Carboplatin Dosing in the Treatment of Ovarian Cancer: An NRG Oncology Group Study

Abstract

Objective

To determine the effects of using National Comprehensive Cancer Network (NCCN) guidelines to estimate renal function on carboplatin dosing and explore adverse effects associated with a more accurate estimation of lower creatinine clearance (CrCl).

Methods

Retrospective data were obtained for 3,830 of 4,312 patients treated on GOG182 (NCT00011986)–a phase III trial of platinum-based chemotherapy for advanced-stage ovarian cancer. Carboplatin dose per patient on GOG182 was determined using the Jelliffe formula. We recalculated CrCl to determine dosing using Modification of Diet in Renal Disease (MDRD) and Cockcroft-Gault (with/without NCCN recommended modifications) formulas. Associations between baseline CrCl and toxicity were described using the area under the receiver operating characteristic curve (AUC). Sensitivity and positive predictive values described the model’s ability to discriminate between subjects with/without the adverse event.

Results

AUC statistics (range, 0.52-0.64) showed log(CrCl_Jelliffe) was not a good predictor of grade ≥3 adverse events (anemia, thrombocytopenia, febrile neutropenia, auditory, renal, metabolic, neurologic). Of 3,830 patients, 628 (16%) had CrCl <60 mL/min. Positive predictive values for adverse events ranged from 1.8%-15%. Using the Cockcroft-Gault, Cockcroft-Gault with NCCN modifications, and MDRD (instead of Jelliffe) formulas to estimate renal function resulted in a >10% decrease in carboplatin dosing in 16%, 32%, and 5.2% of patients, respectively, and a >10% increase in carboplatin dosing in 41%, 9.6% and 12% of patients, respectively.

Conclusion

The formula used to estimate CrCl affects carboplatin dosing. Estimated CrCl <60 mL/min (by Jelliffe) did not accurately predict adverse events. Efforts continue to better predict renal function. Endorsing National Cancer Institute initiatives to broaden study eligibility, our data do not support a minimum threshold CrCl <60 mL/min as an exclusion criterion from clinical trials.

Introduction

Carboplatin-based chemotherapy is the mainstay treatment for many patients with gynecologic malignancies. Accurate assessment of renal function is critical for carboplatin dosing [1]. Historically, renal function has been assessed by measuring the glomerular filtration rate (GFR) of various isotopes, requiring urine collection. This approach was difficult to implement across all areas of practice, and was limited by expense, time, and nuclear exposure.

Carboplatin dose is calculated using one of several formulas, including the Jelliffe, Salazar-Corcoran, and Cockcroft-Gault formulas, which incorporate the patienťs serum creatinine value with or without patient weight [2–5]. The Gynecologic Oncology Group (GOG) previously used the Jelliffe formula, which is normalized for ideal body weight and does not incorporate the patien’s actual weight, for estimating creatinine clearance (CrCl). The Cockcroft-Gault formula is currently used for drug development and carboplatin dosing in ovarian cancer. It accounts for actual patient weight; however, it was developed using data from lean male patients. This limited its accuracy with increasing BMI and female patients until the formula was subsequently validated for these patient demographics [2]. The Modification of Diet in Renal Disease (MDRD) formula was developed to detect chronic renal impairment early, and it estimates GFR rather than CrCl. This formula, however, has not been validated for oncology drug dosing. Indirect estimates of renal function obtained from these methods vary; while they work well in the population used for development, they fail to adapt to common problems such as abnormal renal function.

Optimal carboplatin dosing is unknown for patients with renal dysfunction or low serum creatinine values in the setting of malnutrition and ascites. Our understanding of renal excretion of toxic compounds continues to improve. In recent studies, multidrug and toxic compound extraction proteins (MATE1 and MATE2/2-K) have played a critical role in renal excretion of chemotherapeutics, and these studies suggest CrCl better reflects drug clearance [6].

