Assessment of the Effects of Fruquintinib on Cardiac Safety in Patients with Metastatic Colorectal Cancer

Abstract

Fruquintinib is a highly selective oral inhibitor of vascular endothelial growth factor receptors 1, 2, and 3, approved by the US Food and Drug Administration and the European Commission for the treatment of previously treated metastatic colorectal cancer (mCRC), regardless of biomarker status. This study used concentration QT interval (C‐QTc) modeling, utilizing data from the phase 3 FRESCO‐2 study (NCT04322539) in which patients with mCRC received fruquintinib 5 mg, once daily, or matching placebo in a 28‐day cycle, to evaluate the potential of fruquintinib to delay cardiac repolarization. The primary objectives were to assess the relationship between change from baseline in the QTc and plasma concentrations of fruquintinib and its metabolite M11, and to predict placebo‐corrected change from baseline in the corrected QT interval (ΔΔQTc) associated with clinically relevant fruquintinib or M11 concentrations. The C‐QTc analysis was conducted using 1456 time‐matched concentration‐change from baseline in the population‐based corrected QT interval (ΔQTcP) pairs from 205 patients (fruquintinib n = 137; placebo n = 68). The final C‐QTc model was a linear mixed‐effects model with the effect of M11 concentration on ΔQTcP. This model estimated that the upper bounds of the 90% CI of the mean ΔΔQTcP at steady‐state geometric mean (GM) M11 C_max, and twice the GM M11 C_max were 0.0537 and 4.00 ms, respectively. Additional C‐QTc analysis, including only fruquintinib concentrations, showed no relationship between ΔQTcP and fruquintinib concentrations. The analysis indicated that fruquintinib administered at the approved clinical dose is not anticipated to cause clinically meaningful QT prolongation.

Introduction

Fruquintinib is a highly selective, oral tyrosine kinase inhibitor of vascular endothelial growth factor receptors (VEGFRs ‐1, ‐2, and ‐3), which inhibits angiogenesis, thereby restricting tumor growth and progression. ¹ In preclinical and phase 1 studies, fruquintinib was shown to have enhanced target selectivity and limited off‐target kinase activity, with weak to no inhibitory effects on all other kinases. ¹ , ² M11 is a major metabolite of fruquintinib which is less effective in inhibiting VEGFR2 kinase activity and VEGFR‐induced phosphorylation. ³ The half‐life of fruquintinib is approximately 42 h. ²

The efficacy of fruquintinib was demonstrated in FRESCO (NCT02314819), a phase 3, randomized, double‐blind, placebo‐controlled trial conducted in Chinese patients with metastatic colorectal cancer (mCRC) who received fruquintinib 5 mg or a matching placebo orally, once per day (QD), 3 weeks on, 1 week off over a 28‐day cycle plus best supportive care (BSC). ⁴ Compared with placebo, overall survival (OS) was significantly improved with fruquintinib (median 9.3 versus 6.6 months; hazard ratio [HR]: 0.65; P < .001), as was progression‐free survival (PFS; median 3.7 versus 1.8 months; HR: 0.26; P < .001). ⁴ In September 2018, fruquintinib was approved in China as third or later line of therapy for refractory mCRC. ⁵ In the global, phase 3 FRESCO‐2 study, which was conducted across 14 countries in North America, Europe, Asia, and Australia, patients with refractory mCRC were randomized 2:1 to receive fruquintinib 5 mg or a matching placebo QD, 3 weeks on, 1 week off over a 28‐day cycle plus BSC. ⁶ Fruquintinib plus BSC showed statistically significant and clinically meaningful improvements in OS (median 7.4 versus 4.8 months; HR: 0.66; P < .001) and PFS (median 3.7 versus 1.8 months; HR: 0.32; P < .001) compared with placebo plus BSC, ⁶ with a manageable safety profile and without deterioration in quality of life. ⁷ Based on the results of FRESCO and FRESCO‐2, fruquintinib was approved in November 2023 by the United States Food and Drug Administration for the treatment of adult patients with mCRC who have been previously treated with fluoropyrimidine‐, oxaliplatin‐, and irinotecan‐based chemotherapy, an anti‐VEGF therapy, and, if RAS wild‐type and medically appropriate, an anti‐epidermal growth factor receptor (EGFR) therapy. ⁸ , ⁹ Based on the results of FRESCO‐2, fruquintinib was subsequently approved in Europe for the treatment of adults patients with mCRC who have been previously treated with available standard therapies, including fluoropyrimidine‐, oxaliplatin‐, and irinotecan‐based chemotherapies, anti‐VEGF agents, and anti‐EGFR agents, and who experienced disease progression on, or are intolerant to, treatment with either trifluridine‐tipiracil or regorafenib. ⁶

