The submission-ready companion package to the master protocols.

The retrospective and prospective master protocols reference a family of companion documents that carry the research program from written framework into daily operation. Each document below is a living deliverable maintained by the Data Coordinating Center under formal version control; every amendment is reviewed by the Scientific Steering Committee, the Data Governance Committee, and the Regulatory Core before adoption.

This page presents the current structure and scope of each document so investigators, institutional reviewers, sponsors, and future collaborators can see exactly how the program is engineered end-to-end. Full document PDFs are provided under executed data use agreements on request.

The Manual of Procedures is the operational spine of the program. It translates the scientific protocol into concrete, step-by-step workflows that every coordinator, imaging scientist, laboratory technologist, statistician, and regulatory specialist follows. It exists so that any qualified investigator can be onboarded and become productive without ambiguity about how the study runs.

Structure

• Section 1 — Program governance, roles, decision rights, escalation pathways
• Section 2 — Site activation checklist and readiness criteria
• Section 3 — Participant identification, screening, and eligibility confirmation
• Section 4 — Informed consent process (prospective) and waiver documentation (retrospective)
• Section 5 — Baseline visit choreography and coordination with CT simulation
• Section 6 — Longitudinal visit windows, permissible deviations, and rescheduling logic
• Section 7 — Biospecimen collection, labeling, chain-of-custody, and shipping
• Section 8 — Wearable enrollment, provisioning, participant education, and data sync
• Section 9 — Patient-reported outcome administration (electronic, phone, paper fallback)
• Section 10 — Imaging retrieval, transfer, de-identification, and repository ingestion
• Section 11 — Radiation dosimetry export, cardiac substructure workflow, and QA
• Section 12 — Endpoint capture, adjudication packet assembly, and committee referral
• Section 13 — Deviation reporting, corrective actions, and root-cause analysis
• Section 14 — Retention strategy, contact updating, and re-engagement protocols
• Section 15 — Study close-out, archival, and long-term data stewardship

Governance

The MoP is versioned semantically (major.minor.patch). Patch updates for clarification are approved by the Research Program Director; minor updates affecting workflow are approved by the Scientific Steering Committee; major updates affecting participant experience or data integrity trigger a protocol amendment review and, when applicable, IRB re-review.

SOPs are the atomic, single-task procedures the MoP references. Each SOP follows an identical template — purpose, scope, definitions, responsibilities, materials, step-by-step procedure, quality checks, deviation handling, references, and version history — so investigators encounter the same structure regardless of domain.

Current SOP Library

The Data Dictionary is the authoritative reference for every variable in the repository. It exists in machine-readable form (JSON + CSV) so that CRF logic, REDCap builds, statistical code, and AI feature engineering all bind to a single source of truth.

Variable Metadata Schema

• variable_name — snake_case identifier used in all analytic code
• display_name — human-readable label used on CRFs and dashboards
• description — one-sentence scientific definition
• rationale — why the variable is collected and how it is used analytically
• source_system / source_table / source_field — extraction lineage
• data_type — numeric, integer, categorical, ordinal, datetime, free-text, binary
• permissible_values — enumerated set with codes and labels
• units — measurement units, with harmonization rule if applicable
• collection_frequency — one-time, per-visit, continuous, event-driven
• missing_codes — distinct codes for unknown, not applicable, refused, not obtained
• qc_rules — automated range, logic, and cross-field checks
• transform_rules — derivations (e.g. BMI from weight/height²)
• derived_from — upstream variable dependencies
• stat_class — outcome, exposure, covariate, mediator, sensitivity
• priority — primary, secondary, exploratory, administrative
• version_history — semantic version log with change author and date
• owner — investigator or core responsible for the variable

Domains Covered

Demographics, cancer diagnosis, oncologic treatment, radiation planning, planning CT, dosimetry, cardiovascular history, laboratory, medications, cardiovascular imaging, hospitalizations, encounters, procedures, longitudinal outcomes, mortality, imaging biomarkers, radiomic features, AI outputs, QA metrics, and derived analytic datasets — mirroring the relational domains defined in the master protocol.

REDCap serves as the primary electronic data capture platform for prospectively collected variables and adjudicated outcomes. The specification below is instrument-oriented, event-driven, and designed for longitudinal follow-up.

