Impact of the 2021 CKD-EPI equation on anticancer pharmacotherapy in Black and non-Black cancer patients

Abstract

Cancer and kidney disease disproportionately impact Black patients. The CKD-EPI₂₀₂₁ equation was developed to estimate glomerular filtration rate (eGFR) without using race. We assessed the impact of using CKD-EPI₂₀₂₁ instead of CKD-EPI₂₀₀₉ or Cockcroft-Gault (CG) on dosing and eligibility of anticancer drugs in Black and non-Black patients.

Utilizing the National Cancer Institute Theradex database, deindexed eGFR (mL/min) was calculated for 3931 patients (8.6% Black) using CKD-EPI₂₀₂₁, CKD-EPI₂₀₀₉, and CG. Dosing simulations based on each eGFR were performed for ten anticancer drugs with kidney function-based eligibility or dosing cutoffs.

eGFR differences using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ varied between Black and non-Black patients (p<0.001); on average, Black patients had 10.3 mL/min lower eGFR and non-Black patients had 4.2 mL/min higher eGFR using CKD-EPI₂₀₂₁. This corresponded to a difference in relative odds of cisplatin ineligibility using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉; Black patients had 48% higher odds of ineligibility and non-Black patients had 27% lower odds of ineligibility using CKD-EPI₂₀₂₁ (p<.001). When using CKD-EPI₂₀₂₁ versus CG, eGFR differences were similar between Black and non-Black patients (p=0.679) and relative difference in odds of cisplatin ineligibility did not vary.

Using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ differentially impacts Black versus non-Black cancer patients; Black patients have lower calculated eGFR and are less likely to receive full doses of drug using CKD-EPI₂₀₂₁. From the historical default of CG, adopting CKD-EPI₂₀₂₁ would not disparately impact patients based on race, but would result in Black patients being less likely to receive full doses of drug than if CKD-EPI₂₀₀₉ were used.

1. INTRODUCTION

Cancer disproportionately impacts Black patients in incidence and mortality.^1–3 Accordingly, optimal anticancer pharmacotherapy is critical to improving outcomes in this population. However, Black patients are less likely to receive anticancer pharmacotherapy.^{1, 4, 5} The reasons for these treatment disparities are not fully understood but are likely multifactorial and complex.

Kidney disease is common in cancer patients; about 25% of solid tumor patients have an estimated glomerular filtration rate (GFR; eGFR) <60 mL/min.^{6, 7} Kidney function is a key factor in anticancer drug selection and dosing, as many anticancer drugs are omitted from regimens or dose-reduced on the basis of decreased GFR and consequently decreased drug clearance.⁸ Although some drugs, such as carboplatin, are dosed continuously based on eGFR, the majority of anticancer drugs have categorical renal dose reductions and/or omissions; in other words, once a patient’s eGFR falls below a certain threshold, the patient will receive a lower dose of or become ineligible to receive the drug. Drug eligibility cutoffs can range from 10–60 mL/min, and renal dose reductions can result in over a 50% reduction of dose typically administered.⁸ Therefore, even a relatively small nominal change in calculated eGFR may have significant clinical consequences for a patient’s pharmacotherapy if it causes a patient’s eGFR to cross one of these thresholds. Importantly, Black patients are disproportionately impacted by kidney disease and display increased incidence of GFR <60 mL/min,⁹ a common cutoff for drug dosing and eligibility.

Several bedside equations are available to assess kidney function, including the Cockcroft-Gault (CG) and 2009 Chronic Kidney Disease-Epidemiology Collaboration (CKD-EPI₂₀₀₉) equations.^{10, 11} Although CKD-EPI₂₀₀₉ is the most accurate equation to estimate GFR in cancer patients, CG continues to be routinely utilized in clinical oncology.^12–15 CG calculates estimated creatinine clearance as a surrogate for GFR using serum creatinine, age, sex, and weight. CG was developed in a small study of White male participants and is inaccurate in many patient populations, including Black patients, cancer patients, older patients, and patients of extreme body size.^{11, 13, 16, 17} In contrast, CKD-EPI₂₀₀₉ was developed in a large, diverse population and incorporates serum creatinine, age, sex, body surface area (BSA), and race into its eGFR calculation. All other input variables being equal, CKD-EPI₂₀₀₉ calculates a 15.9% higher eGFR for Black versus non-Black patients to account for the higher serum creatinine observed in Black versus non-Black patients at the same measured GFR, sex, and age.¹⁰

