To the Editor:
An important methodological issue concerning acute kidney injury (AKI) definitions
1
is the choice of “baseline” serum creatinine (SCr).2, 3, 4 The most recent consensus
definition proposes a rolling 48-hour window for AKI ascertainment during hospitalization,
or the use of a baseline value that is “known or presumed to have occurred in the
past 7 days.”
1
However, significant misclassification in assigning AKI status can occur when admission
or nadir inpatient SCr (as has been done in a number of studies) is used rather than
a preadmission outpatient baseline.
4
A well-recognized concern with the use of admission SCr to define baseline kidney
function is that it will be higher than a patient’s true baseline if community-acquired
AKI is present, and therefore community-acquired AKI will be missed if the admission
SCr is used to define baseline. However, animal and human studies have recently shown
that creatinine generation can also quickly fall with acute illness, so falsely low
readings may result.5, 6 It is unknown whether changes in creatinine generation affect
AKI ascertainment. Therefore, to quantitate variation in first inpatient SCr level
and the impact on AKI ascertainment (Figure 1a), we compared preadmission baseline
and first inpatient SCr in a large, population-based, hospitalized cohort. We also
identified predictors of lower first inpatient SCr.
Figure 1
(a) Potential serum creatinine (SCr) trajectories and acute kidney injury (AKI) misclassification.
The lack of a preadmission baseline SCr may lead to a failure to recognize AKI (green
line); in this case, the first admission SCr is used as “baseline,” and criteria for
AKI are not met despite the fact that the individual has community-acquired AKI. The
use of nadir SCr (blue line) or first inpatient (red line) in the absence of a known
baseline may lead to misclassification as AKI when no AKI is present. (b) Lower first
inpatient SCr may lead to misclassification. Here, the use of first inpatient SCr
(red line) in the absence of a known baseline may lead to misclassification as AKI
when no AKI is present. The use of first inpatient SCr (orange line) may also lead
to misclassification of AKI severity.
We identified all hospitalized adults without end-stage renal disease at 21 Kaiser
Permanente Northern California hospitals between 2006 and 2011 (Supplementary Figure S1);
only the first eligible hospitalization per subject was included. Kaiser Permanente
Northern California is a large integrated health care delivery system caring for > 4.1
million persons in the San Francisco Bay Area that is highly representative of the
statewide population.
7
The study was approved by the institutional review boards of the Kaiser Foundation
Research Institute and the University of California, San Francisco.
Baseline SCr was the most recent outpatient SCr from a maximum of 365 days and a minimum
of 7 days preadmission.
8
We selected this as the gold standard because this definition has been used in prior
studies examining the impact of baseline SCr on AKI ascertainment, including the prospective
Assessment, Serial Evaluation, and Subsequent Sequelae of Acute Kidney Injury (ASSESS-AKI)
study.4, 8 A peak inpatient SCr ≥ 50% relative, ≥ 0.3 mg/dl absolute increase from
the outpatient baseline, or need for acute dialysis defined AKI for this analysis.
1
Covariates included demographics, comorbidities, severity of illness,
9
preadmission estimated glomerular filtration rate (eGFR), and proteinuria. Comorbidities
(diabetes, hypertension, cancer, coronary disease, chronic heart failure, prior ischemic
stroke) were ascertained for up to 5 years before hospitalization using previously
validated methods based on inpatient and ambulatory diagnoses and procedures, laboratory
results, and pharmacy databases (codes available upon request).10, 11 We identified
coronary revascularization, sepsis, and acute heart failure occurring during the index
hospitalization using relevant diagnosis and procedure codes. To further describe
acute severity of illness, we determined whether patients were admitted to the intensive
care unit during their stay and calculated the Laboratory-based Acute Physiology Score
(LAPS) and COmorbidity Point Score (COPS), along with a validated predicted mortality
score based on automated inpatient, outpatient and laboratory data.
9
Preadmission eGFR (in ml/min per 1.73 m2) was estimated using the Chronic Kidney Disease
Epidemiology Collaboration (CKD-EPI) equation
12
and baseline SCr information described previously. We ascertained proteinuria based
on a urine dipstick result of 1+ or greater (in the absence of a concomitant urinary
tract infection).
10
The first inpatient SCr was expressed as a percentage change compared with baseline
SCr. We used multivariable logistic regression to identify predictors of a first inpatient
SCr that was < 90% of baseline SCr stratified by AKI status. Analyses were conducted
using SAS version 9.3 (SAS Institute, Cary, NC).
Of 214,802 eligible hospitalizations, 37,827 (17.6%) met our AKI criteria. AKI was
associated with a higher prevalence of sepsis (20% vs. 7%), diabetes mellitus (40.3%
vs. 26.3%), hypertension (79.5% vs. 65.5%), and chronic kidney disease (51.6% vs.
