Serious e-scooter casualties: comparing police and hospital data
Published 24 May 2023
© Crown copyright 2023
This publication is licensed under the terms of the Open Government Licence v3.0 except where otherwise stated. To view this licence, visit nationalarchives.gov.uk/doc/open-government-licence/version/3 or write to the Information Policy Team, The National Archives, Kew, London TW9 4DU, or email: psi@nationalarchives.gov.uk.
Where we have identified any third party copyright information you will need to obtain permission from the copyright holders concerned.
This publication is available at https://www.gov.uk/government/statistics/serious-e-scooter-casualties-comparing-police-and-hospital-data/serious-e-scooter-casualties-comparing-police-and-hospital-data
About this report
This report presents results of linking STATS19 data on vulnerable road user casualties with data on trauma patients from the Trauma Audit and Research Network (TARN) dataset, with a focus on what this suggests about the relative completeness of STATS19 data for e-scooter casualties.
E-scooters are a relatively new mode of transport. The Department are running e-scooter trials in a number of locations around England, but outside these areas e-scooters are illegal to use on public roads. This report does not distinguish between trial and private scooters as the distinction is not recorded within the TARN dataset.
Main findings
Based on data for patients admitted to hospital with serious injuries sustained while riding an e-scooter, as recorded in the TARN dataset, for the period 2020 to June 2022:
- a total of 428 casualties were identified, compared with 6,785 pedal cyclist and 5,996 motorcyclist casualties over the same period
- overall, around 33% were linked to a STATS19 record, compared to 26% for pedal cyclists and 56% for motorcyclists
Trends in police and trauma registry cases for e-scooter casualties are broadly similar since the start of 2020. This suggests we can be somewhat confident that, although incomplete, the police data is capturing trends for the more serious casualties sufficiently well.
Allowing for the characteristics of collisions and casualties, there is no evidence that more severely injured e-scooter casualties are any less likely to be reported to police than pedal cyclists, however both e-scooter user and pedal cyclist casualties in single vehicle collisions are less likely to appear in STATS19 than motorcyclists.
Overall, for the vulnerable road users considered here, the odds of a serious casualty in the TARN dataset being linked to a record in the STATS19 police data are affected by:
- type of collision (no other party involved = less likely to appear in STATS19)
- severity (more serious = more likely)
- collision location (not on road = less likely)
- data quality (precise location recorded = more likely)
These findings are comparable with related research. Further work could extend this to other road user types, explore injury rates (per trip, or per mile travelled) and provide a more detailed analysis of the resulting linked dataset.
Introduction
Following initial work to establish feasibility of linking police data on reported road casualties (collected via the STATS19 system) with data from the Trauma Audit and Research Network (TARN), this analysis explores the value of the linked data in assessing the likely extent to which e-scooter and other vulnerable road user casualties are reported to and by police, and compares how injury severity is captured in each dataset.
E-scooter casualties have been captured via a free text field in STATS19 from 2020, with results published in the Department’s e-scooter factsheets. It has long been known that a considerable proportion of non-fatal road casualties are not reported to police. It is further suspected that non-fatal casualties for e-scooter users are among the most likely to be under-reported in road casualty data since they have no obligation to inform the police of collisions, and are currently not legal on public roads outside of trial areas, but to this point it has not been possible to reliably quantify this.
Aims and objectives
The specific aims of this work are:
- to link police (STATS19) and hospital (TARN) records for more seriously injured e-scooter casualties creating a dataset for analysis of the number and patterns of injuries arising from e-scooter collisions (and compare this to pedal cyclist and motorcyclist collisions)
- to use the linked data to estimate the level of reporting to police for more seriously injured e-scooter user casualties, and compare this to other modes
- to compare the reporting of injury severity for cases appearing in both police and hospital data
This report presents an initial analysis which particularly focuses on the second of these aims.
Datasets and methodology
STATS19 data
STATS19 data forms the basis for the Department’s published road safety statistics, and is collected in accordance with the guidance set out in the STATS20 document. For this analysis, an extract of data covering final data for 2020 and 2021, together with provisional data for 2022. Data was extracted for pedal cyclist, motor cyclist and ‘other road user’ casualties to mirror the TARN dataset.
