Predialysis Health, Dialysis Timing, and Outcomes among Older United States Adults

Abstract

Studies of dialysis initiation timing have not accounted for predialysis clinical factors that could impact postdialysis outcomes. We examined the association of predialysis health with timing of dialysis initiation in older adult patients in the United States and contrasted morbidity and mortality outcomes among patients with early [estimated GFR (eGFR)≥10 ml/min per 1.73 m²] versus later (eGFR<10 ml/min per 1.73 m²) initiation. We included all patients from the US Renal Data System who initiated dialysis between 2006 and 2008, were ≥67 years old, and had ≥2 years of prior Medicare coverage (n=84,654). We calculated patients’ propensity to initiate dialysis early and matched patients by propensity scores. Cox models were used to compare risks of mortality and hospitalization among initiation groups. The majority (58%) of patients initiated dialysis early. Early initiators were more likely to have had AKI, multiple congestive heart failure admissions, and other hospitalizations preceding initiation. Among propensity-matched patients (n=61,930), early initiation associated with greater all-cause (hazard ratio [HR], 1.11; 95% confidence interval [95% CI], 1.08 to 1.14), cardiovascular (CV; HR, 1.13; 95% CI, 1.09 to 1.17), and infectious (HR, 1.13; 95% CI, 1.06 to 1.22) mortality and greater all-cause (HR, 1.03; 95% CI, 1.01 to 1.05) and infectious (HR, 1.10; 95% CI, 1.07 to 1.13) hospitalizations. There was no difference in CV hospitalizations. Among these older adults, early dialysis initiation associates with greater mortality and hospitalizations, even after accounting for predialysis clinical factors. These findings do not support the common practice of early dialysis initiation in the United States.

There has been a global trend toward initiation of dialysis at higher levels of kidney function, particularly among older adults.¹ This trend may have been driven, in part, by clinicians’ beliefs that earlier initiation could better preserve patients’ nutritional status, thereby decreasing their risks of morbidity and mortality.^2–4 However, recent studies have questioned the ability of dialysis to prevent functional decline in institutionalized older adults⁵ and suggest greater mortality with early dialysis initiation in several patient groups.^6–8 Definitive clinical trials have been challenged by the influence of patient and clinician preferences and the unpredictable course that often accompanies renal function decline. The Initiating Dialysis Early and Late (IDEAL) trial, which found no difference in survival, did not achieve the desired difference in kidney function between early and late initiators, because the majority of patients randomized to the late-start group initiated dialysis before reaching the target estimated GFR (eGFR). At initiation of dialysis in the IDEAL trial, the mean eGFR was 12.0 ml/min in the early-start group compared with 9.8 ml/min in the late-start group.⁹ Furthermore, the IDEAL trial may not be generalizable to older adults in the United States, because the study included a relatively healthy ESRD population (e.g., 5.5% of study patients had congestive heart failure [CHF]⁹ compared with 32.5% of incident dialysis patients in the United States in 2007 had CHF¹). Thus, rigorous observational studies still have an important role in elucidating optimal dialysis initiation strategies among older United States adults.

To our knowledge, which was corroborated by a recent meta-analysis,⁷ prior studies of outcomes related to timing of dialysis initiation have not carefully examined the predialysis health course (e.g., incidence of AKI or CHF hospitalization) or specifically focused on older adults, the fastest growing US ESRD population.¹⁰ Furthermore, studies have not examined specific causes of morbidity and mortality experienced by patients after their earlier versus later dialysis initiation.⁷

We examined the association of older United States adult patients’ predialysis health with timing of dialysis initiation and compared morbidity and mortality outcomes after early versus later dialysis initiation using data from predialysis Medicare claims linked to the US Renal Data System (USRDS).

