News|Articles|November 19, 2026

Dose Optimization, Evidentiary Bars, and Patient Selection Shape Trial Design in EGFR- and HER2-Mutant NSCLC

Author(s)OncLive Staff
Fact checked by: Caroline Seymour
Listen
0:00 / 0:00

Key Takeaways

  • Model-based/model-assisted phase 1 designs now dominate, reflecting Project Optimus’ shift from MTD to optimal dosing where efficacy plateaus despite escalating toxicity.
  • Multi-dose expansions can require 20–30 patients per arm, exemplified by tarlatamab where 10 mg outperformed 100 mg in ORR, guiding dose selection.
SHOW MORE

As oncology drug development shifts away from maximum tolerated dose–driven paradigms, early design choices now determine whether an active agent ever reaches patients with non–small cell lung cancer (NSCLC). "These are multimillion dollar decisions, and not just that, they're decisions that influence potentially if you're going to find a drug that can help patients," Joshua Reuss, MD, of MedStar Georgetown University Hospital, said during a recent OncLive® Scientific Interchange and Workshop.1 Over a single day, fellows from 4 academic programs worked as drug development teams, advancing 2 simulated assets—a brain-penetrant EGFR × MET bispecific antibody for EGFR-mutant NSCLC and a HER2-directed antibody-drug conjugate (ADC) for HER2-mutant NSCLC—with faculty from academia, industry, biostatistics, and the FDA weighing in at each stage.

Top Takeaways From an OncLive Workshop on Clinical Trial Design

  • Faculty described maximum tolerated dose–first escalation and the 3+3 design as obsolete under Project Optimus, with model-based designs and multidose expansion now the default.
  • An FDA medical officer said breakthrough therapy decisions are benchmarked to the best available therapy, and that small single-arm datasets rarely clear that bar.
  • Patient selection—biomarker enrichment, mandatory biopsies, and eligibility breadth—was framed as the central trade-off between scientific precision and accrual.

How has Project Optimus changed early-phase dose selection?

Cody Chiuzan, PhD, of Northwell Health, said the 3+3 design has effectively disappeared from protocol review. "In the last two years, not one single protocol has used the 3+3," she said, citing model-based and model-assisted designs as the new standard. Because efficacy often plateaus with novel agents while toxicity continues to rise, Chiuzan advocated designs that model safety and efficacy jointly to identify an optimal dose rather than the maximum tolerated dose. Backfill and expansion cohorts, she added, are pushing phase 1 trials into the hundreds of patients.

Parth Desai, MD, MBBS, of Fox Chase Cancer Center, said carrying several doses into expansion typically requires roughly 20 to 30 patients per arm. He pointed to tarlatamab (Imdelltra), for which the phase 2 DeLLphi-301 trial (NCT05060016) evaluated 10-mg and 100-mg doses; the objective response rate (ORR) was 40% with the 10-mg dose vs 32% with the 100-mg dose, and the lower dose moved forward.2 The FDA finalized its guidance on oncology dosage optimization in August 2024.3

What evidence does the FDA need to see from early-phase trials?

Geetika Srivastava, MD, MSPH, a medical officer in gastrointestinal oncology at the FDA, walked fellows through the 30-day investigational new drug review clock. "If you don't hear from the FDA in 30 days, that is good news," she said. Whereas the bar for fast track designation is low, she said, breakthrough therapy designation is measured against the best available therapy. "We are not going for the lowest common denominator there," Srivastava said, noting that ORR data from approximately 10 patients with wide CIs generally fall short. Asked by Benjamin Levy, MD, of Johns Hopkins Sidney Kimmel Comprehensive Cancer Center, how "substantial improvement" is defined without a comparator, she acknowledged internal debate: "Sometimes the room is literally half and half split…A slam dunk drug, we will get it."

Reuss said accelerated approval decisions can be hard to predict, contrasting the approval of datopotamab deruxtecan (Datroway) in previously treated EGFR-mutant NSCLC on pooled single-arm data with the outcome for ifinatamab deruxtecan in small cell lung cancer. Srivastava attributed such differences to disease rarity and confirmatory trial feasibility, and called pre-IND meetings "an invaluable resource."

How should investigators balance precision against accrual in patient selection?

Reuss framed biomarker strategy as a choice between market share and identifying who benefits. "If you have a drug that looks promising, but you try to enroll in as many people as possible, that phase 2 or phase 3 study might fail," he said. Srivastava presented the regulator's view of the same tension, saying the FDA may ask sponsors, "Do you really want to have a mandatory biopsy? We get it. You need it for your other analyses, but this could really affect your accrual." Desai urged a correlative science plan regardless: "Your drug might fail, your trial might fail, but you should not fail to learn from the study."

The trade-off resurfaced in the team pitches. One team, proposing an EGFR × TROP2 bispecific ADC, chose plasma circulating tumor DNA over tissue sampling after consulting patient advocates, and faculty pressed for a tissue-based response analysis. The second team designed a HER2 ADC with an antimicrotubule payload for patients previously treated with fam-trastuzumab deruxtecan-nxki (Enhertu).

What unmet needs did the panel identify?

Faculty pointed to training gaps around investigator-initiated trials, sponsor partnerships, and data ownership. "It's not just what the investigator thinks, it's about who you're partnering with to do an investigator-initiated study," Reuss said. Access to real-time biostatistics support was flagged as a divide between large and smaller sites. Faculty also cautioned against delegating design judgment to artificial intelligence. "I've received protocols every week generated by AI, but they're general. Your trial is specific," Chiuzan said, urging investigators to consult a statistician before institutional review board submission. "A poorly designed study cannot be fixed with any Bayesian or frequentist in any kind of method in the world."

References

  1. Designing a cancer clinical trial development program. An OncLive Scientific Interchange and Workshop. OncLive. October 7, 2026. Accessed November 19, 2026.
  2. Ahn MJ, Cho BC, Felip E, et al. Tarlatamab for patients with previously treated small-cell lung cancer. N Engl J Med. 2023;389(22):2063-2075. doi:10.1056/NEJMoa2307980
  3. Optimizing the dosage of human prescription drugs and biological products for the treatment of oncologic diseases: guidance for industry. FDA. August 2024. Accessed November 19, 2026. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/optimizing-dosage-human-prescription-drugs-and-biological-products-treatment-oncologic-diseases

Related to this article