**3. Defining AKIs**

6 Will-be-set-by-IN-TECH

ability to diagnose a subsequent rise in a surrogate marker of function (normally creatinine) is assessed. However, we need to explore the early stages of, and prior to, loss of renal function in order to understand the relevant pathophysiology of clinical AKI. We can conceptualise this as an evaluation of the evolution of injury phase which leads to loss of GFR and the early loss of function phase immediately following a GFR decrease (Figure 1). Detection and characterisation of this very early phase in man appears essential for progress. This highlights the need for quantifying the time course of injury biomarkers in relation to change in GFR and for real-time assessment of renal function. Two promising techniques are under development. The ambulatory renal monitor (ARM) is a shielded detector which monitors extracellular excretion of <sup>99</sup>*m*TcDTPA over up to 24 hours following injection and relates this to GFR (Rabito et al. (2010)). The ratiometric fluorescence approach monitors the plasma disappearance of both a rapidly filtered and a poorly filtered fluorescent marker introduced by a single bolus infusion, changes in the ratio of which enable a calculation of GFR (Wang et al. (2010)). Both techniques allow a rapid (5 to 15 min) measure of GFR. Ideally these or similar techniques would allow monitoring of kidney function from prior to the time of insult

**0 6 24 48 72**

**Time from insult (hours)**

Fig. 1. The paradigm of function compared with the paradigm of injury. Following injury the early injury biomarker is elevated prior to change in function (here, a 65% loss of GFR, approximately 6 hours post-injury). The later biomarker takes takes longer to be elevated.

Much of the literature describes a biomarker as "predicting" AKI when it precedes an increase in plasma creatinine. However, a biomarker is only truly predictive if it precedes a decrease in GFR. If it is elevated shortly after such a decrease, then it should be described as *diagnostic* rather than predictive of AKI even though it may predict the latter increase in plasma

Creatinine is only elevated following loss of GFR (following equation 6)

**-70**

**-60**

**-50**

**-40**

**-30**

**GFR change (%)**

**-20**

**-10**

**Creatinine Change in GFR**

**Early Biomarker Later Biomarker** **0**

through the evolution time until change of GFR and beyond.

**0.5**

**1.0**

**1.5**

**Plasma Creatinine (mg/dl)** 

**Biomarker concentrations U/ml**

**1000.0**

**0.0**

creatinine.

**2.0**

**2.5**

In its broadest sense renal failure is simply the rapid loss of renal filtration (ie decrease in GFR). As discussed, current clinical diagnosis of AKI and evaluation of novel biomarkers of kidney injury are largely dependent on observation of changes in creatinine as a surrogate for a change in GFR. An understanding of the kinetics of the relationship between creatinine and GFR provides a basis for understanding the definitions of AKI and their limitations. Plasma cystatin C, an alternative to creatinine (Nejat et al. (2010); Westhuyzen (2006)), follows similar kinetics.

## **3.1 Creatinine and cystatin C kinetics**

Creatinine and cystatin C are generated in tissue outside of the plasma compartment, diffuse into the plasma compartment from where they are lost by renal and non-renal excretion. Because both creatinine and cystatin C are not bound to plasma proteins the exchange between the extravascular and plasma compartments is rapid (compared with rates of production and elimination) allowing us to conflate the two compartments into one compartment with volume of distribution, *V* (Figure 2).

#### Fig. 2. One compartment pharmacokinetics model

The change in total mass (*q*) of creatinine/cystatin C depends on the rate at which it is entering and the rate at which it is leaving the compartment:

$$\frac{dq}{dt} = \text{gain from generation} - (\text{real loss} + \text{non-real loss}) \tag{1}$$

Under normal circumstances non-renal losses are much less than renal losses and may be ignored. The renal loss is the product of the renal elimination rate constant, *kr*, and the total mass, *q*. As the total mass is the product of the concentration *C* and volume of distribution *V*, equation 1 becomes:

$$\mathbf{C}\frac{dV}{dt} + \frac{d\mathbf{C}}{dt}V = \mathbf{G} - k\_r \mathbf{C}V\tag{2}$$

where *G* is the generation rate of creatinine/cystatin C (see Box 3.1). The volume of distribution of creatinine is equal to the total body water (*TBW*) and that of cystatin C to the extracellular fluid (about one third of *TBW*)(see Box 3.2). If this is assumed not to change then equation 2 becomes:

$$\frac{d\mathbb{C}}{dt} = \frac{G}{V} - k\_r \mathbb{C} \tag{3}$$

At equilibrium (*dC*/*dt* = 0 at *t* = 0), prior to any change in GFR the renal elimination rate constant may be determined from equation 3:

$$k\_{r0} = \frac{G}{C\_b V} \tag{4}$$

*Box 3.2.* Estimating the volume of distribution

estimated from the body weight as:

**3.2 Categorical consensus**

Lameire et al. (2006)).

The volume of distribution of creatinine is equal to the total body water (*TBW*) and of cystatin C to the volume of the extracellular fluid which is about 1/3*rd* of the *TBW* (Hansen (2002)). TBW is often

The Metamorphosis of Acute Renal Failure to Acute Kidney Injury 133

*TBW* = 0.6 × *Body weight* (*L*)

Male (*r*<sup>2</sup> <sup>=</sup> 0.704): *TBW* <sup>=</sup> 2.447 <sup>−</sup> 0.09516 <sup>×</sup> *age* <sup>+</sup> 0.1074 <sup>×</sup> *height* <sup>+</sup> 0.3662 <sup>×</sup> *weight* (*L*)