Results

Patient Characteristics by Timing of Dialysis Initiation

A total of 84,654 patients was included in our study. Exclusions are summarized in Figure 1. Mean age of the study population was 76.7 (SD=6.3) years. Median eGFR at dialysis initiation was 11.0 (interquartile range=8.1–14.3) ml/min per 1.73 m². The majority (57.7%) initiated dialysis early (eGFR>10 ml/min per 1.73 m²). Early and later initiators differed substantially (Table 1). Examination of predialysis clinical factors revealed AKI to be more common among early initiators, whereas there were no differences in uremia or hyperkalemia between the groups. Predialysis CHF hospitalizations and total hospital days were more common among early initiators.

Figure 1.

Overview of cohort formation.

Table 1.

Patient characteristics by timing of dialysis initiation before and after propensity score matching

Patient Characteristics	Before Match, % (n=84,654)		Logistic Regression of Predictors of Early Initiationa Odds Ratio (95% CI)	After Match, %b (n=61,930)
	Later (eGFR<10 ml/min per 1.73 m2)	Early (eGFR≥10 ml/min per 1.73 m2)		Later (eGFR<10 ml/min per 1.73 m2)	Early (eGFR≥10 ml/min per 1.73 m2)
n (%)	35,769 (42.3%)	48,885 (57.7%)	—	30,965 (50.0%)	30,965 (50.0%)
Mean age, yr (SD)	76.7 (6.3)	77.5 (6.4)	—	77.0 (6.3)	77.0 (6.4)
Age categories, yr (%)
67–69	15.0	12.0	Reference	13.8	13.7
70–74	26.2	23.4	1.11 (1.05 to 1.16)	25.3	25.3
75–79	25.6	26.1	1.26 (1.20 to 1.32)	25.9	25.9
80–84	20.5	23.0	1.40 (1.33 to 1.47)	21.4	21.5
≥85	12.8	15.4	1.56 (1.47 to 1.65)	13.7	13.6
Men	47.7	58.7	Reference	51.1	50.7
Women	52.3	41.3	0.59 (0.57 to 0.60)	49.0	49.3
Race
White	76.6	78.4	Reference	77.0	76.8
Black	18.4	18.0	1.12 (1.07 to 1.16)	18.6	18.8
Other	5.0	3.6	0.78 (0.73 to 0.84)	4.4	4.4
Primary cause of ESRD (%)
Diabetes mellitus	35.4	42.9	Reference	38.4	38.3
Hypertension	37.5	37.4	0.89 (0.85 to 0.93)	38.2	38.2
GN	11.7	7.0	0.65 (0.62 to 0.69)	9.3	9.3
Other/missing/unknown	15.5	12.7	0.79 (0.75 to 0.83)	14.1	14.2
Year of dialysis initiation
2006	37.0	33.8	Reference	35.7	35.8
2007	32.6	33.1	1.15 (1.11 to 1.19)	32.8	32.8
2008	30.4	33.1	1.27 (1.22 to 1.31)	31.4	31.4
Dialysis modality
Hemodialysis	96.2	95.3	Reference	95.9	95.9
