A comparison of three methods to estimate baseline creatinine for RIFLE classification

Transcription

1 Nephrol Dial Transplant (2010) 25: doi: /ndt/gfp766 Advance Access publication 25 January 2010 Original Articles A comparison of three methods to estimate baseline creatinine for RIFLE classification Jakub Závada 1,2, Eric Hoste 1,3, Rodrigo Cartin-Ceba 4, Paolo Calzavacca 5, Ognjen Gajic 4, Gilles Clermont 1, Rinaldo Bellomo 5, John A. Kellum 1 and for the AKI6 investigators 1 The CRISMA (Clinical Research, Investigation, and Systems Modeling of Acute Illness) Laboratory, Department of Critical Care Medicine, University of Pittsburgh, School of Medicine, Pittsburgh, USA, 2 Department of Nephrology, Charles University in Prague, First Faculty of Medicine and General University Hospital, Prague, Czech Republic, 3 Intensive Care Unit, Ghent University Hospital, De Pintelaan 185, 9000 Ghent, Belgium, 4 Department of Internal Medicine, Division of Pulmonary and Critical Care Medicine, Mayo Epidemiology and Translational Research in Intensive Care (METRIC) and Mayo Clinic College of Medicine, Rochester, Minnesota., USA and 5 Department of Intensive Care and Department of Medicine, Austin Hospital, Melbourne, Australia Correspondence and offprint requests to: John A. Kellum; kellumja@ccm.upmc.edu Abstract Background. A pre-morbid baseline creatinine is required in order to diagnose and stage acute kidney injury (AKI) using the RIFLE classification. Estimation of baseline creatinine by solving the Modification of Diet in Renal Disease (MDRD) equation assuming a glomerular filtration rate of 75 ml/min/1.73 m 2 has been widely used but never validated. Methods. We analysed four cohorts of intensive care unit (ICU) patients from three centres (two from Pittsburgh and one from Mayo and Austin). Three cohorts consisted of preselected patients without AKI (Pittsburgh 1 n = 1048, Mayo n = 737, Austin n = 333), and measured creatinine values in these cohorts were taken to represent baseline creatinine values. The last cohort (Pittsburgh 2 n = 468) consisted of unselected ICU patients with baseline creatinine values recorded within 1 year before ICU admission. Using the Pittsburgh 1 cohort, we derived an equation using the same anthropometric variables as the MDRD equation: baseline creatinine = (if female) (if black) age (in years). We then compared measured creatinine in the Mayo and Austin cohorts and recorded creatinine in the Pittsburgh 2 cohort to the estimated creatinine from: (i) the MDRD equation; (ii) our new equation; (iii) a genderfixed creatinine of 0.8 mg/dl for females and 1.0 mg/dl for males. Results. Using any of the three methods, the median absolute error of the estimates was of the order of mg/dl, and overall accuracy was similar. When the definition of AKI was limited to the severity grades of Injury and Failure, all three methods were able to generate 78 90% reliable results for preselected normal range cohorts, and 63 70% for the unselected cohort of ICU patients. Conclusions. Estimates of incidence of AKI in the critically ill using RIFLE classification can be affected by the bias and limited accuracy of methods to estimate baseline creatinine. Whenever possible, recorded creatinine values should be used as a reference of baseline. The use of the MDRD equation to estimate baseline creatinine when it is unknown may over- or underestimate some mild (Risk) AKI cases but is unlikely to misclassify patients in Injury and Failure. Keywords: acute kidney injury; baseline creatinine; estimation method; RIFLE classification Introduction Acute kidney injury (AKI) is a common condition in critical illness and is associated with a significantly increased risk of death. AKI is classified using RIFLE criteria [1]; although a slight modification has been recently proposed [2]. RIFLE classification provides three grades of increasing severity of AKI (Risk, Injury, Failure) and two outcome classes (Loss, End-stage kidney disease). The severity grades for AKI are based either on urine output or changes in serum creatinine from baseline. However, many patients may present with an elevated creatinine but without any baseline measure of renal function. When they initially proposed the RIFLE criteria, the Acute Dialysis Quality Initiative (ADQI) also proposed that, for patients without recognized chronic kidney disease, an estimated baseline creatinine could be obtained by solving the four-variable Modification of Diet in Renal Disease (MDRD) equation for a low normal glomerular filtration rate (GFR) of 75 ml/min/1.73 m 2 [1]. Although this approach has been used in prior studies [3,5,6,8,9,12,13,16,17], it has never been validated. As of 2008, more than patients have been screened and evaluated for AKI using the RIFLE criteria [3 18]. A substantial number of the larger studies evaluating AKI have relied on the creatinine criteria (while the The Author Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved. For Permissions, please journals.permissions@oxfordjournals.org

