Comparison of Drug Dosing Recommendations Based on Measured GFR and Kidney Function Estimating Equations

Results 

Study Population 

Clinical characteristics of the 5,504 participants included in the study population are listed in Table 2. Mean age of the cohort is 47 ± 15 years. Approximately a third of the cohort was African American, a similar number had diabetes, and 5% were kidney transplant recipients. Mean measured GFR was 75 ± 44 mL/min; eGFR from the MDRD Study equation and estimated creatinine clearances from the CG and CGIBW equations were 69 ± 38, 75 ± 42, and 62 ± 36 mL/min, respectively. All pairwise comparisons between values for eGFR and measured GFR were significantly different from each other (P < 0.001). Table 2 lists values for measured GFR and estimated kidney function by using the 3 equations across subgroups.

Table 2. Clinical Characteristics

	Name	Total	Measured and Estimated Kidney Function(mL/min)
			mGFR	MDRD Study	CG	CGIBW
	All	5,504(100)	75±44	69±38	75±42	62±36
	Age(y)	47±15	—	—	—	—
	<40	2,058(37)	100±46	91±41	102±43	87±37
	40-65	2,751(50)	65±36	60±31	64±33	51±25
	>65	695(13)	45±26	45±23	42±20	33±15
	Sex
	Women	2,391(43)	72±42	65±36	74±41	60±34
	Men	3,113(57)	77±45	73±39	76±42	65±37
	Race
	African American	1,740(32)	65±32	62±30	62±30	49±30
	White or other	3,764(68)	80±47	73±41	82±45	69±38
	Weight(kg)	82±20	—	—	—	—
	<60	590(11)	64±40	58±35	58±34	54±33
	60-90	3,296(60)	76±44	70±39	74±41	63±36
	>90	1,618(29)	78±43	72±37	84±43	64±34
	Diabetes
	Yes	1,581(29)	100±48	91±44	100±46	86±41
	No	3,923(71)	65±37	61±32	65±35	53±28
	Transplant
	Yes	251(5)	51±27	52±27	59±31	48±24
	No	5,253(95)	76±44	70±38	76±42	63±36
	Mean body surface area(1.73 m2)	1.93±0.24	—	—	—	—
	Mean serum creatinine(mg/dL)	1.65±1.15	—	—	—	—
	Mean body mass index(kg/m2)	28±6	—	—	—	—

Note: Values expressed as number (percent) or mean ± SD. Conversion factors for units: GFR from mL/min to mL/s, ×0.0167; serum creatinine in mg/dL to μmol/L, ×88.4. Abbreviations: CG, Cockcroft-Gault equation using actual body weight; CGIBW, Cockcroft-Gault equation using ideal body weight; MDRD, Modification of Diet in Renal Disease Study equation; mGFR, measured glomerular filtration rate.

Assignment to FDA Guidance to Industry Kidney Function Category 

Comparison With Measured GFR 

Table 3 lists concordance and discordance between each estimating equation and measured GFR with respect to the assigned kidney function categories defined in the FDA Guidance to Industry. The MDRD Study equation showed the highest (78%) and the CGIBW showed the lowest (66%) concordance with measured GFR (P < 0.001). The direction of discordance was different for the 3 equations. The CG equation assigned a higher kidney function category compared with measured GFR for 16% of people compared with 5% for the CGIBW equation and 8% for the MDRD Study equation. Conversely, CGIBW assigned a lower kidney function category in 29% of people compared with 12% for the CG equation and 14% for the MDRD Study equation.

Table 3. Concordance Between Kidney Function Categories Assigned Using mGFR Versus Estimated Kidney Function

	Equation	Concordant (%)⁎	Discordant (%)
			< mGFR	> mGFR
	MDRD Study	78	14	8
	CG	73	12	16
	CGIBW	66	29	5

Abbreviations: CG, Cockcroft-Gault equation using actual body weight; CGIBW, Cockcroft-Gault equation using ideal body weight; MDRD, Modification of Diet in Renal Disease Study equation; mGFR, measured glomerular filtration rate.

⁎ P < 0.001 for the difference in concordance among all equations.

