Senior Biostatistician — FDA-Oriented AI Medical Device / Diagnostic Validation Study

Senior Biostatistician Needed — Biomedical Clinical Research - AI Software / Medical Device / Diagnostic Validation Study Job Description Seeking an experienced senior biostatistician or boutique biostatistical consulting group to support Phase I evidence development for an AI-enabled medical software program in pancreatic endoscopic ultrasound (EUS). The project analyzes archived content to generate a examination-completeness assessment. The system produces outputs such as complete, incomplete, indeterminate, and non-assessable, then aggregates these into an auditable quality metric. We are seeking a hands-on statistical expert—not a general data analyst—to provide independent design review, formal statistical-analysis planning, and potentially subsequent analysis support. Experience with FDA-facing medical devices, SaMD, diagnostic performance, imaging AI, reader studies, and correlated clinical data is strongly preferred. This initial engagement should be structured as a defined, milestone-based consulting project. The successful consultant may be retained for analysis execution, publication support, FDA pre-submission preparation, and Phase II multi-site validation planning. Initial Scope The selected consultant will review our current Phase I research strategy and help finalize a rigorous, pre-specified statistical framework. Expected work includes: - Review the clinical validation design, endpoint definitions, estimands, and analysis populations. - Advise on diagnostic-performance analyses for segment-level outputs against blinded expert-reader consensus. - Confirm or refine methods for per-segment sensitivity, specificity, PPV, NPV, and 95% confidence intervals. - Develop an approach to account for correlation among segments, frames, readers, operators, platforms, and studies/patients. - Review and refine the planned use of study-level clustered bootstrap confidence intervals. - Advise on inter-reader agreement, algorithm-to-consensus agreement, consensus adjudication, and prevalence-sensitive agreement metrics, including Cohen’s kappa and Gwet’s AC1. - Review the blinded-reader study design, including case allocation, balanced rotation, reader training, adjudication, counterbalancing, washout, and paired assisted-versus-unassisted feasibility-study methods. - Develop or review sample-size and precision simulations for a retrospective held-out validation dataset and a future reader study. - Pre-specify handling and reporting of indeterminate, non-assessable, paused, out-of-operating-envelope, artifact-affected, and missing-data cases. - Advise on analysis of generalizability across operators, endoscopy platforms, indications, and imaging conditions. - Review proposed mixed-effects models for descriptive operator/platform/context effects. - Develop a formal Statistical Analysis Plan (SAP). - Develop mock tables, listings, and figures appropriate for grant reporting, publication, investor diligence, clinical collaboration, and future FDA interaction. - Ensure all methods, assumptions, analysis code, and outputs are reproducible and sponsor-owned. Current Study Context The initial study is a retrospective, multi-platform organ validation program. Key characteristics include: - Corpus of approximately 600 distinct de-id patient studies. - Patient-disjoint dataset partitions, including training, calibration, and a held-out validation set. - Blinded independent scoring by board-certified clinicians. - Each held-out case scored by at least two clinicians, with a defined subset scored by three readers to estimate pairwise agreement. - A pre-specified consensus/adjudication process. - Primary focus on estimation precision and confidence intervals, rather than hypothesis testing alone. - Later feasibility work involving matched, counterbalanced, assisted-versus-unassisted. - Planned reporting of model scope, failures, artifacts, uncertainty, and indeterminate outputs—not just favorable performance metrics. Required Qualifications Please apply only if you have substantial experience in several of the following areas: - REQUIRED: Biostatistics for FDA-facing medical devices, diagnostics, IVDs, SaMD, or clinical AI/ML products. - PREFERRED: Diagnostic-test accuracy and performance evaluation; Medical-imaging, endoscopy, ultrasound, radiology, pathology, or other image/video-based clinical studies.; Multi-reader/multi-case, reader-performance, or human-factors reader studies.; Correlated or hierarchical clinical data, including clustered bootstrap methods, mixed-effects models, GEE, repeated measures, or nested data structures.; Inter-rater agreement and reliability methods, including Cohen’s kappa, Gwet’s AC1/AC2, ICC where applicable, and prevalence effects.; Sample-size and precision simulation for diagnostic-performance or reader studies.; Statistical Analysis Plan development, pre-specified estimands, mock TLFs, and regulatory-quality documentation.; FDA pre-submission support, 510(k), De Novo, PMA, or equivalent medical-device evidence programs.; Reproducible statistical programming in R, SAS, Python, or another validated environment.; Preferred Qualifications - Direct work on AI/ML-enabled medical-device, imaging-AI, computer-aided detection/diagnosis, or clinical-video analysis programs. - Experience with MRMC methods, ROC/AUC analysis, JAFROC, Obuchowski-Rockette, Dorfman-Berbaum-Metz, or comparable reader-study approaches. - Experience supporting NIH research projects, SBIR/STTR projects, NCI-funded studies, early-stage MedTech companies, or investigator-initiated validation studies. - Experience distinguishing exploratory engineering analyses from locked clinical-validation analyses. - Comfortable serving as a named consulting biostatistician for grant, protocol, publication, or regulatory materials. Deliverables The initial engagement is expected to produce: - A written statistical-design review identifying material risks, gaps, and recommended changes to the current validation plan. - A finalized statistical framework, including primary and secondary estimands, analysis populations, endpoint definitions, and handling rules for difficult cases. - A formal Statistical Analysis Plan suitable for a clinical-AI/medical-device validation program. - Sample-size and/or precision simulations supporting the held-out validation set and future reader study. - Draft shells for tables, listings, and figures. - A reproducible analysis-plan outline, including proposed software environment, data specifications, code-validation approach, and sponsor ownership/transfer requirements. - Participation in a limited number of technical meetings with the internal team. - Analysis execution, final reporting, publication support, and Phase II design would be scoped separately after successful completion of the initial engagement. Proposal Requirements Please include all of the following in your proposal: - Your CV or LinkedIn profile, including education and statistical credentials. - The name and CV/biography of the specific senior statistician who would lead the work. - Two to three directly comparable projects involving medical devices, diagnostics, clinical AI/ML, imaging, reader studies, or FDA-facing evidence generation. Please do not disclose confidential client information. - NIH biosketch if you have one - A brief description of how you would approach: Correlated segment-level diagnostic-performance data; Blinded reader consensus and inter-reader agreement; Indeterminate or non-assessable model outputs; Precision-based sample-size planning for the proposed held-out validation set. - Your availability over the next 3–6 months. - Confirmation that we will receive all final work products, analysis plans, code, documentation, and intellectual-property rights related to the engagement.