2 3912 J. Závada et al. Table 1. Overview of the approaches to determine baseline creatinine in the application of RIFLE classification in previous studies Study by No of patients analysed Multi/single centre Criteria used Method to determine baseline creatinine % recorded % estimated Bagshaw [6] Multi cr+uo Estimated by MDRD formula Ostermann [9] Multi cr Estimated by MDRD formula Uchino [12] Single cr Retrieved from hospital database, N/A N/A Bell [16] 8152 Single cr+uo Retrieved from hospital database, N/A N/A Hoste [13] 5383 Single cr+uo Estimated by MDRD formula, N/A N/A or admission creatinine value, whatever was lower Ali [4] 5321 Multi cr Retrieved from hospital database, or admission creatinine value Cruz [5] 2164 Multi cr+uo Retrieved from hospital database, Perez-Valdivieso [17] 1008 Single cr Estimated by MDRD formula Kuitunen [14] 813 Single cr+uo Preoperative value Coca [18] 304 Single cr The lowest s-creatinine value in the first five hospital days Arnaoutakis [7] 267 Single N/A N/A N/A N/A Abosaif [15] 247 Single cr+uo Retrieved from hospital database, or admission creatinine value Maccariello [3] 214 Multi cr+uo Retrieved from hospital database, N/A N/A Jeng [8] 134 Single cr+uo Admission creatinine value, cr creatinine criteria, uo urine output criteria, N/A not available. measurements of urine output were not available), and the majority of these studies used the calculated estimates of baseline creatinine to apply the RIFLE classification (see Table 1.). Hence, most current data concerning AKI defined by RIFLE criteria are based on the assumption that baseline creatinine for respective patients could be estimated by solving the MDRD equation using a low normal GFR of 75 ml/min/1.73 m 2. However, this assumption may not be valid. The MDRD formula was derived from a population of outpatients with renal disease [19]. It is considered to perform poorly in patients with GFR above 60 ml/min/1.73 m 2 [20], in severely ill hospitalized patients [21] and in malnourished patients [22]. There are few data showing the results of GFRs (either determined by estimation equations or direct measurement) in critically ill patients. Thus, we sought to test the performance of the MDRDbased approach and explore alternative methods of estimating unknown baseline creatinine using data from three geographically distinct cohorts of critically ill patients. Subjects and methods Sources of data Derivation set: Using a previously published database [13] from the University of Pittsburgh Medical Center, we constructed a cohort of patients with creatinine values within the normal range (using the definition of normal range according to the manufacturer of the creatinine assay used in Pittsburgh centre, or s-creatinine <1.5 mg/dl for males, s-creatinine <1.2 mg/dl for females) and not changing more than 0.2 mg/dl (serum creatinine in milligram per decilitre may be converted to micromole per litre by multiplying by 88.4) or 30%, whichever is larger, during the ICU stay (n = 1048). Thus, by definition, these patients did not have AKI by creatinine criteria (increase in serum creatinine <0.3 mg/dl and/or by 50%), and their measured creatinine values were taken to represent the baseline creatinine values. Validation set 1: Using the same criteria, we constructed two new cohorts including patients admitted between 1 January and 31 December 2006; one cohort from the Mayo Clinic (Rochester, Minnesota) (n = 737) and the other from Austin Hospital (Melbourne, Australia) (n = 333). Moreover, a prospectively created unselected cohort of ICU patients from the Pittsburgh Medical Center with baseline creatinine values (n = 468) recorded within 1 year before ICU admission was included as validation set 2. Estimation methods To predict the baseline creatinine value, three estimation methods were tested: Firstly, ( MDRD-based estimation method ) we solved the MDRD formula for GFR 75 ml/min/1.73 m 2 as suggested by ADQI: Serumcreatinine = ð75=½186 ðage 0:203 Þ ð0:742if femaleþ ð1:21if blackþš 0:887 : Secondly, ( three-variable equation ) using the same demographic and anthropometric variables included in MDRD formula (i.e. gender, age and race), we derived from the Pittsburgh non-aki cohort (derivation set) by means of multivariable linear regression analysis the following equation: Serumcreatinine = 0:74 0:2ðif femaleþ + 0:08ðif blackþ + 0:003 age ðinyearsþ Thirdly, we tested an average gender-fixed creatinine (derived from the entire non-aki cohort) of 1.0 mg/dl for males and 0.8 mg/dl for females. Bias, precision and accuracy Mean relative error and absolute error were used as indices of bias and precision of the estimates. Relative error was calculated as the estimate minus

