Mortality and Recovery Associated with Kidney Failure due to Acute Kidney Injury

Abstract

Visual Abstract

Background and objectives 

AKI requiring dialysis is a contributor to the growing burden of kidney failure, yet little is known about the frequency and patterns of recovery of AKI and its effect on outcomes in patients on incident dialysis.

Design, setting, participants, & measurements 

Using the US Renal Data System, we evaluated a cohort of 1,045,540 patients on incident dialysis from January 1, 2005 to December 31, 2014, retrospectively. We examined the association of kidney failure due to AKI with the outcome of all-cause mortality and the associations of sex and race with kidney recovery.

Results 

Mean age was 63±15 years, and 32,598 (3%) patients on incident dialysis had kidney failure due to AKI. Compared with kidney failure due to diabetes mellitus, kidney failure attributed to AKI was associated with a higher mortality in the first 0–3 months following dialysis initiation (adjusted hazard ratio, 1.28; 95% confidence interval, 1.24 to 1.32) and 3–6 months (adjusted hazard ratio, 1.16; 95% confidence interval, 1.11 to 1.20). Of the patients with kidney failure due to AKI, 11,498 (35%) eventually recovered their kidney function, 95% of those within 12 months. Women had a lower likelihood of kidney recovery than men (adjusted hazard ratio, 0.86; 95% confidence interval, 0.83 to 0.90). Compared with whites, blacks (adjusted hazard ratio, 0.68; 95% confidence interval, 0.64 to 0.72), Asians (adjusted hazard ratio, 0.82; 95% confidence interval, 0.69 to 0.96), Hispanics (adjusted hazard ratio, 0.82; 95% confidence interval, 0.76 to 0.89), and Native Americans (adjusted hazard ratio, 0.72; 95% confidence interval, 0.54 to 0.95) had lower likelihoods of kidney recovery.

Conclusions 

Kidney failure due to AKI confers a higher risk of mortality in the first 6 months compared with kidney failure due to diabetes or other causes. Recovery within 12 months is common, although less so among women than men and among black, Asian, Hispanic, and Native American patients than white patients.

Introduction

In the United States, >726,000 people are on maintenance dialysis, and kidney disease ranks as the ninth leading cause of death (^1–234). AKI requiring dialysis is a devastating complication during acute hospitalizations, and it is associated with significant mortality (^5–6789). Epidemiologic studies indicate that AKI is associated with a higher risk of incident and progressive CKD, and dialysis-dependent AKI contributes to kidney failure (^10–1112). Morbidity, mortality, and costs of care are among the highest during the transition to maintenance dialysis, particularly in the first year of dialysis care (¹³). However, whether these outcomes differ across patients with kidney failure due to AKI compared with those due to diabetes or other causes is not well understood. Another key element in the natural history of patients with kidney failure due to AKI is the prospect of recovery and nonrecovery. Yet, information about recovery is derived largely from small, single-center studies, where the rates of nonrecovery from AKI requiring dialysis vary significantly between 0% and 40% (^{14–15161718}).

Among patients undergoing maintenance dialysis, sex and race disparities in dialysis care, vascular access, and mortality are well reported in the literature. For instance, women have a lower likelihood of permanent vascular access, and after initiating dialysis, white patients have a higher mortality than nonwhite races (¹⁹,²⁰). Yet, there is limited information regarding the association of sex and race with kidney recovery from dialysis-dependent AKI.

Thus, that current literature contains key knowledge gaps regarding kidney failure due to AKI that impede our ability to deliver appropriate care to these patients on incident dialysis and to potentially promote kidney recovery. The objective of this study was to evaluate differences in mortality between patients with kidney failure due to AKI and those with kidney failure due to diabetes or other causes and to examine factors associated with kidney recovery, particularly sex and race, using the largest national administrative dataset in the United States, the US Renal Data System (USRDS).

Materials and Methods

Study Population and Data Sources

The study cohort was composed of 1,045,540 adults listed in the USRDS who initiated dialysis between January 1, 2005 and December 31, 2014. Total follow-up time for all patients was 3.3 million person-years, averaging 3.1 years per patient. Patients with missing information on sex, race, dialysis modality, or dialysis access were excluded. Figure 1 illustrates the study cohort derivation. Patients with unavailable information on other covariates were categorized into a “missing” group for that covariate (Table 1). The University of Cincinnati Institutional Review Board committee deemed the study exempt because the data were deidentified.

