Non-Dialytic Care vs. Dialysis Transition on Hospitalization Outcomes in Veterans With Advanced Chronic Kidney Disease

Abstract

Objective:

To compare non-dialytic care vs. dialysis on hospitalization outcomes in advanced CKD, given that dialysis, as the predominant treatment paradigm for advanced CKD patients transitioning to end-stage kidney disease, may be associated with high early mortality, healthcare utilization, and withdrawal.

Patients and Methods:

We examined a national cohort of Veterans with advanced CKD (≥2 estimated glomerular filtration rates [eGFRs] <25ml/min/1.73m² separated by ≥90 days from 10/1/2010–9/30/2019) categorized according to receipt of dialysis vs. non-dialytic care, defined as those who did vs. did not receive dialysis within two-years of their index eGFR. Propensity score (PS)-matching was used to account for differences in baseline characteristics across exposure groups. We examined associations of non-dialytic care vs. dialysis with hospitalization rates and rates of days hospitalized (separately) using Poisson regression. Among hospitalized patients, we examined associations with lengths of stay (LOS) using linear mixed effects models.

Results:

In the 1:1 PS-matched cohort of 13,020 non-dialytic and 13,020 dialysis patients, dialysis transition was associated with higher rates of hospitalization and days hospitalized (ref: non-dialytic care): IRRs (95%CIs) 1.64 (1.61, 1.66) and 1.76 (1.73, 1.78], respectively. Dialysis was also associated with longer average LOS vs. non-dialytic care: 8.6 vs. 8.1 days, respectively (P<.001). In secondary analyses of timing of dialysis transition, incrementally earlier dialysis transitions were associated with increasingly higher hospitalization rates vs. non-dialytic care, with similar patterns observed for rates of days hospitalized and LOS.

Conclusion:

Compared with non-dialytic care, dialysis was associated with higher rates of hospitalization and days hospitalized and longer LOS.

Introduction

In the United States (US), each year ~125,000 patients with advanced chronic kidney disease (CKD) transition to dialysis as the dominant treatment paradigm for end-stage kidney disease (ESKD), particularly among those ineligible for kidney transplantation.¹ While the 1972 Medicare End-Stage Renal Disease (ESRD) program enabled near-universal access to dialysis for patients progressing to ESKD,² population-based studies show that advanced CKD patients transitioning to dialysis have high mortality particularly in the first year of treatment.^3–5 Evidence also suggests that among elderly and comorbid patients, two of the fastest growing ESKD subgroups,^6–10 transition to dialysis is associated with frequent hospitalizations,^11–13 poor health-related quality of life (HRQOL),¹⁴ loss of independence,¹⁵ functional decline,¹⁶ and high withdrawal rates.¹⁷ These observations have led to pressing urgency to examine conservative non-dialytic care as an alternative treatment strategy for advanced CKD.^18,19

While national professional societies have issued guidance on conservative non-dialytic care as a treatment option for select populations,^20,21 this management strategy has remained under-utilized. This is in part due to need for stronger evidence to inform the development and implementation of non-dialytic care models. Although there are data showing that dialysis may not improve survival^22–24 and patient-centered outcomes¹⁶ in certain populations, there is sparse data specifically examining the comparative effectiveness of non-dialytic care vs. dialysis transition. Existing studies are limited by small sample size, restriction to single centers, focus on non-US cohorts that are not generalizable to US CKD patients,^37–47 lack of a dialysis comparator group, and presence of length- and lead-time bias from non-consideration of differences in CKD progression and estimated glomerular filtration rate (eGFR) at dialysis transition.⁴⁸

To address these knowledge gaps, we conducted a retrospective study examining the comparative effectiveness of non-dialytic care vs. dialysis transition on hospitalization outcomes in US Veterans. We additionally sought to evaluate whether timing of dialysis transition may have a bearing on these comparative outcomes using national Veterans Affairs (VA) data.

Methods

Source Population

We conducted an observational cohort study among Veterans as part of the National Institutes of Health (NIH)-funded “Defining Optimal Transitions of Care in Advanced Kidney Disease: Non-dialytic care vs. Dialysis Approaches” (OPTIMAL) study. Our source population was derived from the national VA database linked to US Renal Data System (USRDS) and Center for Medicare and Medicaid Services (CMS) data. In this study, we focused on Veterans with incident advanced CKD, who were comprised of CKD patients with eGFRs <25ml/min/1.73m², a threshold used in CKD trials (Modification of Diet in Renal Disease²⁵ and Replicating Evidence of Preserved Renal Function: An Investigation of Tolvaptan Safety and Efficacy in ADPKD²⁶ studies) and in the NIH U01 “Transitions of Care in CKD” study.^7–10 Patients were included provided that they (1) received care within the VA healthcare system, (2) had at least two or more eGFR values <25 ml/min/1.73m² separated by ≥90 days over 10/1/2010–9/30/2019 (in which the first eGFR <25 ml/min/1.73m² was defined as the index eGFR), and (3) had an index eGFR value from 2010–2017 (to allow for an adequate follow-up period through the administrative censoring date) (Supplemental Figure 1). Patients were excluded if they (1) were missing core demographic data (date of birth, sex, race), (2) had an intervening eGFR value ≥25 ml/min/1.73m² between their first (index) and second eGFR values <25 ml/min/1.73², (3) received dialysis on or before their index eGFR date, (4) received a kidney transplant on or before their second eGFR <25 ml/min/1.73m², or (5) had non-sensical follow-up data. The study was approved by the Institutional Review Boards of the VA Long Beach Healthcare System.