In 2020, the use of race as a covariate in GFR-estimating equations was reconsidered over concerns that it may contribute to implicit bias in medicine.¹⁸ As a result, some health systems advocated for simply omitting the race term from CKD-EPI₂₀₀₉ calculations (i.e., assigning the non-Black value for all patients; CKD-EPI_{2009, without race}). We previously reported that using CKD-EPI_{2009, without race} to determine anticancer therapy would calculate a lower eGFR for Black cancer patients, thus excluding more Black patients from receiving full doses of anticancer drugs.¹⁹ In 2021, a new version of the CKD-EPI equation that excludes race (by re-deriving the regression equation without race as a covariate) was published (CKD-EPI₂₀₂₁) and recommended for immediate implementation by the nephrology community.^{20, 21} CKD-EPI₂₀₂₁ calculates eGFR on the basis of serum creatinine, age, and sex.

To our knowledge, there are no reports investigating the potential clinical consequences for patients’ anticancer drug eligibility and/or dosing that may be caused by health systems adopting the CKD-EPI₂₀₂₁ equation in place of either CKD-EPI₂₀₀₉ (the equation recommended for use in cancer patients)¹² or CG (the equation most often utilized for anticancer drug eligibility and dosing).¹⁵ Furthermore, it is unknown whether such potential clinical consequences would be similar for both Black and non-Black patients. The aim of the current study was to perform a complementary analysis to our prior findings by determining the impact of using CKD-EPI₂₀₂₁ in place of CKD-EPI₂₀₀₉ or CG on dosing and eligibility of ten anticancer drugs in Black and non-Black patients.

2. MATERIALS AND METHODS

2.1. Dataset and eGFR calculations

The dataset was extracted from the National Cancer Institute Theradex database and has been described previously.^{19, 22} Estimates of GFR using CKD-EPI₂₀₂₁²⁰ and CKD-EPI₂₀₀₉,¹⁰ and creatinine clearance using CG¹¹ were calculated (estimates will be collectively referred to as eGFR for brevity). Kidney disease stage was assigned according to clinical guidelines²³ and stage discordance (i.e., patients assigned to a different kidney disease stage when using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ or CG) was determined. Further details are in Supplementary Methods.

2.2. Drug dosing simulations

A simulation study was performed as previously described¹⁹ to compare eligibility and dosing recommendations for ten anticancer drugs based on eGFR calculated by CKD-EPI₂₀₂₁, CKD-EPI₂₀₀₉, and CG. Further details are in Supplementary Methods.

2.3. Statistical analysis

Differences between Black and non-Black patients were compared using two sample t-test for continuous variables and Fisher exact test or χ² test for categorical variables. Agreement in drug eligibility and dosing recommendations based on CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG was assessed by κ statistic with linear weighting.²⁴

We further assessed the difference in eGFR calculated by the different equations using linear mixed effects modeling to account for correlations between eGFR values from the same patient. Differences in eGFR between equations are reported with the 95% confidence interval (CI). We also assessed the relative difference in drug ineligibility using generalized estimating equation logistic regression to account for correlations between drug ineligibility from the same patient. Relative differences in drug ineligibility between equations are reported as odds ratio (OR) with the 95% CI. The impact of sex, race, age or BSA on the difference in eGFR and drug ineligibility by each eGFR equation was examined by considering the interaction between each covariate and each eGFR equation. Statistical significance was specified as p<0.05, and all tests were two-sided. Further details are in Supplementary Methods.

3. RESULTS

3.1. Study population

The dataset included 4118 patients. After removing patients without race data (n=1), without kg as weight units (n=136), and without height (n=36) or height <100 cm (n=14), 3931 patients remained. 8.6% of the population was Black and most of the non-Black population was White (88.3% of total population). Patient characteristics are reported in Suppl.Table 1.

