ThresholdBrain injury and justice Stan Gilmour and Huw Williams

What the evidence says, in twelve statements

Each statement below is a reading of the corpus. It carries a status, the records it rests on, and a link to the wider set in the evidence base. Where the field disagrees, the statement says so and names the dispute. Nothing here is a policy recommendation. Those are authored separately and will be kept visibly apart from the evidence.

12statementsDrawn from the report's Introduction and Chapter 1
61records citedEach checked against the corpus when the site is built
48of them tier oneLandmark or definitive in the field's own judgement
17carry a contested flagThe dispute is named in the record and summarised here
Draft synthesis, for review by Stan Gilmour and Huw Williams Statements follow the Introduction and Chapter 1 of the report. Each rests on the records listed under it, which the site checks against the corpus at build time. A statement is a reading of the evidence, not a policy recommendation. As of 17 September 2026.
01 Settled in direction

Brain injury is several times more common among people in contact with the justice system than in the general population, and the elevation holds on every method of counting.

The largest synthesis pooled 64 studies and 52,540 participants and put lifetime prevalence among adults in contact with the criminal legal system at 45.8 per cent. The most conservative method, linking hospital records to the whole Scottish prison population, counts only injuries severe enough to reach a ward and still finds 24.7 per cent against 18.2 per cent in a matched general population, twice the odds. Structured interview in specific populations returns higher figures again: 86 per cent of men in two Scottish prisons and 78 per cent of women across four. Among young people in custody the summary odds ratio against controls is above three.

The records this rests on

Browse the 38 works in strand A, adult custody prevalence, 22 of them tier one.

02 Unsettled in magnitude

The precise prevalence figure depends on how injury is defined and detected, and pooling across methods has reached the limit of what it can tell us.

The 45.8 per cent pooled estimate comes with a 95 per cent prediction interval running from 5.5 to 92.5 per cent, which is an admission that the underlying studies do not agree. Definition and method of detection are the moderators that matter. The 60 per cent figure from the first meta-analysis is still quoted in policy documents and is treated by later work as an overestimate driven by definitional heterogeneity. Self-report is more reliable than commonly assumed: in one validation, hospital records supported 70 per cent of the injuries prisoners reported. The responsible way to state prevalence is within ascertainment strata, not across them, and the most recent development in the field is a proposal to change the classification of injury itself.

The records this rests on

Browse the 53 works in the case-definition theme, 32 of them tier one.

03 Settled

Most injured people in custody do not know they have a brain injury, and the system does not routinely ask.

Of the women in one United Kingdom prison who reported an injury meeting the screening threshold, 28.5 per cent believed they had sustained a brain injury. On the other side of the desk, fewer than 2 per cent of 307 detainees referred to a forensic medical examiner across five London boroughs were thought to need an appropriate adult, and the Bradley review found the same pattern in the same period. Of 57 clinical practice guidelines for traumatic brain injury, 14 mention the criminal justice system and one makes a concrete recommendation. A safeguard that depends on a custody officer noticing, or on a detained person disclosing, will not operate for a population in which neither reliably happens.

The records this rests on

Browse the 80 works in the under-identification theme, 38 of them tier one.

04 Contested

Whether brain injury causes offending is not settled, and the choice of comparison group decides the answer.

Against the general population the association is large: a Swedish register cohort of 22,914 people with hospital-recorded injury gave an adjusted odds ratio of 3.3 for later violent crime, and an Ontario cohort of 1.4 million found people with a prior injury about 2.5 times more likely to be imprisoned. Against other injured people it is small or absent: in a United Kingdom birth cohort an orthopaedic injury raised the odds of offending as much as a brain injury did. Against siblings it is smaller again, and the result depends on the exposure. Within-family comparison attenuated but did not remove the association for hospital-recorded injury of any severity, and removed it for mild injury diagnosed before the age of ten. Both sides now say so in print, and neither disputes the other's estimate.

The study the corpus lacks is a confounder-controlled comparison of moderate to severe and repeated injury, with ascertainment that does not depend on hospital attendance, in a population whose justice outcomes are observed. Until one exists the field will keep comparing estimates derived from exposures that are not the same exposure.

The records this rests on

Browse the 14 works in contested findings in strand E, trajectories and recidivism, 12 of them tier one.

05 Settled

Once inside the system, injured people accumulate worse outcomes at each stage, and this does not depend on the injury having caused the offence.

The effects of injury present as something else, most often intoxication, and staff describe punishing injury-related behaviour as ordinary misconduct because they do not recognise where it comes from. In a statewide cohort followed over 11.5 years, men with a medically attended injury had a rate ratio of 1.86 for violent infractions, and in 12,038 people serving a first Canadian federal sentence the relative risk of a serious disciplinary charge remained elevated after adjustment. Discipline governs regime status, programme access and release. After release, 70 per cent of 2,389 people referred to case management with a recorded injury failed to keep contact with services. Over-representation at the deep end of the system therefore has two possible sources: a higher rate of entry, which is disputed, and a higher rate of accumulation once inside, which is comparatively well evidenced.

The records this rests on

Browse the 39 works in strand F, custody and supervision, 19 of them tier one.

06 Settled

For women in custody, a brain injury history is usually the physical record of violence already suffered, most often from a partner and usually before the first offence.

Among 173 women screened at HMP and YOI Drake Hall, 64 per cent reported a history indicative of brain injury, 96 per cent reported domestic abuse, and of those with an injury history 62 per cent attributed it to that abuse. Across four Scottish prisons, 84 per cent of women with a significant head injury had been injured repeatedly, in most cases through domestic abuse over years. The pattern was first described outside custody in 2003, when almost three quarters of 99 battered women had sustained a partner-related brain injury. Strangulation, which injures by a different mechanism and leaves little to photograph, became a standalone offence in England and Wales only in 2022. Reading injury only as a possible contributor to offending misses half of what these studies record.