In 2004 the Acute Dialysis Quality Initiative Group (ADQI) developed a consensus definition for AKI and severity staging (RIFLE: Risk, Injury, Failure, Loss, End Stage) (Bellomo et al. (2004)). The scheme involved both a creatinine based classification and a urine output based classification (Table 1). A decrease in GFR of more than 25% or increase in serum creatinine of 50% was deemed sufficient to diagnose AKI. Unfortunately, there was an error in calculating this relationship which was not identified until 2009 (Pickering & Endre (2009a)). A 50% increase in creatinine is equivalent to a one third decrease in GFR (see equation 8) not a 25% decrease. Similarly, for RIFLE stage F a 200% increase in creatinine is equivalent to a two thirds decrease in GFR not a 75% decrease. Whilst plasma creatinine rather than a GFR measure, or an estimation with creatinine clearance, is the analyte of choice in AKI studies, creatinine is but a surrogate for GFR and GFR should remain the principal diagnostic parameter of AKI (Pickering & Endre (2009b)). The ADQI group recommended AKI be defined as "sustained" (lasting at least 24hrs) and "abrupt" (1-7 days) (http://www.ccm.upmc.edu/adqi/ADQI2/ADQI2g1.pdf). Whilst duration did not appear in the seminal RIFLE publication it was included in later publications (Hoste et al. (2006);

On the back of new evidence that even minor changes in serum creatinine are associated with poor outcomes (eg Chertow et al. (2005); Lassnigg et al. (2004)), the Acute Kidney Injury Network modified the RIFLE definition to include a small absolute rise in creatinine (0.3 mg/dl or 26.4 *μ*mol/l). Further modifications included requiring the change to occur within 48 hours for the definition of AKI, and removing RIFLE stages L and E in preference to all patients requiring renal replacement therapy to be assigned to severity stage III (Table 1). Both definitions have received broad support and there is considerable evidence for an association of increased mortality with increased RIFLE or AKIN stages (Bagshaw et al. (2008); Ricci et al. (2008)). More recently, duration of AKI (using the AKIN criteria) has been shown to be independently associated with long-term mortality (Brown et al. (2010); Coca et al. (2010); Goldberg et al. (2009)). KDIGO (Kidney Disease for Improving Global Outcomes:

More accurate formulae have been derived from population studies (Watson et al. (1980)):

Female (*r*<sup>2</sup> <sup>=</sup> 0.736): *TBW* <sup>=</sup> 2.097 <sup>+</sup> 0.1069 <sup>×</sup> *height* <sup>+</sup> 0.2466 <sup>×</sup> *weight* (*L*)

Where height is not available, the slightly less robust equations may be used:

Male (*r*<sup>2</sup> <sup>=</sup> 0.689): *TBW* <sup>=</sup> 20.03 <sup>−</sup> 0.1183 <sup>×</sup> *age* <sup>+</sup> 0.3626 <sup>×</sup> *weight* (*L*)

where *height* is measured in cm, *weight* in kg, and *age* in years.

Female (*r*<sup>2</sup> <sup>=</sup> 0.717): *TBW* <sup>=</sup> 14.46 <sup>+</sup> 0.2549 <sup>×</sup> *weight* (*L*)

where *Cb* is the baseline concentration (at time *t* = 0). *kr*<sup>0</sup> may also be estimated from the renal clearance (*Cl*) which is simply the product *kr*<sup>0</sup> and *V*. For creatinine, this may be measured in critically ill paitients using a short duration creatinine clearance and an estimate of an individual's volume of distribution (see below). From the EARLYARF study (Endre et al. (2010)) where 4-h creatinine clearances were measured on entry to ICU in 484 patients we were able to calculate a median (interquartile range) for *kr*<sup>0</sup> of 0.11 (0.07-0.17) h-1.

Following a loss in GFR (Δ*g*%), the renal elimination rate constant becomes:

$$k\_{\!\!\!F} = (1 - \frac{\Delta \!\!g}{100}) k\_{r0} \tag{5}$$

and equation 3 may be solved numerically, or, if we assume of *G*, *V*, or *kr* are not varying, may be integrated to give the concentration as a function of time (Chiou & Hsu (1975); Chow (1985)):

$$\mathcal{C}(t) = \frac{G}{k\_I V} (1 - e^{-k\_I t}) + \mathbb{C}\_b e^{-k\_I t} \tag{6}$$

As *t* → ∞ the concentration asymptotically approaches a new steady state, *Css* which from equation 6 is:

$$\mathcal{L}\_{\rm ss} = \frac{G}{k\_r V} \tag{7}$$

Mathematically the concentration is within 5% of the new steady state in 4.4 half lives (Half life: *t*1/2 = *ln*(2)/*kr*). In practice this is well within the uncertainty in creatinine measurements. The new steady state may also be determined by substituting equations 5 and 4 into equation 7:

$$\mathcal{C}\_{\rm ss} = \frac{\mathcal{C}\_{\rm b}}{(1 - \frac{\rm Ag}{100})} \tag{8}$$

From this equation we may determine the equivalences between a decline in GFR and a rise in creatinine used in the RIFLE definition of AKI (Table 1).

*Box 3.1.* Creatinine and Cystatin C production rates

*Creatinine production* (Bjornsson (1979)):

Male (*r*<sup>2</sup> <sup>=</sup> 0.919): *<sup>G</sup>*<sup>0</sup> = (<sup>27</sup> <sup>−</sup> 0.173 <sup>×</sup> *age*) <sup>×</sup> *weight*/24 (*mg*/*h*) Female (*r*<sup>2</sup> <sup>=</sup> 0.966): *<sup>G</sup>*<sup>0</sup> = (<sup>25</sup> <sup>−</sup> 0.175 <sup>×</sup> *age*) <sup>×</sup> *weight*/24 (*mg*/*h*)