3 A simple gender-based estimate of baseline creatinine 3913 the measured value and expressed as a percentage of measured creatinine values. Positive values thus reflected tendency of the prediction method to overestimate, while negative values tendency to underestimate the level of baseline creatinine. The absolute error was calculated as the absolute value of the estimate minus the measured value and expressed in milligram per decilitre. Accuracy was tested as follows: RIFLE severity grades are based on the multiplication (by a factor of 1.5, 2 or 3) of the baseline creatinine values. Whenever the estimated baseline creatinine value is lower than the real (but unknown) baseline, the level (or presence) of AKI tends to be overestimated. Conversely, whenever the estimated baseline is higher than the real baseline, the level (or presence) of AKI could be underestimated. Because of the multiplicative principle of RIFLE classification, it is convenient to think in terms of the ratio between the estimated and real baseline creatinine value. If estimated/measured creatinine is <1/2 (0.5), 2/3 (0.66) or <3/4 (0.75), a false positive AKI of the RIFLE severity class F, I or R, respectively, will occur. Whenever the estimated/measured ratio is more than 1, 4/3 (1.33) or 2, a false negative corresponding to R, I and F occurs. Any systematical bias in the prediction of the baseline function could be thus transformed in a systematical error of the RIFLE classification. Thus, we evaluated the theoretical accuracy of the RIFLE criteria based on the estimated creatinine values as the percentage of ratios between the estimated and measured creatinine values (estimated baseline creatinine/measured baseline creatinine) within predetermined intervals of and ( ). These intervals ensure that subsequent RIFLE classification would not miss any patient with at least 1.5- or 2-fold increase from their true baseline nor ascribe AKI (grade Risk or more, or Injury or more) to any patients with a creatinine increase of <1.5- and 2-fold. Patients with ratios <0.75 were considered false positive for AKI and >1.00 falsely negative for AKI. Determination of plasma creatinine Plasma concentrations of creatinine were determined in Pittsburgh by an enzymatic assay (Vitros 950), in Mayo Clinic by an enzymatic assay (Roche/Hitachi 747) and in Austin Hospital by Jaffe colorimetric method (Beckman Synchron LX). Re-calibration of original creatinine values to isotope-dilution mass spectrometry-traceable values To enable comparison between centres, we recalibrated the measured creatinine values to isotope-dilution mass spectrometry (IDMS)-traceable values using creatinine standardization correlation equations provided by the manufacturer of the respective assays. We have also checked these re-calibration equations by externally available data [32, 33]. To recalibrate the values of creatinine by Jaffe Beckman LX assay, we used the equation: Recalibrated serum creatinine = measured serum creatinine (in milligram per decilitre). To recalibrate the values of creatinine by the Vitros 950 enzymatic method, we used the equation: Recalibrated serum creatinine = measured serum creatinine (in milligram per decilitre). The values measured by Roche/Hitachi 747 enzymatic assay are considered to be very close to IDMS values, and were not recalibrated. [33-34]. For IDMS-recalibrated creatinine values, we used the reexpressed MDRD formula: serum creatinine = (75 / [175 (age ) (0.742 if female) (1.21 if black)]) and re-expressed version of the three-variable equation: serum creatinine = (if female) (if black) age (in years). Statistical evaluation The central tendency for continuous data is expressed as the mean ± standard deviation or the median (interquartile range). We tested continuous variables for normality by distribution plots and the Kolmogorov Smirnov test. We compared means using the Student's t-test when normally distributed, and the Mann Whitney U-test when not. Comparisons across multiple groups were performed using the F-test, with Bonferroni correction for multiple comparisons. When data were not normally distributed, we used the Kruskal Wallis H analysis of variance test. Chisquare test was used to compare categorical outcomes between groups. A value of P < 0.05 was considered significant. In order to develop an estimation method based on the available anthropometric values, we used both stepwise and simultaneous multivariable linear regression analysis. All analyses were performed on a statistical computer package (SPSS version 15, Illinois, USA). Table 2. Characteristics of the derivation and validation sets Serum creatinine Centre Number of patients (mg/dl) (mg/dl) (ml/min/1.73 m 2 ) (ml/min/1.73 m 2 ) (years) Female Black GFR calculated by re-expressed MDRD equation Age GFR calculated by original MDRD equation Serum IDMS creatinine Pittsburgh 1 a (0.75, 0.95) 0.69 (0.64, 0.83) 93 (76, 112) 103 (83, 126) 61 (48, 72) 40% 8% Mayo b (0.75, 1.05) 0.90 (0.75, 1.05) 80 (67, 98) 75 (64, 92) 65 (51, 75) 44% 0.50% Austin b (0.77, 1.01) 0.82 (0.71, 0.95) 85 (73, 101) 86 (74, 104) 60 (46, 71) 38% 0% Pittsburgh 2 b (0.80, 1.20) 0.90 (0.72, 0.94) 68 (56, 90) 74 (61, 97) 67 (53, 76) 41% 7.7% Continuous variables are expressed as medians (interquartile range). Serum creatinine values in Pittsburgh 1, Mayo and Austin cohorts refer to the average of measured values during the ICU stay, while those in Pittsburgh 2 cohort refer to the recorded creatinine values within 1 year before ICU admission. Note: Serum creatinine in milligram per decilitre may be converted to micromole per litre by multiplying by a Derivation set. b Validation sets.

