# Early-Phase Clinical Pharmacology — Biostatistics Deliverables Pipeline

**Protocol review → CSR draft · every document, dataset, and statistical output · with the hybrid-AI routing overlay**

> **Scope & how to read this.** This is the complete biometrics deliverables map for a Phase 1 clinical-pharmacology study (e.g., First-in-Human SAD/MAD with PK/PD, NCA + popPK, safety). 136 deliverables across five stages, each with its owner, inputs, outputs, governing standard, QC method, and how AI is (or is not) used. The at-a-glance tables are below; the **full register with every field is the companion `.xlsx`**, and the **one-page visual is the `.pdf` map**. Standards are current as of mid-2026 (CDISC SDTMIG v3.4 / ADaMIG v1.3 / Define-XML v2.1; ICH E3, E6(R3), E9/E9(R1), M11; 21 CFR Part 11; ISPE GAMP 5).

## How the pipeline flows

Five sequential stages, each closed by a gate. A **CDISC data-standards track** (SDTM → ADaM → Define-XML + reviewer guides) runs through Stages 3–5, and a **hybrid-AI governance track** (data-classification routing, model-freeze records, engine-of-record register) runs across all five.

| Stage | Focus | Gate |
|---|---|---|
| **S1 — Study Design & Protocol** | Biostatistics shapes the design: estimands, endpoints, sample-size rationale, randomization, and the statistical sections of the protocol. | Protocol approved / finalized |
| **S2 — Planning & Specifications** | Everything is specified before data: the SAP, TLF shells, ADaM/SDTM specs, the randomization schedule, the data-review and QC plans. | SAP & specifications final |
| **S3 — Data & CDISC** | Raw data becomes standardized, conformant, traceable CDISC datasets (SDTM → ADaM) with reviewer guides and Define-XML, under full data integrity controls. | Database lock |
| **S4 — Analysis & QC** | TLFs and PK/PD analyses are produced and independently double-programmed; reported numbers come from validated engines (Phoenix WinNonlin for NCA), never an LLM. | All outputs QC-passed |
| **S5 — Reporting & CSR** | Results become the CSR — statistical methods and results text, in-text tables/figures, appendices, the submission package, and cross-document consistency. | CSR draft delivered |


## S1 — Study Design & Protocol

Biostatistics shapes the design: estimands, endpoints, sample-size rationale, randomization, and the statistical sections of the protocol.

| ID | Deliverable | Owner | Standard | QC | AI |
|---|---|---|---|---|---|
| S1-01 | **Biostatistics input to protocol — design rationale & study design selection** | Lead biostatistician | ICH M11 (CeSHarP protocol template) design section; ICH E9 (design principles, bias control); ICH E6(R3) GCP; FDA/EMA FIH guidance | Independent biostatistician peer review; cross-check design vs objectives/estimands; medical/clin-pharm/regulatory review; protocol team consensus | GREEN |
| S1-02 | **Objectives → estimands → endpoints alignment (ICH E9(R1) estimand framework)** | Lead biostatistician | ICH E9(R1) estimand addendum; ICH E9; ICH M11 (estimand placement in protocol); ICH E3 | Independent biostatistician review of estimand completeness (all 5 attributes, every ICE addressed); clinical & clin-pharm alignment review; consistency check objectives↔estimands↔endpoints↔analysis sets | GREEN |
| S1-03 | **Statistical considerations / analysis-overview section of protocol** | Lead biostatistician | ICH E9 / E9(R1); ICH M11 statistical-considerations section; ICH E3 (consistency with future CSR methods); ICH E6(R3) | Peer review by independent biostatistician; alignment check vs estimands and vs planned SAP scope; cross-functional protocol review | GREEN |
| S1-04 | **Analysis populations / analysis-set definitions (protocol-level)** | Lead biostatistician | ICH E9 / E9(R1); ICH M11; CDISC ADaMIG v1.3 (population-flag conventions, e.g., SAFFL/PKFL); ICH E3 | Independent review of set definitions for mutual consistency and estimand linkage; clin-pharm review of PK-evaluable exclusion rules; traceability check to ADSL flag plan | GREEN |
| S1-05 | **Sample-size / power justification (or documented rationale for no formal sizing in FIH)** | Lead biostatistician | ICH E9 (sample size & assumptions); ICH M11; FDA BA/BE & food-effect guidance; EMA bioequivalence/early-phase guidance; ICH E9(R1) (sizing tied to estimand) | Independent re-derivation/double-check of sample-size or precision calculation (independent code/program run); review of assumptions vs source data; documentation of method and seed/software version | AMBER |
| S1-06 | **Randomization design & specification (incl. sentinel dosing & active:placebo allocation)** | Lead biostatistician | ICH E9 (randomization & blinding); ICH E6(R3) GCP; ICH M11; 21 CFR Part 11 (controlled, audit-trailed list generation); sponsor randomization SOP | Independent biostatistician review of design spec; verification that block/ratio/sentinel logic enforces intended allocation; segregation-of-duties check (spec author ≠ schedule generator); blinding-integrity review | GREEN |
| S1-07 | **DSMB/SRC/SEC charter — statistical decision rules & dose-escalation criteria** | Lead biostatistician | ICH E9 (data monitoring, decision rules); ICH E6(R3) GCP; FDA guidance on Data Monitoring Committees; FIH guidance (escalation/stopping); ICH M11 | Independent review of decision-rule logic for unambiguity and safety conservatism; consistency check vs protocol dose levels and stopping rules; committee/regulatory review; (if model-based) operating-characteristics simulation review | AMBER |
| S1-08 | **Dose-escalation operating-characteristics simulation (design-stage)** | Lead biostatistician | ICH E9 / E9(R1); FDA adaptive-design & FIH guidance; reproducibility per 21 CFR Part 11 (versioned, seeded code); sponsor simulation SOP | Independent re-run/replication of simulation with documented seed & environment; code review; scenario-grid completeness review; sanity-check of OC against design intuition | AMBER |
| S1-09 | **Interim / adaptive-design elements specification (if any)** | Lead biostatistician | ICH E9 / E9(R1); FDA Adaptive Designs for Clinical Trials guidance; EMA reflection paper on adaptive designs; ICH M11; 21 CFR Part 11 | Independent statistical review of adaptation rules & error-rate control; simulation of operating characteristics; firewall/independence review; documentation of 'N/A' decision where no adaptation | AMBER |
| S1-10 | **PK precision / variability assumptions memo (CV%, BLQ/LLOQ & sampling-scheme adequacy)** | Lead biostatistician | ICH E9; FDA/EMA PK and BA/BE guidance; sponsor NCA/PK SOP; reproducible assumptions | Joint biostatistician–PK scientist review; cross-check sampling adequacy vs predicted t1/2; sensitivity of precision to CV% assumptions; documentation of assumption sources | GREEN |
| S1-11 | **Biostatistician protocol review & sign-off** | Lead biostatistician | ICH E9 / E9(R1); ICH M11; ICH E6(R3) GCP; 21 CFR Part 11 (electronic signature & audit trail); sponsor protocol-approval SOP | Comment-tracking & resolution log; independent (second-biostatistician) consistency check on objectives↔estimands↔endpoints; signed/audited approval; amendment version control | GREEN |
| S1-12 | **Emergency unblinding / code-break plan (design-stage definition)** | Lead biostatistician | ICH E6(R3) GCP; ICH E9 (blinding integrity); 21 CFR Part 11 (audit trail of unblinding); pharmacovigilance regs; sponsor unblinding SOP | Independent review of firewall & segregation-of-duties; verification unblinding pathway is auditable and minimally disruptive to the blind; cross-check vs DSMB charter and randomization spec | GREEN |
| S1-13 | **Blinded Sample-Size Re-Estimation (BSSR) Plan — design-stage provision** | Lead biostatistician | ICH E9 / E9(R1); ICH E6(R3) GCP | Biostatistician review and sign-off; no unblinded data touched | GREEN |
| S1-14 | **Estimand-to-design traceability & assumptions log (intercurrent-event strategy register)** | Lead biostatistician | ICH E9(R1) estimand framework | Biostatistician + medical reviewer sign-off | GREEN |

