Correlation of Measured and Estimated Creatinine Clearance in Hospitalized Elderly Patients: A Retrospective Cohort Study

Abstract

Background: Accurate assessment of renal function is essential in hospitalized elderly patients. Few studies have examined the accuracy of Cockcroft-Gault (C-G) estimates of creatinine clearance (CrCl) compared with measured clearance in these patients. Objective: The objective of this study was to determine the correlation between C-G estimates of CrCl and measured CrCl in hospitalized elderly patients. Methods: This Institutional Review Board–approved, single-center retrospective observational cohort study included all patients who were 65 years and older admitted to our medical center in January to September 2018 with either an 8- or 24-hour urine collected during admission. The primary outcome was correlation, bias, and precision of C-G estimates of CrCl versus measured CrCl using Pearson correlation, Spearman linear regression, and Bland-Altman analysis. Outliers were determined using a cut-off of ±20%. Data are presented as median (interquartile range) or percentages. Results: A total of 108 urine collections from 90 unique patients were included in the study. The patients were 51% female, median age was 71 (68-77) years, and median body mass index was 26.6 (22.8-31) kg/m². Most collections were over 24 hours (66.7%), and 38% were performed while patients were in an intensive care unit. Median blood urea nitrogen (BUN) was 24.5 (17-36) mg/dL and median serum creatinine was 0.71 (0.55-1.09) mg/dL. The median C-G estimation was 75.4 (48.2-110.6) mL/min, and the median measured CrCl was 79.1 (38.1-99.5) mL/min, r² = .56 (P < .001). Bland-Altman analysis showed large limits of agreement (-75.5-57.7 mL/min), with a bias of −8.9 and precision (standard deviation of bias) of 34 mL/min. Outliers were common, with 38% of C-G estimation values >120% of measured CrCl, and 18% of C-G estimates <80% of measured CrCl. Conclusions: Measured CrCl varied significantly from C-G estimates in hospitalized elderly patients. It is important to recognize characteristics of patients who may benefit from measurement of CrCl. Future studies should examine the impact of this variance on clinical outcomes.

Introduction

The Cockcroft-Gault (C-G) equation is commonly used to determine renal function and dose renally adjusted medications, but has significant limitations. ¹ In the original paper published by Cockcroft and Gault in 1976, 96% of patients tested were male, and those with widely varying serum creatinine (SCr) values were excluded from analysis. In addition, the same patients were used to derive the equation and subsequently test its fit. Perhaps most surprisingly, the now-ubiquitous use of an 85% correction factor for females was not evaluated in the study, but simply proposed within the discussion section. ² Despite these issues, there have been few studies regarding the correlation between estimated (via C-G) and measured creatinine clearance (CrCl) in hospitalized elderly patients and its potential effect on drug dosing.

Previous analyses have suggested that the C-G equation has a relatively weak correlation with true clearance; correlation tends to weaken with larger populations and more even sex distributions. Some authors have proposed that the equation may be more accurate for females without the 85% correction factor.^3,4 However, most previously published correlative studies have included small populations of elderly patients or limited their analysis to healthy subjects. Renal function can change dramatically in hospitalized and critically ill patients; these are also the patients most likely to be receiving important renally adjusted medications such as antimicrobials and anti-arrhythmic drugs. The correlation between measured and estimated CrCl in the hospitalized elderly is not well understood.

Regarding methods of urine collection for measured CrCl, 8-hour collections appear to be a valid assessment of true clearance. Values for clearance from 8-hour collections largely fall within 20% of 24-hour urine collection values, and the differences between values have not been found to be statistically significant. ⁵ Urine collections may be useful for determining true clearance in certain patient populations with abnormal SCr values, but are still limited by the accuracy of the collection time, volume, and potential interference of diuretic drugs. At our institution, 24-hour urine collections are part of the standard weekly order set for patients on parenteral nutrition and are used to determine nitrogen balance. Eight-hour urine collections are most commonly ordered by pharmacy specialists or other providers for the purpose of determining renally adjusted medication dosing. Pharmacists at our institution are credentialed as providers to independently make renal dose adjustments for admitted patients.

