Official Statistics

Impact of drug and alcohol treatment on reoffending: methodology

Published 9 July 2026

Applies to England

Introduction

This technical report describes the methodology we used to analyse the effect of community-based drug and alcohol treatment on the reoffending behaviour of offenders in treatment.

The work combined data from the Police National Computer (PNC), managed by the Ministry of Justice (MOJ), with the National Drug Treatment Monitoring System (NDTMS), managed by the Department of Health and Social Care (DHSC).

This technical report sits alongside the main statistical report and provides transparency on:

• data governance
• data sources
• data linking methods
• analytical methods
• definitions

This technical report is intended for analysts and researchers working in criminal justice and community substance misuse treatment.

Information governance

Data protection impact assessment

DHSC developed the data protection impact assessment (DPIA) (see glossary), which was approved by the MOJ PNC team and the DHSC Data Protection Officer.

The DPIA identified how the data would be processed, stored, used and deleted. It also assessed that the amount of personal information used would be necessary and evaluated potential risks, such as data breaches. To mitigate these risks, the DPIA stated that the access to the PNC and NDTMS linked information asset will be strictly controlled and any outputs will be quality assured to mitigate statistical disclosure risks.

Data sharing agreement

A cross-government data sharing agreement (DSA) (see glossary) was established and signed by senior management representatives across DHSC and MOJ. The agreement sets out the purpose of the data processing, roles and responsibilities, security requirements and retention and destruction arrangements.

Caldicot Guardian approval

Since the data was hosted by the UK Health Security Agency (UKHSA), approval for processing the data was obtained from the UKHSA Caldicott Guardian. This ensured we adhered to all 8 Caldicott Principles, consistent with standards for handling confidential health information. You can find out more about these in The Caldicott Principles.

Legal basis

UK General Data Protection Regulation

Processing was undertaken in line with Articles 6 and 9 of the UK General Data Protection Regulation, relying on the following legal bases:

• public interest
• public health purposes
• scientific and statistical research

Data Protection Act 2018

Processing also met the conditions set out in schedule 1, part 1 of the Data Protection Act 2018, specifically:

• health or social care
• public health
• research

Common law duty of confidentiality

People accessing specialist drug and alcohol treatment in England are asked to provide consent for their information to be shared with NDTMS. Approximately 98% of individuals recorded in NDTMS provide consent for their treatment data to be used for service planning and performance monitoring.

This satisfies DHSC’s common law duty of confidentiality and allows DHSC to carry out data linkage internally. Identifiable NDTMS data is not shared with other government departments. All linkage and analysis was conducted exclusively by DHSC staff.

Refer to NDTMS: consent and privacy notice for further details.

Confidentiality and security controls

Data was stored on accredited, secure government networks, with access restricted to authorised analysts who had been appropriately vetted. All staff completed mandatory information governance training. Outputs were aggregated to prevent individuals from being identified and direct identifiers were removed or pseudonymised (see glossary) after linkage.

For further information, see How DHSC processes special category data.

Data linkage methodology

Linkage approach

NDTMS and PNC do not share a common unique identifier to enable a simple linkage of the 2 systems. While the PNC contains the full name, NDTMS only contains the initials of the person in treatment. However, both systems contain the date or birth, sex and local authority of residence. This necessitates linking using either probabilistic or deterministic (see glossary) linkage approaches.

Due to the DSA preventing some methods of probabilistic linkage, we adopted a deterministic linkage strategy like that in The effect of drug and alcohol treatment on re-offending, released by Public Health England (PHE) and MOJ in 2017.

This approach involved 3 rounds of exact matching to maximise linkage between the NDTMS and PNC data sets.

In round one, records were matched on initials, date of birth, sex and local authority (with no alias data (see glossary)), and only unique one-to-one matches were accepted as true links. Ambiguous cases (one-to-many or many-to-many) were discarded, and any unmatched records moved to the next round.

Round 2 incorporated PNC alias data and repeated the matching under the same rules, carrying forward only those still unmatched.

In round 3, the local authority variable was dropped to relax the matching criteria, again accepting only clear one-to-one matches.

The final matched cohort consisted of all the one-to-one matches identified across these 3 rounds.

Potential issues with linkage

The deterministic methodology relies on the identifier fields being identical, so any data errors will cause a match to fail. Also, some identifiers change over time, such as a person changing names or moving to a different local authority area. This could result in them failing to match.

There may also be false matches between different individuals where people share the same initial, date of birth, sex and local authority of residence.

Match rates

Information was received on 9,388,754 individuals from the PNC, which was then matched with data on 1,140,999 individuals in community treatment from NDTMS.

