Categories of Hospital-Associated Acute Kidney Injury: Time Course of Changes in Serum Creatinine Values

Abstract

Hospital-associated acute kidney injury (HA-AKI) is associated with increased inpatient mortality. Our objective was to categorize HA-AKI based on the timing of minimum and peak inpatient serum creatinine and describe the association with inpatient mortality.

Materials and Methods

Retrospective analysis of an administrative data set for adults admitted to a single medical center over 4 years. Changes and timing of the minimum and peak serum creatinine were used to define HA-AKI categories.

Results

Peak creatinine followed minimum creatinine for HA-AKI, and preceded the minimum value for transient hospital-acquired AKI (THA-AKI). A subset of patients developed HA-AKI after recovering from THA-AKI. Multivariable Cox regression analyses examined the association between these categories and 28-day inpatient mortality, adjusting for age, sex, race, Charlson comorbidity index, baseline kidney function, AKI recovery and renal replacement therapy. There were 50,601 patients included in the analyses, and 29,996 (59%) did not have AKI. There were 2,440 deaths; HA-AKI had a 2.24-fold (95% CI: 1.99–2.51) increased risk, while TH-AKI group (12,101) had a 1.23-fold (95% CI: 1.09–1.40) increased risk for inpatient mortality. THA-AKI patients who recovered and then developed HA-AKI had the same mortality risk as THA-AKI (1.27-fold (95% CI: 1.07–1.51)) but longer hospitalization and less recovery from AKI.

Conclusions

Risk of short-term inpatient mortality is associated with AKI, and this risk is attenuated with recovery of kidney function in the hospital. Systematic surveillance with repeated inpatient sCr values is needed to assess the short- and long-term consequences of HA-AKI.

Introduction

Acute Kidney Injury (AKI) has significant sequelae including increased inpatient mortality, and long-term risk of renal and cardiovascular events and death [1]. AKI is common among critically ill patients [2-5]. Once AKI was ascertained based on changes in serum creatinine values rather than diagnostic related groups, the striking incidence of AKI became apparent [6].

The definition of AKI is based on a change in serum creatinine (sCr) from baseline values, or a surrogate value if baseline values are not available [7,8]. Hospital-associated AKI (HA-AKI) is characterized by maximal (“peak”) sCr values compared to the baseline value [3,9]. “Transient azotaemia” has been used to describe elevated sCr that normalizes within 3 days of admission without renal replacement therapy [10] in contrast to patients with HA-AKI who often have persistently elevated sCr at the time of discharge.[10-12] Transient HA-AKI (THA-AKI) has been described an independent predictor of hospital mortality (odds ratio 2.3 compared to patients without AKI), while persisting AKI was associated with a 6.1-fold risk for inpatient mortality [10].

We retrospectively analyzed an administrative dataset from a single large referral center (University of Alabama Hospital at Birmingham (UAHB)) to examine the classification of AKI and its association with inpatient mortality, using the minimum sCr during the admission as the surrogate for baseline sCr. We hypothesized that the timing of minimum sCr relative to the peak sCr could distinguish between THA-AKI and HA-AKI. Accordingly, our objectives were to 1) distinguish between AKI categories, using the temporal relation between minimum and peak sCr, and 2) describe the associations of the AKI categories with inpatient mortality.

Materials and Methods

We examined hospital discharges from UAHB, a 1,121-bed academic health care center between October 1, 2009 and September 30, 2013, using demographic information, admissions and discharge dates, admission site location code, hospital inpatient service location, discharge diagnoses and procedure codes, and discharge vital status. Each admission was linked to UAHB's central laboratory, and sCr values were extracted with date-time stamps and service locations. The Institutional Review Board at UAHB approved the use of de-identified patient information for this study, and waived the requirement for informed consent at the patient level.

There were 109,456 patients (age ≥18 years) discharged from UAHB (181,574 admissions), with 5,452 inpatient deaths (3.0%) during the ascertainment interval. Patients with chronic dialysis, or kidney transplantation, or less than two sCr measurements were excluded from the analysis. Cases were excluded for sCr <0.3 mg/dL, estimated glomerular filtration rates (eGFR) <10 ml/min/1.73 m², and length of stay <1 day [4]. Four cases were excluded with apparent miscoding of inpatient renal replacement therapy, based on <0.2 mg/dL difference between first sCr, peak sCr, minimum sCr and last sCr.