Exposure Ascertainment

The objective of the study was to compare hospitalization outcomes among Veterans with incident advanced CKD who underwent non-dialytic care vs. dialysis transition. In primary analyses, patients were categorized according to receipt of non-dialytic care, defined as those who did not receive dialysis within two-years of the index eGFR, vs. dialysis transition, defined as those with a diagnostic/procedural dialysis code or presence of a first service date for dialysis using linked VA-USRDS data within two-years of the index eGFR date.

In secondary analyses, we compared non-dialytic care to timing of dialysis transition based on the most proximate eGFR 30-days prior to this date. We first categorized patients according to receipt of non-dialytic care vs. dialysis, with the latter group parsed into earlier vs. intermediate+later dialysis transition according to initiation of dialysis with an eGFR ≥15 vs. <15 ml/min/1.73m². We also compared non-dialytic care with finer gradations of timing of dialysis transition, categorized as earlier vs. intermediate vs. later transition (eGFRs ≥15 vs. 10-<15 vs. <10 ml/min/1.73², respectively). In the event of missing eGFR data within 30-days of dialysis transition, we imputed eGFR values using eGFR slopes generated by linear mixed models of longitudinal eGFR values within 180-days of dialysis transition.

Outcome Ascertainment

Our primary outcome of interest was the rate of hospitalizations within two-years of the index eGFR date. In secondary analyses, we examined the rate of days hospitalized within two-years of the index eGFR date (cumulative days hospitalized over a two-year period). In a subcohort of patients who were hospitalized within two-years of the index eGFR, we examined the average length of stay (LOS) per hospitalization. All-cause hospitalization (hospitalizations irrespective of cause) data was obtained from VA data (Medical SAS and Corporate Data Warehouse) and CMS records. Patients were followed for the outcomes of interest following the day after study entry (index eGFR date) until occurrence of a censoring event (death, kidney transplant, loss to follow-up, or last available date in the cohort), end of the study period (9/30/2019), or after two years following the index eGFR date, whichever occurred first. Information on censoring events and their associated dates were obtained from VA database, as well as USRDS and CMS data linked to national VA data.

Socio-demographic, Comorbidity, and Other Treatment Related Information

Data from the VA, USRDS, and CMS were used to determine sociodemographic information. Marital status was ascertained from VA data. Information regarding comorbidities were extracted from the VA and CMS datasets using International Classification of Diseases, Ninth and Tenth Revision, Clinical Modification (ICD-9/10) diagnostic and procedure codes and Current Procedural Terminology (CPT) codes. Charlson Comorbidity Index (CCI) scores were estimated using the Deyo modification for administrative datasets without including kidney disease. The VA Frailty Index, an electronic frailty index used to measure frailty in Veterans that has been associated with clinical outcomes, was calculated using ICD-9/10 diagnosis and procedure codes from the VA database. Body mass index (BMI) data were obtained from the VA Vital Status file. Medication data were ascertained from VA and CMS data. Laboratory data except serum creatinine were obtained from the VA Decision Support System National Data Extracts Laboratory Results. Patients’ residential status in metropolitan vs. nonmetropolitan areas based on Rural-Urban Continuum Codes was ascertained from VA, USRDS, and CMS zip code data that were converted to Federal Information Processing Series codes. VA and CMS data were used to determine if patients received nephrology care.

Statistical Analyses

We accounted for differences in baseline characteristics across exposure groups using propensity score (PS) matching in a 1:1 ratio (1:1 match between non-dialytic care and dialysis transition groups). In sensitivity analyses, we also conducted PS matching in a 2:1 ratio of non-dialytic care to dialysis transition patients. The PS matching caliper was 0.2 times the standard deviation of the PS in the dialysis group. The caliper is a set threshold for matching the PS values between non-dialytic care and dialysis transition patient (PS for a non-dialytic care patient to a dialysis transition patient), in which a smaller caliper threshold represents a more stringent matching criterion. Adequacy of matching was assessed by checking the standardized difference for each covariate between exposure groups in the PS-matched cohort. PS’s were estimated using logistic regression models that included the following baseline covariates:

1. Demographics: Age, sex, race, ethnicity, marital status, BMI (on or up to one-year prior to the index eGFR date), and residential status in a metropolitan/nonmetropolitan area;

2. Comorbidity status: CCI score, VA frailty index score, and presence vs. absence of the following comorbidities: prior Alzheimer’s Disease and related disorders, atrial fibrillation, CKD, chronic obstructive pulmonary disease and bronchiectasis, heart failure, diabetes, ischemic heart disease, depression, stroke, colorectal cancer, prostate cancer, lung cancer, anemia, asthma, hyperlipidemia, or hypertension;

3. Medications: use or non-use of aspirin, erythropoietin, opioid analgesics, non-opioid analgesics, nonsteroidal anti-inflammatory drugs, antidepressants, beta-blockers, alpha-blockers, calcium channel blockers, antianginals, antiarrhythmics, statins, non-statin lipid-lowering drugs, vasodilator antihypertensives, thiazides, loop diuretics, potassium-sparing diuretics, renin-angiotensin-aldosterone system inhibitors, bicarbonate, insulin, oral hypoglycemics, thyroid supplements, iron, centrally acting alpha agonists, metformin, sodium-glucose cotransporter-2 (SGLT2) inhibitors, dipeptidyl peptidase-4 (DPP-4) inhibitors, glucagon-like peptide-1 (GLP-1) receptor agonists, niacin, fibrates, and mineralocorticoid receptor antagonists;

4. Laboratory test values (on or up to one-year prior to the index eGFR date): serum albumin, hemoglobin, sodium, potassium, calcium, phosphate, bicarbonate, total cholesterol, triglycerides, glucose, and eGFR slope;

5. Other treatment factors (on or up to one-year prior to the index eGFR date): prior hospitalization and nephrology care.