4 3914 J. Závada et al. Table 3a. Performance of the (i) MDRD-based method, (ii) three-variable equation and (iii) gender-fixed approach with respect to age group (tested on validation set 1) Age groups (years) Method 0 24 (n = 42) (n = 216) (n = 553) (n = 259) P* i Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 38 (22, 60) 21 (7, 42) 5 ( 8, 24) 4 ( 13, 8) Absolute error (mg/dl) 0.33 (0.20, 0.47) 0.19 (0.09, 0.31) 0.15 (0.06, 0.22) 0.12 (0.06, 0.20) ii Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 15 ( 29, 3) 9 ( 20, 3) 7 ( 20, 9) 6 ( 15, 6) Absolute error (mg/dl) 0.14 (0.05, 0.27) 0.11 (0.05, 0.20) 0.13 (0.06, 0.23) 0.11 (0.06, 0.20) 0.06 iii Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 10 ( 4, 24) 7 ( 6, 23) 0 ( 13, 17) 5 ( 13, 7) Absolute error (mg/dl) 0.15 (0.05, 020) 0.15 (0.05, 0.20) 0.15 (0.05, 0.20) 0.10 (0.05, 0.220) 0.27 Accuracy within (0.75 1) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades, including Risk. Accuracy within ( ) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades Injury or Failure. Relative error is a relative (%) difference between estimated and measured creatinine, and absolute error is an absolute value of difference between estimated and measured creatinine (in milligram per decilitre). Continuous variables are expressed as medians (interquartile range). *P indicates the result of the chi-square/trend test for the difference in categorical outcomes (accuracy) and Kruskal Wallis test for the difference in continuous outcomes (errors) between age groups and genders. Note: Serum creatinine in milligram per decilitre may be converted to micromole per litre by multiplying by Table 3b. Performance of the (i) MDRD-based method, (ii) three-variable equation and (iii) gender-fixed approach with respect to gender (tested on validation set 1) Method Parameter Male (n = 613) Female (n = 457) P* Gender i Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 9 ( 7, 28) 4 ( 8, 24) Absolute error (mg/dl) 0.16 (0.08, 0.27) 0.12 (0.05, 0.21) ii Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 6 ( 19, 8) 9 ( 19, 3) Absolute error (mg/dl) 0.14 (0.06, 0.24) 0.10 (0.06, 0.19) iii Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 0 ( 13, 17) 0 ( 11, 23) Absolute error (mg/dl) 0.15 (0.05, 0.24) 0.10 (0.05, 0.15) Accuracy within (0.75 1) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades, including Risk. Accuracy within ( ) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades Injury or Failure. Relative error is a relative (%) difference between estimated and measured creatinine, and absolute error is an absolute value of difference between estimated and measured creatinine (in milligram per decilitre). Continuous variables are expressed as medians (interquartile range). *P indicates the result of the chi-square/trend test for the difference in categorical outcomes (accuracy) and Kruskal Wallis test for the difference in continuous outcomes (errors) between age groups and genders. Note: Serum creatinine in milligram per decilitre may be converted to micromole per litre by multiplying by Results Descriptive statistics Patient characteristics of the derivation and validation datasets are shown in Table 2. Both the mean creatinine value and the mean MDRD-GFR differed significantly between centres. After indirect recalibration efforts, the difference between these mean values persisted or even increased. Blacks were under-represented in the Pittsburgh cohorts and virtually absent in the cohorts from Mayo and Austin hospital. Bias of the estimates Median relative error and accuracy (percentage of acceptably accurate estimates) of the estimation methods for particular age quartiles are shown in Table 3a. As indicated, none of the estimation methods performed equally across all age groups. However, Method i (MDRD-based) was most heavily biased by age, overestimating the level of serum creatinine by almost 40% in the young (below 25 years). Although the relative errors of Method i (MDRDbased) and ii (three-variable equation) were slightly (but statistically significantly) different between both genders

