A comprehensive discrete-event simulation model for multiple myeloma disease outcomes and treatment pathways, developed through collaboration between Monash University's Centre for Health Economics and Transfusion Research Unit.
- Calibrated Transport methods: out-of-trial outcome prediction (e.g. DVd at L2) via the
transport_dvdanalysis - Common Random Numbers (CRN): aligned RNG across treatment arms for variance-reduced cost-effectiveness comparisons
- Per-line overall survival: OS is now a separate parametric model for each line/stage of therapy, clocked from that line's own start, replacing the single from-diagnosis survival curve — removing an accumulated-time bias that inflated survival for poor responders at later lines
- Individual comorbidities: the OS and ASCT-eligibility equations now carry four individual comorbidity flags (renal impairment, cardiac, pulmonary, diabetes) instead of a single combined comorbidity score
- Standardised CSV exports: machine-readable result surface for downstream and programmatic use
- Vectorised engine (incorporated from v2.1): Mata vector/matrix rewrite for dramatically faster large-scale simulation
- Rebrand: project renamed from EpiMAP Myeloma to the Monash Myeloma Model
The model simulates the complete treatment journey of multiple myeloma patients using 50 evidence-based risk equations derived from the Australia and New Zealand Myeloma and Related Diseases Registry (MRDR).
Discrete-event structure: patients flow from diagnosis through successive lines of therapy (LoT 1 … LoT N). At each line the model simulates regimen, best clinical response (BCR) and treatment duration, with ASCT intent/receipt, maintenance, treatment-free intervals and death (green = simulated outcomes).
- Patient Characteristics: Age, sex, ECOG performance score, R-ISS staging & four individual co-morbidities (renal, cardiac, pulmonary, diabetes)
- Treatment Pathways: Comprehensive modelling of up to 9 lines of therapy
- Clinical Outcomes: Best Clinical Response (BCR) and Overall Survival (OS), the latter modelled per line of therapy
- ASCT Modelling: Separate pathways for transplant-eligible patients
- Maintenance Therapy: Post-induction treatment modelling
- Parametric Survival Models: Time-to-event analysis for all outcomes
- High Performance: Mata implementation for efficient large-scale simulations
- Stata 15.0 or higher (valid licence required)
- Windows, macOS, or Linux operating system
git clone https://github.com/CHE-Monash/Myeloma-Model.git
cd Myeloma-ModelEach analysis is driven by its own dispatcher in analyses/<name>/. Configure a run by editing the globals at the top of the dispatcher, then run it from Stata:
cd "path/to/myeloma-model"
do "analyses/default/simulate.do"The dispatcher's configuration block sets the run via globals. Interactive runs need nothing more; run.do and the HPC array jobs instead pass a few of these as optional positional arguments (boot, min_bs, max_bs, and — where an analysis has scenarios — scenario), which override the globals of the same name:
| Global | Description | Example |
|---|---|---|
analysis |
Analysis name (folder under analyses/) |
default |
int |
Intervention | all, VRd, SoC, DVd |
line |
Line of therapy assessed (1–9; 0 = all) |
0 |
coeffs |
Coefficient set | full, train |
data |
Patient cohort | synthetic, train, test, predicted |
min_year / max_year |
Diagnosis-year range | 1995 / 2040 |
min_id / max_id |
Patient ID range | 1 / 101212 |
cost_year |
Price year for costs (AUD) | 2025 |
drate |
Annual discount rate (PBAC = 5%) | 0.05 |
report |
Generate PDF report (0/1) | 0 |
scenario |
Scenario label | Base |
boot |
Bootstrap flag (0/1) | 0 |
min_bs / max_bs |
Bootstrap iteration range | 1 / 100 |
| Dispatcher | Focus |
|---|---|
analyses/default/simulate.do |
The reference analysis — projection (all regimens, current practice) by default; out-of-sample (70/30) validation with $scenario outsample |
analyses/transport_dvd/simulate.do |
DVd via Calibrated Transport |
default is one model run two ways (see analyses/default/README.md): $scenario "" = full-registry fit on the synthetic incidence population (projection); $scenario "outsample" = 70%-train fit on the held-out real 30% (validation).
Results are written to analyses/<analysis>/simulated/.
