Survival analytics
Kaplan–Meier curves, number-at-risk tables, Cox models, hazard ratios, log-rank tests, RMST and censoring diagnostics for adaptive clinical trials.
Open Survival AnalyticsBayesian decision copilot · synthetic sandbox
Adaptive trials, meet Bayesian intelligence.
Adaptrials is a human-in-the-loop decision workspace for biostatisticians and clinical biometrics teams. Explore adaptive designs, simulate interim looks and review posterior evidence in a synthetic sandbox. Production use requires sponsor validation, prospective approval and controlled deployment.
- ICH-E9(R1) estimands
- Human authorization gates
- Seeded synthetic simulations
- CDISC-oriented data model
Trial ADAPT-204 · Phase II oncology
4-arm Bayesian adaptive randomization · IA2 of 3
Posterior probability density · P(Δ | data)n = 16
PriorLikelihoodPosteriorEff. region (Δ > 0.30)
Posterior mean0.37P(efficacy)91.6%95% credible interval[0.27, 0.47]
Explore the update live
Observed effect (x̄)0.34
Interim sample size (n)16
Prior strength (pseudo-n)9
P(Δ > 0.30)
91.6%
posterior prob.
Decision
Continue accrual
PP ≥ 0.95 threshold
Randomization
2 : 1 : 0.4 : 0
adapted to posterior
From design assumptions to a human-authorized decision.
Explore the analytical engines behind adaptive clinical-trial planning. Change an assumption, inspect the evidence, and see where qualified review remains mandatory.
Automatically advancing every 2 seconds; pauses while you interact.
Survival Analytics & Kaplan–Meier Engine
Explore synthetic time-to-event evidence with Kaplan–Meier estimates, at-risk tables, Cox hazard ratios, log-rank tests, RMST summaries, censoring review and proportional-hazards diagnostics.
Kaplan–Meier overall survival
Stratified log-rank test · Cox model · Greenwood confidence intervals
Number at risk0m6m12m18mADP-401320251181113Control32023514965
Proportional-hazards assumption needs review
Schoenfeld residual test p=0.012; inspect a time-varying effect and RMST sensitivity before interpretation.
- Hazard ratio
- 0.74
- 95% CI
- 0.58–0.95
- Log-rank p
- 0.018
- RMST difference
- +2.1 mo
Decision support only. Qualified statisticians and authorized governance bodies retain responsibility. This public workspace uses synthetic examples and does not authorize a clinical action.
One connected statistical workspace
Each engine has a distinct purpose, but its assumptions, diagnostics and outputs remain connected to the protocol, SAP, interim charter and review record.
SAP Studio
A connected SAP authoring workspace for estimands, analysis populations, missing-data methods, interim rules, survival analyses, mock TLFs and versioned review evidence.
Open SAP StudioSample size
Clinical-trial sample-size and power assistance with sensitivity analysis, attrition adjustment, information fractions and simulation-ready adaptive-design assumptions.
Open Sample SizeInterim analysis
Interim-analysis planning with alpha spending, efficacy and futility boundaries, conditional power, predictive probability, DMC access separation and human authorization.
Open Interim AnalysisOutcomes
Transparent decision support for design teams.
Illustrative synthetic benchmarks for product exploration — not clinical validation, regulatory performance claims or a substitute for sponsor qualification.
90%
structured design workspace
estimands, priors and adaptation rules
32%
seeded simulations
inspect power, type I error and expected n
<2s
reviewable posterior updates
assumptions and thresholds remain visible
100%
versioned decision records
human rationale captured at each step
Governed decision architecture
AI recommends. Qualified humans decide.
The Adaptrials decision engine separates autonomous computation from human authorization. Every recommendation carries uncertainty, alternatives and provenance before a reviewer can approve, reject, modify or request re-analysis.
Scroll the complete 12-stage pipeline
Selected stage · 07
Human-in-the-Loop
A qualified statistician or authorized governance body must interpret context and control the next step. The stage remains visible in the decision record so a reviewer can reconstruct what the system knew and which control applied.
Current dispositionAwaiting qualified review
Mandatory human decision gate
Choose a disposition to preview how the accountable reviewer’s action is captured. These controls are a product demonstration; clinical decisions remain sponsor-governed.