Creatinine production may decrease during critical illness (Griffiths (1996)). If the reduction is at constant rate *m*%, (eg 2% per day as suggested by Griffiths) then:

$$G(t) = G\_0 e^{-mt} \quad \text{and} \quad \frac{dG}{dt} = -mG$$

*Cystatin C production*

Only one study has measured the rate constant of Cystatin C (Sjostrom et al. (2005)). There was no significant difference with age, sex or lean body mass. The rate constant per 1.73*m*<sup>2</sup> of body surface area was:

$$G\_{\rm CysC} = 7.44 \quad (mg/h/1.73m^2)$$

*Box 3.2.* Estimating the volume of distribution

8 Will-be-set-by-IN-TECH

where *Cb* is the baseline concentration (at time *t* = 0). *kr*<sup>0</sup> may also be estimated from the renal clearance (*Cl*) which is simply the product *kr*<sup>0</sup> and *V*. For creatinine, this may be measured in critically ill paitients using a short duration creatinine clearance and an estimate of an individual's volume of distribution (see below). From the EARLYARF study (Endre et al. (2010)) where 4-h creatinine clearances were measured on entry to ICU in 484 patients we

were able to calculate a median (interquartile range) for *kr*<sup>0</sup> of 0.11 (0.07-0.17) h-1. Following a loss in GFR (Δ*g*%), the renal elimination rate constant becomes:

*<sup>C</sup>*(*t*) = *<sup>G</sup>*

in creatinine used in the RIFLE definition of AKI (Table 1).

constant rate *m*%, (eg 2% per day as suggested by Griffiths) then:

Male (*r*<sup>2</sup> <sup>=</sup> 0.919): *<sup>G</sup>*<sup>0</sup> = (<sup>27</sup> <sup>−</sup> 0.173 <sup>×</sup> *age*) <sup>×</sup> *weight*/24 (*mg*/*h*) Female (*r*<sup>2</sup> <sup>=</sup> 0.966): *<sup>G</sup>*<sup>0</sup> = (<sup>25</sup> <sup>−</sup> 0.175 <sup>×</sup> *age*) <sup>×</sup> *weight*/24 (*mg*/*h*)

*G*(*t*) = *G*0*e*

*Box 3.1.* Creatinine and Cystatin C production rates

*Creatinine production* (Bjornsson (1979)):

(1985)):

equation 6 is:

and 4 into equation 7:

*Cystatin C production*

area was:

*kr* = (<sup>1</sup> <sup>−</sup> <sup>Δ</sup>*<sup>g</sup>*

*krV* (<sup>1</sup> <sup>−</sup> *<sup>e</sup>*

and equation 3 may be solved numerically, or, if we assume of *G*, *V*, or *kr* are not varying, may be integrated to give the concentration as a function of time (Chiou & Hsu (1975); Chow

As *t* → ∞ the concentration asymptotically approaches a new steady state, *Css* which from

*Css* <sup>=</sup> *<sup>G</sup>*

Mathematically the concentration is within 5% of the new steady state in 4.4 half lives (Half life: *t*1/2 = *ln*(2)/*kr*). In practice this is well within the uncertainty in creatinine measurements. The new steady state may also be determined by substituting equations 5

*Css* <sup>=</sup> *Cb*

(<sup>1</sup> <sup>−</sup> <sup>Δ</sup>*<sup>g</sup>*

From this equation we may determine the equivalences between a decline in GFR and a rise

Creatinine production may decrease during critical illness (Griffiths (1996)). If the reduction is at

<sup>−</sup>*mt* and *dG*

Only one study has measured the rate constant of Cystatin C (Sjostrom et al. (2005)). There was no significant difference with age, sex or lean body mass. The rate constant per 1.73*m*<sup>2</sup> of body surface

*GCysC* = 7.44 (*mg*/*h*/1.73*m*2)

*dt* <sup>=</sup> <sup>−</sup>*mG*

−*krt*

) + *Cbe*

<sup>100</sup> )*kr*<sup>0</sup> (5)

*krV* (7)

<sup>100</sup> ) (8)

<sup>−</sup>*krt* (6)

The volume of distribution of creatinine is equal to the total body water (*TBW*) and of cystatin C to the volume of the extracellular fluid which is about 1/3*rd* of the *TBW* (Hansen (2002)). TBW is often estimated from the body weight as:

$$TBW = 0.6 \times Body\ weight \quad (L)$$

More accurate formulae have been derived from population studies (Watson et al. (1980)):

Male (*r*<sup>2</sup> <sup>=</sup> 0.704): *TBW* <sup>=</sup> 2.447 <sup>−</sup> 0.09516 <sup>×</sup> *age* <sup>+</sup> 0.1074 <sup>×</sup> *height* <sup>+</sup> 0.3662 <sup>×</sup> *weight* (*L*) Female (*r*<sup>2</sup> <sup>=</sup> 0.736): *TBW* <sup>=</sup> 2.097 <sup>+</sup> 0.1069 <sup>×</sup> *height* <sup>+</sup> 0.2466 <sup>×</sup> *weight* (*L*)

where *height* is measured in cm, *weight* in kg, and *age* in years.

Where height is not available, the slightly less robust equations may be used:

Male (*r*<sup>2</sup> <sup>=</sup> 0.689): *TBW* <sup>=</sup> 20.03 <sup>−</sup> 0.1183 <sup>×</sup> *age* <sup>+</sup> 0.3626 <sup>×</sup> *weight* (*L*) Female (*r*<sup>2</sup> <sup>=</sup> 0.717): *TBW* <sup>=</sup> 14.46 <sup>+</sup> 0.2549 <sup>×</sup> *weight* (*L*)

#### **3.2 Categorical consensus**

In 2004 the Acute Dialysis Quality Initiative Group (ADQI) developed a consensus definition for AKI and severity staging (RIFLE: Risk, Injury, Failure, Loss, End Stage) (Bellomo et al. (2004)). The scheme involved both a creatinine based classification and a urine output based classification (Table 1). A decrease in GFR of more than 25% or increase in serum creatinine of 50% was deemed sufficient to diagnose AKI. Unfortunately, there was an error in calculating this relationship which was not identified until 2009 (Pickering & Endre (2009a)). A 50% increase in creatinine is equivalent to a one third decrease in GFR (see equation 8) not a 25% decrease. Similarly, for RIFLE stage F a 200% increase in creatinine is equivalent to a two thirds decrease in GFR not a 75% decrease. Whilst plasma creatinine rather than a GFR measure, or an estimation with creatinine clearance, is the analyte of choice in AKI studies, creatinine is but a surrogate for GFR and GFR should remain the principal diagnostic parameter of AKI (Pickering & Endre (2009b)). The ADQI group recommended AKI be defined as "sustained" (lasting at least 24hrs) and "abrupt" (1-7 days) (http://www.ccm.upmc.edu/adqi/ADQI2/ADQI2g1.pdf). Whilst duration did not appear in the seminal RIFLE publication it was included in later publications (Hoste et al. (2006); Lameire et al. (2006)).