Provisional data used in linking was based on that supplied by police forces as at 1 November 2022. This data is expected to be over 95% complete, though this will vary by police force area and month (with some entire months of data missing in some police force areas). Consequently, the analysis is provisional and subject to revision as final data becomes available (though this is unlikely to change the main findings).
TARN data
The TARN data for this work was supplied by the TARN team at the University of Manchester, under project reference 220703.
TARN data is based on specified inclusion criteria which includes all trauma patients admitted for 3 or more nights, admitted to a High Dependency Unit or dying in hospital, and having particular types of injuries.
The data provided covered all TARN patients in England from the start of 2020 (the point at which e-scooters began to be identified in TARN) where the injury mechanism was recorded as ‘vehicle collision’ and the position in vehicle was motorcyclist, pedal cyclist or ‘powered personal transporter or e-scooter’. Data were provided in January 2023 and included all records for 2022 available at that point, a total of 14,266 records. Data to mid-2022 are expected to be substantially complete.
Figure 1: TARN STROBE diagram
Linkage methodology
The TARN and STATS19 datasets do not share common unique identifiers, but do contain a number of common (or closely related) variables including age and sex of casualty, and date, time and location of incident. Therefore a probabilistic method was used to link the two datasets, as outlined in the previous feasibility study. This was based on the well established Fellegi-Sunter method. Further details of the precise approach used are given separately and the R code used is available.
TARN data on vulnerable road users
In itself, TARN provides a rich source of information on more seriously injured road casualties in England, including vulnerable road users, and has been used for many studies including recently on e-scooters.
This analysis is focused on comparing TARN with STATS19, so the the following presents a high level summary for context. The findings are in line with the academic research which provides a more detailed analysis of TARN data in isolation.
For this study, TARN casualty type was based on what was coded in the ‘position’ variable. While there are distinct categories for pedal cyclists and motorcyclists, e-scooters are included within the ‘powered personal transporter or e-scooter’ category. Therefore the incident description was used to identify e-scooter casualties, as far as possible. Records were classified as ‘e-scooter’ where the description explicitly mentioned this or a related word (such as ‘electric’) but not where it was solely ‘scooter’ as in these cases it was not possible to distinguish between e-scooters and petrol scooters from the information available. A small number of records where the position variable was recorded as pedal cycle or motorcycle were subsequently recoded as e-scooter based on the incident description.
The ‘other’ category shown in the tables covers the remainder of the ‘powered personal transporter or e-scooter’ group, and includes records which could not definitively be assigned to one of the other groups. This is likely to include some e-scooter records which could not be identified with certainty from the information available, as well as electric bicycles, mobility scooters and other vehicles.
Incident variables
Table 1 shows the number of TARN records by year and road user type for the data analysed. E-scooter casualties recorded within TARN have grown relative to the other modes over this period; this may reflect better recording as well as growth in the use of e-scooters in England. From 2021 onward, the number of e-scooter trauma patients was broadly around a tenth of the number of pedal cyclists.
The table also shows some of the other information available about the nature of the incident recorded in TARN which, in contrast to STATS19, covers collisions which occur away from the public highway. Unlike STATS19, precise incident location is not always available, with the best indication being the incident postcode though this is not always recorded.
While TARN does not routinely categorise the nature of the collision, some information can be gleaned from the free text incident description. This was used to assess whether another party (another vehicle, or a pedestrian) was involved in the collision, or whether it was the result of a fall or a collision with an object.