Figure 1.: Study cohort derivation of 1,045,540 adults who initiated dialysis between January 1, 2005 and December 31, 2014. USRDS, US Renal Data System.

Table 1. - Characteristics of patients initiating dialysis by cause of kidney failure from 2005 to 2014 included in the US Renal Data System

Characteristics	All, n=1,045,540	AKI, n=32,598 (3%)	Diabetes Mellitus, n=480,444 (46%)	Others, n=532,498 (51%)
Demographics
Age, yr	63 (15)	66 (14)	63 (13)	63 (17)
18–30	3	2	0.9	4
30–40	5	4	4	6
40–50	11	8	10	11
50–60	20	17	23	17
60–70	25	25	30	21
70–80	23	27	23	23
80+	15	19	10	19
Sex
Women	43	42	45	42
Race/ethnicity
Asian	4	2	5	4
Black	28	15	26	30
Hispanic	14	7	19	10
Native American	0.9	0.5	1.5	0.5
White	54	76	49	56
Body mass index (99% nonmissing), kg/m2	29 (8)	29 (8)	31 (8)	28 (8)
<18.5	3	4	2	5
18.5–25	30	31	23	36
25–30	28	27	28	28
>30	38	37	47	30
Missing	1	1	0.9	0.9
Neighborhood poverty, %
<13.8	59	68	56	61
13.8 to <20	18	16	19	18
20 to <40	20	14	2	19
≥40	2	1	2	2
Missing	1	1	2	1
Region
Midwest	23	31	21	23
Northeast	19	22	18	19
South	39	31	39	39
West	20	16	22	19
Unknown	0	0	0	0
Comorbidities
Congestive heart failure	32	35	37	27
Atherosclerotic heart disease + other	33	43	36	29
Hypertension	86	73	88	85
Diabetes mellitus	55	40	91	24
Cancer	8	14	5	10
Amputations	3	3	6	1
Peripheral vascular disease	13	15	17	10
Cerebrovascular accident/transient ischemic attack	9	10	11	8
Chronic obstructive pulmonary disease	9	15	9	9
Poor functional status	14	26	16	12
Laboratory values
Albumin (74% nonmissing), g/dl	3.1 (0.7)	2.8 (0.8)	3.1 (0.7)	3.2 (0.8)
<3.5	48	60	52	45
≥3.5	26	14	23	30
Missing	26	26	26	26
Hemoglobin (90% nonmissing), g/dl	9.8 (1.7)	9.9 (1.8)	9.8 (1.6)	9.8 (1.8)
<11	69	69	71	68
11–12	13	12	12	13
>12	8	9	7	9
Missing	10	11	10	11
Care
History of nursing home	7	17	7	6
Nephrology care, mo
None	29	62	25	31
<12	33	18	36	31
>12	26	7	27	25
Unknown	13	13	13	13
Dialysis access/modality
Arteriovenous fistula/arteriovenous graft	17	2	18	17
Catheter	76	97	75	75
Peritoneal dialysis	7	1	7	8
Mortality
Cause of mortality (anytime)
Cardiovascular	39	28	42	35
Infection	9	10	10	9
Malignancy	4	5	2	5
Withdrawal of dialysis	12	11	11	12
Unknown/other	37	47	35	39
Time
90 d	8	15	7	9
12 mo	22	35	20	23
Anytime during follow-up	61	66	63	58
Kidney recovery
Recovered function	6	35	4	7
Transplant
Kidney transplant	9	2	7	11

Data are represented as mean (SD) or proportion where appropriate.

The USRDS patient file was used to obtain information on the date of maintenance dialysis incidence (identified by date of first dialysis service), cause of kidney failure, age, race, sex, date and cause of death, first kidney transplant date, initial dialysis modality, and dialysis networks classified into geographic regions (²¹). The Centers for Medicare and Medicaid Services (CMS) Form-2728 was used to obtain information on body mass index, dialysis modality, vascular access type, comorbidities, predialysis nephrology care, history of nursing home, laboratory data, and poor functional status. We used the residence file to obtain information on patients’ zip codes and determined neighborhood socioeconomic status, defined as percentage of zip code residents living below the federal poverty level (^22–2324). Kidney recovery was determined from the USRDS Rxhist file (³).

