calculate hazard ratios (95% confidence intervals). Adjustment variables were age, sex, race, Hispanic ethnicity, cause of ESRD, ischemic heart disease, GFR, predialysis nephrology care, dialysis, and access. Po0.001, unless otherwise stated. a PX0.05 b 0.01pPo0.05 c 0.001pPo0.01
coverage and who die before 3 months. Our findings are consistent with the hypothesis that registration of patients at high risk of early death may be incomplete. If this is the case, it may be appropriate to wait until approximately 6 weeks have elapsed in studies in which death is a principal outcome. Our study is not without limitations and shares all the imperfections of retrospective, registry-based studies, including lack of prospective validation of key data elements. Care should be exerted when attempting to generalize the findings 396
from this study to other countries. This was a nonexperimental study of associations, and causes and effects cannot be separated with confidence. As this was a study of incident dialysis patients, many of the characteristics described may reflect care in the years preceding dialysis initiation. Although information regarding some markers of nephrology care was collected, complete and comprehensive descriptors were not available. A particular limitation of studies purporting to start follow-up at the beginning of maintenance dialysis is Kidney International (2014) 86, 392–398
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
Table 4 | Adjusted hazard ratios of baseline characteristics for death between 7–12 and 13–52 weeks (inclusive) according to initial dialysis therapy Death, 7–12 weeks
Death, 13–52 weeks
HD, n ¼ 451,928; 4.8% dead
PD, n ¼ 29,449; 1.3% dead
HD, n ¼ 430,792; 17.5% dead
PD, n ¼ 29,085; 8.3% dead
Age, years o40 40–64 X65
1 (Referent) 2.36 (2.13–2.62) 5.83 (5.26–6.46)
1 (Referent) 1.59 (0.92–2.74)a 6.53 (3.88–10.99)
1 (Referent) 1.98 (1.89–2.07) 4.24 (4.04–4.44)
1 (Referent) 2.86 (2.24–3.67) 9.36 (7.35–11.91)
Sex Men Women
1 (Referent) 1.07 (1.04–1.1)
1 (Referent) 0.91 (0.74–1.12)a
1 (Referent) 1.05 (1.03–1.06)
1 (Referent) 0.99 (0.91–1.07)a
Race White Black Other
1 (Referent) 0.74 (0.72–0.77) 0.59 (0.54–0.63)
1 (Referent) 0.59 (0.44–0.79) 0.46 (0.26–0.8)c
1 (Referent) 0.77 (0.76–0.78) 0.61 (0.59–0.64)
1 (Referent) 0.59 (0.53–0.66) 0.50 (0.41–0.62)
Hispanic ethnicity No Yes
1 (Referent) 0.68 (0.64–0.71)
1 (Referent) 0.60 (0.41–0.89)b
1 (Referent) 0.67 (0.65–0.69)
1 (Referent) 0.52 (0.44–0.61)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
1 (Referent) 1.17 (1.13–1.21) 0.80 (0.73–0.86) 0.55 (0.45–0.66) 1.60 (1.51–1.71) 1.71 (1.64–1.77)
1 (Referent) 1.71 (1.34–2.18) 0.54 (0.35–0.85)c 0.26 (0.11–0.64)c 7.33 (3.23–16.61) 1.89 (1.42–2.5)
1 (Referent) 1.04 (1.02–1.06) 0.71 (0.68–0.74) 0.42 (0.38–0.47) 1.21 (1.17–1.26) 1.36 (1.33–1.38)
1 (Referent) 1.03 (0.94–1.14)a 0.32 (0.27–0.39) 0.20 (0.14–0.28) 2.49 (1.52–4.08) 1.02 (0.91–1.15)a
Ischemic heart disease No Yes
1 (Referent) 1.16 (1.12–1.19)
1 (Referent) 1.91 (1.51–2.41)
1 (Referent) 1.21 (1.19–1.23)
1 (Referent) 2.24 (2.05–2.45)
GFR, ml/min per 1.73 m2 o15 X15
1 (Referent) 1.53 (1.49–1.58)
1 (Referent) 1.98 (1.57–2.51)
1 (Referent) 1.51 (1.48–1.53)
1 (Referent) 2.02 (1.84–2.22)
Nephrology care, months 412 6–12 0–5
1 (Referent) 1.25 (1.19–1.31) 1.66 (1.59–1.73)
1 (Referent) 1.54 (1.19–2.00)c 1.78 (1.39–2.29)
1 (Referent) 1.12 (1.09–1.14) 1.33 (1.30–1.35)
1 (Referent) 1.16 (1.05–1.28)c 1.22 (1.11–1.35)
Dialysis and access HD, fistula HD, graft HD, catheter PD
1 (Referent) 1.50 (1.34–1.68) 2.73 (2.56–2.92) —
— — — —
1 (Referent) 1.45 (1.38–1.52) 2.10 (2.04–2.16) —
— — — —
Characteristics
Abbreviations: ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Cox regression was used to calculate hazard ratios (95% confidence intervals). Adjustment variables were age, sex, race, Hispanic ethnicity, cause of ESRD, ischemic heart disease, GFR, predialysis nephrology care, dialysis, and access. Po0.001, unless otherwise stated. a PX0.05 b 0.01pPo0.05 c 0.001pPo0.01
the inability to identify pure cases of acute kidney injury incorrectly labeled as ESRD, and vice versa. Similarly, there is no simple way to calculate the exact number of true ESRD patients who die shortly after starting RRT and are never registered with the USRDS. Clearly, large prospective studies beginning long before ESRD onset are needed to confirm the hypotheses suggested in this study. Limitations notwithstanding, this study may provide some useful information. The sample size was large, which is a Kidney International (2014) 86, 392–398
useful feature when attempting to study sequential mortality rates in short increments of time. Studies examining survival of incident dialysis patients that rely completely on the first day of RRT to begin follow-up may not be completely reliable, and our study suggests that follow-up from approximately 6 weeks may be another useful strategy. It also highlights modifiable associations of early mortality, such as predialysis nephrologist care, choice of dialysis modality, and vascular access for hemodialysis. If our 397
clinical investigation
