Official Statistics

Linking STATS19 fatalities and ONS death registration data

Published 25 September 2025

Applies to England and Wales

About this report

This report links STATS19 police-reported road fatalities with Office for National Statistics (ONS) death registrations data for England and Wales, primarily to validate the STATS19 data on those killed in road collisions, that is to assess how well STATS19 captures fatal road incidents. This represents the first stage of what is planned to be further work to compare reported road deaths in STATS19 with death registrations data, and should be treated as an initial feasibility study which it is hoped will be developed further.

Main findings

This work suggests that, on the whole, STATS19 captures most of the road deaths which are within the scope of the collection, though there are some areas where further work to establish this could be useful. The analysis also highlights deaths outside STATS19’s scope, such as railway incidents and suicides on the road network.

The main findings are as follows:

Completeness of STATS19 fatality records: Most road-related deaths within the scope of STATS19 are successfully captured and matched to ONS death registrations data. The majority of deaths in the ONS data not linked to STATS19 could be outside scope (such as occurring away from the public highway, or resulting from medical episodes). A small number of potentially in-scope fatalities, such as single-vehicle collisions or deaths with ambiguous circumstances, may be missing, though it is hard to draw definitive conclusions based on the information available.

Out of scope records included in STATS19: A small proportion of STATS19 fatalities were later identified as suicides in ONS data (around 3% in 2021 and 2% in 2022). These deaths are considered out of scope of STATS19, but may remain in the dataset if coroner reports are not available at the time at which the annual statistics are produced. Additionally, a sample of STATS19 records not matched to ONS data revealed that some had non-traffic cause codes, including assault and circulatory disorders, indicating further possible mis-classification although more work is needed to explore this.

Deaths outside the scope of STATS19: This work suggests that the majority of road-related deaths are within scope of STATS19, however unmatched ONS records included fatalities clearly outside STATS19 scope, such as incidents involving horse riders (likely on private land), and non-traffic accidents. In addition, based on this data, we estimate that between around 80 and 200 suicides may have occurred on the road network in 2021, depending on classification assumptions. This should be treated as a very tentative finding, with again further work needed to explore further.

Proposed next steps include:

• further work with the ONS to improve the completeness of the death registration data used for this analysis, and further validate the results
• work to improve data linkage to reduce reliance on sensitive data, which would allow comparison of data for earlier years
• improving STATS19 guidance provided to police forces, particularly around off-road and single-vehicle collisions

Introduction

This report summarises the findings of a data linkage exercise between STATS19 police-reported road fatalities and death registration records from the Office for National Statistics (ONS) for England and Wales, covering the years 2021 to 2023. The linkage was undertaken to better understand how well STATS19 captures fatal road incidents, and to explore the nature of deaths that are either missing from or incorrectly included in the dataset.

By comparing STATS19 with ONS death registrations, the analysis provides insight into the completeness and accuracy of police-reported data, and highlights the types of fatalities that fall outside the scope of STATS19 but may still be relevant for road safety and public health.

Aims

The analysis focused on three key questions:

• completeness of STATS19 fatality data – Are there in-scope deaths missing from STATS19?
• accuracy of STATS19 fatality data coverage – Are there fatality records included STATS19 which are not within scope, such as suicides or deaths with non-traffic causes?
• road-related deaths outside the scope of STATS19 – Can we identify deaths that fall outside the scope of STATS19, such as suicides or off-road incidents?

The data received provides some insight into these questions, although not definitive answers.

Data Sources

STATS19: STATS19 is a dataset of police-reported road collisions and casualties, broadly covering collisions occurring on the public highway, involving at least one vehicle and in which there was at least one casualty. Further details of the coverage of STATS19 are available in the STATS20 document, which provides guidance to reporting police officers. For fatalities, known suicides and deaths from medical episodes within a vehicle are out of scope, but are only removed from the STATS19 dataset if the cause of death has been confirmed by a coroner. Figures are published in the September following the year to which the data relates and are not revised, so if the coroner’s verdict is not received by the time of publication, records for possible (but not confirmed) suicides will remain in the data. Fatality names are not a standard part of STATS19 but are provided separately to the department by police forces for the purpose of linking to coroner data.