Admission point of origin was determined from the site location code included in the discharge summary [13]. There were 11,065 patients transferred to UAHB from other hospitals, and 939 from other health care facilities (e.g., skilled nursing homes, hospice units, psychiatry unit, rehabilitation unit, etc.); these 12,004 patients are defined as admissions originating from health care facilities. There were 38,576 patients admitted to UAHB from community locations (including 20,931 from home or office, 12,821 from outpatient physician offices or clinics and 4,725 patients from UAHB Emergency Department). There were 122 patients with unknown point of origin who were excluded from analysis. The follow-up period was limited to 28-day maximum length of stay. The final analytic data set included 50,580 patients who were admitted once during the evaluation period (eTable 1).

The admission date-time stamp was assigned to that of the first sCr, and was adjusted by 24 hours for each day that the first sCr measurement was obtained after the admission date. HA-AKI was defined as cases where the minimum sCr preceding the highest measured (“peak”) sCr value for the admission. We used an increment of 0.3 mg/dL as the criteria for AKI based on prior studies of the minimum increment independently associated with risk of 28-day inpatient mortality [3,9,14]. THA-AKI was defined with the highest (“peak”) sCr preceding the minimum sCr, with a subsequent decrease sCr ≥0.3 mg/dL from the peak sCr. Some THA-AKI patients subsequently developed HA-AKI following recovery to the minimum sCr value, and were then classified as 2HA-AKI. Patients without a clinically significant change (± 0.3 mg/dL) between peak and minimum sCr were classified as not having AKI (No-AKI).

For time-to-event regression analysis, AKI category was the effect variable, and the exposure period was length of stay (>1and ≤28 days). The covariates included age stratified at the median value of 55 years, gender, black or non-black race, Charlson comorbidity index [15], and eGFR (stratified at 60 ml/min/1.73 m², where the CKD-EPI eGFR [16] was calculated from the minimum inpatient sCr, provision of inpatient renal replacement therapy, and covariates that described the extent of recovery from peak sCr. The outcome measure was inpatient mortality, ascertained as vital status at time of discharge or at the 28^th day of stay. We calculated the hazard ratios (±95% confidence intervals (CI)) for inpatient mortality among cases with No-AKI (reference), and AKI categories. We compared baseline characteristics of patients using chi-squared test and ANOVA. Dunnett tests were used for multiple comparisons, with minimal statistical significance set at P≤0.05. Mean values are presented as ±1 standard deviation (SD), and medians with 25^th and 75^th-centiles. Survivor functions for multivariable-adjusted Cox analysis with point-wise confidence intervals were generated as described [17]. Relative integrated discrimination improvement (rIDI) was used with an a priori criterion for clinical significance of 10% [18,19]. Statistical analyses were performed with Stata version 13.1 (Stata Corp, College Station, TX), using the following functions: group, IDIsurv, median test, pwmean, prtest, stcox (with tvc option), and survci.

Results

The patient characteristics, classified by AKI category are shown in Table 1: 29,989 (59%) did not develop AKI (No-AKI); 12,101 (24%) had THA-AK; 2,243 (4.4%) developed HA-AKI after recovering from THA-AKI (2HA-AKI); and 6,247 (12%) developed HA-AKI. Compared to the No-AKI reference group, patients with AKI were older, more likely to be male, black, had longer lengths of stay, higher Charlson comorbidity scores, more likely to spend at least one day in an intensive care unit, and more likely to be transferred from another health care facility to UAHB.

Table 1.

Characteristics of 50,580 Inpatients (2,500 deaths (4.9%)) without, or with Acute Kidney Injury, Including Discharge Diagnosis Groups

