predictors of mortality and the provision of dialysis in patients with

7
BRIEF COMMUNICATION Predictors of Mortality and the Provision of Dialysis in Patients with Acute Tubular Necrosis GLENN M. CHERTOW,* J. MICHAEL LAZARUS,* EMIL P. PAGANINI,t ROBIN L. ALLGREN, RICHARD A. LAFAYETTE, and MOHAMED H. SAYEGH,* FOR THE AURICULIN#{174} ANARITIDE ACUTE RENAL FAILURE STUDY GROUP *Re,icil Division, Department of Medicine, Brigham and Women ‘s Hospital. Harvard Medical School, Boston, Massachusetts; Section of Dialysis (hid Extracorporeal Therapy. Cleveland Clinic Foundation, Cleveland, Ohio; Dis’isioiz of Nephrologv, Department of Medicine, Stanford University Medical Center, Palo Alto, California; (t1(i .Scios, Jar,, !‘ilou,itain Vie%i’, Caiift’r,iia. fl/)%(#{149}.(#{149}#{149}( 1. To explore the lItLlril1 history of critically i II l)atienls with ilc’tlIC relLil IUI lure titie to acute tubular flct..’fl)SIS. we CVlILliltC(1 256 J)al ieflts enrol led in the pIaceh arni of a ran- doiniicd clinical trial . l)eLth and the c()lnpositc OUICOIUC. death or the provision f lIj were determined with lollow-up to 60 d. ‘I’he relative risks ( RR ) and 95X confidence intervals (()5(/( (‘1 ) UssoCialeti with routinely available (len)graphic. ci illical , ilfl(l laboratory variables were esti niated using propor- I it ma I liiarls regression . N I nety -I lii’ee ( 36X ) deaths were doe- uluenteci; an additiotal 52 (20f ) patients who survived re- cci ved dialysis. Predictors of lnort4llity included nude geiler (RR. 2.01 ; t)5(/, (‘J, I .2 1 to 3.3(. oliguria (RR. 2.25; 95% Cl. I .43 to 3.55 ), niec’lianical ventilation ( HR. I .86; 95X (‘1. 1 . I8 IC) 2.93 ), actile niyocardial infarction ( RR. 3. I 4; t)5(% Cl, I .85 to 5.3 I ). acute stroke or seizure (KR. 3.08: 95% Cl, I .56 to 6.06), chronic ilulunosuppressit)n ( RR. 2.37: 95/ (‘I. I . I6 to 4.88 ). hyperhiliruhineinia ( RR. I .06; 95(4 (], I .03 o I .08 per I nig/di increase in total hilirtihii. all letaholie acidosis (RR. 0.’)5; 95(/( CJ 0.9() to (1.99 per I mEq/L increase in serum bicarbonate concentration ). Predictors of death or the provision 01 dialysis were oligtiria (RR. 5,95; 95(4 Cl. 3.96 to 8,95). mechanical ventilation (RR. I .53: 95’7 CI. I .07 to 2.2 1 ). acute Inyocardial inlaretion ( RR. I .95; ‘)5* Cl. I .24 to 3.07 . ar- rhyihlnia (RR. I .5 1 ‘. 95”f (‘1. I .04 to 2. I 9), and hypoalhtIlin- eflula (RR. 0.56: 95(7( Cl. 0.42 to 0.74 per I g/dl increase in sertiin 4ilhuniin concentration). Neither mortality nor the pro- vision Of dialysis was related to patient age. These observations Clfl he used to estiniIe risk early in the course of acute tubular necrosis. Furtherniore, these aliti related niodels may he used to adjust fr case-nh x variat ion in qutli ity ilprovelent efforts, LIfl(.l to objectively strati l’y pat jents in future intervention trials aile(l at favorably altering the course of hospital-acquired acute renal failure. (J Am SOC Nephrol 9: 692-698, 1998) Acute renal failure (ARF) occurs in approxilately 2 to 5* of liospitalii.ed patients (I .2 ). Hospital-acquired ARF is associ- ated with a 25 It) 90/( risk of in-hospital iuortal ity. de)elflliflg. at least IU part. 011 case flu ,( and the severity of renal injury ( 322 ) . A nuniber ol’ previts studies have attelnj)ted to identify clinical risk factors associated with adverse outconies ifl this patient 1)PIIltifl. IViost have been derived Iron) single iflstittltU)fl eXpelience. which Iiiits generalii.ahility. and the niajority live iLItI insufficient s1tiple site to perfi’irni iii1ti- variable analyses. 1-kIrthernore. the lnjority of’ studies were retrosl)ect i ye. and w ithout hI inded recorti reviewers. int1 there- kre subjeci It) illf()rnst ion bias. Vv’e recent ly PLl’t icipated in il placebo-control led. nult icenter R’ceived July 2’). I 1)1)7, Accept’tl ( )oIIher I . I lhi 5’IOFk ‘I’I P1’CS’t1I’(l I Ii Il)SII’b.t I ttiii #{149}Il il)’ 28th AItIiIIaI v1eet PUg I (I The An)L’ricIn S1)tiCIy of Nephrlllogy. Sin I )iego, C‘A. Nivetuhcr S lu 8, I 995 C)(It’S1)(1Ii(l’Ii(.’ III I )r. ( il’iiii 1vI ( ‘heiit. I )ialss I tilt Adini,iiitssi v’ Oi t’ie’. l3riglsiiii intl \Voi’n s I lospilal . 75 Francis Slr’el. l3osion, V1A 02 II I O46607 /‘#)4()(i’)2$() (X)/() .11 1tIrI)Il i d’ Ihe Antericitn Si ciet v iI N’Plir11l( ( yrigl1l () I 1)98 t)Y h’ A l1h’li.(Ii Sis:k’iy ol Nephrolllgy II’kII. the prinlary t*ijcctive of’ which was to evaluate the effect of’ intravenous Auriculinok Anaritide ( synthetic atrial natriuretic peptide. ANP1 l’(, ()l the need for dialysis in 504 people with acute tubular necrosis (ATN) of’ ischemic and/or toxic origin. For the purpose of this report. we restricted our analysis It) placebo recipients (n = 256). This strategy affor(led us the opportunity tC) examine the i nipact of’ deniographic, clinical ( historic and current ). int1 laboratory variables, derived f’roni lBLllt i-illstittltit)llal experience, on the ivitural history of pn’- gressive ARF due to ATN. Materials and Methods Palleits POtI.’flti.tI 5lLid subjects were #{149}ilults with ARF tltic to ATN. who lil a progressive risc in sertun ercatinine concentration of at least I .0 tiig/tIl 0VC 24 II) 48 h. without CVi(Ietlc’e Of recovery or stabili/alion, Microscopic urinary SCdiICIU exLInutlatiC)n, renal ultrasonography. LU1C.l lCtCt’I11i flLII 101) 1)1 I he Itact innul exeret ion of’ soditum weuc OfiiI)t) the diaguiC)stic iiethods tused by investigators IC) identify stuhjecls with ATN. Subjects whose ARF was titue to eatises other than ATN. stich 8% prereuial itotenhia. urinary tract obstruction. glonieruloncphritis. interstit ial nephritis. CIheuoeuiihol Ic liscasc. nialignant hypertension.

