Transcript The Malthus Programme

Clinical impact of a discrete event simulation
model for radiotherapy demand
Raj Jena
University of Cambridge
COMODO
Computation | Modelling | Dose Calculation
Disclosures
• This research programme was funded by the
National Cancer Action Team
• I receive funding from:
Overview
•
•
•
•
•
•
•
Radiotherapy : cost effective cancer cure
Cottier Report : stand up and be counted
NRAG 2007 Model : so near yet so far
Malthus & Multi-scale modelling
Implementation
Impact : curing cancer with computation
Next steps & conclusion
Radiotherapy
• Effective spatially and
biologically targeted
anti-cancer therapy
• Used in treatment of
40% of patients cured of
cancer
• Cost effective : £2500 for
course of treatment
Pre-2K era
• 1997 – 25% of RT machines
aged 10 years or older
• 15% increase in treatment
fractions year on year
• 28% patient waiting more
than 4 weeks to start RT
• RT services deemed
inadequate
• 5 year investment programme
in RT hardware (NOF / DOH
Funding)
2003 Cottier report
• Survey of 57 NHS radiotherapy centres
• 25% linacs under 3 years old, but 39% over 10
years old.
• Nationally running at just over 50% predicted
number of linacs
• Only 39% of centres reaching target of 4 linacs
per million population
Equipment, Workload and Staffing for Radiotherapy in
the UK 1997–2002 Ref No: BFCO(03)3
The Royal College of Radiologists
RT Utilisation Models
•
•
•
•
CCORE 2003
SRAG 2006
WCSCG 2006
NRAG 2007 – model of RT demand
– 63% increase in RT activity required from 30,000 to 48,000
fractions/million/year
– Projected activity of 54,000 fractions per million by 2016
• Aspirational targets used in 2007 Cancer Plan
Garbage In :: Garbage Out
• Poor assessment of data quality
• Usage of data from different healthcare
systems
• Applied to a generic population
• Poor robustness of computational models
• Poor fit to local data : commissioning ‘blight’
• We could do a lot better!
Multi-scale models
DNA
• PARTRAC (Fortran)
• Monte Carlo transport codes (Fortran)
Cell
• CelCyMUS : Cell cycle model (Fortran)
• Virtual Petri Dish (C++)
Tissue
• CAMUS : Cellular Automaton (Pascal)
• Ayatana : 3D Spheroid (F#)
Patient
Population
• BJJK : Discrete event simulation (Fortran)
• ??? (Discrete event simulation)
“Dear Prof Richards. We can
do this properly. Please can
we have some money…”
Malthus was born
Monte Carlo application for local radiotherapy treatment & hospital usage statistics
Discrete event simulation
• Create virtual cancer patient who acquires a cancer diagnosis, treatment events, and
radiotherapy treatment
• Use locale specific base data (population data and cancer incidence)
• Incorporate population and cancer burden projection to 2030
• User facing tool : Windows executable allowing user modification of model
Model architecture
User select PCT / Region and
disease sites for simulation
Disease
Stage
Age
Co-morbidity
Virtual
population of
patients
Curated
incidence data
feeds from
NCAT server
Evidence based trees
Consensus based trees
Breast
Lung
H&N
Summary stats
Detailed report
Urology …
Typically 2000 passes
through the decision tree
for each cancer patient
Decision trees
Incidence
data for
given
disease
stage
Selecting
between major
treatment
groups (esp
surgery vs RT)
Taking disease and
patient specific factors
into account for choice
of treatment
Details of the available treatment options
Standards based architecture
• Simple GUI interface to
run and modify model
• Decision trees encoded
in XML
• Base data in Excel files
• OLE automation to
generate reports in MS
Office
Advantages over NRAG model
•
•
Study local variation in breast cancer incidence
Breast Cancer : Evidence based fractionation, Haringey PCT vs Torbay PCT
Haringey
9025 per M
Simulation summary : Haringey
Total number of fractions in the selected population :2032
Total population of selected PCTS :225138
Fraction burden per million of populations :9025.58
Access rate is 75.27%
Simulation performed using NCAT validated decision trees
Simulation completed at 19:17:07
Simulation summary : Torbay
Total number of fractions in the selected population :2286
Total population of selected PCTS :133749
Fraction burden per million of populations :17091.72
Access rate is 75.46%
Simulation performed using NCAT validated decision trees
Simulation completed at 19:19:01
Torbay
17091 per M
Double the population, roughly the same
number of fractions…
Advantages over NRAG model
• Study effect of changes in non-RT related management
• Prostate cancer, England, no retreatment : change in divide from
surveillance and EBRT
Scenario 1 : Low risk prostate Cancer
• Surveillance 25%
• Surgery 20%
• Brachytherapy 15%
• EBRT 40%
Scenario 2 : Low risk prostate Cancer
• Surveillance 70%
• Surgery 10%
• Brachytherapy 5%
• EBRT 15%
Simulation summary
Total number of fractions in the selected population :480479
Total population of selected PCTS :51111574
Fraction burden per million of populations :9400.59
Access rate is 57.16%
Simulation performed using NCAT validated decision trees
Simulation completed at 19:34:51
Simulation summary
Total number of fractions in the selected population :384736
Total population of selected PCTS :51111574
Fraction burden per million of populations :7527.38
Access rate is 48.05%
Simulation performed using NCAT validated decision trees
Simulation completed at 19:36:41
Sense check of data
REAL WORLD IMPACT
Decision support | Influencing Policy
Installed User Base
• Over 400 registered users
• 2012 : Every commissioning lead, RT service
manager
• 2013 : Malthus Cymru developed for Welsh
CSAG
• Canada, Australia, France, Germany
• Facilitates dialogue between providers and
purchasers of RT
Impact case studies
• Support of numerous business cases for new
treatment units (e.g. Sheffield, Norwich,
Oxford, Brighton)
• Satellite centres in Peterborough, Manchester
• Evaluation of new technologies (stereotactic
radiotherapy in south-west)
• WIPT : Workflow planning tool for medical
physics
Future plans
• Complexity level of
treatment
• Specialised models for new
Malt hus ram p - up
technologies
Modelling suite for scenario based planning of proton beam Therapy
– Proton Beam Therapy
– Radiosurgery
• BI integration
– GIS data for travel isochrones
– Cost effectiveness analysis
(Jean-Marc Bourque, King’s)
Acknowledgements
University of Manchester : Norman Kirkby, Karen Kirkby
University of Surrey : Tom Mee
Addenbrooke’s : Mike Williams
King’s : Jean-Marc Bourque
NCAT : Mike Richards, Tim Cooper
COMODO
Computation | Modelling | Dose Calculation
Models
encode
knowledge
Data
empowers
models
Knowledge
informs
decision