There were no missing values for age, sex, race, comorbidities, medications, and treatment factors. The remaining covariates had ≤11% missing data, which included ethnicity (<1%), marital status (<1%), BMI (2%), metropolitan/nonmetropolitan (<1%), serum albumin (8%), hemoglobin (7%), sodium (<1%), potassium (<1%), calcium (4%), phosphorus (1%), bicarbonate (2%), total cholesterol (11%), triglycerides (11%), and glucose (<1%). The main analyses utilized a complete case analysis approach. In sensitivity analyses, missing data were addressed using multiple imputation with 10 imputed datasets.

Poisson regression models were used to estimate incident rate ratios (IRRs) for rates of hospitalization and rates of days hospitalized, and linear mixed effects models were used to estimate hospitalization LOS using the PS-matched cohorts. Sensitivity analyses were conducted in which models were doubly-adjusted for the PS covariates. Statistical analyses and graphical results were conducted using SAS Enterprise Guide version 8.2 (SAS Institute Inc., Cary, NC, USA), Stata version 17.0 (Stata Corp., College Station, TX), and SigmaPlot version 13 (Systat Software, San Jose, CA).

Results

Study Population

There were 91,598 patients who met eligibility criteria, among whom 66,189 received non-dialytic care and 25,409 received dialysis (Exposure Group 0; Supplemental Figure 1). In analyses of 1:1 matched sets examining rates of hospitalization (primary outcome) and rates of days hospitalized (secondary outcome) using complete case analysis, the PS-matched cohort was comprised 13,020 pairs of individuals who underwent non-dialytic care or dialysis (total N=26,040 patients). In contrast to the unmatched cohort, there were no meaningful differences between the exposure groups across any of the baseline characteristics in the matched cohort (Table 1). Among the dialysis group, 51.4% vs. 48.6% of patients transitioned to dialysis in the inpatient vs. outpatient settings, respectively (Supplemental Table 1), and 92.1% vs. 7.6% vs. 0.3% initiated hemodialysis, peritoneal dialysis, vs. had an unknown modality type, respectively.

Table 1.

Baseline characteristics of the study cohort examining non-dialytic care vs. dialysis and rates of hospitalization and rates of days hospitalized using complete case analysis (1:1 matched sets): Non-dialytic care vs. Dialysis Transition (Exposure Group 0).