Historically, serum creatinine was determined using a variety of reagents and assays, resulting in significant variability between laboratories. In 2006, the National Kidney Disease Education Program (NKDEP) published recommendations to standardize serum creatinine assays by using the isotope dilution mass spectrometry (IDMS) traceable reference method [7]. The IDMS assay, however, was not universally adopted because it may require the purchase of new equipment. The Cockcroft–Gault and Jelliffe formulas, as well as other formulas used to calculate CrCl, were developed and validated using non-IDMS creatinine values. The MDRD formula was developed to identify patients with poor renal function rather than drug dose calculation, but it has been validated using both non-IDMS and IDMS methods [8].

In patients with normal renal function, creatinine values determined via IDMS are on average 10%-20% lower than non-IDMS values, leading to significantly higher CrCl estimates with the Jelliffe or Cockcroft-Gault models translating to higher than desired carboplatin doses [9, 10]. Importantly, the Jelliffe formula also over estimates CrCl, which subsequently leads to higher calculated carboplatin dosing [11]. As such, in discussion with the Food and Drug Administration (FDA), the National Cancer Institute (NCI) and Cancer Therapy Evaluation Program (CTEP) recommended capping the maximum allowable CrCl and carboplatin dose for desired exposure of drug [9, 12, 13]. In October 2010, NCI/CTEP issued an action letter describing the change to IDMS creatinine, and prompted the GOG to revise its guidelines for carboplatin dosing in 2011 [9, 14]. The GOG followed by the National Comprehensive Cancer Network (NCCN) created guidelines that recommend the Cockcroft-Gault formula instead of the Jelliffe, as well as a minimum allowable creatinine value of 0.7 mg/dL when calculating carboplatin dose. Based on literature assessing obesity and its impact on CrCl estimation, the GOG and NCCN recommended using adjusted body weight rather than actual body weight for patients with a body mass index (BMI) ≥25 kg/m² to further reduce the risk of CrCl overestimation in obese patients [14, 15].

We sought to determine how the revised 2011 GOG and NCCN guidelines affect carboplatin dosing in a large cohort of patients with ovarian cancer. Our primary objective was to compare carboplatin dose calculated using the Cockcroft-Gault formula (with and without the GOG/NCCN-specified limits) and the MDRD formula against the Jelliffe formula. Our secondary objective was to explore the relationship between reported toxicity and baseline CrCl used for carboplatin dosing, and to determine whether CrCl <60 mL/min might lead to increased toxicity.

Methods

Patient Selection

We obtained retrospective data from patients enrolled on GOG 182 [16]. GOG 182 compared the efficacy of four experimental platinum-based doublet or triplet arms (carboplatin/paclitaxel, carboplatin/paclitaxel/gemcitabine, carboplatin/paclitaxel/liposomal doxorubicin, carboplatin/topotecan to carboplatin/paclitaxel, and carboplatin/gemcitabine) against the reference arm of carboplatin and paclitaxel in patients with newly diagnosed stage III or IV epithelial ovarian or primary peritoneal cancer. Patients had a GOG performance status of symptomatic (i.e., in bed <50% of the time or better). Eligibility criteria included non-IDMS serum creatinine values ≤1.5x the institutional upper limit of normal. Low serum creatinine values are common in this patient population, and for patient safety, the protocol assigned a lower limit of serum creatinine (0.6 mg/dL) for use in CrCl estimation. GOG 182 trial accrued patients between January 2001 and September 2004, predating universal adoption of IDMS serum creatinine assay.

The Jelliffe formula was used to estimate CrCl on GOG 182. We re-calculated the CrCl for each patient using the Cockcroft-Gault formula with and without the revised GOG/NCCN-specified limits. The limits used a minimum serum creatinine value of 0.7 mg/dL and an adjusted body weight in patients with a BMI ≥25 kg/m². GOG 182 permitted the use of a prior baseline creatinine value (pre-cancer diagnosis) if determined to be the more medically appropriate value for the patient. Recent surgery with or without aggressive intravenous hydration can lead to erroneously low serum creatinine values, so investigators were advised to consider, if applicable, using a more appropriate (higher) value from the pre-operative period when estimating GFR [14]. We also estimated the GFR using the MDRD formula. Renal function estimates were then used to re-calculate the carboplatin dose for each patient.