ONS death registrations data: the Office for National Statistics (ONS) death registration dataset provides detailed information on deaths that have been formally registered in England and Wales. These data are derived from death certificates and include demographic details of the deceased, the date and place of death, and the medically certified underlying cause of death, coded using the International Classification of Diseases (ICD-10). Further information about the data is available from the death statistics published by ONS.

For this analysis, a subset of the ONS death registration data, based on deaths occurring in years up to 2023 and registered up to the end of 2024 was provided by ONS, focusing specifically on deaths attributed to land transport accidents (coded with ICD10-codes V01 to V89) and suicides. These were matched against deaths recorded in the STATS19 dataset. This data was provided under section 42 of the Statistics and Registration Services Act 2007, with approval provided by the Department for Health and Social Care (DHSC).

Linkage method and results

Both STATS19 and ONS datasets should, in theory, contain records for those people killed in road collisions on the public highway. There is no common identifier, but some common variables are held in each dataset which allow a fuzzy matching method to be developed. For this analysis, data for 2021 to 2023 were linked as for these years fatality names are available within STATS19.

Linkage method

To link the STATS19 and ONS datasets, a multi-stage matching process was used to account for variations in data quality and completeness, where:

• the first stage involved exact matches on first name, surname, and age, with the date of death falling within a window of 7 days before to 4 months after the collision
• the second stage allowed for fuzzy matching of names — accounting for minor spelling differences — and a one-year age tolerance, with a narrower date of death window of 7 days before to 2 months after the collision
• the third stage addressed cases where names were reordered or included middle names, or reversing forename and surname when age and date of death aligned. - the fourth stage broadened the fuzzy matching criteria for names but required an exact age match, again using the -7 days to +2 months date window
• the fifth stage enabled matches where names were missing from STATS19, relying instead on age, gender, and a death occurring within 3 days of the accident

In the first instance, STATS19 data was linked to death registrations recorded as transport accidents, with suicides explored separately.

Validation

Although a multi-stage approach was used to allow for them to some extent, several data quality issues affected the matching process itself. Variations in name formatting, such as inconsistent spelling, reduced match rates, particularly where names were absent or incomplete in STATS19. Age discrepancies between the two datasets also contributed to missed matches, especially in cases where age was recorded incorrectly.

However, the record linkage was validated by a manual review, based on data for the first half of 2021, which suggested a very small number of discrepancies (3 out of around 500 records reviewed) which provides confidence that the approach should be sufficiently robust for this feasibility study.

Limitations

Several factors affected the success and interpretation of the data linkage between STATS19 and ONS death registration records. These limitations help explain discrepancies in fatality counts and matching outcomes across the datasets.

One of the most prominent issues is the consistent difference in the number of fatalities recorded. For instance, in 2021, STATS19 reported approximately 1,400 road-related deaths in England and Wales, while the ONS dataset included only around 1,200 transport accident fatalities (see chart 1). This gap can be attributed to three key factors:

Scope and inclusion criteria: The ONS dataset used for this analysis includes only deaths that have been formally registered and assigned a medically certified underlying cause. In this analysis, only deaths coded as transport accidents were included. Deaths with ICD-10 code U50.9, typically used for cases involving an adjourned coroner inquest, were excluded. STATS19, by contrast, includes all police-reported road traffic fatalities, regardless of registration status or cause determination.

Cause of death misclassification: Some fatalities recorded in STATS19 were registered under a different cause of death in the ONS dataset. A sample of 40 STATS19 fatalities from 2021 was reviewed by ONS. The majority of these had an underlying cause code of U50.9, which as noted above, relates to awaiting coroner inquest and thus means the cause of death was not available for this analysis. However, around a quarter (10 of 40) had other ICD-10 codes. This suggests that some road-related deaths may be classified under non-transport causes, such as medical events or undetermined external factors, which fall outside the scope of this analysis, but that pending investigations are a larger factor in the non-matching cases. Further work is planned to explore this issue in more detail.