The primary objective of this study was to correlate and determine bias, accuracy, and precision for measured CrCl by 8- or 24-hour urine collection as compared with C-G estimates in elderly hospitalized patients. Secondary analyses explored the percentage of measurements within ±20% of each other (a standard used to better understand correlation in other studies) and discordance in dosing of renally adjusted medications. ² We hypothesized that the correlation between measured and estimated CrCl in the hospitalized elderly would be relatively weak, implying a potentially significant difference in drug-dosing recommendations for selected renally adjusted medications.

Methods

This was a single-center retrospective observational cohort study conducted at a 1506-bed academic medical center. Patients who were 65 years and older who were admitted between January 2018 and September 2018 with an 8- or 24-hour urine collection during their stay and at least one active order for a selected renally dosed medication on the day of urine collection were included. Subjects were excluded if they were incarcerated, had an ileal conduit, had a history of renal transplant or if specified admission was for renal transplant, were on renal replacement therapy in the 72 hours prior to urine collection, received a diuretic within 8 hours prior to or during the specified urine collection, or if estimated CrCl could not be calculated (no SCr collected within 24 hours of specified urine collection, no measured height and/or weight during admission, height <60 inches, or undetectable SCr value [<0.2 mg/dL]). The University Office of Responsible Research Practice deemed this study to be exempt from full Institutional Review Board review.

Data obtained from the electronic medical record (EMR) included age; sex; height; body weight; race; accommodation code; SCr and corresponding blood urea nitrogen (BUN); type, time/date, and value of creatinine for each urine collection; past medical history of paraplegia/quadriplegia or chronic kidney disease; and active order for selected renally adjusted medications. The SCr collected was the actual value used within the EMR for estimated CrCl calculation and was required to have been collected within 24 hours prior to the specified urine collection. The EMR (Epic Systems Corporation, Verona, WI) automatically calculated the CrCl, using the actual body weight if it was less than the ideal body weight. If the actual body weight was ≥120% of the ideal body weight, adjusted body weight was used for calculation using 40% of the difference added to the ideal body weight. All other situations used ideal body weight. The presence of acute kidney injury (AKI) was later determined and was defined as at least one of the following: (1) SCr increase of ≥0.3 mg/dL within the 48-hour time period prior to index SCr collection, (2) SCr 1.5 times the lowest SCr (estimated baseline) measured during admission, (3) urine volume (of specified collection) less than 0.5 mL/kg/hr. We also collected data on the incidence of augmented renal clearance (ARC), defined as a measured CrCl >130 mL/min based on urine collection. A selection of renally adjusted medications was chosen based on their relevance in our study population. Dosing breakpoints were defined for each medication; these breakpoints were based on institutional policy (when available) or package insert recommendations. Dosing breakpoints were determined for each active medication order based on both estimated and measured CrCl. Selected renal medications and their dosing breakpoints are given in Table 1.

Table 1.

Renally Adjusted Medications Included in Analysis.

Medication	CrCl dosing breakpoints, mL/min
Aminoglycosides	≥60	40-59	20-39	<20
Ampicillin/sulbactam	≥50	30-49	15-29	<15
Cefazolin	≥30	10-29	<10
Cefepime	≥60	30-59	<30
Ceftaroline	≥50	30-49	15-29	<15
Ceftolozane/tazobactam	≥50	30-49	<30
Daptomycin	≥30	<30
Enoxaparin	≥30	<30
Lacosamide	≥30	<30
Levetiracetam	>80	50-80	30-49	<30
Levofloxacin	>50	20-50	<20
Meropenem	≥50	26-49	10-25	<10
Piperacillin/tazobactam	≥20	<20
Rivaroxaban (atrial fibrillation)	>50	15-50	<15
Vancomycin	≥60	30-59	<30

Note. Renal dosing adjustments provided in the package insert are different when the drug is used for the indication of atrial fibrillation (AFib) versus other indications. All patients included in the study were taking rivaroxaban for the indication of AFib.

