Individuals involved in the criminal justice system are at higher risk of suicide than the general population. This is supported by international research on prisoners showing elevated suicide rates versus the general population, with rate ratios typically higher than 3 in male prisoners and 9 in female prisoners when accounting for age [1, 2]. This elevated risk is not limited to time served in custody, with evidence from an international meta-analysis indicating the risk of suicide is also substantially increased for released prisoners [3–5]. Figures from England and Wales in recent years have further exposed this issue where 2016 saw the highest number of self-inflicted deaths in prisons since records began and in 2019 the rate remains above 1 in 1,000 prisoners annually [6, 7]. Suicides following police custody have also continued to rise [8].

Several countries have included prisoners in their national suicide prevention strategies as a priority (US [9], Guyana [10], Ireland [11]), to work toward reducing high suicide rates in prisons globally. In England, the fourth progress report of the cross-government outcomes strategy to save lives highlights that the prevention of suicide among people in contact with the criminal justice system (CJS) remains a high priority, reiterating the importance of lessons that can be learned when deaths have occurred in police, prison or other custodial settings [12]. Political concerns regarding deaths occurring in custody, though, routinely outweigh any attention paid to suicides of offenders that occur in non-custodial settings, even though responsibility still rests under the umbrella of the CJS (i.e. individuals engaged in community police contact, court processes and probation supervision). This stands despite evidence that individuals within this group of ‘community-based offenders’ are also at a higher risk of suicide¹⁴, where more than one in ten of general population suicides had community criminal justice contact before death [13].

It has been suggested that deaths outside custodial settings by people involved in the criminal justice system are less scrutinised than deaths occurring in prison custody [14], potentially influenced by a number of sociological and policy factors as well as methodological issues in collecting this data. One of the key methodological factors is the difficulty in simply identifying deaths under community supervision, where recording and reporting practices may be unreliable compared to the frequent and official reviews of deaths in custodial settings [14]. This leads to consideration of developing new methods of accurately ascertaining the prevalence of such deaths in community settings. Each contact with a CJS organisation is an opportunity for intervention, representing significant potential for suicide prevention. It follows that a clearer understanding is required of the prevalence of suicides of all individuals across the criminal justice pathway. One solution potentially lies in data that is already being routinely collected.

All CJS organisations increasingly use electronic systems to record key and relevant details about individuals who come into contact with their services, with many such organisations accessing and entering information into more than one system as part of their routine record-keeping. For example, after arrest, an individual’s details may be entered by a police custody sergeant into both local and national police databases, then by the courts’ mental health liaison and diversion team into a clinical database. In many cases they would also be recorded by a third-sector worker into their electronic system, as is the case for some substance misuse services. Within the UK’s prison and probation services, multiple systems are used to record service user data and only a small number of these systems ‘talk to each other’. This fractured technological landscape is not limited to the UK; the US lacks a comprehensive joined up framework for criminal justice data [15] and studies of the Australian CJS have established a number of legal, ethical, organisational and technical challenges of using routine administrative data for research purposes [16, 17].

However, linkage on a large scale has been, and continues to be, successful in Western Australia where the West Australia Data Linkage System (WALDS) has been running for over 20 years and now spans over 50 health and government datasets [18]. Linkage in Australia more specifically between CJS and mortality data has been established for research purposes with Registrar-General Death records [19] and the National Death Index [20]. In the UK, smaller scale linkage with police data has been successfully implemented recently with substance misuse data [21, 22].

Effective identification and monitoring are key themes in improving outcomes for individuals at risk of suicide [23], supporting the rationale for investigating ways of developing a more connected knowledge base.

The objective of the current study was to review the electronic systems used by the UK’s CJS organisations (i.e. police, prison and probation) and scope out the methodological considerations, permission pathways and technical challenges of linking all datasets with ONS mortality data, at the level of the individual. The ONS already routinely records details of all registered deaths by suicide in the UK (sex, gender, age, geographical area and method of suicide) which would provide the required identifiers for linkage with CJS systems.

Such a linkage would potentially provide a comprehensive insight into individuals who die by suicide who have had contact with the CJS. Details would include demographics of the population, contact with services, mental health needs, their chronological pathway through the CJS, assessed level of risk and, importantly, discrepancies between datasets (e.g. the prison system had recorded a suicide risk but this was not recorded after release on probation systems).