Registration delays: The ONS dataset is based on death registrations, not occurrences. Deaths that occurred in a given year but were subject to significant registration delays may not be included in the dataset, though this is more of an issue for the more recent years. In particular, for 2023, this resulted in a pronounced drop in ONS records not observed in STATS19, significantly impacting the success of data linkage for that year and the trend shown in chart 1.

Chart 1: Fatality counts in STATS19 compared to transport accident fatalities in ONS death registrations, England and Wales, by year

Data linkage outcomes

Across the three years analysed (2021 to 2023), the proportion of STATS19 fatalities successfully matched to ONS death registrations with an underlying cause of transport accident varied:

• in 2021 there were 1,026 matched records, leaving 338 unmatched ONS fatalities and 446 unmatched STATS19 records
• in 2022 matched records remained stable at 1,035, with a modest increase in unmatched records from both datasets
• in 2023 matched records dropped to 682, while unmatched STATS19 records rose to 768, indicating that reduced data availability from ONS impacted linkage success (likely related to potential registration delays as noted above)

The figures for each year are shown in chart 2, with the above limitations kept in mind when interpreting these results. While it is possible that missed matches could contribute to both unlinked ONS and unlinked STATS19 counts (thus introducing a degree of double-counting), a manual review of 2021 records suggests that many of these unlinked cases are genuinely different records within the two datasets. However, some uncertainty remains, and further work would help to confirm this finding.

Chart 2: Number of matched and unmatched fatalities, 2021 to 2023

Completeness of STATS19 fatality data

This section relates to the first aim of this work, and assesses how well STATS19 captures road-related fatalities by comparing it with ONS death registration data. The focus is on identifying deaths within the ONS dataset recorded as from transport accidents that might fall within the scope of STATS19 but are not linked.

Potentially in-scope deaths in ONS

To explore whether STATS19 is missing in-scope fatalities, unmatched ONS records were manually reviewed, in particular looking at the underlying cause of death, which for transport accidents has an ICD-10 code starting with V (‘V codes’).

Of the 2021 unmatched records, 265 had ICD-10 V codes that may fall within STATS19 scope. Based on text from the coroner’s inquest (where available) these records were classified into 3 categories including:

• potentially in scope – no clear evidence to exclude them from STATS19 (though this does not necessarily mean that they were, as the coroners notes may not provide all relevant information)
• out of scope – included indicators such as medical episodes, off-road incidents, or deaths occurring more than 30 days post-collision
• inconclusive – lacked sufficient detail to determine inclusion

Table 1 summarises the top 5 V-codes reviewed. The most frequent was V89.2, with 95 records, though only 12 had enough information to assess scope. V47.5 had the highest proportion of potentially in-scope fatalities, suggesting that guidance on single-vehicle collisions may need clarification, although the number of records here would still be a small proportion of single-vehicle collisions within STATS19 and as noted.

These findings highlight the challenges of determining STATS19 relevance from death registration data, particularly where coroner notes or contextual details are limited.

Table 1: Manual review of most common potentially in-scope ICD-10 V-codes for 2021 data

ICD10 code	Description	Total records	Potentially in scope	Out of scope	Inconclusive	Notes
V89.2	Person injured in unspecified motor-vehicle accident, traffic	95	9	3	83	Out-of-scope reasons: 2 deaths >30 days post-accident, 1 on driveway. Remaining records lack coroner notes.
V48.4	Car occupant injured in non-collision transport accident while boarding or alighting	22	1	21	0	Out-of-scope due to parked vehicle or medical episode.
V47.5	Car occupant injured in collision with fixed or stationary object: driver injured in traffic accident	15	11	4	0	Out-of-scope due to medical episode or incident on private roads.
V03.1	Pedestrian injured in collision with car, pick-up truck or van: unspecified whether traffic or non-traffic	13	7	3	3	Out-of-scope reasons: 2 deaths >30 days post-accident, 1 in parking area.
V18.4	Pedal cyclist injured in non-collision transport accident: driver injured in traffic accident	11	1	4	6	Out-of-scope due to off-road incidents or medical episodes.

