3. Overview
• Mapping 101 using John Snow’s Cholera discovery in 1854
• Trend mapping for KPIs to look for opportunities for improvements
• Localised mapping of a pharmacy to look at market penetration
• Specific medication mapping using trend and penetration mapping
5. Thematic Mapping
• Boundaries are arbitrary
• Boundaries don’t change
as the underlying data
changes
• Boundaries are often
related to non-healthcare
functions (eg voting,
rating)
• Uncommunicative and
statistically invalid
29. Advantages of Surface Analysis
• Computation is not particularly intensive
• Easy to understand
• Easy to compare different data sets
• Anonymise the data (without the need for meshblocks)
• Communicative
• Can be statistically validated
• Automated and served up from the cloud
Notas del editor
Geospatial consultants. We provide geospatial advice, design and services.
People love maps – they’re very communicative
Even among professionals the awareness of what can be achieved with maps is relatively low. Across all industries.
A map isn’t or shouldn’t be a handcrafted product but almost always is
A map should be a report, a piece of analysis. A good map is simple – it highlights a problem and suggests a solution, and because it is a map it tells you where you can take action.
Yesterday David talked about the way we are drowning in data and the way forward is about how we use the data we’ve got. While Graeme challenged whether we were being smart enough and maximising the use of the data.
We think that there are some real opportunities in the health sector to make better business decisions, to get better health outcomes using data from which locational intelligence can be gleaned.
I’ll also thank Neal Johnston who is doing a masters in geographic information science at Canterbury – he prepared the analysis of the pharmacy data as an internship under our direction and mentoring.
One of the most underused techniques is Surface analysis. Use surface analysis to look at 3 areas – Service delivery, Market penetration, and Medication effectiveness.
First I’ll give a quick lesson in mapping of surfaces
Then the Service Delivery Analysis will be looking at KPI performance in the Emergency Services sector but the principle will equally apply to business problems and outcomes in the community care sector.
The market penetration analysis will be using Pharmacy prescription data to get a feel for the pharmacy’s catchment and performance within that catchment. But equally the principles could be applied to any health programme that is being rolled out into the community.
The third analysis will be looking at the prescription of two different drugs that address smoking and what can be learnt from a policy perspective.
The purpose will be to demonstrate the techniques rather than to highlight any particular insights.
Some of you will be familiar with John Snow’s 1854 map of the London Cholera outbreak around the Broad St pump. Its regarded as one of the first examples of using spatial data to solve a problem and transform health.
The dominant infection theory at the time was transmission through ‘bad air’. Snow was a skeptic so he plotted the data on a map and noticed a correlation between the water pump and the outbreak. The pump handle was removed and the outbreak brought under control.
John Snow didn’t have the tools we have today so he did his Analysis between the ears by staring at the data. 160 years later this is still a very common practise. The Google push pin is still the datatype of choice and the preferred analysis method is still to stare hard at the data.
It is becoming more common to see data presented spatially using thematic mapping. We saw one yesterday of hospital mortality rates by English County.
This example is a Waikato map of social wellbeing
Here aggregated data is displayed by colour flooding polygons based on an attribute.
I find these to be problematic.
English counties don’t deliver hospital care so why would you map hospital mortality by County?
It’s difficult to determine what the map means or what actions need to be taken. Larger areas with low populations can dominate the map or large areas are colour flooded based on data that exists only in one corner of the polygon
When financial analysts look at an organisation, they don’t look at the transactions they look at the summaries and trends. The Cholera incidents you see here are the transactions
To view the trends it is best done through surfaces.
A Surface is a single wobbly piece of 3 dimensional data interpolated from sparse points
The cholera incidents were the transactions. The surface is the summary. This surface was created using Hotspotting. As you can see the red hotspot coincides with the Broad Street Pump
We wouldn’t recommend that anyone actually uses Hotspotting as a technique. It is just a very crude clustering technique.
The reason that it works on John Snows map is that John Snow got lucky – doubly lucky.
1 there was a single problem with a single cause
2 there was no transportation – buckets of water are heavy to carry so everyone went to the nearest well. That doesn’t happen today.
We’re much more mobile. The Pharmacy I use is as likely to be the one that is open late or near my GP as it is to be near my house and the Pharmacy data we analysed certainly bears this out.