Table 1: TARN patient counts by road user type and incident variables
Note: E-scooter records coded from ‘powered personal transporter or e-scooter’ category, using information in free-text incident description
| Variable | Value | E-scooter | Pedal cycle | Motorcycle | Other |
|---|---|---|---|---|---|
| Year | 2020 | 46 (10.7) | 2932 (43.2) | 2185 (36.4) | 23 (10.9) |
| 2021 | 245 (57.2) | 2649 (39.0) | 2435 (40.6) | 115 (54.5) | |
| 2022 to Jun | 137 (32.0) | 1204 (17.7) | 1376 (22.9) | 73 (34.6) | |
| Postcode? | Postcode | 176 (41.1) | 2605 (38.4) | 2687 (44.8) | 78 (37.0) |
| No postcode | 252 (58.9) | 4180 (61.6) | 3309 (55.2) | 133 (63.0) | |
| Location | Road | 296 (69.2) | 4791 (70.6) | 5041 (84.1) | 140 (66.4) |
| Pavement | 72 (16.8) | 311 (4.6) | 24 (0.4) | 23 (10.9) | |
| Public area | 30 (7.0) | 1009 (14.9) | 439 (7.3) | 23 (10.9) | |
| Other | 30 (7.0) | 674 (9.9) | 492 (8.2) | 25 (11.8) | |
| Other party? | Yes | 156 (36.4) | 2130 (31.4) | 4883 (81.4) | 0 (0.0) |
| No | 272 (63.6) | 4655 (68.6) | 1113 (18.6) | 211 (100.0) |
Casualty rates
Casualty rates for road collisions per mile travelled are calculated based on STATS19 data and published as part of the department’s road casualty statistics.
Equivalent casualty rates can be calculated using TARN data, to show the number of hospitalised casualties per mile, or per trip. There are however, more limitations in these calculations as the TARN data includes those injured away from the public highway (for example at home, or off road such as in a car park or field) where the activity would not be captured within the traffic data used to calculate casualty rates. In addition, robust exposure data for e-scooters does not currently exist.
Consequently, comparative casualty rates have not been presented here but this will be explored further as new data becomes available.
Demographic variables
Both TARN and STATS19 contain information about the age and sex of those injured (table 2). Of trauma patients appearing in the TARN data, compared to pedal cyclists and motorcyclists e-scooter users are more likely to be female, and on average younger. As noted, a more comprehensive analysis is available in the academic research.
Table 2: TARN patient counts and distribution by road user type and demographic variables
SD denotes standard deviation.
| Variable | Value | E-scooter | Pedal cycle | Motorcycle | Other |
|---|---|---|---|---|---|
| Gender | Female | 82 (19.2) | 986 (14.5) | 305 (5.1) | 24 (11.4) |
| Male | 346 (80.8) | 5799 (85.5) | 5691 (94.9) | 187 (88.6) | |
| Age | Mean (SD) | 34.7 (15.5) | 47.8 (18.3) | 39.9 (16.7) | 41.7 (19.1) |
| Age group | Under 20 | 93 (21.7) | 692 (10.2) | 741 (12.4) | 31 (14.7) |
| 20 to 29 | 82 (19.2) | 594 (8.8) | 1356 (22.6) | 31 (14.7) | |
| 30 to 39 | 92 (21.5) | 803 (11.8) | 1132 (18.9) | 41 (19.4) | |
| 40 to 49 | 86 (20.1) | 1178 (17.4) | 831 (13.9) | 40 (19.0) | |
| 50 plus | 75 (17.5) | 3518 (51.8) | 1936 (32.3) | 68 (32.2) |
Injury type and severity
Compared with STATS19, TARN provides much more detailed information on the clinical outcomes of those that are relatively more severely injured (that is, meeting the criteria to be captured in TARN). However, TARN does not capture all fatalities (in particular, those that die at the scene of a collision will not appear) or those less severely injured.
TARN records severity of injury using the Abbreviated Injury Scale (AIS), and the associated Injury Severity Score (ISS).
Table 3 shows that comparing e-scooter users with pedal cyclists and motorcyclists, the overall distribution of injuries by severity (for example, AIS score) is similar. A higher proportion of e-scooter casualties had the most serious injuries to the head or limbs than pedal cyclists or motorcyclists, with a lower proportion of chest injuries. Again, the academic research referenced above contains a more comprehensive analysis.