Implications for data completeness

The presence of potentially in-scope deaths in ONS records that do not appear in STATS19 suggests that some road-related fatalities may be missing from police-reported data. These gaps could arise from a range of factors, including the misinterpretation of inclusion criteria, though they could also reflect inaccurate coding within the death registration data (for example, in relation to whether was on or away from the public highway).

However, the overall findings indicate that STATS19 performs reasonably well in capturing road traffic fatalities. The majority of deaths that fall within its scope are successfully recorded and matched to ONS registrations. While there is room for improvement, particularly in clarifying guidance and reducing ambiguity around edge cases, the data suggests that STATS19 provides a reliable account of fatal road collisions in England and Wales.

Accuracy of STATS19 fatality data coverage

This section relates to the second aim of the work, and explores the accuracy of STATS19, specifically whether there are records in STATS19 that may fall outside the scope of STATS19 such as suicides or those with non-traffic causes which remain in the dataset once it has been finalised.

Matched suicides

Deaths on the road network that are later determined by a coroner to be suicides are considered out of scope of STATS19. These records are removed from the dataset once the coroner’s report is received. However, if the report arrives after the department’s annual reporting deadline, no further amendments are made to that year’s data. As a result, some suicide-related deaths may remain in STATS19.

To estimate the scale of this issue, unmatched STATS19 records were linked to deaths in the ONS dataset which were officially registered with suicide as the underlying cause. The results show that:

• in 2021, approximately 3.13% of fatalities in STATS19 were linked to ONS fatalities with “suicide” as the final underlying cause of death
• in 2022, approximately 2.11% of fatalities in STATS19 were linked to ONS fatalities with “suicide” as the final underlying cause of death

These proportions suggest that while the number of suicide-related deaths retained in STATS19 is relatively small, it is not negligible. Their presence may slightly inflate fatality counts and could affect trend analysis, particularly in categories such as single-vehicle collisions.

Other potential misclassifications

In addition to suicides, other types of misclassification may affect STATS19 data quality. A random sample of 40 of the 446 unmatched STATS19 records for 2021 was reviewed against ONS death registrations. The majority of these were found to have a code of U50.9, denoting that a coroner inquest has been adjourned and therefore the underlying cause code is not available.

However, 10 of these were found to have non-traffic ICD-10 cause codes, such as COVID-19, assault-related incidents and circulatory system disorders. These cases may reflect deaths that occurred on the road network but were primarily caused by medical or external factors not directly related to traffic collisions. They warrant further investigation to determine whether they were correctly included in STATS19 or should potentially have been excluded.

Conclusion

This analysis suggests that there are fatalities recorded within STATS19 which are likely to be out of scope, and provides a potential basis to broadly quantify this in future. Depending on user feedback, this information could also be used to make revisions to STATS19 fatality figures for future years, where cases are known to be out of scope.

Road-related deaths outside the scope of STATS19

This section examines fatalities recorded in the ONS death registration dataset for land transport accidents that fall outside the scope of STATS19. These deaths are not expected to be captured in police-reported road traffic statistics due to the nature of the incident or location, but this analysis may provide a broad estimate of the number of such cases relative to the number of deaths in collisions on the public highway. In particular, it may allow a broad estimate of the number of road-related suicides to be made.

Types of out-of-scope fatalities

In the un-matched ONS records for 2021, there are 73 records assigned ICD-10 codes that are out of scope of STATS19. Table 2 presents the ICD-10 V codes most frequently assigned to unmatched ONS death registration records from 2021 that are out of scope of STATS19. These codes primarily relate to incidents involving trains, horse riders, and cyclists in non-traffic accidents. Their presence in the unmatched records highlights the types of fatalities that fall outside the scope of STATS19 reporting and their prevalence.