3.2. eGFR calculations

eGFR values for Black and non-Black patients are reported in Suppl.Table 2. Black patients exhibited a mean 10.5 mL/min decrease in eGFR using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉, whereas non-Black patients exhibited a mean 4.1 mL/min increase (p<0.001). All Black patients exhibited a decrease in eGFR and the majority of non-Black patients exhibited an increase in eGFR using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ (Figure 1A). CKD stage discordance using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ was 15.3% for Black patients and 9.9% for non-Black patients (p=0.002) (Suppl.Table 2). Black patients exhibited a mean 3.6 mL/min decrease in eGFR using CKD-EPI₂₀₂₁ versus CG, whereas non-Black patients exhibited a mean 0.5 mL/min decrease (p<0.01).

Figure 1. Change in eGFR based on CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG for Black and non-Black patients.

(A) eGFR was calculated using CKD-EPI₂₀₂₁, CKD-EPI₂₀₀₉, and CG for Black (gold, n=340) and non-Black (blue, n=3591) patients. The percent change in eGFR between CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ (left) and CKD-EPI₂₀₂₁ versus CG (right, note larger Y-axis range compared to left panel) was calculated. (B) A linear mixed-effect random intercept model was constructed to assess the impact of eGFR equation (CKD-EPI₂₀₂₁, reference; CKD-EPI₂₀₀₉, blue; CG, gold) on eGFR, controlling for race, sex, BSA, and age. Details of the model are given in Suppl.Table 3. Reference: eGFR calculated by CKD-EPI₂₀₂₁ for a non-Black male of average age and BSA.

Asterisks indicate significance of the interaction between eGFR equation and covariate terms, *p<.05, **p<.01, ***p<.001. Error bars indicate the 95% confidence intervals of the β coefficient point estimate.

Abbreviations: BSA, body surface area; CG, Cockcroft-Gault equation; CKD-EPI₂₀₀₉, 2009 Chronic Kidney Disease-Epidemiology Collaboration equation; CKD-EPI₂₀₂₁, 2021 Chronic Kidney Disease-Epidemiology Collaboration equation; eGFR, estimated glomerular filtration rate

Difference in calculated eGFR depending on eGFR equation is depicted by race, sex, age, and BSA in Figure 1B (details of the model are given in Suppl.Table 3). The interaction between CKD-EPI₂₀₀₉ and race was significant (p<0.001), indicating that the difference in eGFR using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ differs significantly between Black and non-Black patients. For instance, a Black male of average age and BSA has 10.3 mL/min (95% confidence interval [CI]: 10.1–10.5 mL/min) lower eGFR using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉, whereas a non-Black counterpart has 4.2 mL/min (95% CI: 4.1–4.3 mL/min) higher eGFR. In contrast, the difference in eGFR using CKD-EPI₂₀₂₁ versus CG was similar between Black and non-Black patients (p=0.679). The difference in eGFR using CKD-EPI₂₀₂₁ versus both CKD-EPI₂₀₀₉ and CG varied by sex, age, and BSA (Figure 2).

Figure 2. Change in eGFR based on CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG for Black and non-Black patients of varying sex, age, and BSA.

n=732 patients were simulated with a combination of various patient characteristics (Black [gold] versus non-Black [blue]; male [solid line] versus female [dashed line]; age [20–80 years]; and BSA [low=1.6 m², average=1.9 m², high=2.2 m²]). The change in eGFR calculated by CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ (top row) and CG (bottom row) was calculated for each simulated patient based on the linear mixed effect model estimates described in Figure 1 and Suppl.Table 3.

Abbreviations: BSA, body surface area; CG, Cockcroft-Gault equation; CKD-EPI₂₀₀₉, 2009 Chronic Kidney Disease-Epidemiology Collaboration equation; CKD-EPI₂₀₂₁, 2021 Chronic Kidney Disease-Epidemiology Collaboration equation; eGFR, estimated glomerular filtration rate