Table 3: TARN patient counts and distribution by road user type and clinical outcomes
| Variable | Value | E-scooter | Pedal cycle | Motorcycle | Other |
|---|---|---|---|---|---|
| Outcome | Alive | 417 (97.4) | 6628 (97.7) | 5845 (97.5) | 209 (99.1) |
| Dead | 11 (2.6) | 157 (2.3) | 151 (2.5) | 2 (0.9) | |
| ISS score | Mean (SD) | 15.2 (9.3) | 14.8 (9.7) | 17.1 (11.4) | 15.2 (9.7) |
| ISS band | ISS 1 - 8 | 47 (11.0) | 1038 (15.3) | 715 (11.9) | 26 (12.3) |
| ISS 9 - 15 | 205 (47.9) | 3167 (46.7) | 2579 (43.0) | 99 (46.9) | |
| ISS > 15 | 176 (41.1) | 2580 (38.0) | 2702 (45.1) | 86 (40.8) | |
| AIS | 1 | 2 (0.5) | 7 (0.1) | 4 (0.1) | 1 (0.5) |
| 2 | 47 (11.0) | 1152 (17.0) | 823 (13.7) | 26 (12.3) | |
| 3 | 222 (51.9) | 3380 (49.8) | 3068 (51.2) | 106 (50.2) | |
| 4 | 89 (20.8) | 1488 (21.9) | 1426 (23.8) | 48 (22.7) | |
| 5 | 68 (15.9) | 753 (11.1) | 672 (11.2) | 30 (14.2) | |
| 6 | 0 (0.0) | 4 (0.1) | 3 (0.1) | 0 (0.0) | |
| Body region | Limbs | 171 (40.0) | 2227 (32.8) | 2088 (34.8) | 65 (30.8) |
| Head | 150 (35.0) | 1304 (19.2) | 612 (10.2) | 63 (29.9) | |
| Chest | 35 (8.2) | 1563 (23.0) | 1564 (26.1) | 42 (19.9) | |
| Multiple | 26 (6.1) | 692 (10.2) | 1090 (18.2) | 17 (8.1) | |
| Other | 46 (10.7) | 999 (14.7) | 642 (10.7) | 24 (11.4) | |
| Length of stay | Mean (SD) | 11.0 (14.7) | 9.3 (13.7) | 12.8 (16.9) | 9.7 (10.2) |
| Critical care | Yes | 90 (21.0) | 1133 (16.7) | 1447 (24.1) | 48 (22.7) |
| No | 338 (79.0) | 5652 (83.3) | 4549 (75.9) | 163 (77.3) |
Other variables
For patients admitted following collisions, TARN contains some information related to the road user behaviour, for example whether there was evidence of alcohol or drug intoxication (recorded via an ‘additional information’ field) and any protection used which in the case of e-scooter users, cyclists and motorcyclists largely records whether a helmet was worn. In STATS19, helmet wearing is currently only recorded for pedal cyclist casualties.
It is already known from the previously published research, relative to pedal cyclists, casualties riding e-scooters are more likely to be intoxicated and less likely to wear a helmet. It should be kept in mind that TARN covers all casualties, including those injured in off-road incidents.
Linkage rates and levels of reporting to police
Linking TARN to STATS19 can provide insight into the proportion of casualties which are known to the police, though some important points should be noted.
Firstly, as noted above, a probabilistic matching method was used for the data linking, which means that both incorrect and missed matches are possible. In particular, where the linkage variables are missing or inaccurate on one or both of the datasets, it is likely to be difficult to establish a linkage even if one exists. While incorrect matches are also possible (for example, where there are two similar casualties within a collision), the linkage method has been designed to minimise these. Therefore, the linkage rates shown below are on the whole likely to be underestimates of the true proportion of common records.
Secondly, the coverage of the two datasets is different. Notably, TARN will include incidents which are outside the scope of STATS19, for example those occurring away from the public highway. While TARN records the incident location (as for example road, pavement or public area) this is not always precise. Again, this is more likely to lead to under-estimation of the proportion of linked records.
Therefore, while this analysis can provide some insight into patterns and trends, the achieved linkage rates should be interpreted with some caution.
Overall matching rate
To assess the completeness of police data for the more serious casualties (as are covered by TARN), we look at the proportion of TARN casualties linked to STATS19. The overall linkage rates are shown in table 4.
This shows that overall, a minority of e-scooter and pedal cyclist trauma patients in TARN are captured in STATS19 - even for this group of relatively more seriously injured casualties. Factors influencing the propensity of a casualty appearing in STATS19 are explored below.