Human accountability is the control. The recommendation can be rerun; the disposition, rationale and authorization cannot be silently replaced.
AI support · human authorityPlatform capabilities
Everything a design decision has to survive
Adaptrials runs the full design lifecycle — authoring, governance, statistics and evidence — so a trial decision arrives with the record that justifies it.
26-module design workflow
Clinical strategy through to regulatory evidence, each module carrying its own status, version history, approvals and comments.
Open the workflow5 roles, enforced server-side
Biostatisticians author and submit; DMC members and clinical leads approve. Permissions are checked on every action, not just hidden in the UI.
Sign in to try a roleBayesian engine (demo-qualified functions)
Core beta-binomial, posterior and decision-rule functions are covered by reference-value tests in this demo. They are not a production validation package.
Run an interim analysisSeeded simulation
Seeded Monte Carlo demonstration scenarios. Reproducible results do not by themselves establish clinical validity or operating-characteristic acceptance.
Simulate a designVersioned history (demo)
Every save writes a new version with an attributed reason in the workspace. Production retention, immutable storage and change control require sponsor qualification.
See module historyReviewable audit events (demo)
Authentication, edits, submissions and governance decisions are recorded with actor, role, before/after state and rationale. Production audit-trail security, retention, access review and inspection export require validated controls.
View the audit trail
The Adaptrials workflow
From estimand to reviewable decision, in four moves.
Each step produces a versioned workspace record for human review. In a regulated deployment, the sponsor must lock the protocol and SAP, validate the software, and document monitoring, access, change control and decision rationale.
Average cycle~6 days · end-to-end
01
Specify
Encode your estimand & prior
Declare treatment effects, intercurrent events and priors in the design workspace. The demo captures a protocol/SAP-ready specification; sponsor review and controlled document generation remain required.
ICH-E9(R1) estimandsSkeptical priorsADaM shell
02
Simulate
Stress-test adaptation rules
Run seeded demonstration operating-characteristic scenarios and inspect type I error, power, expected sample size and decision boundaries. Production thresholds require independent programming and approval.
10k trial simsOC dashboardType I / power trade-offs
03
Adapt
Update randomization live
The copilot can illustrate posterior updates and candidate adaptations. It never authorizes or executes a clinical action; qualified reviewers control any decision and implementation.
BAR updatesDrop futile armsSSR triggers
04
Decide
Assemble reviewable evidence
Assemble a review package with assumptions, results, rationale and version history. Submission artifacts, electronic signatures and validation evidence require a qualified sponsor/CRO process.
Posterior reportTLF galleryBriefing book export
Every step emits a reviewable record — protocol inputs, simulation settings, candidate recommendation and decision rationale — versioned for human review. Part 11 qualification and e-signatures require controlled deployment.
Live sandbox · no sign-up
The Bayesian cockpit for adaptive trials.
Drag the observed effect size. Watch the illustrative posterior, decision boundary and randomization ratios update in real time. This synthetic sandbox does not represent a live trial or authorize an adaptation.
Trial ADAPT-204 · 4-arm Bayesian
Posterior mean Δ̂
0.465
95% CrI ± 0.206
P(Δ > 0.30)
0.942
efficacy probability
P(Δ ≤ 0)
1.000
futility probability
Enrolled / planned
184 / 312
59% · IA2 of 3
Observed effect size Δ̂0.55
Subjects enrolled (n)184
Posterior density · P(Δ | data) · Normal–Normal conjugate updateΔ̂ = 0.465 · σ̂ = 0.105
Update rule: posterior precision = prior precision + data precision. This deterministic demo recomputes the illustrative posterior; a controlled production workflow must independently review each interim data sync and decision.
Copilot recommendation
Continue enrollment
P(efficacy)
0.942
P(futility)
1.000
Decision Copilot
The Bayesian agent that reasons with you.
Ask about the synthetic trial — interim logic, prior sensitivity and regulatory concepts. The copilot shows an illustrative reasoning trace; qualified humans must verify the source data, assumptions and any proposed action.
Reasoning traceCited to protocolADaM-groundedSuggests next moves
Adaptrials Copilot
Trial context · ADAPT-204 · Phase II oncology
Try
Press Enter to send · Shift+Enter for newline · deterministic demo responses; verify before use
Built for the biometrics team
One platform. Four audiences. Zero compromise.