3.3. Drug dosing simulations

Impact on eligibility

The proportion of Black patients ineligible for drug was up to 10.0% using CKD-EPI₂₀₂₁, 7.4% using CKD-EPI₂₀₀₉, and 12.7% using CG. The proportion of non-Black patients ineligible for drug was up to 8.6% using CKD-EPI₂₀₂₁, 11.5% using CKD-EPI₂₀₀₉, and 14.0% using CG. (Suppl.Table 4). For Black and non-Black patients, eligibility agreement was almost perfect between CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ (κ=0.8187–0.9228) and moderate-to-substantial between CKD-EPI₂₀₂₁ and CG (κ=0.5786–0.7515) (Suppl.Table 5). This corresponded to similar eligibility discordance rates for Black and non-Black patients using CKD-EPI₂₀₂₁ versus both CKD-EPI₂₀₀₉ and CG (Table 1). However, more Black patients were ineligible using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉, whereas more non-Black patients were eligible using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ (Suppl.Table 4).

Table 1.

Rates of drug eligibility and dosing discordance in Black and non-Black cancer patients based on eGFR calculated using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG

	CKD-EPI2021 vs CKD-EPI2009			CKD-EPI2021 vs CG
	Black (n=340)	Non-Black (n=3591)	p-value	Black (n=340)	Non-Black (n=3591)	p-value
Drugs with renal eligibility recommendations a
Cisplatin	9 [2.7]	102 [2.8]	0.837c	17 [5.0]	280 [7.8]	0.062c
Pemetrexed	1 [0.3]	20 [0.6]	1.000d	7 [2.1]	70 [2.0]	0.889c
Bendamustine	1 [0.3]	17 [0.5]	1.000d	2 [0.6]	32 [0.9]	0.765d
Mitomycin	2 [0.6]	2 [0.1]	0.04d	1 [0.3]	8 [0.2]	0.557d
Drugs with renal dosing recommendations b
Oxaliplatin	2 [0.6]	2 [0.1]	0.04d	1 [0.3]	8 [0.2]	0.557d
Capecitabine	9 [2.7]	57 [1.6]	0.146c	11 [3.2]	151 [4.2]	0.390c
Etoposide	7 [2.1]	55 [1.5]	0.456c	10 [2.9]	143 [4.0]	0.343c
Topotecan	1 [0.3]	19 [0.5]	1.000d	2 [0.6]	34 [1.0]	0.766d
Fludarabine	31 [9.1]	262 [7.3]	0.222c	5 [16.2]	623 [17.4]	0.584c
Bleomycin	10 [2.9]	61 [1.7]	0.1000c	14 [4.1]	161 [4.5]	0.755c

Reported as

^an [%] of patients with different eligibility recommendations, or

^bn [%] of patients recommended to receive different dose of drug, depending on the eGFR equation used

Calculated by

^cχ² test or

^dFisher exact test

Relative difference in drug ineligibility for cisplatin, pemetrexed, and bendamustine depending on eGFR equation is depicted by race, sex, age and BSA in Figure 3 (details of model are given in Suppl.Table 6). For cisplatin, the interaction between CKD-EPI₂₀₀₉ and race was significant (p<0.001), indicating that relative difference in OR of cisplatin ineligibility using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ differs between Black and non-Black patients. For example, a Black male of average age and BSA has 48% higher odds of ineligibility for cisplatin using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉. Conversely, a non-Black counterpart has 27% lower odds for cisplatin ineligibility using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉. Pemetrexed exhibited a similar trend; Black patients had 16% higher odds and non-Black patients had 23% lower odds for pemetrexed ineligibility using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉, with the interaction approaching significance (p=0.056). No significant differences were observed for the relative difference in OR of bendamustine ineligibility using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ between Black and non-Black patients. Similarly, no difference in relative difference in ORs for cisplatin, pemetrexed, and bendamustine ineligibility using CKD-EPI₂₀₂₁ versus CG were observed between Black and non-Black patients. The OR of cisplatin, pemetrexed, and bendamustine ineligibility using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG varied by sex, age, and BSA (Figure 4).

Figure 3. Odds ratio of drug ineligibility based on CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG for Black and non-Black patients.

