NESS 2026 · Full-day short course · Hybrid (in-person + online)

Bayesian Statistics (Bayesics), Its History, Rationale, and Future in Clinical Trials

A one-day journey from Bayes' Theorem to the FDA's January-2026 Bayesian draft guidance — built for statisticians and clinical-trial professionals who want to understand what Bayesian inference actually does for drug development, not just memorize how to run it.

Motivated by the recent release of FDA's draft guidance on the Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products (12 January 2026), this short course gives a concise yet rigorous introduction to Bayesian statistics — what we call Bayesics — covering its history, its rationale, and its current and future role in clinical trials.

InstructorYuan Ji, PhD · University of Chicago
DateWednesday, 27 May 2026
Time9:00 AM – 5:00 PM ET
FormatFull day · McHugh Hall + online
Registration is open through the NESS 2026 symposium portal. Seats are limited, and the course is hybrid — attend live at McHugh Hall or join from anywhere online. A course completion certificate is provided.
Register at NESS 2026 →

01Why this course

FDA Jan-2026 draft guidance Project Optimus dose optimization BESS sample-size estimation C3TI & tipping-point analysis Hierarchical borrowing Platform / adaptive trials

Bayesian statistics rest on two probability statements — one for the data (the likelihood, shared with frequentist inference) and one for the parameter (the prior, unique to Bayesian). Through Bayes' Theorem we get the posterior — a direct probability statement on the question a clinician or regulator actually wants answered: given what we observed, how likely is it that this drug helps?

This short course consists of three parts: (1) Bayesian thinking — the philosophy and human-logic intuition behind Bayes' Theorem; (2) Bayesian modeling and computation — priors, likelihoods, posteriors, MCMC, and what to do when models outgrow conjugacy; and (3) Bayesian clinical trials — early-phase dose finding and optimization, late-phase sample sizing and decision making, and the practical sponsor workflow under the new FDA draft guidance.

Examples include questions every working biostatistician encounters: what is the difference between type I error and type I error rate? How does Bayes quantify the chance of a wrong decision? When does c = 0.975 not mean what you think it means? The emphasis is on "why" rather than "how" — attendees should leave able to defend or critique a Bayesian SAP in a sponsor or regulatory meeting.

Why this course, why now

The FDA released its first comprehensive Bayesian draft guidance for drugs and biologics on 12 January 2026. Sponsors who can write a regulator-ready Bayesian SAP — pre-specified priors, robust safeguards, operating characteristics by simulation, computational reproducibility — will be the ones moving programs fastest in 2026 and beyond. Project Optimus, C3TI, and the rapid growth of hybrid and platform designs make Bayesian fluency a practical advantage, not a theoretical curiosity.

02By the numbers

170+
peer-reviewed publications by the instructor across medical and statistical journals
5
named Bayesian designs invented by the instructor (mTPI, i3+3, BESS, ADOPT, PEDOOP)
200+
clinical trials worldwide using i3+3 or mTPI-family dose-finding designs
3
major regulators where these designs have been accepted: FDA, NMPA, EMA
8 hr
of intensive content across 4 sessions, 2 coffee breaks, and a 75-minute lunch
4
worked real-world case studies pulled from FDA-reviewed submissions

03Who should attend

Industry & Regulatory

Pharma / biotech biostatisticians, clinical pharmacologists, clinical scientists, regulatory affairs, and medical writers working on adaptive trials, dose-optimization studies, hybrid / platform designs, or sponsor briefings to FDA, NMPA, or EMA. Calibrated to give a senior biostatistician something new while remaining accessible to a first-year industry hire.

Academic & Students

Graduate students (MS / PhD) in statistics, biostatistics, pharmacometrics, and clinical-trials methodology, plus faculty who want a current view of regulator-aware Bayesian practice. Background in graduate-level probability and inference is assumed; no prior clinical-trials experience is required.

04What you will leave with

A working Bayesian vocabulary. Confidently distinguish posterior probability, predictive probability, assurance, and credible interval — and know when each one belongs in a clinical context.
Philosophical clarity. Know the difference between type I error and type I error rate, and why posterior probability is the natural language for clinical and regulatory decisions.
A modern design playbook. Recognize when Bayesian dose finding (i3+3, BOIN, mTPI-2), dose optimization (ADOPT, backfill + randomized comparison, PEDOOP), and BESS sample sizing are the right tools.
Regulatory fluency. Walk through what FDA's January-2026 Bayesian draft asks for — pre-specified priors, robust safeguards, operating characteristics by simulation, computational reproducibility.
Concrete case studies. Read Phase III hierarchical-borrowing examples (Rebyota / PUNCH CD3) and Phase I optimization examples (Project-Optimus-aligned ADOPT) the way a regulator would.
An opinion of your own. The course teaches the “why”, not a recipe. You should leave able to defend (or push back on) a Bayesian design proposal in your next protocol-review meeting.