The clinical development of new drugs with systemic bioavailability requires robust characterization of the potential effect of the drug to delay cardiac repolarization, as measured by prolongation of the corrected QT interval (QTc) on the surface electrocardiogram (ECG). ¹⁰ Concentration‐QTc (C‐QTc) modeling of QTc data can be applied to estimate a drug's effect on cardiac repolarization. ¹¹ In this analysis, a subset of patients enrolled in FRESCO‐2 underwent a cardiovascular safety analysis with centrally‐read, triplicate ECG assessment and time‐matched pharmacokinetic (PK) sampling to evaluate the effect of fruquintinib versus placebo on cardiac electrophysiology at the clinical dose of 5 mg once daily (QD) for 21 days of a 28‐day cycle. The maximum tolerated dose of fruquintinib was 6 mg QD for 21 days of a 28‐day cycle, based on data from six patients with advanced solid tumors. ² One patient experienced a dose‐limiting toxicity (DLT) at this dose, while no DLTs were observed at the 5 mg dose level. As a result, supratherapeutic doses could not be given. Accordingly, a positive control (e.g., moxifloxacin) was not included in the assessment, which is consistent with approaches used in evaluating effects of anti‐cancer agents on QTc. ¹² The primary objective was to characterize the relationship between the change from baseline in the corrected QT interval (ΔQTc) and plasma concentrations of fruquintinib and its major metabolite M11 in patients with refractory mCRC, and to predict the placebo‐corrected change from baseline in the corrected QT interval (ΔΔQTc) associated with clinically relevant clinical fruquintinib or M11 concentrations. Other objectives were to perform categorical and by‐time point analyses of ECG parameters.

Methods

Overview of Data

This cardiac safety analysis was based on data from a subset of patients with mCRC enrolled in FRESCO‐2 who received fruquintinib 5 mg or matching placebo orally QD, on days 1‐21 in 28‐day cycles plus BSC and underwent 12‐lead ECGs extracted in triplicate from continuous (Holter) recordings, hereafter referred to as the ECG subset (n = 137 fruquintinib; n = 68 placebo). The study excluded patients with a QTcF >480 ms, those with any other factors that could prolong the QTc interval or increase the risk of arrhythmic events, those receiving concomitant QTc prolonging medication, and prohibited the use of concomitant medications known to be associated with QTc prolongation. Plasma concentrations of fruquintinib and M11 were collected at the same nominal time as ECG measurements from patients. All patients provided written informed consent. The study protocol was approved by review boards at each center, and the clinical trials were conducted in accordance with applicable regulatory standards and the International Conference on Harmonization Guideline for Good Clinical Practice. ¹³

For variables that involved baseline ECG values (i.e., change from baseline parameters, ΔΔQTc, and categorical analysis for all variables except QTc), the analysis population for the central tendency, by‐time point, and categorical analyses included all patients in the ECG subset who received at least one dose of fruquintinib or placebo with a baseline ECG measurement and at least one postbaseline ECG measurement. All ECG data for these patients were included. All patients who received at least one dose of fruquintinib or placebo and had at least one postbaseline ECG measurement were included in the categorical analysis for QTc; a baseline QTc measurement was not required.

The analysis population for the C‐QTc analysis included all patients who received at least one dose of fruquintinib or placebo with at least one valid nominal time‐matched postdose fruquintinib concentration‐ΔQTc or M11 concentration‐ΔQTc pair, as applicable. All ECG data from patients receiving placebo were included in the C‐QTc analysis if there was at least one valid ΔQTc value, and these patients were assumed to have a concentration (fruquintinib or M11) of 0 at each nominal time point.

Cardiac Safety Analyses

The methods described by Garnett et al ¹¹ and the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use E14 were used to conduct the analyses. ¹⁰ , ¹⁴ , ¹⁵ The C‐QTc analysis and creation of graphs and summary statistics were performed using R version 4.2.0. Modeling was performed using the nlme package (version 3.1‐157) in R. ¹⁶

The primary objective of this analysis was to characterize the relationship between ΔQTc and plasma concentrations of fruquintinib and M11 in patients with refractory mCRC, and to predict ΔΔQTc associated with clinically relevant fruquintinib or M11 concentrations. The other objectives were to perform categorical and by‐time point analyses of ECG parameters.

QTcF and QTcP

12‐lead Holter ECG data were captured in triplicate and processed centrally by eResearch Technology, Inc. As the QT interval is negatively correlated to heart rate, the measured QT interval was corrected for heart rate using Fridericia's formula for each replicate, according to the following equation:

$$\mathrm{QTcF}=\frac{\mathrm{QT}}{\sqrt[3]{\mathrm{RR}}}$$

where QTcF is the corrected QT interval using Fridericia's formula and RR is in seconds.