	Before PS-Matching (n=68,336)				After PS-Matching (n=26,040)
	Non-dialytic care (n=48,925)	Dialysis(n=19,411)			Non-dialytic care (n=13,020)	Dialysis (n=13,020)
	Mean (SD)/No. (%)	Mean (SD)/No. (%)	Standard Difference	P-value	Mean (SD)/No. (%)	Mean (SD)/No. (%)	Standard Difference	P-value
Age, years	75.1 (10.9)	66.2 (10.3)	−0.85	<.001	68.0 (10.2)	68.5 (9.9)	0.05	<.001
Male	47,695 (97)	18,946 (98)	0.008	.37	12,737 (98)	12,724 (98)	−0.007	.58
Female	1,230 (3)	465 (2)			283 (2)	296 (2)
White	34,107 (70)	10,263 (53)	0.39	<.001	7,692 (59)	7,376 (57)	0.11	<.001
Black	9,488 (19)	6,505 (34)			3,552 (27)	3,991 (31)
Asian/Pacific islander	2,339 (5)	1,500 (8)			924 (7)	901 (7)
American Indian/missing/unknown	2,991 (6)	1,143 (6)			852 (7)	752 (6)
Non-Hispanic	46,061 (94)	17,762 (92)	0.10	<.001	12,042 (92)	12,030 (92)	0.004	.78
Hispanic	2,864 (6)	1,649 (9)			978 (8)	990 (8)
Single/divorced/widowed/unknown or missing	21,513 (44)	9,000 (46)	−0.05	<.001	5,766 (44)	5,879 (45)	−0.02	.16
Married	27,412 (56)	10,411 (54)			7,254 (56)	7,141 (55)
Diabetes	32,974 (67)	15,857 (82)	0.33	<.001	10,513 (81)	10,337 (79)	−0.03	.006
Body Mass Index, kg/m2	29.8 (6.4)	31.2 (6.8)	0.20	<.001	31.3 (7.1)	31.0 (6.7)	−0.04	<.001
Charlson Comorbidity Index	7.2 (3.0)	7.8 (3.1)	0.20	<.001	7.7 (3.0)	7.7 (3.1)	−0.01	.23
Index eGFR, mL/min/1.73m2	21.2 (3.4)	18.8 (5.0)	−0.58	<.001	19.7 (4.3)	19.7 (4.4)	−0.003	.83
Serum albumin, g/dl	3.7 (0.5)	3.3 (0.6)	−0.62	<.001	3.5 (0.6)	3.5 (0.5)	0.003	.81
Serum hemoglobin, g/dl	11.9 (1.8)	11.3 (1.8)	−0.33	<.001	11.4 (1.8)	11.5 (1.7)	0.009	.48
Serum sodium, meq/l	139.5 (3.2)	139.3 (3.4)	−0.07	<.001	139.3 (3.4)	139.3 (3.4)	0.004	.77
Serum potassium, meq/l	4.6 (0.6)	4.5 (0.7)	−0.07	<.001	4.5 (0.6)	4.5 (0.6)	0.004	.78
Serum calcium, mg/dl	9.1 (0.6)	8.8 (0.6)	−0.51	<.001	8.9 (0.6)	8.9 (0.6)	0.003	.82
Serum phosphorus, mg/dl	4.4 (0.7)	4.5 (0.7)	0.09	<.001	4.5 (0.7)	4.5 (0.7)	0.004	.74
Serum bicarbonate, meq/l	24.5 (3.7)	23.6 (3.9)	−0.26	<.001	23.9 (3.9)	23.9 (3.8)	0.007	.59
Serum total cholesterol, mg/dl	158.7 (43.9)	170.5 (54.5)	0.24	<.001	165.3 (50.6)	165.7 (50.9)	0.007	.57
Serum total triglyceride, mg/dl	159.6 (114.2)	176.7 (133.1)	0.14	<.001	174.3 (134.5)	173.2 (130.8)	−0.009	.48
Serum glucose, mg/dl	132.2 (59.7)	142.4 (69.0)	0.16	<.001	142.3 (67.6)	140.7 (67.6)	−0.02	.05
Veterans Affairs Frailty Index	0.4 (0.1)	0.4 (0.1)	0.1219	<.001	0.4 (0.1)	0.4 (0.1)	−0.01	.37
eGFR slope, per year	−8.3 (5.4)	−14.3 (7.1)	−0.95	<.001	−12.0 (8.0)	−12.5 (5.5)	−0.07	<.001
Alzheimer’s disease	8,817 (18)	2,622 (14)	−0.12	<.001	1,803 (14)	1,873 (14)	0.02	.21
Atrial fibrillation	16,017 (33)	5,849 (30)	−0.06	<.001	4,111 (32)	4,023 (31)	−0.01	.24
CKD	47,446 (97)	19,009 (98)	0.06	<.001	12,730 (98)	12,714 (98)	−0.008	.51
COPD	22,367 (46)	9,095 (47)	0.02	.007	6,148 (47)	6,125 (47)	−0.004	.78
Heart failure	27,503 (56)	12,383 (64)	0.16	<.001	8,190 (63)	8,063 (62)	−0.02	.10
Ischemic heart disease	33,293 (68)	13,447 (69)	0.03	.002	9,105 (70)	8,987 (69)	−0.02	.11
Depression	17,047 (35)	8,859 (46)	0.22	<.001	5,750 (44)	5,586 (43)	−0.03	.04
Stroke	11,920 (24)	4,584 (24)	−0.02	.04	3,083 (24)	3,079 (24)	−0.0007	.95
Colorectal cancer	2,596 (5)	744 (4)	−0.07	<.001	537 (4)	560 (4)	0.009	.48
Prostate cancer	8,033 (16)	2,104 (11)	−0.16	<.001	1,606 (12)	1,629 (13)	0.005	.67
Lung cancer	1,530 (3)	569 (3)	−0.01	.18	412 (3)	410 (3)	−0.0009	.94
Anemia	36,683 (75)	15,826 (82)	0.16	<.001	10,442 (80)	10,369 (80)	−0.01	.26
Asthma	5,376 (11)	2,136 (11)	<.001	.95	1,443 (11)	1,444 (11)	0.0002	.98
Hyperlipidemia	44,728 (91)	17,816 (92)	0.01	.13	11,992 (92)	11,932 (92)	−0.02	.17
Hypertension	48,195 (99)	19,219 (99)	0.05	<.001	12,869 (99)	12,874 (99)	0.004	.77
Aspirin use	24,244 (50)	11,126 (57)	0.16	<.001	7,375 (57)	7,250 (56)	−0.02	.12
Erythropoietin use	4,622 (9)	2,640 (14)	0.13	<.001	1,668 (13)	1,579 (12)	−0.02	.10
Opioid use	34,539 (71)	14,674 (76)	0.11	<.001	9,871 (76)	9,755 (75)	−0.02	.10
Non-opioid use	38,788 (79)	16,237 (84)	0.11	<.001	10,843 (83)	10,804 (83)	−0.008	.52
NSAID use	33,813 (69)	14,483 (75)	0.12	<.001	9,704 (75)	9,627 (74)	−0.01	.28
Antidepressant use	21,594 (44)	10,206 (53)	0.17	<.001	6,730 (52)	6,636 (51)	−0.01	.24
Beta-blocker use	38,421 (79)	15,960 (82)	0.09	<.001	10,706 (82)	10,633 (82)	−0.01	.24
Alpha-blocker use	20,328 (42)	7,385 (38)	−0.07	<.001	5,298 (41)	5,149 (40)	−0.02	.06
Calcium channel blocker use	37,082 (76)	16,251 (84)	0.20	<.001	10,868 (83)	10,724 (82)	−0.03	.02
Antianginal use	17,296 (35)	6,833 (35)	−0.003	.71	4,816 (37)	4,724 (36)	−0.01	.24
Anti-arrhythmia use	4,531 (9)	1,214 (6)	−0.11	<.001	909 (7)	957 (7)	0.01	.25
Statin use	41,147 (84)	16,397 (84)	0.01	.23	11,081 (85)	11,029 (85)	−0.01	.37
Non-statin use	15,256 (31)	6,987 (36)	0.10	<.001	4,730 (36)	4,625 (36)	−0.02	.18
Vasodilator use	14,949 (31)	8,288 (43)	0.25	<.001	5,454 (42)	5,233 (40)	−0.03	.005
Thiazide use	29,296 (60)	12,999 (67)	0.15	<.001	8,718 (67)	8,550 (66)	−0.03	.03
Loop diuretic use	31,366 (64)	13,568 (70)	0.12	<.001	9,135 (70)	8,997 (69)	−0.02	.06
Potassium-sparing diuretic use	9,116 (19)	4,286 (22)	0.09	<.001	2,974 (23)	2,852 (22)	−0.02	.07
RAAS inhibitor use	41,660 (85)	17,317 (89)	0.12	<.001	11,589 (89)	11,513 (88)	−0.02	.14
Bicarbonate use	4,222 (9)	2,215 (11)	0.09	<.001	1,464 (11)	1,427 (11)	−0.009	.47
Insulin use	20,976 (43)	11,794 (61)	0.36	<.001	7,723 (59)	7,490 (58)	−0.04	.003
Oral hypoglycemics use	22,966 (47)	11,627 (60)	0.26	<.001	7,703 (59)	7,540 (58)	−0.03	.04
Thyroid supplement use	8,814 (18)	2,567 (13)	−0.13	<.001	1,892 (15)	1,849 (14)	−0.009	.45
Iron use	16,463 (34)	7,325 (38)	0.09	<.001	5,014 (39)	4,854 (37)	−0.03	.04
Alpha-agonist use	10,901 (22)	6,206 (32)	0.22	<.001	3,960 (30)	3,843 (30)	−0.02	.11
Metformin use	12,893 (26)	7,817 (40)	0.30	<.001	5,036 (39)	4,867 (37)	−0.03	.03
SGLT2 inhibitor use	13 (0.03)	7 (0.04)	0.005	.51	8 (0.06)	5 (0.04)	−0.01	.41
DPP4 inhibitor use	1,835 (4)	687 (4)	−0.01	.19	498 (4)	499 (4)	0.0004	.97
GLP1 receptor agonist use	399 (1)	246 (1)	0.04	<.001	167 (1)	163 (1)	−0.003	.82
Niacin use	6,739 (14)	3,065 (16)	0.06	<.001	2,117 (16)	2,026 (16)	−0.02	.12
Fibrate use	7,900 (16)	3,863 (20)	0.10	<.001	2,582 (20)	2,541 (20)	−0.008	.52
MRA use	6,725 (14)	3,373 (17)	0.10	<.001	2,344 (18)	2,212 (17)	−0.03	.03
Hospitalization prior to index eGFR	46,734 (96)	19,195 (99)	0.21	<.001	12,840 (99)	12,820 (98)	−0.01	.30
Nephrology care	43,420 (89)	18,658 (96)	0.28	<.001	12,423 (95)	12,363 (95)	−0.02	.08
Metropolitan residence	38,366 (78)	15,649 (81)	0.05	<.001	10,439 (80)	10,449 (80)	0.002	.88