Generalized estimating equation logistic regression models were constructed to assess the impact of eGFR equation (CKD-EPI₂₀₂₁, reference; CKD-EPI₂₀₀₉, blue; CG, gold) on the odds of ineligibility for cisplatin, pemetrexed, and bendamustine for patients, controlling for race, sex, BSA, and age. Details of the models are given in Suppl.Table 6. Reference: Odds using eGFR calculated by CKD-EPI₂₀₂₁ for a non-Black male of average age and BSA.

Asterisks indicate significance of the interaction between eGFR equation and covariate terms, *p<.05, **p<.01, ***p<.001. Error bars indicate the 95% confidence intervals of the β coefficient point estimate.

Abbreviations: BSA, body surface area; CG, Cockcroft-Gault equation; CKD-EPI₂₀₀₉, 2009 Chronic Kidney Disease-Epidemiology Collaboration equation; CKD-EPI₂₀₂₁, 2021 Chronic Kidney Disease-Epidemiology Collaboration equation; eGFR, estimated glomerular filtration rate

Figure 4. Odds ratio of drug ineligibility based on CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG for Black and non-Black patients of varying sex, age, and BSA.

n=732 patients were simulated with a combination of various patient characteristics (Black [gold] versus non-Black [blue]; male [solid line] versus female [dashed line]; age [20–80 years]; and BSA [low=1.6 m², average=1.9 m², high=2.2 m²]). The odds ratio of ineligibility for cisplatin, pemetrexed, and bendamustine based on eGFR calculated by CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ (top rows) and CG (bottom rows) were calculated for each simulated patient based on the generalized estimating equation logistic regression model estimates described in Figure 3 and Suppl.Table 6.

Abbreviations: BSA, body surface area; CG, Cockcroft-Gault equation; CKD-EPI₂₀₀₉, 2009 Chronic Kidney Disease-Epidemiology Collaboration equation; CKD-EPI₂₀₂₁, 2021 Chronic Kidney Disease-Epidemiology Collaboration equation; eGFR, estimated glomerular filtration rate; OR, odds ratio

Impact on dosing

The proportion of Black patients requiring dose reduction ranged up to 28.2% using CKD-EPI_2021, 21.8% using CKD-EPI₂₀₀₉, and 34.4% using CG. The proportion of non-Black patients requiring dose reduction ranged from up to 27.9% using CKD-EPI₂₀₂₁, 33.9% using CKD-EPI₂₀₀₉, and 38.1% using CG (Suppl.Table 4). Dosing agreement between CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ was substantial-to-almost perfect for Black patients (κ=0.68–0.86) and almost perfect for non-Black patients (κ=0.82–0.92). Dosing agreement between CKD-EPI₂₀₂₁ and CG was substantial for Black patients (κ=0.67–0.75) and moderate-to-substantial for non-Black patients (κ=0.59–0.71) (Suppl.Table 5), corresponding to similar rates of dosing discordance for Black and non-Black patients using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and CG (Table 1). However, more Black patients were recommended to receive a full dose of drug using CKD-EPI₂₀₀₉ versus CKD-EPI₂₀₂₁, whereas more non-Black patients were recommended to receive a full dose of drug using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ (Suppl.Table 4).

4. DISCUSSION

The CKD-EPI₂₀₂₁ equation was intended to accurately estimate GFR with consequences that do not disproportionately impact any specific group of individuals.²¹ Both CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ are accurate GFR-estimating equations and calculate eGFR within 30% of measured GFR for ≥85% for Black, non-Black, and cancer patients.^{13, 20, 25} The similar accuracy of CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ is reflected in comparable agreement and discordance rates observed for drug eligibility and dosing recommendations for Black and non-Black patients (κ≥0.58 and 0.59 for Black and non-Black patients, respectively).

However, these favorable agreement and discordance rates are dominated by the majority of patients who have eGFRs significantly above the thresholds for drug eligibility and dosing adjustment; the mean eGFR in our cohort was >90 mL/min and <15% of patients had eGFR <60 mL/min. Overall agreement and discordance rates do not account for directionality (i.e., changing from eligible to ineligible or to a higher versus lower dose), nor do they reflect trends and changes that occur at the individual level for subgroups of patients (i.e., for Black versus non-Black patients at high versus low GFR). Indeed, there are disproportionate consequences of using CKD-EPI₂₀₂₁ in place of CKD-EPI₂₀₀₉ to calculate eGFR for Black versus non-Black patients, as demonstrated by the results of our linear mixed effect model of eGFR. The interaction term between race and equation (CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉₎ was significant (p<0.001), indicating that using CKD-EPI₂₀₂₁ in place of CKD-EPI₂₀₀₉ impacts a patient’s calculated eGFR differently depending on race. Whereas Black patients in our cohort exhibited a mean 10.5 mL/min decrease in eGFR using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉, non-Black patients exhibited a mean 4.1 mL/min increase. The explanation for this result may lay in the differential bias of CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ for Black versus non-Black patients.