5 A simple gender-based estimate of baseline creatinine 3915 Table 3c. Performance of the (i) MDRD-based method, (ii) three-variable equation and (iii) gender-fixed approach with respect to race (tested on validation set 2) Race Method Parameter White and other (n = 432) Black (n = 36) P* i Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 5 ( 15, 28) 1 ( 50, 20) 0.09 Absolute error (mg/dl) 0.21 (0.09, 0.39) 0.34 (0.13, 1.17) 0.02 ii Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 6 ( 25, 10) 18 ( 58, 3) 0.04 Absolute error (mg/dl) 0.17 (0.07, 0.34) 0.25 (0.10, 1.32) 0.01 iii Accuracy within (0.75 1) (%) Accuracy within ( ) (%) Relative error (%) 0 ( 17, 14) 17 ( 58, 3) Absolute error (mg/dl) 0.20 (0.10, 0.30) 0.30 (0.10, 1.40) Accuracy within (0.75 1) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades, including Risk. Accuracy within ( ) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades Injury or Failure. Relative error is a relative (%) difference between estimated and measured creatinine, and absolute error is an absolute value of difference between estimated and measured creatinine (in milligram per decilitre). Continuous variables are expressed as medians (interquartile range). *P indicates the result of the chi-square/trend test for the difference in categorical outcomes (accuracy) and Kruskal Wallis test for the difference in continuous outcomes (errors) between age groups and genders. Note: Serum creatinine in milligram per decilitre may be converted to micromole per litre by multiplying by (see Table 3b), and the relative error of Method ii and iii (gender-fixed creatinine) between black and other races (see Table 3c), these biases rarely translated into significant differences in respective accuracies. The relative error was significantly different (P < 0.005) for all the tested methods when applied across centres (see Table 4). Precision and accuracy of the estimates The median absolute error of the estimates was of the order of mg/dl using any of the three methods (see Tables 3a, 3b, 3c and 4). As shown in Table 4, the frequency of estimates sufficiently accurate to prevent ascribing false negative and false positive AKI according to RIFLE (using all severity grades including Risk) was rather low and in most occasions, <50%. When the definition of AKI was limited to the severity grades of Injury and Failure, all three methods were able to generate 78 90% reliable results for preselected normal range cohorts, and 63 70% for the unselected cohort of ICU patients. Overall, in terms of accuracy and precision, Method ii performed slightly better than Method i and iii. Discussion The primary finding of our study is that the methods used to ascertain the baseline creatinine have an important impact on subsequent detection and classification of AKI. We found that significant variation exists between centres in terms of mean serum creatinine values for patients, and this fact limits a generalizable method of estimating the baseline serum creatinine when it is unavailable. Most laboratories report values that deviate substantially from reference methods like IDMS [23]. Therefore, attention to the problems arising from inter-laboratory calibration differences must be taken into account in using any method estimating unknown baseline creatinines. To overcome the problem of inter-laboratory creatinine assay calibration differences, we asked the manufacturers of the respective assays to provide the IDMS standardization graphs and recalibration equations. The parameters of the recalibration equations were checked by externally available information [24,25] and found to fit well. It was beyond the scope of our study to perform direct comparison of the assays. However, even after recalibration, significant differences in terms of mean serum creatinine values between centres persisted or even increased. These results suggest either a failure of the indirect calibration efforts or real differences between cohorts. Neither the use of IDMS-recalibrated creatinine values or re-expressed versions of the MDRD formula or the newly derived equation helped to overcome centre-specific bias. Serum creatinine was chosen by the ADQI investigators as the main reference measure of kidney function, along with urine output, because it is readily available in clinical practice, and no better substitute is currently available. However, reliable information on baseline serum creatinine may not be available as many patients may have no previous creatinine measurement. This may be especially true in studies using large hospital databases. Furthermore, even if a previous creatinine can be found, if it was measured in the remote past, or by a different laboratory, the value may be influenced by progression of chronic disease, change in the muscle mass or different calibration of the creatinine assay. Because AKI may be present at admission (according to one study [13], 22% of patients had AKI at ICU admission), the initial creatinine values are also unreliable measures of baseline renal function. Since serum creatinine level reflects mainly the balance of creatinine released from the skeletal muscle and that removed by the kidneys, any estimation method should include variables generally

