Loading...
A prospective workflow study at a top-tier NCI-designated cancer center accelerator program, co-designed with a former pharma CMO. A 1,200-chart retrospective accuracy study at a US NCI-designated comprehensive cancer center. A postmarket pharmacovigilance pilot launching 2027. Pre-registered endpoints. Peer-reviewed target venues. External co-authorship. The validation roadmap is the procurement asset.
Target venues: JCO, JAMIA, JCO Clinical Cancer Informatics, Annals of Oncology. Pre-print on medRxiv within 60 days of database lock. Independent IRB or HREC oversight on every prospective study. Two patents filed. CancerX member. FDA EDSTP application in review. 25 advisors across 5 councils.
The validation roadmap is the moat.
In flight. A prospective workflow study, a 1,200-chart retrospective accuracy study, and a postmarket pharmacovigilance pilot launching 2027.
JCO, JAMIA, JCO Clinical Cancer Informatics, and Annals of Oncology. Pre-print on medRxiv within 60 days of database lock on every study.
Declared up front. Pre-registered endpoints, peer-reviewed venues, IRB or HREC oversight, conservative reporting, external co-authorship, pre-print within 60 days, and failed studies still publish.
Every Burna validation study has pre-registered endpoints, peer-reviewed target venues, independent oversight, conservative reporting, external co-authorship, a medRxiv pre-print within 60 days of database lock, and a commitment that failed studies still publish. Declared before the first data point is collected.
Primary and secondary outcomes are declared in writing before data collection begins. Endpoints do not move once data collection starts. Every prospective study is published as a design document before kickoff.
Studies are designed to JCO, JAMIA, JCO Clinical Cancer Informatics, and Annals of Oncology requirements, not to marketing tolerances. The target journal shapes the protocol, not the other way around.
Independent IRB or HREC review on every prospective study. Data monitoring plans filed before launch. The external review is a structural property of the study, not a procedural step that can be skipped.
Studies report what was found, not what was hoped for. Specific accuracy percentages remain under NDA until peer-reviewed publication. The external framing is "strong agreement with expert clinicians in ongoing internal testing." The peer-reviewed manuscript is where the numbers land.
Site collaborators are scientific partners, not vendor testimonial fodder. Co-authorship is first-author eligible. The paper is a partnership asset; public naming follows written approval and confidentiality terms.
medRxiv pre-print within 60 days of database lock is standard. Conference posters are a summary, not a substitute. The science is in the public record on a timeline that the field can plan against.
A study that does not meet its pre-registered endpoints publishes as a negative result. The clinical AI field has a file-drawer problem. Burna does not contribute to it. Non-negotiable.
Each was scoped against a publication target before kickoff. Each is structured for independent review.
JCO Clinical Cancer Informatics, 2026 to 2027. Pre-print on medRxiv within 60 days of database lock.
Bio-IT World 2026 poster (April). JCO or JCO CCI full manuscript, 2026 to 2027. Pre-print on medRxiv within 60 days of database lock.
Regulatory science journal (TBD), 2027 to 2028.
“The validation study endpoints I co-designed for Burna are the endpoints I would have wanted to see as a sponsor's CMO before pilot kickoff. Pre-registered, peer-reviewed target venues, paired-grader timing, multi-drug attribution agreement. The science is the procurement asset.”
Dr. Andrea PirzkallFormerly Chief Medical Officer at Replimune and Asher Biotherapeutics; Executive Director Clinical Development at BeiGene; Principal Medical Director at Genentech.
Conference posters precede full manuscripts (Bio-IT World 2026 for the retrospective accuracy study; ASCO, ESMO, EORTC abstract submissions for downstream readouts). Pre-prints on medRxiv land within 60 days of database lock on every study.
Specific accuracy percentages, kappa coefficients, and inter-rater agreement numbers are not published on this page. The peer-reviewed manuscript is where the numbers go.
Conservative reporting until peer review supports the claim. This is not a marketing constraint. It is the standard the field expects of any clinical AI platform claiming evidence.
The methods that make the studies independently reviewable. Each card below is a structural commitment, baked into every protocol. Deeper detail in the Validation Methodology Brief under NDA.
Every prospective study has a published design document filed before data collection begins. The document covers endpoint definitions, statistical analysis plan, inclusion and exclusion criteria, adjudication procedures, and pre-specified subgroup analyses.
Filed prior to kickoff · locked under version control
Coordinators and expert graders work the same cases. Inter-rater patterns are reported by System Organ Class, MedDRA Preferred Term, and CTCAE criterion. The disagreement analysis is part of the manuscript.
SOC · MedDRA PT · CTCAE criterion reporting
The institution running each study is the ethics-review authority, not Burna. Data monitoring plans filed before launch. External review is a structural property of the protocol, not an audit applied at the end.
IRB / HREC review · DSMB where required
Model freeze at study start. Validation-dataset text is never sent to a model that has seen or will see that text during training. Architectural separation, not procedural separation.
Model freeze · held-out architectural boundary
Independent third-grader review where the protocol requires it. Adjudication outcomes are reported in the manuscript; they do not retroactively change the inter-rater agreement on the original two-grader comparison.