Table 2: Top out-of-scope ICD-10 V-codes assigned to unmatched ONS records in 2021

ICD10 code	Description	Count 2021
V05.9	Pedestrian injured in collision with railway train or railway vehicle: unspecified whether traffic or nontraffic accident	17
V80.0	Rider or occupant injured by fall from or being thrown from animal or animal-drawn vehicle in non-collision accident	12
V18.0	Pedal cyclist injured in non-collision transport accident: driver injured in nontraffic accident	5
V03.0	Pedestrian injured in collision with car, pick-up truck or van: nontraffic accident	4
V09.0	Pedestrian injured in nontraffic accident involving other and unspecified motor vehicles	3
V83.5	Driver of special industrial vehicle injured in nontraffic accident	3
V17.0	Pedal cyclist injured in collision with fixed or stationary object: driver injured in nontraffic accident	2
V28.0	Motorcycle rider injured in non-collision transport accident: driver injured in nontraffic accident	2
V68.0	Occupant of heavy transport vehicle injured in non-collision transport accident: driver injured in nontraffic accident	2
V83.7	Person on outside of special industrial vehicle injured in nontraffic accident	2
V84.7	Person on outside of special agricultural vehicle injured in nontraffic accident	2

Suicides on the road network

While suicides are considered out of scope for STATS19, some occur on the road network and may be relevant for broader safety and policy discussions. This section estimates the number of suicides occurring on roads using matched STATS19 records and the available ONS death registration data. This should be considered as an initial analysis which would benefit from further refinement, and the limitations of the data noted above should be kept in mind - in particular, there may be cases where the coroner verdict has not been recorded within the ONS dataset, which would mean that these figures would likely be under-estimates.

Estimates of road-related suicides can be made based on the underlying cause of date and the place of occurrence recorded in the ONS data. As noted above, a number of suicides are linked to records in STATS19 and these can be used to make some assumptions as to which codes might be relevant. However, as some of the codes refer to unspecified categories, it is not possible to produce a definitive estimate, and instead upper and lower bounds are estimated.

For example, between 2021 and 2023, 76% suicides matched to STATS19 records had a recorded place of occurrence as street and highway in the ONS death registration data. This suggests that location, when used in conjunction with ICD-10 codes, can support the identification of suicides likely to have occurred on the road network.

Table 3: Counts of matched suicides by place of accident, 2021 to 2023

Place of occurrence	Count
Street and highway	67
Other	20
Home	1

Similarly, the cause code can be examined for the matched suicide records. The most frequently observed ICD-10 code among matched suicides is X84 (intentional self-harm by unspecified means), with 18 instances occurring on streets and highways. However, this code also appears in records with unknown or other locations, introducing uncertainty when estimating an upper bound.

Table 4: Top 10 unique combinations of ICD-10 codes and location for matched suicides, 2021 to 2023

ICD10 code	Place of accident	Count
X84: Intentional self-harm by unspecified means	Street and highway	18
X81: Intentional self-harm by jumping or lying before moving object	Street and highway	15
X82: Intentional self-harm by crashing of motor vehicle	Street and highway	14
X84: Intentional self-harm by unspecified means	Unknown	6
Y31: Falling, lying or running before or into moving object, undetermined intent	Street and highway	6
X80: Intentional self-harm by jumping from a high place	Street and highway	5
X84: Intentional self-harm by unspecified means	Other	5
Y32: Crashing of motor vehicle, undetermined intent	Street and highway	4
X82: Intentional self-harm by crashing of motor vehicle	Unknown	3
X82: Intentional self-harm by crashing of motor vehicle	Other	2

Based on this, together with manual review of the text descriptions where provided, 3 estimates were produced to reflect different assumptions about scope, namely:

• lower bound: includes matched STATS19 suicides and unmatched ONS records with ICD-10 codes X82, X81, or X80, and place of occurrence as street and highway
• realistic upper bound: adds unmatched ONS records with ICD-10 codes X82 or Y32, regardless of location, and any records with relevant ICD-10 codes occurring on street and highway
• extreme upper bound: further includes unmatched ONS records with ICD-10 code X84 and unknown or undefined location

These estimates suggest for 2021:

• lower bound: 78 suicides
• realistic upper bound: 110
• extreme upper bound: 203

And equivalently for 2022:

• lower bound: 65 suicides
• realistic upper bound: 101
• extreme upper bound: 200

Conclusion

These figures provide insight into the potential scale of suicides occurring on the road network, which are excluded from STATS19 but may be relevant for understanding broader road safety and public health risks, though further review of the data, together with other sources, would be valuable to assess their validity. If these estimates could be refined, then this approach may provide a basis for estimating the annual number of road-related suicides in future.