Peritoneal dialysis	3.8	4.7	1.26 (1.10 to 1.45)	4.1	4.1
Vascular access
Arteriovenous fistula	13.7	13.7	Reference	14.1	14.1
Arteriovenous graft	4.0	4.2	1.08 (0.99 to 1.17)	4.1	4.3
Catheter	77.4	76.1	0.89 (0.86 to 0.93)	76.5	76.3
Comorbid conditions			c
CHF	61.0	76.1	1.32 (1.27 to 1.38)	66.6	67.0
Atherosclerotic heart disease	58.3	72.3	1.16 (1.12 to 1.20)	63.5	63.6
Cerebrovascular disease	29.1	34.9	1.04 (1.01 to 1.08)	31.0	31.1
Diabetes mellitus	58.5	68.2	1.26 (1.21 to 1.31)	62.4	62.5
Chronic obstructive pulmonary disease	32.9	42.9	1.19 (1.16 to 1.23)	35.4	35.5
Peripheral vascular disease	45.2	54.2	1.08 (1.04 to 1.11)	48.1	48.2
Other CV disease	50.2	63.2	1.17 (1.13 to 1.21)	54.3	54.4
Dysrhythmia	39.9	54.4	1.23 (1.19 to 1.27)	43.8	43.8
Cancer	24.8	23.4	0.91 (0.88 to 0.94)	24.0	24.0
Gastrointestinal bleeding	15.9	20.0	1.04 (1.00 to 1.08)	17.0	17.0
Liver disease	2.8	4.1	1.37 (1.26 to 1.49)	3.0	3.1
Nursing home resident	8.7	11.6	1.17 (1.11 to 1.23)	9.4	9.5
Functional limitation	16.7	20.0	1.04 (1.00 to 1.08)	17.6	17.7
Mean serum albumin, g/dl (SD)	3.2 (0.7)	3.2 (0.7)	—	3.2 (0.7)	3.2 (0.7)
<3.0	33.5	32.1	0.96 (0.92 to 1.01)	32.9	33.1
3.0–3.4	17.5	17.9	Reference	17.6	17.6
>3.4	27.3	27.2	1.04 (0.99 to 1.08)	27.5	27.4
Mean hemoglobin, g/dl (SD)	10.0 (1.6)	10.4 (1.5)	—	10.1 (1.6)	10.3 (1.5)
<11	66.9	61.6	0.74 (0.71 to 0.77)	65.2	65.1
11.0–12.0	15.6	18.7	Reference	16.6	16.8
>12.0	9.3	11.4	1.03 (0.98 to 1.09)	10.0	9.9
BMI, kg/m2 (SD)	27.1 (6.5)	27.5 (6.8)	—	27.2 (6.6)	27.4 (6.9)
<18.5	4.6	4.2	1.10 (1.03 to 1.19)	4.4	4.4
18.5–25	37.6	36.1	Reference	36.8	36.6
25–30	30.9	30.0	0.95 (0.92 to 0.99)	30.6	30.6
>30	25.8	28.4	1.12 (1.07 to 1.16)	27.0	27.2
Predialysis clinical factorsd			c
AKI	61.0	63.9	0.77 (0.74 to 0.80)	60.9	61.0
Uremia	36.1	36.5	0.94 (0.91 to 0.96)	36.0	36.0
Hyperkalemia	32.0	32.0	0.87 (0.84 to 0.90)	31.7	31.8
CHF hospital admissions
None	65.2	50.4	Reference	61.3	61.2
1	22.2	25.1	1.07 (1.03 to 1.12)	24.3	24.2
1–2	7.8	12.8	1.33 (1.25 to 1.41)	8.9	9.0
>2	4.8	11.7	1.68 (1.57 to 1.81)	5.6	5.6
Total hospital days
0	55.4	43.2	Reference	52.5	52.3
<6	18.1	18.1	1.12 (1.07 to 1.17)	18.5	18.4
6–15	14.9	18.4	1.25 (1.19 to 1.31)	16.0	16.1
>15	11.6	20.4	1.52 (1.44 to 1.61)	13.0	13.2