Abbreviations: CKD, chronic kidney disease; COPD, Chronic obstructive pulmonary disease; DPP4, Dipeptidyl peptidase 4; eGFR, GLP1, Glucagon-like peptide-1; estimated glomerular filtration ratio; MRA, mineralocorticoid receptor antagonists; NSAID, Nonsteroidal anti-inflammatory drugs; RAAS, Renin-Angiotensin-Aldosterone System; SGLT2, Sodium-Glucose Transport Protein 2.

In the unmatched dialysis group, 4590 vs. 18,482 patients underwent an earlier vs. intermediate+later dialysis transition (Exposure Group 1; Supplemental Figure 1). In granular analyses of timing of dialysis transition, there were 4590, 7216, vs. 11,266 patients who underwent an earlier, intermediate, vs. very late dialysis transition (Exposure Group 2). In the dialysis group, there were 2337 without eGFR data on or within 180-days prior to dialysis transition for whom dialysis transition timing could not be ascertained. In PS-matched cohorts of these dialysis transition and non-dialytic care groups, there were no meaningful differences in the baseline characteristics across exposure categories (Supplemental Tables 2–5).

In the subcohort of patients among whom hospitalization LOS (tertiary outcome) were examined, the 1:1 PS-matched cohort was comprised of 10,108 pairs of individuals who underwent non-dialytic care or dialysis (total N=20,216 patients) (Supplemental Figure 2). Similarly, there were no meaningful differences between the exposure groups across any baseline characteristics (Supplemental Table 6). Balanced characteristics were also observed in secondary analyses of granular dialysis transition groups (Exposure Groups 1 and 2; Supplemental Tables 7–10) and in sensitivity analyses that utilized 2:1 matching (2:1 ratios of non-dialytic care to dialysis patients) or that used multiple imputation (data not shown).

Non-Dialytic care vs. Dialysis Transition and Rates of Hospitalization

In the 1:1 PS-matched cohort using complete case analysis, patients contributed a total of 47,239 person-years of follow-up during which there were 67,432 hospitalization events (crude rates of hospitalization and death over the follow-up period shown in Table 2). Crude rates of hospitalization were 1427.5, 1073.9, and 1758.6 hospitalizations per 1000 person-years of follow-up for the overall, non-dialytic care, and dialysis groups, respectively. Median (IQR) at-risk time was 2.0 (2.0, 2.0) years.

Table 2.

Crude rates of hospitalization and death over a two-year follow-up period.

