Anybody Have a Map?
Boot Disk
Before we introduce ourselves, let’s take a bit of time to hear from you:
- Name
- Role
- Organisation
Who do I work for?
- National partnerships between academia and NHS
- Undertake collaborative, applied research
- We sit across the Universities of Exeter, Plymouth and Bristol
We are measured on the impact we generate
Who am I?
- Ex NHS data scientist
- MSc health data science, 2021
- Postdoctoral research fellow for 3 years
- Author of software for simulation visualisation and geographic optimization
- Trainer on the core HSMA programme
HSMA Approach to Teaching
We’re teaching you complex, valuable stuff in a (hopefully) fun and lighthearted way
- We spend far too much of our lives being very serious
- I could give you a very dry lecture for three hours
- But you’ll actually learn and remember more this way
- We respect your time in coming here today; we’re here to deliver serious learning, sometimes unseriously
The Lambda pathway
The HSMA pathway
OR vs DS Fight
“We have lots of data on readmissions.
Can we teach a machine to automatically identify which patients are likely to be readmitted?”
“We want to make these changes to our process for triaging patients.
What do we predict the impact will be?
What resources will we need to ensure the process is efficient?”
Hat Dan is talking about data science
Fonz Dan is talking about operational research
The Analytics Pyramid
What HSMA Builds To
The full picture
Discussion
Where do you feel like your organisation is in this progression?
- Do you know if you have someone with the job title ‘data scientist’ or something similar?
- Do the splits between backward and forward reflect your experience?
A Mental Model for DES
Discrete Event Simulation is like Lego
We give you a few bricks - the core components of discrete event simulation - and suddenly you can build pretty much anything you can imagine
A Mental Model for Geographic
Geographic modelling is more like getting into DIY and buying yourself an empty toolbox.
You start with your core geographic knowledge and map building skills.
This is your hammer.
DES vs Geographic
And it will take time to build up your collection of tools…
But HSMA gets people started with some really powerful techniques that can help with data-driven decision making.
The power of maps
In recent years, we’ve experienced the power of the digital mapping revolution.
It’s now entirely normal to us that we can generate routes in seconds, with accurate travel times, traffic data, points of interest…
The past…
This was revolutionary!
Tobler’s First Law
But to interpret things about areas correctly, we need to consider this key geographic principle from the one of the most influential geographer’s of the 20th century, Waldo Tobler.
Everything is related to everything else, but near things are more related than distant things
Tobler’s First Law in Action
What does this mean in practice? Two maps of the same nine deprivation deciles across Devon — one is the real pattern, one is the same values shuffled randomly across the same areas.
Which is real — left or right? Vote in the poll.
Indices of Deprivation - randomly spread
Indices of Deprivation - real spread
Hotspots
Choropleths can tell us how a metric (like demand, equity, or incidents) varies.
But using statistical methods for identifying hotspots, coldspots and outliers can help us identify significant patterns.
Travel Times - Good Enough?
Travel Times - Limitations
Isochrones
Travel times open up more plot types
Coverage
Layering multiple isochrones is one way we can start to get an idea of where there are areas of poorer access that could be targeted.
Catchment
Taking this one step further, 2SFCA (2 step floating-catchment area) combines both demand and capacity to give a more detailed look at not just where there is access (which we define as a particular travel time banding), but where there is sufficient capacity.
Being close to a clinic doesn’t guarantee you can get seen there — if lots of other people can also reach it, it’s effectively less available to you.
And more kinds of hotspots!
We can combine travel data to identify high priority areas, like high deprivation and long journey times.
Remember this one?
What if we opened a new site here?
Location optimization is how we answer it properly.
It’s the first power tool we hand HSMAs once they’ve got the mapping fundamentals.
What the optimiser found
It compared all 14 combinations against 5 different measures.
Several of them are defensible — nothing else beats them across the board. Each one is the best answer to a different question.
Mapping of Violent Crime
- HSMAs used their skills to isolate hotspots of violent crime to target.
Routing and Scheduling
Most of us primarily encounter routing and scheduling in the context of parcel deliveries!
But there are potential applications in health data science projects too.
Efficient Routes
You may have heard of the ‘travelling salesman problem’ - where the goal is to find the most efficient route between a series of points.
This is actually a hard problem for computers to solve as the number of points increases - and even more so when you expand it to multiple vehicles (e.g. community nurses) with constraints (e.g. patient continuity, different skill mixes).
But computers are, overall, better at it than humans - and can do it at scale, automatically.
PenCHORD Project Showcase - Dialysis Transport during COVID-19
This PenCHORD project was tasked with making sure dialysis transport could continue safely and efficiently during the COVID-19 Pandemic.
By combining simulation of patient disease status with routing, they were able to show the impact of different patients-per-vehicle limits on overall driving time required.
PenCHORD Project Showcase - Dialysis Transport during COVID-19
Modelling multiple disease spread scenarios, they were able to show a significant reduction in travel hours required by relaxing policies to allow two patients to travel in the same vehicle simultaneously, balancing infection control needs with limited resource and essential treatments.
Boundary Optimization
Boundary optimization algorithms, a bit like location optimization - can help you redraw boundaries in a way that is fairer.
They can try out hundreds of thousands of combinations, scoring against metrics like balance of assigned workload across regions, or how much the region differs from an ideal workload per team.
HSMA Project Showcase - Boundary Optimization for North West Ambulance Service
The North West Ambulance service used advanced location optimization approaches to update historic ambulance dispatch boundaries, which were now imbalanced and inefficient after population shifts since their original setup.
After HSMA 7
After completion of the project, we’ll follow up with you to see what happened as a result of the work. This is important for us as we need to capture impact from our work.