The records this rests on

Browse the 28 works in strand C, women and girls, 14 of them tier one.

07 Settled

Brain injury in custody travels with suicide risk, substance use and mental ill health, which argues for screening that covers more than one condition.

In 665 men entering a category B prison in Wales, a history of brain injury was associated with 3.3 times the odds of a previous suicide attempt. In two Scottish prisons, disability after head injury was associated with problematic substance use, clinical anxiety and clinical depression. In a Victorian cohort followed after injury, substance use before and after the injury raised the risk of imprisonment for violence far more than any other factor measured. The Finnish birth cohort made the same link between childhood injury, adult psychiatric disorder and heavy alcohol use a quarter of a century ago. The finding cuts both ways. It is cited on the sceptical side of the causal debate, and it is the reason a screening programme confined to brain injury alone would miss most of what matters.

The records this rests on

Browse the 81 works in the mental health co-occurrence theme, 29 of them tier one.

08 Obligation settled, delivery not

The legal duty to identify vulnerability already exists. What is missing is a routine, evaluated way of finding brain injury at any point in the pathway.

PACE Code C defines vulnerability functionally, in terms of difficulty understanding the implications of detention and a tendency to confusion, suggestibility or undue compliance, and requires an appropriate adult once it is identified. Effective participation has been a requirement under Article 6 since 1999 and the Law Commission built its 2016 recommendations on it. Neither test needs a prior finding about brain injury and crime. The obstacle is identification. One custody screening tool, tested against clinical interview in one force, detected head injury with 72 per cent sensitivity against 17 per cent for the tool previously evaluated. No United Kingdom jurisdiction routinely screens at any point in the pathway, and the law on unfitness to plead does not reach the magistrates' courts, where most cases are heard.

The records this rests on

Browse the 49 works in strand G, courts and rights, 29 of them tier one.

09 Thin evidence

Almost nothing has been tested as an intervention in a justice setting: two linkworker evaluations without a control group, and one randomised trial of 91 people.

The service model with most exposure in England and Wales is the brain injury linkworker. It has been evaluated twice, at HMP Drake Hall and across two Welsh prisons, and both evaluations report falls in anxiety and depression scores that their own authors decline to attribute to the intervention, because there was no comparison group and, in Wales, only a third of participants completed the follow-up measures. Against that, the single randomised trial in the corpus reports that three years after release 17 per cent of the intervention arm had returned to prison against 46 per cent receiving treatment as usual, in one United States programme. One trial of that size settles nothing. A difference of that size, in an outcome as expensive as reimprisonment, is a reason to run more. Sixty-four studies have been pooled to count the problem. One trial has tested whether anything can be done about it.

The records this rests on

Browse the 35 works in strand H, interventions, 11 of them tier one.

10 Settled, in the negative

No neurological finding identifies who will offend, and none has emerged in twenty-five years of looking.

The association between frontal lobe dysfunction and aggressive or antisocial behaviour is supported, and the Vietnam Head Injury Study gave it its most rigorous empirical footing. The systematic review of 2001 found that association real and the predictive claim unsupported, and nothing since has displaced that. The sport literature, where imaging and neuropathology have advanced furthest, has produced a lesson about what selected samples can support rather than a mechanism that transfers to justice populations. A reader looking for a scan or a test that picks out future offenders will not find one in this corpus, and the absence is not a gap in the search.

The records this rests on

Browse the 41 works in claims from 1996 to 2005 that still hold, 21 of them tier one.

11 Emerging

The justice workforce carries its own burden of brain injury.

A 2026 meta-analysis put pooled lifetime prevalence of traumatic brain injury among law enforcement and corrections officers at 58.8 per cent, with 17.2 per cent sustained in the course of the job, on five studies with very high heterogeneity. Firefighters report 62 to 77 per cent. A concussion management framework for law enforcement exists in the clinical literature and has no counterpart in United Kingdom policing guidance in this corpus. The people asked to recognise brain injury in the people they detain are a population at elevated risk of it themselves.

The records this rests on

Browse the 39 works in strand J, sport, military and occupational injury, 18 of them tier one.

12 Absent from the literature

Three absences shape what can be said: no United Kingdom linkage study of a justice outcome after injury, no economic evaluation of screening, and no United Kingdom mortality study in an injured justice cohort.

Both halves of the mortality study exist in Scotland. A cohort of 76,627 first-time prisoners linked to death records found an age-standardised mortality rate ratio of 3.3 in men and 7.6 in women, and hospitalised head injury in homeless people carried a standardised mortality ratio more than twice that of the uninjured. Nobody has crossed the two. That linkage can answer this class of question is not in doubt: a New South Wales study linked health and offending records for 7,743 people with a psychotic disorder and measured the effect of court diversion on reoffending, and Ontario has done the same for brain injury and imprisonment. No screening programme in any justice setting has been costed against what it finds, which is what stops the case being made in the terms commissioners use.

The records this rests on

Browse the 73 works in the evidence-gap theme, 27 of them tier one. See also the six research gaps, each as a fundable question.

How these statements are made

A statement is written from records, not from the reviewer memos, and every figure in it appears in a record's key findings. The status is the reviewer's judgement against the later evidence in the corpus: settled means the direction has survived later and better designs, contested means the literature disputes it now and the dispute is named, and absent means the corpus establishes that the study does not exist. The judgement is recorded so it can be argued with, and the methodology sets out what this review does not yet meet.