	AKI Category
	No-AKI: 29,989 (59%)b	THA-AKI: 12,101 (24%)b	2HA-AKI: 2,243 (4.4%)b	HA-AKI: 6,247 (12%)b
Age (years)	53±18	55±18	56±18c	58±17c
Male sex	15,028 (50%)	7,270 (60%)	1,408 (63%)	3,408 (55%)c
Black raced	8,673 (29%)	3,933 (33%)	465 (36%)	1,998 (32%)c
Baseline eGFRe	100 ±27	94±35	95±40	87±34c
HCF Transfers	6,336 (21%)	2,941 (24%)	722 (32%)	2,005 (32%)c
LOS (days)	3.7±3.2	6.9±5.2	12±6.5	8.1±6.1
Charlson Score	1.2±1.7	1.6±1.9	2.0±2.1	2.1±2.1c
Diabetes (%)	5,089 (17%)	2,697 (22%)	466 (21%)	1,553 (25%)c
COPD (%)	4,459 (15%)	1.875 (16%)	397 (18%)	1,140 (18%)c
CHF (%)	2,104 (7.0%)	1,429 (12%)	434 (19%)	1,312 (21%)c
Stroke (%)	2,534 (8.5%)	994 (8.2%)	241 (11%)	631 (10%)c
Renal Disease (%)	1,139 (3.8%)	1,681 (14%)	492 (22%)	970 (16%)
AMI (%)	1,893 (6.3%)	1,057 (8.7%)	265 (12%)	842 (14%)
Cancer (%)	2,263 (7.6%)	854 (7.1%)	154 (6.9%)	549 (8.8%)c
Met Cancer (%)	1,394 (4.7%)	590 (4.9%)	87 (3.8%)	361 (5.8%)c
Inpatient RRT	0	287 (2.4%)	246 (11%)	256 (4.1%)
Ever in ICU (%)	5,280 (18%)	4,564 (38%)	1,425 (64%)	3,028 (49%)c
Inpatient deaths (%)	506 (1.7%)	606 (5.1%)	267 (12%)	1,121 (18%)c

Unless otherwise indicated, values are given as number (percentage) or means ± 1 SD; medians are given with (25^th –75^th-centile).

Abbreviations: AKI, acute kidney injury; 2HA-AKI, second episode of AKI in a the course of a single hospital admission; eGFR, estimated glomerular filtration rate; HA-AKI, hospital-associated AKI; ICU, intensive care unit; IRR, incident rate ratio (incident rates express as events per 28 patient-days); sCr, serum creatinine; THA-AKI, transient hospital-associated AKI.

^a P<0.001 for comparisons of characteristics between each group by ANOVA for means, signed rank test for medians, and Chi Squared test for proportions; shown as bold text, relative to No-AKI which served as the reference group.

^bRow percentages; all others are column percentages.

^cP<0.05 for comparison between THA-AKI, 2HA-AKI and HA-AKI (Dunnett test for multiple comparisons) or Chi Squared test or proportions.

^dRace or ethnic group was self-reported, and further classified as black or non-black.

^eeGFR calculated with minimum serum creatinine

There were differences in the Charlson co-morbidities between AKI subgroups, especially for diabetes, congestive heart failure, and acute myocardial infarction. The prevalence of chronic renal disease was higher, which was reflected in lower baseline eGFR values for each patient. The inpatient mortality rates were: 1.7% for No-AKI; 5.1% for THA-AKI; 12% for 2HA-AKI; and 18% for HA-AKI.

Figure 1 illustrates the time-course of sCr changes for the AKI categories. HA-AKI, 2HA-AKI and THA-AKI were defined based on the timing of the peak and minimum sCr; the peak sCr preceded the minimum sCr for THA-AKI (blue) and 2HA-AKI (purple), while the minimum sCr preceded the peak sCr for HA-AKI (red). The inset shows elapsed time (hours) relative to the estimated time of admission.

Figure 1. Time-Course for acute kidney injury (AKI) types.

The analysis pertains to 50,601 patients with a single admission to the University of Alabama at Birmingham Hospital between October 1, 2009 and September 30, 2013; first, peak, minimum and last serum creatinine values, connected by dashed lines, stratified by AKI type. Abbreviations: #1, first serum creatinine recorded for admission; AKI, acute kidney injury; 2HA-AKI, transient hospital associated AKI that resolves, followed by a second episode of AKI; THA-AKI, transient hospital associated AKI, defined as peak sCr ≥0.3 mg/dL above the minimum sCr with peak sCr preceding the minimum sCr; HA-AKI, hospital-acquired AKI, defined as peak sCr ≥0.3 mg/dL above the minimum sCr with minimum sCr preceding the peak sCr; last, last serum creatinine recorded for admission; No-AKI, no AKI during admission, defined as peak sCr <0.3 mg/dL above the minimum sCr; sCr, serum creatinine (mg/dL). The elapsed hours relative to admission are listed in the inset.

For THA-AKI and 2HA-AKI, the peak sCr was recorded 38 ± 47 and 30 ± 36 hours after admission, while the minimum sCr values were observed at for 115 ± 88 and 136 ±97 hours after admission. The peak sCr for HA-AKI occurred at 127 ± 105 hours, and followed the minimum sCr recorded at 42 ± 59 hours. For all AKI categories, the discharge time was 7 to 10 hours after the last recorded sCr (Figure 1 insert).