The MDRD Study equation has a greater rate of concordance with measured GFR for all subgroups tested (Fig 1). Other than for transplant recipients, the CG equation had a greater rate of concordance with measured GFR than the CGIBW equation. Large differences in concordance rates among equations were observed in many subgroups.

View Large Image Download to PowerPoint

Figure 1. Concordance (Conc) of Modification of Diet in Renal Disease (MDRD) Study, Cockcroft-Gault (CG), and CG adjusted for ideal body weight (CGIBW) equations with measured glomerular filtration rate (GFR) for assignment of kidney function categories by patient subgroup. Each bar indicates the percentage of concordance to measured GFR for each of the 43 different equations. (A) Age (<40, 40 to 65, or >65 years), (B) sex, (C) race (African American or other), (D) weight (<60, 60 to 90, or >90 kg), (E) presence or absence of diabetes, and (F) presence or absence of kidney transplant. The rate of concordance to measured GFR for the CG and CGIBW equations was significantly different (P < 0.001) from concordance to measured GFR for the MDRD Study equation for all subgroups except weight of 60 to 90 kg (CG), weight greater than 90 kg (CGIBW), diabetes (CG), and transplant recipients (CGIBW).

Comparison to the MDRD Study Equation 

The CG equation was concordant with the MDRD Study equation in 78% of cases, whereas the CGIBW had a slightly lower rate of concordance at 75% (P < 0.001). When discordance was observed with the MDRD Study equation, the CG and CGIBW equations differed in their patterns. Relative to the MDRD Study equation, the CG equation was more likely to predict assignment to a higher kidney function category (16%) than a lower kidney function category (6%). By contrast, the CGIBW equation was more likely to predict assignment to a lower kidney function category (22%) than a higher kidney function category (3%). Among subgroups, the CG and CGIBW equations both showed variable rates of concordance with the MDRD Study equation (71% to 86% and 55% to 89%, respectively; Fig 2).

View Large Image Download to PowerPoint

Figure 2. Concordance (Conc) of the Cockcroft-Gault (CG) and CG adjusted for ideal body weight (CGIBW) equations with the Modification of Diet in Renal Disease (MDRD) Study equation for assignment of kidney function categories by patient subgroups. Each bar indicates the percentage of concordance to the MDRD Study equation for the 2 equations. (A) Age (<40, 40 to 65, or >65 years), (B) sex, (C) race (African American or other), (D) weight (<60, 60 to 90, or >90 kg), (E) presence or absence of diabetes, and (F) presence or absence of kidney transplant. *P < 0.0001 for comparisons of each equation to the MDRD Study equation for each subgroup.

Recommended Drug Dosages 

Comparison to Measured GFR 

Compared with measured GFR, the average concordance rate for the specific drug dosing recommendations was 88% for the MDRD Study equation compared with 85% for the CG equation (P < 0.001) and 82% for the CGIBW equation (P < 0.001; Table 4). Concordance for all equations was lower for drugs that have a greater number of kidney function categories for dose adjustment. As observed with the FDA-assigned kidney function categories, use of the CG equation was most likely to translate into greater recommended drug dosages, whereas the CGIBW equation was most likely to translate into lower recommended drug dosages.

Table 4. Concordance Between Drug Dosing Recommendations Using Measured Versus Estimated Kidney Function

	Drug	MDRD Study			CG			CGIBW
		Concordant	Discordant		Concordant	Discordant		Concordant	Discordant
			< mGFR	> mGFR		< mGFR	> mGFR		< mGFR	> mGFR
	2 Dosing levels
	Enoxaparin	95	2	3	93	1	5	93	4	3
	Eptifibatide	93	4	3	90	4	7	87	11	2
	Ranitidine	93	4	3	90	4	7	87	11	2
	Average	94	3	3	91	3	6	89	9	2
	3 Dosing levels
	Acylovir	95	2	3	93	1	6	94	3	3
	Atenolol	91	4	6	88	2	10	85	12	3
	Cefazolin	92	3	5	90	2	8	90	6	4
	Digoxin	91	5	4	88	4	8	85	12	3
	Levofloxacin	88	5	7	84	4	12	82	12	6
	Tenofovir	86	5	9	82	4	14	79	14	8
	Tramadol	93	2	4	92	2	7	92	4	4
	Average	91	4	5	88	3	9	87	9	4
	4 Dosing levels
	Allopurinol	79	14	6	77	11	12	69	27	4
	Gabapentin	84	9	8	79	7	14	74	19	6
	Sotolol	85	8	7	81	7	12	77	18	5
	Average	83	10	7	79	8	13	73	21	5
	5 Dosing levels
	Disopyramide	75	15	10	72	11	17	65	27	8
	Lamivudine	85	7	8	80	5	15	77	16	7
	Average	80	11	9	76	8	16	71	22	8
	Overall average	88	6	6	85	5	10	82	13	4