Project Configuration

• Longitudinal project with defined arms (retrospective, prospective, adjudication)
• Events: Screening, Baseline, Mid-RT, End-RT, 3-mo, 6-mo, 12-mo, Annual
• Randomization module reserved (disabled) for future embedded trials
• Data Access Groups by site for future multicenter expansion
• Audit logging enabled; nightly encrypted backups; SOC 2-aligned hosting
• External Modules: Auto-continue, Field Notes, Data Quality, API Playground

Core Instruments

• Screening & Eligibility
• Consent Tracking (prospective) / Waiver Log (retrospective)
• Demographics & Social History
• Cancer Diagnosis & Staging
• Oncologic Treatment History
• Radiation Planning Summary
• Cardiovascular History
• Medications
• Laboratory Values (auto-populated via EHR feed)
• Cardiovascular Imaging Log
• Biospecimen Collection & Chain of Custody
• Wearable Enrollment & Sync Log
• Patient-Reported Outcomes Battery
• Adverse Events / Cardiovascular Events
• Endpoint Adjudication Packet
• Mortality Ascertainment
• Protocol Deviations
• Data Quality Queries

API & Automation

A read-only REDCap API token is issued to the Statistical Core for nightly analytic dataset builds. A write-scoped token, restricted to specific instruments, powers ETL from the institutional EDW. All API traffic is logged and reviewed quarterly.

Each REDCap instrument is backed by a printable CRF used for source verification, paper fallback, and IRB submission packets. CRFs are laid out for one-page readability wherever possible and mirror the variable order defined in the Data Dictionary.

CRF Design Principles

• Every field maps 1:1 to a Data Dictionary variable_name
• Skip logic printed inline so paper fallback preserves branching
• Explicit checkboxes for 'not applicable' vs 'not obtained' vs 'unknown'
• Date fields use ISO 8601 (YYYY-MM-DD) with partial-date handling
• Adjudicated fields include a signature line for the adjudicator
• Version and effective date printed in the footer of every page

Full CRF Catalog

CRF-001 Screening · CRF-002 Consent · CRF-003 Demographics · CRF-004 Cancer Diagnosis · CRF-005 Oncologic Treatment · CRF-006 Radiation Planning · CRF-007 Cardiovascular History · CRF-008 Medications · CRF-009 Laboratory · CRF-010 Cardiovascular Imaging · CRF-011 Biospecimen · CRF-012 Wearables · CRF-013 PROs · CRF-014 Cardiovascular Events · CRF-015 Adjudication · CRF-016 Mortality · CRF-017 Deviations · CRF-018 Study Exit.

The SAP prespecifies the analytic strategy for the primary aims of the master protocol and establishes a template for endpoint-specific addenda accompanying each manuscript or grant application.

Structure

• 1. Study design summary and analytic populations (ITT-equivalent, per-imaging, sensitivity cohorts)
• 2. Estimands framework — target population, treatment/exposure, endpoint, handling of intercurrent events, summary measure
• 3. Descriptive summaries — continuous (mean/SD, median/IQR) and categorical (n, %) with visualization strategy
• 4. Missing data — mechanism assessment, primary handling (multiple imputation), sensitivity (complete case, tipping-point)
• 5. Primary analyses — Cox with time-varying covariates, competing-risk regression for cardiovascular endpoints
• 6. Secondary analyses — mixed models for longitudinal imaging biomarkers, joint models where appropriate
• 7. Prediction modeling — training/validation split strategy, discrimination (C-index, AUROC), calibration (Brier, calibration slope, ICI), decision curves
• 8. Multiplicity — hierarchical testing for primary/secondary; Benjamini–Hochberg for exploratory biomarker screens
• 9. Subgroup analyses — prespecified interactions with disease site, sex, age, race/ethnicity, radiation modality
• 10. Sensitivity analyses — alternative model specifications, exclusion of protocol deviators, unmeasured confounding (E-value)
• 11. Software & reproducibility — R (primary), Python/PyTorch (AI), version-pinned environments, code archived with each manuscript
• 12. Interim & administrative analyses — enrollment, follow-up, data completeness reports

Reporting Standards

Manuscripts follow STROBE for observational studies, TRIPOD+AI for prediction models, CHARMS for prognostic reviews, and CONSORT extensions when embedded pragmatic trials are activated. All reporting checklists are appended to submitted manuscripts.