Overall conclusions

This data linkage exercise has explored how well STATS19 captures road-related fatalities and where gaps or misclassifications may occur. The analysis shows that STATS19 performs well overall, with most in-scope deaths successfully recorded and matched to ONS death registrations. However, a small number of potentially relevant fatalities — such as single-vehicle collisions or deaths with ambiguous circumstances — may be missing due to scope interpretation or reporting inconsistencies (in either dataset).

Conversely, a limited number of records remain in STATS19 that may fall outside its intended scope. These include suicides, which are excluded only if the coroner’s report is received before the reporting deadline, and deaths with non-traffic causes such as medically related. While suicides are clearly out of scope, the inclusion of other non-traffic causes is less straightforward and warrants further investigation to determine whether these deaths were appropriately recorded or should be excluded.

The analysis also identified fatalities that fall entirely outside the scope of STATS19, such as off-road incidents, railway collisions, and suicides on the road network. Although these are not captured in police-reported data, they may be relevant for broader road safety and public health discussions.

Further work

Further work and improvements to the linkage process and data quality are planned, to build on this initial feasibility study. These include:

• introducing fuzzy matching on postcodes to reduce reliance on names, which are sensitive and often inconsistently recorded, which would allow a longer time series of data to be considered
• further exploring the records in STATS19 which have not been linked to transport accidents or suicides in the ONS dataset
• further exploring the scope to refine estimates of road-related suicides
• considering whether it is possible to clarify guidance provided to police forces — for example around single-vehicle collisions that begin on-road but end off-road

These steps could improve understanding of the quality of the important element of STATS19 data related to road fatalities, and improve consistency in how such cases are recorded in STATS19 in order to help ensure the dataset continues to provide a robust and reliable picture of fatal road collisions in England and Wales.

Feedback

Any feedback on this initial analysis is welcome, and can be provided to the road safety statistics team by email to roadacc.stats@dft.gov.uk.

Annex: Data linking methodology, further detail

To link STATS19 police-reported fatalities with ONS death registration records, both datasets were cleaned and a multi-stage matching process was applied to account for variations in data quality and formatting.

Matching stages:

• match 1: Exact match on first name, surname, and age. Date of death within -7 days to +4 months of the accident
• match 2: Fuzzy match on first name and surname, with age ±1 year. Date of death within -7 days to +2 months
• match 3: Reordered or extended names (for example, “John Turner” matched to “John Matthew Turner”), or reversed forename and surname combinations, with age and date alignment
• match 4: Increased fuzzy match distance on names, requiring exact age match. Date of death within -7 days to +2 months
• match 5: Records with missing names in STATS19 matched using age, gender, and date of death within 3 days of the accident

Results for 2021:

• ONS had 1,248 road traffic death records (post-cleaning)
• after Match 1, the match rate was 64%
• match 2 added 69 records, increasing the match rate to 70%
• match 3 added 6 records
• match 4 added 38 records, bringing the total to 920 matches (73%)
• match 5 added 13 records, resulting in a final match rate of 75%

Instructions for printing and saving

Depending on which browser you use and the type of device you use (such as a mobile or laptop) these instructions may vary.

You will find your print and save options in your browser’s menu. You may also have other options available on your device. Tablets and mobile device instructions will be specific to the make and model of the device.

How to search

Select Ctrl and F on a Windows laptop or Command and F on a Mac.

This will open a search box in the top right-hand corner of the page. Type the word you are looking for in the search bar and press enter.

Your browser will highlight the word, usually in yellow, wherever it appears on the page. Press enter to move to the next place it appears.