Outcomes and Exposures

The outcome was all-cause mortality in patients on dialysis. The exposure was kidney failure cause grouped into AKI, diabetes, and all other causes. For this study, patients with kidney failure due to AKI were those with the primary cause of kidney failure listed as “tubular necrosis (no recovery)” on the CMS Form-2728. We evaluated differences in mortality in patients with kidney failure due to AKI, kidney failure due to diabetes, and kidney failure due to other causes. In patients with kidney failure due to AKI, we examined frequency and patterns of kidney recovery and the associations of sex and race with kidney recovery. Kidney recovery was defined as discontinuation of dialysis due to recovered kidney function, and kidney recovery was restricted to the first 12 months postkidney failure in the survival models. We also determined the effect of recovery on all-cause mortality, with recovery treated as a time-dependent covariate.

Statistical Analyses

Summary statistics are presented as percentages for categorical variables and mean ± SD for continuous variables. Cox proportional hazard models were used to model all-cause mortality over the entire study period. Because the proportional hazards assumption is strongly violated for the association of kidney failure causes with mortality (a linear nonproportionality effect across follow-up time yielded a P<0.001), we estimated multiple models with mortality as the outcome during different postkidney failure periods (0–3, 3–6, 6–12, and 12 months thereafter). Observation time was censored for kidney transplant, end of study follow-up, or follow-up window being examined (for each time period, only those patients who survived the previous time period were eligible to be followed for the subsequent time period). Models were nonparsimonious and, in addition to sex, race, and kidney failure cause, included the following covariates: body mass index, comorbidities, dialysis access, predialysis nephrology care, laboratory data, history of nursing home, functional status, neighborhood poverty, and region. We collected baseline information for patients overall and for those with kidney failure due to AKI, entering each of these follow-up windows because of possible changes in the cohort due to dropping patients across these follow-up periods (Supplemental Table 1). Because of the possible misclassification bias, we performed sensitivity analyses by excluding patients with AKI who initiated dialysis with arteriovenous access and patients without AKI who eventually recovered kidney function. For the models predicting recovery, we used Fine and Gray cumulative incidence models to determine factors associated with kidney recovery in patients with kidney failure due to AKI, with death as a competing risk, using the same covariates as above and with observation time censored for kidney transplant or after 12 months because 95% of the recoveries happened within that time (²⁵,²⁶). Kaplan–Meier curves were used to present unadjusted survival probabilities, starting from dialysis initiation, across the entire follow-up period, with censoring for kidney transplant, for patients with kidney failure due to AKI, diabetes, and for other causes. Cumulative incidence probabilities were computed for kidney recovery with death considered as a competing risk, as well as from the submodel of death itself, unadjusted, and censored for kidney transplant. Sensitivity models repeated the models of recovery using death as a censoring event rather than as a competing risk. To assess the effect of kidney recovery in the patients with AKI, we estimated Cox proportional hazards models of mortality using recovered function as a time-dependent covariate with and without the previously mentioned covariates. All P values are two tailed, and α for all tests was 0.05, unadjusted for multiple tests. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC).

Results

Baseline Characteristics of the Study Cohort

The mean age was 63±15 years. In the study cohort, 57% were men, and 54% were white. In 3% of patients (n=32,598), AKI was the cause of kidney failure; in 46% (n=480,444), it was diabetes; and in 51% (n=532,498), it was one of the other causes of kidney failure. Table 1 shows the clinical characteristics of the study cohort overall and by kidney failure cause. The proportion of whites was higher among patients with kidney failure due to AKI than among diabetes or other causes (76% versus 49% versus 56%, respectively; P<0.001). Patients with kidney failure due to AKI were more likely to start dialysis with a central venous catheter (as opposed to arteriovenous access or peritoneal dialysis) than were patients with kidney failure due to diabetes or other causes. Patients with kidney failure due to AKI had higher proportions of cancer, poor functional status, and nursing home stay.