The AI Analysis Plan governs how deep learning and multimodal models are developed, validated, documented, and — eventually — considered for clinical decision support. It is written to satisfy TRIPOD+AI, FUTURE-AI, and CLAIM reporting expectations from day one.

Model Lifecycle

• 1. Task definition — clinical question, intended use, target population, deployment constraints
• 2. Dataset construction — inclusion, exclusion, splitting strategy (patient-level), leakage checks
• 3. Preprocessing — DICOM normalization, resampling, windowing, augmentation policy
• 4. Model selection — nnU-Net / MONAI baselines for segmentation; transformer & multimodal fusion for outcome models
• 5. Training — reproducible seeds, cross-validation, early stopping, compute budget logged
• 6. Validation — internal (repeated CV, bootstrap), temporal (held-out era), and external (multicenter) validation
• 7. Performance — discrimination, calibration, subgroup performance, fairness metrics across sex/age/race
• 8. Explainability — Grad-CAM / saliency for imaging, SHAP for tabular, uncertainty estimates (deep ensembles, MC-dropout)
• 9. Robustness — scanner, protocol, dose, and adversarial perturbation testing
• 10. Governance — model card, dataset card, training environment lockfile, weight registry, and audit log
• 11. Deployment readiness — shadow evaluation, human-in-the-loop workflow, drift monitoring plan

Model Registry

Every candidate model receives a permanent identifier (MODEL-YYYY-NNN), an immutable weights hash, and a signed model card linking training data lineage, validation results, and known limitations. Models not meeting the release checklist remain in research-only status and are never exposed through clinical decision-support surfaces.

The Radiomics Analysis Plan operationalizes IBSI-compliant feature extraction and downstream signature discovery so results are reproducible across scanners, institutions, and time.

Extraction Pipeline

• Standardized resampling to isotropic voxel spacing per region of interest
• Fixed bin-width discretization with sensitivity analyses across bin widths
• IBSI-compliant feature families: first-order, shape, GLCM, GLRLM, GLSZM, NGTDM, GLDM
• Wavelet and Laplacian-of-Gaussian filter derivatives with fixed parameters
• Feature provenance stored per case: extractor version, config hash, ROI hash
• Reproducibility QC: test–retest coefficient, ComBat harmonization for scanner effects

Signature Development

• Unsupervised discovery — hierarchical clustering, UMAP, phenotype identification
• Supervised signatures — elastic net, gradient boosting, random survival forests
• Stability selection with bootstrap resampling to control false discovery
• External validation required before any signature is reported in a manuscript
• Comparison against clinical baseline (e.g. traditional risk score) with decision-curve analysis

The Imaging Core Manual documents every step from PACS retrieval to analytic feature delivery. It is the reference used to onboard imaging scientists and to qualify future collaborating sites.

Sections

• 1. Roles, workstations, and secure imaging environment architecture
• 2. PACS/DICOM retrieval workflows with audit logging
• 3. De-identification profile (DICOM Supplement 142) and pixel-level burned-in text scrub
• 4. Repository ingestion, hashing, and duplicate detection
• 5. Image quality assessment — motion, truncation, contrast timing, artifact scoring
• 6. Segmentation protocol — cardiac substructures, great vessels, pulmonary tissue, body composition
• 7. AI-assisted segmentation review, edit tracking, and reviewer certification
• 8. Radiomics extraction and dose overlay procedures
• 9. Inter- and intra-reader reliability program (annual)
• 10. Repository backups, disaster recovery, and long-term archival

Site Qualification

Prospective collaborating institutions complete a qualification packet: sample anonymized cases, scanner inventory, radiation planning system version, PACS connectivity test, and reviewer certification exam. Sites are activated only after passing all qualification steps.

The Biospecimen Manual defines collection, processing, storage, chain-of-custody, and future retrieval procedures — with the explicit goal that specimens collected today remain scientifically valuable for analytical technologies not yet invented.