Not all categorical variables sum to 100% because of missing data.

^aAll other variables are included in each logistic regression model.

^bP values comparing later and early groups after matching were all >0.05.

^cReference group is patients without the condition.

^dPredialysis clinical factors were each measured in 6 months preceding dialysis initiation.

Many characteristics were independently associated with early initiation after adjustment for all other variables (Table 1). Among predialysis factors, AKI, uremia, and hyperkalemia were negatively associated with early initiation. In contrast, CHF admissions and total hospital days were positively associated with early dialysis initiation.

Propensity Matching of Cohort

A total of 61,930 (71.5%) patients was propensity matched (Figure 1). Median eGFR was 13.3 ml/min per 1.73 m² for early initiators and 7.8 ml/min per 1.73 m² for later initiators. Patient characteristics were well balanced after propensity matching (Table 1). Most (78.9%) of unmatched patients were early initiators, with a median eGFR of 13.0 ml/min per 1.73 m², and larger proportions had experienced AKI (67.5%), greater than two CHF hospitalizations (17.6%), and/or greater than 15 total hospital days (26.4%) in the 6 months before dialysis initiation compared with the matched patients. Forty-six percent of unmatched patients (n=10,396) died during follow-up.

All-Cause and Cause-Specific Mortality

A total of 25,151 (40.6%) patients in our matched cohort died during follow-up. Estimated median survival was 23.4 months (later initiators=24.9 months; early initiators=22.1 months; P<0.001) (Figure 2). Overall, death from a cardiovascular (CV) cause was more common than infectious causes of death (175.7 versus 47.5 per 1000 person-years). Early initiation was associated with greater risk of all-cause, CV, and infectious mortality compared with later initiation (Table 2). Subgroup analyses of all-cause mortality generally revealed similar findings across subgroups (P interaction>0.05 for all) (Figure 3). However, among patients with an arteriovenous fistula (AVF) or arteriovenous graft (AVG) as their dialysis access, there was no statistically significant difference in survival comparing early with later initiation (hazard ratio [HR], 1.05; 95% confidence interval [95% CI], 0.98 to 1.13).

Figure 2.

Proportion of patients surviving from time of dialysis initiation, among those initiating at eGFR <10 and ≥10 ml/min per 1.73 m².

Table 2.

Clinical outcomes by timing of dialysis initiation

Outcome	N for Analysisa		Number of Events		Incidence Rate per 1000 person-yrs		Relative Risk of Early Versus Later (Referent) Timing of Dialysis Initiation HR (95% CI)
	Early	Later	Early	Later	Early	Later
All-cause mortality	30,965	30,965	12,999	12,152	402.2	360.6	1.11 (1.08 to 1.14)
CV mortality	30,965	30,965	6049	5551	187.2	164.7	1.13 (1.09 to 1.17)
Infectious mortality	30,965	30,965	1637	1496	50.7	44.4	1.13 (1.06 to 1.22)
All-cause hospitalization	30,731	30,771	62,967	63,857	2079.8	2017.8	1.03 (1.01 to 1.05)
CV hospitalization	30,731	30,771	19,512	20,552	623.7	629.9	0.99 (0.96 to 1.01)
Infectious hospitalization	30,731	30,771	16,829	15,911	538.5	487.5	1.10 (1.07 to 1.13)

^aN for analysis for hospitalizations is unbalanced between early and later initiators because of patients being followed from day 8 after dialysis initiation for hospitalization analyses.

Figure 3.

Hazard of mortality was statistically significantly greater for early as compared to later dialysis initiators across all patient subgroups except those patients with an AV fistula or graft. AV, arteriovenous.

All-Cause and Cause-Specific Hospitalizations

Over two thirds of patients (n=42,912) were hospitalized at least one time in the follow-up period. Risk of all-cause hospitalization was greater for early initiators (HR, 1.03; 95% CI, 1.01 to 1.05); risk of infectious hospitalization was also greater (HR, 1.10; 95% CI, 1.07 to 1.13) (Table 2). However, there was no difference in risk of CV hospitalization between groups (HR, 0.99; 95% CI, 0.96 to 1.01). We analyzed the risk of CHF hospitalization specifically, which also revealed no difference between groups (HR, 0.98; 95% CI, 0.93 to 1.03). Results for all-cause hospitalization in the subgroups tested showed no difference or greater risk of hospitalization among early initiators (Supplemental Appendix A). The Andersen–Gill approach with hospitalization-dependent hazard yielded the same results.

