In phase 1b, the FAS (N = 45) included patients without swallowing dysfunction (alpelisib 300 mg/d + cetuximab, n = 16; alpelisib 400 mg/d + cetuximab, n = 5); those with swallowing dysfunction were allocated to arm B (alpelisib 300 mg/d + cetuximab, n = 18) or C (alpelisib 300 mg/d + cetuximab, n = 6).

Of 106 cetuximab-naïve patients in the phase 2 randomised part, 71 received alpelisib + cetuximab and 35 received cetuximab monotherapy. Of these 35 patients, 16 (45.7%) switched to combination therapy (arm 2B) after disease progression; 19 did not (Table 1).

Patients in phase 1b and phase 2 discontinued the study treatment primarily due to disease progression or adverse events (AEs) (Table 2). Approximately, half of the patients in phase 1b and the non-randomised arm of phase 2 and 35% in randomised part of phase 2 had stage 3/4 disease at initial diagnosis. The demographic characteristics matched between treatment arms, except for ECOG status 0/1 and tumour lysis syndrome (TLS) risk (both higher in combination arm versus cetuximab arm) (Table 1). In the phase 2 part, PIK3CA alterations in archival tumour tissue by Foundation One next-generation sequencing was found in 8/44 samples and 5/20 samples analysed in the alpelisib + cetuximab and cetuximab monotherapy arms, respectively. Due to the low frequency of PIK3CA alterations in this patient population, a subset analysis by PIK3CA alteration was not deemed meaningful (Fig. 2).

a Best percentage change from baseline in sum of longest diameters and best overall response (full analysis set). Phase 1b. Phase 2 randomised part. Treatment arm—alpelisib 300 mg + cetuximab. Phase 2 randomised part. Treatment arm—cetuximab. n, number of patients with a baseline and ≥ 1 post-baseline assessment of target lesions based on investigator’s assessment. #Patients missing best percentage change from baseline are not included. b Waterfall plot of percent change from baseline in sum of longest diameters as per central radiology review by treatment (full analysis set). Phase 2 randomised part. n, number of patients with a baseline and ≥ 1 post-baseline assessment of target lesions based on central radiology review. CR complete response, PD progressive disease, PR partial response, SD stable disease, UNK unknown

Patients who did not receive ≥ 21 doses of alpelisib and ≥ 3 doses of cetuximab, i.e. 75% of the planned doses, during cycle 1 without experiencing a DLT were excluded from the dose-determining set (DDS) per protocol. Based on these criteria, 15 out of 45 patients have been excluded from the DDS. Two patients in the combination treatment arm died before the initiation of study treatment and were excluded from the safety set.

Of 30 patients, 9 had a DLT within 28 d of study initiation (1 on alpelisib 300 mg/d tablet, 2 on alpelisib 400 mg/d tablet, 4 on drinkable suspension and 2 on dispersible alpelisib). Per BLRM inference, alpelisib 300/350 mg/d whole-tablet cetuximab showed a 1.5% and 15.3% risk of overdosing for 300 and 350 mg/d, respectively. In patients on drinkable suspension (arm B) of alpelisib 300 mg/d, the posterior probability of the DLT rate lying within the targeted toxicity range (16–35%) for alpelisib 300 mg/d was 66.4%, with a 6.9% risk of overdosing (i.e. the posterior probability of a DLT rate of > 35%). In patients on dispersible tablets (arm C), the posterior probability of DLT rate lying within the targeted toxicity range (16–35%) for 300 mg alpelisib–cetuximab was 59.5%, with a higher risk of overdose (25.9%).

Of 45 patients enrolled in phase 1b study, most patients (n = 44, 97.8%) experienced ≥ 1 any-grade AE, regardless of relationship to the study treatment (Table 3). The most commonly reported all-grade AEs were hyperglycaemia, hypomagnesaemia, stomatitis, diarrhoea, weight decrease, loss of appetite, rash and fatigue. The most common reason for discontinuation was AEs (15 patients [33%] in phase 1b and 18 [17%] in phase 2), with the most frequent being hyperglycaemia (9% of patients in phase 1b, 3% in phase 2). The incidence of AEs leading to study treatment discontinuation was higher in the randomized combination arm versus the cetuximab arm (15 patients [21.7%] versus 3 [8.6%]) in phase 2 (Table 4). The incidence of clinically notable AEs for phase 1b studies are presented in Table S1.

In phase 1b, all patients experienced AEs that were suspected to be related to study treatment. The most frequently reported AEs suspected to be related to study treatment included hyperglycaemia, hypomagnesaemia, diarrhoea and rash. The most frequently reported serious adverse events (SAEs) were pneumonia, hyperglycaemia and pyrexia. Grade 3/4 AEs were lower in treatment arm B (n = 12; 66.7%) compared with arm A (n = 13; 86.7%) and arm C (n = 5; 83.3%).

In phase 2, the incidence of any-grade AEs was higher (difference ≥ 10%) in the combination treatment arm compared with the cetuximab monotherapy arm. Patients experiencing grade 3/4 AEs and SAEs were higher in the combination treatment arm (87% and 58%, respectively) compared with the cetuximab monotherapy arm (57.1% and 42.9%, respectively). The most frequently reported AEs suspected to be study treatment related in the combination treatment arm were hyperglycaemia, stomatitis, diarrhoea, rash, decreased appetite, dermatitis acneiform, fatigue, weight decrease, hypomagnesaemia, paronychia, dry skin and nausea.