Mortality by Cause of Kidney Failure

Overall, mortality was 0.19 deaths per person-year (PPY; n=21,411). Patients with kidney failure due to AKI had both higher 12-month mortality (35% [n=11,465] or 0.45 deaths PPY; P<0.001) and 3-month mortality (15% [n=4990] or 0.66 deaths PPY; P<0.001) than did those with kidney failure due to diabetes or other causes. Supplemental Table 2 shows mortality in PPY for various follow-up periods by cause of kidney failure. In the unadjusted analysis, patients with kidney failure due to AKI compared with those due to diabetes had a higher hazard of death in 0–3, 3–6, 6–12, and 12–24 months but subsequently demonstrated a lower hazard of death until the end of follow-up. In the adjusted analyses, compared with kidney failure due to diabetes, kidney failure due to AKI was associated with higher adjusted hazards of mortality in the first 0–3 (hazard ratio [HR], 1.28; 95% confidence interval [95% CI], 1.24 to 1.32) and 3–6 months (HR, 1.16; 95% CI, 1.11 to 1.20), followed by decreasing adjusted hazards of mortality at 6–12 months and beyond. Similarly, patients with kidney failure due to AKI had a higher hazard of death compared with those with kidney failure due to other causes through the first 6 months of follow-up, followed by a lower hazard of death for the time period from 12 months and beyond (Table 2). Figure 2 shows the HRs of mortality for the first 36 months comparing three groups of kidney failure. The later follow-up periods, which included patients surviving the longest, had lower rates of comorbidities (Supplemental Table 1). Figure 3 shows Kaplan–Meier survival estimates comparing patients by cause of kidney failure. In this time-to-event analysis, kidney failure due to AKI was associated with lower survival (log-rank test P<0.001). In the sensitivity analyses, by excluding patients with AKI who initiated dialysis with arteriovenous access and patients without AKI who recovered kidney function, hazards of death were comparable with the main model, tending to attenuate by <10% in magnitude (Supplemental Table 3).

Table 2. - Associations of AKI as cause of kidney failure with mortality after initiating dialysis according to time from dialysis initiation

Follow-Up, mo	Mortality a			Unadjusted Hazard Ratio (95% Confidence Interval)		Adjusted Hazard Ratio (95% Confidence Interval)
	AKI	Diabetes Mellitus	Others	AKI versus Diabetes Mellitus	AKI versus Others	AKI versus Diabetes Mellitus	AKI versus Others
0–3	4990/32,598 (15%)	31,448/480,444 (7%)	46,160/532,498 (9%)	2.45 (2.38 to 2.52)	1.82 (1.77 to 1.88)	1.28 (1.24 to 1.32)	1.06 (1.03 to 1.09)
3–6	3206/27,576 (12%)	26,033/447,979 (6%)	32,683/483,338 (7%)	2.07 (1.99 to 2.14)	1.76 (1.70 to 1.88)	1.16 (1.11 to 1.20)	1.09 (1.05 to 1.13)
6–12	3269/24,346 (13%)	37,886/420,381 (9%)	42,053/446,769 (9%)	1.53 (1.47 to 1.58)	1.45 (1.40 to 1.50)	0.94 (0.90 to 0.97)	0.97 (0.94 to 1.01)
12–24	3511/20,988 (17%)	59,367/378,559 (16%)	57,663/396,844 (15%)	1.07 (1.03 to 1.11)	1.15 (1.11 to 1.19)	0.74 (0.72 to 0.77)	0.86 (0.83 to 0.89)
24–36	2182/16,621 (13%)	46,011/296,067 (16%)	40,680/310,765 (13%)	0.82 (0.78 to 0.85)	0.98 (0.94 to 1.02)	0.61 (0.58 to 0.64)	0.77 (0.73 to 0.80)

^aNumber of patients died during time period/number of patients alive at the beginning of the time period (percentage died).

Figure 2.: Kidney failure due to AKI is associated with a higher risk of mortality in the first 6 months following dialysis initiation compared with kidney failure due to diabetes or other causes. Cox proportional hazard models showing hazard ratios for mortality in the kidney failure due to AKI group versus mortality in the diabetes mellitus group, and mortality in the kidney failure due to AKI group versus the mortality in the other group for various postdialysis periods, censored for transplant. (A) Unadjusted hazard ratios and (B) adjusted hazard ratio for all covariates.

Figure 3.: Kaplan-Meier survival estimates for patients show that relative survival probabilities for patients with different kidney failure causes changes across follow-up time. Months from initial dialysis to death, unadjusted and censored for transplant or end of study period, for three cause of kidney failure groups, with numbers of patients at risk.