Upload: hoangtuong

Post on 21-Jan-2017

216 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Predictors of Mortality and the Provision of Dialysis in Patients with

BRIEF COMMUNICATION

Predictors of Mortality and the Provision of Dialysis in

Patients with Acute Tubular Necrosis

GLENN M. CHERTOW,* J. MICHAEL LAZARUS,* EMIL P. PAGANINI,t

ROBIN L. ALLGREN,� RICHARD A. LAFAYETTE,� and MOHAMED H. SAYEGH,*

FOR THE AURICULIN#{174} ANARITIDE ACUTE RENAL FAILURE STUDY GROUP

*Re,icil Division, Department of Medicine, Brigham and Women ‘s Hospital. Harvard Medical School, Boston,

Massachusetts; �Section of Dialysis (hid Extracorporeal Therapy. Cleveland Clinic Foundation, Cleveland,

Ohio; �Dis’isioiz of Nephrologv, Department of Medicine, Stanford University Medical Center, Palo Alto,

California; (t�1(i�.S�cios, Jar,, !‘ilou,itainVie%i’, Caiift’r,iia.

fl/)%�(�#{149}.(#{149}#{149}(�1. To explore the l�ItLlril1 history of critically i II l)atienls

with ilc’tlIC relLil IUI lure titie to acute tubular flct..’fl)SIS. we

CV�lILliltC(1 256 J)al ieflts enrol led in the pIaceh� arni of a ran-

doiniicd clinical trial . l)eLth and the c()lnpositc OUICOIUC. death

or the provision f l�Ij were determined with lollow-up to

60 d. ‘I’he relative risks ( RR ) and 95�X confidence intervals

(()5(/(� (‘1 ) UssoCialeti with routinely available (len�)graphic.

ci illical , ilfl(l laboratory variables were esti niated using propor-

I it ma I l�iiar�ls regression . N I nety -I lii’ee ( 36�X ) deaths were doe-

uluenteci; an additiot�al 52 (20�f ) patients who survived re-

cci ved dialysis. Predictors of lnort4llity included nude gei�ler

(RR. 2.01 ; t)5(�/�, (‘J, I .2 1 to 3.3(�. oliguria (RR. 2.25; 95% Cl.

I .43 to 3.55 ), niec’lianical ventilation ( HR. I .86; 95�X (‘1. 1 . I 8

IC) 2.93 ), actile niyocardial infarction ( RR. 3. I 4; t)5(% Cl, I .85

to 5.3 I ). acute stroke or seizure (KR. 3.08: 95% Cl, I .56 to

6.06), chronic ilul�unosuppressit)n ( RR. 2.37: 95�/ (‘I. I . I 6 to

4.88 ). hyperhiliruhineinia ( RR. I .06; 95(4 (], I .03 �o I .08 per

I nig/di increase in total hilirtihii�. al�l l�etaholie acidosis (RR.0.’)5; 95(/( CJ� 0.9() to (1.99 per I mEq/L increase in serum

bicarbonate concentration ). Predictors of death or the provision

01 dialysis were oligtiria (RR. 5,95; 95(4 Cl. 3.96 to 8,95).

mechanical ventilation (RR. I .53: 95’7 CI. I .07 to 2.2 1 ). acute

Inyocardial inlaretion ( RR. I .95; ‘)5�* Cl. I .24 to 3.07 . ar-

rhyihlnia (RR. I .5 1 ‘. 95”f (‘1. I .04 to 2. I 9), and hypoalhtIl�in-

eflula (RR. 0.56: 95(7( Cl. 0.42 to 0.74 per I g/dl increase in

sertiin 4ilhuniin concentration). Neither mortality nor the pro-

vision Of dialysis was related to patient age. These observations

C�lfl he used to estin�iIe risk early in the course of acute tubular

necrosis. Furtherniore, these aliti related niodels may he used to

adjust f�r case-nh x variat ion in qutli ity il�provel�ent efforts,

LIfl(.l to objectively strati l’y pat jents in future intervention trials

ail�e(l at favorably altering the course of hospital-acquired

acute renal failure. (J Am SOC Nephrol 9: 692-698, 1998)

