Evidence grades (GRADE-adapted): A high — multiple well-conducted RCTs or systematic reviews converge. B moderate — single pivotal RCT or consistent observational evidence. C limited — single observational study, mechanistic, or expert consensus. D preclinical / hypothesis-generating.
Why adaptive platform trials
Traditional phase III randomized trials test a single intervention against a single control in a fixed-size cohort. They are statistically rigorous but slow, expensive, and committed to enrollment criteria fixed at study initiation. In oncology, where biomarker-driven precision approaches are increasing, traditional designs are limited:
- A drug effective in a biomarker subset may fail an all-comer trial that dilutes the signal
- Multiple investigational drugs require separate trials with their own infrastructure
- Promising signals require new trials for confirmation rather than seamless escalation
- Enrollment of competing trials in the same disease creates inefficiency
Adaptive platform trials address these limitations by allowing pre-specified adaptations (arm dropping, accrual adjustment, biomarker-subgroup enrichment), shared control arm across investigational arms, and continuous addition of new investigational drugs as old ones complete.
I-SPY 2 — design and methodology
I-SPY 2 (Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging And Molecular Analysis 2) launched in 2010 as a neoadjuvant platform trial in high-risk early-stage breast cancer, with TNBC as one of the principal enrollment populations[1]A. Key design elements:
Biomarker subtype stratification
Patients are classified into eight subtypes by HR, HER2, and MammaPrint risk-score combinations, with TNBC as one principal class (HR−/HER2−). Investigational arms can be matched to specific subtypes or tested across multiple subtypes.
Bayesian adaptive randomization
As patients enroll and pCR data accumulate, the randomization probability for each investigational arm is updated based on the arm's predicted pCR rate in each subtype. Arms with positive emerging signals receive more patients in the relevant subtypes; arms with weak signals receive fewer.
Drug-graduation methodology
An investigational arm "graduates" from I-SPY 2 when its Bayesian-predicted probability of success in a hypothetical 300-patient phase III trial exceeds a pre-specified threshold (typically 85%). Graduation signals that the drug is ready for phase III confirmatory testing, having been validated in I-SPY 2's adaptive design. Drugs can also "fail" out of I-SPY 2 if their predicted phase III success probability drops below a futility threshold.
Common neoadjuvant chemotherapy backbone
All patients receive standard neoadjuvant taxane → anthracycline; investigational arms add the test drug to this backbone. This common backbone enables shared control-arm comparisons across investigational arms.
Endpoint — pCR
pCR is the trial's primary endpoint, enabling fast readout per the CTNeoBC-based regulatory framework (see pCR as endpoint synthesis).
I-SPY 2 graduates and TNBC
Over 20 investigational drugs / drug combinations have been tested in I-SPY 2 since 2010. Notable TNBC-relevant graduates and outcomes:
- Veliparib + carboplatin — graduated in TNBC with promising signal; subsequent phase III BrighTNess trial confirmed carboplatin benefit but not veliparib benefit. Lesson: I-SPY signals are biological hypothesis-generating but require confirmatory phase III.
- Pembrolizumab — graduated in TNBC and HR+/HER2−; the TNBC graduation supported the KEYNOTE-522 confirmatory trial that established pembrolizumab + chemo as standard.
- Datopotamab deruxtecan + durvalumab — testing in TNBC arm with promising preliminary readouts.
- Tucatinib + trastuzumab — tested in HER2-positive arm; not TNBC-relevant.
- Sacituzumab govitecan — tested in TNBC arm; outcomes contributing to the adjuvant and earlier-line development programs.
The I-SPY 2 evidence base has informed multiple drug-development decisions in TNBC and has demonstrated proof-of-concept for platform-trial methodology that other consortia have adopted.
Operational characteristics
I-SPY 2 has approximately 20 participating cancer centers in the US, with enrollment of about 100–200 patients per year. Drug-development sponsors contribute investigational arms; the trial infrastructure (control arm, biomarker stratification, statistical methodology) is provided by the consortium. This shared-infrastructure model substantially reduces per-drug development cost compared with standalone phase II trials.
Operational considerations include:
- Pharmaceutical-sponsor agreements regarding drug intellectual property and graduation criteria
- Real-time central biomarker assessment for subtype assignment
- Centralized adaptive randomization with statistical review
- Patient enrollment infrastructure across multiple participating centers
- Long-term follow-up coordination for event-time endpoints
Related platform trials in TNBC
- I-SPY 2.2 — evolution of I-SPY 2 with refined biomarker stratification and additional drug arms; continues active enrollment.