Based on a graphical assessment of the QTcF versus RR relationship (Figure S1), QTcF did not provide adequate heart rate correction. Therefore, change from baseline in the population‐based corrected QT interval (ΔQTcP) was derived using baseline (cycle 1 day 1 predose) individual replicates of the QT and RR intervals and using a logarithmic‐logarithmic model. The following equation was applied to the baseline data of all patients:

$$\begin{array}{c}\hfill \ln (\mathrm{QT})=\alpha +\beta \times \ln (\mathrm{RR})\end{array}$$

where RR is in milliseconds. No random effects were included in the model. The estimate of β was then used to derive the population‐based corrected QT interval (QTcP) for all data (baseline and postbaseline) as follows:

$$\mathrm{QTcP}=\frac{\mathrm{QT}}{\mathrm{R}{\mathrm{R}}^{\beta }}$$

where RR is in milliseconds.

QTcP was used for the main analysis, and a supportive analysis was carried out with QTcF.

Change from baseline in QTc and ΔΔQTc was defined as follows:

$$\mathrm{\Delta }\mathrm{QTcind}(\mathrm{t})=\mathrm{QTcind}(\mathrm{t})-\mathrm{QTcind},\mathrm{baseline}$$ (1)

where QTc_ind(t) is the individual QTc at time t, QTc_ind, baseline is the individual baseline QTc (defined as the individual predose measurement), and ΔQTc_ind(t) is the individual ΔQTc at time t.

$$\begin{array}{ccc}\hfill \mathrm{\Delta }\mathrm{\Delta }\mathrm{QTcind}(\mathrm{t})& =& \mathrm{\Delta }\mathrm{QTcind}(\mathrm{t})\hfill \\ & & -\mathrm{mean}(\mathrm{\Delta }\mathrm{QTcplacebo}(\mathrm{t}))\hfill \end{array}$$ (2)

where ΔQTc_ind(t) is the individual ΔQTc at time t, mean (ΔQTc_placebo(t)) is the mean ΔQTc for placebo at time t, and ΔΔQTc_ind(t) is the individual ΔΔQTc at time t.

The ΔQTc was matched to the plasma fruquintinib and/or M11 concentration collected at the same nominal time point. The ΔΔQTc calculated as described above was not used for modeling but only for graphical presentation in exploratory plots.

Heart rate, PR interval, and QRS interval parameters, along with the corresponding changes from baseline, were calculated for summarization in the categorical analysis.

C‐QTc Model

C‐QTc modeling was performed using the methods described by Darpo et al ¹⁷ and Garnett et al ¹¹ The primary analysis was based on a linear mixed‐effects model, with ΔQTc as the dependent variable. The continuous covariates in the initial model structures included plasma concentrations of fruquintinib, M11, or both and baseline QTc. The patient was included as an additive random effect on intercept and slope. The initial model was described as follows:

$$\begin{array}{ccc}\hfill \mathrm{\Delta }\mathrm{Q}{\mathrm{Tc}}_{\mathrm{ijk}}& =& ({\theta }_0+{\eta }_{0,\mathrm{i}})+{\theta }_1{\mathrm{TRT}}_{\mathrm{i}}\hfill \\ & & +({\theta }_2+{\eta }_{2,\mathrm{i}}){\mathrm{C}}_{\mathrm{ijk}}+{\theta }_3{\mathrm{Time}}_{\mathrm{j}}+{\theta }_4{\mathrm{Visit}}_{\mathrm{k}}\hfill \\ & & +{\theta }_5\left({\mathrm{QTc}}_{0,\mathrm{i}}-\overline{{\mathrm{QTc}}_0}\right)+{\epsilon }_{\mathrm{ijk}}\hfill \end{array}$$

where ΔQTc_ijk is the QTc of patient i at nominal time j at visit k (cycle 1 day 1 or 21), ${\theta }_0$ is the population mean intercept, η_0,i is the random effect associated with the intercept term, ${\theta }_1$ is the fixed effect associated with treatment TRT_i received by patient i (TRT_i = 0 for placebo and 1 for active drug; the θ ₁ term has no physiological interpretation but allows flexibility in the event of model specification), ¹⁸ ${\theta }_2$ is the population mean slope of the assumed linear association between concentration and ΔQTc_ijk, η_2,i is the random effect associated with the slope term, C_ijk is the plasma concentration of fruquintinib or M11 for patient i at nominal time j at visit k, ${\theta }_3$ is the fixed effect associated with nominal time (one value estimated nonreference per nominal time point), ${\theta }_4$ is the fixed effect associated with cycle 1 day 21 (taking cycle 1 day 1 as the reference visit), ${\theta }_5$ is the fixed effect associated with baseline QTc, QTc_0,i is the baseline QTc of patient i, $\overline{\mathrm{QT}{\mathrm{c}}_0}$ is the overall mean baseline QTc, and ε_ijk is the residual error for patient i at nominal time j at visit k. Random effects were assumed to have a normal distribution with a mean of 0 and an unstructured covariance matrix. Residuals were assumed to be normally distributed with a mean of 0 and a variance R.