findings are valid, they may help inform decision making and trials of interventions targeting the extremely high mortality rates seen early in the course of dialysis treatment. MATERIALS AND METHODS Objectives Among adults initiating maintenance dialysis in the United States between 2005 and 2009, the main objectives of this study were as follows: 1. To enumerate weekly mortality rates during the first year of RRT, the main objective; 2. As a measure of the duration of a possible initial ascertainment bias associated with non-registration of ESRD cases in which death occurs early after RRT initiation, to determine the week at which peak mortality rates occurred; 3. To calculate hazard ratios for early and later mortality, defined as death occurring during weeks 7–12, inclusive, and weeks 13–51, inclusive, respectively. Subjects In this retrospective study, we studied US adults aged X18 years who initiated dialysis between 2005 and 2009 and registered with the most recent (2005) version of the USRDS Medical Evidence Report (n ¼ 498,566), which differs from previous versions by collecting data on predialysis nephrology care and type of vascular access used at initial hemodialysis therapy. Patient characteristics at dialysis initiation were determined from the Medical Evidence Report, a federal requirement for all new maintenance dialysis patients in the United States.7 Survival analyses in the USRDS depend on a combination of the ESRD Death Notification (form CMS-2746) and death information from the Medicare Enrollment Data Base (EDB). According to CMS policy, form CMS-2746 must be submitted by dialysis or transplant providers within 30 days of a patient’s death. This primary source of death information identifies about 90% of deaths. The EDB, which is populated from death data from the Social Security Administration’s Death Master File, provides the remaining mortality information. Analysis Poisson regression was used to calculate weekly mortality rates in the first year after starting dialysis therapy. Findings were similar whether or not we censored at any change of dialysis modality or at transplant, and we present findings in which follow-up ended at the earliest occurrence of change in RRT, at death, or at 1 year of followup. For cause-specific mortality, the groupings used on the 2004
398
RN Foley et al.: Early mortality in hemodialysis
Death Notification were adopted and deaths were classified as cardiovascular or non-cardiovascular.8 As an indicator of the length of the posited ascertainment bias, we calculated the time required for mortality rates to peak, based on the expectation that mortality rates should improve continuously in the initial weeks of dialysis therapy.3 Cox regression was used to calculate hazard ratios for early (weeks 7–12, inclusive) and later (weeks 13–51, inclusive) mortality. SAS, v9.1.3 (Cary, NC) was used for data analysis. DISCLOSURE
All the authors declared no competing interests.
ACKNOWLEDGMENTS
We thank the United States Renal Data System colleagues Dana Knopic, AAS, for manuscript preparation, and Nan Booth, MSW, MPH, ELS, for manuscript editing. This study was performed as a deliverable under contract no. HHSN267200715002C (National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, Maryland).
REFERENCES 1.
2.
3.
4.
5.
6.
7.
8.
Besarab A, Bolton WK, Browne JK et al. The effects of normal as compared with low hematocrit values in patients with cardiac disease who are receiving hemodialysis and epoetin. N Engl J Med 1998; 339: 584–590. Parfrey PS, Foley RN, Wittreich BH et al. Double-blind comparison of full and partial anemia correction in incident hemodialysis patients without symptomatic heart disease. J Am Soc Nephrol 2005; 16: 2180–2189. Chan KE, Maddux FW, Tolkoff-Rubin N et al. Early outcomes among those initiating chronic dialysis in the United States. Clin J Am Soc Nephrol 2011; 6: 2642–2649. Bradbury BD, Fissell RB, Albert JM et al. Predictors of early mortality among incident US hemodialysis patients in the Dialysis Outcomes and Practice Patterns Study (DOPPS). Clin J Am Soc Nephrol 2007; 2: 89–99. U.S. Department of Health and Human Services. End-Stage Renal Disease Medical Evidence Report Medicare Entitlement and/or Patient Registration: Form CMS-2728-U3. Centers for Medicare and Medicaid Services. 2006. Available at http://www.cms.gov/Medicare/CMS-Forms/CMS-Forms/ downloads/CMS2728.pdf, Accessed 9 October 2013. U.S. Department of Health and Human Services. End-Stage Renal Disease Death Notification: Form CMS-2746-U3. Centers for Medicare and Medicaid Services. 2006. Available at http://www.cms.gov/Medicare/CMS-Forms/ CMS-Forms/downloads/CMS2746.pdf. Accessed 9 October 2013. U.S. Department of Health and Human Services. End Stage Renal Disease Medical Evidence Report Medicare Entitlement and/or Patient Registration: Form CMS-2728-U3. Centers for Medicare and Medicaid Services. 2004. Available at http://www.usrds.org/2008/rg/forms/03_2728_2005.pdf, Accessed 9 October 2013. Foley RN, Gilbertson DT, Murray T et al. Long interdialytic interval and mortality among patients receiving hemodialysis. N Engl J Med 2011; 365: 1099–1107.
