Patidegib topical gel stopped in non-Gorlin basal cell carcinoma when the endpoint ran out of events

What was tried

PellePharm designed and ran NCT04155190, a multicenter, randomized, vehicle-controlled phase 2 study of patidegib topical gel 2 percent in adults with non-Gorlin high-frequency basal cell carcinoma. Sol-Gel Technologies is the current responsible party and posted the results. Participants applied patidegib gel or matching vehicle twice daily to the whole face for 9 months. Allocation was 1 to 1, stratified by gender, with participants, care providers, and investigators masked.

Entry required at least 10 clinically typical basal cell carcinomas within the 24 months before screening, at least 3 of them on the face, plus at least 2 facial lesions under 5 mm at baseline. A confirmed or suspected diagnosis of Gorlin syndrome was an exclusion, as was a family history of medulloblastoma. The primary endpoint was the number of new surgically eligible basal cell carcinomas, written nSEB, developing on the face across the 9 month period.

The study started on 2019-12-20 and reached actual primary completion on 2021-04-23. Fifty participants enrolled, 47 formed the intent to treat and dosed set, split at baseline into 23 on patidegib and 24 on vehicle. Twenty completed and 30 did not. The posted reason for stopping reads: "Terminated early due to low blinded event rate."

The biological hypothesis

Basal cell carcinoma is the clearest example in oncology of a tumor driven by one pathway. Loss of PTCH1 or activating mutation of SMO releases Smoothened from inhibition and drives constitutive Hedgehog signaling. Open Targets records an association score of 0.3885 between SMO and skin basal cell carcinoma, carried by literature and somatic mutation evidence, and lists 10 clinical or approved agents against the target. Patidegib, ChEMBL identifier CHEMBL538867 and formerly IPI-926, is a Smoothened antagonist that reached phase 3.

Systemic Smoothened antagonists are approved for advanced disease, but their class toxicity limits chronic preventive use. openFDA holds 8,768 adverse event reports naming vismodegib, of which 3,739 are serious, with muscle spasms (2,042 reports), alopecia (1,392), ageusia (1,002), and dysgeusia (745) leading the list. Sonidegib shows the same ranking across 1,477 reports. Patients with high lesion burden need years of therapy, and few tolerate that profile.

The topical hypothesis followed directly. Deliver a Smoothened antagonist into skin at concentrations sufficient to suppress Hedgehog signaling in nascent tumors while keeping plasma exposure low enough to avoid the class effects. A phase IIA trial in Gorlin syndrome supported the pharmacology, reporting very low blood concentrations, minimal adverse effects, and post hoc reductions in both new surgically eligible lesions and Hedgehog signaling.

What actually happened

The pivotal test came first in Gorlin syndrome. NCT03703310 randomized 174 participants 1 to 1 and treated for 12 months. Posted results give mean new basal cell carcinomas per participant of 2.84 with a standard deviation of 3.39 on patidegib against 4.03 with a standard deviation of 4.76 on vehicle, in analysis populations of 58 and 65. For the secondary nSEB endpoint the means were 0.71 (SD 1.28) against 0.91 (SD 1.61). Serious adverse events were 5 of 87 against 7 of 87. The registry posts no p value or confidence interval for either comparison.

NCT04155190 then moved the same drug into a non-Gorlin population with high sporadic lesion burden and promoted nSEB from secondary to primary. It never produced an arm-level readout. Posted results carry a limitations note stating that PellePharm terminated the study because of the low blinded event rate, and that participant data were collected and reported only in aggregate for participant flow, outcome measures, mortality, and adverse events. Eighteen of 47 participants had a treatment-emergent adverse event. Two serious events were reported across the pooled population, one respiratory syncytial virus infection and one myocardial infarction. There were no deaths.

Failure mechanism, best guess

The archetype here is translational mismatch, and the mismatch is between the population used to estimate the endpoint rate and the population enrolled. Two features of the design compounded each other.

First, the endpoint was rare by construction. Using the posted Gorlin phase 3 means, surgically eligible lesions were 0.91 of 4.03 new basal cell carcinomas in the vehicle arm, or 23 percent of all new lesions. Gorlin syndrome is the highest lesion burden setting available. Moving to a non-Gorlin population lowers the underlying rate, and shortening the observation window from 12 months to 9 lowers it again.