*Stage gate: **Protocol approved / finalized.** 14 deliverables.*

## S2 — Planning & Specifications

Everything is specified before data: the SAP, TLF shells, ADaM/SDTM specs, the randomization schedule, the data-review and QC plans.

| ID | Deliverable | Owner | Standard | QC | AI |
|---|---|---|---|---|---|
| S2-01 | **Statistical Analysis Plan (SAP)** | Lead biostatistician | ICH E9 + E9(R1) estimands, ICH E3 (alignment with CSR), ICH E6(R3) GCP, ICH M11; version-controlled, dated, signed | Senior biostatistician peer review; cross-check vs protocol & estimands; documented sign-off; formal version control of amendments | GREEN |
| S2-02 | **SAP Amendment & Version-Control Log** | Lead biostatistician | ICH E9, ICH E3, ICH E6(R3), 21 CFR Part 11 (audit trail), ALCOA++ | Independent review of each change; timestamp vs lock/unblinding verified; Part 11 audit trail retained | GREEN |
| S2-03 | **Analysis-Set / Population Definitions** | Lead biostatistician | ADaMIG v1.3 (population flags), ICH E9(R1), CDISC CT | Independent reviewer confirms flag logic vs SAP; double-programmed in ADSL; reconciled at blinded data review | GREEN |
| S2-04 | **TLF Shells / Mock-Up Shells (Tables, Listings, Figures)** | Lead biostatistician | ICH E3 (in-text/appendix structure), sponsor output standards, CDISC CT for terminology | Peer/medical-writer/clinical review of shells; titles/footnotes vs SAP; dummy-data render test for layout | GREEN |
| S2-05 | **TLF Inventory / Table of Contents** | Lead biostatistician | ICH E3, sponsor standards | Cross-check vs SAP analyses (no orphan/missing outputs); peer review | GREEN |
| S2-06 | **ADaM Dataset Specifications (ADSL)** | Stat programmer | ADaMIG v1.3, CDISC CT, Define-XML v2.1, traceability to SDTM | Spec peer review; downstream independent double-programming of ADSL; Pinnacle 21 ADaM conformance | GREEN |
| S2-07 | **ADaM Dataset Specification (ADPC — PK Concentrations)** | Stat programmer | ADaMIG v1.3, CDISC CT, Define-XML v2.1, ICH PK reporting practice | Spec review vs PK plan (BLQ/time rules); independent double-programming of ADPC; reconciliation to SDTM PC | GREEN |
| S2-08 | **ADaM Dataset Specification (ADPP — PK Parameters)** | Stat programmer | ADaMIG v1.3, CDISC CT (PK parameters), Define-XML v2.1 | Spec review vs parameter list; ADPP reconciled to validated Phoenix output; independent double-programming of structure | AMBER |
| S2-09 | **ADaM Dataset Specification (ADAE — Adverse Events)** | Stat programmer | ADaMIG v1.3 (OCCDS), MedDRA, CDISC CT, Define-XML v2.1 | Spec review of TEAE logic; independent double-programming; reconcile vs SDTM AE | GREEN |
| S2-10 | **ADaM Dataset Specifications (ADLB / ADVS / ADEG — Labs, Vitals, ECG)** | Stat programmer | ADaMIG v1.3, CDISC CT, Define-XML v2.1 | Spec review of baseline/window/unit logic; independent double-programming; SDTM reconciliation | GREEN |
| S2-11 | **ADaM Dataset Specification (ADPD — Pharmacodynamics)** | Stat programmer | ADaMIG v1.3, CDISC CT, Define-XML v2.1 | Spec review vs PD plan; independent double-programming; SDTM reconciliation | GREEN |
| S2-12 | **SDTM Mapping Specification** | Stat programmer | CDISC SDTMIG v3.4, Controlled Terminology, Define-XML v2.1 | Independent review vs SDTMIG/CT; Pinnacle 21 rule mapping; traceability check raw->SDTM | AMBER |
| S2-13 | **Annotated CRF (aCRF) — Preparation & Review** | Stat programmer | CDISC SDTMIG v3.4, Define-XML v2.1, CDISC aCRF conventions | Page-by-page review vs mapping spec; consistency with Define-XML origin; peer review | AMBER |
| S2-14 | **Programming Specifications / Derivation Specs** | Stat programmer | ADaMIG v1.3, Define-XML v2.1 (value-level), CDISC CT | Independent spec review; basis for double-programming reconciliation; traceability to SAP | GREEN |
| S2-15 | **PK Analysis Plan (NCA detail)** | PK scientist | ICH PK reporting practice, regulatory NCA guidance, Phoenix WinNonlin (validated NCA engine) | PK scientist + biostatistician review; rules cross-checked vs SAP; Phoenix template qualified | AMBER |
| S2-16 | **popPK / Exposure-Response Modelling Analysis Plan (PMAP)** | Pharmacometrician | Regulatory popPK/E-R guidance, ICH E9(R1), software validation; NONMEM/Monolix as analysis engines | Independent pharmacometrics review; model-qualification criteria pre-specified; run records reproducible | GREEN |
| S2-17 | **NONMEM/Monolix Analysis Dataset Specification (PK/PD modelling dataset)** | Pharmacometrician | popPK dataset conventions, traceability to ADaM, software validation | Independent check vs ADaM source; structure/covariate completeness verified; reproducibility | AMBER |
| S2-18 | **Randomization Schedule Generation** | Lead biostatistician | ICH E6(R3), ICH E9, 21 CFR Part 11 (audit trail, access control), ALCOA++ | Independent QC of allocation vs spec (ratios/blocks/strata); seed/reproducibility verified; secured storage | RED |
| S2-19 | **Randomization Specification** | Lead biostatistician | ICH E6(R3), ICH E9, 21 CFR Part 11 | Independent review vs protocol; IRT UAT alignment; documented approval | GREEN |
| S2-20 | **Unblinding / Code-Break Plan** | Lead biostatistician | ICH E6(R3), ICH E9, 21 CFR Part 11 (audit trail), ALCOA++ | SOP/charter alignment review; role/firewall verification; emergency-unblinding log tested | RED |
| S2-21 | **Data Review Plan / Blinded Data-Review Plan** | Lead biostatistician | ICH E9 (pre-unblinding decisions), ICH E6(R3), ALCOA++ | Peer review; decisions recorded with timestamp before unblinding; consistency with SAP | GREEN |
| S2-22 | **Define-XML Planning / Metadata Strategy** | Stat programmer | Define-XML v2.1, CDISC SDTMIG v3.4 / ADaMIG v1.3, Controlled Terminology | Metadata completeness vs specs; Pinnacle 21 define checks; value-level metadata review | AMBER |
| S2-23 | **Controlled Terminology Plan / CT Version Pinning** | Stat programmer | CDISC Controlled Terminology, SDTMIG v3.4, ADaMIG v1.3, Define-XML v2.1 | Version-pin verification; Pinnacle 21 CT checks; extensible-codelist review | GREEN |
| S2-24 | **Validation / QC Plan (Independent Double-Programming Strategy)** | Lead biostatistician | ICH E6(R3), 21 CFR Part 11, ALCOA++, CDISC conformance; sponsor validation SOPs | Risk-based assignment reviewed/approved; double-programming mandatory for key outputs; audit-ready | GREEN |
| S2-25 | **Hybrid-AI Data-Classification & Routing Plan** | Lead biostatistician | 21 CFR Part 11, ALCOA++, ICH E6(R3); sponsor AI/data-governance policy; HIPAA (API+BAA, not consumer) | Governance review of routing assignments; zero-egress verification for local tier; retention/audit-trail check | GREEN |
| S2-26 | **Pseudo / Dummy Data for Shell Testing** | Stat programmer | CDISC structures (for realism), sponsor standards; non-PHI by construction | Structure conformance to ADaM specs; coverage of edge cases (BLQ, missing, windows) | AMBER |
| S2-27 | **Data-Transfer Specifications (DTS / interface review)** | Data mgmt | CDISC SDTMIG v3.4 (target), 21 CFR Part 11, ALCOA++, sponsor DM SOPs | Test-transfer review; key/format/unit verification; reconciliation procedure agreed | GREEN |
| S2-28 | **Dose-Escalation / Operating-Characteristics Simulation Plan & Outputs** | Lead biostatistician | ICH E9 + E9(R1), reproducibility/software validation | Independent review of assumptions/code; seed/reproducibility documented; results sanity-checked | GREEN |
| S2-29 | **Interim Analysis Plan / DSMB statistical output specifications & firewall (SAP-level)** | Independent unblinded statistician + lead biostatistician | ICH E9; ICH E6(R3); FDA/EMA DSMB guidance | Independent unblinded statistician; firewall enforced | AMBER |
| S2-30 | **Bioanalytical data-transfer & PK sample reconciliation specification (LC-MS/MS lab interface)** | PK scientist + biostat programming + bioanalytical lab | CDISC SDTMIG PC domain; FDA Bioanalytical Method Validation guidance; ALCOA++ | Double-check vs lab dictionary; PK scientist sign-off | AMBER |
| S2-31 | **Value-Level Metadata (VLM) specification for Define-XML** | Biostat programming / metadata lead | Define-XML v2.1; CDISC ADaMIG v1.3 / SDTMIG v3.4 | Independent review vs datasets; Pinnacle 21 Define validation | AMBER |
| S2-32 | **Software / computing environment validation & installation qualification (IQ/OQ) records** | Biometrics systems / QA + statistical programming | 21 CFR Part 11; GAMP 5; ICH E6(R3) | QA review; reported numbers always from validated engines | GREEN |

