Loading...
Please wait, while we are loading the content...
Similar Documents
A More Accurate Method To Estimate Glomerular Filtration Rate from Serum Creatinine: A New Prediction Equation
| Content Provider | Semantic Scholar |
|---|---|
| Author | Levey, Andrew S. Bosch, Juan P. Lewis, Julia Breyer Greene, Tom Rogers, Nancy Roth, D. |
| Copyright Year | 1999 |
| Abstract | The glomerular filtration rate (GFR) is traditionally considered the best overall index of renal function in health and disease (1). Because GFR is difficult to measure in clinical practice, most clinicians estimate the GFR from the serum creatinine concentration. However, the accuracy of this estimate is limited because the serum creatinine concentration is affected by factors other than creatinine filtration (2, 3). To circumvent these limitations, several formulas have been developed to estimate creatinine clearance from serum creatinine concentration, age, sex, and body size (4-12). Despite more recent studies that have related serum creatinine concentration to GFR (13-24), no formula is more widely used to predict creatinine clearance than that proposed by Cockcroft and Gault (4). This formula is used to detect the onset of renal insufficiency, to adjust the dose of drugs excreted by the kidney, and to evaluate the effectiveness of therapy for progressive renal disease. More recently, it has been used to document eligibility for reimbursement from the Medicare End Stage Renal Disease Program (25) and for accrual of points for patients on the waiting list for cadaveric renal transplantation (26). Major clinical decisions in general medicine, geriatrics, and oncology (as well as nephrology) are made by using the Cockcroft-Gault formula and other formulas to predict the level of renal function. Therefore, these formulas must predict GFR as accurately as possible. The Modification of Diet in Renal Disease (MDRD) Study, a multicenter, controlled trial, evaluated the effect of dietary protein restriction and strict blood pressure control on the progression of renal disease (27-30). During the baseline period, GFR, serum creatinine, and several variables that affect the relation between them were measured in patients with chronic renal disease. The purpose of our study was to develop an equation from MDRD Study data that could improve the prediction of GFR from serum creatinine concentration. Methods Baseline Cohort and Measurement Methods in the Modification of Diet in Renal Disease Study The overall study design and methods of recruitment for the MDRD Study have been described elsewhere (31, 32). A total of 1785 patients entered the baseline period. Of these patients, 1628 (91%) also underwent measurement of GFR and the other variables described below; these patients constitute the study group for these analyses. Glomerular filtration rate was measured as the renal clearance of 125I-iothalamate (33, 34). Creatinine clearance was computed from creatinine excretion in a 24-hour urine collection and a single measurement of serum creatinine. Serum and urine creatinine were measured by using a kinetic alkaline picrate assay with a normal range in serum of 62 to 124 mol/L (0.7 to 1.4 mg/dL) (35). Glomerular filtration rate and creatinine clearance were expressed per 1.73 m2 of body surface area by multiplying measured values by 1.73/body surface area (36). The serum and urine specimens were also used for other measurements, including serum albumin (bromcresol green method [35]), serum urea nitrogen (urease method [35]), and urine urea nitrogen (urease method [35]). Protein intake (g/d) was estimated as 6.25 [UUN (g/d) + 0.031 (g/kg per day) SBW (kg)], where UUN is urine urea nitrogen, SBW is standard body weight, and 0.031 g/kg per day is a constant reflecting the rate of excretion of nitrogen in compounds other than urine urea (37, 38). The diagnosis of diabetes and the cause of renal disease were assigned on the basis of chart review at the clinical center (39). Statistical Analysis Descriptive Statistics The relation of renal function measurements to other baseline characteristics was assessed by using contingency tables, t-tests, analysis of variance, and linear regression, as appropriate. Nonparametric tests (Wilcoxon rank-sum tests and Kruskal-Wallis tests) gave consistent results. A P value less than 0.01 was considered statistically significant. Multivariable Analysis of Glomerular Filtration Rate We used stepwise multiple regression to determine a set of variables that jointly predicted GFR. The stepwise regression models were developed by using a training sample consisting of a random sample of 1070 of the 1628 patients. We found that the variability of the difference between the observed and predicted GFR values was greater for higher GFR values. This increase was eliminated by performing multiple regressions on log-transformed data. To facilitate clinical interpretation, the results were re-expressed in terms of the original units. Consequently, the prediction equation is a multiplicative model; regression coefficients refer to the change in geometric mean GFR associated with unit changes in the independent variable. Predicted GFR is expressed in mL/min per 1.73 m2. The following variables were considered for possible inclusion in the regression model: weight, height, sex, ethnicity, age, diagnosis of diabetes, serum creatinine concentration, serum urea nitrogen level, serum albumin level, serum phosphorus level, serum calcium level, mean arterial pressure, urine creatinine level, urine urea nitrogen level, urine protein level, and