Body Mass Index (BMI):

$$BMI=\frac{weigh{t}_{kg}}{heigh{t}_m^2}=\frac{weigh{t}_{lbs}·703}{heigh{t}_{in}^2}$$

Ideal Body Weight (for women) (IBW):

$$\begin{gathered} IB{M}_{female}={45.5}_{kg}+{2.3}_{kg}\left(heigh{t}_{in}-60\right) \\ ={45.5}_{kg}+{2.3}_{kg}\left(\frac{heigh{t}_{cm}}{2.54}-60\right) \end{gathered}$$

Adjusted Body Weight (ABW):

$$AB{W}_{kg}=IB{W}_{kg}+0.4\left(Weigh{t}_{kg}-IB{M}_{kg}\right)$$

Jelliffe formula:

$$CrC{l}_{Jel}=\frac{\left[98-0.8\left(age-20\right)\right]\left({0.90}_{female}\right)}{SCr}$$

Cockcroft-Gault formula:

$$CrC{l}_{CG}=\frac{\left(140-age\right)·Weigh{t}_{kg}·\left({0.85}_{female}\right)}{72·SCr}$$

Modification of Diet in Renal Disease (MDRD) formula:

$$CrC{l}_{mdrd}=175·SC{r}^{-1.154}·Ag{e}^{-0.203}·\left({0.742}_{female}\right)\left(I_{white}+1.212·I_{af.am.}\right)$$

Calvert dose estimate formula:

$$dos{e}_{mg}=AUC·\left(CrCl+25\right)$$

Statistical Design

We considered relationships between Jelliffe-derived baseline CrCl estimates and the incidence of grade 4 or 5 hematologic toxicities, or grade 3, 4, or 5 auditory, genitourinary, fever, metabolic, and neurologic toxicities. National Cancer Institute Common Toxicity Criteria (CTC) version 2.0 was used to grade these adverse events in the GOG 182 protocol [16]. These relationships were quantified using univariable logistic regression models to predict the presence of a grade 4 or 5 toxicity (yes vs no) as a function of the baseline log(CrCl). Receiver operating characteristic (ROC) curves were used to describe the ability of the log(CrCl) predictor to distinguish patients who experienced specific adverse events from those who did not. The area under ROC curve (AUC) is a descriptive measure of how well log(CrCl) distinguishes between patients with a grade 4 or 5 adverse event and those without. AUC values >0.8 are generally considered clinically useful for predicting outcomes in individual patients, while an AUC of 0.5 suggests the predictor has no discrimination ability. Patients with lower CrCl values were expected to have higher adverse event rates. Adverse event information was summarized for each patient by the maximum grade observed within the toxicity group. Sensitivity and positive predictive values describe the classification accuracy of CrCl_Jel <60 mL/min for the specified adverse events. The positive predictive value was defined as the percentage of patients with CrCl_Jel <60 mL/min who had the specific adverse event.

Results

Demographics of the 3830 evaluable patients are shown in Table 1. Patients had a mean age of 58.7 years and a mean BMI of 26.8 kg/m². Carboplatin dose was calculated using the Calvert formula; CrCl was calculated using the Jelliffe formula and non-IDMS serum creatinine assays in the original GOG 182 trial. The mean baseline CrCl_Jel was 81.9 mL/min (range, 23.4-239.5 mL/min). Supplemental Figures 1 and 2 demonstrate the distribution of CrCl by treatment arm in the trial and graphical representation of reported vs. estimated CrCl_Jel. The AUC statistics (range, 0.52-0.64) for grade ≥3 adverse events are reported in Table 2. The log (CrCl_Jel) was not a good predictor of grade ≤3 adverse events (i.e., anemia, leukopenia, thrombocytopenia, febrile neutropenia, auditory, genitourinary/renal, metabolic, neurologic, and peripheral neurologic toxicities). Of the 3,830 patients included in our study, 628 (16%) had CrCl <60 mL/min.