C‐QTc Model Selection and Evaluation

The final model was selected based on diagnostic plots, statistical estimators of goodness of fit, and clinical relevance. The appropriateness of the model was assessed using diagnostic plots of observed ΔQTc versus model‐predicted ΔQTc; quantile–quantile plots of standardized residuals; and plots of standardized residuals versus fitted ΔQTcP, fruquintinib and/or M11 concentration, nominal time, visit (cycle 1 days 1 and 21), and baseline QTc. The model was selected based on the Akaike information criterion (AIC) value, with a lower value signifying an improved model fit. The estimated parameter values and the associated 95% confidence intervals (CIs) and relative standard errors were also inspected.

A plot of the observed mean (90% CI) ΔQTcP versus quantiles of observed M11 concentrations overlaid with the model predictions (mean and 90% CI) was generated to assess the final model. The upper bound of the two‐sided 90% CI for the mean ΔQTcP predicted at the geometric mean (GM) maximum plasma concentration (C_max) at clinically relevant doses was compared to the thresholds of regulatory concern of 10 and 20 ms (commonly used for oncology indications). Values <20 ms would indicate a lack of clinically significant QT prolongation.

Based on the final model, the mean ΔΔQTc and the associated two‐sided 90% CI over the clinical concentration range were predicted for the average C_max and twice the average C_max for relevant dosing regimens. The upper bounds of the two‐sided 90% CI for the mean ΔΔQTc predictions were compared to 10 and 20 ms.

The mean ΔΔQTc was calculated as follows, based on the final C‐QTc model:

$$\begin{array}{ccc}& & \mathrm{Mean}\mathrm{\Delta }\mathrm{\Delta }\mathrm{QTc}=\mathrm{Mean}(\mathrm{\Delta }\mathrm{QTcij}|\mathrm{TRTi}=1,\mathrm{Cij}=\mathrm{C})\hfill \\ & & -\mathrm{Mean}(\mathrm{\Delta }\mathrm{QTcij}|\mathrm{TRTi}=0,\mathrm{Cij}=0)\hfill \end{array}$$

where Cij is the plasma concentration of fruquintinib or M11 for patient I at time nominal time j.

The mean ΔΔQTc (90% CI) was derived using the contrast R package (version 0.24.2).

Categorical, by‐Time Point, and Central Tendency Analyses

Categorical summaries were performed to determine the number and percentage of patients by treatment who met the criteria in Table 1. For the by‐time point analysis, the mean ΔQTc (for the selected QT correction variable) and ΔΔQTc were calculated at each time point, and the upper bounds of two‐sided 90% CI of ΔΔQTc on cycle 1 day 21 were compared to clinically relevant thresholds (10 and 20 ms). This was done using the intersection union test (using the methodology described by Hutmacher et al ¹⁹ ). For the central tendency analysis, mean ΔQTc and ΔΔQTc at each time point were summarized (mean, standard deviation, coefficient of variation percentage, median and range, and two‐sided 90% CI).

Table 1.

Categorical Analysis: QTc and ΔQTc

Categories: QTc	Placebo (N = 80)	5 mg – All Data (N = 163)	5 mg – Postbaseline (N = 162)
QTcP >450 ms	16 (20.0%)	24 (14.72%)	24 (14.81%)
QTcP >450 and ≤480 ms	15 (18.8%)	20 (12.3%)	20 (12.3%)
QTcP >480 ms	1 (1.2%)	4 (2.5%)	4 (2.5%)
QTcP >480 and ≤500 ms	1 (1.2%)	3 (1.8%)	3 (1.9%)
QTcP >500 ms	0 (0%)	1 (0.6%)	1 (0.6%)
QTcF >450 ms	4 (5%)	18 (11.04%)	17 (10.49%)
QTcF >450 and ≤480 ms	3 (3.8%)	14 (8.6%)	13 (8%)
QTcF >480 ms	1 (1.2%)	4 (2.5%)	4 (2.5%)
QTcF >480 and ≤500 ms	1 (1.2%)	3 (1.8%)	3 (1.9%)
QTcF >500 ms	0 (0%)	1 (0.6%)	1 (0.6%)