Sensitivity Analyses

We conducted multiple sensitivity analyses to test our findings. In an analysis in which we did not include variables reflecting patients’ predialysis health status (i.e., AKI, uremia, hyperkalemia, CHF admissions, and all-cause hospital days), findings were very similar to our primary analysis, although they yielded a slightly greater association between early initiation and mortality. For example, our sensitivity analysis, which did not include the predialysis variables, yielded a relative HR of all-cause mortality among those patients initiating dialysis early versus later of 1.12 (95% CI, 1.09 to 1.15), which was slightly greater in magnitude than our primary findings (HR, 1.11; 95% CI, 1.08 to 1.14). We observed similar findings for CV disease mortality (HR, 1.15; 95% CI, 1.11 to 1.20) for analyses not including the predialysis variables compared with our primary analyses, which yielded an HR of 1.13 (95% CI, 1.09 to 1.17). In a second analysis restricted to hemodialysis patients, findings were similar to our primary analysis (HR for all-cause mortality, 1.11; 95% CI, 1.08 to 1.14; HR for all-cause hospitalization, 1.03; 95% CI, 1.01 to 1.05). A third analysis, in which we included patients who recovered kidney function, yielded findings similar to our primary analysis (HR for all-cause mortality, 1.10; 95% CI, 1.08 to 1.13; and HR for all-cause mortality and all-cause hospitalization, 1.03; 95% CI, 1.01 to 1.05, comparing early to later initiators). A fourth analysis among patients presumed to be well nourished (i.e., serum albumin≥3.5 mg/dl or body mass index [BMI]≥25 kg/m²) and thus, less likely than those patients with poorer nutritional status to have their GFR falsely estimated because of lower creatinine production¹¹ also yielded similar findings to our primary analysis. Among patients with serum albumin≥3.5 mg/dl (n=17,016), HRs for all-cause mortality and all-cause hospitalization were 1.10 (95% CI, 1.04 to 1.16) and 1.03 (95% CI, 0.99 to 1.07) for early versus later initiators, respectively. Among patients with a BMI≥25 kg/m² (n=35,737), the HRs for all-cause mortality and all-cause hospitalization were 1.11 (95% CI, 1.07 to 1.15) and 1.04 (95% CI, 1.01 to 1.06) for early versus later initiators, respectively.

Because creatinine-based estimates of GFR could be particularly unstable in the setting of AKI,¹² we performed a fifth analysis to explore whether our findings were consistent among patients without AKI in the days leading up to dialysis initiation. In this analysis, we excluded 32,797 patients who had documented AKI (de novo or AKI on CKD) in the 30 days up to the date of dialysis initiation (53% of our matched cohort) and patients with acute tubular necrosis (ATN) as their cause of ESRD (n=1859). Among the remaining 28,782 patients, we found similar findings to our primary analysis (HR, 1.17; 95% CI, 1.12 to 1.22 for all-cause mortality; HR, 1.02; 95% CI, 0.99 to 1.05 for all-cause hospitalizations) comparing early with later initiators.

Finally, to confirm that our findings were consistent among those patients with adequate predialysis nephrology care for advanced CKD and without AKI as their cause of ESRD, we performed a sixth analysis and included patients who had (1) at least 12 months of outpatient nephrology care before initiation, (2) an ESRD cause other than ATN, and (3) an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code of 585.4 (CKD stage 4) during the 2 years before initiation. Of 84,654 patients, 27,098 patients met our criteria for adequate predialysis care, and 19,554 (72.2%) patients were successfully matched. Similar to our primary analysis, risks for all-cause mortality and all-cause hospitalization were greater for early compared with later initiators (HR, 1.12; 95% CI, 1.07 to 1.18; HR, 1.04; 95% CI, 1.00 to 1.08, respectively).

Discussion

In this national study, we found that early dialysis initiation was common among older United States adults and associated with their greater mortality and hospitalizations, even after accounting for their predialysis health status and health care use. Our findings were consistent across multiple subgroups, including those patients presumed to have had adequate predialysis care and those patients without AKI in the month preceding dialysis initiation.

To our knowledge, no previous study has examined outcomes associated with early versus later dialysis initiation among older adults.⁷ Previous observational studies in other populations have shown conflicting findings, with the majority of studies showing greater mortality risk with early dialysis initiation^8,13–25 and some studies showing no difference²⁶ or improved survival for earlier initiators.^27–29 These studies have incorporated limited or no information on critical predialysis factors, which could heavily influence the timing of dialysis initiation and morbidity and mortality after dialysis, including nephrology care, comorbidity, and common reasons for early initiation (e.g., AKI, frequent hospitalization, and CHF exacerbations). These factors are especially important for older adults, because they suffer a greater burden of comorbid illness compared with their younger counterparts,³⁰ often progress to ESRD after episodes of AKI,³¹ and may be more likely than younger persons to initiate dialysis under suboptimal circumstances (e.g., emergent inpatient settings).³² In our study, we identified previously unreported independent risk factors for early dialysis initiation (CHF hospital admissions and all-cause hospitalization days in the 6 months preceding initiation) and found that the consideration of these factors and other predialysis factors served to attenuate, at least somewhat, the risk of poor outcomes associated with early initiation.

