AYA HRQoL Tri-Layer Hybrid Model

Hybrid ABM + Network + System Dynamics model of quality of life after adolescent and young adult cancer. MATLAB R2024b, no toolboxes, deterministic under fixed seeds.

The repository is private during the dissertation; public release (with license) is pending. Contact for access.

Implementation Details Request Access

Quickstart

% MATLAB R2024b, no toolboxes
cd model_v2/coupling_mvp
verify_interfaces                 % interface checks (incl. no-leak)
Run_Ablation_Ladder(20)           % ablation ladder + macro trend check
Run_Measurement_Validation(20)    % observation layer: calibrate/validate split
make_figures(20)                  % regenerate the figure pack -> figures/
recreate_preprint_figures(20)     % regenerate the EORTC preprint figures

% Standalone benchmark vs pre-existing SD / ABM models:
cd ../benchmark_comparison
run_benchmark                     % one command reproduces every output

Repository Structure

.
├── README.md
├── CITATION.cff                               machine-readable citation
├── docs/
│   ├── INTERFACE_SPECIFICATION_TEMPLATE.md    the formal six-interface spec
│   ├── MODEL_TIMELINE_AND_EQUATIONS.md        timeline, SD-model comparison, equations
│   ├── AYA_parameters_and_equations.xlsx      parameters / equations workbook
│   ├── PARAMETER_PROVENANCE.md                every parameter traced to source
│   └── MODEL_CREDIBILITY_AND_VALIDATION.md    verification vs validation
├── model_v2/
│   ├── coupling_mvp/                          the coupled model (no toolboxes)
│   │   ├── AYA_MVP_Coupled.m                  six-interface coupled model
│   │   ├── coupling_switches.m                interface switches + ablation cells
│   │   ├── Run_Ablation_Ladder.m              ablation + macro trend check
│   │   ├── Run_Measurement_Validation.m       calibrate/validate split
│   │   ├── Run_Adaptation_Sensitivity.m       anti-overfitting sweep
│   │   ├── verify_interfaces.m                no-leak verification
│   │   └── make_figures.m                     regenerates the figure pack
│   ├── benchmark_comparison/                  standalone docking benchmark
│   │   ├── run_benchmark.m                    reproduces every output
│   │   └── models/                            SD-only, Q-PRIMA, hybrid arms
│   └── diagrams/                              CLD + UML (Mermaid, Graphviz, Vensim)
├── website/                                   collaborator site (this AYA page)
└── .github/workflows/deploy-pages.yml         auto-deploy to GitHub Pages

Core Scripts

AYA_MVP_Coupled.m

The coupled tri-layer model: 100 heterogeneous agents, an evolving support network, and a system-dynamics service queue exchanging state through six switchable interfaces over a 24-month, diagnosis-anchored horizon.

coupling_switches.m

The six interface switches (SW_NA, SW_SA, SW_AS, SW_NS, SW_SN, SW_AN) and the ablation cells they define. Every coupling can be turned off independently.

verify_interfaces.m

No-leak verification: proves each off switch removes exactly one effect, so the ablation ladder cleanly measures the contribution of each layer.

Run_Ablation_Ladder.m

Runs the ablation ladder across coupling cells plus a macro trend check, the evidence that each layer earns inclusion.

Run_Measurement_Validation.m

Observation layer with a calibrate/validate split against empirical anchors (EORTC QLQ-C30 reference values and AYA cohort patterns).

run_benchmark.m

Standalone docking benchmark against pre-existing models (SD-only, Q-PRIMA, ancestor hybrid). Full coupled model: 7.0% validation error (seeds 1-20).

Model Specification

Agents            100 heterogeneous AYA patients (age, diagnosis, HRQoL state)
Time              0 to +730 days (24 months), anchored at diagnosis (t = 0)
Network           Watts-Strogatz; mean degree drawn PER RUN from the empirical
                  AYA support-network distribution (Cloyes 2022):
                  mean 6, SD 2.22, range 2-10 -> even values {2,4,6,8,10}
Service layer     1 SD service stock with queue + per-agent states
Coupling          6 switchable interfaces: SW_NA SW_SA SW_AS SW_NS SW_SN SW_AN
Reproducibility   deterministic under fixed seed; independent RNG stream for
                  the network-size draw (a run drawing k = 6 reproduces the
                  previous fixed-6 result exactly)
Provenance        every parameter Cited / Derived / flagged Assumption

Parameters & Equations Workbook Equations on the Model Site

Verification & Validation Workflow

1. No-leak verification

Each interface switch removes exactly one effect, verified across all six interfaces.

2. Ablation ladder

Layer-by-layer contribution measurement across coupling cells.

3. Measurement validation

Calibrate/validate split against EORTC QLQ-C30 anchors and AYA cohort patterns.

4. Adaptation sensitivity

Anti-overfitting sensitivity sweep over adaptation parameters.

5. Stress scenarios

Network stress tests (isolation, reduced ties) with degree-matched comparisons.

6. Benchmark docking

Structural docking against SD-only and Q-PRIMA arms; seed-robustness table (20 seeds).

Requirements

MATLAB R2024b
No external toolboxes required
Deterministic: fixed seeds reproduce every figure and table

Empirical Anchors

Data	Source	Citation
Incidence / Survival	SEER	seer.cancer.gov
HRQoL Norms	EORTC QLQ-C30	Hinz (2014), Eur J Cancer
Support-Network Size	AYA cohort	Cloyes (2022), mean 6, range 2 to 10
Network Effects	Meta-analysis	Pinquart (2010), 87 studies
Unmet Needs	AYA HOPE Study	Keegan (2012)

Full provenance, every parameter traced to a source or a flagged assumption, is documented in the repository (docs/PARAMETER_PROVENANCE.md) and summarized on the model site.