3.1 Population Pharmacokinetic Model

The pooled PopPK analysis included 4746 quantifiable cabozantinib PK samples obtained from 1745 patients and healthy volunteers, including 205 PK samples from 101 patients in the COSMIC-311 trial who received cabozantinib and had measurable cabozantinib concentrations. Fourteen subjects were excluded from the PopPK analysis due to missing information. Demographic and covariate information for the 101 patients in COSMIC-311 receiving cabozantinib who had quantifiable PK samples are summarized in Table 1. Demographic and covariate information for all 1745 individuals (1682 patients with cancer and 63 healthy volunteers) included in the analysis are summarized in Supplementary Table S2.

Table 1 Summary of covariate information for patients receiving cabozantinib in COSMIC-311 included in population pharmacokinetic model3.1.1 Base PopPK Model Development

Model development was initiated with PK data from a single-dose study in healthy volunteers receiving 20, 40, and 60 mg of cabozantinib (study XL184-020). The primary absorption process describing the initial absorption phase was best characterized by a model with 4 transit compartments. The secondary absorption process included a lag time, which described the increase in cabozantinib exposure at 24 h post-dose relative to the cabozantinib exposure observed at 14 h post-dose in XL184-020.

After including PK data from the other clinical trials, the structure of the residual error (RE) component of the PK model was best described by a RE term for healthy volunteers and a separate RE term for patients with various cancer types. This was considered the base model (see Supplementary Materials for base model equations).

3.1.2 Final Model Development

A full model was developed by incorporating potential effects of sex and body weight on CL/F and body weight on Vc/F (see Supplementary Material for model equations details). The full model was determined to be the final model upon achieving acceptable predictive performance by pcVPC evaluation. Final (full) model PK parameter estimates are reported in Table 2. Predicted mean exposure measures at steady state in adults following 60 mg cabozantinib QD based on the final PK model are summarized by study in Supplementary Table S3.

Table 2 Pharmacokinetic parameter estimates for the final (full) PK model

The effects of body weight and sex on the following PK parameters were assessed: steady state area under the concentration-time profile during one dosing interval (AUC0–24,ss), steady state maximum concentrations (Cmax,ss), pre-dose plasma concentrations at steady state (Cmin,ss), CL/F, and Vc/F. The reference patient was a 70-kg male with DTC receiving 60 mg cabozantinib QD. Body weight had minimal impact on cabozantinib exposure based on predicted AUC0–24,ss (< 6% change), Cmax,ss (< 14% change) and Cmin,ss (< 4% change) values, but had a notable impact on Vc/F (Supplementary Fig. S1). Lighter weight (53 kg, 5th percentile) patients had approximately 40% lower Vc/F versus a reference patient, while heavier weight (106 kg, 95th percentile) patients had an over 2-fold larger Vc/F. Females were estimated to have an approximately 20% lower CL/F than males, which corresponded to a 27% higher AUC0–24,ss, 23% higher Cmax,ss, and 29% higher Cmin,ss values. There was a lower (not clinically significant) CL/F and Vc/F in the Asian population compared to the White population. The CL/F and Vc/F range was overlapping in DTC patients and other cancer types.

3.1.3 Model-Based Simulations in Adolescents

Figure 1 shows the predicted steady-state AUC0–24,ss, Cmax,ss, and Cmin,ss exposure for adolescent patients with DTC receiving 60 mg QD and 40 mg QD, stratified by weight group, based on stochastic simulations using the final PK model and the CDC weight ranges for this age group. Most of the interquartile ranges of predicted cabozantinib exposures based on these PK parameters for adolescent patients with DTC receiving 60 mg QD fell within the simulated adult exposure associated with a daily dose of 60 mg, except Cmax,ss for adolescent patients with body weight < 40 kg. The predicted Cmax,ss by weight with daily dose of 60 mg for these patients was on the high side of the range observed for adult patients receiving 60 mg QD. The predicted steady state AUC0–24,ss, Cmax,ss and Cmin,ss by weight for adolescent patients with DTC receiving 40 mg QD tended to be on the low side of the range relative to adult patients with DTC receiving 60 mg QD.