Every analysis folder under analyses/<name>/ follows the same layout, so the same commands work across analyses:
| File / folder | Role |
|---|---|
run.do |
Analysis runbook — the full ordered pipeline (prep → simulate → validate) as runnable do lines, plus the bootstrap HPC plumbing (the canonical record of the sbatch/rsync sequence, kept in a trailing block comment). Start here to reproduce an analysis. |
simulate.do |
Dispatcher — one simulation run, configured by the globals block at the top. Also accepts the optional positionals boot min_bs max_bs [scenario] (used by run.do and the HPC arrays). |
outcomes/txr_<coeffs>.do |
Per-line regimen code lists ($TXR_L1..L9) for the coefficient set; prep/risk_equations.do's gen_txr builds the regimen dummies from them. |
outcomes/ |
Any analysis-specific outcome overrides (e.g. sim_bcr_override.do) and coefficient generators (e.g. calibrated_transport.do). |
coefficients/ |
Fitted coefficients coefficients_<coeffs>.mmat (+ bootstrap/ for the 500 resamples). |
patients/ |
The simulation cohort .dta (git-ignored; restricted data). |
simulated/ |
Run outputs (per-scenario subfolders where an analysis has scenarios; bootstrap/ for resamples). |
results/ |
Analysis-level CSVs/figures + a results.md — the canonical read surface for downstream consumers. |
README.md |
Analysis-specific notes (research question, methodology, quirks). |
Shared inputs are built once by prep/ (multiple imputation, population cohorts) and fitted per-analysis by prep/risk_equations.do <analysis> <coeffs> …, which loads that analysis's outcomes/txr_<coeffs>.do. The bootstrap steps run on the MASSIVE cluster via the generic array jobs in hpc/ (multiple_imputation.script, risk_equations.script, simulate.script); per-analysis arguments ride on each sbatch --export= line, recorded in that analysis's run.do.
Starting a new analysis? Copy analyses/template/ — a documented skeleton already in this layout (dispatcher, runbook, txr regimen list, and optional BCR/TXR outcome-override stubs following transport_dvd).
Each simulation can emit a PDF report ($report) and machine-readable flat CSV outputs for downstream use (R/Python post-processing, dashboards, manuscript drafting). CSV export is produced for point-estimate runs and skipped under bootstrap.
- Engine-level —
core/export_results.dowrites the CSVs every analysis needs (per-patient summary, BCR distribution, mean cost/QALY/LY) intosimulated/<scenario>/. - Analysis-level — outputs specific to one analysis, plus cross-scenario aggregation, live under
analyses/<name>/results/with aresults.mdsummarising the key figures — the canonical read surface for downstream consumers.
See each analysis README (e.g. analyses/transport_dvd/README.md) for specifics.
Myeloma-Model/
├── core/ # Shared simulation engine
│ ├── load_patients.do # Patient data loading
│ ├── mata_setup.do # Mata vector/matrix setup
│ ├── mata_functions.do # Mata utility functions
│ ├── simulation_engine.do # Discrete-event simulation core
│ ├── rng_slots.do # Common-random-number slot registry
│ ├── run_pipeline.do # Shared engine pass used by every analysis
│ ├── process_data.do # Post-simulation processing
│ ├── export_results.do # Machine-readable CSV exports
│ ├── generate_report.do # PDF report
│ ├── outcomes/ # Outcome (risk-equation) modules
│ └── tests/ # Engine verification: unit tests + extreme-value
├── analyses/ # Per-analysis dispatchers, data & results
│ ├── template/ # Copy-me skeleton for a new analysis
│ ├── default/ # Reference analysis: projection ($scenario "") + out-of-sample ($scenario "outsample")
│ └── transport_dvd/ # DVd Calibrated Transport
├── prep/ # MRDR → model inputs: imputation, risk equations, cohorts, benchmarks
├── patients/ # Synthetic incidence cohorts (.dta; git-ignored)
├── docs/ # Technical documentation
├── hpc/ # MASSIVE M3 cluster scripts
├── README.md
├── CHANGELOG.md
├── LICENSE # PolyForm Noncommercial 1.0.0 (source-available)
└── LICENSING.md # dual-licence model + commercial enquiries
The vectorised implementation provides significant performance benefits:
- Faster Execution: Matrix operations process all patients simultaneously
- Better Memory Efficiency: Reduced overhead from loop-based processing
- Scalability: Handles large cohorts (10,000+ patients) more efficiently
- Maintainability: Cleaner code structure with modular vector setup
The model has been comprehensively validated:
- Published Validation: See Irving et al. (2024) in PLOS ONE
- Out-of-Sample Testing: 70/30 split validation with 500 bootstrap iterations
- Survival Curve Accuracy: No significant difference in 98.3% (118 of 120) of monthly comparisons over the 10 years post-diagnosis for the current per-line OS model (up from 90% for the published 2024 model)
- Engine Verification: unit tests + an extreme-value stress test in
core/tests/(the stress test passes 7/7) - Vectorisation Validation: Comprehensive test suite confirms identical results to original implementation
Out-of-sample validation (per-line OS model): whole-population overall survival for the held-out cohort (95% CI) vs. the simulated cohort (95% CI) over 10 years, with the monthly two-sample p-value (blue, right axis). The curves are statistically indistinguishable in 118 of 120 months.