Table 1:

Summary statistics for calculated creatinine clearance and carboplatin dosing estimates.

Demographics
Variable	N = 3,830
Age, Median (Range)	59 (18.2-87.5) years
Race
American Indian/Alaskan Native	9 (<1%)
Asian	67 (1.7%)
Black/African American	166 (4%)
Native Hawaiian/Pacific Islander	6 (<1%)
White	3,503 (92%)
Unknown	79 (2%)
Stage
III	3,218 (84%)
IV	612 (16%)
Body Mass Index, Median (Mean)	25.6 (26.8) kg/m2
Baseline Serum Creatinine, Median (Range)	0.8 (0.3-2.1) mg/dL
CrCl Jelliffe, Median (Range)	81.9 (23.4-239.5) mL/min

Creatinine Clearance Estimates
Variable	N	N Missing	Mean (mL/min)	Median (mL/min)	Minimum (mL/min)	Maximum (mL/min)
CrCl Jell	3,830	52	81.9	79.6	23.4	239.5
CrCl CG1	3,789	93	91.2	85.3	20.9	306.0
CrCl CG2	3,662	220	80.1	76.8	20.9	234.3
CrCl CG3	3,789	93	89.1	84.8	20.9	262.7
CrCl CG4	3,789	93	85.4	81.6	20.9	237.7
CrCl CG5	3,662	220	75.0	74.5	20.9	151.1
CrCl MDRD	3,830	52	82.0	78.4	23.8	287.3
Carboplatin Dose Estimates (AUC=6)
Variable	N	N Missing	Mean (mg)	Median (mg)	Minimum (mg)	Maximum (mg)
Dose Jell	3,830	52	641.5	627.7	290.5	1586.9
Dose CG1	3,789	93	697.1	662.0	275.4	1986.1
Dose CG2	3,662	220	630.6	611.0	275.4	1555.7
Dose CG3	3,789	93	684.9	658.7	275.4	1726.5
Dose CG4	3,789	93	662.4	639.4	275.4	1576.4
Dose CG5	3,662	220	599.7	596.9	275.4	1056.6
Dose MDRD	3,830	52	642.3	620.2	292.9	1873.8

CrCl, creatinine clearance; AUC, area under receiver operator curve; BMI, body mass index; Jell, Jelliffe; CG1, Cockcroft-Gault formula (actual body weight); CG2, Cockcroft-Gault formula (adjusted body weight if BMI ≥25 kg/m²); CG3, Cockcroft-Gault formula (actual body weight, minimum CrCl = 0.6 mg/dL); CG4, Cockcroft-Gault formula (actual body weight, minimum CrCl = 0.7 mg/dL); CG5, Cockcroft-Gault formula (adjusted body weight if BMI ≥25 kg/m², minimum CrCl = 0.7 mg/dL); MDRD, Modification of Diet in Renal Disease

Table 2:

Summary statistics predicting grade ≥3 adverse events for patients with creatinine clearance <60 mL/min.

Adverse Event	Area Under Receiver Operator Curve	Positive Predictive Value (%)	Sensitivity (%)
Hemoglobin	0.52	1.8	15
Platelets	0.54	7.3	13
Granulocytes	0.52	69	18
Auditory	0.54	0.6	29
Genitourinary/Renal	0.56	1.3	19
Infection/Fever	0.53	15	21
Metabolic	0.54	8	21
Neurologic	0.57	6.2	23
Peripheral Neurologic	0.64	8	28

GOG 182 eligibility required serum creatinine values to be ≤1.5x the institutional upper limit of normal at study entry. In that context, supplemental Figures 3–11 illustrate the probability of observing various adverse events as a function of CrCl_Jel among patients with CrCl <60 mL/min and those with CrCl ≤60 mL/min. If leukopenia and auditory adverse events are excluded, the range of positive predictive values for the adverse events examined in patients with CrCl <60 mL/min was 1.8%-15.3%. The positive predictive value of observing a grade 4 or 5 thrombocytopenia was 7.31% (chi-square test of independence, P = .047) (Supplemental Figure S4). The best positive predictive value for observing an adverse event was 68.8% (neutropenia/granulocytopenia) (Supplemental Figure S5). The worst positive predictive value for observing an adverse event was 0.64% (auditory) (Supplemental Figure S6). Each supplementary figure 3–11 also describes the adverse event rates by patients with CrCl <60 mL/min and CrCl ≤60 mL/min.