Fig. 1

Predicted steady-state adolescent cabozantinib exposure (AUC0–24,ss [A], Cmax,ss [B], and Cmin,ss [C]) using the final (full) model (60 mg QD and 40 mg QD doses). The final (full) model was developed based on data from 7 cabozantinib studies. Lower and upper boundaries of the box represent the 1st quartile (Q1) and 3rd quartile (Q3), respectively; median is shown as a line inside the box and labelled as the value inside the box; whiskers represent minimum and maximum values that are within 1.5-times the inter-quartile range (IQR) below Q1 and above Q3, respectively; black circles represent outliers (values >1.5-times IQR below Q1 or above Q3); the gray shaded region represents the 90% prediction interval of adult reference (based on a 60 mg daily dose); solid line through the gray region is the predicted median adult reference. AUC0–24,ss steady state area under the concentration-time profile during one dosing interval (ng*h/mL), Cmax,ss steady state maximum plasma drug concentration (ng/mL), Cmin,ss pre-dose plasma drug concentrations at steady state (ng/mL), PI prediction interval

3.1.4 Allometric Scaling Simulations in Adolescents

Results of the allometric scaling simulations are shown in Fig. 2. Box plots in this figure illustrate the predicted steady-state AUC0–24,ss, Cmax,ss, and Cmin,ss based on these parameter estimates and allometric exponents, stratified by weight group.

Fig. 2

Predicted steady-state adolescent cabozantinib exposure (AUC0–24,ss [A], Cmax,ss [B], and Cmin,ss [C]) using the final (full) model and allometric scaling (60 mg QD and 40 mg QD doses). The final (full) model was developed based on data from 7 cabozantinib studies. Lower and upper boundaries of the box represent the 1st quartile (Q1) and 3rd quartile (Q3), respectively; median is shown as a line inside the box and labelled as the value inside the box; whiskers represent minimum and maximum values that are within 1.5-times the inter-quartile range (IQR) below Q1 and above Q3, respectively; black circles represent outliers (values >1.5-times IQR below Q1 or above Q3); the gray shaded region represents the 90% prediction interval of adult reference (based on a 60 mg daily dose); solid line through the gray region is the predicted median adult reference. AUC0–24,ss steady state area under the concentration-time profile during one dosing interval (ng*h/mL), Cmax,ss steady state maximum plasma drug concentration (ng/mL), Cmin,ss pre-dose plasma drug concentrations at steady state (ng/mL), PI prediction interval

Adolescents with DTC and body weight < 40 kg were predicted to have approximately 1.7-fold higher median AUC0–24,ss, Cmax,ss, and Cmin,ss for a daily 60-mg regimen relative to that of adults with DTC receiving the same dose. Exposure from 40 mg QD dosing in adolescents < 40 kg was similar (within 10% in median exposure) to that of adults receiving 60 mg QD.

For adolescents with DTC weighing ≥ 40 kg, the median exposure with daily doses of 60 mg QD was higher than in adults, but the interquartile range (IQR) was within the 90% prediction interval. As with the model-based simulation, predicted exposures with daily doses of 40 mg for adolescents weighing ≥40 kg were on the lower side of the range of exposures observed for adults receiving 60 mg QD.

3.2 Exposure Response Analyses

The number of patients with PFS or safety events and the total number of patients at risk in the COSMIC-311 trial are listed in Supplementary Table S4. In total, 125 patients had at least one documented cabozantinib dose, of whom 115 had at least one measurable concentration and were included in the exposure–response analysis (14 patients who were excluded from the PopPK analysis due to missing information were assigned typical PopPK parameters for the exposure–response analyses). The PK base model was used to generate the exposures for the 115 patients included in the exposure–response analyses.