Kidney International (2014) 86, 392–398
http://www.kidney-international.org & 2014 International Society of Nephrology
see commentary on page 238
Early mortality in patients starting dialysis appears to go unregistered Robert N. Foley1,2, Shu-Cheng Chen1, Craig A. Solid1, David T. Gilbertson1 and Allan J. Collins1,2 1
United States Renal Data System, Minneapolis Medical Research Foundation, Minneapolis, Minnesota, USA and 2Department of Medicine, University of Minnesota, Minneapolis, Minnesota, USA
Clinical experience suggests a heightened risk associated with the transition to maintenance dialysis but few national studies have systematically examined early mortality trajectories. Here we calculated weekly mortality rates in the first year of treatment for 498,566 adults initiating maintenance dialysis in the United States (2005–2009). Mortality rates were initially unexpectedly low, peaked at 37.0 per 100 person-years in week 6, and declined steadily to 14.8 by week 51. In both early (weeks 7–12) and later (weeks 13–51) time frames, multivariate mortality associations included older age, female, Caucasian, non-Hispanic ethnicity, end-stage renal disease (ESRD) from hypertension and acute tubular necrosis, ischemic heart disease, estimated glomerular filtration rate of 15 ml/min per 1.73 m2 or more, shorter duration of nephrologist care, and hemodialysis, especially with a catheter. For early mortality risk, adjusted hazard ratios of 2 or more were seen with age over 65 (5.80 vs. under 40 years), hemodialysis with a catheter (2.73 vs. fistula), and age 40–64 (2.33). For later mortality risk, adjusted hazard ratios of 2 or more were seen with age over 65 (4.32 vs. under 40 years), hemodialysis with a catheter (2.10 vs. fistula), and age 40–64 (2.00). Thus, low initial mortality rates question the accuracy of data collected and are consistent with deaths occurring in the early weeks after starting dialysis not being registered with the United States Renal Data System. Kidney International (2014) 86, 392–398; doi:10.1038/ki.2014.15; published online 12 February 2014 KEYWORDS: ascertainment bias; dialysis; incident; mortality
Correspondence: Robert N. Foley, United States Renal Data System, Minneapolis Medical Research Foundation, 914 South 8th Street, Suite S4.100, Minneapolis, Minnesota 55404, USA. E-mail: [email protected] Received 15 March 2013; revised 4 December 2013; accepted 19 December 2013; published online 12 February 2014 392
The initial weeks of maintenance dialysis treatment are a time of clinical uncertainty. Although end-stage renal disease (ESRD) is associated with high mortality in and of itself, initiation of dialysis often adds features that could also put patients at risk. For example, initiation of in-center hemodialysis could heighten the risk of blood-borne infection, dialysis-related hypotension, hypokalemia, and loss of intrinsic renal function. Similarly, initiation of dialysis remains the most common time for introducing typical treatments for ESRD that may be dangerous if used inappropriately, with overaggressive use of erythropoiesisstimulating agents being a notable example.1,2 Although it is not known with certainty whether mortality risks associated with noninitiation of dialysis outweigh those associated with the dialysis procedure itself, at least one large study of patients initiating hemodialysis in a large chain of dialysis units in the United States suggested that this may be the case, as weekly mortality rates were maximal at the time of dialysis initiation.3 Dialysis initiation is a period of great flux, and clinical experience suggests that successful navigation through this period has long-term ramifications; however, the evolution of mortality risk early in the course of dialysis treatment remains poorly studied. This lack of information is surprising, as more information could help with clinical decision making, especially for patients making judgments about when, whether, and how to initiate renal replacement therapy (RRT). Accurate knowledge of survival expectations early in the course of dialysis treatment could be especially enlightening. Regarding mortality patterns at that time, large national dialysis patient registries have inherent attractions and potential caveats. For example, registries are typically representative, and large sample size allows for precise risk estimates and examination of findings within subgroups. On the other hand, registries may encompass potential biases, especially in the period surrounding the transition to RRT. In particular, there is concern that an early ascertainment bias may be present, with ESRD patients who die soon after dialysis initiation not being registered in national registries. In this scenario, the desire to understand outcomes ever earlier in RRT would have to be tempered against the ever-increasing likelihood of invalid findings. In addition, the fact that Medicare coverage in the Kidney International (2014) 86, 392–398
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
United States requires 90 days to have elapsed from initiation of hemodialysis may have created the impression that survival rates in the first 90 days cannot be studied when, in actuality, registration in the United States Renal Data System (USRDS) has been required for all new RRT patients for decades. As few large national studies have systematically examined early mortality trajectories in dialysis patients, we set out to address this issue among adult patients in the USRDS data.
