Method 1: Hospital Episode Statistics accident and emergency attendance data

We identified this method in our protocol as our intended method of calculating resident catchment populations.

Using HES A&E attendance data, each LSOA was allocated to a single ED. For each LSOA, of all first attendances (in contrast to ‘follow-up attendances’) from the LSOA to any ED, the ED with most first attendances from the LSOA was selected. The LSOA that was recorded was that of the patient’s residence.

Method 2: Department for Transport road travel time data

A list of type 1 A&E department (ED) locations (in England, Wales and the south of Scotland) was prepared for the Department for Transport (DfT). Journey times were modelled by the DfT using their journey time statistics methodology⁵⁹ (minute level accuracy), from the centroid of every LSOA in England to every listed ED. The DfT provided details of the 10 shortest travel times from each LSOA. For each LSOA, the ED with the shortest road travel time was selected.

Method 3: ambulance service dispatch data

Using ambulance service dispatch data, each LSOA was allocated to a single ED. For each LSOA, of all conveyances from the LSOA to any ED, the ED with most emergency conveyances from the LSOA was selected. The LSOA recorded was that of the location from where the patient was conveyed (the scene of the incident).

Method 4: ambulance service estimated catchment areas

Ambulance services were provided with a map of the area surrounding the relevant ED sites (and the road network) and were asked to draw a boundary (or boundaries) to indicate the area(s) in which they would most probably convey patients to the relevant EDs. The annotated maps were digitised and the set of LSOAs intersecting with the bounded area(s) were selected as the catchment area(s) for the ED(s).

Method 5: straight-line (Euclidean) distance

The straight-line distances from every LSOA centroid to every identified ED were calculated. For each LSOA, the ED with the shortest straight-line distance was selected.

Method 1: Hospital Episode Statistics accident and emergency attendance data

The project proposal stated that we would identify resident catchment areas using attendance data to each ED site from the HES A&E data set. It was found that the HES A&E data set only reliably records activity at the acute trust level,⁹ and so it was not possible to directly identify site-level activity for trusts operating more than one ED. It was only possible to use this method for one site. The project management team believed that it would be better to use a single, consistent method to identify catchment areas for all EDs.

Method 2: Department for Transport road travel time data (the selected method)

This method was possible for all areas and used a single, consistently recorded data source. An additional benefit was that the data included intuitive ‘dose’ data (the excess travel time from the LSOA to the next-nearest ED) for each of the LSOAs in the catchment area of the closure ED.

The identified ED closures did not take place simultaneously, thus catchment areas were calculated using the configuration of EDs 2 years prior to the closure of the relevant ED. In practice, this only required two different sets of ED catchment areas for all control and intervention areas.

Method 3: ambulance service computer-aided dispatch data

There were issues with smaller than anticipated numbers in the data set within a 1-year period for some LSOAs, and LSOAs that contained hospitals required corrections (as many of the ambulance conveyances from the LSOA were probably from the hospital to other hospital sites). Moreover, ambulance conveyances are likely to represent the more-severe attendances at an ED rather than attendances as a whole.

Method 4: ambulance service estimated catchment areas

The boundaries that were returned from ambulance services were of variable quality, with some services returning boundaries that formed catchment areas highly similar to those found using computer-aided dispatch (CAD) data (method 3). However, other services provided boundaries that produced catchment areas that varied markedly from those found using the CAD data (method 3). Possible reasons for such variation include staff turnover – some closures had taken place 6 years earlier.

Method 5: straight-line (Euclidean) distance

This method is similar to the DfT road travel time method but is not sensitive to unusual road geographies and so was considered inferior to the DfT method.

Method 6: Hospital Episode Statistics admitted patient care emergency admissions (not investigated)

It would be possible to uniquely identify hospital sites within the HES admitted patient care (APC) data set and to construct catchment areas by examining the residence location of emergency admission patients as opposed to A&E first attendance patients (method 1); however, there are some difficulties with this method.

• Some trusts use multiple codes for the same site and these would all have to be mapped for all hospitals with significant numbers of emergency admissions.
• Emergency admissions are not necessarily representative of ED use as emergency admissions are likely to represent more-severe cases and also include other, direct admissions that did not go through the ED.