Kidney Recovery Rates and Factors Associated with Kidney Recovery

Overall, 35% (n=11,498) of patients with kidney failure due to AKI recovered kidney function (34% [n=10,928] within 12 months) for an overall recovery rate of 0.3 PPY and a 12-month recovery rate of 0.6 PPY. In comparison, 4% of patients with kidney failure due to diabetes and 7% of patients with kidney failure due to other causes recovered kidney function (4% and 6%, respectively) within 12 months. Supplemental Table 4 shows kidney recovery rates in PPY for various follow-up periods. Table 3 shows characteristics of patients with kidney failure due to AKI by kidney recovery status within 12 months of kidney failure. Patients with recovered kidney function at 12 months were younger on average than patients with nonrecovered kidney function (62 versus 68 years). Octogenarians had a lower rate of 12-month kidney recovery than nonoctogenarians (20% versus 37%; P<0.001). Women had a lower rate of 12-month kidney recovery than men (30% versus 36%; P<0.001). Also, blacks and Asians had significantly lower 12-month recovery rates than Hispanics, Native Americans, or whites (29% versus 29% versus 32% versus 33% versus 35%, respectively; P<0.001). Figure 4 shows the unadjusted cumulative incidence probabilities for kidney recovery.

Table 3. - Patient characteristics by kidney recovery status within 12 mo of ESKD in patients with ESKD due to AKI

Characteristics	Recovered, n=10,928 (34%)	Not Recovered, n=21,670 (66%)
Demographics
Age, yr	62 (15)	68 (13)
18–30	3	1
30–40	6	2
40–50	11	6
50–60	21	15
60–70	25	25
70–80	23	29
80+	11	22
Sex
Women	38	44
Race/ethnicity
Asian	1	2
Black	13	16
Hispanic	7	7
Native American	0.5	0.5
White	78	75
Body mass index (99% nonmissing), kg/m2	30 (8)	29 (9)
<18.5	3	5
18.5–25	28	33
25–30	29	26
>30	40	35
Missing	1	1
Neighborhood poverty, %
<13.8	69	68
13.8 to <20	16	15
20 to <40	13	14
≥40	0.8	1
Missing	1	1
Region
Midwest	32	31
Northeast	18	24
South	35	29
West	16	16
Unknown	<0.1	<0.1
Comorbidities
Congestive heart failure	24	40
Atherosclerotic heart disease + other	36	46
Hypertension	68	75
Diabetes mellitus	36	42
Cancer	11	15
Amputations	21	3
Peripheral vascular disease	12	16
Cerebrovascular accident/transient ischemic attack	8	11
Chronic obstructive pulmonary disease	12	16
Poor functional status	19	30
Laboratory values
Albumin (74% nonmissing), g/dl	2.8 (0.8)	2.8 (0.8)
<3.5	61	60
≥3.5	13	14
Missing	26	26
Hemoglobin (90% nonmissing), g/dl	10.2 (2.0)	9.8 (1.7)
<11	63	71
11–12	13	11
>12	14	7
Missing	11	10
Care
History of nursing home	12	19
Nephrology care, mo
None	73	56
<12	12	22
>12	4	9
Unknown	11	14
Dialysis access/modality
Arteriovenous fistula/arteriovenous graft	1.0	2
Catheter	99	96
Peritoneal dialysis	0.2	1.0
Mortality
Time
90 d	2	23
12 mo	11	49
Anytime during follow-up	40	80
Transplant
Kidney transplant	0.4	3

Data are represented as mean (SD) or proportion where appropriate.

Figure 4.: Cumulative probabilities for mortality and recovery of kidney function show that recovery is confined primarily to the first 6 months postdialysis initiation. From unadjusted Fine and Gray cumulative incidence model of recovery with censoring for transplant and treating death as a competing risk and from the submodel of mortality with censoring for transplant.

Table 4 elucidates the Fine and Gray model with death as a competing risk, showing factors associated with 12-month kidney recovery in patients with kidney failure due to AKI. Women had lower adjusted hazards of 12-month kidney recovery than men (HR, 0.86; 95% CI, 0.83 to 0.90). Blacks (HR, 0.68; 95% CI, 0.64 to 0.72), Asians (HR, 0.82; 95% CI, 0.69 to 0.96), Hispanics (HR, 0.82; 95% CI, 0.76 to 0.89), and Native Americans (HR, 0.72; 95% CI, 0.54 to 0.95) also had lower hazards of kidney recovery compared with whites. The likelihood of recovery decreased steadily with older age at the onset of kidney failure. The comorbidities of diabetes, congestive heart failure, amputation, cancer, atherosclerotic heart disease, and poor functional status were associated with lower likelihoods of kidney recovery. Patients with a history of predialysis nephrology care had a lower likelihood of kidney recovery than those who did not receive predialysis nephrology care. The results from the Cox proportional hazards model with death modeled as a censoring event rather than as a competing risk were consistent with the main model, including the HRs for race and sex (Supplemental Table 5).