The change in bias from CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ differs between Black versus non-Black patients in terms of the direction of bias (i.e., over- or underestimation) and patient subgroup (i.e., for patients at high versus low GFR). In Black patients, CKD-EPI₂₀₀₉ overestimates GFR by 3.7 mL/min, whereas CKD-EPI₂₀₂₁ underestimates GFR by 3.6 mL/min.²⁰ This shift from over- to underestimation led to a decrease in eGFR for all Black patients in our cohort when using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉. Importantly, underestimation of GFR by CKD-EPI₂₀₂₁ in Black patients is more prominent as GFR decreases. Therefore, the change in bias resulting from switching from CKD-EPI₂₀₀₉ to CKD-EPI₂₀₂₁ is more likely to impact patients who are already at risk of falling below drug eligibility and dosing thresholds (i.e., patients with already low GFR). This could potentially inappropriately exclude Black cancer patients from therapy.

Conversely, in non-Black patients, CKD-EPI₂₀₀₉ underestimates GFR by 0.5 mL/min and CKD-EPI₂₀₂₁ overestimates by 3.9 mL/min.²⁰ This differential bias resulted in the majority of non-Black patients experiencing an increase in eGFR when using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉. Importantly, overestimation of GFR by CKD-EPI₂₀₂₁ in non-Black patients is more prominent as GFR increases. Therefore, the change in bias resulting from switching from CKD-EPI₂₀₀₉ to CKD-EPI₂₀₂₁ is likely to affect non-Black patients who are not at risk of falling below drug eligibility and dosing thresholds (i.e., patients with a high GFR), and is therefore unlikely to exclude additional patients from therapy.²⁰

These trends in bias are reflected in the results of our dosing simulations; the proportions of patients recommended to be ineligible or to receive a decreased dose of drug using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ is higher for Black patients and lower for non-Black patients. Although similar proportions of Black and non-Black patients displayed cisplatin eligibility discordance between CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ (2.7% versus 2.8%, respectively), the direction of that discordance differs – Black patients had 48% higher odds and non-Black patients had 27% lower odds of cisplatin ineligibility using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉, and the interaction term between race and equation (CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉₎ was significant for cisplatin (p<0.001)_. Therefore, using CKD-EPI₂₀₂₁ in place of CKD-EPI₂₀₀₉ to determine cisplatin ineligibility will impact Black versus non-Black patients differently depending on race. Paradoxically, unilaterally using eGFR calculated by CKD-EPI₂₀₂₁ instead of CKD-EPI₂₀₀₉ to determine drug eligibility and dosing may actually exacerbate already existing treatment disparities and worsen anticancer outcomes for Black patients. Our findings align with previous reports that the CKD-EPI₂₀₂₁ equation calculates a lower eGFR for Black cancer patients, increasing exclusion of Black patients from oncology clinical trials.²⁶

It would be ideal to utilize the eGFR equation with the most favorable profile of accuracy, precision, and bias, particularly in cancer patients who are likely to be impacted by drug eligibility and dosing recommendations (i.e., those with GFR <60 mL/min). Unfortunately, the performance of CKD-EPI₂₀₂₁ specifically in Black and non-Black cancer patients with GFR <60 mL/min has not been reported to our knowledge. Furthermore, performance of CKD-EPI₂₀₂₁ has not been reported in a US or international cancer population. In a multiracial cohort of Brazilian cancer patients, both CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ tended to overestimate GFR. CKD-EPI₂₀₂₁ had lower median bias than CKD-EPI₂₀₀₉ in Black Brazilian cancer patients (3.5 and 12.9 mL/min/1.73 m², respectively), but the race-specific bias at different levels of GFR was not reported.^{13, 25} In a cohort of Danish cancer patients, CKD-EPI₂₀₂₁ had lower absolute bias than CKD-EPI₂₀₀₉, though that appears to be driven by superior performance of CKD-EPI₂₀₂₁ at higher GFRs (>60 mL/min). Additionally, the cohort was presumed to be predominantly white.²⁷