Exposure Categorization	Exposure Group*	No. of patients	No. of hospitalizations	Total follow-up time, years	Crude hospitalization rate, per 1000-person years	No. of deaths	Crude mortality rate, per 1000-person year
Exposure 0
	Non-dialytic care (ref.)	13,020	24,530	22,843	1073.9	3316	145.2
	Dialysis	13,020	42,902	24,396	1758.6	2417	99.1
Exposure 1
	Non-dialytic care (ref.)	3,628	7,716	6,273	1229.9	1044	166.5
	Earlier dialysis	3,628	14,357	6,556	2190.0	1007	153.7
	Non-dialytic care (ref.)	9,833	18,957	17,190	1102.8	2504	145.7
	Intermediate+ Later dialysis	9,833	32,565	18,528	1757.6	1597	86.2
Exposure 2
	Non-dialytic care (ref.)	3,628	7,716	6,273	1229.9	1044	166.5
	Earlier dialysis	3,628	14,357	6,556	2190.0	1007	153.7
	Non-dialytic care (ref.)	5,443	11,347	9,404	1206.7	1467	156.1
	Intermediate dialysis	5,443	19,470	10,159	1916.6	1014	99.8
	Non-dialytic care (ref.)	6,473	12,552	11,294	1111.4	1627	144.1
	Later dialysis	6,473	21,252	12,299	1728.0	890	72.4

*Exposure 1: Earlier dialysis (defined as dialysis transition at eGFR ≥15 ml/min/1.73m²); Intermediate+later dialysis (defined as dialysis transition at eGFR <15 ml/min/1.73m²).

*Exposure 2: Earlier dialysis (defined as dialysis transition at eGFR ≥15 ml/min/1.73m²); Intermediate dialysis (defined as dialysis transition at 10–<15 ml/min/1.73m²); Later dialysis (defined as dialysis transition at eGFR <10 ml/min/1.73²).

In primary analyses of the 1:1 PS-matched cohort using complete case analysis, compared to non-dialytic care, dialysis was associated with higher hospitalization rates in unadjusted and doubly-adjusted Poisson models: IRRs (95%CIs) 1.64 (1.61, 1.66) and 1.61 (1.59, 1.64), respectively (Figure 1A and Supplemental Table 11). Similar findings were observed in the 1:1 matched cohort using multiple imputation (Supplemental Figure 3A and Supplemental Table 11) and the 2:1 matched cohorts using complete case analysis (Supplemental Figure 4A and Supplemental Table 12) and multiple imputation (Supplemental Figure 5A and Supplemental Table 12).

Figure 1. Association between non-dialytic care vs. dialysis transition and rates of hospitalization estimated by Poisson regression using complete case analysis (1:1 matched sets).

Abbreviations: ED, earlier dialysis; LD, later dialysis; ID, intermediate dialysis; VLD, very late dialysis.

In secondary analyses that parsed the dialysis group according to timing of dialysis transition, in the 1:1 PS-matched cohort using complete case analysis, the earlier dialysis group had higher rates of hospitalization in unadjusted and doubly-adjusted Poisson analyses (ref: non-dialytic care): IRRs (95%CIs) 1.78 (1.73, 1.83) and 1.75 (1.70, 1.80), respectively (Figure 1B and Supplemental Table 11). Similarly, the intermediate+later dialysis group had higher rates of hospitalization than non-dialytic care in unadjusted and doubly adjusted analyses, albeit with slightly weaker point estimates: IRRs (95%CIs) 1.59 (1.57, 1.62) and 1.56 (1.54, 1.59), respectively. Similar findings were observed in the 1:1 matched cohort using multiple imputation (Supplemental Figure 3B and Supplemental Table 11) and the 2:1 matched cohorts using complete case analysis (Supplemental Figure 4B and Supplemental Table 12) and multiple imputation (Supplemental Figure 5B and Supplemental Table 12).

In analyses examining granular categories of timing of dialysis transition, in the 1:1 PS-matched cohort using complete case analysis, compared to non-dialytic care incrementally earlier dialysis transitions were associated with increasingly higher rates of hospitalization in unadjusted Poisson analyses: IRRs (95%CIs) 1.78 (1.73, 1.83), 1.59 (1.55, 1.63), and 1.56 (1.52, 1.59) for earlier, intermediate, and later dialysis transition, respectively (Figure 1C and Supplemental Table 11). Similar findings were observed in the doubly-adjusted analyses (Figure 1C and Supplemental Table 11), the 1:1 matched cohort using multiple imputation (Supplemental Figure 3C and Supplemental Table 11), and the 2:1 matched cohorts using complete case analysis (Supplemental Figure 4C and Supplemental Table 12) and multiple imputation (Supplemental Figure 5C and Supplemental Table 12).

Non-Dialytic Care vs. Dialysis Transition and Rates of Days Hospitalized

In the 1:1 PS-matched cohort using complete case analysis, there were a total of 556,561 days of hospitalization (Supplemental Table 13). Crude rates of days hospitalized were 11.8, 8.6, and 14.7 days hospitalized per 1000 person-years of follow-up for the overall, non-dialytic care, and dialysis cohorts, respectively.

In primary analyses of the 1:1 PS-matched cohort using complete case analysis, compared to non-dialytic care, dialysis was associated with higher rates of days hospitalized in unadjusted and doubly-adjusted Poisson models: IRRs (95%CIs) 1.71 (1.70, 1.72) and 1.67 (1.66, 1.68), respectively (Figure 2A and Supplemental Table 14). Similar findings were observed in the 1:1 matched cohort using multiple imputation (Supplemental Figure 6A and Supplemental Table 14) and the 2:1 matched cohorts using complete case analysis (Supplemental Figure 7A and Supplemental Table 15) and multiple imputation (Supplemental Figure 8A and Supplemental Table 15).