On the back of new evidence that even minor changes in serum creatinine are associated with poor outcomes (eg Chertow et al. (2005); Lassnigg et al. (2004)), the Acute Kidney Injury Network modified the RIFLE definition to include a small absolute rise in creatinine (0.3 mg/dl or 26.4 *μ*mol/l). Further modifications included requiring the change to occur within 48 hours for the definition of AKI, and removing RIFLE stages L and E in preference to all patients requiring renal replacement therapy to be assigned to severity stage III (Table 1).

Both definitions have received broad support and there is considerable evidence for an association of increased mortality with increased RIFLE or AKIN stages (Bagshaw et al. (2008); Ricci et al. (2008)). More recently, duration of AKI (using the AKIN criteria) has been shown to be independently associated with long-term mortality (Brown et al. (2010); Coca et al. (2010); Goldberg et al. (2009)). KDIGO (Kidney Disease for Improving Global Outcomes:

**0 6 24 48 72**

**(C5, t5)**

**D E**

**(C6, t6)**

**Time from insult (hours)**

Fig. 3. The Relative Average Creatinine by the trapezoidal rule is the sum of the areas A to E

function (ie reduced the reduction in GFR). The more efficacious the treatment, the less the decrease in GFR. Creatinine profiles were calculated using equation 6 (Pickering et al. (2009)). AKIN, RIFLE and RAVC as outcome variables were compared. At low treatment efficacy, the categorical outcomes underestimated and at high treatment efficacy overestimated the effect of treatment. These effects were exaggerated when the population contained a high proportion of patients with more severe AKI. The RAVC, on the other hand, responded in an almost linear fashion across treatment efficacies. Importantly, when the efficacy was low it was best able to distinguish between placebo and treatment arms. The advantage of the RAVC over the categorial metrics is two fold, first it includes the effect of treatment on those patients who had mild kidney injury which in normal circumstances would not exceed the diagnostic threshold for AKI according to a categorical definition, but which may result in a small increase (eg 20-30%) in creatinine and, second, it measures function over a (pre-determined) time period. As discussed previously, the length of time creatinine is elevated is independently of the maximum elevation associated with mortality. The RAVC captures both severity *and* duration of injury and was used as the primary outcome in the EARLYARF trial, which was the first randomised control trial to use an injury biomarker to

All creatinine based definitions of AKI and the RAVC depend on knowing the normal, or "baseline", plasma creatinine for each individual. For patients undergoing elective surgery

**0.5**

**3.4 The baseline issue**

**1.0**

**(C1, t1)**

**(C2, t2)**

divided by (*t*<sup>6</sup> − *t*1)(following equation 10).

**A**

**1.5**

**Plasma Creatinine (mg/dl)** 

**2.0**

**2.5**

**Measured creatinine Baseline creatinine (Cb)**

**(C4, t4)**

The Metamorphosis of Acute Renal Failure to Acute Kidney Injury 135

**C**

triage patients to placebo or high-dose erythropoietin (Endre et al. (2010)).

**(C3, t3)**

**B**


www.kdigo.org ) has recently reviewed the use of the AKIN and RIFLE criteria and is shortly to release a new consensus definition which combines the two definitions.

\*,\*\* corrected from 25% and 75% see Pickering & Endre (2009a)

Table 1. Severity Staging: Consensus definitions

#### **3.3 A continuum needing continuous variables**

One of the intentions of ADQI in setting up RIFLE was that it would provide common outcomes in clinical trials. During the four year period (2005-08) only 36% of published AKI (non Contrast Induced Nephropathy, CIN) trials used RIFLE or AKIN as an outcome variable, and the use was not consistent in terms of timing and duration of injury (Endre & Pickering (2010)). Amongst CIN intervention trials only 13% used RIFLE or AKIN, whereas most continued to use an increase in creatinine of 25% and/or 0.5 mg/dl (44.2 *μ*mol/l) to diagnose CIN. This later definition is slightly anomalous as only those with pre-existing kidney disease (creatinine > 2.0 mg/dl) can have a greater than 0.5 mg/dl elevation in creatinine that is less than 25%. We have recommended that all CIN studies adopt the AKIN or RIFLE definition of AKI (Endre & Pickering (2010)).

We investigated whether a continuous variable measure of kidney function, the Relative Average Creatinine (RAVC), performed better than the RIFLE and AKIN categorical definitions as an outcome variable in AKI prevention or intervention trials. The RAVC is the integral of the area under the plasma creatinine curve above baseline creatine divided by the total time and baseline creatinine:

$$RAVC(\%) = \frac{100}{\mathcal{C}\_b t} \int\_0^t (\mathcal{C}(t) - \mathcal{C}\_b) dt \tag{9}$$

which in practice is calculated using the trapezoidal rule (Figure 3):

$$RAV\mathbb{C}(\%) = \frac{100}{\mathbb{C}\_b(t\_N - t\_1)} \sum\_{0 < n < N-1} \frac{(\mathbb{C}\_{n+1} - \mathbb{C}\_b) + (\mathbb{C}\_n - \mathbb{C}\_b)}{2} (t\_{n+1} - t\_n) \tag{10}$$

where *N* is the number of creatinine measurements.

