Modelling small-area rates of self-harm in London

The purpose of data processing is to complete a research project examining how rates of self-harm vary between different neighbourhoods and populations within London, as outlined in more detail below. The legal justification for this comes from Articles 6(1)(e) and 9(2)(j) of the GDPR.

The purpose for processing meets the requirements for Article 6(1)(e) as the data is necessary to complete the project outlined and doing so is in the public interest as it will produce findings that will usefully inform our understanding of population and area-level predictors of self-harm rates. As outlined in section 5c, these findings will be shared with audiences in mental health services and public health departments within London in order to inform provision of preventative interventions and services for those who self-harm, promoting the public interest.

In terms of Article 9(2)(j), the project described below represents a scientific research purpose which the processing described is necessary to complete. The data requested and processing used is proportionate in that it only involves data relevant to the scientific question being investigated and it uses an anonymised data set and reports findings at area and group level such that individual subjects are not identifiable to the researcher carrying out the work. Additional measures taken to ensure the security of the data are outlined below.

South London and Maudsley NHS Trust and King’s College London require HES Admitted Patient Care (APC) data for use in a project entitled ‘Understanding variations in self-harm rates between deprived areas in London’.

This study forms the final part of a PhD project examining variations in the rates of self-harm between small-areas in London. The applicant is employed by King’s College London (KCL) and registered as a PhD student within the Department of Psychological Medicine at KCL. They also do clinical work as a psychiatrist for South London and Maudsley NHS Foundation Trust (SLaM) and has a clinical addendum to their contract with KCL which establishes an honorary contract for this work.

The National Institute for Health Research (NIHR) Maudsley Biomedical Research Centre (BRC) is part of SLaM and works in partnership with the Institute of Psychiatry, Psychology & Neuroscience at King's College London. Together they aim to develop more individualised treatments and support advances in the prevention, diagnosis, treatment and care of mental ill health and dementia. To do this, they bring together researchers, clinicians, allied health professionals and service users from across the University/Trust partnership to work together better in order to meet the challenges of finding better treatments and improved care for patients. Staff working within the BRC are SLaM employees. From May 2021, all data for BRC projects is to be held on the Microsoft Azure Cloud and is only accessible to substantive employees of SLaM and individuals employed outside SLaM that hold honorary contracts with SLaM. Data is currently stored on physical servers at SLaM but is to be destroyed once the move of data to the cloud is complete (scheduled May), at which point, a certificate of destruction will be sent to NHS Digital. No data will be stored by KCL.

The study was devised by the applicant as part of the work towards the PhD they are completing at KCL. The PhD is funded through a clinical research training fellowship awarded to the applicant by the Wellcome Trust.

Work prior to this study is currently using a clinical dataset already held by the Maudsley BRC, of non-NHS Digital data, covering an area of South East London. This work is testing associations between area exposures and rates of hospital admission for self-harm for people aged 11 and over living in four London boroughs. Subsequent work will build a model that aims to predict rates of self-harm based on area-level variables.

Under the original agreement, the study aimed to:

1) Describe the distribution of self-harm hospital admission rates across small-areas of Greater London and over time and compare them to the distribution of all admissions to identify any self-harm specific spatial patterning.

2) Explore the impact of using different definitions of self-harm, for exampling including injuries and poisonings coded as of undetermined intent or accidental, on associations with self-harm rates and the geographical patterning of self-harm rates.

3) Test the validity of a model using area and population level exposures to predict areas with high and low rates of self-harm.

As the research progressed, patterning of rates of self-harm were identified by analysis of the data, comparing different neighbourhood and communities as described in the original Agreement, but it was found that they do not produce a predictive model. Hence, aims (1) and (2) described above were met but aim (3) was abandoned. Aside from the predicative model being dropped, all other purposes stated in the original application remain the same. No additional purposes of data dissemination have been added.

The data provided by NHS Digital is pseudonymised Hospital Episode Statistics Admitted Patient Care data. This will be used to identify the outcome of interest for the study: admission to hospital for self-harm. The NHS Digital data will not be linked with any other data sets and there will be no attempt to identify individuals. Data is required for individuals living in Greater London at the time of admission, for the years 2008-2017 as this is the study area and period of interest. Data for all individuals aged 11 or older is being requested so that the dataset the model is being tested on matches the clinical data that was used to build the model in this regard. Data for younger individuals is not being requested as incidents of self-harm are rare prior to adolescence. Data would be at admission level to allow the calculation of admission rates. Fields required include age in years and sex to allow standardisation of rates, lower super output area (LSOA) of residence to allow small area rates to be calculated, ethnicity, date of admission, hospital of admission and ICD-10 diagnostic codes. Efforts have been made in data minimisation to meet the minimum level of data required to be able to carry out the study. Only fields necessary to perform the study, such as those relating to self-harm-related admissions, have been selected. Any fields irrelevant to the study have been removed from the selection. In addition, all planned admissions have been removed from the request, on the assumption that the same proportions hold for Greater London 2008/09-2017/18 as nationally for 2017/18, this will minimise the number of Admitted Patient Care records significantly.