The distinguishing characteristic of the 2HA-AKI patients was progression to a secondary peak sCr (peak2) that occurred after they had recovered from the initial elevation of sCr present at admission (purple line, Figure 1). The peak2 sCr value needed for classification as 2HA-AKI was defined as ≥0.3 mg/dL above the individual minimum sCr, and occurred at 219 ± 126 hours (136±97 hours) for the 2HA-AKI patients. As seen in Table 1, the baseline characteristics of these patients were more similar to the HA-AKI patients than the THA-AKI patients, and they were more likely to receive inpatient RRT, had longer lengths of stay, and spent more days in an ICU than the other categories. Despite these differences, the 2HA-AKI patients had better short-term mortality outcomes than the HA-AKI patients (Table 2).

Table 2.

Characteristics of 50,580 Inpatients (2,500 deaths (4.9%)) without, or with Acute Kidney Injury, Including Serum Creatinine Values (mg/dL)

	AKI Category
	No-AKI: 29,989 (59%)b	THA-AKI: 12,101 (24%)b	2HA-AKI: 2,243 (4%)b	HA-AKI: 6,247 (12%)b
LOS (days)	3.7±3.2	6.9±5.2	12±6.5	8.1±6.1
Age >Median ( 55 years)	14,262 (48%)	6,666 (55%)	1,268 (57%)c	3,798 (61%)
HCF Transfers	6,336 (21%)	3,237 (25%)	426 (33%)	2,005 (32%)c
Baseline eGFR < 60e	2,494 (8.3%)	2,290 (19%)	469 (21%)	1,447 (23%)c
First sCr	0.9±0.3	1.7±1.6	2.1±2.4c	1.2±0.8c
Minimum sCr	0.8±0.3	1.0±0.7	1.1±1.0c	1.1±0.7c
Peak sCr	1.0±0.3	1.9±1.7	2.3±2.6c	2.0±1.5c
2nd Peak sCr		0.9±0.4	1.7±1.5
Last sCr	0.9±0.3	1.0±0.7	1.5±1.4c	1.6±0.3c
sCr: Number/day	1.4±0.7	1.5±0.8	1.6±0.9	1.4±0.9
Percent Recovery (%)	N/A	88±18	100	35±35
> 75% Recoveryd	N/A	9,643 (80%)	100	1,140 (18%)
<33% Recovery	N/A	66 (0.6%)	0	3,232 (52%)
>75% Recovery2	N/A		108 (4.8%)
<33% Recovery2	N/A		1,594 (71%)
Charlson Index	1 (0 – 2)	1 (0 – 2)	1 (0 – 2)	2 (0 – 2)
Inpatient RRT	0	287 (2.4%)	246 (11%)	256 (4.1%)
Total ICU Days	4.8±3.6	9.2±6.0	13±6.6c	9.8±6.6c
Inpatient Deaths (%)	506 (1.7%)	606 (5.0%)	267 (12%)	1,121 (18%)c
IR: Death	0.128 (0.117–0.140)	0.205 (0.189 – 0.222)	0.278 (0.247 – 0.314)	0.622 (0.587 – 0.660)
IRR: Deaths	1.00 (Reference)	1.60 (1.42–1.80)	2.18 (1.88–2.52)	4.86 (4.38–5.40)c
HR for Deathf	1.00 (Reference)	1.23 (1.09–1.40)	1.27 (1.07–1.51)c	2.24 (1.99–2.51)c

Unless otherwise indicated, values are given as number (percentage) or means ± 1 SD, medians are given with (25^th –75^th-centile).

Abbreviations: AKI, acute kidney injury; 2HA-AKI, second episode of AKI in a the course of a single hospital admission; Baseline eGFR<60, baseline eGFR calculated from minimal sCr; risk category defined by patients with baseline eGFR <60 m/min/1>73 m²; eGFR, estimated glomerular filtration rate; HA-AKI, hospital-associated AKI; HR, Hazard Ratio; ICU, intensive care unit; IRR, incident rate ratio (incident rates express as events per 28 patient-days); LOS, length of stay (days); RRT, inpatient renal replacement therapy; sCr, serum creatinine; THA-AKI, transient hospital-associated AKI.

^a P<0.001 for comparisons of characteristics between each group by ANOVA for means, signed rank test for medians, and Chi Squared test for proportions; shown as bold text, relative to No-AKI which served as the reference group.

^bRow percentages; all others are column percentages.