We created a population of 10,000 Virtual-In-Patients (VIPs) whose baseline creatinine and changes in GFR were based on real ICU populations. Placebo controlled trials were simulated by randomly assigning half the VIPs to treatments which ameliorated loss of renal 10 Will-be-set-by-IN-TECH

www.kdigo.org ) has recently reviewed the use of the AKIN and RIFLE criteria and is shortly

Injury (I) 100% *>* 50% II 100% *<* 0.5 mg/kg/h

*>* 66.7%∗∗ III 200% or

One of the intentions of ADQI in setting up RIFLE was that it would provide common outcomes in clinical trials. During the four year period (2005-08) only 36% of published AKI (non Contrast Induced Nephropathy, CIN) trials used RIFLE or AKIN as an outcome variable, and the use was not consistent in terms of timing and duration of injury (Endre & Pickering (2010)). Amongst CIN intervention trials only 13% used RIFLE or AKIN, whereas most continued to use an increase in creatinine of 25% and/or 0.5 mg/dl (44.2 *μ*mol/l) to diagnose CIN. This later definition is slightly anomalous as only those with pre-existing kidney disease (creatinine > 2.0 mg/dl) can have a greater than 0.5 mg/dl elevation in creatinine that is less than 25%. We have recommended that all CIN studies adopt the AKIN

We investigated whether a continuous variable measure of kidney function, the Relative Average Creatinine (RAVC), performed better than the RIFLE and AKIN categorical definitions as an outcome variable in AKI prevention or intervention trials. The RAVC is the integral of the area under the plasma creatinine curve above baseline creatine divided by

*Cbt*

We created a population of 10,000 Virtual-In-Patients (VIPs) whose baseline creatinine and changes in GFR were based on real ICU populations. Placebo controlled trials were simulated by randomly assigning half the VIPs to treatments which ameliorated loss of renal

 *t* 0

(*Cn*+<sup>1</sup> − *Cb*)+(*Cn* − *Cb*)

*RAVC*(%) = <sup>100</sup>

0*<n<N*−1

which in practice is calculated using the trapezoidal rule (Figure 3):

*Cb*(*tN* <sup>−</sup> *<sup>t</sup>*1) <sup>∑</sup>

RIFLE AKIN RIFLE and AKIN

Stage Creatinine increase

50%

 0.5 mg/dl and above 4.0 mg/dl

(*C*(*t*) − *Cb*)*dt* (9)

<sup>2</sup> (*tn*+<sup>1</sup> <sup>−</sup> *tn*) (10)

Urine output

*<* 0.5 mg/kg/h

*<* 0.3 mg/kg/h for 24 h or anuria for 12 h

for 6 h

for 12 h

to release a new consensus definition which combines the two definitions.

GFR decrease

\*,\*\* corrected from 25% and 75% see Pickering & Endre (2009a)

Risk (R) 50% *>* 33.3%∗ I 0.3 mg/dl or

Stage Creatinine

Failure (F) 200% or

increase

 0.5 mg/dl and above 4.0 mg/dl

Table 1. Severity Staging: Consensus definitions

**3.3 A continuum needing continuous variables**

or RIFLE definition of AKI (Endre & Pickering (2010)).

the total time and baseline creatinine:

*RAVC*(%) = <sup>100</sup>

where *N* is the number of creatinine measurements.

Fig. 3. The Relative Average Creatinine by the trapezoidal rule is the sum of the areas A to E divided by (*t*<sup>6</sup> − *t*1)(following equation 10).

function (ie reduced the reduction in GFR). The more efficacious the treatment, the less the decrease in GFR. Creatinine profiles were calculated using equation 6 (Pickering et al. (2009)). AKIN, RIFLE and RAVC as outcome variables were compared. At low treatment efficacy, the categorical outcomes underestimated and at high treatment efficacy overestimated the effect of treatment. These effects were exaggerated when the population contained a high proportion of patients with more severe AKI. The RAVC, on the other hand, responded in an almost linear fashion across treatment efficacies. Importantly, when the efficacy was low it was best able to distinguish between placebo and treatment arms. The advantage of the RAVC over the categorial metrics is two fold, first it includes the effect of treatment on those patients who had mild kidney injury which in normal circumstances would not exceed the diagnostic threshold for AKI according to a categorical definition, but which may result in a small increase (eg 20-30%) in creatinine and, second, it measures function over a (pre-determined) time period. As discussed previously, the length of time creatinine is elevated is independently of the maximum elevation associated with mortality. The RAVC captures both severity *and* duration of injury and was used as the primary outcome in the EARLYARF trial, which was the first randomised control trial to use an injury biomarker to triage patients to placebo or high-dose erythropoietin (Endre et al. (2010)).

#### **3.4 The baseline issue**

All creatinine based definitions of AKI and the RAVC depend on knowing the normal, or "baseline", plasma creatinine for each individual. For patients undergoing elective surgery

Situation Timing of baseline creatinine sample

2 All Pre-hospital outpatient or prior admission

developed during that period.

will not (yet) have become elevated.