Acute renal failure (ARF) occurs in approxil�ately 2 to 5�* of

liospitalii.ed patients ( I .2 ). Hospital-acquired ARF is associ-

ated with a 25 It) 90�/( risk of in-hospital iuortal ity. de�)elflliflg.

at least IU part. 011 case flu ,( and the severity of renal injury

( 3�22 ) . A nuniber ol’ previ�ts studies have attelnj)ted to

identify clinical risk factors associated with adverse outconies

ifl this patient 1)�PIIl�ti�fl. IViost have been derived Iron) single

iflstittltU)fl eXpelience. which Iii�its generalii.ahility. and the

niajority l�ive iLItI insufficient s1ti�ple site to perfi’irni i�ii1ti-

variable analyses. 1-kIrthern�ore. the ln��jority of’ studies were

retrosl)ect i ye. and w ithout hI inded recorti reviewers. �int1 there-

kre subjeci It) illf()rn�st ion bias.

Vv’e recent ly P�Ll’t icipated in il placebo-control led. n�ult icenter

R�’ceived July 2’). I 1)1)7, Accept�’tl ( )oIIher I . I

lhi� 5’IOFk ‘��I’I P1’CS�’t1I�’(l I Ii Il)SII’�b.t I t�tiii #{149}Ilil)�’ 28th AItIiIIaI �v1eet PUg I (I TheAn)L’ricIn S1)tiCIy of Nephrlllogy. Sin I )iego, C‘A. Nivetuhcr S lu 8, I 995

C‘� )(It’S1)(1Ii(l�’Ii�(.’ III I )r. ( il�’iiii 1�vI� ( ‘heiit��. I )ial�ss� I tilt Adini,ii�itssi v�’ Oi�

t’ie�’. l3riglsiiii intl \�Voi��’n � s I lospilal . 75 Francis Slr�’el. l3osion, �V1A 02 I I

I O46�607 �/‘#)4()(i’)2�$�() �(X)/()

.111tIrI)�Il i d’ Ihe Antericitn Si �ciet v i�I N�’Plir11l( �

( ��yrigl1l () I 1)98 t)Y h�’ A l1h’li�.(Ii Sis:k’iy ol Nephrolllgy

II’kII. the prinlary t*ijcctive of’ which was to evaluate the effect

of’ intravenous Auriculinok Anaritide ( synthetic atrial natriuretic

peptide. ANP1� l’(,� ()l� the need for dialysis in 504 people

with acute tubular necrosis (ATN) of’ ischemic and/or toxic

origin. For the purpose of this report. we restricted our analysis

It) placebo recipients (n = 256). This strategy affor(led us the

opportunity tC) examine the i nipact of’ deniographic, clinical

( historic and current ). �int1 laboratory variables, derived f’roni

lBLllt i-illstittltit)llal experience, on the ivitural history of pn’-

gressive ARF due to ATN.

Materials and MethodsPalle�its

POtI.’flti.tI 5lLid� subjects were #{149}i�lultswith ARF tltic to ATN. who

l�i�l a progressive risc in sertun ercatinine concentration of at least I .0

tiig/tIl 0VC� 24 II) 48 h. without CVi(Ietlc’e Of recovery or stabili/alion,

Microscopic urinary SCdiI�CIU exLInutlatiC)n, renal ultrasonography.

LU1C.l�lCtCt’I11i flLII101) 1)1 I he Itact innul exeret ion of’ soditum weuc OfiiI)t)�

the diaguiC)stic �iiethods tused by investigators IC) identify stuhjecls with

ATN. Subjects whose ARF was titue to eatises other than ATN. stich

8% prereuial �itotenhia. urinary tract obstruction. glonieruloncphritis.

interstit ial nephritis. �CIheuoeuiihol Ic liscasc. nialignant hypertension.

Page 2: Predictors of Mortality and the Provision of Dialysis in Patients with

Mortality and Dialysis in Acute Tubular Necrosis 693

renovasculan thrombosis on dissection, on hepatonenal syndrome, were

excluded. Subjects with chronic renal failure (usual serum creatinine

greaten than 3.0 mg/dl). prior renal transplantation. circulatory shock

(systolic BP less than 90 mmHg with presson support). and those in

whom dialysis was anticipated within 24 h or who were deemed

unsuitable candidates for dialysis were also excluded from the study.

Diuretic agents. vasoactive agents, hyperalimentation. and all other

therapies were administered at the discretion of the treating physician.

The need for dialysis and its timing. modality, membrane, duration,

and/on intensity were determined on a case-by-case basis by thetreating nephrologist. Outcomes were assessed for up to 60 d after

randomization. Oligunia was defined as an average urine output of less

than 0.4 L /d at the time of study entry.

Statistical Analyses

Two-by-two contingency tables were evaluated with Fisher’s exact

test. Correlation among variables was described with the Pearson

product moment correlation coefficient. Time-to-event analyses were

performed using proportional hazards regression, with censoring at

day 60 after study entry. Continuous and categorical variables were

examined for univariate associations with mortality and the need for

dialysis. Variables with univaniate associations at the P < 0.05 level

were considered candidates for multivariable analysis. Proportional

hazards regression was used to simultaneously adjust for multiple

covaniates. using stepwise selection and entry and exit criteria set at

the P < 0.05 level (23). Variables not included by the automated

technique were reentered individually to evaluate for residual con-

founding and model fit. Multiplicative interaction terms with oliguria

were considered for all other model covariates. Relative risks (RR)

and 95% confidence limits (95% CI) were calculated based on the

model parameter coefficients and standard errors, respectively. Plots

of log (-log lsurvival rate]) against log (survival time) were per-

formed to establish the validity of the proportionality assumption (24).