Careful consideration has been made to ensure only the minimum level of data has been requested and that there is no alternative, less-intrusive way to achieve the purpose of this study and the subsequent benefits to healthcare.

Data flows:

The data requested is pseudonymised and is not linked to any other data. Hence there will not need to be any data provided to NHS Digital. HES APC data has previously been provided by NHS Digital to SLaM. It is currently being stored within the SLaM network and there has been no subsequent flow of data from there. As of May 2021, the data will be moved from this current storage location at SLaM to the Microsoft Azure Cloud. Data at the SLaM location will be permanently deleted and data on the Cloud will be accessed via individual user accounts and only by individual's who are substantively employed by SLaM. No data will be received or stored by KCL and KCL will not have access to the data. Data will not be transferred to other locations and will not be made available to others within SLaM who are not working on the project described in this application. SLaM will not link the data disseminated by NHS Digital to any other data they may already hold.

The HES data is currently hosted by SLaM within the Maudsley BRC. Pseudonymised data is stored on the SLaM server in an SQL server database on a storage area network and secured using an active directory user group, with access restricted to named individuals according to SLaM’s security policy. Remote access to the database is permitted, but only via a virtual private network accessed using a secure token, such that actual data processing is still carried out on site. Local downloading and printing of data is disabled when it is being accessed via VPN. This Data Sharing Agreement will allow the data storage to be relocated to the Microsoft Azure Cloud, with access restricted to named individuals according to SLaM’s security policy.

Who can access the data:

Technical assistance with data downloading and processing will be provided by staff within the Maudsley BRC with expertise in database management and information governance who are authorised to access the data for the purpose(s) described, all of whom are substantive employees of SLaM.

All other processing activities will be undertaken by the applicant or research assistants / postgraduate students under her direct supervision. These supervised individuals have not yet been appointed but it is expected that there will be three research assistants / postgraduate students with access to the data under the applicant's supervision. The processing activities of these research assistants / postgraduate students under the applicant’s supervision will be data analyses describes in points (1) and (2) under the planned analysis section described later in section 5(b). The applicant is employed by King’s College London (KCL) and registered as a PhD student within the Department of Psychological Medicine at KCL. They also do clinical work as a psychiatrist for SLaM and has a clinical addendum to their contract with KCL which establishes an honorary contract for this work. It includes provisions to allow sharing of information regarding their performance, conduct and health between SLaM and KCL, a responsibility on the applicant to comply with all relevant NHS policies and procedures and provision for their employment with KCL to be reviewed or terminated if they are in breach of this. This honorary contract has been extended to March, directly after which, the applicant will hold a substantive clinical contract with SLaM. ​Any research assistant or postgraduate student using the data under the applicant's supervision will be employed by or registered as a student with KCL and will also first have to obtain an honorary contract with SLaM with the same provisions as those described for the applicant. They will also have to have completed information governance training prior to any access to the data.

Only staff that have received information governance training will be permitted access to the data.

What will be done with the data:

Analyses will be performed, and results reported at lower super output area (LSOA) level. This is a standard census geography available within the HES data and defined by NHS Digital as a non-identifiable field. Individual level data will be used to allow the standardisation of the LSOA rates calculated for age and sex, both of which are strongly associated with rates of self-harm.

Data for analyses will be extracted from the SQL database aggregated into age, sex and year strata within individual LSOAs. The data extracted will be filtered by age and individual diagnostic codes to define cohorts of interest. Similar control/comparison cohorts may also be established.

Extracted data will be imported into R, a programme for statistical analysis, from which OpenBUGS, a programme required for the planned Bayesian analyses, will be called. Analyses will be conducted at LSOA level and will involve the calculation of spatially smoothed rates of admission with and without standardisation for age and sex, and adjustment for area level exposures of interest as described in the purpose section above.

No record level data will be linked to this dataset, but it will be combined with publicly available data at Lower Layer Super Output Areas (LSOA) level including demographic data to provide denominators for age and sex standardisation, and data relating to the exposures of interest, for example area-level deprivation, area level exposures calculated from census data and area level environmental measures such as air pollution measurements.

All outputs are aggregated with small number suppression in line with the HES Analysis Guide. The output of analyses will be descriptive tables for the cohort as a whole, estimates of standardised admission ratios for areas, estimates of the probability that area rates differ from the overall rate for the study area and estimates of the effect sizes for the association between the exposures of interest and self-harm (rate ratios). Thus all will be based on aggregate data. Descriptive statistics will be reported in line with the HES analysis guide, suppressing results if necessary due to cells having low counts. Area-level outputs will be imported into ArcMap, a geographical information system, in order to visualise the results. This will be done at area level with no individuals being identifiable from any of the reported output.