The Metamorphosis of Acute Renal Failure to Acute Kidney Injury 137

of experimental studies of nephrotoxic AKI biomarkers, serum creatinine and blood urea nitrogen (BUN) were the poorest predictors of histologically determined injury compared with numerous urinary biomarkers (Dieterle et al. (2010); Ozer et al. (2010); Vaidya et al. (2010); Yu et al. (2010)). In what may turn out to be a seminal publication Haase et al. (2011) demonstrated that across 10 studies patients with a positive NGAL, yet negative plasma creatinine, for AKI had worse outcomes (mortality, need for dialysis, and ICU length of stay)

Biomarker clinical utility is most often quantified by the area under the receiver operator characteristic curve (AUC or c-statistic, see Box 3.3). The AUC is crude estimation of the ability of the biomarker to distinguish between those with and without AKI. However, a high AUC does not necessarily imply clinical utility. Clinical utility depends on what alternative biomarkers there are, what treatments are available, and the risks and costs involved with false negatives or false positives. The calculation of the sensitivity, specificity and particularly the negative and positive predictive values at either an established cut-off from the literature, or a cut-off chosen for clinical reasons, or derived mathematically from the ROC. The latter is usually either the cut-off closest to a Sensitivity and Specificity of 1 or the Youden index. Comparing AUCs between different studies is problematical. AUCs for AKI are highly dependent on the AKI definition. Typically a definition requiring greater injury (eg RIFLE sustained for 24-h, compared with an increase of 0.3 mg/dl within 48-h) will result in higher AUCs. Within a study when more than one biomarker is being measured they should be compared using the method of DeLong (DeLong et al. (1988)). One biomarker should not be described as "better" than another unless the difference between the two is statistically

Preferably this is between 7 and 90 days prior to admission. 7 days avoids a period which may reflect changing renal function. Less than 90 days is preferable because 90 days is the period usually used to diagnose CKD. Up to 365 days may be used if there is little likelihood of CKD having

If a pre-hospital value is not available and the time between insult and the first measurement is short (< 2h). Within a short time frame the creatinine

If a pre-hospital value is not available and the time between insult and the first measurement is short (< 2h) and crush-injuries are not involved.

Within 90 days of admission to avoid capturing the development of CKD.

1 Elective surgery Prior to surgery

3 Cardiac Arrest First hospital

4 Trauma First hospital

significant at *p <* 0.05.

5 Post ICU discharge

6 Lowest of first hospital or final ICU

Table 2. Hierarchical determination of baseline creatinine

than those with both negative NGAL and negative creatinine.

this is easily obtainable prior to surgery. However, about half of the patients entering the ICU have no previous record of plasma creatinine to serve as a baseline. For trials this requires a retrospective determination of renal function. The ADQI recommend assuming a normal (eg 75 ml/min) GFR for all these patients and "back-calculating" a plasma creatinine using the MDRD equation. Unfortunately, this has proved erroneous. We showed that in our VIP population and in an ICU population that this approach seriously overestimates the proportion of patients with AKI using either the AKIN or RIFLE definitions (Pickering & Endre (2010); Pickering et al. (2009)). Randomly assigning a baseline creatinine produced just as accurate results as back-calculation. In a population already with AKI the presence of patients with CKD was seen to be driving the overestimation (Bagshaw et al. (2009)). Using the emergency department creatinine as an alternative baseline underestimated AKI (AKIN) and lowered the sensitivity (Siew et al. (2010)).

In the EARLYARF trial we overcame this problem by using an adjudicated hierarchical approach to choose in each patient the measured plasma creatinine that best represented normal renal function. Outpatient plasma creatinine prior to admission were considered the most likely to represent true baseline function. Whilst CKD can be diagnosed over three months, up to twelve months appears to be a reasonable time period prior to admission in which to ascertain baseline creatinine as it reduces misclassification of AKI (Lafrance & Miller (2010)). Amongst patients with no pre-admission creatinine measurement, a post-discharge measurement is the next best option if it is stable. Increasing creatinine may be indicative of developing CKD. Hence, it is preferable that a post-discharge creatinine is within three months of the insult. As a last resort the lower of the first hospital or final hospital (when there is recovery) may be used. We have presented our recommended hierarchical approach in table 2. This differs from the earlier approach in that we now consider that for cardiac arrest and trauma patients where there is no baseline prior to admission available the first hospital sample is likely to be the best estimate of baseline function if it is measured close to the time of renal insult. In our experience this is less than 2-h for most cardiac arrest and trauma patients. For the creatinine to have increased considerably in this time frame the loss of GFR would have had to have been substantial.

#### **3.5 Quantifying function in a dilute environment**

Fluid resuscitation dilutes plasma creatinine concentrations, which in turn may lead to delayed diagnosis or severity underestimation. This may explain why the only successful AKI intervention in the ICU has been early consultation with a nephrologist (Mehta et al. (2002)). The effect of fluids may be estimated by adjusting plasma creatinine for fluid balance (Macedo et al. (2010)):

$$\mathcal{C}\_{\text{adjusted}} = \mathcal{C}\_{\text{measured}} \times \frac{\text{admission weight (kg)} \times 0.6 + \sum \text{daily cumulative fluid balance (L)}}{\text{admission weight (kg)} \times 0.6} \tag{11}$$

#### **3.6 Injury meets function**

Current evaluation of novel biomarkers of renal injury is largely confined to evaluation of their performance to predict increases in plasma creatinine which lead to a diagnosis of AKI according to RIFLE or AKIN. This runs the risk of missing significant injury because creatinine did not increase beyond the diagnostic threshold of 50% or 0.3 mg/dl. In a large series 12 Will-be-set-by-IN-TECH

this is easily obtainable prior to surgery. However, about half of the patients entering the ICU have no previous record of plasma creatinine to serve as a baseline. For trials this requires a retrospective determination of renal function. The ADQI recommend assuming a normal (eg 75 ml/min) GFR for all these patients and "back-calculating" a plasma creatinine using the MDRD equation. Unfortunately, this has proved erroneous. We showed that in our VIP population and in an ICU population that this approach seriously overestimates the proportion of patients with AKI using either the AKIN or RIFLE definitions (Pickering & Endre (2010); Pickering et al. (2009)). Randomly assigning a baseline creatinine produced just as accurate results as back-calculation. In a population already with AKI the presence of patients with CKD was seen to be driving the overestimation (Bagshaw et al. (2009)). Using the emergency department creatinine as an alternative baseline underestimated AKI (AKIN)