Note: Units are mL/min; conversion factor from mL/min to mL/s, ×0.0167. Abbreviations: CG, Cockcroft-Gault equation using actual body weight; CGIBW, Cockcroft-Gault equation using ideal body weight; MDRD, Modification of Diet in Renal Disease Study equation; mGFR, measured glomerular filtration rate.

Comparison to the MDRD Study Equation 

The concordance rate between the MDRD Study and CG equations was 89%, with the MDRD Study equation recommending lower drug dosages in 9% of the study population. The concordance rate between the MDRD Study and CGIBW equations was 88%, with the MDRD Study equation recommending higher drug dosages in 10% of the study population. Concordance was lower for drugs with a greater number of kidney function categories (ranging from 90% for drugs with 2 dosing levels to 81% with 5 dosing levels).

Discussion 

Accurate estimates of kidney function are essential for optimal dosing of drugs cleared by the kidney. Overestimates of kidney function may lead to administration of inappropriately large doses and possible toxicity, and conversely, underestimates may lead to subtherapeutic dosing, treatment failures, and prolonged illness. In this study, we showed that the MDRD Study equation had the greatest rate of concordance with measured GFR for both assignment of kidney function categories recommended by the FDA Guidance for Industry and adjustment of specific drug dosing. For specific drug dosing, concordance rates among equations were high, with lower concordance for drugs with a greater number of dosing levels.

The CG equation, published in 1976, estimates creatinine clearance and therefore overestimates measured GFR because of creatinine secretion. Even after correcting for this overestimation, substantial imprecision remains.36 Modifications of the CG equation, such as the use of ideal body weight, were developed in an attempt to overcome the imprecision with the use of measured body weight. However, as shown here and previously, this modification results in substantially worse performance compared with measured GFR.37, 38, 39 Use of standardized serum creatinine values leads to another source of error for the CG equation. In previous analyses of these same data, we showed that the CG equation kidney function estimates were 11.4% greater than measured GFR with standardized creatinine compared with 2% greater with nonstandardized values.2 Serum samples are not available to enable reexpression of the CG equation for standardized serum creatinine. Altogether, these considerations do not support continued sole reliance on the CG equation for estimating kidney function for drug dosing adjustments.

Our finding of 11% to 29% discordance between the MDRD Study and CG equations overall and in subgroups is consistent with some previous studies that showed discordance rates of approximately 20% to 40% between the equations.20, 21, 22, 23, 24 Possible explanations for the variation in reported discordance rates may be related to true differences in the accuracy of equations among study populations included in the different reports or to variations in the methods used to estimate kidney function (eg, actual versus ideal body weight for the CG equation), units of kidney function (ie, adjustment versus no adjustment for BSA), or the presence or absence of calibration of the creatinine assay. Our results also are consistent with 1 study that compared carboplatin doses determined by means of nuclear imaging of the kidneys to doses calculated using estimates based on the MDRD Study and CG equations and showed that the MDRD Study equation resulted in more accurate dosing.40

Strengths of our approach include a large and diverse population; inclusion of measured GFR determined by means of urinary clearance of iodine-125–iothalamate as the gold standard, calibrated serum creatinine in all studies, and standard units for all equations; and inclusion of medications representative of those commonly used in both inpatient and outpatient settings and that have narrow therapeutic windows or are associated commonly with dosing errors or adverse drug events.