To do Surface Analysis you need to be using more sophisticated and robust approaches.
Now we’re going to summarise large quantities of data and look for trends, We’ll be converting thousands of points into several surfaces and comparing the relationships between those surfaces mathematically
Auckland region
St John gets funded by MOH based on how quickly they turn out to incidents.
For example 50% of life threatening incidents have to be responded to within 6 minutes in urban areas.
Overall they hit their targets nationally but there is less certainty about regional variations, or where there is pressure on meeting their kpis as demand for their services increases. Because it affects their funding it is a very real business problem.
One months data of life threatening incidents
Slightly dated now
Once again These are the transactions
If the incident was responded to within the kpi target we assigned a value of 1. Coloured pink
If it missed we assigned a value of 0 Coloured black
By and large its pretty good – St John already know that.
There are some areas that are pretty good while others are a bit patchy.
To create a summary, what we can do now is to calculate a % surface so we can tell where they are hitting their targets and where they are routinely missing them.
To do this you simply stand at a point and look at all the nearby points and average them. You then move your position say 100m and recalculate it. And you keep going until the entire area has been summarised
So here we have our wobbly 3 dimensional percentage surface
(Action song)
The Red hilltops are 100% so every patient has been responded to within the kpi time frame
The yellow troughs are 0% - not one patient was responded to within the KPI timeframe
The orange slopes are somewhere in between
First thing that you will notice is that surface analysis immediately anonymises the data. This is really important in a health context
We could look at a yellow area to see if we can improve things, but just because it is yellow it doesn’t mean it is a problem it could be that there are very few incidents in those areas.
So we need to be able to compare this data with the overall incident traffic.
Density Analysis
Where are the most jobs
This is much the same process except in this case instead of averaging the 1’s and 0’s this time we have counted them
The darker cells are the peaks in the surface
Where can St John make a meaningful difference to both the business outcomes by minimising the risk to its funding, and also the health outcomes by reducing the risk to the patients?
The great thing about surface analysis is the ability to do comparisons between surfaces.
Maths can be conducted between the corresponding cells of different surfaces.
So in this case we are going to try and stretch the values by looking for the poorest performance in the busiest areas and find those locations where we can make the biggest difference.
This is a good illustration of the properties of a good map.
Simple Map that highlights a problem, suggests an action and tells you where to take it
Because it is a 3d surface we can contour it. The darker blue represents the areas where the biggest opportunity exists to improve the kpi’s for life threatening jobs. And of course this is just one of many kpi’s that St John needs to achieve and others would also need to be taken into account.
Indeed surface modelling could be used to determine whether the kpis themselves are in sync or whether they are confounding each other.
If you are simply using a tool like excel the only changes you can make are policy changes. Surface analysis allows for targeted tactical changes.
If a trial was rolled out in say South Auckland, the model can be very quickly recalculated to see if the situation has improved or deteriorated. Such analysis can be automated as a reproducible process.
Because you can do maths between surfaces you can compare surfaces of different dates to look for where temporal trends are taking place. You can subtract this months surface from last months to see whether and where things have got better or worse.
It is also possible to model ‘what if’ scenarios. What if We added a new station or relocated a station. What might the KPI’s look like as a result?
In this scenario we’re going to be comparing business data with some demographic data.
We’ll also be exploring cost-distance modelling and how this can be used. This is particularly relevant when you are looking at certain targeted sectors of the population and whether they have adequate access to the services that are needed.
Population Density surface calculated from the Census data. It was calculated for the entire country – this is the population density for Chch.
It has been contoured with the densest areas shown darker.
You can see the central city and industrial areas have far fewer residents as does the eastern suburbs which have been depopulated due to the red zoning following the earthquakes.
The red star is the location of Bastins Pharmacy in St Albans
This is another density surface this time of Bastins Patient data
As you can see, for various reasons a percentage of the patients come from well beyond the immediate vicinity
The 3 red shapes represent 500 m driving distances – out to 1.5 km, our initial analysis would suggest that the average catchment for Bastins would be about 1.25 km. The green dots are neighbouring pharmacies
Comparing the patient density with the population density gives us the market penetration
The green dots represent the location of other pharmacies
Again the surface has been contoured and we can start to see where Bastins is doing really well and where they are subject to greater competition.