CG continues to be utilized in oncology for drug eligibility and dosing, despite its inaccuracy relative to contemporary eGFR equations, including CKD-EPI₂₀₀₉.^12–15 CKD-EPI₂₀₂₁ has improved accuracy and precision versus CG in cancer patients.²⁵ Although the average change in eGFR calculated by CKD-EPI₂₀₂₁ versus CG in our patient cohort was not of a clinically significant magnitude for either Black or non-Black patients (<4 mL/min), the difference in eGFR for individual patients could be quite large (>25%, even in patients with eGFR <60 mL/min). In contrast, the difference in eGFR calculated by CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉ was within ~20% for all patients. This is likely a function of the relative imprecision of CG compared to both CKD-EPI₂₀₂₁ and CKD-EPI₂₀₀₉. Using CKD-EPI₂₀₂₁ to calculate eGFR led to a greater proportion of patients eligible for drug or receiving full dose of drug versus using CG.

Importantly, there was no difference between Black and non-Black patients in the eligibility or dosing discordance rates or the directionality of eligibility discordance using CKD-EPI₂₀₂₁ versus CG. The interaction terms between equation (CKD-EPI₂₀₂₁ versus CG) and race were not significant in either our eGFR model or drug ineligibility models, suggesting that replacing CG with CKD-EPI₂₀₂₁ impacts patients equally irrespective of race. Therefore, implementing use of deindexed CKD-EPI₂₀₂₁ in place of CG would already represent a major improvement in clinical onconephrology by modernizing GFR estimation methods, increasing the proportion of patients eligible for drug or receiving full doses of drugs, and achieving harmonization of kidney function estimation methodologies across clinical arenas, while avoiding disparate impact on Black versus non-Black patients. However, implementing CKD-EPI₂₀₂₁ would result in Black patients having lower calculated eGFR and less likely to receive full doses of drug than if CKD-EPI₂₀₀₉ were used.

Prior to publication of the CKD-EPI₂₀₂₁ equation, some health systems were simply omitting the race multiplier from CKD-EPI₂₀₀₉ eGFR calculations (i.e., CKD-EPI_{2009, without race}). In a previous analysis of this dataset, we reported that using CKD-EPI_{2009, without race} versus CKD-EPI₂₀₀₉ leads to a median 14 mL/min decrease in calculated eGFR for Black cancer patients, corresponding to up to 5% and 18% of Black patients receiving a discordant recommendation for drug eligibility and dosing, respectively.¹⁹ In comparison, we demonstrate here that using the CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ leads to a 10.5 mL/min decrease in Black cancer patients. This corresponds to approximately half of the rate of eligibility and dosing discordance as seen when using CKD-EPI_{2009, without race} (up to 2.7% and 9.1%, respectively). Therefore, within the current framework of avoiding the use of race in medical algorithms and clinical decision-making, implementing CKD-EPI₂₀₂₁ instead of CKD-EPI_{2009, without race} is preferable because it decreases the proportion of Black patients who would receive discordant drug eligibility (from eligible to ineligible) or dosing (from a higher to a lower dose) recommendations.

Our analysis has several limitations. Firstly, there were a low number of events (i.e., patients ineligible for a certain drug) in Black patients for pemetrexed and bendamustine. Therefore, the statistical power of the models may be limited in detecting a significant difference in drug ineligibility between groups. Secondly, we did not measure GFR to compare the accuracy and bias of eGFR calculations. Thirdly, our study population included patients enrolled in phase 1 studies and therefore did not reflect the prevalence of kidney disease in the general cancer population.^{6, 7} Finally, the study was simulation-based and therefore did not represent a true change to clinical care. However, the findings of the study are illustrative of the potential impact of using CKD-EPI₂₀₂₁ to determine drug eligibility and dosing for Black and non-Black cancer patients.