Statistical Analysis

Study data were collected and managed using REDCap electronic data capture tools hosted at our institution. ⁶ Sample size was not calculated as there was a paucity of data comparing measured CrCl to C-G in elderly hospitalized patients. Statistical calculations were performed using Statistical Package for Social Science version 26.0 (SPSS Inc., Chicago, IL). The primary outcome of correlation was assessed using Spearman linear regression to compare measured CrCl using 8-hour or 24-hour urine collection to estimated CrCl from the C-G equation. Bias, precision, and accuracy of the 2 equations were determined via the Bland-Altman analysis, a validated method used to determine the degree of agreement between 2 quantitative methods of measurement. A subset analysis of those without AKI was also performed using Bland-Altman analysis. Discordance in drug dosing was defined using previous renal drug dosing studies and was reported as a percentage of total collections that would result in a difference in drug dosing for at least one drug using measured CrCl versus estimated CrCl. ⁷ Descriptive statistics were used to describe the patient population, with normally distributed data presented as mean ± standard deviation and non-normally distributed data as median (25%-75% interquartile range).

Results

A total of 532 individual urine collections were screened for inclusion in the study, and 108 collections from 90 unique patients met the study criteria (Figure 1). Collections from male patients comprised 48.9% of the study set, and patients were a median age of 71 (68-77) years at the time of urine collection. The median admission weight was 77.3 (67.2-91.2) kg, height was 66.5 (63.2-69.7) inches, and body mass index (BMI) was 26.6 (22.8-31) kg/m². The median SCr was 0.71 (0.55-1.09) mg/dL and median BUN was 24.5 (17-36) mg/dL. Most of the patients had 24-hour urine collections performed (66.7 %). The median estimated CrCl was 75.4 mL/min (48.2-110.6), and the median measured CrCl was 79.1 mL/min (38.1-99.5). Acute kidney injury and ARC were present at the time of 38% and 11.1% of collections, respectively. Patients were located in an intensive care unit at the time of 38.9% of collections. In those without AKI, the median estimated CrCl was 86 mL/min (58-128.6) and the median measured CrCl was 85.7 mL/min (67.5-109.4).

Figure 1.

Inclusion criteria.

Note. RRT = rental replacement therapy.

^aPatients could be excluded for multiple reasons listed.

Spearman linear regression found a coefficient of determination of r² = .565 (P < .001) between estimated and measured CrCl. Using Bland-Altman analysis, the mean bias between the 2 methods of determination was found to be −8.9 ± 34 mL/min. Precision (the standard deviation of bias) was thus 34 mL/min. Agreement limits were set at ±2 standard deviations of the bias and fell at +57.7 mL/min (upper) and −75.5 mL/min (lower) (Figure 2). For all sets of clearance values, 44% fell within ±20% of each other; 18% of estimated CrCl values were >20% lower than the related measured clearance value, and the remaining 38% fell >20% higher than the associated measured value. In the subset of patients without AKI, the Spearman linear regression coefficient of determination was r² = .49 (P < .001). The mean bias was −2.9 + 32.2 mL/min, and agreement limits were +60.2 mL/min (upper) and −66 mL/min (lower) (Figure 2).

Figure 2.

Bland-Altman analyses: (A) for all patients and (2B) for patients without acute kidney injury.

Note. Bland-Altman analyses. Mean ([x + y]/2) is represented on the horizontal axis and difference (x-y) is represented on the vertical axis, where x = estimated CrCl and y = measured CrCl. The bias (mean difference) is represented by the middle horizontal line on the graph. The upper and lower lines represent the agreement limits, set at +2 and −2 standard deviations of the bias. AKI = acute kidney injury.

Enoxaparin was the most common active medication order on the day of urine collection and was associated with 38 of the 108 urine collections, followed by vancomycin (32), piperacillin-tazobactam (23), and cefepime (22). Regarding drug dosing, 19 of the 108 urine collections (17.6%) showed discordance, meaning that at least one medication would have been dosed differently depending on the method of clearance determination used. The drug most commonly associated with discordance was cefepime; dosing of cefepime was discordant in 6 of the 19 (31.6%) urine collections.