The consistency of our findings across numerous subgroups should be considered in conjunction with the recent study by Bao et al.⁸ reporting that frailty was associated with early initiation and attenuated the relation of early dialysis initiation and mortality among a subset of 1576 USRDS registrants (adjusted HR, 1.08; 95% CI, 0.98 to 1.19 per 5 ml/min per 1.73 m²). In our study, we found that nursing home residence was independently associated with early initiation among older adults but that mortality was statistically significantly greater for early initiators regardless of nursing home or functional status, likely owing to our larger sample size, older adult population, and administrative ascertainment of our measures of functional limitation. However, when we examined risk of hospitalization, there was no difference between early and later initiators either residing in a nursing home or with functional limitations (Supplemental Appendix A). Considered together, our study and the study by Bao et al.⁸ suggest that early dialysis initiation does not benefit patients with functional limitations. Our findings among vascular access subgroups also deserve comment. We found no statistical difference in survival or hospitalizations among patients with an AVF or AVG (less than 25% of patients). This finding could be attributed to several potential reasons, including these patients’ greater likelihood to have received predialysis care, their potential better health, which is reflected by vasculature able to support access creation, or small numbers of patients in this group, yielding too few patients to observe statistical differences in clinical outcomes. Patients with worse or unstable health might be less likely to obtain early vascular access and start dialysis earlier because of factors that we may not have been able to completely account for in our analysis.

Our examination of the association of timing of dialysis initiation with cause-specific morbidity and mortality underscores mechanisms through which early dialysis initiation may be harmful. Patients with preserved urine output at 1 year, for example, have greater survival and less inflammation compared with patients with anuria.^33,34 Because over 50% of endogenous renal function may be lost in the first few months of hemodialysis treatment,³⁵ patients initiated early may experience a more rapid change in their inflammatory state relative to those patients initiated later who have lost residual function more gradually, which could increase early initiators’ risk of morbidity and mortality. Additionally, this rapid loss of residual renal function could make early initiators more susceptible to deleterious electrolyte shifts during dialysis, which might explain our finding that risk of CV mortality (e.g., sudden cardiac death³⁶) was especially high among this population.

A minority of patients remained unmatched after our propensity analysis. They likely represent older US adults who do not have a choice regarding timing of dialysis initiation because of comorbid disease severity. Our study findings are, therefore, least applicable to this population. In contrast, our finding of deleterious outcomes associated with early dialysis initiation when we analyzed patients more likely to have had a choice (e.g., those patients with predialysis nephrology care and separately, those patients without AKI) may have important clinical implications. Our findings are most applicable to these patients who may benefit from improved efforts to engage in shared decision making about dialysis initiation,³⁷ because many older adults with kidney failure may choose to forgo dialysis altogether.³⁸

Our study has limitations. First, our approach did not address lead time bias (i.e., the erroneous observation that earlier dialysis improves survival, because patients beginning dialysis at a higher eGFR entered our analysis earlier than those patients beginning later). Survivor bias is also possible (i.e., early initiators may not have been able to survive to become later initiators). Future studies that anchor patient recruitment and analyses at a common eGFR well above the level that patients are typically initiated on dialysis are urgently needed. Second, we used eGFR to define timing of dialysis initiation, which is not the only measure used by nephrologists.^39–41 Also, creatinine-based measures of eGFR may misclassify patients with poor nutritional status and/or sarcopenia,^11,42 which is often seen among older dialysis patients.⁴³ Furthermore, creatinine-based measures of eGFR have not been well studied among Asian Americans,^44,45 which could have influenced our results. In an attempt to account for these limitations, we carefully matched patients on numerous predialysis factors, markers of nutritional status, and comorbidities. We also conducted a sensitivity analysis among patients presumed to be well nourished and found results consistent with our primary analysis, similar to a prior study.²¹ Third, although our analysis included propensity matching on 28 variables, residual confounding remains plausible in our study, such as in other observational studies. Fourth, despite our sensitivity analysis excluding patients with AKI, we lacked a precise measure of the circumstances surrounding dialysis initiation (i.e., elective versus urgent initiation⁴⁶). Fifth, we did not assess patient-reported outcomes such as quality of life, which is an area worthy of future study.⁴⁷ The limitations of our study are balanced by our inclusion of all older adult incident dialysis patients in the United States, ascertainment of predialysis health status using claims data, and cause-specific mortality assessment through a linkage to the National Death Index (NDI; the gold standard of US mortality databases⁴⁸).