Competing models were compared with the log likelihood test. Lo-

gistic regression analysis (yielding odds ratio [OR] as the analogous

effect estimate) was performed for the mortality outcome, using death

at 30 d as the dependent variable. The area under the receiver

operating characteristic (ROC) curve was calculated to assess model

discrimination. There were no missing data for the categorical van-

ables analyzed. Five percent or less of the continuous observations

were missing. Median values of missing continuous variables were

imputed in multivariable models. Statistical analyses were conducted

using SAS (The SAS Institute, Cany, NC). All P values are two-tailed.

ResultsThe mean age of study subjects was 62.0 ± 16.9 yr. Twenty-

five percent of the study sample was age 75 on above. Thirty-

five percent of subjects were women. There were 1 82 (7 1 %)

Caucasian, 43 (17%) African-American, 25 (10%) Hispanic,

and six (3%) people of Asian-American or other race or eth-

nicity. The large majority (85%) of patients were cared for in

intensive care units at the time of study entry. Table I outlines

the baseline clinical characteristics of the study population.

The mean serum ereatinine and urea nitrogen concentrations at

study enrollment were 4.6 ± 2.0 mg/dl and 66.2 ± 28.5 mg/dl,

respectively, indicating a marked degree of renal dysfunction

in the majority of eases.

Three patients (1 %) were lost to follow-up: one each at I I,

22, and 33 d. Ninety-three (36%) deaths were documented over

the 60-d study period. An additional 52 (20%) patients who

survived received dialysis.

Univariate Analyses

The relations among demographic, clinical, and laboratory

variables, and the outcomes of interest (mortality, and the

composite outcome death or the provision of dialysis) were

initially assessed using unadjusted proportional hazards regres-

sion. Table 2 lists those variables significantly associated with

mortality on univaniate analysis. Although there were too few

patients with leukemia (ti = 4) or lymphoma (ii = 2) to

accurately estimate RR, it should be noted that six of six

(100%) patients with hematologie malignancy died within 21 d

of study entry (P = 0.001).

Outcome: Mortality or Diah’sis

It would be desirable to identify risk factors for requiring

dialysis among those patients with early ATN. The difficulty in

doing so is related to the fact that some individuals with ARF

may die before their renal failure is so severe as to require the

initiation of dialysis. As a result, an “adverse” risk factor for

the outcome “provision of dialysis” might be favorable with

regard to survival; i.e., the provision ofdialysis is dependent on

surviving the early days of ATN. Therefore, we fit our models

using the composite outcome: death on dialysis. Table 3 lists

those variables significantly associated with mortality or dial-

ysis on univariate analysis.

Multivariable Analyses

To determine significant multivariable predictors of the out-

comes of interest, we fit proportional hazards regression mod-

els, using time to death, on time to death or dialysis, as the

dependent variables. Several key explanatory variables were

highly correlated, as expected (e.g. , serum albumin and cal-

cium concentrations, r = 0.53, P < 0.0001 ; serum bicarbonate

and chloride concentrations, r = -0.43, P < 0.0001 ; meehan-

ical ventilation and sepsis, r = 0.35, P < 0.0001; arrhythmias

and acute congestive heart failure, r = 0.25, P < 0.0001). The

potential for collineanity was accounted for in the model build-

ing process.

The results of the multivaniable proportional hazards model

derived for mortality are outlined in Table 4. Although the

association between albumin and mortality did not reach con-

ventional statistical significance, it was included in the final

model due to its large effect estimate, and improved model fit.

The results of the multivaniable proportional hazards model

derived for the composite outcome death or dialysis are out-

lined in Table 5. The coefficients of the regression models are

summarized in the Appendix.

The multivaniable models were validated using the bootstrap

procedure. The bootstrap randomly selects, with replacement, a

predetermined number of observations from the original data

set; /3-coefficients and their respective SEM are then reesti-

mated with each successive procedure (25). Each successive

iteration of the bootstrap yields a sample of distinct composi-

tion, on which the derived model is tested. In theory, this

process is akin to prospective validation of a model, assuming

Page 3: Predictors of Mortality and the Provision of Dialysis in Patients with

Category Value

Table 1. Demographic

characteristics

and baseline clinical and laboratory Table 2. Significant predictors of mortality: univariate

of study subjects analysisa

Parameter RR 95% CI

Male gender 2.20 1.35 to 3.57

Oliguria 2.20 1.43 to 3.38

Mechanical ventilation 2.43 1 .58 to 3.74

Acute myocardial infarction 2.02 1 .23 to 3.31

Arrhythmias 1.77 1.16 to 2.69

Acute congestive heart failure 2. 14 1 . 1 1 to 4.13

Gastrointestinal bleeding 2.68 1 .61 to 4.45

Infection 2.16 1.42 to 3.29

Sepsis 1.72 1.14 to 2.61

Acute stroke or seizure 2.80 1 .49 to 5.26

Chronic liver disease 2.42 1.22 to 4.81

History of hypertension 0.64 0.43 to 0.97

Chronic immunosuppression 2.57 1.29 to 5.11

Albumin (per g/dl increase) 0.54 0.38 to 0.77

Bicarbonate (per mEqfL increase) 0.93 0.89 to 0.97

Calcium (per mg/dl increase) 0.75 0.60 to 0.94

Chloride (per mEqIL increase) 1.04 1.01 to 1.07

Creatinine (per mg/dl increase) 0.83 0.72 to 0.96

Glucose (per mg/dl increase) 1.13 1.01 to 1.27

Sodium (per mEqIL increase) 1 .05 1 .0 1 to 1.08

Total bilirubin (per mg/dl increase) 1 .06 1 .03 to 1.06

Total protein (per g/dl increase) 0.79 0.63 to 0.98

Urea nitrogen (pen mg/dl increase) 1.08 1.01 to 1.16

a RR, relative risk; CI, confidence interval.