Planned analyses under the original application:

1) Indirectly 5-year-age and sex standardised rates of all-cause admission and admission for self-harm (ICD-10 codes X60-X84) will be calculated for each LSOA using Bayesian spatio-temporal disease mapping models. The distribution of rates across the Greater London area and over time will be described and compared using overall model parameters, tests for space-time interactions and by mapping rates for different time periods.

2) Sensitivity analyses will be performed for the effect of using broader definitions of likely self-harm. Standardised rates will be calculated for the standard definition of self-harm (ICD-10 Cause codes X60-X84) and broadened definitions including ‘Event of undetermined intent’ (ICD-10 cause codes Y10-Y34), and accident codes that relate to similar mechanisms of injury to those seen in self-harm (e.g. X40-X49, Accidental poisoning by and exposure to noxious substances). Standardised rates , associations with individual and area-level measures and spatial distribution of rates will be calculated and compared using the methods described in 1).

3) Areas with above and below average rates of self-harm admission, before and after adjustment for deprivation will be identified.

4) Predicted rates for the LSOAs will be modelled based on findings from previous work, using publicly available area and population level exposure data. Predicted rates will be mapped and areas predicted to have above or below average rates of self-harm admission identified.

5) Performance of the model in predicting self-harm admission rates will be checked against the clinical data. Reasons for any discrepancies between the model and the clinical data will be explored.

Of these initial planned analyses, points 3-5 have already been conducted, while points 1 and 2 have not happened due to delays caused by COVID. This analyses is expected to happen during the remainder of the agreement.

The data will also be used to support work on future outreach events, workshops and seminars conducted, based on the project's findings, after the PhD's paper and thesis have been reviewed and published.

What will not be done with the data:

* The data will not be linked with any record level data.

* There will be no requirement nor attempt to re-identify individuals from the data.

* The data will not be made available to any third parties except in the form of aggregated outputs with small numbers suppressed in line with the HES Analysis Guide.

Data required:

Only HES APC data has been requested because other datasets do not contain sufficiently accurate and complete coding of self-harm to be used for the study aims.

Data has been requested for a 10 year period from 2008-2017. This will allow trends in rates of self-harm by area over time to be examined.

Data was only requested for individuals who were resident within the Greater London region at the time of hospital admission, as this is the study area of interest.

Data was only requested for individuals aged 11 or older at the date of admission in order to match the age range in the data that will be used to build the model being tested. Data for younger individuals is not being used as incidents of self-harm are rare prior to adolescence. For further details of the attempts made towards data minimisation, see the previous section, 5(a).

Filters have not been applied to specific conditions in order to allow sensitivity analyses of different ways self-harm may have been coded to be carried out, for example how the inclusion of acts coded as of undetermined intent, or accidents affects the results seen and to allow comparison to be made to rates of all-cause admission.

No identifiable or sensitive data items were requested.

All organisations party to this agreement must comply with the Data Sharing Framework Contract requirements, including those regarding the use (and purposes of that use) by “Personnel” (as defined within the Data Sharing Framework Contract i.e.: employees, agents and contractors of the Data Recipient who may have access to that data).

Post-migration, the datasets used for the research project will continue to be stored in shared network drives within the SLaM network with data being extracted from Microsoft Azure Cloud to create them.

Aside from Microsoft Azure Cloud providing a hosting infrastructure, SLaM will continue to be the only organisation able to process the data and the use of the MS Azure infrastructure is covered within SLaM IT and IG policies and the SLaM Data Security and Protection Toolkit.

The proposed project forms part of a larger PhD project examining the reasons for variations in the rates of self-harm between small areas in London, and area level factors that may be associated with them. It arose from a literature review that found that the rates and predictors of self-harm in London appear to differ from those in England as a whole.

Work on the PhD project has been delayed due to COVID-19 as the applicant had to be deployed back into the NHS workforce as part of the COVID response during 2020. As such, the predicted submission date of May 2020 for the final paper has not been possible. A paper based on the findings of the PhD using the requested data has been submitted to a journal and is currently under review. The focus of the applicant's PhD changed during its development such that the studies using the data describe the patterning of rates of self-harm, comparing different neighbourhood and communities as described in the original Agreement, but do not produce a predictive model.

It was initially estimated that the applicant's PhD thesis would be submitted September 2020. This deadline has been missed due to delays caused by the pandemic. The thesis was instead submitted in January 2021 and is set to be published via the KCL repository following the applicant's viva - scheduled March 2021 - and any required corrections from this. The data is still required to support any of these required corrections after review, as well as for the planned analyses described previously in section 5b relating to 5-year-age and sex standardised rates of all-cause admission and admission for self-harm and sensitivity analyses for the effect of using broader definitions of likely self-harm.