This paper also considers challenges in the use of routinely collected CJS data for research purposes internationally. Further, it could demonstrate the value of a linked resource specifically for suicide prevention strategies by offering data on potential intervention timepoints for this population (e.g. how long before death was last contact and with which service?), and observations on disparity of information held across different organisations (e.g. did police custody record a self-harm marker which was not recorded on reception to prison?), which could indicate where more effective information-sharing policies are required to assist risk management.

In 2018 and 2019, the research team investigated the feasibility of linking the ONS suicide data with the following five UK criminal justice datasets:

Detailed scoping discussions were initiated with the respective data teams/research correspondents through either enquiry mailboxes, contact details on publications or, where available, use of existing professional networks within the wider research team. Email was the primary method for correspondence, though this was supplemented with telephone calls for further clarification. Drafts of information were sent to teams for accuracy checking and feedback before inclusion as definitive findings.

After explaining the purpose of the study, the CJS data teams were first asked if a version of data linkage was at the very least possible. If so, full lists of data items were requested; where these were not available to be shared, a list of examples of required data items compiled by the wider research team were sent to clarify whether or not they were recorded. Data items of potential interest from each dataset were collated and refined as full lists were made available or individual data items were confirmed as extractable. Supplementary Appendix 1 shows the resultant list of available data items across each of the datasets considered.

Clarification was also sought from the data teams on which identifiers would be needed for linkage and what unique identifier is used within each dataset. Details were sought on the full process of linkage, including any approval processes for access, costs associated, understanding of available data items, timescales, transfer of data and secure data storage. Each data team was also questioned about the quality of the data held (i.e. completeness of fields, date ranges, if they are mandatory, if they rely on self-report or are validated in any way).

Contact was successfully made with all CJS data teams, with all teams expressing that, in principle, a version of the proposed data linkage was possible. To begin with, permission to access ONS UK mortality data would need to be secured via the Research Accreditation Service [24].

The potential linkage of five criminal justice datasets in England and Wales was investigated, with a focus upon three key considerations: feasibility, value, and cost. Regarding feasibility, police data would provide the minimum required linkage. The current scoping exercise clarified that linkage of ONS mortality data with only Perito or the PNC extract would likely exclude a significant proportion of relevant individuals (deaths not referred to the IOPC and arrests not resulting in caution or conviction, respectively), therefore both Perito and PNC should be included within the linkage exercise. Linkage with the ‘spine’ would also provide unique identifiers for subsequent data linkage with the OASys, nDelius and P-NOMIS datasets, thus making record-matching much more efficient, with fewer false-positives or one-to-many matches. We conclude that the proposed linkage with these five datasets would be eminently feasible.

In considering the value of such a data linkage exercise, a comprehensive list of available data fields could not be obtained for all datasets. However, confirmation was received from each of the CJS data teams that, as a minimum, the provisional requested data items are populated and could be included in a shared extract. Supplementary Appendix 1 details this provisional list of variables and from which dataset each variable could be extracted; illustrating a breadth of key risk factors that this linkage could report on (e.g. days from release from custody to death, self-harm history). Measures of data completeness were not available for most datasets, but teams offered a general response of high quality for the data fields requested, with a few exceptions (i.e. prison visits were noted as not consistently recorded by establishments and the collection of some demographic items were not mandatory). On this issue of feasibility this is generally promising, however obtaining reliable quantified measures of data completeness for key variables would be a priority in the commencement of such a linkage, given the integral importance to the value of the output. Of note with the data items is that all datasets record basic demographics, indicating duplication of inputting for different services in relation to the same individual. There are obvious implications in terms of potential for inaccuracies and use of resource, leading to a suggestion that solutions such as shared databases or improved ‘communication’ between different CJS datasets could significantly streamline parts of the process.