*Stage gate: **SAP & specifications final.** 32 deliverables.*

## S3 — Data & CDISC

Raw data becomes standardized, conformant, traceable CDISC datasets (SDTM → ADaM) with reviewer guides and Define-XML, under full data integrity controls.

| ID | Deliverable | Owner | Standard | QC | AI |
|---|---|---|---|---|---|
| S3-01 | **EDC / raw data extracts & Data Transfer Agreement (DTA) interface** | Data mgmt (biostat/stat programmer co-reviewer) | 21 CFR Part 11, ALCOA++, ICH E6(R3); CDISC CT for code lists; sponsor DTA SOP | Transfer-spec conformance check on each load; record counts & checksum reconciliation; test-transfer (dry-run) before first production transfer | GREEN |
| S3-02 | **Biostat review of edit checks / data queries / data-review listings** | Lead biostatistician / Stat programmer (with Data mgmt) | ICH E6(R3) GCP, 21 CFR Part 11, ALCOA++; Data Review Plan | Independent review of listing logic; query resolution tracked to closure; reconciliation against SAP-defined critical variables | AMBER |
| S3-03 | **Annotated CRF (aCRF)** | Stat programmer (CDISC/SDTM lead) | CDISC SDTMIG v3.4, Define-XML v2.1, Controlled Terminology; CDISC aCRF conventions | Peer review against final CRF and SDTM specs; Pinnacle 21 / Define linkage check that every annotated origin resolves | AMBER |
| S3-04 | **SDTM datasets — DM (Demographics)** | Stat programmer (CDISC/SDTM lead) | CDISC SDTMIG v3.4, Controlled Terminology, Define-XML v2.1 | Independent double-programming + Pinnacle 21 conformance; reconciliation to raw and randomization | RED |
| S3-05 | **SDTM datasets — PC (Pharmacokinetic Concentrations)** | Stat programmer (CDISC/SDTM lead) with PK scientist | CDISC SDTMIG v3.4 (Findings), Define-XML v2.1, value-level metadata; CT for units/analytes | Independent double-programming + Pinnacle 21; 100% reconciliation to bioanalytical transfer (record count, value, BLQ flags); nominal-vs-actual time check | RED |
| S3-06 | **SDTM datasets — PP (Pharmacokinetic Parameters)** | Stat programmer (CDISC/SDTM lead) with PK scientist | CDISC SDTMIG v3.4 (PP domain), CT for PK parameters (PKPARMCD), Define-XML v2.1, RELREC linkage to PC | Reconciliation to Phoenix output (the engine of record); independent double-programming of the SDTM structure; Pinnacle 21 | RED |
| S3-07 | **SDTM datasets — EX (Exposure)** | Stat programmer (CDISC/SDTM lead) | CDISC SDTMIG v3.4 (Interventions), Define-XML v2.1, Controlled Terminology | Independent double-programming + Pinnacle 21; reconciliation to randomization and to PC time-after-dose derivation | RED |
| S3-08 | **SDTM datasets — AE (Adverse Events)** | Stat programmer (CDISC/SDTM lead) with Data mgmt (MedDRA coding) | CDISC SDTMIG v3.4 (Events), MedDRA (versioned), Define-XML v2.1, CT | Independent double-programming + Pinnacle 21; MedDRA coding QC; reconciliation to safety database/SAE recon | RED |
| S3-09 | **SDTM datasets — LB (Laboratory)** | Stat programmer (CDISC/SDTM lead) | CDISC SDTMIG v3.4 (Findings), CT/LOINC, Define-XML v2.1 value-level metadata, unit standardization | Independent double-programming + Pinnacle 21; 100% reconciliation to lab transfer; unit conversion QC | RED |
| S3-10 | **SDTM datasets — VS (Vital Signs)** | Stat programmer (CDISC/SDTM lead) | CDISC SDTMIG v3.4 (Findings), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; reconciliation to raw | RED |
| S3-11 | **SDTM datasets — EG (ECG)** | Stat programmer (CDISC/SDTM lead) with PK/cardiac-safety scientist | CDISC SDTMIG v3.4 (Findings), CT, Define-XML v2.1; time-matched to PC for C-QTc | Independent double-programming + Pinnacle 21; reconciliation to ECG core-lab transfer; PK time-matching QC | RED |
| S3-12 | **SDTM datasets — DS (Disposition)** | Stat programmer (CDISC/SDTM lead) | CDISC SDTMIG v3.4 (Events), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; reconciliation to enrollment/randomization counts | RED |
| S3-13 | **SDTM datasets — MH (Medical History)** | Stat programmer (CDISC/SDTM lead) with Data mgmt (coding) | CDISC SDTMIG v3.4 (Events), MedDRA, CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; coding QC | RED |
| S3-14 | **SDTM datasets — CM (Concomitant/Prior Medications)** | Stat programmer (CDISC/SDTM lead) with Data mgmt (WHODrug coding) | CDISC SDTMIG v3.4 (Interventions), WHODrug (versioned), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; coding QC | RED |
| S3-15 | **SDTM datasets — PD/DV and other supporting domains (PD/PE/SU/QS, DV, CO, SE/TA/TV/TE trial design, SUPPQUAL, RELREC)** | Stat programmer (CDISC/SDTM lead) with PD/PK scientists | CDISC SDTMIG v3.4 (Findings/Events/Special-Purpose/Trial-Design/SUPP/RELREC), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; RELREC integrity check (PC<->PP); deviation-classification reconciliation | RED |
| S3-16 | **SDTM Define-XML (define.xml + stylesheet)** | Stat programmer (CDISC/SDTM lead) | CDISC Define-XML v2.1, SDTMIG v3.4, Controlled Terminology | Pinnacle 21 Define conformance + dataset-define consistency; peer review; origin/aCRF cross-link check | AMBER |
| S3-17 | **SDTM Reviewer Guide (cSDRG)** | Stat programmer (CDISC/SDTM lead) / Lead biostatistician | PHUSE cSDRG template, CDISC SDTMIG v3.4, Define-XML v2.1 | Peer review; consistency with Define-XML and Pinnacle 21 explanations; sign-off | GREEN |
| S3-18 | **ADaM datasets — ADSL (Participant-Level Analysis)** | Stat programmer (ADaM lead) / Lead biostatistician | CDISC ADaMIG v1.3, Define-XML v2.1, traceability to SDTM, CT | Independent double-programming + Pinnacle 21 ADaM checks; population-flag reconciliation to disposition/randomization | RED |
| S3-19 | **ADaM datasets — ADPC (PK Concentrations Analysis)** | Stat programmer (ADaM lead) with PK scientist | CDISC ADaMIG v1.3 (BDS / PK ADaM), Define-XML v2.1, traceability to PC/EX, CT | Independent double-programming + Pinnacle 21; time-after-dose & BLQ-rule QC; reconciliation to PC | RED |
| S3-20 | **ADaM datasets — ADPP (PK Parameters Analysis)** | Stat programmer (ADaM lead) with PK scientist | CDISC ADaMIG v1.3, Define-XML v2.1, traceability to PP/Phoenix, CT (PK params) | Reconciliation to Phoenix (engine of record); independent double-programming; Pinnacle 21 | RED |
| S3-21 | **ADaM datasets — ADAE (Adverse Events Analysis)** | Stat programmer (ADaM lead) / Lead biostatistician | CDISC ADaMIG v1.3 OCCDS, MedDRA, Define-XML v2.1, traceability to AE, CT | Independent double-programming + Pinnacle 21; TRTEMFL logic QC; reconciliation to SDTM AE & SAE database | RED |