Access previous versions via Git tags:
- v3.0: Current version (
main) — Calibrated Transport & CRN methods, per-line overall survival, individual comorbidity covariates, standardised CSV exports, rebrand to Monash Myeloma Model (incorporates the earlier vectorised Mata engine) - v2.0: Reorganised architecture with extended treatment options —
git checkout v2.0 - v1.0: Initial public release (August 2024) —
git checkout v1.0
Irving A, Petrie D, Harris A, Fanning L, Wood EM, Moore E, et al. Developing and validating a discrete-event simulation model of multiple myeloma disease outcomes and treatment pathways using a national clinical registry. PLOS ONE. 2024;19(8):e0308812. doi:10.1371/journal.pone.0308812
@software{monash_myeloma_model_v3_0,
title = {Monash Myeloma Model (originally published as EpiMAP Myeloma)},
author = {Irving, Adam and Petrie, Dennis and Harris, Anthony and Fanning, Laura and
Wood, Erica M and Moore, Elizabeth and Wellard, Cameron and Waters, Neil and
Augustson, Bradley and Cook, Gordon and Gay, Francesca and McCaughan, Georgia and
Mollee, Peter and Spencer, Andrew and McQuilten, Zoe K},
version = {3.0},
year = {2026},
url = {https://github.com/CHE-Monash/Myeloma-Model},
doi = {10.1371/journal.pone.0308812},
institution = {Monash University}
}Health Economists: Adam Irving, Dennis Petrie, Anthony Harris, Laura Fanning
Clinical Experts: Zoe K McQuilten, Erica M Wood, Bradley Augustson, Gordon Cook, Francesca Gay, Georgia McCaughan, Peter Mollee, Andrew Spencer
Registry Team: Elizabeth Moore, Cameron Wellard, Neil Waters
Consumer Representative: Andrew Marks
For access to genuine patient data from the Australia and New Zealand Myeloma and Related Diseases Registry: - Submit applications to the MRDR Steering Committee - Visit: mrdr.net.au
- Model Questions: [email protected]
- Technical Issues: Create an issue
- Collaboration Enquiries: Contact the research team
Open-science, source-available, dual-licensed — following the principle "as open as possible, as closed as necessary." The full code and model documentation are public for inspection and reproduction.
- Noncommercial use (academic, educational, public-research, government, personal) is free under the PolyForm Noncommercial License 1.0.0.
- Commercial use — including industry-sponsored regulatory or reimbursement (e.g. PBAC) submissions — requires a separate commercial licence from Monash University.
- Registry-derived data and fitted parameters are additionally governed by the MRDR data agreement.
See LICENSING.md for the full model and commercial-licence enquiries. This is source-available, not "open source" in the OSI sense (it restricts commercial use).
- Australia and New Zealand Myeloma and Related Diseases Registry
- Monash Centre for Health Economics
- Monash Transfusion Research Unit
The Monash Myeloma Model project (originally EpiMAP Myeloma) was supported by Medical Research Future Fund GNT1200706 and GNT2017480 and by National Health and Medical Research Council GNT1189490, GNT2024876 and GNT2036025. We thank patients, clinicians, and research staff at participating centres for their invaluable contributions to the MRDR.
Important: This model is designed for research purposes. Clinical decisions should always involve qualified healthcare professionals and consider individual patient circumstances.