Adaptrials is shaped from the day-to-day reality of each persona on the biometrics team — from the biostatistician drafting the SAP to the CRO bid lead simulating a sponsor's design space.
For biostatisticians
Skip the SAS boilerplate. Ship Bayesian evidence.
Adaptrials gives biostatisticians a structured workspace for priors, estimands and adaptation rules. The copilot accelerates exploration and evidence assembly; qualified statisticians remain responsible for the analysis, interpretation and controlled production code.
8hsaved per interim analysis
What you get
- Typed estimand & prior DSL with live type-checking
- CDISC-oriented dataset and TLF planning fields
- Analysis specifications for R / SAS / Python hand-off
- Posterior predictive check gallery (visual + tabular)
- Human-reviewed reproducibility checklist per interim look
- Versioned workspace history for SAP revisions
Regulatory mapping & validation boundary
Designed for regulated development — subject to qualification.
The workflow maps design concepts to FDA, EMA and ICH expectations and makes the evidence chain visible. This local sandbox is synthetic and is not a validated GxP system; sponsor procedures, qualification, security controls and independent review are required before regulated use.
FDA Adaptive Design Guidance — current (2019) final
Design reference only: the workflow exposes the simulation, pre-specification and decision-rationale concepts discussed in FDA's adaptive-design guidance. A sponsor must perform its own protocol/SAP review, simulation validation and regulatory engagement.
Guidance referencePre-specificationBayesian & complex designsIndependent simulation review
ICH E9(R1) Estimands & Sensitivity
The M02 workspace provides structured fields for the ICH E9(R1) estimand attributes and sensitivity-analysis planning. It does not automatically prove estimand consistency or generate validated TLFs; qualified statisticians must review the specification.
ICH E9(R1)Treatment policyCompositeWhile-on-treatmentPrincipal stratum
ICH E6(R3) Good Clinical Practice (2025)
The workflow is designed to support quality-by-design, risk-proportionate oversight and critical-to-quality thinking described in ICH E6(R3). It is a planning aid, not evidence that a sponsor's GCP system, vendor controls or study conduct is compliant.
ICH E6(R3) — 2025Quality by DesignRisk-proportionateCritical-to-quality data
21 CFR Part 11 — Electronic Records & Signatures
The demo records basic actor, role, version and rationale fields. It does not provide cryptographic signatures, immutable/WORM retention, validated e-signature manifestations or an inspection export; those Part 11 controls require a qualified deployment.
Control boundaryHuman approvalValidation requiredNo e-signature in demo
CDISC SDTM & ADaM (current versions)
The workspace uses CDISC-oriented labels and planning fields for SDTM/ADaM deliverables. It does not bundle validated dataset generation, define.xml production, Pinnacle 21 execution or a one-click submission package.
CDISC-orientedSDTM/ADaM planningGeneration not bundledValidation required
EMA Bayesian Methodology & Audit Trail
The workspace makes prior assumptions, posterior summaries and convergence fields visible for human review. This is conceptually aligned with transparent statistical practice; it is not an EMA qualification or a substitute for controlled evidence and inspection-ready records.
Pre-specified priorsTransparent posteriorConvergence diagnosticsQualification required
Aligned with
FDA·EMA·ICH-E9(R1)·ICH-E6(R3)·21 CFR Part 11·CDISC SDTM·CDISC ADaM
Alignment with the guidance above is a design property, not a claim of certification. Adaptrials holds no SOC 2, ISO 27001, HIPAA or GxP certification yet; a certification programme is under consideration and preparation. Security & Compliance
Knowledge Center
See all resourcesField notes from the Bayesian frontier.
White papers, guides, and playbooks written by biostatisticians, for biostatisticians. No fluff — just the math, the regulatory framing, and the worked examples.
Two ways to start
Design your first adaptive trial in under an hour.
Spin up a sandbox trial with synthetic Phase II data, or book a regulator walkthrough with a biostatistician on our team. No procurement call, no enterprise contract, no SOW — start now.
14-day full-feature sandboxNo credit card requiredExport to ADaM, define.xml, QuartoSOC 2 Type II · ISO 27001