Table 1 also shows summary statistics for CrCl estimates that were calculated using the 3 formulas, and the resulting carboplatin doses. The Jelliffe and MDRD formulas are presented in comparison to the Cockcroft-Gault formula with various modifications (CG1-CG5). The median CrCl estimates using the Jelliffe, Cockcroft-Gault without modification, and MDRD formulas were 79.6 mL/min, 85.3 mL/min, and 78.4 mL/min, respectively. The median carboplatin dosing estimates using an AUC of 6 for the Jelliffe, Cockcroft-Gault without modification, and MDRD formulas were 628 mg, 662 mg, and 620 mg, respectively.

The median CrCl estimate for the Cockcroft-Gault formula with modification (adjusted body weight if BMI ≤25 kg/m², minimum CrCl = 0.7 mL/min; CG5) was 74.5 mL/min, ranging from 20.9-151.1 mL/min. The median carboplatin dosing estimate using an AUC of 6 for the CG5 formula was 597 mg, ranging from 275 mg to 1057 mg (Table 1).

Figures 1–3 depict the estimated CrCl and resulting carboplatin dosing using the Cockcroft-Gault and MDRD formulas compared to the Jelliffe. The largest percentage change in carboplatin dosing was noted when comparing the Cockcroft-Gault (actual body weight; CG1) formula and the Jelliffe. The CG1 formula resulted in an increased carboplatin dose among 41% of patients, with a mean dose increase of 28% (Figure 1). The smallest percentage increase in carboplatin dosing was noted when comparing the MDRD formula to the Jelliffe (Figure 3). Using the MDRD formula resulted in an increased carboplatin dose in 12% of patients, with a mean increase of 15.4% (Figure 3). Compared to the Jelliffe formula, using the CG5 formula resulted in the largest percentage decrease in dosing, and the MDRD formula resulted in the smallest percentage decrease. Using the CG5 formula resulted in a decrease in carboplatin dosing among 32% of patients, with a mean dose decrease of 19.5% compared to Jelliffe (Figure 2). Using the MDRD formula resulted in a decrease in carboplatin dose in 5.2% of patients, with a mean dose decrease of 11.7% compared to Jelliffe (Figure 3).

Figure 1: Estimated creatinine clearance and carboplatin dose using the Cockcroft-Gault (using actual body weight) vs the Jelliffe formula.

CrCl, creatinine clearance;CG1, Cockcroft-Gault formula (using actual body weight); PATID, patient identification; Pct N, percent N; Std Err, standard error; pctl, percentile; AUC, area under the receiver operating characteristic curve

Figure 3: Estimated creatinine clearance and carboplatin dose using the MDRD vs the Jelliffe formula.

CrCl, creatinine clearance; MDRD, Modification of Diet in Renal Disease formula; PATID, patient identification; Pct N, percent N; Std Err, standard error; pctl, percentile; AUC, area under the receiver operating characteristic curve

Figure 2: Estimated creatinine clearance and carboplatin dose using the Cockcroft-Gault (using adjusted body weight if BMI ≥25 kg/m², minimum CrCl = 0.7 mg/dL) vs the Jelliffe formula.

BMI, body mass index; CrCl, creatinine clearance; CG5, Cockcroft-Gault formula (using adjusted body weight if BMI ≥25 kg/m², minimum CrCl = 0.7 mg/dL); PATID, patient identification; Pct N, percent N; Std Err, standard error; pctl, percentile; AUC, area under the receiver operating characteristic curve

Discussion

Our data demonstrate the variability in estimated CrCl and subsequent carboplatin dosing when using standard CrCl estimating formulas. Consistent with previous literature, we found that CrCl_jel estimates were a poor predictor of adverse effects from carboplatin [9, 11, 17, 18].