RESULTS
Of the study population, 49.7% of subjects were aged 65 years or older, 28.8% were black, 13.7% were Hispanic, and diabetes was the cause of ESRD in 45.4% (Table 1). Peritoneal dialysis was the initial mode of RRT for 6.0% of patients, and in 77.7% hemodialysis was initiated with a catheter. Table 1 also compares patients according to duration of predialysis nephrologist care and mode of dialysis. With adjusted logistic regression, nephrologist care for o6 months was associated
Table 1 | Study population characteristics at initiation of dialysis (n ¼ 498,566), compared by duration of predialysis nephrologist care and mode of dialysis therapy Predialysis nephrologist care, months Characteristics
All
Percent of population
X6
0–5
45.6
54.4
AOR, 0–5 (vs. X6)
Mode of dialysis HD
PD
94.0
6.0
AOR, PD (vs. HD)
Age, years o40 40–64 X65
7.9 42.4 49.7
7.2 43.1 49.7
8.9 43.0 48.1
1 (Referent) 0.88 (0.86–0.91) 0.78 (0.76–0.80)
7.6 41.8 50.6
13.0 51.6 35.4
1 (Referent) 0.70 (0.67–0.73) 0.38 (0.36–0.39)
Sex Men Women
56.1 43.9
55.5 44.5
56.3 43.7
1 (Referent) 0.94 (0.93–0.95)
56.2 43.8
54.6 45.4
1 (Referent) 1.10 (1.07–1.13)
Race White Black Other
65.7 28.8 5.5
65.8 28.4 5.9
64.0 30.5 5.5
1 (Referent) 1.20 (1.19–1.22) 1.20 (1.17–1.24)
65.3 29.3 5.4
72.1 21.2 6.7
1 (Referent) 0.54 (0.53–0.56) 0.96 (0.91–1.01)
Hispanic ethnicity No Yes
86.3 13.7
85.5 14.5
85.1 14.9
1 (Referent) 1.35 (1.32–1.37)
86.2 13.8
88.4 11.6
1 (Referent) 0.67 (0.65–0.70)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
45.4 28.4 6.5 2.0 3.0 14.7
49.8 26.9 7.6 3.0 1.0 11.7
41.8 29.6 5.6 1.2 4.6 17.2
1 (Referent) 1.34 (1.32–1.36) 0.93 (0.90–0.95) 0.69 (0.66–0.72) 4.57 (4.36–4.79) 1.67 (1.64–1.70)
45.6 28.6 6.1 1.8 3.2 14.8
42.7 24.6 13.1 6.1 0.3 13.3
1 (Referent) 1.10 (1.06–1.13) 1.80 (1.73–1.87) 2.38 (2.25–2.52) 0.13 (0.11–0.17) 0.97 (0.93–1.01)
Ischemic heart disease No Yes
78.3 21.7
77.2 22.8
80.8 19.2
1 (Referent) 0.75 (0.74–0.76)
78 22
84.4 15.6
1 (Referent) 0.7 (0.68–0.72)
GFR, ml/min per 1.73 m2 o15 X15
82.6 17.4
82.9 17.1
81.4 18.6
1 (Referent) 1.17 (1.15–1.19)
82.4 17.6
84.9 15.1
1 (Referent) 1.03 (1.00–1.07)
Nephrologist care, months 412 6–12 0–5
23.2 22.4 54.4
50.9 49.1 0
— — —
22.1 22.0 55.9
40.4 29.5 30.1
1 (Referent) 0.77 (0.75–0.79) 0.32 (0.31–0.33)
Dialysis and access HD, fistula HD, graft HD, catheter PD
12.7 3.5 77.7 6
15.9 4.5 71.7 7.9
1 (Referent) 1.64 (1.58–1.7) 4.58 (4.49–4.67) 1.29 (1.25–1.33)
13.6 3.8 82.7 0
0 0 100
5.8 2.3 88.6 3.4
0 0 0 100
— — — —
Abbreviations: AOR, adjusted odds ratio; ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Data are presented as column percent or odds ratio (95% confidence interval). Logistic regression was used to calculate odds ratios, with adjustment for all the variables shown in the first column. Po0.001 for all comparisons, except AOR of GFR 415 ml/min per 1.73 m2 and association with mode of dialysis (P ¼ 0.08). Missing data: dialysis and access, 1.1%; GFR, 1.9%.
Kidney International (2014) 86, 392–398
393
Mortality rate, per 100 patient-years
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
50 40 Care < 6 Months Hemodialysis All Care . 6 Months Peritoneal dialysis
30 20 10 0 0
20 Weeks
40
Figure 1 | Weekly mortality rates during weeks 1–51, inclusive, following initiation of dialysis therapy. Care refers to duration of nephrologist care before dialysis initiation.
with younger age, male sex, black race, ESRD from hypertension and acute tubular necrosis, lack of ischemic heart disease, estimated glomerular filtration rate X15 ml/min per 1.73 m2, and hemodialysis with a catheter. Associations of initial peritoneal dialysis treatment, as opposed to hemodialysis, included younger age, female sex, white race, nonHispanic ethnicity, ESRD from glomerulonephritis and cystic disease, lack of ischemic heart disease, and longer duration of nephrologist care. During the 1-year study (mean follow-up 0.86 years), 2.2% of patients were censored because of change in RRT, whereas 23.4% of patients died from cardiac causes (35.8%), withdrawal of dialysis (9.6%), and other causes (54.6%). Figure 1 shows weekly mortality rate intervals following hemodialysis initiation. Mortality rates in the dialysis population were 5.9 per 100 person-years in week 1, peaked at 37.0 per 100 person-years in week 6, and declined steadily to intermediate levels by week 51 (14.8 per 100 person-years). Figure 1 also shows weekly mortality rates in subgroups defined by duration of nephrologist care and initial mode of dialysis. At all time points, rates were highest with nephrologist care o6 months followed by hemodialysis, nephrologist care X6 months, and peritoneal dialysis. Mortality rates peaked after 4 weeks in all subgroups and gradually declined in all subgroups except in peritoneal dialysis patients, for whom rates remained stable. Table 2 shows hazard ratios for mortality in two time frames, 7–12 weeks (inclusive) and 13–51 weeks (inclusive). In each of these time frames, multivariate analysis showed that mortality was associated with older age, female sex, white race, non-Hispanic ethnicity, ESRD from hypertension and acute tubular necrosis, ischemic heart disease, estimated glomerular filtration rate X15 ml/min per 1.73 m2, shorter duration of nephrologist care, and hemodialysis, especially with a catheter. For early mortality risk, adjusted hazard ratios (AHRs) X2 were seen with age 465 years (AHR 5.80 vs.o40 years), hemodialysis with a catheter (AHR 2.73 vs. hemodialysis with a fistula), and age 40–64 years (AHR 2.33); for later mortality, AHRs X2 were seen with age 465 years 394