Figure 2. Association between non-dialytic care vs. dialysis transition and rates of days hospitalized estimated by Poisson regression using complete case analysis (1:1 matched sets).

Abbreviations: ED, earlier dialysis; LD, later dialysis; ID, intermediate dialysis; VLD, very late dialysis.

In secondary analyses in which the dialysis group was parsed by timing of dialysis transition, in the 1:1 PS-matched cohort using complete case analysis, earlier dialysis was associated with higher rates of days hospitalized in unadjusted and doubly adjusted Poisson regression analyses (ref: non-dialytic care): IRRs (95%CIs) 2.12 (2.10, 2.14) and 2.08 (2.06, 2.10), respectively (Figure 2B and Supplemental Table 14). Similarly, intermediate+later dialysis was associated with higher rates of hospitalization vs. non-dialytic care in unadjusted and doubly adjusted analyses, albeit with slightly weaker point estimates: IRRs (95%CIs) 1.54 (1.53, 1.55) and 1.50 (1.49, 1.51), respectively. Similar findings were observed in the 1:1 matched cohort using multiple imputation (Supplemental Figure 6B and Supplemental Table 14) and the 2:1 matched cohorts using complete case analysis (Supplemental Figure 7B and Supplemental Table 15) and multiple imputation (Supplemental Figure 8B and Supplemental Table 15).

In analyses examining finer categories of dialysis transition timing, in the 1:1 PS-matched cohort using complete case analysis, compared to non-dialytic care incrementally earlier dialysis transitions were associated with increasingly higher rates of days hospitalized in unadjusted Poisson analyses: IRRs (95%CIs) 2.12 (2.10, 2.14), 1.58 (1.57, 1.59), and 1.45 (1.44, 1.46) for earlier, intermediate, and later dialysis transition, respectively (Figure 2C and Supplemental Table 14). Similar findings were observed in the doubly-adjusted analyses (Figure 2C and Supplemental Table 14), the 1:1 matched cohort using multiple imputation (Supplemental Figure 6C and Supplemental Table 14), and the 2:1 matched cohorts using complete case analysis (Supplemental Figure 7C and Supplemental Table 15) and multiple imputation (Supplemental Figure 8C and Supplemental Table 15).

Non-Dialytic Care vs. Dialysis Transition and Hospitalization LOS

In a subcohort of patients hospitalized within two-years of the index eGFR, those in the dialysis transition group vs. the non-dialytic care group had longer LOS in the 1:1 matched complete case cohort: 8.6 vs. 8.1 days, respectively; LOS difference (Δ) +0.5 days (Figure 3A and Supplemental Table 16). In secondary analyses, compared with non-dialytic care, patients with earlier dialysis transition had a longer LOS (Δ+1.9 days) whereas intermediate+later dialysis transition had a similar LOS (Δ+0.1 days) (Figure 3B and Supplemental Table 16). In analyses of finer gradations of dialysis transition, earlier dialysis also had a longer LOS (Δ+1.9 days) whereas intermediate dialysis had similar LOS (Δ+0.2 days) and later dialysis had slightly shorter LOS (Δ−0.4 days) (Figure 3C and Supplemental Table 16). The overall pattern of findings was similar in the 1:1 matched cohort using multiple imputation (Supplemental Table 16) and the 2:1 matched cohorts using complete case analysis (Supplemental Figure 9A–C and Supplemental Table 17) and multiple imputation (Supplemental Table 17).

Figure 3.

Association of non-dialytic care vs. dialysis transition with hospitalization length of stay estimated by linear mixed effects models using complete case analysis (1:1 matched sets).

Discussion

In a national cohort of Veterans with advanced CKD, patients who transitioned to dialysis had higher rates of hospitalization and rates of days hospitalized vs. those who received non-dialytic care. In analyses that granularly examined timing of dialysis transition, incrementally earlier transitions to dialysis were also associated with increasingly higher rates of hospitalization and rates of days hospitalized vs. non-dialytic care. Among hospitalized patients, the dialysis transition group also had longer hospitalization LOS compared to non-dialytic care. These findings were robust to multiple secondary and sensitivity analyses that doubly-adjusted for PS covariates, used complete case analysis or multiple imputation to address missing data, and used variable matching ratios.

Population-based data show that Veterans transitioning to dialysis experience high mortality rates in their first few months of treatment.^7–10 These observations mimic patterns observed in the broader ESKD population.^7–10 While a number of studies have sought to elucidate the determinants contributing to the high mortality of this fragile transition period,^27–56 there has been growing interest in the role of conservative dialysis-free management as a viable treatment option for advanced CKD patients. However, there have remained major knowledge gaps regarding the trajectory and outcomes of advanced CKD patients who do not undergo dialysis transition. This includes the unclear comparative effectiveness of non-dialytic care vs. dialysis on hospitalization outcomes including admission rates and LOS, which have important implications on functional decline, subsequent institutionalization, and healthcare utilization.