The applicant has presented work based on the data to psychiatrists working in South London February 2020. Subsequent planned presentations were largely cancelled due to the COVID pandemic and the applicant's redeployment but it is projected that the applicant will be submitting abstracts to conferences in 2021 (for example the European Psychiatric Association Section of Epidemiology meeting which was postponed in September 2020) as and when they are announced.

The original project plans were "to share findings with local mental health services which work with individuals presenting with self-harm, and with public health departments, who write local suicide and self-harm prevention plans and commission mental health services, through seminars and presentations at team meetings and more widely across London through the Public Health England London Mental Health Network between March and September 2020." This work was also delayed due to COVID. The plan was to also disseminate lay summaries of the findings to Health Inequalities Research Network (HERON) engagement events and website and through the Maudsley BRC’s engagement events and website. This has not been able to happen due to COVID.

In addition, lay summaries of the research will be disseminated to a wider lay audience. This dissemination will take place in collaboration with community organisations with an interest in mental health, who the PhD student has been working with during earlier work as part of the PhD project, and through the Health Inequalities Research Network (HERON) engagement events and website and through the Maudsley BRC’s engagement events and website. The results relating to the data requested in this proposal were expected to be disseminated between March and September 2020. Due to COVID, the publish date of the paper has been delayed. Subsequently, this public engagement work has also been delayed and will follow the publication of the paper currently under review.

The plan was to further disseminate the PhD findings more broadly across London public health teams and mental health services. This was planned between March and September 2020, with support from local contacts and by using Public Health England London Mental Health Network, however this work has been delayed and will now be conducted following the publication of the paper currently under review.

All outputs will contain only data that is aggregated with small numbers suppressed in line with the HES Analysis Guide.

it is projected that the applicant will be submitting abstracts to conferences in 2021 (for example the European Psychiatric Association Section of Epidemiology meeting which was postponed in September 2020) as and when they are announced.

The applicant anticipates that much of the further dissemination described in section 5c will take place over the next year, although the course of this is likely to be shaped by the impact of the current pandemic on capacity within both service providers and public health departments to engage with non-COVID issues which is not entirely predictable at this stage.

Wider dissemination of the work beyond local institutions such as KCL has started through the applicant's work with the Maudsley Cultural Psychiatry Group and involvement in the Data and Research subgroup of the Patient and Carer Race Equality Framework pilot at South London and Maudsley NHS Foundation Trust. In both of these roles, the applicant is feeding the findings relating to the patterning of self-harm rates within the communities the trust serves, and their relationship to population race and ethnicity in particular, into efforts to improve access to and outcomes in services and the training of practitioners. This collaborative work with community and service user groups as well as academic partners within London will provide future opportunities for dissemination of the findings through writing blogs and presenting at workshops and seminars.

All expected outputs remain the same as stated in the original application, albeit delayed due to the pandemic, with the exception of the abandonment of the predictive model. No additional outputs have been added from the original application.

Self-harm results in over 100,000 admissions to hospital in England each year. For individuals, self-harm requiring medical attention represents mental distress and usually disorder, damage to physical health and is the strongest single risk factor for future suicide. For health care services it represents a substantial proportion of presentations for emergency care, especially among young people.

At a population level, preventing and responding to self-harm and enhancing community based support have been identified as key priorities within Public Health England’s suicide prevention strategy. Local authorities’ public health departments are tasked with drawing up local suicide prevention plans based on these priorities. Targeting of preventative interventions and support services for self-harm at a local level could be enhanced by a better understanding of where areas of increased risk are likely to be. However, statistics on service use for self-harm are not routinely available at a small area level, and the statistical techniques required to reliably interpret spatial patterns across areas with relatively low counts are complex. Furthermore, area-level factors that are known to be predictive of self-harm rates nationally, especially deprivation, have been shown to be less strongly predictive of self-harm rates within London, making them less useful as a basis for targeting services.

The project has established contacts with the public health departments local to KCL and with local community groups working in mental health who are keen to make use of data such as this in their local suicide prevention plans. Further dissemination more broadly across London public health teams and mental health services were planned between March and September 2020, with support from local contacts and by using Public Health England London Mental Health Network, to ensure these benefits are realised. As explained above, these have been delayed due to the pandemic.

This project originally aimed to validate a model to predict areas likely to have higher or lower than average rates of self-harm within Greater London. After conducting research around this, such a model was unable to be produced, however, the other expected benefits to the Health and Social Care system already described remain key to the purpose of this project.

There have been no benefits to the Health and Social Care System yielded to date due to delays to the submission date of the PhD paper and thesis caused by the pandemic. Any positive changes to the way the Health and Social Care System operates yielded as a result of the findings of this PhD project will become more apparent once the paper and thesis have been reviewed and subsequently published.