Estimation of CrCl

Despite other areas of oncology using the Cockcroft-Gault formula for CrCl estimation and carboplatin dosing, historically, the GOG migrated toward the Jelliffe CrCl estimation because it provided a simple formula, requiring only age and serum creatinine value. The introduction of the IDMS method for serum creatinine assays, in 2009, exacerbated the overestimation of CrCl by the Jelliffe formula [9, 12–14]. In 2011, the GOG/NCCN revised its guidelines for carboplatin dosing to recommend the Cockcroft-Gault formula with modifications – a minimum allowable serum creatinine value of 0.7 mg/dL and the use of an adjusted body weight for patients with a BMI ≤25 kg/m² [9]. NCI/CTEP also capped carboplatin dosing for an AUC of 6, 5, and 4 at 900 mg, 750 mg, and 600 mg, respectively [9, 12, 13].

Our data demonstrate large variability in both CrCl estimation and subsequent carboplatin dosing when using the Jelliffe, Cockcroft-Gault (with and without GOG/NCCN modifications), or MDRD formulas. Prior studies have also demonstrated large variability in estimated CrCl using these standard formulas [11, 17–21]. Specifically, many prior studies of patients with cancer have noted difficulty in using standard formulas for CrCl estimation [22–24]. In their review article, Collins et al discussed the concern for patient toxicity as a possible outcome with changing creatinine values and subsequent variability in carboplatin dosing. They concluded that using radiolabeled nucleotide measurement of GFR is safest for patients whose CrCl estimates are likely to be inaccurate; for example, in patients who are obese, very young or elderly, have low serum creatinine values, or have low muscle mass [19]. Radiolabeled nucleotide testing for GFR, however, is expensive, difficult to perform, exposes patients to radiation, and is rarely used in clinical practice today. As such, newer methodologies, including inulin clearance, serum cystatin C levels, and CamGFR are being investigated [8, 25].

The NKDEP and the National Institute of Diabetes and Digestive and Kidney Diseases released a consensus statement in 2009 recommending the use of estimated GFR and CrCl over gold-standard measurements of GFR for drug dosing [26]. In current gynecologic oncology practice, carboplatin is dosed per the 2011 GOG/NCCN guidelines, which recommend using the Cockcroft-Gault formula with GOG/NCCN-specified modifications to estimate CrCl. With the switch to IDMS creatinine serum values there was concern for overestimation of CrCl and resultant higher carboplatin dosing. Our results demonstrate that using the Cockcroft-Gault formula with appropriate GOG/NCCN modifications in today’s practice, demonstrates a propensity for patients to be prescribed lower doses of carboplatin.

Carboplatin Dosing

Carboplatin dosing is a function of the CrCl estimate. Figure 4 shows the variability in carboplatin dosing resulting from different CrCl estimation methods. Ainsworth et al performed a retrospective analysis of oncology patients with GFR measured via standard chromium-51-ethylenediaminetetraacetic acid (51 Cr-EDTA) testing compared to GFR estimated by the Cockcroft-Gault, Jelliffe, Wright, and MDRD formulas [11]. In their analysis, the Cockcroft-Gault formula produced the smallest bias and highest precision, and was the most accurate for carboplatin dosing [11]. Our study, however, demonstrates the largest decrease in estimated carboplatin dosing using the Cockcroft-Gault formula with GOG/NCCN modifications compared to the Jelliffe formula; we observed a decrease in up to 32% of patients, with a mean decrease in dose of 19.5%.

Figure 4: Flowchart demonstrating variability in factors that contribute to calculating estimated GFR and subsequent carboplatin dosing.