To date, few studies have compared the impact of non-dialytic care vs. dialysis on hospitalization outcomes, which have been mixed. In a single-center study of 202 elderly patients with ESKD from the United Kingdom, compared to conservative care, those who opted for kidney replacement therapy had higher rates of hospitalization and inpatient deaths, with a large proportion of their surviving days spent in the hospital setting.⁵⁷ In a community-based cohort of 262 elderly patients from the Netherlands, those who chose dialysis had higher rates of hospitalization and in-hospital days per year compared to those receiving non-dialytic care, as well as greater intensive hospital utilization near the end of life.⁵⁸ In another study of 968 elderly patients with stage 5 CKD from Canada, patients who underwent dialysis spent more in-hospital days per person-year and had higher rates of intensive care unit admissions, although did not have higher rates of hospitalizations overall vs. those who received dialysis care.⁵⁹ With respect to US data, in a case series of Veterans with advanced CKD among whom 12,756 received dialysis and 503 opted not to pursue dialysis prior to death, the latter group was less likely to be hospitalized and to receive intensive patterns of end-of-life care in the last month prior to dying.⁶⁰ While these data support the role of non-dialytic management of advanced CKD, inference from these findings are limited by the small samples of medically-managed patients, residual confounding by indication, and unclear generalizability beyond single centers and/or non-US populations.

To our knowledge, this is the largest study to date that has examined the comparative effectiveness of non-dialytic care vs. dialysis transition upon hospitalization endpoints. Using national VA data linked to the USRDS and CMS databases, we were able to follow the longitudinal trajectories of a large cohort of advanced CKD patients who transitioned vs. did not transition to dialysis from one of the largest US healthcare systems. By applying PS-matching to identify comparable non-dialytic care vs. dialysis patient groups using the national VA database with exceptional capture of hospitalization data, we were also able to mitigate potential missing hospitalization endpoints (given the highly integrated and centralized nature of the national VA healthcare system, in which Veterans typically receive standardized care in VA-based centers and are less likely to switch healthcare systems) and confounding by indication and non-uniform health access. Using this rigorous approach, we found that patients who received dialysis had 1.6-fold higher hospitalization rates, 1.7-fold higher rates of days hospitalized, but marginally higher average LOS (~0.5 days) vs. patients managed conservatively. These data suggest that there is a trade-off between extended survival, greater healthcare utilization, and the increased time spent in the hospital within the context of dialysis treatment.

Another noteworthy finding in our study was the variation in hospitalization outcomes when comparing different periods of dialysis transition with non-dialytic care. In secondary analyses, incrementally earlier dialysis transitions were associated with increasingly higher hospitalization rates and rates of days hospitalized. Notably, while the earlier dialysis transition group had longer average LOS vs. the non-dialytic care group, those with intermediate+later dialysis transitions had similar lengths of stay. Given the attenuated hospitalization risk among patients with later dialysis transitions, these findings suggest that conservative and preservative measures to delay dialysis initiation may be beneficial, even for those who eventually pursue renal replacement therapy.^18,19,61,62

The strengths of our study include its examination of a large national cohort of advanced CKD patients with comprehensive capture of detailed clinical data; linkage of the VA database to CMS and USRDS data, enabling access to high-dimensional data and longitudinal tracking of outcomes; and rigorous use of PS-matching to reduce confounding by indication and ensure balance in characteristics across exposure groups. However, several limitations bear mention. First, our study solely focused on Veterans, who are predominantly male and tend to have higher comorbidity burden and lower socio-economic status. Although this limits the generalizability of findings to non-Veteran populations, Veterans comprise a significant portion of the US incident ESKD population (~13%), and these findings could have important implications for clinical practice and public health within a large segment of the kidney community.^7–10,63 Second, while the non-dialytic care group was classified based on the non-receipt of dialysis, we were unable to precisely determine which of these patients decided to forgo dialysis, nor the non-dialytic treatments they received for advanced CKD given the retrospective nature of the study. Third, among the dialytic care group, there was a tendency towards dialysis transition at higher eGFR levels than observed in the broader US and non-US based ESKD populations. While ~14% of the incident US ESKD population has had an eGFR ≥15 mL/min/1.73m² at the time of transition to kidney replacement therapy,^1,64 a higher prior proportion has been observed in US Veterans.⁶⁵ While we were unable to directly ascertain the indications for earlier dialysis initiation in this retrospective study, it is possible that this may have been due to higher comorbidity burden, leading to earlier onset of uremic symptoms, volume overload, and/or poor nutritional status, prompting clinicians to start dialysis earlier, and/or greater frailty and lower creatinine generation, leading to a numerically higher eGFR when actual kidney function is lower. Fourth, it bears mention that the hospitalization analyses in the dialysis group considered the entire peri-dialytic period (i.e., both the time preceding and following dialysis initiation as a clinically relevant period when patients are particularly vulnerable to hospitalization⁹) in order to capture the overall hospitalization burden associated with progression to and initiation of dialysis, rather than isolate risk during maintenance dialysis alone which may underestimate the burden of hospitalizations surrounding dialysis initiation. Fifth, although we used PS-matching to adjust for numerous confounders between the non-dialytic care and dialysis groups, residual confounding from unmeasured factors cannot be excluded.

Conclusion

In conclusion, among a national cohort of Veterans with advanced CKD, dialysis initiation was associated with higher rates of hospitalization and rates of days hospitalized, particularly with earlier dialysis transitions, as well as longer average lengths of stay compared to non-dialytic care. Further studies are needed to evaluate the comparative effectiveness of non-dialytic care vs. dialysis transition on other clinically relevant endpoints, and to identify which patients are ideally suited for these different treatment approaches.

List of abbreviations

BMI

Body mass index

CKD

chronic kidney disease

CMS

Center for Medicare and Medicaid Services

ESKD

end-stage kidney disease

eGFR

estimated glomerular filtration rate

IRRs

incident rate ratios

LOS

length of stay

NIH

National Institutes of Health

PS

propensity score

USRDS

US Renal Data System

VA

Veterans Affairs

Data Sharing Statement

Due to the nature of the research and data restrictions (i.e., data containing information that could compromise the privacy of research participants), supporting data is not available.