- FUTURE series (FUSCC, China) — biomarker-stratified umbrella trial for refractory metastatic TNBC, using the FUSCC molecular subtype framework to assign patients to subtype-matched targeted therapy arms (see Burstein synthesis for FUSCC framework).
- TRAMS-Breast / ASCENT-04 / similar — more focused trials with adaptive elements but not full platform designs.
Lessons learned from a decade of I-SPY 2
- Platform trials accelerate drug-development decisions. The pembrolizumab-TNBC graduation directly informed the KEYNOTE-522 phase III; the time savings vs traditional sequential phase II/III development is meaningful.
- I-SPY 2 graduates require confirmatory phase III. The BrighTNess veliparib failure despite I-SPY 2 graduation demonstrated that adaptive trial signals are biological hypotheses requiring confirmation, not standalone definitive evidence.
- Subtype stratification matters in TNBC. Several drugs showed differential signals across subtypes that all-comer trials would have missed.
- Operational complexity is substantial. Running a platform trial requires investment in central infrastructure; not all institutions can participate without external support.
- Regulatory acceptance is increasing. FDA has accepted adaptive-trial evidence for accelerated approval in several contexts; the precedent informs new platform-trial design.
Evidence table
| Trial / Drug | TNBC outcome | Status |
|---|---|---|
| I-SPY 2 veliparib + carbo | Graduated in TNBC | BrighTNess phase III confirmed carbo only |
| I-SPY 2 pembrolizumab | Graduated in TNBC | KEYNOTE-522 phase III confirmed; approved |
| I-SPY 2 datopotamab deruxtecan + durvalumab | Testing in TNBC arm | Ongoing |
| I-SPY 2 sacituzumab govitecan | Testing in TNBC arm | Informing adjuvant program |
| FUTURE-A (FUSCC LAR) | Pyrotinib + capecitabine | Higher RR than historical controls |
| FUTURE-B (FUSCC IM) | Camrelizumab + chemo | Improved outcomes |
Open questions and active investigation
- Will adaptive trials directly support FDA approval? Currently, I-SPY 2 graduates require confirmatory phase III. Whether adaptive-trial evidence alone can support full approval (not just accelerated approval) is being explored in some indications.
- Biomarker refinement within TNBC. I-SPY 2 treats TNBC as a single subtype; sub-stratification within TNBC (basal-like, LAR, immune-rich, BRCA-mutant) could improve precision. I-SPY 2.2 incorporates some refinements.
- Adaptive trial methodology in metastatic TNBC. I-SPY 2 focused on neoadjuvant; metastatic-setting adaptive trials face different challenges (heterogeneous prior therapy, varied biomarkers) but are being developed.
- Multi-arm adjuvant trials. Platform trial methodology applied to adjuvant therapy (residual-disease setting after KEYNOTE-522) could accelerate evaluation of multiple adjuvant agents in parallel.
- Patient and clinician acceptance. Adaptive randomization with subtype stratification is operationally complex from the enrollment perspective; patient and clinician decision aids would improve enrollment.
- Cost-effectiveness and access. Platform trials require central infrastructure that may not be available in all health systems globally. Adapting the model for resource-limited settings is an active question.
For the pCR endpoint that I-SPY 2 uses for its primary readout, see the pCR endpoint synthesis and the endpoint design synthesis. For biomarker-stratified trial designs generally, see the (forthcoming) biomarker-stratified design synthesis.
References
Each citation links to the original publication via DOI. The same records are searchable in the evidence library by title or DOI.
- Barker AD, Sigman CC, Kelloff GJ, et al. I-SPY 2: an adaptive breast cancer trial design in the setting of neoadjuvant chemotherapy. Clin Pharmacol Ther. 2009;86(1):97–100. doi:10.1038/clpt.2009.68. ↩
- Rugo HS, Olopade OI, DeMichele A, et al. Adaptive randomization of veliparib-carboplatin treatment in breast cancer. N Engl J Med. 2016;375(1):23–34. doi:10.1056/NEJMoa1513749. ↩
- Nanda R, Liu MC, Yau C, et al. Effect of pembrolizumab plus neoadjuvant chemotherapy on pathologic complete response in women with early-stage breast cancer (I-SPY 2). JAMA Oncol. 2020;6(5):676–684. doi:10.1001/jamaoncol.2019.6650. ↩
Last reviewed: 2026-06-04. Researcher-layer synthesis page. Evidence grades follow the GRADE-adapted rubric defined at the top of this page.