The first stage involved matching individuals using available common identifiers (initials, date of birth, sex and local authority), resulting in 439,957 one-to-one links between the data sets. In addition, 97,327 individuals were identified as having uncertain matches (referred to as ‘multiple matches’). These occur when:

• one individual in NDTMS links to multiple individuals in the PNC
• multiple individuals in the PNC link to one individual in NDTMS
• multiple individuals in both systems link to each other

These multiple matches were excluded from further analysis.

The second matching rule was applied to individuals not previously linked or removed. This involved using aliases that offenders may have used at the point of arrest, as recorded in the PNC. This stage produced an additional 26,501 one-to-one links, along with 6,503 further multiple matches, which were again excluded from further analysis.

The third and final matching rule was applied to the remaining unlinked individuals, using only initials, date of birth and sex. This resulted in a further 94,950 one-to-one links and 197,140 multiple matches, which were also removed from further analysis.

At the conclusion of the process, a total of 561,408 individuals who accessed treatment between 2013 and 2023 had at least one corresponding record on the PNC, while 278,621 individuals did not have a link. Overall, this indicates that 67% of the available cohort - after excluding all individuals with multiple matches - had at least one record on the PNC between 2000 and 2024.

Data preparation and exclusions

Preparing the NDTMS data

An extract of NDTMS data was used in which a person’s treatment information was included if their triage date fell between 1 April 2013 and 31 March 2023. Records outside this period were excluded.

Individuals were excluded if they:

• were aged 8 and under or aged 99 and over
• had a recorded sex other than male or female
• had a local authority of residence outside England
• had invalid or inconsistent dates, for example where a treatment intervention start date preceded the triage date

During the analysis period, local authority geographies changed due to mergers and splits over time. To ensure consistency, some local authorities were mapped to a common geography. For example, where Northamptonshire later split into 2 authorities, records were mapped back to the original single upper-tier local authority.

As a result of this geographic harmonisation, annual counts may not exactly align with published official statistics. This is because some individuals may be assigned to a different local authority and treatment periods may vary slightly.

The latest official statistics on substance misuse treatment are published separately for adults and for children (aged 17 and under):

• Substance misuse treatment for adults
• Substance misuse treatment for children

Preparing the PNC data

The same local authority mapping was applied to the PNC data to ensure consistency with the NDTMS data set. Offences with a missing offence start date were excluded, as these could not be aligned with periods of substance misuse treatment.

To simplify analysis, PNC offence codes were mapped to offence groups and records with missing offence codes were excluded. The analysis data set was further restricted to offences occurring between 1 April 2000 and 31 March 2024. The only offences included were those with adjudication outcomes:

• warning
• caution (see glossary)
• guilty

Analysis data sets

Given the size of the linked data, 2 analysis data sets were created to meet different analytical needs.

Data set for descriptive statistics

This data set was designed to replicate the approach used in the 2017 PHE and MOJ experimental publication. It is offence-based and focuses on the earliest treatment journey within each financial year. For each person, offences occurring up to 10 years before and up to 10 years after the start of treatment were linked to the relevant NDTMS treatment journey, including time spent in treatment before and after that start date.

The data set was constructed by identifying the earliest treatment start in each financial year for individuals in the NDTMS community data set with treatment journeys starting between April 2013 and March 2023. Offending data was then linked to this NDTMS cohort using the deterministic linkage described above. Offences were grouped by Home Office offence group and assigned to time bands of between 1 and 10 years before and after treatment start. This allows offences to be analysed by NDTMS characteristics such as drug treatment group, treatment outcome status or region.

This data set only collects information on new treatment journeys, so it does not accurately reflect the entire treatment population. A person retained in treatment year on year was counted once, whereas a person who came in and out of treatment was counted multiple times.

Data set for inferential statistics

This data set includes all treatment journeys in the NDTMS community data set, rather than only the earliest journey in a financial year. It is treatment journey based, with offences (by offence group) aggregated to each treatment journey for the one year before and one year after treatment start. As a result, the same offence may be associated with more than one treatment journey.

Offending data (after the exclusions described earlier) were linked to all NDTMS treatment journeys starting between April 2013 and March 2023 using the lookup table from the one-to-one matching process.

Definitions of variables

To enable this analysis some specific methodologies were used, which are listed below.

Prolific offenders

The methodology for prolific offenders was based on the number of previous offences and an individual’s age. Individuals who met the following criteria were flagged as prolific offenders for this treatment journey:

• aged 17 years and under with 4 or more offences
• aged 18 to 20 years with 8 or more offences
• aged 21 years and over with 16 or more offences

According to MOJ, if a person is classified as a prolific offender before the age of 21, they retain this classification even if they go on to commit no further offences.