^cP<0.05 for comparison between THA-AKI and HA-AKI (Dunnett test for multiple comparisons) or Chi Squared test of proportions.

^dPercent recovery calculated as 100*(peak sCr – minimum sCr)/(peak sCr – last sCr)

^eeGFR calculated with minimum serum creatinine

^fHazard Ratio (HD) for inpatient mortality (Cox Model; adjusted for median Age, Gender, Race, Charlson comorbidity index, Admission Source, eGFR<60, percent recovery to 75%, percent recover <35%, and inpatient renal replacement therapy).

Table 2 summarizes recovery for patients with THA-AKI, 2HA-AKI and HA-AKI. By definition, 100% of the 2,243 2HA-AKI patients recovered from the initial peak sCr to the minimum sCr, and thereafter developed a subsequent elevation of sCr that was ≥0.3 mg/dL above their individual minimum sCr value. Of the 12,101 THA-AKI patients, 80% had >75% decrease from the peak sCr compared to their last sCr, and <1% had less than 33% improvement from their peak sCr value. In contrast, the improvement from the secondary peak sCr for 2HA-AKI patients, and from peak sCr for HA-AKI patients was blunted, with < 20% of these patients showing 75% improvement from their peak sCr values. The majority of 2HA-AKI (71%) and HA-AKI (52%) patients had less than modest improvement (>33%) from their peak sCr values.

The relevant sCr values for each AKI category are summarized in Table 2, as well as other baseline characteristics and inpatient mortality comparisons. For those patients who received care in an intensive care unit, the average number of days in those units was 4.8±3.6 days for patients with No-AKI, compared to >9 days for patients with AKI. The inpatient mortality rates (No-AKI, 1.7%; THA-AKI, 5%; 2HA-AKI, 12%; HA-AKI, 18%) increased in parallel with the number of ICU days, as well as length of stay across the AKI categories. The last row in Table 2 shows the hazard ratios for the AKI categories, obtained with Cox multivariable regression analysis.

Table 3 summarizes the Cox proportional hazard models for inpatient mortality, with sequential addition of covariates and analysis of their time-dependence, as reflected in the correlation (rho) of the smoothed Schoenfeld residuals with time and the significant interactions with ln (length of stay). When the time-dependent covariates were include in the Cox model, the hazard ratio for THA-AKI was 1.088 (95% CI: 0.952 – 1.242; P=0.217) and the hazard ratio for 2HA-AKI was 1.015 (95% CI: 0.84 – 1.223; P=0.868). The hazard ratio for HA-AKI remained significantly different from zero and the other AKI hazard ratios (Table 3). Even though the time-averaged hazard ratio for THA-AKI and 2HA–AKI were not significantly from the reference category (No-AKI), there was a significant interaction between 2HA-AKI and elapsed time in the hospital.

Table 3.

Cox Multiple Regression Models: Univariate Analysis

Univariate Analysis	HR	Lower 95% CI	Upper 95% CI	P Value	Rho (P Value)	rIDI (%)
NoAKI	1.00	-	-
THA-AKI	1.38	1.22	1.55	<0.001	0.03 (0.102)
2HA-AKI	1.56	1.34	1.82	<0.001	0.06 (0.003)
HA-AKI	3.86	3.45	4.30	<0.001	−0.05 (0.006)
Age> Median	1.91	1.76	2.08	<0.001	0.06 (0.006)	22.5%
Male Gender	1.03	0.95	1.12	0.431	−0.01 (0.818)	−0.1%
Black Race	0.98	0.90	1.07	0.684	0.02 (0.301)	0.1%
HCF Transfer	1.70	1.57	1.84	<0.001	−0.16 (<0.001)	8.3%
eGFR<60	3.59	3.31	3.90	<0.001	−0.22 (<0.001)	18.2%
Charlson Index	1.82	1.72	1.92	<0.001	0.03 (0.205)	13.3%
Recovery <75%	1.76	1.63	1.91	<0.001	−0.12 (<0.001)	5.1%
Recovery2 <75%	0.49	0.24	0.97	0.042	0.04 (0.039)	0.1%
Inpatient RRT	3.44	3.06	3.87	<0.001	−0.02 (0.221)	−5.9%