| S3-22 | **ADaM datasets — ADLB (Laboratory Analysis)** | Stat programmer (ADaM lead) | CDISC ADaMIG v1.3 BDS, Define-XML v2.1 value-level metadata, traceability to LB, CT | Independent double-programming + Pinnacle 21; baseline/change & category-derivation QC; reconciliation to LB | RED |
| S3-23 | **ADaM datasets — ADVS (Vital Signs Analysis)** | Stat programmer (ADaM lead) | CDISC ADaMIG v1.3 BDS, Define-XML v2.1, traceability to VS, CT | Independent double-programming + Pinnacle 21; baseline/change QC; reconciliation to VS | RED |
| S3-24 | **ADaM datasets — ADEG (ECG Analysis)** | Stat programmer (ADaM lead) with cardiac-safety/PK scientist | CDISC ADaMIG v1.3 BDS, Define-XML v2.1, traceability to EG/PC, CT; ICH E14/concentration-QTc practice | Independent double-programming + Pinnacle 21; QTc category & time-matching QC; reconciliation to EG | RED |
| S3-25 | **ADaM datasets — ADPD and other analysis datasets (ADPD, ADCM, ADMH, deviation/exposure analysis as applicable)** | Stat programmer (ADaM lead) with PD/PK scientists | CDISC ADaMIG v1.3 BDS/OCCDS, Define-XML v2.1, traceability to SDTM, CT | Independent double-programming + Pinnacle 21; PK-PD linkage & baseline/change QC | RED |
| S3-26 | **ADaM Define-XML (define.xml + stylesheet)** | Stat programmer (ADaM lead) | CDISC Define-XML v2.1, ADaMIG v1.3, Controlled Terminology | Pinnacle 21 Define conformance + dataset-define consistency; derivation-method review; traceability check | AMBER |
| S3-27 | **ADaM Reviewer Guide (ADRG)** | Stat programmer (ADaM lead) / Lead biostatistician | PHUSE ADRG template, CDISC ADaMIG v1.3, Define-XML v2.1 | Peer review; consistency with Define-XML, SAP and Pinnacle 21 explanations; sign-off | GREEN |
| S3-28 | **CDISC conformance / validation (Pinnacle 21) report — SDTM & ADaM** | Stat programmer (CDISC lead) / Lead biostatistician | CDISC conformance rules, SDTMIG v3.4, ADaMIG v1.3, Define-XML v2.1, FDA/PMDA validator rules | Validated Pinnacle 21 engine; documented triage/disposition of every finding; reviewer-guide reconciliation | AMBER |
| S3-29 | **PK bioanalytical sample reconciliation (concentration data load & BLQ handling)** | PK scientist / Stat programmer (with bioanalytical lab & Data mgmt) | FDA/ICH M10 bioanalytical method validation, 21 CFR Part 11, ALCOA++; PK analysis plan BLQ rules | 100% sample-accountability reconciliation; independent check of BLQ/LLOQ flagging; record-count/value match to transfer | AMBER |
| S3-30 | **Blinded data review outputs (data-review listings & blinded review meeting package)** | Lead biostatistician (with Data mgmt, Clinical, PK) | ICH E9/E9(R1), ICH E6(R3); Data Review Plan; 21 CFR Part 11 | Independent review of listings; documented blinded decisions (minutes) prior to unblinding; reconciliation to SAP rules | AMBER |
| S3-31 | **Dry-run / test datasets & dry-run TLF execution** | Stat programmer / Lead biostatistician | Sponsor validation/QC plan; CDISC SDTMIG v3.4/ADaMIG v1.3; 21 CFR Part 11 (program version control) | Independent double-programming on dry-run; shell-vs-output reconciliation; defect tracking to closure | AMBER |
| S3-32 | **Data cleaning / soft lock** | Data mgmt (with Lead biostatistician sign-off) | ICH E6(R3), 21 CFR Part 11, ALCOA++; sponsor lock SOP | Lock checklist completion; cross-functional reconciliation sign-off; audit-trail review | GREEN |
| S3-33 | **Database hard lock** | Data mgmt (cross-functional sign-off incl. Lead biostatistician) | ICH E6(R3), 21 CFR Part 11 (audit trail/e-records), ALCOA++; sponsor lock SOP | Formal hard-lock checklist & multi-party sign-off; audit-trail/lock verification; snapshot integrity check | RED |
| S3-34 | **Traceability (raw -> SDTM -> ADaM) matrix** | Stat programmer (CDISC lead) / Lead biostatistician | CDISC SDTMIG v3.4/ADaMIG v1.3/Define-XML v2.1 traceability principles; ICH E6(R3), ALCOA++ | Cross-link verification raw<->SDTM<->ADaM; peer review; consistency with define/reviewer guides | AMBER |
| S3-35 | **Controlled Terminology & coding-dictionary version management (CDISC CT, MedDRA, WHODrug)** | Data mgmt / Stat programmer (CDISC lead) | CDISC Controlled Terminology (versioned), MedDRA, WHODrug; Define-XML v2.1; 21 CFR Part 11 | Version-lock verification; CT-conformance via Pinnacle 21; coding QC against locked dictionaries | GREEN |
| S3-36 | **Hybrid-AI data-classification routing log & engine-of-record register (Stage 3)** | Lead biostatistician (biometrics AI governance owner) | 21 CFR Part 11, ALCOA++, ICH E6(R3); sponsor AI-governance SOP; hybrid-AI routing rules | Periodic governance review; routing-vs-classification reconciliation; model-freeze/version verification | GREEN |
| S3-37 | **Randomization QC / verification report (schedule vs IXRS/EDC reconciliation)** | Independent statistician (not the schedule generator) | ICH E6(R3); 21 CFR Part 11; ALCOA++ | Independent double-check; segregation of duties from generator | RED |
| S3-38 | **Emergency unblinding / code-break event log & reconciliation** | Unblinded pharmacovigilance / IXRS admin + biostat | ICH E6(R3); 21 CFR Part 11 audit trail | Independent verification; segregation from blinded team | RED |
| S3-39 | **Protocol-deviation classification & reconciliation log (important vs non-important)** | Clinical ops + biostat (blinded review) | ICH E3; ICH E6(R3); ICH E9(R1) (impact on populations/estimands) | Blinded review meeting; medical/biostat adjudication documented | AMBER |
| S3-40 | **Analysis-population derivation & participant-disposition reconciliation (population-flag QC)** | Lead biostatistician + programming | CDISC ADaMIG v1.3 (ADSL population flags); ICH E9(R1) | Independent double-programming + blinded allocation sign-off | AMBER |
| S3-41 | **Data Review Meeting minutes & blinded data-review sign-off (pre-lock)** | Cross-functional (DM, biostat, clinical, PK) — blinded | ICH E6(R3); ALCOA++; 21 CFR Part 11 | Attendees confirm minutes; signatures retained | GREEN |
| S3-42 | **Database lock authorization & 21 CFR Part 11 audit-trail review / system records** | Data management + QA + biostat | 21 CFR Part 11; ALCOA++; ICH E6(R3) | QA review of audit trail; independent lock approval | GREEN |
| S3-43 | **PK sampling-time deviation & actual-vs-nominal time reconciliation (PK timing dataset QC)** | PK scientist + biostat programming | CDISC PC/PP domains; FDA PK guidance; ALCOA++ | Independent check vs source; PK scientist sign-off | AMBER |