Regardless of the equation used, eGFR is only one of many datapoints to be used for assessing the risk-benefit ratio of a given anticancer drug in a given patient scenario.^{8, 28} Clinicians should always follow a patient-centered approach to determining anticancer drug eligibility and dosing, which assesses the potential clinical benefit, goal of therapy, and patient-specific toxicity risk. If a more precise estimate of GFR would clarify the risk-benefit ratio of a given anticancer drug for a given patient, then including cystatin C in eGFR calculations or directly measuring GFR may be warranted.^{20, 21} Although there has been discourse regarding the validity of cystatin C as a marker of GFR in cancer patients,¹² adding cystatin C improved the performance of the CKD-EPI₂₀₀₉ and CKD-EPI₂₀₂₁ equations in Black and non-Black cancer patients.^{13, 25} However, currently there exist several logistical barriers to widespread adoption of cystatin C in clinical practice (i.e., assay standardization and availability).²⁹

In summary, this analysis demonstrates that utilizing CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ may differentially impact Black versus non-Black cancer patients, corresponding to differential likelihood of being deemed ineligible for some anticancer drugs based on eGFR. Importantly, drug eligibility and dosing recommendations based on either CKD-EPI₂₀₂₁ or CKD-EPI₂₀₀₉ lead to more patients deemed eligible for or to receive full dose of drug than when utilizing CG. Furthermore, using CKD-EPI₂₀₂₁ versus CG impacts the calculation of eGFR and likelihood of drug ineligibility similarly in Black versus non-Black patients. Thus, from the historical default of CG, adopting CKD-EPI₂₀₂₁ would not disparately impact patients based on race, but would result in Black patients having lower calculated eGFR and less likely to receive full doses of drug then if CKD-EPI₂₀₀₉ were used, though the latter requires the input of race as a variable. As always, eGFR should be utilized as a tool for assessing risk-benefit ratio of cancer pharmacotherapy, and clinical judgment and collaborative interdisciplinary decision-making is paramount for optimizing oncology care.

Translational Relevance.

Assessing kidney function is a key consideration when prescribing anticancer pharmacotherapy. This is typically achieved by calculating estimated glomerular filtration rate (eGFR) using one of several available bedside equations, including the Cockcroft-Gault, CKD-EPI_2009, or CKD-EPI₂₀₂₁ equations. While CKD-EPI₂₀₀₉ includes a race variable in its calculations, CKD-EPI₂₀₂₁ was developed explicitly without race as an input variable over concerns of implicit bias in medicine. We explored the impact of using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ and Cockcroft-Gault on anticancer drug dosing and eligibility in Black and non-Black patients. eGFR using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉ was lower (10.3 mL/min) for Black patients and higher (4.2 mL/min) for non-Black patients (p<0.001). Consequently, Black patients had 48% higher and non-Black patients had 27% lower relative odds of cisplatin ineligibility using CKD-EPI₂₀₂₁ (p<0.001). eGFR difference and relative odds of cisplatin ineligibility using CKD-EPI₂₀₂₁ versus Cockcroft-Gault did not vary between Black and non-Black patients. Black patients are less likely to receive full doses of drug using CKD-EPI₂₀₂₁ versus CKD-EPI₂₀₀₉, but not CG. Depending on the kidney function estimating formula used at any particular institution, existing healthcare disparities may be worsened while avoiding race as a variable.

Highlights:

• CKD-EPI₂₀₂₁ removes race from eGFR calculations over concerns of implicit bias

• Nephrologists recommend replacing CKD-EPI₂₀₀₉ with CKD-EPI₂₀₂₁ to calculate eGFR

• eGFR calculated by CKD-EPI₂₀₂₁ is lower for Black and higher for non-Black patients

• Black patients had 48% higher odds of cisplatin ineligibility using CKD-EPI₂₀₂₁

• Black patients are less likely receive full doses of drug using CKD-EPI₂₀₂₁

DATA AVAILABILITY

The data underlying this article were provided by NCI by permission. Data will be shared on request to the corresponding author with permission of NCI.