Time-Dependent Hazard Ratio Analysis
Time-dependent Hazard Ratios	Main Analysis				Interactions with ln(time)
	HR	Lower 95% CI	Upper 95% CI	P Value	HR	95% CI	P Value
AKI	1.29	1.21	1.39	<0.001	1.00	0.96 - 1.03	0.860
Age> Median	0.78	0.65	0.93	0.006	1.32	1.20 - 1.45	<0.001
Male Gender	1.12	0.96	1.33	0.139	0.96	0.88 - 1.04	0.285
Black Race	0.92	0.77	1.10	0.378	1.02	0.93- 1.12	0.647
HFC Transfer	2.42	2.06	2.84	<0.001	0.76	0.70 - 0.82	<0.001
eGFR60	6.86	5.69	8.26	<0.001	0.51	0.47 - 0.57	<0.001
Charlson Index	1.02	0.92	1.14	0.705	1.20	1.13 - 1.28	<0.001
Recovery <75%	2.49	2.09	2.95	<0.001	0.79	0.72 - 0.86	<0.001
Recovery2 <75%	0.02	0.01	1.40	0.072	3.45	0.8 - 14.9	0.095
Inpatient RRT	1.18	0.87	1.61	0.282	1.30	1.14 - 1.49	<0.001

Adjusted for Time-Dependent Covariates
	HR	Lower 95% CI	Upper 95% CI	P Value	Rho (P Value)
No AKI	1.00	Reference
THA-AKI	1.088	0.952	1.242	0.217	0.031 (0.123)
2HA-AKI	1.015	0.844	1.223	0.868	0.055 (0.004)
HA-AKI	2.120	1.881	2.389	<0.001	−0.021 (0.286)

Abbreviations: AKI, acute kidney injury; 2HA-AKI, second episode of AKI in a the course of a single hospital admission; Age> median; risk category for age > 55 years (median value for entire cohort); CI, confidence interval; eGFR<60, baseline eGFR calculated from minimal sCr; risk category defined by patients with baseline eGFR <60 m/min/1>73 m²; HA-AKI, hospital-associated AKI; HCF Transfer, patient admitted to the University of Alabama Hospital for another health care facility; HR, Hazard Ratio; ICU, intensive care unit; IRR, incident rate ratio (incident rates express as events per 28 patient-days); LOS, length of stay (days); Rho, correlation coefficient for smoothed Schoenfeld residuals with time; rIDI, relative integrated discrimination improvement; RRT, inpatient renal replacement therapy; sCr, serum creatinine; THA-AKI, transient hospital-associated AKI

P<0.001 for comparisons of characteristics between each group by ANOVA for means, signed rank test for medians, and Chi Squared test for proportions; shown as bold text, relative to No-AKI which served as the reference group.

Percent recovery calculated as 100*(peak sCr – minimum sCr)/(peak sCr – last sCr)

Criteria for Inclusion of Covariates: Age, Race, Gender and others with: univariate HR >3.0; significant correlation coefficient (rho) for Schoenfeld residual (P<0.05); or rIDI >10%.

Figure 2 shows the survival functions for the AKI categories with point-wise 95% CI for the survival curves [17]. While the survival curve for HA-AKI clearly separated from the other categories early during the course of hospitalization, the survival curves for THA-AKI and 2HA-AKI only became evident for patients with prolonged length of stays. The curves for THA-AKI and 2HA-AKI were similar over time.

Figure 2. Survival curves for acute kidney injury (AKI) types.

The analysis pertains to 50,601 patients with a single admission to the University of Alabama at Birmingham Hospital between October 1, 2009 and September 30, 2013. The survival curves were generated from Cox regression models, using point-wise confidence intervals for the covariate-adjusted survivor functions generated from Cox regression analysis [17].

Abbreviations: #1, first serum creatinine recorded for admission; 2HA-AKI, hospital-associated AKI, defined by second peak sCr ≥0.3 mg/dL above the minimum sCr with second peak sCr following the minimum sCr; AKI, acute kidney injury; HA-AKI, hospital-acquired AKI, defined as peak sCr ≥0.3 mg/dL above the minimum sCr with minimum sCr preceding the peak sCr; last, last serum creatinine recorded for admission; No-AKI, no AKI during admission, defined as peak sCr <0.3 mg/dL above the minimum sCr; sCr, serum creatinine (mg/dL); THA-AKI, transient-associated AKI, defined by peak sCr ≥0.3 mg/dL above the minimum sCr with peak sCr preceding the minimum sCr;

The patients at risk for each category relative to admission date are listed in the inset.