Collection & Processing

• Tube types, order of draw, and volume targets by visit
• Processing time windows (target: within 2 hours of collection)
• Centrifugation protocols (spin, temperature, duration) per specimen type
• Aliquoting scheme optimized for freeze–thaw minimization
• Barcode labeling with double-verification at every handoff

Storage & Monitoring

• Redundant −80°C and vapor-phase LN₂ storage with independent alarms
• Continuous temperature logging with 24/7 escalation contact tree
• Quarterly freezer inventory reconciliation
• Chain-of-custody log for every specimen movement

Access & Requests

Specimen requests are reviewed by the Biospecimen Committee against a published scoring rubric (scientific merit, statistical power, specimen impact, translational potential). Approved requests execute under a Material Transfer Agreement with a return-of-data clause so derived assays enrich the repository.

The Regulatory Binder is maintained electronically with role-based access and full audit history. It is the artifact any auditor, sponsor, or IRB monitor inspects to confirm the study operates within its approvals.

Contents

• Section A — Approved protocol(s) and all amendments with tracked changes
• Section B — IRB approvals, continuing review, and correspondence
• Section C — HIPAA waiver documentation and Privacy Board correspondence
• Section D — Investigator CVs, licenses, and financial disclosures
• Section E — CITI/GCP/HIPAA training certificates for all study personnel
• Section F — Delegation of Authority log with dated signatures
• Section G — Data Use Agreements, MTAs, and Business Associate Agreements
• Section H — Vendor qualification (REDCap host, cloud, wearables, sequencing)
• Section I — Monitoring reports and CAPA log
• Section J — Adverse events, protocol deviations, and unanticipated problems
• Section K — Publications, presentations, and press releases
• Section L — Correspondence with sponsors and collaborators

During the observational phase, a Data & Safety Monitoring (DSM) Plan — rather than a formal DSMB — provides proportional oversight. When embedded pragmatic interventional studies activate, a formal DSMB is chartered with independent members.

Observational-Phase DSM Plan

• Quarterly Scientific Steering Committee review of enrollment, retention, and data quality
• Semiannual Data Governance Committee review of privacy, security, and access logs
• Annual independent audit sampling consent, source verification, and adjudication
• Continuous automated dashboards for freezer temperature, backup success, and pipeline failures
• Immediate escalation pathway for any privacy incident with 24-hour notification target

Governance Committees

Scientific Steering · Data Governance · Imaging Core · Artificial Intelligence · Radiomics Working Group · Radiation Dosimetry · Statistical Analysis · Clinical Outcomes Adjudication · Publications · External Scientific Advisory Board · Trainee Education · Community & Patient Engagement. Each committee operates under a written charter defining membership, quorum, voting, and conflict-of-interest handling.

Training materials are curated as a role-based curriculum. Every team member completes a general onboarding track plus role-specific modules before touching participant data.

General Onboarding (all roles)

• Program mission, governance, and code of conduct
• Human Subjects Protection (CITI) and HIPAA privacy training
• Good Clinical Practice (ICH E6 R3)
• Responsible AI development and algorithmic bias
• Data security, phishing awareness, and incident reporting
• Publication ethics, authorship, and conflict of interest

Role-Specific Modules

• Coordinators — consent, REDCap, deviation reporting, retention
• Imaging scientists — segmentation, AI-assisted review, radiomics QC
• Machine learning engineers — model registry, validation, release checklist
• Statisticians — analytic dataset construction, missing data, reproducibility
• Laboratory technologists — biospecimen SOPs, chain-of-custody, freezer response
• Regulatory specialists — IRB submission, HIPAA waiver, audit preparation
• Faculty investigators — governance, endpoint adjudication, publication policy

Competency & Recertification

Each module concludes with a competency assessment; passing scores are logged in the Regulatory Binder. Annual recertification is required for all direct-contact and data-access roles, with rolling deadlines managed by the Coordination Center.

Every companion document is version-controlled in the program's document repository with immutable release tags, cryptographic hashes, and a signed changelog. Investigators and collaborating institutions may request the current release package under an executed data use agreement through the Contact page.

Together, these companion documents transform the retrospective and prospective master protocols into a submission-ready package for IRB review, grant applications, and multi-decade program implementation — advancing the collaboration among Dartmouth Radiation Oncology, HIER Institute, and Heart Spark RII toward a nationally recognized precision cardiovascular oncology enterprise.