Table 3. Significant predictors of mortality or dialysis:

univariate analysisa

Parameter RR 95% CI

Oliguria 4.81 3.35 to 6.89

Mechanical ventilation 2.06 1 .47 to 2.89

Acute myocardial infarction 1.58 1.04 to 2.41

Arrhythmias 1.70 1.20 to 2.38

Acute congestive heart failure 1 .90 1 .09 to 3.31

Albumin (pen g/dl increase) 0.61 0.46 to 0.79

Bicarbonate (per mEqfL increase) 0.96 0.92 to 0.99

Calcium (per mg/dl increase) 0.80 0.67 to 0.96

Phosphorus (per mg/dl increase) 1 . 10 1 .00 to 1.21

Potassium (per mEqfL increase) 1 .25 1 .02 to 1.53

Total bilirubin (per mg/dl increase) 1.04 1.02 to 1.06

Total protein (per g/dl increase) 0.78 0.65 to 0.93

Urea nitrogen (per mg/dl increase) 1 .09 1.02 to 1.15

a Abbreviations as in Table 2.

that the sample from which the model was derived and the

sample on which the model was tested are drawn from the

same population. One-hundred bootstrapped samples per out-

come measure, each with 256 observations, were analyzed.

The point estimates of RR were ±5% compared with the RR

estimates derived from the original data set. The 95% CI were

694 Journal of the American Society of Nephrology

a Variable highly skewed, expressed as median (intenquartile

range), otherwise mean ± SD.

Demographic factors

age (yr)

gender (% female)

race or ethnicity (%)

Caucasian

African-American

Hispanic

Asian-American

other

Acute medical conditions

mechanical ventilation (%)infection, with on without sepsis (%)sepsis (%)

arrhythmias (%)oliguria (%)

acute myocandial infarction (%)gastrointestinal bleeding (%)

acute congestive heart failure (%)acute stroke on seizure (%)

panereatitis (%)Chronic medical conditions

hypertension (%)

coronary artery disease (%)congestive heart failure (%)diabetes mellitus (%)liver disease (%)immunosuppression (%)

obstructive lung disease (%)

Solid nonhematogenous cancer (%)leukemia or lymphoma (%)

Laboratory values

albumin (g/dl)

alanine aminotnansferase (UIL)

aspartate aminotnansfenase (UIL)

bicarbonate (mEq/L)

calcium (mg/dl)

chloride (mEqfL)

eneatinine (mg/dl)

glucose (mg/dl)

lactate dehydrogenase (U/L)

phosphorus (mg/dl)

potassium (mEqfL)

sodium (mEqIL)

total bilirubin (mg/dl)

total protein (g/dl)

urea nitrogen (mg/dl)

leukocyte count (Xl000)

hematoenit (%)

platelet count (Xl000)

62.0 ± 16.9

35

71

17

10

2

50

4730

29

23

15

11

7

6

5

59

48

29

28

6

6

4

3

2

2.7 ± 0.7

41 (19 to 87y’

58 (28 to l45)�

21.4 ± 4.8

7.9 ± 0.9

101.2 ± 7.6

4.6 ± 2.0

167.5 ± 81.8

460 (280 to 9l6)�

5.2 ± 1.9

4.5 ± 0.7136.1 ± 5.9

3.5 ± 6.6

5.3 ± 1.0

66.2 ± 28.5

13.7 ± 8.9

29.7 ± 5.0

154 ± 96

Page 4: Predictors of Mortality and the Provision of Dialysis in Patients with

Mortality and Dialysis in Acute Tubular Necrosis 695

Table 4. Multivariable proportional hazards regression analysis: mortality’1

Parameter RR 95% CI P Value

Total bilirubin 1 .06 1 .03 to 1 .08 18.42 <0.0001

Acute myocardial infarction 3.14 1.85 to 5.31 18.11 <0.0001

Oligunia 2.25 1.43 to 3.55 12.26 0.0005

Acute stroke on seizure 3.08 1.56 to 6.06 10.57 0.001

Mechanical ventilation 1 .86 1 . 1 8 to 2.93 7.24 0.007

Male gender 2.01 1.21 to 3.36 7.15 0.008

Chronic immunosuppression 2.37 1.16 to 4.88 5.53 0.02

Bicarbonate 0.95 0.90 to 0.99 4.82 0.03

Albumin 0.73 0.51 to 1.04 3.04 0.08

a Total bilinubin, per mg/dl increase; bicarbonate, per mEqfL increase; albumin, pen g/dl increase; other factors (yes/no). x2 ranked by

level of statistical significance after simultaneous adjustment for other model covariates. Abbreviations as in Table 2.

