How Faster Address Confirmation Improves Ambulance Performance

Author: Marcus Poulton, Birkbeck, University of London, August 2017

This is a simulation notebook for the London Ambulance Service. This notebook explores what happens to arrival time performance when you change the characteristics of the call handling process. Specifically, this notebook explores what happens when you change the profile of the gazetteer component in the call-handling process. Throught this notebook we only consider Red 1 calls.

First, we characterise the time taken toconfirm a location by examining 15,000 Red 1 calls, using LAS data from March 2015 onwards.

The histogram shows that the call-taker can confirm the location of 75% of all calls within 53 seconds (blue dotted line), and 90% of calls within 81 seconds. This curve can be approximately estimated using a log-normal distribution with mean log of 3.5 and a standard deviation log of 0.7. Log-normal distributions have a long right hand tail, which in our case means that some locations are complex and take a considerable amount of time to find. This effect pushes the dotted lines to the right.

We can extend this concept across the call-taking, dispatch and ambulance process to estimate current performance. We use similar curves for AMPDS, dispatch and journey time and combine them to predict arrival time. Using existing data, we compare our model to actual data and find that it is accurate (enough). The chart below shows these curves. The colours are as follows:

• the grey curve is the gazetteer time (the same as above),
• the light red is the AMPDS time,
• the green is the dispatch time and ,
• the blue is the arrival time
• dotted lines show the 75% quantile time for each of the curves.

We also plot a dotted blue line at the time where 75% of ambulances will arrive on scene. Here, the arrival time is 540 seconds, 1 minute above the 480 seconds performance target, resulting in a 65% achievement for calls to arrive within 8 minutes.

Assume we have new improved gazetteer. Again, we can postulate a new performance curve for this gazetteer. Here we assume that the new gazetteer can achieve better performance with a mean log of 3 and a standard deviation log of 0.65. Here is what the curve looks like..

With this new improved gazetteer, 75% of the locations can be confirmed within 31 seconds instead of 53 seconds. So how does performance improve? We take the existing data and characterise the AMPS, dispatch and arrival times and insert our new gazetteer.

Now we can see that performance is now increased by 4%.