When is an address not an address? Working with A&E data on violent incident locations
This is a question the GLA Intelligence Unit’s Strategic Crime Analysis team have had to consider in relation to the locations of violent incidents provided by victims on attending hospital.
This data is part of the Home Office Information Sharing for Tackling Violence (ISTV) project, where hospitals are encouraged to record additional information at their A&E receptions around the injuries suffered by victims of violence. This is with the aim of then sharing with other public safety bodies to enrich ongoing preventative work and identify new priorities.
The GLA’s well-respected SafeStats data centre was chosen by the Home Office to source, store, process and visualise data from over 25 hospitals in London to enable analysis by colleagues in fields from preventative health, emergency medicine and health analytics. By hosting this data within SafeStats, it can be viewed and analysed by users across London alongside over 15 million SafeStats records from the Metropolitan Police, London Ambulance Service, Transport for London, British Transport Police, and the London Fire Brigade.
This form of comparative analysis has the potential to identify previously unknown hotspots of violence not reported to the police, something commonly seen in youth, domestic and gang violence.
Of the additional data recorded at hospital receptions, the location of the violent incident is therefore key to assisting this process. However due to the understandable pressures existing on victims and reception staff, the freetext locations recorded are often of varying quality, and therefore cannot be mapped using one single process.
Over recent months and after inspiration from the Metropolitan Police Service’s Stop and Search address coding methods, the Strategic Crime Analysis team have devised an automated methodology to geocode the ISTV data. Using the FME data processing software provided by SAFE, the process can be run easily and objectively and is transparent in its methods; allowing address information to be categorised based on the address elements located (eg. street, postcode) and then geocoded using different methods accordingly.
The aim of the process is to maximise the likelihood of an accurate match, be it at point or area polygon level (depending on the detail of the source data). For potential point matches, online searches have their benefits; however their methodologies are often not fully understood or editable. We therefore aim to geocode as much point data as possible using in-house lookups. This is done by manually creating lookup tables of map co-ordinates based on frequently occurring locations in the data (eg. schools, hospitals, and a wide range of landmarks) and creating processes that can ‘fuzzy match’ the source data against these locations. The ongoing aim is therefore to continue to shift the balance of geocoding away from online searches and towards in-house lookups to maximise the likelihood of an accurate match.
The outputs provide three levels of confidence – point (actual and street centrepoints), electoral ward and local authority – allowing users to view data and carry out analyses at a confidence level that suits them. A secure mapping interface has also been developed to visualise and filter the geocoded outputs, available to a wide range of health professionals from those involved in health analytics right across to the reception staff that record the data in the first.
This process has so far allowed over 70,000 records from 28 hospitals to be brought to life, when previously they were potentially sat in spreadsheets in individual hospitals, requiring manual geocoding in order to glean information. The data are now fit for analyses at a pan-London level on its own or alongside comparable data sources.
This process is not restricted to only London or the ISTV project; it is transferable for similar projects elsewhere in the country, and also for other projects outside of this field battling with the same problem.
If you have any questions on this work, are interested in replicating this process, or are a health analyst interested in setting up an account, please contact email@example.com