Table 5. Multivaniable proportional hazards regression analysis: mortality or dialysisa

Parameter RR 95% CI x� P Value

Oliguria 5.95 3.96 to 8.95 73.43 <0.0001

Albumin 0.56 0.42 to 0.74 15.74 <0.0001

Acute myocardial infarction 1.95 1 .24 to 3.07 8.29 0.004

Mechanical ventilation 1.53 1.07 to 2.21 5.30 0.02

Arrhythmias 1.51 1.04 to 2.19 4.71 0.03

a Albumin, pen g/dl increase; other factors (yes/no). � ranked by level of statistical significance after simultaneous adjustment for other

model covaniates. Abbreviations as in Table 2.

Table 6. Multivariable logistic regression analysis: mortality�’

Parameter OR 95% CI P Value

Male gender 3.70 1.75 to 7.82 15.67 <0.0001

Mechanical ventilation 2.95 1.53 to 5.68 15.53 <0.0001

Oliguria 4.39 2.09 to 9.24 13.29 0.0003

Acute myocardial infarction 5.90 2.43 to 14.4 1 1 .07 0.0009

Acute stroke or seizure 7.35 1.92 to 28.1 8.32 0.004

History of hypertension 0.44 0.23 to 0.86 5.68 0.02

Bicarbonate 0.93 0.86 to 1.00 4.22 0.04

a Bicarbonate, pen mEqfL increase; other factors (yes/no). x2 ranked by level of statistical significance after simultaneous adjustment for

other model covariates. OR, odds ratio. Other abbreviations as in Table 2.

up to 5% wider than in the primary regression models. No

covaniates were eliminated from the bootstrapped predictive

models due to nonsignificance.

To test whether the predictors were influenced by the chosen

analytical method and to determine model discrimination,

companion logistic regression models were fit using 30-d mor-

tality as the dependent variable. Seventy-four patients died

within 30 d. Multivariable logistic regression yielded a model

with seven significant explanatory variables, outlined in Table

6. The area under the ROC curve was 0.8 1 , indicating very

good model discrimination. The logistic regression model was

qualitatively similar to the model based on proportional haz-

ards regression. Total bilirubin was a prominent feature of the

hazard function, but not significantly associated with 30-d

mortality using the multivaniable logit function. This disenep-

ancy suggests that elevated levels of total bilirubin were asso-

ciated with death early in the post-ARF course.

Validation of the Cleveland Clinic Model

In an effort to refine outcome prediction in this population,

we aimed to validate an existing model on these data. We

chose the Cleveland Clinic Foundation Mortality in Acute

Renal Failure model (CCF Model) of Paganini et a!. (26),

because: (1) it captured a similar array of variables; (2) it had

been developed using logistic regression analysis; and, most

importantly, (3) it had performed extremely well in its own

institution. Because the CCF Model was restricted to patients

who required dialysis, we arbitrarily modified the points as-

Page 5: Predictors of Mortality and the Provision of Dialysis in Patients with

696 Journal of the American Society of Nephrology

Table 7. Validation of the modified

et al.’1

CCF model of Paganini

SconeNo. of Deaths/

Total No. in GroupCCF Derivation (%)

No. of Deathsfl’otalNo. in Group ANP

Validation (%)

0 to 4 9/37 (24) 1/29 (3)

5 to 7 49/99 (49) 14/87 (16)

8 to 14 245/325 (75) 42/1 18 (36)

�15 40/45 (89) 17/22 (77)

a CCF, Cleveland Clinic Foundation; ANP, atrial natniuretic

peptide.

signed to the ereatinine and urea nitrogen variables (i.e. , pa-

tients with serum ereatinine �2 mg/dl were assigned one rather

than two points, and patients with blood urea nitrogen >75

mg/dl were assigned one point; all others were assigned 0

points). This resulted in a score range of 0 to 1 8 (rather than 0

to 20 for the original CCF Model). For other variables, points

were assigned according to those of Paganini et a!. Patients

were grouped in a similar manner (i.e. , 0 to 4 points, 5 to 7

points, 8 to 14 points, �15 points).

Using the modified CCF Model, the number of points as-

signed was 8.7 ± 3.5 (median. 9; range, 3 to 16). Table 7

shows the predictive performance of the modified CCF Model

on the ANP study data, compared with the original derivation

set reported by Paganini et al. Although the overall mortality

rates were lower in the ANP Study (in which all patients did

not require dialysis), the modified CCF Model performed

extremely well on these data. The area under the ROC curve

for the modified CCF score was 0.75.

DiscussionPrognostic stratification in ARF has been the subject of

numerous previous reports (see above). An ARF-speeifie in-

strument is needed, because generic severity-of-illness mea-

sures, such as the Acute Physiology and Chronic Health Eval-

uation (APACHE) II scone, have not consistently predicted

short-term mortality in critically ill patients with ARF

(14,18,22). Furthermore, such an index could be used to pro-

mote quality improvement in ARF management by providing

an objective means of comparing differences in case mix and

outcomes across multiple institutions.

Several observations from the mortality analysis are note-

worthy. First, there was a striking difference in risk by gender,

with men experiencing approximately twice the mortality nate

of women, even after covariate adjustment. The reasons for this

increased risk are unclear. Second, several factors that were

associated with mortality on univaniate analysis (arrhythmias,

acute congestive heart failure, gastrointestinal bleeding,

chronic liver disease, infection, and numerous laboratory van-

ables) were not independently associated with mortality after

adjustment for the model covariates. This observation does not

suggest that these variables were unimportant, but rather that

they have insignificant predictive power over and above that of

other related factors, such as mechanical ventilation and total

bilirubin. Finally, advanced age was notably absent from both

the univariate results and the multivariable regression models.

Despite a broad age range (18 to 93, interquartile range, 50 to

75) and relatively large sample size, we were unable to detect

any relation between advanced age and the RR of death.