Table 4. - Patient characteristics associated with recovery of kidney function within 12 mo of initiating dialysis for AKI

Characteristics	N	Adjusted Hazard Ratio (95% Confidence Interval)	P Value
Demographics
Year of dialysis initiation		0.98 (0.98 to 1.00)	<0.001
Age, yr			<0.001
18–29	294/519 (57%)	Reference
30–39	652/1131 (58%)	0.96 (0.83 to 1.12)
40–49	1251/2486 (50%)	0.79 (0.69 to 0.91)
50–59	2263/5459 (41%)	0.64 (0.56 to 0.73)
60–69	2741/8076 (34%)	0.54 (0.47 to 0.62)
70–79	2483/8826 (28%)	0.45 (0.39 to 0.52)
80+	1244/6101 (20%)	0.33 (0.29 to 0.38)
Sex			<0.001
Women	4116/13,672 (30%)	0.86 (0.83 to 0. 90)
Men	6812/18,926 (36%)	Reference
Race/ethnicity			<0.001
Asian	154/534 (29%)	0.82 (0.69 to 0.96)
Black	1406/4880 (29%)	0.68 (0.64 to 0.72)
Hispanic	748/2312 (32%)	0.82 (0.76 to 0.89)
Native American	52/157 (33%)	0.72 (0.54 to 0.95)
White	8568/24,715 (35%)	Reference
Body mass index, kg/m2			<0.001
<18.5	347/1369 (25%)	0.78 (0.70 to 0.87)
18.5–24.99	3009/10,130 (30%)	Reference
25–29.99	3145/8844 (36%)	1.25 (1.19 to 1.32)
≥30	4320/11,915 (36%)	1.29 (1.24 to 1.36)
Missing	107/340 (31%)	1.07 (0.88 to 1.30)
Neighborhood poverty, %			0.29
<13.8	7500/22,247 (34%)	Reference
13.8 to <20	1777/5129 (35%)	1.00 (0.95 to 1.05)
20 to <40	1438/4475 (32%)	0.96 (0.90 to 1.02)
≥40	91/362 (25%)	0.83 (0.67 to 1.02)
Missing	122/385 (32%)	0.94 (0.78 to 1.13)
Region			<0.001
Midwest	3463/10,145 (34%)	0.91 (0.87 to 0.96)
Northeast	1923/7170 (27%)	0.76 (0.72 to 0.80)
South	3770/10,088 (37%)	Reference
West	1770/5191 (34%)	0.90 (0.84 to 0.95)
Unknown	2/4 (50%)	1.60 (0.44 to 5.88)
Comorbidities
Congestive heart failure			<0.001
Yes	2657/11,402 (23%)	0.66 (0.63 to 0.70)
No	8271/21,196 (39%)	Reference
Atherosclerotic heart disease + other			0.05
Yes	3896/13,910 (28%)	0.96 (0.92 to 1.00)
No	7032/18,688 (38%)	Reference
Hypertension			<0.001
Yes	7416/23,667 (31%)	0.93 (0.89 to 0.97)
No	3512/8931 (39%)	Reference
Diabetes mellitus			<0.001
Yes	3883/12,933 (30%)	0.91 (0.87 to 0.95)
No	7045/19,665 (36%)	Reference
Cancer			<0.001
Yes	1246/4430 (28%)	0.78 (0.74 to 0.83)
No	9682/28,168 (34%)	Reference
Amputations			0.007
Yes	258/958 (27%)	0.84 (0.74 to 0.95)
No	10,670/31,640 (34%)	Reference
Peripheral vascular disease			0.33
Yes	1285/4775 (27%)	0.98 (0.92 to 1.04)
No	9643/27,823 (35%)	Reference
Cerebrovascular accident/transient ischemic attack			0.04
Yes	845/3210 (26%)	0.93 (0.86 to 1.00)
No	10,083/29,388 (34%)	Reference
Chronic obstructive pulmonary disease			0.002
Yes	1304/4826 (27%)	0.91 (0.86 to 0.97)
No	9624/27,772 (35%)	Reference
Poor functional status			<0.001
Yes	2027/8.440 (24%)	0.70 (0.66 to 0.74)
No	8901/24,158 (37%)	Reference
Laboratory values
Albumin, g/dl			0.004
<3.5	6656/19,650 (34%)	1.11 (1.04 to 1.17)
≥3.5	1428/4490 (32%)	Reference
Missing	2844/8458 (34%)	1.08 (1.01 to 1.16)
Hemoglobin, g/dl			<0.001
<11	6904/22,396 (31%)	0.64 (0.61 to 0.68)
11–12	1375/3738 (37%)	0.81 (0.75 to 0.87)
>12	1483/3040 (49%)	Reference
Missing	1166/3424 (34%)	0.67 (0.61 to 0.72)
Care
History of nursing home			0.03
Yes	1345/5562 (24%)	0.93 (0.87 to 0.99)
No	9583/27,063 (35%)	Reference
Nephrology care, mo			<0.001
None	8013/20,053 (40%)	Reference
0–12	1310/5990 (22%)	0.51 (0.49 to 0.55)
>12	402/2409 (17%)	0.42 (0.38 to 0.47)
Unknown	1203/4146 (29%)	0.72 (0.68 to 0.76)