*Stage gate: **Database lock.** 43 deliverables.*

## S4 — Analysis & QC

TLFs and PK/PD analyses are produced and independently double-programmed; reported numbers come from validated engines (Phoenix WinNonlin for NCA), never an LLM.

| ID | Deliverable | Owner | Standard | QC | AI |
|---|---|---|---|---|---|
| S4-01 | **Analysis environment qualification & program setup (folder structure, validated software versions, run/seed control)** | Stat programmer | 21 CFR Part 11; ALCOA++; ICH E6(R3) (computerized systems / software validation); CSV/IQ-OQ-PQ | Installation/operational qualification records; environment qualification checklist; version log; reviewer sign-off | AMBER |
| S4-02 | **Disposition tables (screening, randomization, dosed, completed/discontinued; analysis-set attribution)** | Stat programmer | ICH E3; ADaMIG v1.3 (ADSL); SAP | Independent double-programming + reconciliation against shells; population counts cross-checked to ADSL | AMBER |
| S4-03 | **Demographics & baseline characteristics tables** | Stat programmer | ICH E3; ADaMIG v1.3; CDISC CT; SAP | Independent double-programming + reconciliation; descriptive stats verified against ADSL/raw | AMBER |
| S4-04 | **Exposure / dosing & treatment-compliance tables** | Stat programmer | ICH E3; SDTMIG v3.4 (EX); ADaMIG v1.3; SAP | Independent double-programming + reconciliation; dosing records cross-checked to EX/EC | AMBER |
| S4-05 | **Protocol-deviation summary tables (with importance classification)** | Stat programmer | ICH E3; ICH E6(R3); SDTMIG v3.4 (DV); SAP | Independent double-programming + reconciliation; classification reviewed at blinded data review | AMBER |
| S4-06 | **PK concentration summary tables + mean concentration-time figures (linear & semi-log)** | Stat programmer | ICH E3; ADaMIG v1.3 (ADPC); SDTMIG v3.4 (PC); SAP / PK analysis plan | Independent double-programming + reconciliation; BLQ handling verified; figures QC'd vs tables | AMBER |
| S4-07 | **NCA execution in validated Phoenix WinNonlin (PK parameter derivation)** | PK scientist | Validated NCA in Phoenix WinNonlin; ICH E3; ADaMIG v1.3 (ADPP); SAP / PK analysis plan | Phoenix is the validated engine of record; independent re-derivation of key parameters (second analyst / second tool) + reconciliation; lambda_z selection reviewed; run records archived | RED |
| S4-08 | **PK parameter (NCA) summary tables (Cmax, Tmax, AUC, t1/2, CL/F, Vz/F)** | Stat programmer | ICH E3; ADaMIG v1.3 (ADPP); SAP | Independent double-programming of the summary table + reconciliation; values traceable to validated Phoenix ADPP | AMBER |
| S4-09 | **Dose-proportionality & accumulation analyses** | Stat programmer | ICH E3; SAP / PK analysis plan; relevant FDA/EMA dose-proportionality guidance | Independent double-programming + reconciliation; model specification reviewed; outputs traced to validated ADPP | AMBER |
| S4-10 | **Food-effect / DDI / relative BA-BE statistical analysis (ANOVA, 90% CI, GMR) — if applicable** | Stat programmer | ICH E3; FDA/EMA BE & food-effect & DDI guidances; SAP | Independent double-programming of mixed-model/GMR/90% CI + reconciliation; model statement (fixed/random effects) reviewed; values traced to validated ADPP | AMBER |
| S4-11 | **Concentration-QTc (C-QTc) / QT-QTc analysis — if applicable** | Pharmacometrician | ICH E14 + E14 Q&A (Q&A 5/6); ICH E3; ADaMIG v1.3 (ADEG); SAP | Independent re-modelling / double-programming + reconciliation; C-QTc model assumptions reviewed; run records archived | AMBER |
| S4-12 | **popPK / exposure-response modelling (NONMEM/Monolix) with run records** | Pharmacometrician | Validated NONMEM/Monolix; popPK analysis plan; FDA/EMA popPK & exposure-response guidance; ICH E3 | Run records (control streams/seeds/logs) archived; independent review of model code & assumptions; key results reproduced; VPC/GOF reviewed | RED |
| S4-13 | **Safety TLFs — TEAEs (overview, by SOC/PT, severity, relationship, by dose)** | Stat programmer | ICH E3; MedDRA; ADaMIG v1.3 (ADAE); SAP | Independent double-programming + reconciliation; TEAE flag/derivation verified; counts cross-checked to ADAE | AMBER |
| S4-14 | **Safety TLFs — deaths, SAEs, and AEs leading to discontinuation** | Stat programmer | ICH E3; ICH E2A/E6(R3); MedDRA; SAP | Independent double-programming + reconciliation; SAE/death flags reconciled with safety database/PV | AMBER |
| S4-15 | **Safety TLFs — clinical laboratory analyses (summary, change-from-baseline, shift tables, outliers, Hy's law)** | Stat programmer | ICH E3; CDISC CT; ADaMIG v1.3 (ADLB); SAP | Independent double-programming + reconciliation; reference-range/grade derivations verified | AMBER |
| S4-16 | **Safety TLFs — vital signs, ECG (12-lead parameters), physical exam** | Stat programmer | ICH E3; ICH E14 (ECG categorical); ADaMIG v1.3 (ADVS/ADEG); SAP | Independent double-programming + reconciliation; baseline/CFB and outlier criteria verified | AMBER |
| S4-17 | **PD endpoint analyses (biomarker / target-engagement summaries and PK/PD)** | Pharmacometrician | ICH E3; ADaMIG v1.3 (ADPD); SAP | Independent double-programming + reconciliation; derivations/units verified; PK/PD models reviewed | AMBER |
| S4-18 | **Immunogenicity / anti-drug antibody (ADA) analyses — if applicable** | Stat programmer | ICH E3; FDA/EMA immunogenicity guidance; CDISC IS; SAP | Independent double-programming + reconciliation; TE-ADA derivation verified | AMBER |
| S4-19 | **Exploratory / post-hoc analyses** | Lead biostatistician | ICH E3; ICH E9; SAP (exploratory designation) | Risk-based QC (key exploratory outputs double-programmed); exploratory flagging reviewed | AMBER |
| S4-20 | **Statistical analysis execution (run-all production of SAP-specified analyses)** | Stat programmer | ICH E9; ICH E3; 21 CFR Part 11; ALCOA++; SAP | Log review (no errors/warnings); output-vs-shell check; independent double-programming reconciliation gate before release | AMBER |
| S4-21 | **Independent double-programming QC + reconciliation (discrepancy resolution)** | Stat programmer | ICH E6(R3); 21 CFR Part 11; ALCOA++; validation/QC plan; SAP | This IS the QC step — independent double-programming + documented reconciliation; lead biostatistician sign-off | AMBER |
| S4-22 | **TLF validation / QC documentation & discrepancy log** | Stat programmer | 21 CFR Part 11; ALCOA++; ICH E6(R3); validation/QC plan | Completeness/consistency review of QC records; lead biostatistician sign-off | GREEN |
| S4-23 | **Blinded / blind data-review meeting & documented decisions** | Lead biostatistician | ICH E9; ICH E6(R3); ICH E3; SAP / Data Review Plan | Decisions documented before unblinding; cross-functional sign-off; decisions locked | GREEN |
| S4-24 | **Unblinding / code-break execution** | Lead biostatistician | ICH E6(R3); 21 CFR Part 11 (audit trail/access); unblinding/code-break SOP | Authorization & timing documented; access-controlled execution; audit trail; segregation of duties verified | RED |
| S4-25 | **Pharmacometrics / popPK analysis report** | Pharmacometrician | FDA/EMA popPK & exposure-response guidance; ICH E3; modelling analysis plan | Independent review of model & report; key results reproduced; run-record completeness check | GREEN |
| S4-26 | **QC documentation / validation records package (analysis stage)** | Lead biostatistician | 21 CFR Part 11; ALCOA++; ICH E6(R3); validation/QC plan | Completeness/traceability review; lead biostatistician sign-off; archived per retention policy | GREEN |
| S4-27 | **Hybrid-AI governance log for analysis stage (data-classification routing, engine-of-record, model-freeze & prompt/output retention)** | Lead biostatistician | 21 CFR Part 11 (audit trail/records); ALCOA++; internal hybrid-AI governance SOP | Routing/engine-of-record verified per output; freeze hashes recorded; retention completeness checked | GREEN |
| S4-28 | **Phoenix WinNonlin NCA parameter-derivation QC / independent NCA verification record** | PK scientist + independent PK QC reviewer | Phoenix WinNonlin (validated NCA); FDA PK guidance; 21 CFR Part 11 | Independent PK reviewer; documented parameter verification | RED |
| S4-29 | **Pharmacometrics model qualification & validation records (covariate selection, GOF, VPC, bootstrap, run log)** | Pharmacometrician + independent reviewer | NONMEM/Monolix; FDA popPK guidance; 21 CFR Part 11 | Independent pharmacometrics review; reproducible run records | AMBER |
| S4-30 | **Risk-based QC plan execution & QC-level assignment matrix (per-output QC tiering)** | Statistical programming QC lead | ICH E6(R3) risk-based quality management; ICH E8(R1) | QC lead sign-off; mandatory independent double-programming for RED outputs | AMBER |
| S4-31 | **Concentration-QTc (C-QTc) model assumption-check & QC record** | Pharmacometrician / cardiac-safety statistician | ICH E14 / E14-S7B Q&A; FDA C-QTc guidance | Independent cardiac-safety statistician review | AMBER |