Eight on-treatment deaths were reported during phase 1b: disease progression (n = 3), AEs including pneumonia (n = 3), septic shock and TLS (n = 1 each); only TLS was considered treatment related. In phase 2, 15 on-treatment deaths (21.7%) were reported in the randomized combination arm and 8 (22.9%) in the cetuximab arm (4 patients on cetuximab monotherapy who did not cross over and 4 who switched to combination treatment); 3 deaths (10.3%) were reported in the non-randomised arm. Of the 15 patients who died in the combination treatment arm, 7 patients died due to the study indication or squamous cell carcinoma. Eight patients died due to AEs: pneumonia (n = 4), sepsis (n = 1), pulmonary haemorrhage (n = 1), respiratory failure (n = 1) and tumour haemorrhage (n = 1). Pneumonia in one patient was suspected to be related to study treatment. Of the four patients randomized to cetuximab monotherapy who died after the cross-over, two patients died due to study indication and two patients died due to AEs: one patient from sepsis and one patient from respiratory distress, neither event was suspected to be related to study treatment.

In the phase 1b study, the ORR [95% confidence interval (CI)] was 8.9% (2.5–21.2%), and the overall DCR (95% CI) was 51.1% (35.8–66.3%). Four patients (25%) treated with whole-tablet alpelisib 300 mg/d + cetuximab achieved partial response (PR) per RECIST version 1.1; 18 (40.0%) had stable disease (SD) (Table 5). No arm B patients who received an oral suspension of alpelisib crushed tablets and cetuximab achieved a partial response/complete response.

In the phase 2 randomised part, a trend indicating higher ORR and DCR in the combination arm was observed. The ORR (95% CI) was 9.9% (4.1–19.3%) in the combination arm and 5.7% (0.7–19.2%) in the cetuximab arm (Table 5). The DCR (95% CI), including CR, PR, SD or non-CR/non-PD, was 52.1% (39.9–64.1%) in the combination arm and 57.1% (39.4–73.7%) in the cetuximab arm. Of the nine patients in the cross-over part, four showed reductions in tumour burden; none achieved PR/CR (ORR 10.3%, 95% CI 2.2–27.4%; DCR 10.3%, 95% CI 20.7–57.7%).

Of the 23 patients who had tumour reduction in phase 1b, 11 had tumour shrinkage ≥ 30%. In the phase 2 part, tumour shrinkage was observed in 31 of 44 patients in the combination arm and in 9 of 21 patients in the cetuximab arm. There was a higher number of patients with tumour shrinkage ≥ 30% in the combination arm (12 of 44 patients) when compared with the cetuximab arm (2 of 21 patients (Fig. 2)).

Based on central review data, the mean of posterior distribution HR of PFS between combination treatment and cetuximab monotherapy estimated using the Bayesian Cox proportional-hazards model was 0.991 (95% credible interval, 0.643–1.529). The posterior probability of an HR of > 1 was 48.4%. As the posterior summaries did not meet the predefined criteria for success (interval probabilities: substantial efficacy 0.550%, moderate efficacy 5%, slight efficacy 46.06%, no efficacy 48.39%), combination therapy was not declared superior to cetuximab monotherapy. Median PFS (95% CI) per central review was 86 d (73–135 d) in the combination arm and 87 d (49–133 d) in the cetuximab arm. Median PFS for the 16 patients who crossed over from cetuximab to combination treatment was 43 d (95% CI 27.0–88.0 d). The median of the posterior distribution of PFS in cetuximab-resistant patients was 3.896, and the posterior probability of the median PFS being > 2.5 months was 99.69%, indicating moderate efficacy in the non-randomised arm.

The Cox regression model (frequentist approach) showed no significant difference between the combination and cetuximab arms for PFS (unadjusted HR 1.12; 95% CI 0.69–1.82) (Fig. 3). In the model adjusted for selected baseline covariates (sum of the longest diameters [SLD]) from central review/data, haemoglobin and white blood cell (WBC), the adjusted HR was 0.54, indicating a significant difference in favour of the combination arm (adjusted HR 0.54; 95% CI 0.30–0.97).

a Kaplan–Meier plot of progression-free survival as per central review for phase 2 randomised part (full analysis set). b Overall survival—phase 2 randomised part (full analysis set). CI confidence interval, HGB haemoglobin, SLD (C) sum of the longest diameters from central review/data, SLD (L) sum of the longest diameters from investigator/local data, WBC white blood cell

The analysis of PFS based on investigational site assessment resulted in an unadjusted HR of 0.76, trending in favour of the combination arm [not significant (NS)]. In the model adjusted for selected baseline covariates, the HR reduced to 0.64 in favour of the combination arm (non-significant).

The median OS (95% CI) was 173 d (142–249) in the combination arm and 263d  (181–444) in the cetuximab arm. Unadjusted Cox proportional-hazards model data represented a non-statistically significant trend of higher death risk in the combination arm (unadjusted HR 1.28; 95% CI 0.80–2.05). However, the model adjusted for selected baseline covariates resulted in an adjusted HR of 1.00, suggesting that the initial trend observed in the unadjusted analysis might be attributed to unbalanced baseline covariates across the two groups (Fig. 3). The pharmacokinetics and biomarker data are shown in the supplementary information (Supplementary Fig. S1, Table S2).