In conclusion, early dialysis initiation among US older adults is common and associated with greater mortality and hospitalizations compared with later initiation. This association was present even after rigorous adjustment for patients’ predialysis health status and health care use, and it was consistent across multiple subgroups. These findings do not support the common practice of early dialysis initiation among older adults in the United States.

Concise Methods

Overview

As part of the Agency for Healthcare Research and Quality Developing Evidence to Inform Decisions about Effectiveness Network Patient Outcomes in ESRD Study,⁴⁹ we performed a national observational study to compare outcomes related to early versus later dialysis initiation among older adults in the United States.

Data Sources

We used the USRDS ESRD database, a national registry of patients receiving renal replacement therapy. It includes information from the ESRD Medical Evidence Form of the Centers for Medicare and Medicaid Services (CMS; form CMS-2728), which includes demographics, serum creatinine, cause of ESRD, and comorbidities. It also includes the ESRD Death Notification Form (CMS-2746), indicating date and cause of death, as well as Medicare Part A institutional claims and Medicare Part B physician/supplier claims.¹⁰ For patients with Medicare as their primary insurer at dialysis initiation, USRDS also maintains their Parts A and B claims from up to 3 years before dialysis initiation.¹⁰ In this study, we also linked USRDS data to NDI data from the Centers for Disease Control and Prevention.⁴⁸

Study Population

We included 84,654 US dialysis patients ages 67 years or older who initiated peritoneal dialysis or hemodialysis (in-center and home) between January 1, 2006 and December 31, 2008 and had both Medicare Parts A and B coverage as their primary insurer at least 2 years before dialysis initiation. We excluded patients without a serum creatinine recorded on form CMS-2728 or with renal function that recovered within 90 days after dialysis initiation. Follow-up was through December 31, 2008.

Assessment of Predialysis Health Status and Health Care Use

We used predialysis Medicare claims for up to 2 years before dialysis initiation and form CMS-2728 to determine factors related to dialysis timing. Factors considered included (1) diagnosis claims for AKI, uremia, or hyperkalemia in 6 months before dialysis initiation (Supplemental Appendix B), (2) number of hospitalizations for CHF and total all-cause hospital days in 6 months before dialysis initiation (Supplemental Appendix C), and (3) history of outpatient nephrology visits and codes for CKD stage 4 (ICD-9-CM code 585.4; as an indicator of longitudinal nephrology care) in 2 years preceding initiation. Patient comorbidities (CHF, atherosclerotic heart disease, cerebrovascular accident/transient ischemic attack, peripheral vascular disease, dysrhythmia, other cardiac disease, chronic obstructive pulmonary disease, diabetes, cancer, gastrointestinal bleeding, and liver disease) were defined based on form CMS-2728 and 2 years of predialysis claims using previously described methods (Supplemental Appendix B).⁵⁰ Nursing home residency status and presence of functional limitations (which include the inability to ambulate or transfer and/or a need for assistance with daily activities) were also ascertained from form CMS-2728.

Timing of Dialysis Initiation

We calculated each patient’s eGFR at dialysis initiation using serum creatinine from form CMS-2728 and the four-variable Modification of Diet in Renal Disease Study equation.⁵¹ We defined patients’ initiation as early if eGFR was ≥10 ml/min per 1.73 m² and later if eGFR was <10 ml/min per 1.73 m².