urine phosphorus level. The cause of renal disease was not included because in clinical practice, the cause may be unknown or clinicians may not use the same classification method as the investigators in the MDRD Study. A P value less than 0.001 was used as the criterion for entry of a variable into the model. Because of the difficulty in collecting complete 24-hour urine samples in clinical practice, an additional stepwise regression was performed to develop a prediction model that did not include urine biochemistry variables. Finally, because of the interest in developing a prediction equation to assess eligibility for Medicare reimbursement and listing for cadaveric renal transplantation, we repeated the analysis restricting the population to the subgroup of patients with higher serum creatinine concentrations (>221 mol/L [2.5 mg/dL]; n=509 in the training sample). Methods for Comparing Equations To Predict Glomerular Filtration Rate We first developed coefficients for each prediction equation (including the selection of the predictor variables for the stepwise regressions) using the data from the training sample to predict log GFR. Each prediction equation also included a multiplicative constant to account for any consistent bias in the application of that equation in the MDRD Study Group. This was particularly important for equations that are intended to estimate creatinine clearance, which is known to be higher than GFR. The regression coefficients determined in the training sample were then applied to obtain predicted GFRs in a separate validation sample consisting of the remaining 558 patients (172 patients with serum creatinine concentration>221 mol/L [2.5 mg/dL]). These predicted GFR values were compared with the actual GFRs in the validation sample to evaluate the performance of each prediction equation. In this way, separate data sets were used to construct the equations and assess their accuracy after removal of systematic bias. For each equation, we computed overall R 2 (percentage of variability in log GFR explained by the regression model) and the 50th, 75th, and 90th percentiles of the distribution of the percentage absolute difference between measured and predicted GFRs in the validation sample. The 50th percentiles indicate the typical size of the errors in prediction of GFR, and the 75th and 90th percentiles assess the sizes of the larger errors that occurred for each model. Development of Final Prediction Equations To improve the accuracy of the final MDRD Study prediction equations, the regression coefficients derived from the training sample were updated on the basis of data from all 1628 patients. As a result, the standard errors of the regression coefficients in the final MDRD Study prediction equations are slightly smaller than those derived from the training sample; thus, the accuracy of the final prediction equations may be slightly better (by about 0.1% to 0.2%) than their accuracy as assessed in the validation sample. Results Demographic and Clinical Characteristics The mean age ( SD) of the cohort was 50.6 12.7 years. Sixty percent of patients were male, 88% were white, and 6% were diabetic. Causes of renal disease were glomerular disease (32%), polycystic kidney disease (22%), tubulointerstitial disease (7%), and other or unknown renal diseases (40%). Mean protein intake was 0.99 0.24 g/kg of body weight per day and mean arterial pressure was 99.4 12.2 mm Hg. Mean weight was 79.6 16.8 kg, body surface area was 1.91 0.23 m2, serum urea nitrogen concentration was 11.4 5.7 mmol/L [32 16 mg/dL], and serum albumin concentration was 40.0 4.0 g/L [4.0 0.4 g/dL], respectively. Glomerular Filtration Rate, Creatinine Clearance, and Serum Creatinine Concentration Renal function measurements for the study group and for various subgroups are shown in Table 1. Mean GFR for the population was 0.38 mL s 2 m 2 (39.8 mL/min per 1.73 m2), with lower values in patients with lower protein intake, white patients compared with black patients, and older patients ( 55 years) compared with younger patients (P<0.01). The mean value of creatinine clearance was 0.81 mL s 2 m 2 (48.6 mL/min per 1.73 m2) and was lower in older patients and patients with lower protein intake (P 0.01). The mean serum creatinine concentration was 203 mol/L (2.3 mg/dL) and was higher in men, patients with lower protein intake, and patients with higher mean arterial pressure (P 0.01). Figure 1 shows the well-known reciprocal relation of serum creatinine concentration to GFR for subgroups based on sex and ethnicity. At any given GFR, the serum creatinine concentration is significantly higher in men than in women and in black persons than in white persons (P<0.001). Table 1. Association of Renal Fu |
| Starting Page | 461 |
| Ending Page | 470 |
| Page Count | 10 |
| File Format | PDF HTM / HTML |
| DOI | 10.7326/0003-4819-130-6-199903160-00002 |
| PubMed reference number | 10075613 |
| Journal | Medline |
| Volume Number | 130 |
| Journal | Annals of Internal Medicine |
| Alternate Webpage(s) | http://vingulmork.no/public/docs/Levey%20AS%20et%20al.%20A%20More%20Accurate%20Method%20To%20Estimate%20GFR.pdf |
| Alternate Webpage(s) | http://dickyricky.com/Medicine/Papers/1999_03%20Ann%20Intern%20Med%20Modification%20of%20Diet%20in%20Renal%20Disease%20(MDRD)%20Study.pdf |
| Alternate Webpage(s) | https://www.loyolamedicine.org/sites/default/files/gme/nephrology/pdfs/ann_intern_med_-_mdrd_equation.pdf |
| Alternate Webpage(s) | https://doi.org/10.7326/0003-4819-130-6-199903160-00002 |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