3.2.1 Progression-Free Survival

The KM plot for PFS by average cabozantinib exposure tertiles (Fig. 3) showed no clear relationship between the fraction of patients with a PFS event and the different exposure tertiles of cabozantinib. The log-rank test (p = 0.763) indicated that there was no statistically significant difference across the cabozantinib exposure tertiles (Table 3). A similar result was observed when the 14 patients missing PK information were excluded (p = 0.61).

Fig. 3

Kaplan–Meier plot for progression-free survival by average exposure tertile. Data derived from 98 patients with at least one measurable cabozantinib concentration, a valid baseline tumor assessment, and at least one evaluable post-baseline tumor assessment in COSMIC-311. Dashed lines represent 95% confidence intervals for each exposure tertile. CAVG0T predicted average cabozantinib concentration from time zero to the event or censoring time, PFS progression-free survival

Table 3 Log-rank tests for efficacy and safety endpoints with exposure tertiles based on data from COSMIC-3113.2.2 Safety Endpoints

The KM plot for cabozantinib dose modification by average cabozantinib exposure based on the predicted average cabozantinib CAVG1W showed no clear relationship between cabozantinib dose modification and cabozantinib exposure (Supplementary Fig. S2). The log-rank test indicated no statistically significant difference across the cabozantinib exposure tertiles (Table 3).

For hypertension assessed by BP data (using the vital signs source data), the KM plot showed a smaller fraction of patients with hypertension of Grade ≥ 3 in the lowest cabozantinib exposure tertile compared to the highest exposure tertile (Supplementary Fig. S3), and the log-rank test (p = 0.027) indicated a statistically significant relationship between cabozantinib exposure and the incidence of hypertension Grade ≥ 3 (Table 3). For hypertension assessed by MedDRA, neither the KM plot nor the log-rank test (Supplementary Fig. S4; Table 3) indicated any clear relationship between hypertension of Grade ≥ 3 and cabozantinib exposure. It should be noted that there were more Grade ≥ 3 hypertension AEs based on the BP source data (24/115 patients) than there were using the MedDRA terms (11/115 patients). While both analyses used the same systolic and diastolic BP cut-offs to determine Grade ≥3 events, systolic and diastolic BP were counted separately in the BP vital signs data, resulting in more safety events.

The KM plot for fatigue/asthenia by average cabozantinib exposure tertiles showed no instances of Grade ≥ 3 in the lowest cabozantinib exposure tertile and similar rates among the higher two exposure tertiles (Supplementary Fig. S5). The log-rank test (p = 0.025) indicated a statistically significant relationship between cabozantinib exposure and the incidence of fatigue/asthenia Grade ≥ 3 (Table 3).

For PPE (Grade ≥ 1), diarrhea (Grade ≥ 3), nausea/vomiting (Grade ≥ 3), mucositis/stomatitis (Grade ≥ 3), and ALT/AST elevations (Grade ≥ 3), there was no statistically significant difference in rates of these AEs across the cabozantinib exposure tertiles based on the log-rank test (Table 3). The KM plot for PPE by average cabozantinib exposure tertiles indicated no significant relationship between cabozantinib exposure and the rate of PPE, although there was a trend of an increasing frequency at the higher cabozantinib exposures after the first month (Supplementary Fig. S6). The KM plot for diarrhea by average cabozantinib exposure tertiles indicated no patients with diarrhea of Grade ≥ 3 in the lowest exposure tertile and similar rates of diarrhea among patients in the two higher exposure tertiles (Supplementary Fig. S7). The KM plots for nausea/vomiting and mucositis/stomatitis (Supplementary Figs. S8 and S9) demonstrated no clear relationship between the rate of these AEs (Grade ≥ 3) and cabozantinib exposure. Likewise, the KM plot for ALT/AST elevation (Grade ≥ 3) did not indicate any relationship with cabozantinib exposure (Supplementary Fig. S10); only one instance of this AE occurred, which was in cabozantinib exposure tertile 2.