*Stage gate: **All outputs QC-passed.** 31 deliverables.*

## S5 — Reporting & CSR

Results become the CSR — statistical methods and results text, in-text tables/figures, appendices, the submission package, and cross-document consistency.

| ID | Deliverable | Owner | Standard | QC | AI |
|---|---|---|---|---|---|
| S5-01 | **CSR Statistical Methods section (Section 9.7 / 11) per ICH E3** | Lead biostatistician | ICH E3 (Section 9.7/11 structure), ICH E9 + E9(R1) estimands, CDISC ADaMIG v1.3 (population/derivation traceability) | Peer review by second biostatistician against final SAP; cross-check that every method maps to a reported output; medical-writer editorial QC; SAP-vs-CSR methods reconciliation checklist | GREEN |
| S5-02 | **CSR Results — in-text (in-body) summary tables and figures** | Stat programmer | ICH E3 (Section 10–12 Results), CDISC ADaMIG v1.3 (source datasets), Phoenix WinNonlin (NCA engine of record) | Independent double-programming or 1:1 reconciliation of every in-text value against the corresponding full TLF and source ADaM; numerical-consistency check; QC log signed off | RED |
| S5-03 | **Biostatistician contribution to and review of the CSR draft (Results text, consistency QC)** | Lead biostatistician | ICH E3, ICH E9/E9(R1), 21 CFR Part 11 (review records/audit trail) | Documented number-by-number consistency check (CSR-vs-TLF-vs-ADaM); second-reviewer spot-check; tracked-changes/comment resolution log retained as a controlled record | AMBER |
| S5-04 | **CSR Appendix 16.1.9 — Documentation of statistical methods incl. final SAP** | Lead biostatistician | ICH E3 (Appendix 16.1.9), ICH E9/E9(R1), 21 CFR Part 11 / ALCOA++ (versioning, audit trail) | Completeness/version reconciliation (approved versions only); cross-check SAP content vs CSR Methods; QA review of appendix assembly | GREEN |
| S5-05 | **CSR Appendix 16.2 — Data listings (by-participant)** | Stat programmer | ICH E3 (Section 16.2), CDISC SDTMIG v3.4 / ADaMIG v1.3 | Independent double-programming or QC review of each listing; reconciliation against summary TLFs; QC log | RED |
| S5-06 | **CSR appendices — Define-XML + reviewer guides (cSDRG/ADRG) and annotated CRF as submission appendices** | Stat programmer | CDISC Define-XML v2.1, cSDRG/ADRG templates, CDISC Controlled Terminology, FDA/PMDA submission requirements | Pinnacle 21 conformance re-run on final datasets; reviewer-guide completeness check; define-vs-dataset consistency QC | AMBER |
| S5-07 | **Patient narrative data support (safety narratives)** | Stat programmer | ICH E3 (Section 12.3.2 / narratives), ICH E6(R3) GCP, 21 CFR Part 11, ALCOA++ | Narrative-to-source data reconciliation per participant; independent check of the trigger list (all deaths/SAEs captured); QC log | AMBER |
| S5-08 | **Participant (patient) profiles** | Stat programmer | ICH E3, CDISC SDTMIG v3.4 / ADaMIG v1.3, 21 CFR Part 11 | Spot-check profiles vs source listings; independent review of profile program; reconciliation against ADaM | RED |
| S5-09 | **Submission datasets + analysis programs as regulatory deliverables (with ADRG/cSDRG)** | Stat programmer | CDISC SDTMIG v3.4, ADaMIG v1.3, Define-XML v2.1, Controlled Terminology; FDA Study Data Technical Conformance Guide; 21 CFR Part 11 | Final Pinnacle 21 conformance; independent double-programming evidence retained; program executability/QC review; traceability raw→SDTM→ADaM→TLF confirmed | RED |
| S5-10 | **Traceability matrix (raw → SDTM → ADaM → TLF/CSR)** | Lead biostatistician | CDISC ADaMIG v1.3 (traceability), Define-XML v2.1 value-level metadata, ICH E3, 21 CFR Part 11 / ALCOA++ | Independent verification that each CSR/TLF number traces to a derivation and source; gap review; QA audit | AMBER |
| S5-11 | **QC of CSR-vs-TLF-vs-dataset numerical consistency** | Lead biostatistician | ICH E3, ICH E9, 21 CFR Part 11 / ALCOA++ (QC records, audit trail) | Independent reviewer performs the check; discrepancy log tracked to closure; second-level QA sign-off | AMBER |
| S5-12 | **Statistical input to clinical study synopsis and briefing documents** | Lead biostatistician | ICH E3 (synopsis), ICH E9/E9(R1), ICH M11 alignment for protocol/briefing context | Consistency check vs CSR body and TLFs (numbers identical); peer/medical review; sign-off | GREEN |
| S5-13 | **Statistical/biometrics sign-off and CSR release for submission** | Lead biostatistician | ICH E3, ICH E6(R3) GCP, 21 CFR Part 11 / ALCOA++ (e-signatures, audit trail), CDISC submission standards | Final QA review; confirmation all discrepancy logs closed; e-signature with audit trail; engine-of-record and model-freeze records attached | RED |
| S5-14 | **eCTD Module 5 submission package assembly & study-data standards conformance (study data reviewer's guide bundle, study-tagging)** | Regulatory + biometrics submission lead | FDA Study Data Technical Conformance Guide; eCTD; Define-XML v2.1 | Submission-readiness review; Pinnacle 21 package validation | GREEN |
| S5-15 | **Analysis Data Reviewer's Guide of record finalization & cross-document consistency (SAP↔Define↔ADRG↔CSR)** | Lead biostatistician + metadata lead | ICH E3; ADaMIG v1.3; Define-XML v2.1; cSDRG/ADRG | Biostatistician sign-off; documented discrepancy closure | AMBER |
| S5-16 | **Hybrid-AI governance dossier — model-freeze records, engine-of-record register, and prompt/output retention archive (submission-level)** | Biometrics AI-governance lead + QA | 21 CFR Part 11; ALCOA++; ICH E6(R3); internal hybrid-AI policy | QA audit; segregation confirms LLM out of reported-number path | GREEN |