Of the 19 urine collections associated with discordance, 11 (58%) were collected from female patients. Discordant collections were associated with a median patient age of 70 years (69-77), BMI of 28.5 (25.4-35.1), SCr of 1.03 mg/dL (0.86-1.59), and BUN of 40 (25-55) mg/dL. Comparisons to the entire study population are shown in Figure 3. Ten of the 19 urine collections were associated with patients in AKI, and 3 of the associated patients had a diagnosis of chronic kidney disease. Estimated clearance via C-G was greater than measured clearance for 12 (63%) of the discordant collections.

Figure 3.

Rates of selected characteristics in discordant patients versus general study population.

Note. Rates of female sex, CKD, AKI, and collections in which estimated clearance was higher than measured were all higher in the discordant population as compared with the general population. Data is presented as a percentage of total collections. CKD = chronic kidney disease; AKI = acute kidney injury.

Discussion

In hospitalized elderly patients, estimated CrCl from the C-G equation only predicted measured CrCl with an r²of .565. Overall, only 44% of measured CrCl values fell within ±20% of estimated CrCl. Discordant drug dosing was present in 17.6% of collections. At a glance, median values for CrCl in this study appear relatively similar. Our Bland-Altman analysis provides an alternate view of the spread of data; wide agreement limits suggest that much variation exists in terms of the spread between clearance values obtained via each method. Our results for all patients, as well as the subset without AKI, are similar to 3 previous studies which reported a mean bias with large standard deviation (precision) between measured CrCl and C-G. These studies reported biases of −3.2 ± 14.2, −3.5 ± 22.5, and −3.6 ± 22.2 mL/min.^{8 -10} There was also a wide agreement limit in those patients without AKI; the difference between C-G and measured CrCl ranged from −75.5 to 57.7 mL/min in all patients and −66 to 60.2 mL/min in that subgroup. We also found that a minority (44%) of paired values fell within the limits of ±20% difference; this also implies greater spread as compared with the original paper by Cockcroft and Gault, in which 67% of values fell within these limits. ² More recent literature has suggested that the C-G equation is most likely to underestimate renal function when compared with other methods of estimation; however, this may not apply to the hospitalized elderly population.^8,11 Our study found that more than one-third of estimated CrCl values were overestimates when compared with measured CrCl.

In elderly patients, it is common practice to round up low SCr values to 0.8 to 1.0 mg/dL to avoid overestimation of clearance.^1,12,13 Recently, McConachie et al ¹ surveyed members of the American College of Clinical Pharmacy and reported that only 34% use actual SCr in elderly patients with low SCr, while the majority round up. Of 299 survey responders, 30% routinely round up to 0.8 mg/dL and 29% round up to 1.0 mg/dL. Although this method can be useful in certain situations, it could lead to underestimation of clearance and thus potential underdosing of renally adjusted medications. Without using this rounding technique in our study, we found that only 18% of estimated CrCl were underestimates, and 44% fell within ±20% difference. While the use of rounding up SCr may be useful in some patients to overcome overestimation, it might be detrimental in others as it can lead to inadequate treatment. This is similar to the results from Dowling et al, which suggest that rounding up of SCr leads to underestimation of renal function and can lead to subtherapeutic doses of critical medications. ⁸ We also found that ARC was present in 11% of collections; rounding up SCr in this population may lead to even more inaccurate dosing. Future analyses could assess whether the rounding method leads to closer correlation between estimated and measured clearance in certain subpopulations of patients.

It is also nearly universal practice to use a correction factor in women when using the C-G equation. Our results are similar to previous studies demonstrating only moderate correlation between measured CrCl and estimated CrCl in elderly patients.^{4,8 -10,14} However, it is important to note that unlike the present analysis, these earlier, smaller studies did not use the 0.85 correction factor for females as suggested in C-G. Pequignot et al and Chauvelier et al used the following formula for calculating C-G:C-G = (140 − age) × weight (kg) × K/SCr, where K = 1.23 for men and 1.04 for women.^9,10 Dowling et al. ⁸ used the same equation we used to save the adjustment for obesity, but only had ~10% with a BMI greater than 30 kg/m². While the studies by Pequignot et al and Chauvelier et al used a different equation for C-G estimation, they did use a correction factor for females.^9,10 Regardless, all 3 studies and ours were slightly more than half female (range = 52%-54%). While the exact percentage of patients who were 65 and older in the original Cockcroft and Gault study is unknown, only 23.7% were 70 years or older and only 4% of all patients were women. As the female correction factor has not been studied in detail, it may lead to inappropriate estimations of CrCl, particularly in elderly female patients.