An additional benefit of linkage of the five CJS datasets is that no costs would likely be incurred for the decision, linkage or delivery of the extract (see timescales and cost subsections in Appendices; note this relates to any fees levied by the data teams for each organisation and does not account for costs associated with researcher time and work). Further to this, governance for the prison and probation datasets is provided through one approval process via the HMPPS National Research Committee (NRC), with no additional research governance approvals required to request access to police data (only the negotiation of access to the Data Sharing Agreements with the data teams). This should minimise the time spent securing access to the data, with a predicted turnaround of no more than 12 weeks once all access agreements are in place.

A foreseeable issue with the data linkage is the common use of aliases within the criminal justice population. For a number of different reasons, including typing errors, intoxication, deliberate misinformation and nicknames [26–28], there is a relatively high incidence of aliases, with Vollm [27] reporting an incidence of between 17% and 46% in a forensic psychiatric sample. A recent Australian study compared the quality of linkage with a youth justice cohort with mortality data including or excluding aliases and concluded that not including aliases excluded 14% of deaths [29]. Their findings illustrate the benefit of lowering the threshold for verification of any future criminal justice data linkage, considering the high prevalence of aliases in this population. It also further supports any recommendation for improved communication between systems or a shared database given that the police, probation and prison systems all have the option to record aliases.

The data linkage proposed here offers the potential to provide a breadth of epidemiological evidence on contact rates with UK CJS services amongst those who die by suicide, as well as characteristics of this group. Making such evidence available to the CJS services, could then inform operational risk management priorities and improve early identification of at-risk individuals. The utility of such a dataset could be expanded to identify risk and protective factors in this population with the inclusion of a comparison group of justice-involved individuals who did not die by suicide. Future research might also consider the best way to identify an appropriate comparison group so that risk factors can be examined.

Successful linkage, as outlined in this paper, could subsequently provide a foundation to extend to include linkage with healthcare and other relevant governmental datasets. Primary healthcare services, secondary healthcare services, and specialist mental health and substance misuse services could significantly supplement the data garnered from the CJS linkage exercise, which would potentially contribute even more detailed and verified information on the individuals’ health, NHS service usage and lifestyle. For instance, although most CJS datasets would offer opportunity to record if an individual has experienced drug or alcohol misuse, this may often rely on self-report and would not reliably contain details of their history or any treatments offered or undertaken. As seen in Supplementary Appendix 1, there are a number of other key health related variables that are not contained within CJS datasets, further supporting a subsequent more comprehensive linkage with healthcare data in the future. Australian data linkage studies including the National Coronial Information System were able to provide further information on cause of death [21]. In the UK, if a case is investigated by a coroner, these details are used as standard in the ONS, however further detail on the causes of death are rarely provided [30]. Inclusion of UK coronial data to further improve research as has been done in Australia could be useful, but this may require a more manual, qualitative approach to enhance the dataset, rather than through data linkage.

In conclusion, the CJS data linkage proposed here offers a realistic plan with a relatively short-term output but leaves plenty of scope to follow up with subsequent healthcare dataset linkage, which may offer a fuller picture of this under-researched population. Further, it could assist in identifying key points at which to target interventions for suicide prevention across the spectrum of services, many of which would likely be vital in identifying this risk and/or delivering the intervention.

No particular permissions were deemed necessary for this scoping exercise, however, it formed part of a wider project funded by the National Institute for Health Research Policy Research Programme; ‘Managing the Risk of Self-Inflicted Death to Support a Psychologically Safe and Rehabilitative Criminal Justice Pathway’, PR-R16-0516-21003.

The authors declare that they have no conflicts of interests.

AD conducted the scoping exercise with all data teams and prepared the manuscript.

DP was the academic lead on the study and provided guidance and advice on the scoping exercise.

JSh, JS, DP, SF, AP and TW developed the protocol for the investigation as part of the funding bid for the wider project.

VW, JS, JSh, SF, AP, TW and DP provided feedback on draft sections of the paper.

JSh was responsible for overall supervision of the study.

Thanks go to the individuals and teams who assisted with queries in the development of this feasibility study, including representatives from:

National Confidential Inquiry into Suicide and Safety in Mental Health (NCISH)

This report is independent research funded by the National Institute for Health Research Policy Research Programme (‘Managing the Risk of Self-Inflicted Death to Support a Psychologically Safe and Rehabilitative Criminal Justice Pathway’, PR-R16-0516-21003). The views expressed in this publication are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.