In the EARLYARF trial we overcame this problem by using an adjudicated hierarchical approach to choose in each patient the measured plasma creatinine that best represented normal renal function. Outpatient plasma creatinine prior to admission were considered the most likely to represent true baseline function. Whilst CKD can be diagnosed over three months, up to twelve months appears to be a reasonable time period prior to admission in which to ascertain baseline creatinine as it reduces misclassification of AKI (Lafrance & Miller (2010)). Amongst patients with no pre-admission creatinine measurement, a post-discharge measurement is the next best option if it is stable. Increasing creatinine may be indicative of developing CKD. Hence, it is preferable that a post-discharge creatinine is within three months of the insult. As a last resort the lower of the first hospital or final hospital (when there is recovery) may be used. We have presented our recommended hierarchical approach in table 2. This differs from the earlier approach in that we now consider that for cardiac arrest and trauma patients where there is no baseline prior to admission available the first hospital sample is likely to be the best estimate of baseline function if it is measured close to the time of renal insult. In our experience this is less than 2-h for most cardiac arrest and trauma patients. For the creatinine to have increased considerably in this time frame the loss of GFR would

Fluid resuscitation dilutes plasma creatinine concentrations, which in turn may lead to delayed diagnosis or severity underestimation. This may explain why the only successful AKI intervention in the ICU has been early consultation with a nephrologist (Mehta et al. (2002)). The effect of fluids may be estimated by adjusting plasma creatinine for fluid balance

*Cadjusted* <sup>=</sup> *Cmeasured* <sup>×</sup> admission weight (kg) <sup>×</sup> 0.6 <sup>+</sup> <sup>∑</sup> daily cumulative fluid balance (L)

Current evaluation of novel biomarkers of renal injury is largely confined to evaluation of their performance to predict increases in plasma creatinine which lead to a diagnosis of AKI according to RIFLE or AKIN. This runs the risk of missing significant injury because creatinine did not increase beyond the diagnostic threshold of 50% or 0.3 mg/dl. In a large series

admission weight (kg) × 0.6

(11)

and lowered the sensitivity (Siew et al. (2010)).

have had to have been substantial.

(Macedo et al. (2010)):

**3.6 Injury meets function**

**3.5 Quantifying function in a dilute environment**


#### Table 2. Hierarchical determination of baseline creatinine

of experimental studies of nephrotoxic AKI biomarkers, serum creatinine and blood urea nitrogen (BUN) were the poorest predictors of histologically determined injury compared with numerous urinary biomarkers (Dieterle et al. (2010); Ozer et al. (2010); Vaidya et al. (2010); Yu et al. (2010)). In what may turn out to be a seminal publication Haase et al. (2011) demonstrated that across 10 studies patients with a positive NGAL, yet negative plasma creatinine, for AKI had worse outcomes (mortality, need for dialysis, and ICU length of stay) than those with both negative NGAL and negative creatinine.

Biomarker clinical utility is most often quantified by the area under the receiver operator characteristic curve (AUC or c-statistic, see Box 3.3). The AUC is crude estimation of the ability of the biomarker to distinguish between those with and without AKI. However, a high AUC does not necessarily imply clinical utility. Clinical utility depends on what alternative biomarkers there are, what treatments are available, and the risks and costs involved with false negatives or false positives. The calculation of the sensitivity, specificity and particularly the negative and positive predictive values at either an established cut-off from the literature, or a cut-off chosen for clinical reasons, or derived mathematically from the ROC. The latter is usually either the cut-off closest to a Sensitivity and Specificity of 1 or the Youden index.

Comparing AUCs between different studies is problematical. AUCs for AKI are highly dependent on the AKI definition. Typically a definition requiring greater injury (eg RIFLE sustained for 24-h, compared with an increase of 0.3 mg/dl within 48-h) will result in higher AUCs. Within a study when more than one biomarker is being measured they should be compared using the method of DeLong (DeLong et al. (1988)). One biomarker should not be described as "better" than another unless the difference between the two is statistically significant at *p <* 0.05.

It is anticipated that panels of biomarkers will be needed to diagnose AKI in heterogeneous populations with multiple AKI aetiologies. Attempts to assess a panel of biomarkers usually involve logistic regression models with two or more biomarkers measured at the same time point. The EARLYARF trial has demonstrated that the efficacy of any one particular biomarker is very much time dependent, and that each biomarker has its own "window of opportunity" during which it has have diagnostic utility (Endre et al. (2011)). Logistic regression models are not a suitable approach for assessment of biomarker panels because of this. Until we know the time courses of biomarkers much better, an "either and/or" approach is likely to yield greater results. For example, in an ICU population either an elevated GGT and/or an elevated NGAL

The Metamorphosis of Acute Renal Failure to Acute Kidney Injury 139

There is a need to move away from assessing injury biomarkers only in relation to function. All trials should report mortality data, even if the incidence is too low for statistical analysis. This will facilitate later meta-analysis of the relationship between a biomarker and mortality. The EARLYARF trial paved the way for future trials of early intervention based on an elevated biomarker. Since the inception of that trial new biomarkers with rapid assay turn-around (necessary in an early intervention trial) have become available. Plasma and urinary NGAL, and KIM-1 head the list along with urinary Cystatin C and GGT which are already routinely available in many hospital laboratories. It is anticipated that the next early intervention trial

Changes in GFR and water handling will change urinary biomarker concentrations independent of injury. Normalising biomarkers to urinary creatinine has been proposed to account for these effects. This process also amplifies the signal soon after a decline in function (Waikar et al. (2010)). This may be advantageous in an early intervention trial if the threshold for intervention is set high enough, but it may distort the analysis of biomarker performance in a biomarker performance study. Whilst there is no consensus on whether biomarkers should or should not be normalised to urinary creatinine we recommend reporting both normalised

Table 3 presents a summary of practical measures to take into account when planning AKI

Most epidemiological studies are retrospective and face the difficulty of missing data, particularly baseline creatinine data. This was discussed in section 3.4. It is important to quantify the severity of AKI, as more severe AKI is likely to have greater long term impact on health resources. Where possible, data on the duration of AKI as well as severity should be captured as this is an independent predictor of outcome. Three areas of epidemiology lack data. First, there are comparatively few good epidemiology studies in countries other than in Europe, North America or Australasia (Cerda et al. (2008)). The incidence of AKI in countries with large populations such as China, India, Indonesia, Nigeria has significance beyond their own borders. Some countries have numbers of particular AKI aetiologies which, if well studied, could provide useful data world-wide. For example, with anti-retroviral therapy

epidemiology, biomarker efficacy studies or prevention or intervention trials.

may be considered diagnostic.

and non-normalised results.