(AHR 4.32), hemodialysis with a catheter (AHR 2.10), and age 40–64 years (AHR 2.00). Table 3 shows adjusted hazards for death in subgroups, defined by duration of predialysis nephrologist care. Although risk association patterns were similar to patterns in the overall population, risk gradients were generally steeper in the subgroup with shorter duration of care, with the exception of initial mode of dialysis and access in which risk gradients were similar. Table 4, summarizing mortality risk in subgroups defined by mode of dialysis, shows that hazard ratios were generally similar for hemodialysis and peritoneal dialysis patients; risk estimates differed noticeably, however, for age X65 years (vs. o40 years), especially for later mortality (AHR 4.24 for hemodialysis vs. 9.36 for peritoneal dialysis patients), and for acute tubular necrosis as cause of ESRD, especially for early mortality (AHRs (vs. diabetes)) 1.60 for hemodialysis and 7.33 for peritoneal dialysis). DISCUSSION
We found that mortality rates in US dialysis patients were approximately six times higher in week 6 than in week 1 after initiation, and declined by approximately twofold between week 6 and week 52. Early mortality rates were especially high in older subgroups, and in patients in whom acute tubular necrosis with incomplete recovery was the cause of ESRD, in patients whose duration of predialysis nephrologist care was short, and in hemodialysis patients, especially those with catheters for vascular access. In addition, the unexpectedly low mortality rates in the initial week of dialysis suggest a possible initial ascertainment bias, with high-risk patients not being registered. If this initial bias applies, mortality rates reported in this study for the first few weeks of RRT will be underestimates. Examining outcomes in short intervals requires large sample sizes for precise risk estimates. Few studies have reported mortality risk evolution in weekly increments following initiation of maintenance hemodialysis. In the Dialysis Outcomes and Practice Patterns Study, Bradbury and colleagues examined mortality risk in US patients who were started on hemodialysis between 1996 and 2004, and found that mortality rates were 25% higher in the first 3 months of treatment than in months 4–12, with greater disparities for cardiovascular than for non-cardiovascular mortality.4 Chan et al.3 examined mortality rates in 303,289 patients who initiated dialysis between 1997 and 2009 in 1733 facilities affiliated with a large US dialysis organization. They found that mortality estimates were highest in the first 2 weeks of treatment and declined successively thereafter. Unlike in our study, in which mortality estimates rose by sixfold between baseline and the 6-week peak, mortality risk was close to maximum in the first week. The observation that the relative mortality of patients on peritoneal dialysis vs. those on hemodialysis widens with the passage of time was also noted in the study by Chan et al.3 Although the cause of this disparity in time-dependent Kidney International (2014) 86, 392–398
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
Table 2 | Hazard ratios of baseline characteristics for death between 7–12 and 13–52 weeks (inclusive) after starting dialysis therapy Death, 7–12 weeks, n ¼ 481,377; 4.6% dead Characteristics
Death, 13–51 weeks, n ¼ 459,877; 16.9% dead
Unadjusted
Adjusted
Unadjusted
Adjusted
Age, years o40 40–64 X65
1 (Referent) 2.36 (2.13–2.6) 6.74 (6.12–7.43)
1 (Referent) 2.33 (2.11–2.58) 5.80 (5.24–6.41)
1 (Referent) 2.07 (1.98–2.17) 5.06 (4.84–5.29)
1 (Referent) 2.00 (1.91–2.09) 4.32 (4.12–4.52)
Sex Men Women
1 (Referent) 1.05 (1.02–1.08)
1 (Referent) 1.06 (1.04–1.09)
1 (Referent) 1.05 (1.03–1.06)
1 (Referent) 1.05 (1.04–1.07)
Race White Black Other
1 (Referent) 0.64 (0.62–0.66) 0.50 (0.46–0.54)
1 (Referent) 0.74 (0.72–0.77) 0.58 (0.54–0.63)
1 (Referent) 0.69 (0.68–0.70) 0.55 (0.53–0.57)
1 (Referent) 0.77 (0.76–0.78) 0.61 (0.59–0.63)
Hispanic ethnicity No Yes
1 (Referent) 0.62 (0.59–0.65)
1 (Referent) 0.68 (0.64–0.71)
1 (Referent) 0.65 (0.63–0.66)
1 (Referent) 0.67 (0.65–0.69)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
1 (Referent) 1.36 (1.32–1.40) 0.56 (0.52–0.61) 0.29 (0.24–0.35) 2.52 (2.38–2.67) 1.83 (1.76–1.89)
1 (Referent) 1.17 (1.13–1.21) 0.79 (0.73–0.86) 0.53 (0.44–0.64) 1.61 (1.52–1.71) 1.71 (1.65–1.78)
1 (Referent) 1.16 (1.14–1.18) 0.52 (0.50–0.54) 0.26 (0.24–0.29) 1.75 (1.69–1.81) 1.38 (1.35–1.4)
1 (Referent) 1.04 (1.02–1.05) 0.70 (0.67–0.72) 0.40 (0.37–0.44) 1.21 (1.16–1.25) 1.35 (1.32–1.37)
Ischemic heart disease No Yes