05Detailed course outline

The day is divided into four 90-minute sessions, mapped to the three-part structure above. Each session opens with a short compass slide (what we're solving and why), uses one real case study, and closes with a concrete take-away.

Session 1

Part 1 — Bayesian thinking: the body, the wings, and Bayes' Theorem

9:00 – 10:30 · 90 min

Why we are all Bayesians (whether we admit it or not), why classroom statistics lacks the probability the clinician actually needs, and how Bayes' Theorem closes the gap.

1.1Logics — the body of statistics

  • What is statistics? Why probability is its main tool
  • The two probability measures: Pr(data | parameter) vs. Pr(parameter | data)
  • Why classroom (frequentist) statistics does not have the probability we actually want

1.2The Bayesian paradigm

  • Bayes' Theorem in plain English and in symbols
  • Posterior inference: median, credible interval, and Pr(δ > a | data)
  • Worked examples on a binomial trial and a normal mean
  • How Bayesics aligns with how humans actually make decisions

1.3Type I error vs. type I error rate

  • What each one assumes (and which question each one answers)
  • How Bayesians quantify the chance of a wrong decision
  • Common criticisms of Bayesian inference — and what they get right and wrong

1.4Where Bayesics lives in clinical practice

  • A short history: Bayes (1763) → Laplace → modern regulatory acceptance (FDA CDRH 2010, ICH E9, FDA 2026 draft)
  • Map of where Bayesian methods are routine today: dose finding, sample sizing, interim decisions, subgroup borrowing
Worked example: Re-analysis of a fragile Phase III oncology trial 2 × 2 table — frequentist p sweeps past 0.05 while the posterior Pr(δ > 0 | data) stays above 0.95. Same data, two stories.
Take-away: A clear mental model of when a Bayesian analysis genuinely changes the conclusion, and when it just rephrases the frequentist one.
Session 2

Part 2 — Bayesian modeling and computation

10:45 – 12:15 · 90 min

From priors and likelihoods to posteriors that scale. Conjugate models for intuition, then MCMC for everything else, and what to check before you trust the output.

2.1Priors, likelihoods, posteriors

  • Conjugate families: Beta-Binomial, Normal-Normal, Gamma-Poisson
  • Weakly informative priors, informative priors, and elicitation
  • Sensitivity grids: optimistic / neutral / skeptical priors

2.2Bayesian computation

  • Monte Carlo integration: the workhorse of applied Bayes
  • Markov chains in 15 minutes — stationarity, irreducibility, aperiodicity, ergodicity
  • Metropolis-Hastings, Gibbs sampling, and modern HMC (Stan, PyMC)
  • Convergence diagnostics — R-hat, effective sample size, trace plots

2.3Hierarchical and borrowing models

  • Why “borrowing strength” is a Bayesian native, not an afterthought
  • Power priors, MAP, robust MAP, commensurate / hierarchical priors
  • Effective sample size of a prior — and how to communicate it to a clinician or a regulator

2.4Modeling pitfalls and pragmatic shortcuts

  • When conjugate intuition is enough vs. when MCMC is mandatory
  • Reading a posterior plot critically: shape, mass, multimodality
  • How much computation is “enough”? Convergence vs. business deadline
Worked example: A robust MAP prior built from three historical control arms, down-weighting automatically when the new trial's control behaves differently than expected — the safeguard the FDA Jan-2026 draft explicitly asks for.
Take-away: A working understanding of when conjugate models suffice, when to reach for MCMC, and how to read a posterior plot critically.
Session 3

Part 3a — Early-phase Bayesian designs (Project Optimus)

13:30 – 15:00 · 90 min

Phase I has shifted from “find an MTD” to “find the optimal dose with desirable efficacy and safety.” We walk through the modern toolset and where regulators stand today.

3.1Phase 1a dose finding

  • Why Phase 1a often does not need a parametric model
  • i3+3, BOIN, mTPI-2 — choosing among them in practice
  • Operating characteristics: how to read them, how to defend them

3.2Phase 1b/2 dose optimization

  • What FDA's Project Optimus actually asks sponsors to deliver
  • Multi-dose escalation, backfill, randomized parallel-dose comparison
  • The ADOPT platform — escalation → expansion → randomized comparison across baskets

3.3Pharmacometrics-integrated dose decisions

  • PEDOOP — fusing each patient's PK / PD signal into the dose-finding model
  • Tolerability on a chronic-dosing timescale, not just the DLT window
  • Endpoint design: safety, efficacy, and the composite-utility view

3.4Practical sponsor workflow for Phase 1b/2

  • A defensible written SAP for a dose-optimization study
  • Pre-spec'd simulation plan and the operating-characteristics report
  • Common reviewer questions — and how to anticipate them
Case study: A U.S. ADOPT trial in oncology that combined dose escalation, backfill, and randomized comparison — successfully reviewed by the FDA and accepted as a primary-analysis Bayesian design.
Take-away: A defensible early-phase plan that produces an indication-by-dose table — what Project Optimus actually asks for.
Session 4

Part 3b — Late-phase Bayesian inference and the sponsor workflow

15:15 – 16:45 · 90 min

How to size a Bayesian study, how to read a subgroup analysis, and how the “Bayesics-first, frequentist-translation” workflow survives a regulatory review.