Categories: ΔQTc	Placebo (N = 68)	5 mg—All Data (N = 141)	5 mg—Postbaseline (N = 141)
ΔQTcP >30 ms	4 (5.9%)	8 (5.7%)	8 (5.7%)
ΔQTcP >30 and ≤60 ms	4 (5.9%)	8 (5.7%)	8 (5.7%)
ΔQTcP >60 ms	0 (0%)	0 (0%)	0 (0%)
ΔQTcF >30 ms	4 (5.9%)	11 (7.8%)	11 (7.8%)
ΔQTcF >30 and ≤60 ms	4 (5.9%)	11 (7.8%)	11 (7.8%)
ΔQTcF >60 ms	0 (0%)	0 (0%)	0 (0%)

ΔQTc, change from baseline in corrected QT interval; ΔQTcF, change from baseline in the corrected QT interval using Fridericia's formula; ΔQTcP, change from baseline in the population‐based corrected QT interval; N, number of patients; QTc, corrected QT interval; QTcF, corrected QT interval using Fridericia's formula; QTcP, population‐based corrected QT interval.

Note: Results are presented as the number (%) of patients in each category who met the criteria for 1 or more data points. Only patients with at least 1 non‐missing QTcP or QTcF value were included in this table.

Results

Summary of Data Used for Analysis

The C‐QTc analysis dataset contained 1954 sets of Holter ECG measurements from 243 patients (n = 163 and n = 80 in the fruquintinib and placebo arms, respectively). Exclusions were made due to missing baseline QTc, ECG measurements without time‐matched concentrations, and ECG and PK sampling performed >30 min apart. Therefore, the C‐QTc analysis was based on 1456 time‐matched concentration‐ΔQTcP pairs from 205 patients: 137 in the fruquintinib arm and 68 in the placebo arm. Of the 936 fruquintinib‐M11 metabolite concentration pairs in the fruquintinib arm of the analysis dataset, 16 (1.7%) fruquintinib concentrations and 326 (34.8%) M11 concentrations were below the limit of quantification (BLQ). Concentrations BLQ were set to zero and were included in the analysis. The mean baseline QTcP and QTcF in the analysis dataset were 419.3 and 409.5 ms, respectively.

C‐QTcP Model and Evaluation

Three sets of models were estimated as follows: ΔQTc versus fruquintinib concentrations (Table S1), ΔQTc versus M11 concentrations, and ΔQTc versus fruquintinib and M11 concentrations (Table S2).

The final C‐QTc model selected was a linear mixed‐effects model with the effect of M11 concentration on ΔQTcP, which had the lowest AIC of 10727.07 (Table S3). This model included intercept, treatment (fruquintinib versus placebo), nominal time, visit, and baseline QTcP terms, with independent between‐subject variability on the intercept and ΔQTcP‐M11 concentration slope. Parameter estimates for the final model are presented in Table 2. A statistically significant slope of 0.0339 (95% CI 0.00516, 0.0625) ms per ng/mL (P = .0212) was estimated. The non‐significant coefficients and the visit term were kept in the model, which aligns with recommendations from the scientific white paper on concentration‐QTc modeling. ¹¹ The nominal time coefficients reflect the drug‐independent diurnal variation of ΔQTc over time. The treatment‐specific intercept term θ₁ is an empirical term allowing the linear mixed effects model to be flexible, for example, in cases where ΔQTc is nonlinear. ¹⁸

Table 2.

Final C‐QTc Model Parameter Estimates—QTcP Model with M11 Concentration

Coefficient	Units	Estimate	SE	RSE%	95 CI%	P‐value
Intercept	ms	2.82	1.35	47.9	0.176, 5.47	0.0372
Treatment	ms	−5.19	1.36	26.2	−7.86, −2.52	0.0002
M11 conc. slope	ms per ng/mL	0.0339	0.0147	43.4	0.00516, 0.0625	0.0212
NTime = 1	ms	−0.25	0.867	347	−1.95, 1.45	0.7733
NTime = 2	ms	1.78	0.867	48.7	0.0873, 3.48	0.0399
NTime = 3	ms	1.84	0.868	47.2	0.144, 3.54	0.0340
NTime = 4	ms	0.570	0.874	153	−1.14, 2.28	0.5150
Cycle 1 day 21 visit	ms	−1.39	0.676	48.6	−2.71, −0.0638	0.0406
Baseline QTcP	‐	−0.102	0.0306	30.0	−0.162, −0.0415	0.0011
BSV SD for intercept	ms	8.23	0.486	5.91	7.33, 9.24	‐
BSV SD of the M11 conc. slope	ms per ng/mL	0.0953	0.0118	12.4	0.0751, 0.121	‐
Residual error SD	ms	7.93	0.165	2.08	7.61, 8.26	‐

BSV, between‐subject variability; C‐QTc, concentration‐corrected QT interval; CI, confidence interval; Conc., concentrations; M11, major metabolite of fruquintinib; NTime, nominal time in hours; QTcP, population‐based corrected QT interval; RSE, relative SE; SD, standard deviation; SE, standard error.