Second, the effect being chased was small in absolute terms. Taking the posted Gorlin nSEB means and standard deviations, the arm difference was 0.20 lesions per participant and the pooled standard deviation was 1.45, which gives a standardized effect of 0.14. A two-sample comparison at 80 percent power and a two-sided alpha of 0.05 needs roughly 16 divided by the square of the standardized effect per arm, or about 840 participants per arm. Those inputs come from NCT03703310 and the arithmetic is shown so it can be checked. A trial of 47 participants was never going to resolve that difference, and the blinded event count made the gap visible before unblinding.

Nothing in the posted record points to a pharmacological failure. Blood levels were low, the phase IIA work found local target engagement, and the safety picture was quiet.

How to prevent this next time

Run a blinded event rate check as a prespecified gate, not as a discovery. The termination language shows the sponsor was watching the pooled event count, which is the correct instrument. Building that check into the protocol with a stated threshold at a fixed enrollment fraction converts a trial closure into a planned design decision.

Anchor the endpoint rate to the population being enrolled. The nSEB rate available at design time came from Gorlin syndrome. Non-Gorlin high-frequency disease is defined by history over 24 months, a different quantity from prospective lesion accrual under trial surveillance in 9 months.

Use total new lesions rather than the surgically eligible subset when the subset is a quarter of the whole. Applying the same power arithmetic to the Gorlin total new lesion endpoint, with a difference of 1.19 and a pooled standard deviation of 4.13, gives a standardized effect of 0.29 and about 190 participants per arm.

The single highest leverage change would have been to size the trial from a prospectively measured nSEB rate in non-Gorlin high-frequency patients, rather than importing a rate from Gorlin syndrome and then discovering the shortfall in blinded data.

What this means for similar programs

SMO carries a modelled risk score of 41 out of 100 in the yellow band, with saturation contributing 12.97 of a possible 15 because 10 clinical or approved programs already exist against the target. Prevention trials in tumor-prone skin share this structure. The binding constraint is rarely pharmacology. It is whether the control arm generates enough events in the observation window to support a comparison, and that number should be measured rather than inherited.

Open questions

What were the observed nSEB counts by arm in NCT04155190? What blinded event rate threshold triggered the stop, and at what enrollment fraction? Was the 24 month historical lesion count ever validated against prospective accrual in a non-Gorlin cohort? Does Sol-Gel intend to test patidegib against a total new lesion endpoint, and in which population?

Sources

1. • ClinicalTrials.gov v2 record and posted results, NCT04155190. https://clinicaltrials.gov/study/NCT04155190 - ClinicalTrials.gov v2 record and posted results, NCT03703310, phase 3 in Gorlin syndrome. https://clinicaltrials.gov/study/NCT03703310 - Topical application of the Hedgehog inhibitor patidegib in patients with Gorlin syndrome: a phase IIA trial. British Journal of Dermatology, 2025. https://- Patidegib in Dermatology: A Current Review. International Journal of Molecular Sciences, 2021. https://- Topical hedgehog inhibitors for basal cell carcinoma: how far away are we? Expert Opinion on Pharmacotherapy, 2022. https://- Topical Delivery of Hedgehog Inhibitors: Current Status and Perspectives. International Journal of Molecular Sciences, 2022. https://- Basal cell carcinoma pathogenesis and therapy involving hedgehog signaling and beyond. Molecular Carcinogenesis, 2017. https://- Evaluation of Hedgehog Pathway Inhibition on Nevoid Basal Cell Carcinoma Syndrome Fibroblasts and Basal Cell Carcinoma Cells. Cancers, 2021. https://- Open Targets Platform, target SMO (ENSG00000128602), association with skin basal cell carcinoma (MONDO_0005341), accessed 2026-07-27. https://platform.opentargets.org/target/ENSG00000128602 - ChEMBL molecule record CHEMBL538867 (patidegib). https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL538867/ - openFDA drug adverse event API, vismodegib and sonidegib queries, accessed 2026-07-27. https://api.fda.gov/drug/event.json.