Fine and Gray cumulative incidence model censored at 12 mo, end of the follow-up period, or transplant, with death as a competing risk. Column N contains the number of patients recovering within 12 mo divided by the total number of patients in that group and the percentage recovered. Comorbidities and history of nursing home indicate that the patient had the comorbidity and/or had been in a nursing home. Adjusted for covariates of year of dialysis initiation, sex, race/ethnicity, body mass index, neighborhood poverty, region, comorbidities, laboratory values, and care. P value for year of dialysis initiation is from a one-degree of freedom test of linear trend. The other P values are from tests with multiple degrees of freedom when more than two groups are present. All P values are two tailed.

The median time to recovery of patients with kidney failure due to AKI was 2 months (interquartile range, 1.2–3.5), with 95% recovered by 12 months. With regard to the effect of kidney recovery on mortality in patients with kidney failure due to AKI, expectedly, recovery as a time-dependent covariate had an unadjusted HR of 0.31 (95% CI, 0.30 to 0.32) and adjusted HR of 0.38 (95% CI, 0.37 to 0.39).

Discussion

In one of the largest studies derived from a national cohort, we report that kidney failure due to nonrecovery of AKI confers a significantly higher adjusted risk of death for the first 0–6 months compared with kidney failure due to diabetes or other causes. Over a third of patients with kidney failure due to AKI recovered kidney function, with the majority of them doing so by 12 months. Moreover, the likelihood of kidney recovery was lower in women than in men, whereas compared with whites, it was lower in all other racial categories.

Medicare in 2017 allowed chronic dialysis facilities to furnish care for patients with AKI, including those eventually deemed as having kidney failure due to AKI (²⁷). Our study shows that after initiation of dialysis, compared with diabetes, kidney failure due to AKI was associated with 28% and 16% higher adjusted hazards of death in 0–3 and 3–6 months postdialysis initiation, respectively. Similarly, compared with kidney failure due to other causes (no patients with diabetes, non-AKI), these higher hazards of death persisted through the first 6 months. These observations strongly caution against implementing a “one-size-fits-all” model of incident dialysis care and suggest the necessity of customized care for those with kidney failure due to AKI. The initial higher risk of death in subjects with kidney failure due to AKI could be conferred by their acute care hospitalizations prior to transitioning to long-term dialysis because of their severity of illness when they initiate dialysis. Additionally, some standardized dialysis care treatments or practices may be potentially harmful to patients with AKI, particularly when implemented during the early post-AKI phase. For instance, patients with AKI may respond differently to ultrafiltration rates or may need better monitoring for intradialytic hypotension. Additionally, patients with AKI may either be on certain drugs or react differently to drugs commonly used in maintenance dialysis care. Prospective studies with patient-level information are necessary to address these critical questions in this vulnerable group of patients.