The analysis of the composite outcome death or dialysis also

provided important prognostic insights. The RR of death or

dialysis was markedly increased with oliguria, to nearly three

times the magnitude of the RR of death alone (sixfold com-

pared with twofold). This finding emphasizes the importance

of oliguria in influencing the timing of dialysis initiation. The

relation between oliguria and outcome is complex, however.

We have shown previously that once dialysis is required, the

presence or absence of oliguria predating the initiation of

dialysis is of little or no prognostic value (22). These findings

suggest that oliguria itself may not increase mortality risk;

rather, the dialysis procedure, on another associated factor

unrelated to the severity of renal disease, may directly influ-

ence outcomes.

There was a striking association between the serum albumin

concentration and the risk of death or dialysis. Although the

association between low serum albumin and mortality in pa-

tients with end-stage renal disease (27-30), and other disease

states (3 1 ,32), has been well established, the prognostic impor-

tance of the serum albumin in ARF has not been previously

demonstrated. Likewise, others have failed to show an association

between metabolic acidosis and adverse outcomes in ARF. The

therapeutic implications of these findings are unknown; survival

or renal recovery may or may not be enhanced with correction of

acidosis, or with administration of hyperalimentation or growth

factors aimed at reversing catabolism (33).

There are several potential limitations to this study. The

exclusion of subjects who were thought to require dialysis

imminently, or who were not considered candidates for dialy-

sis, might have biased the study sample toward a less critically

ill population. Furthermore, if practitioners precluded from

study participation subjects of a particular age, sex, race, on

disease category, particularly those who they thought were at

higher risk, the study sample might not have been representa-

tive. Conversely, restriction to patients with ATN might bias

the study sample toward a more critically ill population, com-

pared with the ARF population at lange. Although ATN is the

most common pathologic correlate of ARF in seriously ill

patients, the predictive models may not be valid for individuals

with ARF due to atheroembolie disease, prerenal azotemia,

interstitial nephritis, and other causes of renal disease. With

regard to the statistical analyses, models were fit with contin-

uous independent variables. This approach assumes a linear

relation between risk factors and outcomes. Categorization of

these variables obviates the linearity assumption, but markedly

reduces the power to detect associations. On the basis of

companion analyses (data not shown), linearity assumptions

were reasonable for the continuous variables of major interest

(i.e. , total bilirubin, bicarbonate, and albumin) over the ranges

encountered in clinical practice. It is noteworthy that results

obtained from proportional hazards and logistic regression

analyses were qualitatively similar.

Page 6: Predictors of Mortality and the Provision of Dialysis in Patients with

Mortality and Dialysis in Acute Tubular Necrosis 697

In summary, using data collected prospectively as part of a

randomized clinical trial, we found that male gender, oliguria,

mechanical ventilation, acute myocardial infarction, acute

stroke or seizure, chronic immunosuppression, hyperbiliru-

binemia, and metabolic acidosis were significantly associated

with the RR of death after progressive ARF due to ATN. A

similar model was developed for the composite outcome, death

or dialysis. An existing model derived at a single institution

(CCF Model) was validated on these data. These observations

suggest that the risk of death or dialysis can be determined

relatively early in the course of ARF. Continued efforts must

be directed at refining predictive models in ARF (1) to provide

accurate prognosis for patients with this devastating complica-

tion; (2) to adjust for ease-mix variation in quality improve-

ment efforts; and (3) to objectively stratify patients in future

intervention trials aimed at favorably altering the course of

hospital-acquired ARF.

AppendixModel 1: Mortality

log hazard ratio

= 0.6991 (MALE) + 0.8128 (OLIGURIA)

+ 0.0557 (TOTAL BILIRUBIN)

+ 0.6215 (VENTILATION)

+ 1 . 1 245 (STROKE OR SEIZURE)

+ 1.1432 (ACUTE MI)

+ 0.8643 (IMMUNOSUPPRESSION)

- 0.0555 (BICARBONATE)

- 0.3139 (ALBUMIN)

Model 2: Mortality or Dialysis

log hazard ratio

= 1.7836 (OLIGURIA)

+ 0.6662 (ACUTE MI)

+ 0.4107 (ARRHYTHMIA)

+ 0.4279 (VENTILATION)

- 0.5851 (ALBUMIN)

References1 . Hou SH, Bushinsky DA, Wish lB. Cohen II, Harrington IT:

Hospital-acquired renal insufficiency: A prospective study. Am J

Med 74: 243-248, 1983

2. Shusterman N, Strom BL, Murray TG, Morrison G, West SL,

Maislin G: Risk factors and outcome of hospital-acquired acute

renal failure. Am J Med 83: 65-71, 1987

3. Anderson RI, Linas SL, Benns AS, Henrich WL, Miller TR,

Gabow PA, Schrien RW: Nonoliguric acute renal failure. N Engl

JMed296: 1134-1138, 1977

4. McMurnay SD, Luft FC, Maxwell DR. Hamburger RJ, Futty D,

Szwed II, Lavelle KJ, Kleit SA: Prevailing patterns and predictor

variables in patients with acute tubular necrosis. Arch mt Med

138: 950-955, 1978

5. Routh GS, Briggs ID, Mone 1G. Ledingham IM: Survival from

acute renal failure with and without multiple organ dysfunction.