MDRD, = Modification of Diet in Renal Disease formula; GFR, = glomerular filtration rate; 51 Cr-EDTA, = standard chromium 51 ethylenediaminetetraacetic acid testing; AUC, = area under the receiver operator curve

The long-standing effects of variability in carboplatin dosing, especially in terms of adverse events and decreased therapeutic effect, are unknown in ovarian cancer. In a cohort of patients with non-gynecologic malignancies, a post-hoc analysis of a recent trial demonstrated that dose reductions of carboplatin of just 10% could result in increased risk of relapse [27]. Conversely, it is known that increased doses of carboplatin can lead to toxicity and worse quality of life [28]. A prospective study comparing a high carboplatin dosing schedule (AUC12) with the standard schedule (AUC6) in patients with ovarian cancer reported increased toxicity without any improvement in survival outcomes [29] That said, relative dose intensity (RDI) has been linked with survival outcomes in ovarian cancer. [30] These conflicting data, along with our current retrospective analysis, highlight the variability in carboplatin dosing as a result of renal function estimation, and ultimately toxicity and efficacy in patients with ovarian cancer.

Obesity poses another challenge with CrCl estimation and carboplatin dosing [15, 31, 32]. Cockcroft-Gault is the only formula in this study that utilizes weight, or adjusted body weight in obese patients per the GOG and NCCN guidelines [14]. Serum creatinine values, however, are increased in patients with high skeletal muscle mass, which may not be the case in patients with a high BMI due to fatty tissue rather than muscle. Elkhart et al demonstrated that even though using an adjusted body weight in the Cockcroft-Gault formula resulted in the best prediction of carboplatin dosing in overweight and obese patients, they recommend using a flat dose of carboplatin based on the population carboplatin clearance of 140 mL/min [31]. Regardless, the 2011 GOG/NCCN guidelines recommend using an adjusted body weight for patients with a BMI ≤25kg/m² when using the Cockcroft-Gault formula to calculate CrCl and carboplatin dosing.

Predicting Adverse Events

Aside from variability in CrCl estimation and carboplatin dosing, our model shows that estimated CrCl using the Jelliffe formula is a poor predictor of adverse events. The AUC statistics show that CrCl is no better at predicting an adverse event than random chance. In most cases, the positive predictive value estimates suggest patients with CrCl <60 mL/min were not likely to experience the adverse events we considered.

After the IDMS serum creatinine assay became the national standard, Lawson et al investigated the subsequent change in carboplatin dosing (after using the Cockcroft-Gault formula for CrCl estimation) and potential adverse events [10]. They observed no increase in adverse events grade ≤3 thrombocytopenia due to IDMS standardization, even with an observed overall decrease in serum creatinine and subsequent increase in carboplatin dosing [10]. In a cohort of patients with breast cancer receiving non-carboplatin-based therapy, pre-therapy renal function, assessed with the Cockcroft-Gault formula, did not predict patients who would receive dose modifications, complete treatment, or experience hematologic toxicity [33]. Lichtman et al concluded that “patients with varying degrees of renal insufficiency should be able to tolerate standard therapies” [33]. However, predictive models have been validated for grade 3-5 chemotherapy toxicities among patients >65 years of age with cancer [28, 34]. Hurria et al demonstrated that renal dysfunction (as a laboratory-based value) is a risk factor for chemotherapy toxicity, especially for patients with CrCl_Jel <34 mL/min calculated using ideal body weight [28].

Establishing Clinical Trial Eligibility Based on CrCl

While the 2011 GOG/NCCN guidelines and prior studies recommend capping the carboplatin dose for patients with a GFR >125 mL/min, few studies have assessed the impact of a lower limit for CrCl [14, 35]. Our results suggest that if CrCl <60 mL/min was used as an eligibility criterion for clinical trials, approximately 16% of incoming patients would be excluded, most of whom would not experience the toxicities considered here.

We recognize that CrCl eligibility criteria has shifted away from a lower limit of 60 mL/min. Recently, ASCO and NCI updated their recommendations for clinical trial inclusion and exclusion criteria [36]. According to their criteria, patients with CrCl or GFR ≥30 mL/min (using the Cockcroft-Gault or MDRD formulas or as reported in the comprehensive metabolic panel/basic metabolic panel [eGFR]), or patients with a serum creatinine value ≤1.5x the institutional upper limit of normal, can be included in clinical trials [36]. These guidelines recommend using calculated CrCl rather than serum creatinine, given that CrCl reflects changes in drug clearance, whereas serum creatinine alone does not.