All parameters were statistically significant, apart from the coefficients for the dummy variables for the 1‐ and 4‐h post‐dose nominal time points. The standard deviations of the between‐subject variability (BSV) on the intercept and the M11 concentration slope were 8.02 ms and 0.116 ms per ng/mL. Figure 1 depicts the model‐predicted relationship between ΔQTcP and M11 concentration overlaid on the observed data. Observed ΔQTcP and ΔQTcF versus fruquintinib concentrations and versus M11 concentrations are shown in Figure S2.

Figure 1.

Observed and model‐predicted ΔQTcP versus M11 concentration. Note: Circles represent individual observations on cycle 1 day 1 and cycle 1 day 21. The blue line represents the model prediction. The shaded region represents the 90% CI of the model prediction, assuming a nominal time of cycle 1 day 21 predose and a baseline QTcP equal to the mean value in the analysis dataset. ΔQTcP, change from baseline in the population‐based corrected QT interval; CI, confidence interval; M11, major metabolite of fruquintinib; QTcP, population‐based corrected QT interval.

Goodness‐of‐fit plots for the final C‐QTc model are presented in Figure S3, indicating an adequate model fit with no bias.

Final Model Predictions

The predicted mean (90% CI) placebo‐corrected change from baseline in the population‐based corrected QT interval (ΔΔQTcP) values at the observed GM C_max of M11 at steady state following fruquintinib 5 mg QD (77 ng/mL) and twice the GM C_max (154 ng/mL) are presented in Table 3. The corresponding upper bounds of the 90% CI of the predicted mean ΔΔQTcP were at steady‐state GM M11 C_max and twice GM M11 C_max 0.0537 and 4.00 ms, respectively. The model predicts that the upper bound of the 90% CI of mean ΔΔQTcP will exceed 10 ms at an M11 concentration of 262 ng/mL, 3.4‐fold higher than the observed steady‐state GM C_max of M11.

Table 3.

Predictions of Mean (90% CI) ΔΔQTcP Based on the Final Model with M11 Concentration

Scenario	M11 Concentration (ng/mL)	Mean ΔΔQTcP (ms)	90% CI of Mean ΔΔQTcP (ms)
GM M11 Cmax at steady state	77	−2.58	−5.22, 0.0537
2×GM M11 Cmax at steady state	154	0.0251	−3.95, 4.00

ΔΔQTcP, placebo‐corrected change from baseline in the population‐based corrected QT interval; CI, confidence interval; C_max, maximum plasma concentration; GM, geometric mean; M11, major metabolite of fruquintinib.

Supportive Analyses

The development trajectory for the C‐QTcF model was similar to that of ΔQTcP, and the selected model had only M11 concentrations with the same structure as that of the model developed using ΔQTcP. Parameter estimates for the final model with ΔQTcF are presented in Table S4. Predictions of placebo‐corrected change from baseline in the corrected QT interval using Fridericia's formula (ΔΔQTcF) based on this model are presented in Table 4. The predicted upper bounds of the 90% CI of the mean ΔΔQTcF for the GM M11 C_max and twice the GM M11 C_max were 3.07 and 8.34 ms, respectively, and was predicted not to exceed 10 ms at M11 concentrations up to 177 ng/mL, 2.3‐fold higher than the observed GM M11 C_max at steady state. Additional C‐QTc analysis including only fruquintinib concentrations showed no relationship between ΔQTcP and fruquintinib concentrations with a slope (95% CI) of 0.00778 (−0.00247, 0.018) ms/ng/mL (P = .1377) (Table S5).

Table 4.

Predictions of Mean (90% CI) ΔΔQTcF Based on the Final Model with M11 Concentration

Scenario	M11 Concentration (ng/mL)	Mean ΔΔQTcF (ms)	90% CI of Mean ΔΔQTcF (ms)
GM M11 Cmax at steady state	77	0.331	−2.41, 3.07
2×GM M11 Cmax at steady state	154	4.00	−0.341, 8.34

ΔΔQTcF, placebo‐corrected change from baseline in the corrected QT interval using Fridericia's formula; CI, confidence interval; C_max, maximum plasma concentration; GM, geometric mean; M11, major metabolite of fruquintinib.