**4. Practical considerations**

**4.1 Epidemiology**

will use one or more of these biomarkers.

**3.7 Quantifying injury in a dilute environment**

#### *Box 3.3.* The Area Under the Curve (AUC)

The receiver operator characteristic curve (ROC) is a plot of sensitivity verse 1-specificity. An AUC of 1 means the biomarker always discriminates between patients with and without the disease (no false negatives and no false positives). An AUC of 0.5 is equivalent to a coin toss. Whilst an AUC of less than 0.5 means the reciprocal of the biomarker is diagnostic. Some statistics packages, however, will always express the AUC as greater than 0.5 by inverting the biomarker concentration where necessary. The AUC from a study is strictly speaking an estimate of the AUC of the population, therefore it should **always** be presented with appropriate confidence intervals (usually 95%). The 95% confidence interval is ±1.96 times the standard error of a proportion:

$$
\widehat{ALU\mathbb{C}} - 1.96\sqrt{\widehat{ALU\mathbb{C}}(1-\widehat{AUG})/n} \quad \text{to} \quad \widehat{AUG} + 1.96\sqrt{\widehat{AUG}(1-\widehat{AUG})/n}
$$

where *n* is the sample size and *AUC* the estimated AUC. This is equation is adequate for sample sizes *>* 30, but for smaller sample sizes a bootstrapping should be used. Only an AUC for which the lower limit of its 95% confidence interval is greater than 0.5 may be described as diagnostic (or prognistic).

Figure 4 is an example of a ROC from the EARLYARF trial (Endre et al. (2011)). It shows the ability of NGAL to diagnose AKI when the sample was taken between 12 and 36 hours following renal insult. The AUC lower limit of the 95% confidence interval, 0.62, is greater than 0.5, therefore it is diagnostic.

14 Will-be-set-by-IN-TECH

The receiver operator characteristic curve (ROC) is a plot of sensitivity verse 1-specificity. An AUC of 1 means the biomarker always discriminates between patients with and without the disease (no false negatives and no false positives). An AUC of 0.5 is equivalent to a coin toss. Whilst an AUC of less than 0.5 means the reciprocal of the biomarker is diagnostic. Some statistics packages, however, will always express the AUC as greater than 0.5 by inverting the biomarker concentration where necessary. The AUC from a study is strictly speaking an estimate of the AUC of the population, therefore it should **always** be presented with appropriate confidence intervals (usually 95%). The 95% confidence interval

)/*n* to *AUC*

*>* 30, but for smaller sample sizes a bootstrapping should be used. Only an AUC for which the lower limit of its 95% confidence interval is greater than 0.5 may be described as diagnostic (or prognistic).

Figure 4 is an example of a ROC from the EARLYARF trial (Endre et al. (2011)). It shows the ability of NGAL to diagnose AKI when the sample was taken between 12 and 36 hours following renal insult. The AUC lower limit of the 95% confidence interval, 0.62, is greater than 0.5, therefore it is diagnostic.

**0.0 0.2 0.4 0.6 0.8 1.0**

**1 - Specificity**

**AUC = 0.72 (95%CI: 0.62 to 0.81)**

+ 1.96

 *AUC* (<sup>1</sup> <sup>−</sup> *AUC*

the estimated AUC. This is equation is adequate for sample sizes

)/*n*

*Box 3.3.* The Area Under the Curve (AUC)

is ±1.96 times the standard error of a proportion:

**0.0**

Fig. 4. An example of a receiver operator characteristic curve

**0.2**

**0.4**

**0.6**

**Sensitivity**

**0.8**

**1.0**

 *AUC* (<sup>1</sup> <sup>−</sup> *AUC*

*AUC* <sup>−</sup> 1.96

where *n* is the sample size and *AUC*

It is anticipated that panels of biomarkers will be needed to diagnose AKI in heterogeneous populations with multiple AKI aetiologies. Attempts to assess a panel of biomarkers usually involve logistic regression models with two or more biomarkers measured at the same time point. The EARLYARF trial has demonstrated that the efficacy of any one particular biomarker is very much time dependent, and that each biomarker has its own "window of opportunity" during which it has have diagnostic utility (Endre et al. (2011)). Logistic regression models are not a suitable approach for assessment of biomarker panels because of this. Until we know the time courses of biomarkers much better, an "either and/or" approach is likely to yield greater results. For example, in an ICU population either an elevated GGT and/or an elevated NGAL may be considered diagnostic.

There is a need to move away from assessing injury biomarkers only in relation to function. All trials should report mortality data, even if the incidence is too low for statistical analysis. This will facilitate later meta-analysis of the relationship between a biomarker and mortality. The EARLYARF trial paved the way for future trials of early intervention based on an elevated biomarker. Since the inception of that trial new biomarkers with rapid assay turn-around (necessary in an early intervention trial) have become available. Plasma and urinary NGAL, and KIM-1 head the list along with urinary Cystatin C and GGT which are already routinely available in many hospital laboratories. It is anticipated that the next early intervention trial will use one or more of these biomarkers.