Discussion

We describe a retrospective analysis with 4 years ascertainment of all initial admissions to a single large academic referral medical center, and have used the timing of peak and minimal sCr values to distinguish between transient and more prolonged AKI during the hospital course. Over 40% of the admissions presented with AKI at the time of admission, or developed AKI after being admitted UAHB. The 24% of all qualifying admissions classified as TH-AKI (including the subset (4%) of patient who presented with TH-AKI, but experienced re-onset of HA-AKI) accounted for 70% of all cases with AKI during the ascertainment interval.

Overall, our findings confirm the previous description of “transient azotaemia“ [10]. We also describe a subset of patients with THA-AKI who develop a second episode of AKI during the same admission, which we have termed “2HA-AKI”. This subgroup had increased inpatient mortality rates compared to THA-AKI patients who returned and remained at the minimum sCr, but lower mortality rates that the larger group of patients with HA-AKI. When adjusted for length of stay and other covariates, the hazard rates for inpatient mortality were identical for the THA-AKI and 2HA-AKI groups, and significantly less than the HA-AKI group. The mortality risk for the 2HA-AKI group may reflect the case-mix, comorbidities and other baseline characteristics and needs to be assessed in other independent datasets.

The subgroup with 2HA-AKI was not addressed in the previous description of transient azotaemia, although a group with THA-AKI who did not improved by the third hospital day was noted and reclassified as “persisting ATN” [10]. We suspect that this group with poor recovery from THA-AKI may well have been the group we now characterized as 2HA-AKI. In future studies, it may be informative to evaluate these classifications of inpatient AKI and the risks for developing AKI as well as all-cause mortality during subsequent hospitalizations [20].

There are different assumptions that may account for the difference in apparent incidence of AKI and the extent of improvement from peak sCr. Uchino et al. defined transient azotemia as resolving within 3 days without renal replacement therapy, while patients receiving renal replacement therapy were classified as persisting ATN [10]. We have not used extent of recovery in the present definitions of THA-AKI, 2HA-AKI and HA-AKI, and observed that 17% of patients with HA-AKI had significant improvement (≥80%) from their peak sCr values before hospital discharge, and thus did not have persisting AKI. Sensitivity analysis showed that exclusion of patients receiving renal replacement therapy did not change the estimates of recovery from AKI (data not shown); rather than reclassifying these patients as HA-AKI, we simply included inpatient renal replacement therapy as a covariate in the time-to-event regression model. The major reason for the apparent increased incidence of AKI in our series was the use of a lower threshold (0.3 mg/dL) for diagnosing AKI [9,14], rather than the higher 0.5 mg/dL used by Uchino et al. [10]. Neither study used prior measurements of sCr as baseline values; we used the minimal sCr for each admission, and Uchino et al.[10] Imputed the baseline sCr based on an estimated eGFR of 75 ml/min/1.73 m² for two-thirds of their patients, an approach likely to overestimate that incidence of AKI [7,8], and thus does not account for the increased incidence of AKI that we describe in the present analysis.

The definition of baseline kidney function is somewhat controversial even with the availability of previous ambulatory determinations of serum creatinine [9,21,22]. Using a pre-admission ambulatory sCr as the baseline has been advocated, with a number of surrogates considered when ambulatory data is not available [21]. Siew et al. [7,8] defined the most recent ambulatory sCr measurement as the baseline sCr, and found AKI incidence rates were underestimated for adults when the first admission sCr was used for the baseline (14%), compared to the AKI incidence rate of 26% based on the last pre-admission sCr. The minimum sCr during admission performed better as a surrogate baseline sCr than the first measured sCr, which may reflect a number of cases with THA-AKI included in their analyses. The misclassification and overestimation of the incidence of AKI that occurred when the minimum sCr was used as the surrogate baseline sCr was most evident in the mildest stages of AKI (i.e., Stage 1).

There are two limitations of the current definitions of baseline sCr that are worth considering [23]: 1) AKI is a global health issue [24,25], and the majority of patients with AKI in developing countries have never had a previous sCr measurement; and 2) Prehospitalization sCr values when available are only relevant to the magnitude of the baseline sCr, but do not address the timing of the baseline sCr, which is the distinctive feature of the current analysis. Not only was the elapsed time from admission to the minimum sCr measurement significantly different, but there was also a statistically significant difference in the magnitude of the minimum sCr for patients with THA-AKI compared to HA-AKI. This observation questions whether there is a single baseline creatinine that is appropriate for all forms of AKI.