6 3916 J. Závada et al. Table 4. Theoretical accuracy of RIFLE classification based on the respective methods False positives (%) False negatives (%) Failure Injury Risk Accuracy (%) within Risk Injury Failure Centre I 0.5 I 0.66 I , , 1.33 I 1 I 1.33 I 2 Median relative error % (IQR) Median absolute error mg/dl (IQR) Estimation Method i: Back-calculated by original MDRD formula from GFR 75ml/min/1.73 m 2 Mayo ( 9, 26) 0.14 (0.07, 0.23) Austin ( 2, 29) 0.15 (0.06, 0.26) Pittsburgh ( 16, 27) 0.21 (0.09, 0.41) Estimation Method ii: Baseline creatinine = (if black) 0.2 (if female) age Mayo ( 20, 6) 0.13 (0.06, 0.23) Austin ( 17, 6) 0.11 (0.05, 0.20) Pittsburgh ( 27, 9) 0.18 (0.07, 0.36) Estimation Method iii: Baseline creatinine = gender-fixed creatinine (creatinine for males = 1.0, creatinine for females = 0.8) Mayo ( 13, 18) 0.15 (0.05, 0.20) Austin ( 7, 18) 0.11 (0.05, 0.19) Pittsburgh ( 20, 14) 0.20 (0.10, 0.40) Accuracy within (0.75 1) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades, including Risk. Accuracy within ( ) indicates the percentage of estimates resulting in no false positive or false negative RIFLE severity grades Injury or Failure. Relative error is a relative (%) difference between estimated and measured creatinine, and absolute error is an absolute value of difference between estimated and measured creatinine (in milligram per decilitre). Continuous variables are expressed as medians (interquartile range). *P indicates the result of the chi-square/ trend test for the difference in categorical outcomes (accuracy) and Kruskal Wallis test for the difference in continuous outcomes (errors) between age groups and gendersnote: Serum creatinine in milligram per decilitre may be converted to micromole per litre by multiplying by 88.4.

7 A simple gender-based estimate of baseline creatinine 3917 accepted to determinate kidney function or muscle mass. However, even when all the traditional factors are known, estimation of creatinine level may be quite inaccurate. In the study by McDonald [26] on 77 patients with chronic kidney disease (CKD), even when GFR was measured by inulin clearance, age, gender, weight and height were known, just 60% of the variance in the creatinine level was explained (R² = 0.6), and among these factors directly measured GFR accounted for >2/3 of their total predictive value. Even though this may not necessarily apply to patients without CKD, our data support the low level of correlation between measured and estimated creatinine values by any known method. We found that the mean MDRD-GFR of ICU patients without AKI or CKD is significantly >75 ml/min/1.73 m 2 and decreases steadily with age (by about 0.8 ml/min/ 1.73 m 2 per year). This is in good agreement with most population based studies indicating that GFR decreases by ml/min/1.73 m 2 per year of age [26 30]. Accordingly, unless the GFR value used in MDRD equation is corrected for age, significant systematical age-dependent bias ensues such that AKI is underestimated in younger patients, and possibly overestimated in the elderly. In adults, serum creatinine in an individual is relatively stable over time, while the age-dependent decrease in GFR is paralleled by a decrease in muscle mass and increase in creatinine excretion. The gender-fixed approach, thus, was less biased by age and in terms of accuracy and precision performed similarly as compared to the MDRD-based approach. When comparing precision and accuracy of the estimates, the results obtained by the three-variable equation performed better than the MDRD-based estimates across all validation sets. However, it is questionable, whether the demonstrated differences in terms of accuracy and precision represent meaningful improvement. Indeed, none of the alternative methods offered a consistent improvement in accuracy compared to MDRDbased estimates. A caveat to our study may be the fact that, as a reference for baseline creatinine in the derivation and one of the validation sets, we have taken the measured values from ICU patients with stable, normal range creatinine, and not from the recorded baselines of all ICU patients. To check for the effect of this selection bias, we used a second validation cohort consisting of unselected ICU