1 (Referent) 1.49 (1.45–1.54)
1 (Referent) 1.16 (1.12–1.19)
1 (Referent) 1.56 (1.54–1.59)
1 (Referent) 1.22 (1.2–1.24)
GFR, ml/min per 1.73 m2 o15 X15
1 (Referent) 1.81 (1.76–1.87)
1 (Referent) 1.53 (1.49–1.58)
1 (Referent) 1.76 (1.73–1.79)
1 (Referent) 1.51 (1.48–1.53)
Nephrology care, months 412 6–12 0–5
1 (Referent) 1.34 (1.28–1.41) 2.05 (1.97–2.13)
1 (Referent) 1.26 (1.2–1.32) 1.66 (1.6–1.73)
1 (Referent) 1.17 (1.14–1.2) 1.50 (1.47–1.53)
1 (Referent) 1.12 (1.1–1.15) 1.33 (1.3–1.35)
Dialysis and access HD, fistula HD, graft HD, catheter PD
1 (Referent) 1.36 (1.22–1.52) 2.83 (2.68–2.99) 0.64 (0.57–0.72)
1 (Referent) 1.50 (1.34–1.68) 2.73 (2.56–2.91) 0.86 (0.76–0.97)a
1 (Referent) 1.41 (1.34–1.48) 2.13 (2.07–2.18) 0.86 (0.82–0.9)
1 (Referent) 1.45 (1.38–1.52) 2.10 (2.04–2.16) 1.07 (1.02–1.12)b
Abbreviations: ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Cox regression was used to calculate hazard ratios (95% confidence intervals). Adjustment variables were age, sex, race, Hispanic ethnicity, cause of ESRD, ischemic heart disease, GFR, predialysis nephrology care, dialysis, and access. Po0.001, unless otherwise stated. a 0.01pPo0.05 b 0.001pPo0.01
mortality cannot be determined from observational studies, it is tempting to speculate that it may relate to the usual practice in which hemodialysis is almost always used when acute declines in kidney function lead to life-threatening complications, or when the burden of comorbid illnesses is large. The interplay between starting maintenance dialysis, Medicare, and the USRDS is complex. Regardless of future intentions to use Medicare as payer, a Centers for Medicare & Medicaid Services (CMS) Medical Evidence Report (form CMS-2728) must be filed within 45 days of the first dialysis Kidney International (2014) 86, 392–398
session.5 Similarly, a Death Notification (form CMS-2746) must be filed when patients die, regardless of payer designation.6 The Medical Evidence Report performs the dual role of tracking the size and composition of the national RRT program and potential enrollment in Medicare for coverage of medical care. New Medicare coverage for incenter patients is available after 3 months of in-center hemodialysis. In this regard, it is plausible that some of the unexpectedly low mortality rates very early in the course of dialysis treatment may reflect the lack of a financial incentive to complete these forms for patients who lack Medicare 395
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
Table 3 | Adjusted hazard ratios of baseline characteristics for death between 7–12 and 13–52 weeks (inclusive), according to duration of predialysis nephrology care Death, 7–12 weeks
Death, 13–52 weeks
Care X6 months, n ¼ 221,698; 3.3% dead
Care 0–5 months, n ¼ 259,679; 5.7% dead
CareX6 months, n ¼ 214,598; 14.3% dead
Care 0–5 months, n ¼ 245,279; 19.1% dead
Age, years o40 40–64 X65
1 (Referent) 2.03 (1.67–2.46) 5.20 (4.31–6.27)
1 (Referent) 2.59 (2.31–2.92) 7.32 (6.53–8.21)
1 (Referent) 2.08 (1.91–2.26) 4.60 (4.23–5.00)
1 (Referent) 2.11 (2.00–2.23) 5.08 (4.82–5.35)
Sex Men Women
1 (Referent) 1.02 (0.97–1.07)a
1 (Referent) 1.10 (1.06–1.13)
1 (Referent) 1.03 (1–1.05)b
1 (Referent) 1.09 (1.07–1.11)
Race White Black Other
1 (Referent) 0.72 (0.68–0.76) 0.57 (0.50–0.66)
1 (Referent) 0.63 (0.60–0.65) 0.50 (0.46–0.55)
1 (Referent) 0.71 (0.69–0.73) 0.62 (0.58–0.66)
1 (Referent) 0.71 (0.70–0.73) 0.55 (0.52–0.57)
Hispanic ethnicity No Yes
1 (Referent) 0.70 (0.64–0.76)
1 (Referent) 0.58 (0.55–0.61)
1 (Referent) 0.69 (0.66–0.71)
1 (Referent) 0.61 (0.59–0.63)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
1 (Referent) 1.18 (1.12–1.25) 0.77 (0.68–0.87) 0.51 (0.40–0.67) 2.02 (1.74–2.34) 1.63 (1.52–1.75)
1 (Referent) 1.28 (1.23–1.33) 0.59 (0.53–0.65) 0.34 (0.26–0.45) 2.02 (1.89–2.16) 1.74 (1.67–1.82)
1 (Referent) 1.06 (1.03–1.09) 0.71 (0.67–0.75) 0.43 (0.38–0.49) 1.48 (1.36–1.61) 1.30 (1.26–1.35)
1 (Referent) 1.09 (1.07–1.12) 0.52 (0.49–0.55) 0.26 (0.22–0.30) 1.49 (1.43–1.55) 1.35 (1.31–1.38)
Ischemic heart disease No Yes
1 (Referent) 1.21 (1.15–1.27)
1 (Referent) 1.52 (1.47–1.58)
1 (Referent) 1.26 (1.23–1.29)
1 (Referent) 1.57 (1.54–1.6)
GFR, ml/min per 1.73 m2 o15 X15
1 (Referent) 1.62 (1.53–1.7)
1 (Referent) 1.71 (1.64–1.77)
1 (Referent) 1.51 (1.47–1.55)
1 (Referent) 1.72 (1.69–1.76)
Nephrology care, months 412 6–12 0–5
1 (Referent) 1.26 (1.20–1.32) —
— — —
1 (Referent) 1.12 (1.10–1.15) —
— — —
Dialysis and access HD, fistula HD, graft HD, catheter PD
1 (Referent) 1.35 (1.16–1.56) 2.49 (2.30–2.69) 0.81 (0.69–0.94)c
1 (Referent) 1.25 (1.06–1.47)c 2.23 (2.05–2.43) 0.57 (0.47–0.68)
1 (Referent) 1.43 (1.35–1.53) 2.10 (2.03–2.17) 1.09 (1.03–1.16)c
1 (Referent) 1.28 (1.19–1.39) 1.75 (1.68–1.83) 0.73 (0.67–0.79)
Characteristics
Abbreviations: ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Cox regression was used to calculate hazard ratios (95% confidence intervals). Adjustment variables were age, sex, race, Hispanic ethnicity, cause of ESRD, ischemic heart disease, GFR, predialysis nephrology care, dialysis, and access. Po0.001, unless otherwise stated. a PX0.05 b 0.01pPo0.05 c 0.001pPo0.01