*Stage gate: **CSR draft delivered.** 16 deliverables.*

## CDISC dataset & metadata inventory

The standardized-data spine that makes the study submission-ready (SDTM → ADaM → Define-XML, with conformance and reviewer guides).

| ID | Deliverable | Standard | QC |
|---|---|---|---|
| S3-04 | SDTM datasets — DM (Demographics) | CDISC SDTMIG v3.4, Controlled Terminology, Define-XML v2.1 | Independent double-programming + Pinnacle 21 conformance; reconciliation to raw and randomization |
| S3-05 | SDTM datasets — PC (Pharmacokinetic Concentrations) | CDISC SDTMIG v3.4 (Findings), Define-XML v2.1, value-level metadata; CT for units/analytes | Independent double-programming + Pinnacle 21; 100% reconciliation to bioanalytical transfer (record count, value, BLQ flags); nominal-vs-actual time check |
| S3-06 | SDTM datasets — PP (Pharmacokinetic Parameters) | CDISC SDTMIG v3.4 (PP domain), CT for PK parameters (PKPARMCD), Define-XML v2.1, RELREC linkage to PC | Reconciliation to Phoenix output (the engine of record); independent double-programming of the SDTM structure; Pinnacle 21 |
| S3-07 | SDTM datasets — EX (Exposure) | CDISC SDTMIG v3.4 (Interventions), Define-XML v2.1, Controlled Terminology | Independent double-programming + Pinnacle 21; reconciliation to randomization and to PC time-after-dose derivation |
| S3-08 | SDTM datasets — AE (Adverse Events) | CDISC SDTMIG v3.4 (Events), MedDRA (versioned), Define-XML v2.1, CT | Independent double-programming + Pinnacle 21; MedDRA coding QC; reconciliation to safety database/SAE recon |
| S3-09 | SDTM datasets — LB (Laboratory) | CDISC SDTMIG v3.4 (Findings), CT/LOINC, Define-XML v2.1 value-level metadata, unit standardization | Independent double-programming + Pinnacle 21; 100% reconciliation to lab transfer; unit conversion QC |
| S3-10 | SDTM datasets — VS (Vital Signs) | CDISC SDTMIG v3.4 (Findings), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; reconciliation to raw |
| S3-11 | SDTM datasets — EG (ECG) | CDISC SDTMIG v3.4 (Findings), CT, Define-XML v2.1; time-matched to PC for C-QTc | Independent double-programming + Pinnacle 21; reconciliation to ECG core-lab transfer; PK time-matching QC |
| S3-12 | SDTM datasets — DS (Disposition) | CDISC SDTMIG v3.4 (Events), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; reconciliation to enrollment/randomization counts |
| S3-13 | SDTM datasets — MH (Medical History) | CDISC SDTMIG v3.4 (Events), MedDRA, CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; coding QC |
| S3-14 | SDTM datasets — CM (Concomitant/Prior Medications) | CDISC SDTMIG v3.4 (Interventions), WHODrug (versioned), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; coding QC |
| S3-15 | SDTM datasets — PD/DV and other supporting domains (PD/PE/SU/QS, DV, CO, SE/TA/TV/TE trial design, SUPPQUAL, RELREC) | CDISC SDTMIG v3.4 (Findings/Events/Special-Purpose/Trial-Design/SUPP/RELREC), CT, Define-XML v2.1 | Independent double-programming + Pinnacle 21; RELREC integrity check (PC<->PP); deviation-classification reconciliation |
| S3-16 | SDTM Define-XML (define.xml + stylesheet) | CDISC Define-XML v2.1, SDTMIG v3.4, Controlled Terminology | Pinnacle 21 Define conformance + dataset-define consistency; peer review; origin/aCRF cross-link check |
| S3-17 | SDTM Reviewer Guide (cSDRG) | PHUSE cSDRG template, CDISC SDTMIG v3.4, Define-XML v2.1 | Peer review; consistency with Define-XML and Pinnacle 21 explanations; sign-off |
| S3-18 | ADaM datasets — ADSL (Participant-Level Analysis) | CDISC ADaMIG v1.3, Define-XML v2.1, traceability to SDTM, CT | Independent double-programming + Pinnacle 21 ADaM checks; population-flag reconciliation to disposition/randomization |
| S3-19 | ADaM datasets — ADPC (PK Concentrations Analysis) | CDISC ADaMIG v1.3 (BDS / PK ADaM), Define-XML v2.1, traceability to PC/EX, CT | Independent double-programming + Pinnacle 21; time-after-dose & BLQ-rule QC; reconciliation to PC |
| S3-20 | ADaM datasets — ADPP (PK Parameters Analysis) | CDISC ADaMIG v1.3, Define-XML v2.1, traceability to PP/Phoenix, CT (PK params) | Reconciliation to Phoenix (engine of record); independent double-programming; Pinnacle 21 |
| S3-21 | ADaM datasets — ADAE (Adverse Events Analysis) | CDISC ADaMIG v1.3 OCCDS, MedDRA, Define-XML v2.1, traceability to AE, CT | Independent double-programming + Pinnacle 21; TRTEMFL logic QC; reconciliation to SDTM AE & SAE database |
| S3-22 | ADaM datasets — ADLB (Laboratory Analysis) | CDISC ADaMIG v1.3 BDS, Define-XML v2.1 value-level metadata, traceability to LB, CT | Independent double-programming + Pinnacle 21; baseline/change & category-derivation QC; reconciliation to LB |
| S3-23 | ADaM datasets — ADVS (Vital Signs Analysis) | CDISC ADaMIG v1.3 BDS, Define-XML v2.1, traceability to VS, CT | Independent double-programming + Pinnacle 21; baseline/change QC; reconciliation to VS |
| S3-24 | ADaM datasets — ADEG (ECG Analysis) | CDISC ADaMIG v1.3 BDS, Define-XML v2.1, traceability to EG/PC, CT; ICH E14/concentration-QTc practice | Independent double-programming + Pinnacle 21; QTc category & time-matching QC; reconciliation to EG |
| S3-25 | ADaM datasets — ADPD and other analysis datasets (ADPD, ADCM, ADMH, deviation/exposure analysis as applicable) | CDISC ADaMIG v1.3 BDS/OCCDS, Define-XML v2.1, traceability to SDTM, CT | Independent double-programming + Pinnacle 21; PK-PD linkage & baseline/change QC |
| S3-26 | ADaM Define-XML (define.xml + stylesheet) | CDISC Define-XML v2.1, ADaMIG v1.3, Controlled Terminology | Pinnacle 21 Define conformance + dataset-define consistency; derivation-method review; traceability check |
| S3-27 | ADaM Reviewer Guide (ADRG) | PHUSE ADRG template, CDISC ADaMIG v1.3, Define-XML v2.1 | Peer review; consistency with Define-XML, SAP and Pinnacle 21 explanations; sign-off |
| S3-28 | CDISC conformance / validation (Pinnacle 21) report — SDTM & ADaM | CDISC conformance rules, SDTMIG v3.4, ADaMIG v1.3, Define-XML v2.1, FDA/PMDA validator rules | Validated Pinnacle 21 engine; documented triage/disposition of every finding; reviewer-guide reconciliation |
| S3-31 | Dry-run / test datasets & dry-run TLF execution | Sponsor validation/QC plan; CDISC SDTMIG v3.4/ADaMIG v1.3; 21 CFR Part 11 (program version control) | Independent double-programming on dry-run; shell-vs-output reconciliation; defect tracking to closure |
| S3-34 | Traceability (raw -> SDTM -> ADaM) matrix | CDISC SDTMIG v3.4/ADaMIG v1.3/Define-XML v2.1 traceability principles; ICH E6(R3), ALCOA++ | Cross-link verification raw<->SDTM<->ADaM; peer review; consistency with define/reviewer guides |
| S3-35 | Controlled Terminology & coding-dictionary version management (CDISC CT, MedDRA, WHODrug) | CDISC Controlled Terminology (versioned), MedDRA, WHODrug; Define-XML v2.1; 21 CFR Part 11 | Version-lock verification; CT-conformance via Pinnacle 21; coding QC against locked dictionaries |
| S3-43 | PK sampling-time deviation & actual-vs-nominal time reconciliation (PK timing dataset QC) | CDISC PC/PP domains; FDA PK guidance; ALCOA++ | Independent check vs source; PK scientist sign-off |