Our rate of drug dosing discordance, while relatively low, may carry important clinical significance. A rate of discordance of 17.6% implies that nearly 1 in 5 hospitalized elderly adults may be prescribed at least 1 drug that would be dosed differently depending on the method of clearance determination. Studies by Dowling et al and Hudson et al examined dosing discordance between C-G and either the Modified Diet in Renal Disease formula or Chronic Kidney Disease Epidemiology Collaboration formula rather than measured CrCl.^7,8 We found a 25% drug discordance rate when comparing C-G to 8-hour urine CrCl in an adult population of 85 unique patients with a median age of 55 (41-70) years. ¹⁵ The rate of discordance might also be partially related to current drug therapy. While enoxaparin was the most commonly prescribed drug, discordance with cefepime was found in approximately one-third of discordant collections. This might be due to differences in cutoffs for renally adjusted dosing regimens at our institution: cefepime is first adjusted at a CrCl of 60 mL/min, whereas enoxaparin is not adjusted until CrCl is less than 30 mL/min (Table 1). In addition, the clinical significance of this discordance is unknown. The ongoing question, then, is how to determine which patients are most likely to present with discordance to optimize drug dosing for all patients, an issue we have attempted to address by examining the demographics of our discordant population more closely.

While this study does use relevant statistical analyses to better understand the correlation between the 2 methods, it has significant limitations. This analysis was not powered to detect a significant difference between measured and estimated CrCl. The study population was relatively small; sample size was limited by the fact that collections which had no active orders for renally adjusted medications on the day of urine collection were excluded from analysis. In addition, 24-hour urine collections were associated with significantly less renally adjusted drugs than 8-hour collections. This is likely based on their utility at our institution: 24-hour urine collections are most commonly used to collect urine urea nitrogen in patients on parenteral nutrition and are rarely collected for the purpose of drug dosing. As a result, there may have been a selection bias. Calculated CrCl values were based on the methods described above; while standardized for our institution, these may not be generalizable to others as they may use different equations or adjustments for obesity. Theoretical modeling of dosing categories for all renally adjusted drugs—when applied to every eligible urine collection—would lead to fewer excluded patients, and thus could provide significantly more data points to better determine which drugs are most likely to be dosed discordantly. This study was not designed to assess the clinical outcomes of using these methods for assessing medication dosing. Finally, on an operational level, there is the potential for errors with collection: the start and end times for urine collection are not reported in the EMR, and accidental emptying of collected urine could occur if communication is not clear between nursing shifts.

Despite flaws with C-G estimation of clearance, newer estimation methods have generally not been found to provide a superior assessment of clearance in elderly adults. A recent study found that all tested equations (CKD-EPI, LMR, BIS 1, and FAS) had significant limitations in predicting glomerular filtration rate (GFR) as measured by renal inulin clearance. ¹⁶ Cystatin C, a serum marker of renal function, has been studied in dosing of various chemotherapeutic agents.^{17 -19} It may be useful in elderly patients as it is not influenced by muscle volume. ²⁰ However, possible mechanisms of nonrenal clearance may limit its utility until this is better understood. ²¹ Currently, cystatin C is not available for routine use at our institution. Other biomarkers of current interest include urinary and serum NGAL, urinary IL-18, and urinary L-FABP. ²² Until new markers or estimation methods are found to better predict renal function, pharmacists and other medical providers must continue to use careful clinical judgment to choose the best drug dose in populations at risk of renal injury.

Conclusions

In this study of hospitalized elderly adults, measured CrCl via urine collection was not found to correlate well with estimates of CrCl as determined via C-G. Further studies are needed to determine which subgroups of patients may benefit clinically from determination of measured CrCl and improved dosing of renally adjusted medications. Further studies should also assess the clinical impact of using each method of clearance determination. Providers should remain aware of the characteristics which place patients at risk of inaccurate clearance estimation to best use alternative methods of clearance determination such as measured urine collection.