coverage and who die before 3 months. Our findings are consistent with the hypothesis that registration of patients at high risk of early death may be incomplete. If this is the case, it may be appropriate to wait until approximately 6 weeks have elapsed in studies in which death is a principal outcome. Our study is not without limitations and shares all the imperfections of retrospective, registry-based studies, including lack of prospective validation of key data elements. Care should be exerted when attempting to generalize the findings 396
from this study to other countries. This was a nonexperimental study of associations, and causes and effects cannot be separated with confidence. As this was a study of incident dialysis patients, many of the characteristics described may reflect care in the years preceding dialysis initiation. Although information regarding some markers of nephrology care was collected, complete and comprehensive descriptors were not available. A particular limitation of studies purporting to start follow-up at the beginning of maintenance dialysis is Kidney International (2014) 86, 392–398
clinical investigation
RN Foley et al.: Early mortality in hemodialysis
Table 4 | Adjusted hazard ratios of baseline characteristics for death between 7–12 and 13–52 weeks (inclusive) according to initial dialysis therapy Death, 7–12 weeks
Death, 13–52 weeks
HD, n ¼ 451,928; 4.8% dead
PD, n ¼ 29,449; 1.3% dead
HD, n ¼ 430,792; 17.5% dead
PD, n ¼ 29,085; 8.3% dead
Age, years o40 40–64 X65
1 (Referent) 2.36 (2.13–2.62) 5.83 (5.26–6.46)
1 (Referent) 1.59 (0.92–2.74)a 6.53 (3.88–10.99)
1 (Referent) 1.98 (1.89–2.07) 4.24 (4.04–4.44)
1 (Referent) 2.86 (2.24–3.67) 9.36 (7.35–11.91)
Sex Men Women
1 (Referent) 1.07 (1.04–1.1)
1 (Referent) 0.91 (0.74–1.12)a
1 (Referent) 1.05 (1.03–1.06)
1 (Referent) 0.99 (0.91–1.07)a
Race White Black Other
1 (Referent) 0.74 (0.72–0.77) 0.59 (0.54–0.63)
1 (Referent) 0.59 (0.44–0.79) 0.46 (0.26–0.8)c
1 (Referent) 0.77 (0.76–0.78) 0.61 (0.59–0.64)
1 (Referent) 0.59 (0.53–0.66) 0.50 (0.41–0.62)
Hispanic ethnicity No Yes
1 (Referent) 0.68 (0.64–0.71)
1 (Referent) 0.60 (0.41–0.89)b
1 (Referent) 0.67 (0.65–0.69)
1 (Referent) 0.52 (0.44–0.61)
Cause of ESRD Diabetes Hypertension Glomerulonephritis Cystic disease ATN Other
1 (Referent) 1.17 (1.13–1.21) 0.80 (0.73–0.86) 0.55 (0.45–0.66) 1.60 (1.51–1.71) 1.71 (1.64–1.77)
1 (Referent) 1.71 (1.34–2.18) 0.54 (0.35–0.85)c 0.26 (0.11–0.64)c 7.33 (3.23–16.61) 1.89 (1.42–2.5)
1 (Referent) 1.04 (1.02–1.06) 0.71 (0.68–0.74) 0.42 (0.38–0.47) 1.21 (1.17–1.26) 1.36 (1.33–1.38)
1 (Referent) 1.03 (0.94–1.14)a 0.32 (0.27–0.39) 0.20 (0.14–0.28) 2.49 (1.52–4.08) 1.02 (0.91–1.15)a
Ischemic heart disease No Yes
1 (Referent) 1.16 (1.12–1.19)
1 (Referent) 1.91 (1.51–2.41)
1 (Referent) 1.21 (1.19–1.23)
1 (Referent) 2.24 (2.05–2.45)
GFR, ml/min per 1.73 m2 o15 X15
1 (Referent) 1.53 (1.49–1.58)
1 (Referent) 1.98 (1.57–2.51)
1 (Referent) 1.51 (1.48–1.53)
1 (Referent) 2.02 (1.84–2.22)
Nephrology care, months 412 6–12 0–5
1 (Referent) 1.25 (1.19–1.31) 1.66 (1.59–1.73)
1 (Referent) 1.54 (1.19–2.00)c 1.78 (1.39–2.29)
1 (Referent) 1.12 (1.09–1.14) 1.33 (1.30–1.35)
1 (Referent) 1.16 (1.05–1.28)c 1.22 (1.11–1.35)
Dialysis and access HD, fistula HD, graft HD, catheter PD
1 (Referent) 1.50 (1.34–1.68) 2.73 (2.56–2.92) —
— — — —
1 (Referent) 1.45 (1.38–1.52) 2.10 (2.04–2.16) —
— — — —
Characteristics
Abbreviations: ATN, acute tubular necrosis; ESRD, end-stage renal disease; GFR, glomerular filtration rate; HD, hemodialysis; PD, peritoneal dialysis. Note: Cox regression was used to calculate hazard ratios (95% confidence intervals). Adjustment variables were age, sex, race, Hispanic ethnicity, cause of ESRD, ischemic heart disease, GFR, predialysis nephrology care, dialysis, and access. Po0.001, unless otherwise stated. a PX0.05 b 0.01pPo0.05 c 0.001pPo0.01
the inability to identify pure cases of acute kidney injury incorrectly labeled as ESRD, and vice versa. Similarly, there is no simple way to calculate the exact number of true ESRD patients who die shortly after starting RRT and are never registered with the USRDS. Clearly, large prospective studies beginning long before ESRD onset are needed to confirm the hypotheses suggested in this study. Limitations notwithstanding, this study may provide some useful information. The sample size was large, which is a Kidney International (2014) 86, 392–398
useful feature when attempting to study sequential mortality rates in short increments of time. Studies examining survival of incident dialysis patients that rely completely on the first day of RRT to begin follow-up may not be completely reliable, and our study suggests that follow-up from approximately 6 weeks may be another useful strategy. It also highlights modifiable associations of early mortality, such as predialysis nephrologist care, choice of dialysis modality, and vascular access for hemodialysis. If our 397
clinical investigation