Categorical and by‐Time point, and Central Tendency Analyses

The categorical analysis did not show any meaningful effect of fruquintinib administration on QTc measurements as the proportion of patients with outlier QTc values was similar for placebo and fruquintinib treatment arms (Table 1). In addition, drug‐induced mean changes in heart rate greater than 10 beats per minute (bpm) are considered to be significant. ¹¹ The absolute value of mean ΔHR were <10 bpm at all time points included in the analysis, indicating that there was no significant heart rate effect (Figure S4). The largest mean ΔPR and ΔQRS was 5.74 and 1.67 ms, respectively, indicating no clinically meaningful effect of fruquintinib on the PR or QRS intervals. The by‐time point analysis also reflected a lack of clinically significant QT prolongation after fruquintinib administration (Table 5, Figure S5).

Table 5.

By‐Time Point Analysis Results for Fruquintinib on Cycle 1 Day 21 ‐ QTcP

C1D21 Time Point (h)	Fruquintinib QTcP LS Mean (ms)	Placebo QTcP LS Mean (ms)	Fruq. LS Mean ΔQTcP (ms)	Placebo LS Mean ΔQTcP (ms)	LS Mean ΔΔQTcP (ms)	90% CI for LS Mean ΔΔQTcP (ms)
0	419.41	423.72	−1.4575	2.8503	−4.3078	−8.598, −0.018
1	421.04	421.74	0.171	0.8741	−0.703	−4.935, 3.529
2	422.49	426.19	1.6199	5.3224	−3.7025	−7.937, 0.532
3	422.24	423.85	1.3777	2.9891	−1.6114	−5.846, 2.623
4	421.19	423.88	0.3231	3.0151	−2.6921	−6.967, 1.583

ΔΔQTcP, placebo‐corrected change from baseline in the population‐based corrected QT interval; ΔQTcP, change from baseline in the population‐based corrected QT interval; C1D21, cycle 1 day 21; CI, confidence interval; LS, least‐squares; QTcP, population‐based corrected QT interval.

Summary statistics for ΔQTcP and ΔΔQTcP by treatment, visit, and nominal time are presented in Tables S6 and S7, respectively.

Discussion

Nonclinical studies evaluated the potential of fruquintinib to inhibit human ether‐à‐go‐go‐related gene (hERG) tail current. The half‐maximal inhibitory concentration (IC50) value was >13.08 µM (>5146 ng/mL). This is >378‐fold greater than the unbound fruquintinib steady‐state peak concentration in patients at the approved clinical dose of 5 mg QD 3/1 in a 28‐day cycle (total concentration, approximately 290 ng/mL; unbound concentration based on protein binding of 95.3%, ²⁰ 13.6 ng/mL). This suggests no significant inhibitory effect of fruquintinib on the hERG potassium channel at clinically relevant exposures. Similarly, the fruquintinib major metabolite M11 was also evaluated in the hERG assay. M11 did not effectively inhibit hERG channels, resulting in an IC50 value of >6.05 µM. An IC50 value of >6.05 µM (>2295 ng/mL) is >1000‐fold greater than the unbound human peak concentration for M11 at 5 mg QD of fruquintinib (total M11 concentration of 77 ng/mL; unbound M11 concentration based on protein binding of 97.7%, 1.77 ng/mL). In addition, in vivo safety pharmacology evaluations revealed no adverse effects on the cardiovascular system, as evidenced by the results of a single dose telemetry study in repeat‐dose toxicity studies in dogs. The free C_max of fruquintinib (12 ng/mL) at the highest dose of 0.12 mg/kg/day in dogs was similar to the clinical free C_max of fruquintinib following multiple dose administration at 5 mg (13 ng/mL), further supporting the low potential of fruquintinib‐related effects on the cardiovascular, respiratory, and central nervous systems. The absence of findings on the cardiovascular system in vivo was consistent with in vitro hERG assays, indicating a low risk for QTc prolongation.

The effect of fruquintinib on cardiac safety was assessed based on triplicate 12‐lead Holter ECG measurements collected following a single dose of fruquintinb (cycle 1 day 1) and at steady‐state (cycle 1 day 21) in a subset of patients in FRESCO‐2. QTcP provided better heart rate correction than QTcF with a flatter QTcP‐RR slope (Figure S1). The correlation between baseline QTcP and RR intervals was not statistically significant, with a small QTcP‐RR slope of −3.73 × 10⁻⁵ and a 90% CI (−0.0102, 0.0101) that included 0. The baseline QTcF‐RR slope was statistically significant and larger in magnitude, with a slope (90% CI) of 0.0493 (0.0393, 0.0592). In addition, the range of heart rates with or without fruquintinib treatment was similar, supporting that QTcP provided an appropriate correction. ¹¹ For this reason, QTcP was selected as the primary correction method. ΔQTcP was used for the primary analysis, and C‐QTc modeling with ΔQTcF was performed as part of the supportive analyses to assess the cardiac electrophysiology of fruquintinib at the recommended clinical dosing regimen.