However, because of a lack of data available to trace a person’s offences during their whole life, we have potentially underestimated the number of prolific offenders.

Treatment outcome

The methodology for treatment outcomes is the outcome recorded when a person’s structured drug or alcohol treatment episode ends. It describes the immediate end-of-treatment outcome and by itself does not indicate longer-term recovery or whether the person re-engages with services.

In the report, discharge reasons are grouped into 4 categories, as described in table 1. Each row also shows the relative proportion of each discharge reason.

Table 1: mapping of original discharge reasons into grouped categories

Original discharge reason	Proportion of original discharge reason	New grouped discharge category
Died	2.3%	Died, dropped out or inconsistent
Dropped out	35.8%	Died, dropped out or inconsistent
Inconsistent	0.2%	Died, dropped out or inconsistent
Not known	0.0%	Died, dropped out or inconsistent
Other	0.0%	Died, dropped out or inconsistent
Prison	2.9%	Prison
Referred on	0.0%	Died, dropped out or inconsistent
Still in treatment	7.3%	Still in treatment
Successful completion	10.3%	Successful completion
Successful completion - no drug or alcohol use	16.3%	Successful completion
Transferred in custody	18.1%	Prison
Transferred not in custody	4.5%	Died, dropped out or inconsistent
Treatment declined	1.8%	Died, dropped out or inconsistent
Treatment withdrawn	0.6%	Died, dropped out or inconsistent
Inconsistent	0.1%	Died, dropped out or inconsistent

IMD score

The methodology for population-weighted index of multiple deprivation (IMD) (see glossary) scores were calculated at postcode sector level for 2010, 2015 and 2019. This was undertaken by importing Office for National Statistics IMD scores and population estimates for each year, alongside geographical lookup tables linking postcode sectors to Lower layer Super Output Areas (LSOAs).

IMD scores were then aggregated from LSOA to postcode sector level using population weights, ensuring that areas with larger populations had proportionate influence. For each postcode sector, the IMD score was calculated as the population-weighted average of the IMD scores of all constituent LSOAs for the years 2010, 2015 and 2019.

Locations were then ranked by IMD score, and quintiles were derived before being incorporated into the analysis data sets by year of treatment.

Analytical methods

Multilevel logistic regression models with random intercepts were used to estimate the associations between characteristics of people in treatment and offending. This approach appropriately accounted for individuals having multiple treatment journeys over the observation period. This was necessary because individuals who offended in the year prior to one treatment journey are more likely to offend in the year prior to subsequent treatment journeys. These journeys are nested within individuals, and this clustering can only be appropriately modelled using multilevel methods.

The analysis assessed the extent to which individual-level characteristics were associated with:

• being an offender in the year prior to starting treatment
• reoffending in the year following the start of treatment

The regression models produced adjusted odds ratios (see glossary), which provided an intuitive measure of how each predictor variable (for example, age or substance type) was associated with the likelihood of offending or reoffending.

Glossary

Alias: an alternative name or identifier used by an individual, for example an alias or nickname. Information on aliases can help data-linking projects match records across data sets even when names differ.

Caution (police caution): a formal warning given by police to an offender who admits a minor offence, used as an alternative to prosecution. Cautions are not full convictions and do not result in a court sentence, but they are recorded by police.

Data protection impact assessment (DPIA): a formal process to systematically identify and minimise privacy and data protection risks in a project involving personal data.

Data sharing agreement (DSA): a legal system outlining the terms, conditions and safeguards for sharing data between organisations. A DSA specifies what data is shared, for what purpose, who can access it and how it must be protected.

Deterministic matching: a data linkage method where records from different data sets are linked only if they have exactly matching identifiers (such as the same full name, date of birth and other personal details).

Index of multiple deprivation (IMD): the measure of relative deprivation for LSOAs in England. IMD ranks LSOAs from most deprived to least deprived, based on factors like income, employment, health and crime.

Multiple match: in data linkage, a non-unique match where one record in a data set could correspond to more than one possible record in the other data set. This can also occur in the opposite direction.

Odds ratio (OR): a statistical measure from logistic regression analysis that quantifies the strength of association between a characteristic and an outcome. It compares the odds of an event occurring in one group with another. An OR of 1.0 indicates no difference between groups. An OR greater than 1 means higher odds of the outcome in the first group, whereas an OR less than 1 means lower odds.

Probabilistic matching: a data linkage approach that uses statistical algorithms to link records that are not exact matches but likely belong to the same person.

Pseudonymised: a technique that replaces, removes or transforms information that identifies a person, and keeps that information separate. For more information, see the Information Commissioner’s Office page on pseudonymisation.