findings are valid, they may help inform decision making and trials of interventions targeting the extremely high mortality rates seen early in the course of dialysis treatment. MATERIALS AND METHODS Objectives Among adults initiating maintenance dialysis in the United States between 2005 and 2009, the main objectives of this study were as follows: 1. To enumerate weekly mortality rates during the first year of RRT, the main objective; 2. As a measure of the duration of a possible initial ascertainment bias associated with non-registration of ESRD cases in which death occurs early after RRT initiation, to determine the week at which peak mortality rates occurred; 3. To calculate hazard ratios for early and later mortality, defined as death occurring during weeks 7–12, inclusive, and weeks 13–51, inclusive, respectively. Subjects In this retrospective study, we studied US adults aged X18 years who initiated dialysis between 2005 and 2009 and registered with the most recent (2005) version of the USRDS Medical Evidence Report (n ¼ 498,566), which differs from previous versions by collecting data on predialysis nephrology care and type of vascular access used at initial hemodialysis therapy. Patient characteristics at dialysis initiation were determined from the Medical Evidence Report, a federal requirement for all new maintenance dialysis patients in the United States.7 Survival analyses in the USRDS depend on a combination of the ESRD Death Notification (form CMS-2746) and death information from the Medicare Enrollment Data Base (EDB). According to CMS policy, form CMS-2746 must be submitted by dialysis or transplant providers within 30 days of a patient’s death. This primary source of death information identifies about 90% of deaths. The EDB, which is populated from death data from the Social Security Administration’s Death Master File, provides the remaining mortality information. Analysis Poisson regression was used to calculate weekly mortality rates in the first year after starting dialysis therapy. Findings were similar whether or not we censored at any change of dialysis modality or at transplant, and we present findings in which follow-up ended at the earliest occurrence of change in RRT, at death, or at 1 year of followup. For cause-specific mortality, the groupings used on the 2004
398
RN Foley et al.: Early mortality in hemodialysis
Death Notification were adopted and deaths were classified as cardiovascular or non-cardiovascular.8 As an indicator of the length of the posited ascertainment bias, we calculated the time required for mortality rates to peak, based on the expectation that mortality rates should improve continuously in the initial weeks of dialysis therapy.3 Cox regression was used to calculate hazard ratios for early (weeks 7–12, inclusive) and later (weeks 13–51, inclusive) mortality. SAS, v9.1.3 (Cary, NC) was used for data analysis. DISCLOSURE
All the authors declared no competing interests.
ACKNOWLEDGMENTS
We thank the United States Renal Data System colleagues Dana Knopic, AAS, for manuscript preparation, and Nan Booth, MSW, MPH, ELS, for manuscript editing. This study was performed as a deliverable under contract no. HHSN267200715002C (National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, Maryland).
REFERENCES 1.
2.
3.
4.
5.
6.
7.
8.
Besarab A, Bolton WK, Browne JK et al. The effects of normal as compared with low hematocrit values in patients with cardiac disease who are receiving hemodialysis and epoetin. N Engl J Med 1998; 339: 584–590. Parfrey PS, Foley RN, Wittreich BH et al. Double-blind comparison of full and partial anemia correction in incident hemodialysis patients without symptomatic heart disease. J Am Soc Nephrol 2005; 16: 2180–2189. Chan KE, Maddux FW, Tolkoff-Rubin N et al. Early outcomes among those initiating chronic dialysis in the United States. Clin J Am Soc Nephrol 2011; 6: 2642–2649. Bradbury BD, Fissell RB, Albert JM et al. Predictors of early mortality among incident US hemodialysis patients in the Dialysis Outcomes and Practice Patterns Study (DOPPS). Clin J Am Soc Nephrol 2007; 2: 89–99. U.S. Department of Health and Human Services. End-Stage Renal Disease Medical Evidence Report Medicare Entitlement and/or Patient Registration: Form CMS-2728-U3. Centers for Medicare and Medicaid Services. 2006. Available at http://www.cms.gov/Medicare/CMS-Forms/CMS-Forms/ downloads/CMS2728.pdf, Accessed 9 October 2013. U.S. Department of Health and Human Services. End-Stage Renal Disease Death Notification: Form CMS-2746-U3. Centers for Medicare and Medicaid Services. 2006. Available at http://www.cms.gov/Medicare/CMS-Forms/ CMS-Forms/downloads/CMS2746.pdf. Accessed 9 October 2013. U.S. Department of Health and Human Services. End Stage Renal Disease Medical Evidence Report Medicare Entitlement and/or Patient Registration: Form CMS-2728-U3. Centers for Medicare and Medicaid Services. 2004. Available at http://www.usrds.org/2008/rg/forms/03_2728_2005.pdf, Accessed 9 October 2013. Foley RN, Gilbertson DT, Murray T et al. Long interdialytic interval and mortality among patients receiving hemodialysis. N Engl J Med 2011; 365: 1099–1107.
Kidney International (2014) 86, 392–398