Table 4: Proportion of TARN records linked to STATS19, by road user type
| Road user type | Proportion linked |
|---|---|
| E-scooter | 0.33 |
| Pedal cycle | 0.26 |
| Motorcycle | 0.56 |
Trends over time
It has long been known that police road casualty data under reports the true level of non-fatal road casualties. However, an important question in relation to the use of police data relates to whether the level of reporting has changed over time.
E-scooter casualties began to be recorded in both TARN and STATS19 from around the start of 2020. Chart 1 shows the number of fatal or serious casualties recorded by the police, compared to the number of TARN patients and those common to both datasets based on the results of the data linking. The former covers a much wider range of injuries so, were all incidents recorded in STATS19, would be a higher total. However, it can be seen that the broad trends over time are similar - bearing in mind relatively small numbers - which provides no clear evidence of any change in reporting levels over this time period.
Chart 1: TARN patients, STATS19 Killed or Serious Injured (KSI) casualties and linked records for e-scooter users, by month, 2020 onwards
Factors influencing linkage rates
The likelihood of a casualty in the TARN dataset being linked to a STATS19 record (by proxy, a serious trauma casualty being known to police) depends on a number of factors.
To explore these, a logistic regression model was fitted to the data. This allows the effect of each factor to be estimated adjusting for other factors in the model (assuming that the model is a sufficiently good fit). The results are summarised in table 5, which presents the odds ratios for different factors. In summary this shows that:
- patients with more serious outcomes - notably death - are more likely to be linked
- collisions recorded in TARN with a location of ‘road’ were more likely to be linked to STATS19 (unsurprising as these are more clearly within scope of the police recording)
- where TARN records did not have detailed location information, they were less likely to be linked (in these cases there is less precise information to determine if records are linked)
- a collision in 2022 is less likely to be linked to STATS19 (where the data were provisional at the time of linking)
- age and sex do not appear to affect linkage rates, once other variables are allowed for
The extent to which casualty and collision type influences the likelihood of appearing in the police data can also be explored. The ‘other party’ variable in the table captures the interaction between road user type and whether another party was involved in the collision, and this is considered further below.
Table 5: Proportion TARN records linked, by whether incident location available and other TARN variables
| Variable | Value | Unlinked records | Linked records | Odds ratio (multivariable) |
|---|---|---|---|---|
| Location | Road | 5099 (50.3) | 5029 (49.7) | - |
| Pavement | 370 (90.9) | 37 (9.1) | 0.26 (0.18-0.37, p<0.001) | |
| Public area | 1398 (94.6) | 80 (5.4) | 0.08 (0.07-0.11, p<0.001) | |
| Other | 1109 (92.7) | 87 (7.3) | 0.10 (0.08-0.13, p<0.001) | |
| Postcode? | Postcode | 2746 (50.2) | 2722 (49.8) | - |
| No postcode | 5230 (67.6) | 2511 (32.4) | 0.50 (0.45-0.54, p<0.001) | |
| Age group | Under 20 | 938 (61.5) | 588 (38.5) | - |
| 20 to 29 | 1047 (51.5) | 985 (48.5) | 1.05 (0.88-1.24, p=0.592) | |
| 30 to 39 | 1141 (56.3) | 886 (43.7) | 1.07 (0.90-1.27, p=0.439) | |
| 40 to 49 | 1349 (64.4) | 746 (35.6) | 1.03 (0.86-1.23, p=0.756) | |
| 50 plus | 3501 (63.3) | 2028 (36.7) | 1.10 (0.94-1.28, p=0.245) | |
| Gender | Female | 941 (68.5) | 432 (31.5) | - |
| Male | 7035 (59.4) | 4801 (40.6) | 1.02 (0.87-1.18, p=0.832) | |
| ISS band | ISS 1 - 8 | 1225 (68.1) | 575 (31.9) | - |
| ISS 9 - 15 | 3938 (66.2) | 2013 (33.8) | 1.08 (0.93-1.25, p=0.317) | |
| ISS > 15 | 2813 (51.5) | 2645 (48.5) | 1.62 (1.39-1.88, p<0.001) | |
| Body region | Limbs | 2775 (61.9) | 1711 (38.1) | - |
| Head | 1083 (52.4) | 983 (47.6) | 1.15 (0.98-1.34, p=0.081) | |