There are several limitations. First, results presented here may not be fully applicable to other populations with characteristics that are different from the present study population. For example, in populations with a lower prevalence of chronic kidney disease, we would expect greater concordance rates because drug dosing adjustment is relevant only to patients with kidney disease. As such, our findings of differences among patients of different characteristics may reflect differences in levels of GFR of study participants, rather than patient characteristics per se. However, this data set is more diverse than prior studies, and performance of the equations here may be more representative of their performance when applied in clinical practice than prior studies. Second, we have not considered the contribution of tubular reabsorption or secretion to renal clearance of drugs. However, pharmacokinetic studies do not measure tubular reabsorption or secretion directly, and tubular handling of creatinine is not likely to reflect tubular handling of many drugs that are secreted actively by a number of transporters primarily along the renal proximal tubule.41 Third, our results are based on simulation rather than pharmacokinetic studies, and we included only a sample of commonly used drugs. Finally, we used drug dosage recommendations as an outcome, rather than observed drug efficacy and safety.

All 3 equations are based on serum creatinine and therefore all have the same irremediable limitations of creatinine as a filtration marker. Serum level of creatinine is determined by factors other than GFR, in particular, tubular secretion, muscle mass, and diet, leading to bias in some populations and imprecision for all.42 This is particularly relevant for populations with reduced muscle mass, including the frail elderly, critically ill, or patients with cancer.43 Finally, kidney function must be at a steady state to use any endogenous filtration markers; thus, estimates must be used cautiously in hospitalized patients.

Adjustment of drug dosages is the most common use of kidney function estimates, and these results have implications for prescriptions of both new and existing drugs. The MDRD Study equation is used commonly as a clinical tool for detection and stratification of kidney disease, is widely available to most clinicians, and currently provides the best approximation of measured GFR.16 Using the same estimate for drug dosing and detection and evaluation of kidney disease likely would facilitate clinical decision making and improve care. The stated intent of the FDA Guidance for Industry is to use measures of kidney function that are “used widely in patient care settings” because such measures are “more practical than other alternatives.”1(p6) For new drug development, we propose that pharmaceutical manufacturers use the MDRD Study equation for pharmacokinetic studies and in dosing recommendations listed in the product label. For drugs currently in use, it is neither practical nor feasible for pharmacokinetic studies to be repeated using the MDRD Study equation. We propose that either the MDRD Study or CG equation using actual body weight be used for determination of drug dosage. If more accurate equations replace the MDRD Study equation for GFR estimation by clinical laboratories, these equations should be used instead and the FDA Guidance for Industry should incorporate this flexibility in its recommendations.

Currently, no equation provides accurate estimates for all patients. Clinicians must use available estimates together with their best judgment to determine drug dosing for individual patients, particularly for medications with a narrow therapeutic index or high toxicity. For individual patients in whom kidney function estimates from different equations vary substantially, the dose should be determined by weighing the risk of toxicity with a greater dose versus the risk of subtherapeutic dose and treatment failure with a lower dose. If both risks are high, it may be prudent to measure GFR or creatinine clearance before administration of the medication. For such medications, it may be prudent to measure GFR or creatinine clearance in all patients at the extremes of muscle mass in whom all creatinine-based estimates are suspected to be inaccurate. For some drugs, monitoring serum concentrations can minimize errors caused by inaccurate dosage adjustment based on kidney function estimates (eg, aminoglycosides, phenytoin, and lithium). Implementation of computerized clinical decision support systems including automated drug dosing is becoming more common, making such individual assessments of kidney function feasible. Such systems also will easily incorporate more accurate equations as they become used in clinical practice and would facilitate conversion of the MDRD Study equation–derived estimates from units of milliliters per minute per 1.73 m2 to units of milliliters per minute, as is recommended for drug dosing.

In conclusion, the MDRD Study equation had greater concordance with measured GFR for recommended drug dosage than the CG equation. Concordance among equations was greater in the context of specific medications. Either the MDRD Study or CG equation could be used for drug dosage adjustments in most circumstances. Patients for whom eGFR from creatinine is likely to be inaccurate require careful consideration. Greater education is needed for physicians, pharmacists, industry, and the public about chronic kidney disease and the interpretation of GFR estimates for use in drug dosing.