4.1Bayesian sample size (BESS)

  • BESS — three pillars of Sample-size, Evidence, and Confidence (Bi & Ji, 2024)
  • BESS vs. group-sequential design: sample-size savings on real examples
  • Predictive probability of success (PPoS) as a Go / No-Go tool — including Phase II → Phase III planning

4.2Subgroup analysis and tipping points

  • Bayesian subgroup analysis (BSA) under FDA's C3TI pilot
  • Tipping-point analysis and unification of evidence across subgroups
  • Effective sample size of borrowed information — how much is “too much”?

4.3Posterior cutoff calibration

  • What the posterior cutoff c actually controls — and why c = 0.975 is not a universal answer
  • Three calibration perspectives (Zhou & Ji, 2023): size, assurance, and the range-calibration fix when borrowing matters
  • The LEAP-002 case: Pr(HR < 1) = 0.977 vs. regulatory bar 0.9815 — a 4-thousandths gap on a rigid α

4.4The sponsor workflow and submission checklist

  • “Bayesics-first, frequentist-translation” — a workflow that survives a regulatory review
  • Software landscape: R (rstanarm, brms, RBesT, BOIN, mTPI, i3+3), Stan / PyMC
  • A printable submission checklist for Bayesian trials
Case study: PUNCH CD3 / Rebyota hierarchical borrowing from PUNCH CD2 — the first FDA-approved fecal-microbiota product; how the Bayesian primary analysis was written, sized, and accepted.
Take-away: A workable Bayesian primary-analysis plan plus a printable submission checklist you can take to your next Type-B meeting.
Closing

Open Q&A and follow-up roadmap

16:45 – 17:00 · 15 min

A short open Q&A on submission, software, and where to take this material next. Course materials and code are shared via a public repository after the course.

06Schedule at a glance

Time (ET)BlockTopic
09:00 – 10:30Session 1 Part 1 — Bayesian thinking, Bayes' Theorem, posterior inference
10:30 – 10:45Break Coffee & Q&A
10:45 – 12:15Session 2 Part 2 — Bayesian modeling, MCMC, hierarchical borrowing
12:15 – 13:30Lunch (75-minute break)
13:30 – 15:00Session 3 Part 3a — Early-phase Bayesian designs & Project Optimus
15:00 – 15:15Break Coffee & Q&A
15:15 – 16:45Session 4 Part 3b — BESS, BSA, posterior calibration, sponsor workflow
16:45 – 17:00Closing Open Q&A and follow-up roadmap

07What you'll receive

PDF Full slide deck 200+ annotated slides, available for review and re-use under attribution.
R Annotated code repository R / Stan / PyMC scripts for every worked example, runnable on a laptop.
Submission checklist One-page Bayesian-trial submission checklist aligned to the FDA Jan-2026 draft.
SAP SAP template Bayesian Statistical Analysis Plan template for early- and late-phase studies.
SIM Simulation harness Reusable Monte-Carlo harness for computing operating characteristics.
CERT Completion certificate Official NESS 2026 course completion certificate from the conference organizers.

08Instructor

Yuan Ji, PhD

Professor of Biostatistics · The University of Chicago
Chair, Biopharmaceutical Section, International Society for Bayesian Analysis (ISBA)

Dr. Ji's research focuses on innovative Bayesian methods for translational cancer research and clinical-trial design. He is the author of more than 170 peer-reviewed papers across medical and statistical journals, and the inventor of widely adopted Bayesian adaptive designs including mTPI, i3+3, BESS, ADOPT, and PEDOOP, used in dose-finding and dose-optimization trials worldwide.

His Phase III Bayesian work has been accepted by the FDA, NMPA, and EMA across oncology, infectious disease, and rare-disease programs. He received the Mitchell Prize in 2015 from the International Society for Bayesian Analysis and is an elected Fellow of the American Statistical Association.

Office: 5841 S. Maryland Ave, MC2000, Chicago, IL 60637  ·  Phone: (773) 834-0214  ·  Email: yji@bsd.uchicago.edu

09Selected references

None are prerequisite — they are listed so attendees can go deeper after the course.