Consistent with Hurria et al.’s findings, ASCO and NCI’s recommendations support increased risk of toxicity for patients with laboratory-based renal dysfunction (CrCl <30-40 mL/min) [28], indicating CrCl <30mL/min as a strict exclusion criterion from clinical trials [36]. However, among older patients with cancer, the expected organ dysfunction (including renal dysfunction) would exclude a large proportion of patients from clinical trials simply based on laboratory test results [37]. Over the past decade, national cancer organizations have expanded laboratory-based clinical trial eligibility criteria to increase patient accrual and diversity, and to ensure generalizability of clinical trial results [38–40]. Similar to the recommendations made by Spira et al, our data do not support excluding patients from trials based on laboratory values of renal function, specifically for an estimated CrCl <60 mL/min.

Strengths and Limitations

This study is supported by the large cohort of patients enrolled on the GOG 182 clinical trial [16]. The available clinical and adverse event data enabled us to estimate CrCl through various standard formulas, calculate subsequent carboplatin dosing, and explore relationships with the incidence of adverse events. This is the largest study of its kind to assess the impact of CrCl estimation on carboplatin dosing in a large cohort of patients with ovarian cancer. The study is limited by the use of a female-only sample of patients in the postoperative setting. GOG 182 was conducted prior to 2009, therefore, the creatinine values used to estimate the CrCl were all non-IDMS. Furthermore, it is a retrospective review and carboplatin dosing was given per the GOG 182 protocol methods, whereby the platinum-containing doublet or triplet regimen varied according to the study arm. The non-platinum agents could have contributed to the toxicity observed in this analysis, and we are unable to statistically adjust for patient treatment arm with relation to toxicity outcomes. Changing from the Jelliffe to the Cockcroft-Gault formula to estimate CrCl impacted carboplatin dosing, but the impact of this change is unknown. Carboplatin dose adjustments or changes in chemotherapy were not assessed. In addition, we cannot ascertain how toxicity might have differed if another formula was used to calculate CrCl in this cohort. Lastly, even with the large number of patients included, the number of patients with CrCl <60mL/min was small. We do not have granular data to evaluate the potential impact of baseline co-morbidities on individual patient CrCl values. Carboplatin dosing in the trial took into consideration the lower estimated CrCl for this subgroup. This limits our ability to assess the true safety of including patients with CrCl <60mL/min in clinical trials.

In conclusion, CrCl <60 mL/min was not an accurate predictor of adverse events. Using this criterion would exclude 16% of patients from clinical trials, though our study shows they would have received therapy without undue toxicity. Our findings are consistent with recent eligibility criteria for clinical trials recommending broader laboratory-based renal function exclusion criteria. Future work is necessary to elucidate the optimal method to estimate and measure renal function in patients with cancer to best achieve target carboplatin AUC dosing, while avoiding toxicity. An ongoing NRG trial (NRG-GY022) is the first of its kind to prospectively evaluate different CrCl estimation formulas, utilize Iohexol to measure renal function and true renal clearance at the time of carboplatin treatment, and follow for adverse events. This study should help clarify the optimal formulas for estimating true CrCl and provide insight into appropriate methods for carboplatin dosing that reach therapeutic goals. Moreover, while this study highlights the safety of including patients with low CrCl (<60mL/min) in clinical trials, future endeavors should assess the safety and necessity to cap carboplatin dosing for patients with high CrCl estimations. While significant progress has been made to standardize CrCl estimation for chemotherapeutic dosing, prospective work is needed to ensure appropriate dosing of chemotherapeutic agents and the inclusion of patients with gynecologic malignancies in future clinical trials.

Highlights:

• The formula used to estimate CrCl affects carboplatin dosing.

• Estimated creatinine clearance (CrCl) <60 mL/min (by Jelliffe) did not accurately predict adverse events.

• Our data do not support a minimum threshold CrCl <60 mL/min as exclusion criteria from clinical trials.