Our study found that, overall, 35% of patients deemed to have kidney failure due to AKI recovered their kidney function, 95% of those within 12 months. Recovery of kidney function in those with kidney failure due to AKI has been reported in prior decades. For instance, Foley et al. (²⁸) had reported a 27.7% recovery rate for patients with kidney failure due to acute tubular necrosis (between 2001 and 2010) compared with 3.5% in those without acute tubular necrosis in the first 3 months. In contrast, a single-center cohort from the United Kingdom between 1984 and 1995 exhibited only a 5.6% frequency of recovery among dialysis-dependent patients with AKI during the 3–18 months following dialysis initiation (²⁹). Our findings suggest that the careful monitoring for recovery should be extended beyond 3 months. Because 75% of recovery occurs within 4 months, prospective studies should focus on this time frame when evaluating possible interventions that may improve clinical outcomes in patients with kidney failure due to AKI. Patterns and practices of dialysis may also be associated with kidney recovery in these patients. In two other small, single-center studies, the probability of kidney recovery was influenced by the presence of fluid overload and the degree of ultrafiltration (¹⁷,¹⁸). We found that the presence of congestive heart failure as a comorbidity was a strong predictor of nonrecovery of kidney function, as was octogenarian status. These findings suggest that there is an imminent need to identify modifiable conditions or dialysis treatment parameters that may impede improvements in kidney function in dialysis-dependent AKI and a need to develop novel therapies to promote tissue recovery.

In contrast to prior studies, our analysis of kidney recovery focused on the associations of sex and race with the recovery prospects. Sex and race disparities are known to influence incident dialysis care. For example, in the Hemodialysis Study, the mean treatment time was shorter in women, and women had a 19% lower mortality risk than men when randomized to the higher dose of dialysis versus the standard dose of dialysis, suggesting that sex-specific characteristics influence the response to the high dose and, possibly, the likelihood of kidney recovery (³⁰). Women have lower nephron mass, with 12% fewer glomeruli than men (³¹). Compared with men, women receive inferior treatment of anemia, undergo late initiation of dialysis, and receive lower dialysis doses that may further delay kidney recovery (^32–333435). Foley et al. (²⁸) showed 31% and 21% lower likelihoods of kidney recovery of blacks versus whites and Hispanic ethnicity versus non-Hispanic ethnicity, respectively, but they did not report recovery rates in Asians and Native Americans. Minorities have a higher likelihood of late initiation of dialysis, which may reflect reduced access to care (³⁵). This study found that, compared with white patients, all other racial categories were less likely to experience recovery. Similarly, women were 14% less likely to experience recovery than men. These differences persisted despite adjusting for traditional factors that may be associated with disparities in dialysis care (e.g., type of dialysis access, socioeconomic status, etc.). The reasons for the effect of sex or race on the likelihood of kidney recovery remain unclear. One possibility could be that the severity of kidney injury or the residual nephron mass may influence the rate of recovery differently across differences in sex and race. Additionally, postincident dialysis practices, such as ultrafiltration rates, intradialytic weight gain, or hypotension, may affect differently on the basis of certain demographic characteristics. Given the differences observed across sex and race, further studies of the possible cultural and social contributors and strategies to improve clinical monitoring of patients with kidney failure due to AKI for kidney recovery may have to be specifically directed to that subgroup of population.

Our analysis has limitations inherent in the use of administrative registries, such as the lack of ability to demonstrate causality and possible misclassifications due to physician bias in reporting the cause of kidney failure due to AKI on the CMS Form-2728. Although our approach allows estimates of mortality risk for kidney failure due to AKI during the early follow-up periods, there is a possibility that these estimates may not have continued in the later years, thereby leading to survivor bias. Because our cohort represents patients with kidney failure due to AKI, these findings may not be generalizable to all patients with dialysis-requiring AKI. However, we do examine a national sample derived from a validated USRDS database, which allows us to study the natural history and patterns of recovery in a robust fashion. Given that the USRDS data begin with incident dialysis information, we were not able to ascertain the level of CKD prior to patients experiencing AKI. To that effect, the variable defining predialysis nephrology care may be considered a surrogate for receiving care for CKD prior to incident dialysis. This may explain our observation that predialysis nephrology care was associated with lower likelihood of kidney recovery.

In conclusion, our study suggests the need for developing customizable treatment strategies for patients with kidney failure due to AKI, in particular focusing on factors promoting kidney recovery. Future studies will need to assess patient-level information to determine the best dialysis options for those with kidney failure due to AKI, while accurately predicting recovery of kidney function with a focus on women and minorities.

Disclosures

All authors have nothing to disclose.

Funding

S. Shah is supported by National Institutes of Health grant 2UL1TR001425-05A1; intramural funds from the Division of Nephrology, University of Cincinnati; and a Dialysis Clinic, Inc. grant S-2719. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.