Third-grader resolution · original kappa preserved
The clinical AI field has a file-drawer problem. Burna does not contribute to it. This is named in the study protocol, the pilot agreement, and the publication contract with every site partner.
Contractually binding · medRxiv pre-print within 60 days
25 advisors across 5 councils. The composition of the advisory bench is itself part of the validation roadmap: senior pharma CMOs, former Celgene and Takeda executives, founding chairs of biomedical informatics, and CRO operators review the study protocols, the endpoint design, and the publication path.
See the full advisor bench →Dr. Andrea Pirzkall (formerly CMO at Replimune and Asher Biotherapeutics, Executive Director Clinical Development at BeiGene, Principal Medical Director at Genentech). Dr. Michel Azoulay (Strategic Advisor to the CEO; Exec Lead for Protocol Safe). Dr. Tehseen Salimi (Global CMO at Fresenius Medical Care).
Dr. Howard Fingert (formerly VP Medical Oncology at ONO Pharma USA; Senior Medical Director at Takeda; trained at Dana-Farber and MGH).
Dr. Joel Saltz (SUNY Distinguished Professor; founding chair of Biomedical Informatics at Stony Brook). Dr. Mary Morison Saltz (Chief Clinical Integration Officer at Stony Brook Medicine; led the N3C national data initiative).
Dr. Stefan Gluck (Lead, formerly VP at Celgene). Presenting Burna at AI in Oncology Germany, ESMO Breast Cancer, and ASCO 2026.
Scott Schliebner (COO at P1 Pratia Oncology, Lead). Dr. John Hayslip (CMO at Verastem Oncology, formerly Director of Clinical Research at Markey Cancer Center).
Spanning clinical operations, clinical trials, pharma, CRO and site networks, regulatory, informatics, governance, product, implementation, brand, and commercial leads. Advisors are listed in their personal capacity. Affiliations are disclosed for context, not to imply institutional endorsement.
The eight we hear most often. The cross-site FAQ has more.
Read the full FAQ →Conservative reporting until peer review supports the claim. Accuracy percentages, kappa coefficients, and inter-rater agreement numbers are available under NDA to pilot scoping conversations, pharma sponsor diligence, regulatory advisors, and investors in diligence. The peer-reviewed manuscript is where the numbers land on the public record.
Published. Every time. A cohort, a regimen, a language, or a workflow that does not perform as hypothesized is reported. The clinical AI field has a file-drawer problem. Burna does not contribute to it. The commitment is in every study protocol, every pilot agreement, and every publication contract.
When a competitor publishes methodology that can be replicated, yes. Where competitors have not published, Burna's endpoints are reported against the best available public benchmark from the peer-reviewed literature (Hong et al., 2020 on inter-rater agreement; Hillman et al., JCO 2010 on attribution change rates; FDA-NCI 2019 workshop on current-state characterization).
Yes. Independent validation is welcome. Pilot agreements include a clause offering site partners first-rights to publish results independently, co-authored, or with Burna as a listed party. The science is a partnership asset, not a marketing asset.
Every validation dataset is held out from training. Model freeze at study start. Validation-dataset text is never sent to a model that has seen or will see that text during training. The separation is architectural, not procedural.
Public institution naming follows written approval. The accelerator program is a top-tier NCI-designated cancer center; Dr. Andrea Pirzkall co-designed the endpoints. The institution names publicly when the manuscript publishes.
The CDER Emerging Drug Safety Technology Program is the FDA mechanism for upstream safety infrastructure of exactly this category. Burna's application is in review. EDSTP eligibility is not FDA validation or FDA approval; it is structured regulatory engagement. The 2019 FDA-NCI workshop on adverse event attribution established the institutional appetite; EDSTP is the mature program response.
Email security@burna.ai for the NDA template, or book a 30-minute conversation and the brief follows the call. Briefs arrive within 48 hours of executed NDA.
“The validation study endpoints I co-designed for Burna are the endpoints I would have wanted to see as a sponsor's CMO before pilot kickoff. The science is the procurement asset.”
Dr. Andrea PirzkallFormerly Chief Medical Officer at Replimune and Asher Biotherapeutics; Executive Director Clinical Development at BeiGene; Principal Medical Director at Genentech.
The Validation Methodology Brief is a per-study document covering pre-registered endpoints, sample size and power calculation, statistical analysis plan, target publication venue, and readout timeline. Available under NDA.
For pilot partners, pharma sponsors, regulatory advisors, and investors. Covers study designs, endpoint rationale, and publication path. The Methodology Brief follows the call.
For diligence at your pace. Available under NDA for each of the three studies. Pre-registered endpoints, statistical analysis plan, target publication venue, and readout timeline.
Validation readouts, design partner cohort updates, and conference timing only. No tracking beacons, no marketing email cadence.
Engine architecture at /engine. Investment thesis at /investors. Cancer center pilot structure at /cancer-centers. Pharma positioning at /pharma. Procurement-grade architecture and sovereignty at /it-and-security.