Postgrad Med J 56: 244-247, 19806. Cioffi WG, Ashikaga T, Gamelli RL: Probability of surviving

postoperative acute nenal failure. Ann Surg 200: 205-2 1 1, 1984

7. Lien I, Chan V: Risk factors influencing survival in acute renal

failure treated by hemodialysis. Arch Intern Med 145: 2067-

2069, 1985

8. Rasmussen HH, Pitt EA, Ibels LS, McNeil DR: Prediction of

outcome in acute renal failure by diseriminant analysis of clinical

variables. Arch Intern Med 145: 2015-2018, 1985

9. Bullock ML, Umen Al, Finkelstein M, Keane WF: The assess-

ment of risk factors in 462 patients with acute renal failure. Am J

Kidney Dis 5: 97-103, 1985

10. Abneo K, Moorthy V. Osborne M: Changing patterns and out-

come of acute renal failure requiring hemodialysis. Arc/i Intern

Med 146: 1338-1341, 1986

1 1 . Wheeler DC, Feehally J, Walls I: High risk acute renal failure. QJ Med 61: 977-984, 1986

12. Conwin HL, Teplick RS, Schreiben Ml, Fang LST, Bonventre IV,

Coggins CH: Prediction of outcome in acute renal failure. Am J

Nephrol 7: 8-12, 1987

13. Lohn 1W, MeFarlane Ml, Gnantham JJ: A clinical index topredict survival in acute renal failure patients requiring dialysis.

AmiKidnevDis 11: 254-259, 198814. Mahen ER, Robinson KN, Scoble JE, Farnimond JG, Bnowne

DRG, Sweny P. Moorhead iF: Prognosis of critically ill patientswith acute renal failure: APACHE II score and other predictive

factors. Q J Med 72: 857-866, 1989

15. Turney JH, Marshall DH, Bnownjohn AM, Ellis CM: The evo-

lution of acute renal failure, 1956-1988. Q J Med 74: 83-104,1991

16. Gnoeneveld ABJ, Tran DD, van den Meulen I, Nauta JJP, Thijs

LG: Acute renal failure in the medical intensive care unit: Pre-

disposing, complicating factors and outcome. Nephron 59: 602-

610, 1991

17. Lanone JJ, Bnunet F, Pochand F, Bellivien F, Dhainaut IF, Vax-

elaine IF, Ginaud T, Dreyfus F, Dreyfus D, Chiehe ID, Monsallien

IF: Hemodialysis for acute renal failure in patients with hema-

tologic malignancies. Crit Care Med 19: 346-351, 1991

I 8. Schaefer JH, Jochimsen F, Keller F, Wegscheiden K, Distler A:

Outcome prediction of acute renal failure in medical intensive

cane. Intensive Care Med 17: 19-24, 1991

19. Barton 1K, Hilton PJ, Taub NA, Warburton FG, Swan AV,

Dwight I, Mason JC: Acute renal failure treated by haemofiltra-

tion: Factors affecting outcome. Q J Med 86: 81-90, 1993

20. Frost L, Pedensen RS, Bentzen 5, Bille H, Hansen HE: Short and

long term outcome in a consecutive series of 419 patients with

acute dialysis-requiring renal failure. Scand J Urol Nephrol 27:

453-462, 1993

21 . Halstenbeng W, Goormastic M, Paganini EP: Risk modeling inacute renal failure: Valuable predictor on mathematical guessing?

[Abstract] J Am Soc Nephrol 4: 3 1 7, 1993

22. Chertow GM, Chnistiansen CL, Cleary PD, Munno C, Lazarus

JM: Prognostic stratification in critically ill patients with acute

Page 7: Predictors of Mortality and the Provision of Dialysis in Patients with

698 Journal of the American Society of Nephrology

renal failure requiring dialysis. Arch Intern Med 155: 1505-

1511, 1995

23. Cox DR: Regression models and life tables. J Royal Stat Soc [B]

74: 187-220, 1972

24. Collett D: Modelling Survival Data in Medical Research, Lon-

don, Chapman and Hall, 1994

25. Efron B, Gong G: A leisurely look at the bootstrap, the jackknife,

and cross validation. American Statistician 37: 36-48, 1983

26. Paganini EP, Halstenbeng WK, Goormastic M: Risk modeling in

acute renal failure requiring dialysis: The introduction of a new

model. Cli,i Nephrol 46: 206-2 1 1 , 199627. Lownie EG, Lew NL: Death risk in hemodialysis patients: The

predictive value of commonly measured variables and an eval-

uation of death rate differences between facilities. Am J Kidney

Dis 15: 458-482, 1990

28. Churchill DN, Taylor DW, Cook Ri, LaPlante P. Bane P. Cartien

P, Fay WP, Goldstein MB, Jindal K, Mandin H, McKenzie 1K,

Muinhead N, Parfrey PS, Posen GA, Slaughter D, Ulan RA,

Wenb R: Canadian hemodialysis morbidity study. Am J Kidney

Dis 19: 214-234, 1992

29. Collins Al, Ma IZ, Umen A, Keshaviah P: Urea index and other

predictors of hemodialysis patient survival. Am JKidney Dis 23:

272-282, 1994

30. Blake PG, Flowendew G, Blake RM, Oreopoulos DG: Serum

albumin in patients on continuous peritoneal dialysis: Predictors

and correlations with outcomes. J Am Soc Nephrol 3: 1501-

1507, 1993

31. Phillips A, Shaper AG, Whincup PH: Association between serum

albumin and mortality from cardiovascular disease, cancer, and

other causes. Lancet I : 1434-1436, 1989

32. Dames B, Ducimetiene P: Serum albumin and mortality. Lancet

335: 350-351, 1990

33. Sponsel H, Congen ID: Is parentenal nutrition therapy of value in

acute renal failure patients? Am J Kidney Dis 25: 96-102, 95