| Chest | 2046 (64.7) | 1116 (35.3) | 0.70 (0.61-0.79, p<0.001) | |
| Multiple | 840 (46.5) | 968 (53.5) | 1.14 (0.99-1.32, p=0.074) | |
| Other | 1232 (73.0) | 455 (27.0) | 0.64 (0.55-0.75, p<0.001) | |
| Year | 2020 | 3123 (60.5) | 2040 (39.5) | - |
| 2021 | 3187 (59.8) | 2142 (40.2) | 1.04 (0.94-1.15, p=0.457) | |
| 2022 | 1666 (61.3) | 1051 (38.7) | 0.78 (0.69-0.87, p<0.001) | |
| Outcome | Alive | 7904 (61.3) | 4986 (38.7) | - |
| Dead | 72 (22.6) | 247 (77.4) | 4.02 (2.90-5.65, p<0.001) | |
| Other party? | E-scooter_Yes | 54 (34.6) | 102 (65.4) | - |
| E-scooter_No | 232 (85.3) | 40 (14.7) | 0.13 (0.08-0.22, p<0.001) | |
| Pedal cycle_Yes | 796 (37.4) | 1334 (62.6) | 0.95 (0.66-1.37, p=0.798) | |
| Pedal cycle_No | 4233 (90.9) | 422 (9.1) | 0.08 (0.05-0.11, p<0.001) | |
| Motorcycle_Yes | 2058 (42.1) | 2825 (57.9) | 0.96 (0.67-1.36, p=0.818) | |
| Motorcycle_No | 603 (54.2) | 510 (45.8) | 0.58 (0.40-0.84, p=0.004) |
In this model, road user type and collision type were considered together. Relative to e-scooter users in collision with another party, the odds of linkage as shown in the table are:
- similar for pedal cyclists and motorcyclists involved in collision with another party
- lower for collisions with no other party, particularly for e-scooter users and pedal cyclists
This data suggests that once the collision type is taken into account, the proportion of e-scooter records linked is similar for all three vehicle types considered here, except that in collisions with no other party involved (that is, someone falling off) motorcyclists are more likely to appear in STATS19.
Chart 2 shows the linkage rates by road user and collision type, which further illustrates this. Here the data is restricted to records where postcode of location is available in TARN, and the location type is recorded as road. This is likely to give a more accurate approximation of the levels of reporting to police, as these are cases where good data for linking exists and which are within scope of STATS19.
On this basis, there is no evidence to conclude that the more severely injured e-scooter casualties (that is, those appearing in TARN) are less likely to appear in the police data than pedal cyclists. In both cases, lower overall linkage rates are due to single vehicle collisions which are less likely to appear in police data, even when recorded as happening on a road.
Chart 2: Proportion TARN records linked by nature of incident.
Records with postcode of incident location recorded and location = ‘road’.
It is important to note that the TARN data covers only the relatively more serious road casualties, and generalising to those with less severe injuries is not possible. Given that within the TARN data increased severity of injury is associated with an increase in the proportion of records linked to STATS19, it is however likely that a lower proportion of the casualties not appearing in TARN would be linked - but this remains an assumption.
Analysis of linked records
The dataset of linked TARN and STATS19 data can be used to explore how the collision circumstances (available in STATS19) relate to clinical outcomes (recorded in TARN). It also allows comparison of related variables which are captured in both datasets, for example related to how the road user type is identified, or severity of injury. The following presents a brief initial analysis which could be developed further.
As above, caveats apply in relation to the use of linked data. In particular, where there are discrepancies between the two datasets, this could reflect different coding but could also arise from a false linkage.
Comparison of road user type recording
As a simple illustration, we can look at how the coding of road user type compares within the two datasets. This shows a good correspondence (though it should be noted that road user type is used as a linking variable which may impact on this), though with a greater disagreement for casualties recorded in TARN as e-scooter users. This is unsurprising given that the mode is new and ways of capturing the data were being established in the period covered by this dataset.