## TLF & analysis-output inventory (Stage 4)

Tables, listings, and figures by domain, plus the PK/PD analyses. Reported numbers are produced by validated engines, then independently double-programmed.

| ID | Deliverable | Owner | QC |
|---|---|---|---|
| S4-02 | Disposition tables (screening, randomization, dosed, completed/discontinued; analysis-set attribution) | Stat programmer | Independent double-programming + reconciliation against shells; population counts cross-checked to ADSL |
| S4-03 | Demographics & baseline characteristics tables | Stat programmer | Independent double-programming + reconciliation; descriptive stats verified against ADSL/raw |
| S4-04 | Exposure / dosing & treatment-compliance tables | Stat programmer | Independent double-programming + reconciliation; dosing records cross-checked to EX/EC |
| S4-05 | Protocol-deviation summary tables (with importance classification) | Stat programmer | Independent double-programming + reconciliation; classification reviewed at blinded data review |
| S4-06 | PK concentration summary tables + mean concentration-time figures (linear & semi-log) | Stat programmer | Independent double-programming + reconciliation; BLQ handling verified; figures QC'd vs tables |
| S4-07 | NCA execution in validated Phoenix WinNonlin (PK parameter derivation) | PK scientist | Phoenix is the validated engine of record; independent re-derivation of key parameters (second analyst / second tool) + reconciliation; lambda_z selection reviewed; run records archived |
| S4-08 | PK parameter (NCA) summary tables (Cmax, Tmax, AUC, t1/2, CL/F, Vz/F) | Stat programmer | Independent double-programming of the summary table + reconciliation; values traceable to validated Phoenix ADPP |
| S4-09 | Dose-proportionality & accumulation analyses | Stat programmer | Independent double-programming + reconciliation; model specification reviewed; outputs traced to validated ADPP |
| S4-11 | Concentration-QTc (C-QTc) / QT-QTc analysis — if applicable | Pharmacometrician | Independent re-modelling / double-programming + reconciliation; C-QTc model assumptions reviewed; run records archived |
| S4-12 | popPK / exposure-response modelling (NONMEM/Monolix) with run records | Pharmacometrician | Run records (control streams/seeds/logs) archived; independent review of model code & assumptions; key results reproduced; VPC/GOF reviewed |
| S4-13 | Safety TLFs — TEAEs (overview, by SOC/PT, severity, relationship, by dose) | Stat programmer | Independent double-programming + reconciliation; TEAE flag/derivation verified; counts cross-checked to ADAE |
| S4-14 | Safety TLFs — deaths, SAEs, and AEs leading to discontinuation | Stat programmer | Independent double-programming + reconciliation; SAE/death flags reconciled with safety database/PV |
| S4-15 | Safety TLFs — clinical laboratory analyses (summary, change-from-baseline, shift tables, outliers, Hy's law) | Stat programmer | Independent double-programming + reconciliation; reference-range/grade derivations verified |
| S4-16 | Safety TLFs — vital signs, ECG (12-lead parameters), physical exam | Stat programmer | Independent double-programming + reconciliation; baseline/CFB and outlier criteria verified |
| S4-18 | Immunogenicity / anti-drug antibody (ADA) analyses — if applicable | Stat programmer | Independent double-programming + reconciliation; TE-ADA derivation verified |
| S4-22 | TLF validation / QC documentation & discrepancy log | Stat programmer | Completeness/consistency review of QC records; lead biostatistician sign-off |
| S4-25 | Pharmacometrics / popPK analysis report | Pharmacometrician | Independent review of model & report; key results reproduced; run-record completeness check |
| S4-28 | Phoenix WinNonlin NCA parameter-derivation QC / independent NCA verification record | PK scientist + independent PK QC reviewer | Independent PK reviewer; documented parameter verification |
| S4-29 | Pharmacometrics model qualification & validation records (covariate selection, GOF, VPC, bootstrap, run log) | Pharmacometrician + independent reviewer | Independent pharmacometrics review; reproducible run records |
| S4-31 | Concentration-QTc (C-QTc) model assumption-check & QC record | Pharmacometrician / cardiac-safety statistician | Independent cardiac-safety statistician review |

## The hybrid-AI routing overlay

Every deliverable is risk-tiered (**Model Risk = influence × consequence**) and routed by data sensitivity. The pattern from the hybrid AI strategy applies here, deliverable by deliverable.

| Tier | Count | What AI does | Engine |
|---|---|---|---|
| 🟢 GREEN | 48 | Drafts, scaffolds, documentation, summaries on de-identified text | Cloud Claude (API + BAA) or local |
| 🟡 AMBER | 54 | Dataset/TLF code; AI as **one side** of double-programming | Local frozen model (participant-level) / cloud (de-identified) |
| 🔴 RED | 34 | **Nothing autonomous** — validated engine produces the number; LLM out of the path | Validated engine (Phoenix WinNonlin / SAS / NONMEM) only |

**The three hard rules (override everything):** (1) PHI / unblinded / randomization data never leaves — always the local frozen model; (2) every **reported number** comes from a validated engine, never an LLM; (3) de-identification is the only thing that unlocks the cloud path (Claude API under BAA — never a consumer Copilot/Max seat).

## Validation & governance (cross-cutting)

- **Independent double-programming** of every submission-grade output; the AI may replace at most one side, never both.
- **21 CFR Part 11 + ALCOA++**: secure, attributable, time-stamped audit trails; e-signatures; reproducibility.
- **Engine-of-record** recorded per deliverable; reported numbers always from a validated, qualified engine.
- **Model-freeze records** for the local open-weight model (fixed weights + config + seed = a re-runnable validation artifact); change control on any model update.
- **Data-classification routing log** and **zero-egress attestation** for sensitive data; consolidated into a submission-level hybrid-AI governance dossier.
- **Software / environment qualification** (IQ/OQ/PQ) for SAS/R, Phoenix WinNonlin, NONMEM/Monolix, Pinnacle 21.

## Closing

This pipeline is the operating map for an in-house, hybrid-AI biometrics function: human-owned and signed at every step, validated engines for every reported number, full CDISC and Part 11 rigor — with AI accelerating the language, code-drafting, and QC-support layers under explicit, deliverable-by-deliverable controls. The complete register (every field, all 136 deliverables) is in the companion spreadsheet; the visual map is the accompanying one-pager.