28.8%) (all P < 0.001) (Supplementary Table S1). Among all patients, 21.7% had a first
inpatient SCr that was ≥ 110% above outpatient baseline (Table 1 and Figure 2). Not
surprisingly, a greater proportion of patients with AKI (74.6%) experienced this pattern.
Figure 2
Change between baseline and first inpatient serum creatinine (SCr) concentration,
overall and stratified by acute kidney injury (AKI) status. Each column represents
the proportion of individuals meeting those criteria.
Table 1
Distribution of change between baseline and first inpatient serum creatinine (SCr)
concentration, overall and stratified by acute kidney injury (AKI) statusa
Ratio between first inpatient and outpatient baseline SCr
Overall(N = 214,802)
AKIb
(n = 37,827)
No AKI(n = 176,975)
< 0.70
17,573 (8.2)
488 (1.3)
17,085 (9.7)
0.70–0.79
32,351 (15.1)
1056 (2.8)
31,295 (17.7)
0.80–0.89
47,202 (22.0)
1978 (5.3)
45,224 (25.6)
0.90–0.99
40,338 (18.8)
2618 (6.9)
37,720 (21.3)
1.00–1.09
30,745 (14.3)
3455 (9.1)
27,290 (15.4)
1.10–1.19
15,379 (7.2)
3459 (9.1)
11,920 (6.7)
1.20–1.29
8876 (4.1)
4015 (10.6)
4861 (2.7)
1.30–1.39
5119 (2.4)
3828 (10.1)
1291 (0.7)
1.40–1.49
3472 (1.6)
3183 (8.4)
289 (0.2)
1.50–1.99
7283 (3.4)
7283 (19.3)
0 (0.0)
≥ 2.00
6464 (3.0)
6464 (17.1)
0 (0.0)
a
Overall P value < 0.001.
b
It should be noted that among those with AKI, a ratio between first inpatient and
outpatient baseline SCr > 1.1 may or may not meet criteria for AKI. For example if
the baseline SCr is 1.0 mg/dl, 110% of baseline would be 1.1 mg/dl and would not meet
criteria for AKI; such an individual might have evolving AKI and a subsequent rise
in SCr that meets criteria for AKI. In contrast, if the baseline SCr is 3.1 mg/dl,
110% of baseline would be 3.41 mg/dl, and this individual would meet criteria for
AKI, which would be community acquired.
In all, 45% of the patients had a first inpatient SCr that was < 90% of the outpatient
baseline; 9.4% of those with AKI experienced this pattern. Regardless of AKI status,
older age, history of cancer, and intensive care unit admission were associated with
a first inpatient SCr that was < 90% of outpatient baseline SCr (Table 2). Patients
with diabetes mellitus, hypertension, sepsis, and greater severity of illness (measured
by predicted mortality) were less likely to have a low first inpatient SCr, reflecting
the fact that many of these patients present with AKI in evolution. It should be noted
that several of these factors (e.g., sepsis and acute illness) are associated with
acute reductions in creatinine generation, so the severity of AKI may be difficult
to ascertain based on changes in SCr alone. In the future, novel biomarkers or real-time
measurement of GFR may better define AKI severity.
Table 2
Correlates of having a first inpatient serum creatinine (SCr) value < 90% of outpatient
baselinea
Characteristic
AKI(n = 37,827)
No AKI(n = 176,975)
Age, yr
< 45
REF
REF
45–74
1.53 (1.28–1.84)
1.33 (1.28–1.37)
≥75
2.06 (1.71–2.50)
1.37 (1.32–1.43)
Male gender
0.95 (0.88–1.02)
0.85 (0.83–0.87)
Race/ethnicity
White
REF
REF
Black/African American
1.03 (0.92–1.16)
0.88 (0.85–0.91)
Asian/Pacific Islander
1.12 (1.00–1.26)
0.97 (0.94–1.00)
Other/unknown
199 (0.5)
891 (0.5)
Medical history
Diabetes mellitus
0.82 (0.76–0.89)
0.93 (0.91–0.95)
Hypertension
0.90 (0.82–0.99)
0.92 (0.90–0.94)
Systemic cancer
1.26 (1.15–1.37)
1.06 (1.04–1.09)
Coronary heart disease
1.12 (0.97–1.28)
1.01 (0.96–1.05)
Chronic heart failure
0.99 (0.87–1.13)
0.91 (0.87–0.96)
Ischemic stroke
0.92 (0.71–1.19)
1.11 (1.03–1.19)
During index hospitalization
Coronary revascularization
1.43 (1.24–1.63)
0.84 (0.80–0.88)
Sepsis
0.88 (0.80–0.97)
0.67 (0.65–0.70)
Heart failure
0.91 (0.78–1.05)
0.50 (0.48–0.53)
Admitted to intensive care unit
2.33 (2.15–2.52)
1.26 (1.23–1.29)
Predicted Mortality Score category
< 0.1%
REF
REF
0.1 to < 0.5%
1.08 (0.87–1.35)
0.94 (0.90–0.97)
0.5 to < 2%
1.04 (0.85–1.28)
0.89 (0.85–0.92)
2 to < 5%
0.83 (0.68–1.03)
0.84 (0.80–0.87)
5 to < 10%
0.61 (0.49–0.76)
0.74 (0.70–0.77)
10 to < 15%
0.44 (0.34–0.57)
0.70 (0.65–0.74)
15 to < 30%
0.39 (0.30–0.50)
0.70 (0.66–0.75)
≥ 30%
0.23 (0.16–0.33)
0.54 (0.48–0.61)
Unknown
0.81 (0.64–1.04)
0.80 (0.76–0.84)
Prior documented proteinuria
0.94 (0.84–1.04)
0.87 (0.84–0.90)
Outpatient baseline eGFR
≥ 60 ml/min per 1.73 m2
1.32 (0.81)
1.00 (0.39)
45–59 ml/min per 1.73 m2
1.03 (0.94–1.14)
1.38 (1.34–1.42)
30–44 ml/min per 1.73 m2
1.03 (0.92–1.14)
1.52 (1.47–1.57)
<30 ml/min per 1.73 m2
0.87 (0.76–0.99)
1.57 (1.49–1.67)
eGFR, estimated glomerular filtration rate; REF, reference.