The final C‐QTc model was a linear mixed effects model with the effect of M11 concentration on ΔQTcP, which also included intercept, treatment (fruquintinib versus placebo), nominal time, visit, and baseline QTcP terms, with independent BSV on the intercept and ΔQTcP‐M11 concentration slope. A statistically significant slope of 0.0339 (95% CI: 0.00516, 0.0625) msec per ng/mL (P = .0212) was estimated. The final model was used to generate predictions of ΔΔQTcP, and the upper bounds of the 90% CI of mean ΔΔQTcP at the GM C_max of M11 (77 ng/mL) and twice the GM C_max (154 ng/mL) were 0.0537 and 4.00 msec, respectively. These results were notably lower than the clinically meaningful increase of 20 ms for oncology drugs. ¹¹ The only and highest dose used in the current concentration‐QT analysis is 5 mg QD for 21‐days in a 28‐day cycle from a subset of patients in a pivotal trial FRESCO‐2, which does not provide ≥2‐fold supratherapeutic exposure coverage, and no positive control was included in the assessment. Therefore, although the upper bound of model‐predicted ΔΔQTc is <10 ms, fruquintinib's QT effects is evaluated under ICH E14 Q&A 6.1. Specifically, ICH E14 Q&A 6.1 allows for excluding large QT effects (≥20 ms) of a drug if it does not prolong the QT interval by >10 ms without a positive control. ¹⁵ Because of the high proportion of M11 concentrations that were BLQ (34.8%), predictions were also generated using the model that included only fruquintinib concentrations. Results were similar, and the upper bound of the 90% CI of the predicted mean ΔΔQTcP at twice the GM steady‐state fruquintinib C_max (580 ng/mL) was 3.96 ms. A supportive analysis performed using ΔQTcF also predicted that the upper bound of the 90% CI of the mean ΔΔQTcF would not exceed 20 ms at M11 concentrations up to 177 ng/mL, 2.3‐fold higher than the observed GM M11 C_max at steady state following the recommended clinical dose of fruquintinib. Thus, the C‐QTc analysis can be considered robust, and the results indicate that fruquintinib administration does not lead to clinically significant QT prolongation. The findings of the primary analysis are further supported by the lack of QTc prolongation in categorical, by‐time point, and central tendency analyses of ECG parameters.

Fruquintinib 6 mg (1.2‐fold of the recommended dose at 5 mg) was also evaluated during clinical development. No clinically significant abnormal or prolonged QT interval and abnormal left ventricular ejection fraction were observed with the higher dose. ⁹ To date, no intrinsic or extrinsic factors have been identified that increase fruquintinib exposure. Therefore, the anticipated high clinical exposure for fruquintinib is the same as the observed clinical exposure. Coadministration of fruquintinib with rifampin, a strong CYP3A inducer, decreases fruquintinib C_max by 12% and increases M11 C_max by approximately 2.3‐fold. Consequently, moderate and strong CYP3A inducers are prohibited during fruquintinib treatment, ⁹ further minimizing the potential risk for increasing M11 plasma concentrations. Overall, the M11 and fruquintinib concentration ranges in the concentration‐QTc dataset were representative of fruquintinib treatment in a clinical setting.

In summary, in vitro hERG assays, in vivo safety pharmacology, concentration‐QT analyses, clinical cardiac safety assessment, and avoidance of concomitant use with moderate and strong CYP3A inducers in the prescribing guidance support the conclusion that fruquintinib does not cause QT prolongation at the recommended dosage.

Conflicts of Interest

Xiaofei Zhou is employed by Takeda Development Center Americas, Inc. (TDCA). Alice Toms reports employment for Certara, USA Inc., and potential ownership of various shares, through index and/or other funds with Certara, USA Inc. Dave Morton and Adekemi Taylor reports employment and ownership of stocks/shares with Certara, Inc. Arvind Dasari has been on an advisory council or committee for Bristol‐Myers Squibb, Exelixis, Illumina, Lantheus, Personalis, Taiho, and Takeda; and has received grants or funds from Eisai, Enterome, Guardant Health, HUTCHMED, Natera, Neogenomics, Personalis, Taiho, and Xencor. Neeraj Gupta is an employee of and reports ownership of stocks/shares in Takeda. Caly Chien is employed by and owns stock in HUTCHMED. Xiaohui Wang declares no conflict of interest.

Funding

Funding for the study used in this research was provided by Takeda Pharmaceuticals U.S.A., Inc. and HUTCHMED.

Supporting information

Supporting Information

Data Availability Statement

The datasets, including the redacted study protocol, redacted statistical analysis plan, and individual subjects’ data supporting the results reported in this article, will be made available within three months from initial request to researchers who provide a methodologically sound proposal. The data will be provided after its de‐identification, in compliance with applicable privacy laws, data protection and requirements for consent and anonymization.