Interestingly near the top of the surface there is an area of very good market penetration despite other pharmacies being closer. This underlines the idea that people don’t always go to the closest pharmacy
Without Surface Analysis it is unlikely that this pattern would be readily available or understood. It allows you to ask what’s happening and why. May be other demographic features such as age are coming into play and can be added to the model.
While this is a commercial setting the same principle can be applied to the coverage for any healthcare service or initiative.
As a technique it can be used for justify initiatives or to measure their success.
Cost distance modelling enables the modelling of service levels
Cost allocation modelling enables the definition of service catchments
You need some source data from which to grow the model and a friction layer that determines how the model will grow
Friction could be distance covered, travel time, $$$ etc
This slideshows 2 surfaces, the driving distance from the nearest pharmacy out to a maximum of 4km, as well as the driving distance from Bastins. Our Analysis showed that there was a consistent steady decrease of customers out to 4km beyond which there was a rapid decline.
Hardly anywhere is the overall penetration likely to drop below 50% which means Christchurch is well served and there aren’t many if any opportunities to set up a new pharmacy without taking an existing business on head on.
It can be used to Model Access to services particularly in relation to demographics such as deprivation, age, transportation means, rurality or ethnicity.
If Bastins wanted to increase their market penetration where might be the best place to target their marketing effectively?
We could create a formula that takes 4 different surfaces into account.
Firstly areas with high populations have more potential customers.
Secondly Areas where the current market penetration is low means more available customers to convert
Thirdly potential customers who are closer to Bastins may be more likely to convert
As will potential Customers who live further from competing pharmacies
The Blue areas are areas where there is reduced opportunities
The Purple areas are where the formula suggests there are greater opportunities.
When the marketing dollar is under pressure the money needs to be targeted well. Perhaps a pamphlet drop in the streets surrounding his pharmacy to the South looks like a good bet.
Two important concepts come out of here.
Firstly sometimes the results are surprising and not obvious. This formula suggests that it would be worthwhile targeting right in a competitors heartland.
Secondly you must always remember that surface analysis is never the truth but always a model and as a model it must always be tested. Rigorously and statistically.
There is enormous potential in the ability to use surface modelling for Policy purposes.
Here we are going to examine the prescription of 2 smoking medications to see if any patterns stand out and whether there might be any demographic reason for the patterns
With the data coming from just a single pharmacy, there isn’t a lot to be read into it which is a significant limitation but it does highlight some possibilities worth exploring.
Urban Population density
Urban Smoking density
(go back and forth)
There definitely seems to be a reduced smoking rate in the more affluent northwestern and hill suburbs
Of course we could keep doing this or we could create a smoking percentage surface.
Percentage Urban smokers
Which bears out the low rates we suspected with the higher rates around the inner city fringe.
Ok remember our Bastin’s pharmacy market penetration map from before? Now we are going to look at the prescriptions for Champix and Habitrol in the areas of Bastin’s highest market penetration.
Analysis of 2 variables
(a-b)/(a+b)
A = habritol prescription density
B = champix prescription density
There are two islands of data which correspond to the areas of greatest market penetration which are contoured.
The Dark green is a clear preference for Habritol, the Purple a clear preference for Champix. There is a very clear pattern present, but what might be causing it?
Deprivation surface - least deprived is blue running through merivale and fendalton
The red is the most deprived
Bastins Pharmacy neatly straddles these demographics
We can see at least visually that some sort of correlation with deprivation may be present
Here its overlaid over a smoking density surface where dark blue is the highest density and red the lowest density. It seems to be showing a pattern where champix is preferred in areas where smoking is less of a problem.
Is this something that you knew already?
Is it something worth investigating?
Are there other demographics such as education levels at play?
Is it telling us something about the way our smoking policies are working or not.
And is the DHB funding for champix being used effectively or not?
Surface modelling gives us the opportunity to ask these questions and look for answers.
(Final slide next)
These models generally only take minutes to run
Every night at five to 7 the public consume temporal surface models on the TV weather forecast, Temparature surfaces, Wind and rainfall surfaces, Air pressure surfaces.
I hope I’ve demonstrated that there are great opportunities to use surface analysis to identify and communicate opportunities for better health outcomes
If you want to catch up and talk further I’ll be at the toniq stand