Outcomes

Our primary outcomes were all-cause mortality and all-cause hospitalization. Secondary outcomes were CV death, infectious death, CV hospitalizations, and infectious hospitalizations. We derived patients’ dates and causes of death from NDI data, which were available on 99.3% of deaths (death data from form CMS-2746 was used for the remainder). We used ICD-9-CM diagnosis codes from Medicare Part A claims to define cause-specific hospitalizations and International Classification of Diseases, Tenth Revision diagnosis codes to define NDI causes of death (Supplemental Appendix C).

For mortality analyses, patients were followed from dialysis initiation to death or censored at the time of kidney transplantation, loss to follow-up, renal function recovery, or end of study follow-up (December 31, 2008). To exclude hospitalizations directly related to inpatient dialysis initiation, patients were followed from day 8 after dialysis initiation for analyses of hospitalizations and additionally censored for death or end of Medicare as the primary insurance payer.

Statistical Analyses

Propensity Matching.

We anticipated that patient characteristics would vary between early and later initiation groups. Therefore, we matched the groups based on their propensity for early initiation calculated from a logistic regression model including 28 predialysis variables (demographics, renal complication history, comorbidities, predialysis AKI, hospitalizations, etc.) as predictors. We then used a greedy match algorithm⁵² to match an early initiation patient to a later initiation patient with a 1:1 ratio and an initial requirement of a propensity difference less than 0.00001. If no match was found, we then released the requirement of 0.00001–0.0001 and so on until the patient was matched or could not be matched with a requirement of 0.1. We compared patient characteristics before and after the match using chi-squared tests.

Outcome Analyses.

After establishing the propensity-matched cohort, we compared outcomes in the early versus later initiation groups using the Kaplan–Meier method and Cox proportional hazards regression models. Because patients’ characteristics were well balanced through propensity matching, we did not adjust these models.⁵³ Proportionality was checked using Schoenfeld residual plots.⁵⁴ For hospitalization analyses, which included repeated events, a Cox proportional hazards regression model with the Andersen–Gill approach⁵⁵ was used. In case the previous hospitalization changed the probability of the next hospitalization, as a sensitivity analysis, different baseline hazards were allowed for different hospitalizations (i.e., the 1st, the 2nd, etc.) up to the 10th hospitalization. For all analyses, robust standard error calculations were performed taking into account potential clustering by dialysis center.

We performed subgroup analyses with tests for interaction for the primary end points of all-cause mortality and hospitalization to examine the possibility of effect modification. Subgroups were defined by diabetes status, race (because of reports of survival differences by race^56,57), dialysis modality, hemodialysis access (catheter or AVF/AVG), nursing home residence, and functional limitation status. Because patient characteristics among the early and later groups were no longer balanced within some of the subgroups, the analyses were adjusted for all other characteristics.

Sensitivity Analyses

We performed sensitivity analyses to test our findings. First, we conducted an analysis without the inclusion of predialysis health status variables. Second, we restricted our analysis to hemodialysis patients and reperformed our propensity match. Third, we repeated our analysis including patients who recovered renal function within 90 days after dialysis initiation. Fourth, we performed two analyses, where we restricted our propensity-matched cohort to those patients presumed to be well nourished (either serum albumin≥3.5 mg/dl or BMI≥25 kg/m²). Fifth, we conducted an analysis restricted to propensity-matched patients without AKI in the 30 days leading up to dialysis initiation and without ATN as their cause of ESRD. Sixth, to confirm that our findings were consistent among those patients with adequate predialysis nephrology care for advanced CKD, we performed an analysis including patients who had (1) at least 12 months of outpatient nephrology care before initiation, (2) an ESRD cause other than ATN, and (3) an ICD-9-CM code of 585.4 (CKD stage 4) during the 2 years before initiation. We then conducted the propensity-matching procedure to establish a matched cohort in this subset and repeated our primary models.

Disclosures

None.