Table 6: Comparing road user type recorded in TARN (rows) and STATS19 (columns)
| STATS19: E-scooter | STATS19: Pedal cycle | STATS19: Motorcycle | STATS19: Other | |
|---|---|---|---|---|
| E-scooter | 110 (77.5) | 8 (5.6) | 17 (12.0) | 7 (4.9) |
| Pedal cycle | 2 (0.1) | 1715 (97.8) | 33 (1.9) | 4 (0.2) |
| Motorcycle | 1 (0.0) | 35 (1.1) | 3256 (97.7) | 39 (1.2) |
| Other | 14 (23.0) | 20 (32.8) | 20 (32.8) | 7 (11.5) |
Comparison of injury severity recording
While TARN contains detailed information on clinical outcomes and injuries sustained, STATS19 has only an overall assessment of severity (fatal, serious or slight injury), and where data is collected via the Collision Recording and Sharing (CRASH) system, information on the most serious injury sustained as recorded by a police officer at the scene or within a short time of the collision.
In theory, all casualties appearing in TARN should, by the nature of the TARN inclusion criteria, be recorded as fatal or serious in STATS19. It is however possible that, for example, a casualty could die after 30 or more days, in which case a TARN patient with outcome of dead would be recorded in STATS19 as a serious casualty.
Given this, and the fact that the linkage is not perfect, there is overall a good correspondence between the severity as coded at a high level for this dataset.
Table 7: Comparing injury severity recorded in TARN (rows) and STATS19 (columns) E-scooter, pedal cycle and motorcycle casualties.
| STATS19: Fatal | STATS19: Serious | STATS19: Slight | |
|---|---|---|---|
| Alive | 7 (0.1) | 4498 (89.2) | 535 (10.6) |
| Dead | 237 (95.6) | 10 (4.0) | 1 (0.4) |
Where the injury type is recorded in STATS19, a slightly more detailed comparison can be made, though as the STATS19 coding is restricted to a list of 20 specified injuries some of which are relatively broad (for example ‘deep cuts or lacerations’) it is not always possible to establish whether the coding is consistent with TARN.
Table 8 presents data for those e-scooter casualties in the linked dataset where injury coding is available within STATS19. This shows that when the body region can be determined from the STATS19 injury, for example in the case of head injuries, the police officer coding appears reasonable. However, as noted, there are a non-trivial number of cases in STATS19 where it is not possible to determine which part of the body was injured.
Table 8: Comparing injury type recorded in TARN (rows) and STATS19 (columns) for e-scooter casualties Based on linked records where STATS19 casualties recorded by forces using the CRASH system.
| STATS19: Limbs | STATS19: Head | STATS19: Multiple | STATS19: Other | STATS19: Total | |
|---|---|---|---|---|---|
| Limbs | 15 | 1 | 4 | 20 | |
| Head | 1 | 25 | 2 | 7 | 35 |
| Chest | 1 | 2 | 3 | ||
| Multiple | 3 | 1 | 3 | 7 | |
| Other | 2 | 2 | |||
| Total | 20 | 26 | 3 | 18 | 67 |
The above presents an initial high level analysis, and more in-depth study of the linked dataset would be possible, as outlined as one of the next steps below.
Next steps
This work represents an initial attempt to link police and hospital data for more severely injured casualties in e-scooter and other vulnerable road user collisions, building on the previous feasibility study which established the methodology for linking TARN and STATS19 data. Further analysis is possible, and could include:
- extending this approach to other road user types (that is, vehicle occupants) and a longer time series of data
- more detailed analysis of the types of injuries occurring in collisions involving e-scooters, compared to other modes, and using linked data, the collision circumstances
- estimating the number of road casualties with injuries of AIS of 3 or greater (MAIS 3+) to aid international comparisons
- exploring whether casualty rates (for example per trip, or mile travelled) can be estimated
- potentially using TARN (and linked) data to assess the costs of e-scooter and other road collisions to the NHS
Feedback
Any feedback on this analysis is welcome and can be sent to the road safety statistics team.