patients with available recorded baseline creatinines. This second validation cohort contained 16% of patients with baseline creatinines outside normal range and was simulating more real-life conditions. Although a cohort restricted to recorded serum creatinine levels may be biased towards more frequent chronic illness or increased number of medical encounters, this cohort nevertheless includes patients with a more realistic spectrum of underlying illness (e.g. with CKD) and risk for AKI. Very recently, Bagshaw et al. [31] analysed the difference in overall occurrence of respective RIFLE classes in patients with severe AKI (by other definition) when using measured or estimated creatinines. In contrast, we sought to evaluate how estimates may bias RIFLE classification in each individual ICU patient. Although our approach is somewhat artificial (by simulating what would happen if AKI developed in patients who in reality had not met RIFLE creatinine criteria for AKI) and subject to the above-mentioned selection bias, we believe it could add further insights into the problem. In conclusion, we believe that any method to estimate unknown baseline creatinine solely from anthropometric variables is likely to be of limited accuracy particular when AKI is mild Risk only. We recommend that, whenever possible, baseline values should be obtained from prior laboratory testing. If recorded serum creatinine values are not available, the MDRD-based estimates can be used as a reference of baseline. However, case identification by the Risk, Injury, Failure, Loss, End-stage kidney disease (RIFLE) criteria classification may be inaccurate for patients in the Risk stratum. Therefore, results of studies using MDRD-based estimated baseline creatinine should be interpreted with caution especially for patients with mild AKI. Conflict of interest statement. None declared. References 1. Bellomo R, Ronco C, Kellum JA et al. Acute Dialysis Quality Initiative workgroup. Acute renal failure definition, outcome measures, animal models, fluid therapy and information technology needs: the Second International Consensus Conference of the Acute Dialysis Quality Initiative (ADQI) Group. Crit Care 2004; 8: R204 R Mehta RL, Kellum JA, Shah SV et al. Acute Kidney Injury Network. Acute Kidney Injury Network: report of an initiative to improve outcomes in acute kidney injury. Crit Care 2007; 11: R31 3. Maccariello E, Soares M, Valente C et al. RIFLE classification in patients with acute kidney injury in need of renal replacement therapy. Intensive Care Med 2007; 33: Ali T, Khan I, Simpson W et al. Incidence and outcomes in acute kidney injury: a comprehensive population-based study. JAmSoc Nephrol 2007; 18: Cruz DN, Bolgan I, Perazella MA et al. North East Italian Prospective Hospital Renal Outcome Survey on Acute Kidney Injury (NEi- PHROS-AKI) Investigators. North East Italian Prospective Hospital Renal Outcome Survey on Acute Kidney Injury (NEiPHROS-AKI): targeting the problem with the RIFLE Criteria. Clin J Am Soc Nephrol 2007; 2: Bagshaw SM, George C, Dinu I et al. A multi-centre evaluation of the RIFLE criteria for early acute kidney injury in critically ill patients. Nephrol Dial Transplant 2008; 23: Arnaoutakis GJ, Bihorac A, Martin TD et al. RIFLE criteria for acute kidney injury in aortic arch surgery. J Thorac Cardiovasc Surg 2007; 134: Discussion Epub 2007 Oct Jenq CC, Tsai MH, Tian YC et al. RIFLE classification can predict short-term prognosis in critically ill cirrhotic patients. Intensive Care Med 2007; 33: Ostermann M, Chang RW. Acute kidney injury in the intensive care unit according to RIFLE. Crit Care Med 2007; 35: Akcan-Arikan A, Zappitelli M, Loftis LL et al. Modified RIFLE criteria in critically ill children with acute kidney injury. Kidney Int 2007; 71: Tallgren M, Niemi T, Pöyhiä R et al. Acute renal injury and dysfunction following elective abdominal aortic surgery. Eur J Vasc Endovasc Surg 2007; 33: Uchino S, Bellomo R, Goldsmith D et al. An assessment of the RIFLE criteria for acute renal failure in hospitalized patients. Crit Care Med 2006; 34: Hoste EA, Clermont G, Kersten A et al. RIFLE criteria for acute kidney injury are associated with hospital mortality in critically ill patients: a cohort analysis. Crit Care 2006; 10: R Kuitunen A, Vento A, Suojaranta-Ylinen R et al. Acute renal failure after cardiac surgery: evaluation of the RIFLE classification. Ann Thorac Surg 2006; 81:

8 3918 D. Franke et al. 15. Abosaif NY, Tolba YA, Heap M et al. The outcome of acute renal failure in the intensive care unit according to RIFLE: model application, sensitivity, and predictability. Am J Kidney Dis 2005; 46: Bell M, Liljestam E, Granath F et al. Optimal follow-up time after continuous renal replacement therapy in actual renal failure patients stratified with the RIFLE criteria. Nephrol Dial Transplant 2005; 20: Perez-Valdivieso JR, Bes-Rastrollo M, Monedero P et al. Prognosis and serum creatinine levels in acute renal failure at the time of nephrology consultation: an observational cohort study. BMC Nephrol 2007; 8: Coca SG, Bauling P, Schifftner T et al. Contribution of acute kidney injury toward morbidity and mortality in burns: a contemporary analysis. Am J Kidney Dis 2007; 49: Levey AS, Bosch JP, Lewis JB et al. A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group. Ann Intern Med 1999; 130: Froissart M, Rossert J, Jacquot C et al. Predictive performance of the modification of diet in renal disease and Cockcroft Gault equations for estimating renal function. J Am Soc Nephrol 2005; 16: Poggio ED, Wang X, Greene T et al. Performance of the modification of diet in renal disease and Cockcroft Gault equations in the estimation of GFR in health and in chronic kidney disease. J Am Soc Nephrol 2005; 16: Beddhu S, Samore MH, Roberts MS et al. Creatinine production, nutrition, and glomerular filtration rate estimation. J Am Soc Nephrol 2003; 14: Myers GL, Miller WG, Coresh J et al. National Kidney Disease Education Program Laboratory Working Group. Recommendations for improving serum creatinine measurement: a report from the Laboratory Working Group of the National Kidney Disease Education Program. Clin Chem 2006; 52: Peake M, Whiting M. Measurement of serum creatinine current status and future goals. Clin Biochem Rev 2006; 27: Miller WG, Myers GL, Ashwood ER et al. Creatinine measurement: state of the art in accuracy and interlaboratory harmonization. Arch Pathol Lab Med 2005; 129: Macdonald JH, Marcora SM, Kumwenda MJ et al. The relationship between estimated glomerular filtration rate, demographic and anthropometric variables is mediated by muscle mass in non-diabetic patients with chronic kidney disease. Nephrol Dial Transplant 2006; 21: Berg UB. Differences in decline in GFR with age between males and females. Reference data on clearances of inulin and PAH in potential kidney donors. Nephrol Dial Transplant 2006; 21: Clark WF, Macnab JJ, Chen SJ et al. Evaluation of GFR estimating equations in the general community: implications for screening. Clin J Am Soc Nephrol 2006; 1: Wetzels JF, Kiemeney LA, Swinkels DW et al. Age- and genderspecific reference values of estimated GFR in Caucasians: the Nijmegen Biomedical Study. Kidney Int 2007; 72: Epub 2007 Jun Cirillo M, Anastasio P, De Santo NG. Relationship of gender, age, and body mass index to errors in predicted kidney function. Nephrol Dial Transplant 2005; 20: Bagshaw SM, Uchino S, Cruz D et al. A comparison of observed versus estimated baseline creatinine for determination of RIFLE class in patients with acute kidney injury. Nephrol Dial Transplant 2009; 24: Received for publication: ; Accepted in revised form: Nephrol Dial Transplant (2010) 25: doi: /ndt/gfq300 Advance Access publication 31 May 2010 Prematurity, small for gestational age and perinatal parameters in children with congenital, hereditary and acquired chronic kidney disease Doris Franke 1, Sina Völker 1, Sanny Haase 1, Leo Pavičić 3, Uwe Querfeld 2, Jochen H.H. Ehrich 1 and Miroslav Živičnjak 1 1 Department of Paediatric Kidney, Liver and Metabolic Diseases, Children s Hospital, Hannover Medical School, Carl-Neuberg-Str. 1, Hannover, Germany, 2 Department of Paediatric Nephrology, Charité University Hospital, Augustenburger Platz 1, Berlin, Germany and 3 Faculty of Kinesiology, University of Zagreb, Horvačanski zavoj 15, Zagreb, Croatia Correspondence and offprint requests to: Miroslav Živičnjak; zivicnjak.miroslav@mh-hannover.de Abstract Background. Low birth weight has been identified as a risk factor for chronic kidney disease (CKD). Methods. We analysed perinatal parameters taken from the National Birth Certificates of 435 children with CKD stages 3 5 of different aetiology and time of onset of CKD. Diseases were classified as congenital with onset of renal disease during fetal life (n = 260; 60%), hereditary as genetically determined with onset after 3 months of life (n = 93; 21%) and acquired CKD (n = 82; 19%). The Author Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved. For Permissions, please journals.permissions@oxfordjournals.org