Boldface data represent statistically significant associations.
a
Multivariable logistic regression was used to identify predictors of a first inpatient
SCr < 90% of baseline and are expressed as odds ratios (95% confidence intervals).
Our results have important implications for AKI ascertainment. Of the patients, 45%
had a first inpatient SCr that was < 90% of the outpatient baseline. Consequently,
using the first inpatient SCr in place of the outpatient baseline may misclassify
some individuals as having AKI when no AKI is, in fact, present (as the SCr rises
back to the actual baseline) (Figure 1b). In fact, in our study population, 6605 individuals
would have met AKI criteria, had the first inpatient SCr been used in place of the
outpatient baseline (Figure 3). The inclusion of patients without actual AKI because
of this misclassification can bias associations between AKI and true AKI risk factors
towards the null. Even among those who met our criteria for AKI, nearly 1 in 12 had
a first inpatient SCr that was < 90% of outpatient baseline; here, using an inpatient
SCr that is lower than outpatient baseline may lead to misclassification of AKI severity.
In contrast, using the outpatient baseline SCr identified 21,864 individuals who were
not identified as having AKI using the first inpatient SCr, likely due to the presence
of community-acquired AKI. A total of 15,963 individuals were identified as having
AKI using either the outpatient baseline or first inpatient SCr. Thus, the overall
incidence of AKI was higher when the outpatient baseline SCr was used (37,827 vs.
22,568).
Figure 3
Differences in acute kidney injury (AKI) ascertainment when the outpatient baseline
or first inpatient serum creatinine (SCr) are used to ascertain AKI status. A total
of 37,827 individuals met criteria for AKI using the outpatient baseline SCr as described
(red circle); of these individuals, only 15,963 would have been identified, had the
first inpatient SCr been used to define baseline due to the presence of AKI at hospital
admission (overlap between blue and red circle). Due to the variation described in
this analysis (with a large proportion of individuals presenting with an SCr below
their outpatient baseline), an additional 6605 individuals would have been misclassified
as having AKI, had the first inpatient SCr been used to define AKI.
We note that, in some populations (i.e., those of older age and with cancer), the
first inpatient SCr was more likely to be <90% of preadmission baseline, as might
be expected, as these conditions are more likely to be associated with reductions
in muscle mass over time. Further studies are needed to better understand whether
these differences reflect ongoing malnutrition due to chronic illness, and, if so,
how best to estimate baseline SCr in these populations. In clinical practice, careful
evaluation of both the outpatient baseline SCr and first inpatient SCr in these populations
who are more likely to have a low first inpatient SCr is warranted to help with the
clinical ascertainment of AKI. Due to the presence of community-acquired AKI at admission,
a significant proportion of patients who met our AKI definition would not have met
criteria for AKI had the first inpatient SCr been used as baseline (Figure 3).
Study strengths include the large number of patients with outpatient baseline SCr
measurements within an integrated health care delivery system, which allowed for analysis
of patient-level factors stratified by AKI status, an important determinant of admission
SCr. Kaiser Permanente Northern California is highly representative of the local and
statewide population, so results are likely more generalizable than those from specialized
populations (e.g., the predominantly male Veterans Affairs population). A potential
limitation is that we focused on individuals with an available outpatient SCr, so
no inferences can be made about how to estimate baseline SCr without outpatient data.
In those circumstances, given the importance of baseline SCr for AKI classification
4
and given the variation between first inpatient and outpatient baseline SCr that we
have shown, clinicians should use all available resources to identify an outpatient
baseline SCr.
Disclosure
All the authors declared no competing interests.