Finally, we report that there is only a modest risk for inpatient mortality associated with THA-AKI and 2HA-AKI (Figure 2). If the regression analyses used time-averaged covariates, then the significance of the associate of inpatient mortality with THA-AKI and 2HA-AKI was greatly attenuated (Table 3). The time-dependence of hazards associated with AKI has been described in several recent publications [26-28]. Even though the hazard ratio for inpatient mortality was identical for THA-AKI and 2HA-AKI, the longer-term outcomes may diverge, especially noting that the THA-AKI had nearly complete recovery from AKI, while the 2HA-AKI group had severely reduced recovery from AKI, similar to the HA-AKI group.

Other reports have described AKI acquired before admission, which resolved in the hospital setting as “community-acquired” AKI[22,29-33]. We have operationally defined THA-AKI as cases where the peak sCr values precede the subsequent minimal sCr for that admission, and surmise that the rapid decrease in sCr may have reflected dehydration prior to admission that was promptly and appropriately corrected with fluid administration. Of the 14,344 patients with THA-AKI and 2HA-AKI in our analysis, 10,681 (75%) were admitted to UAHB from a community-based setting, but 25% were transferred/admitted to UAHB from other hospitals and health-care facilities, so THA-AKI and community-acquired AKI are not entirely synonymous. In contrast, a higher percentage of HA-AKI patients (32%) were transferred to UAHB from other health care facilities.

Others have described inpatient recovery of AKI but the criteria for recovery have not been standardized [34]. Recovery to the baseline serum creatinine measured before AKI developed has been used [5], often with the focus on severely ill patients in the critical care setting [35]. We report that the average improvement in elevated sCr for patients with THA-AKI was 88±18%, with 80% achieving 75% improvement from the peak sCr before hospital discharge. In contrast, only 4.8% of the 2HA-AKI and 17% of the HA-AKI achieved a 75% reduction from the peak sCr before hospital discharge.

Our study has several limitations, and is based on retrospective analyses of an administrative data set reflecting the 4-year experience at a single tertiary academic referral medical center. Information about pre-hospitalization serum creatinine was not available, nor can the etiology of AKI, urine output or fluid administration be determined from the administrative dataset. Follow-up status after discharge was not included, and many patients, especially with HA-AKI and 2HA-AKI were discharged before they may have had more complete recover of their renal function. We cannot describe their subsequent course due to the limitations of our data set. The density and timing of inpatient serum creatinine data reflect case ascertainment that is dependent on clinical decision-making and laboratory testing ordering practices [3], which introduces several sources of potential confounding and bias [36].

Changes in serum creatinine depend on factors that modify creatinine generation (body composition, muscle wasting, and acute muscle injury), [37] and changes in the volume of distribution of creatinine (dehydration or volume expansion) may occur, especially in elderly [11] and critically ill patients with prolonged hospitalizations [38,39]. Most admissions were not associated with significant changes (≤0.3 mg/dL) in serum creatinine; these admissions were defined as the reference group with No-AKI for these analyses. We do not attribute the observed decreases in sCr to changes in creatinine generation rate because we censored each case at 28 days length of stay, and the majority of the cases of THA-AKI were not exposed to an intensive care setting. Nevertheless, there is certainly a subset of critically ill patients who may have decreased creatinine generation rates, and net cumulative fluid balance, which could dilute their serum creatinine values [40].

The approach described herein is readily applicable to retrospective analyses of large administrative inpatient data sets, and does not used diagnosis related groups to define AKI categories. Further work is be needed to directly compare the minimum inpatient sCr with previous values, to determine if the minimum inpatient sCr value can be used as a real-time tool in assessing the development and recovery from AKI during an acute hospitalization, and to compare the minimum inpatient sCr value to sCr values obtained for patients with previous hospital admissions.

In conclusion, we conducted a retrospective analysis of a large administrative set from a single academic medical center, and described 3 unique subgroups of AKI that were differentiated by the time course of the changes in sCr with reference to the minimal and maximal values obtained during a single hospitalization. While the associate risk for inpatient mortality was attenuated for THA-AKI and 2HA-AKI, the finding that approximately 20% of patients initially classified as THA-AKI experienced a second episode of AKI suggests that continued surveillance of inpatient sCr even after “complete” recovery from THA-AKI would be informative and could potentially lead to clinical initiatives targeted at improving outcomes for this high risk group of patients. Finally, approximately 20% of patients with HA-AKI have significant improvement from their peak sCr values during the hospitalization. More information is need to better characterize recovery from peak sCr in patients with AKI, with the goals of better surveillance and implementation of pro-active treatment programs that could positively impact on patient outcomes and length of stay [23].