The prison industrial complex
May 27, 2025
What is terrorism?
May 27, 2025
Corporation Prisons
More than 11 million individuals are imprisoned worldwide. People who pass through prisons often have poor health profiles, including high frequencies of self-harm. Self-harm is a leading cause of morbidity in prisoners; the annual prevalence of self-harm in prison has been estimated to be 5–6% in men and 20–24% in women, which greatly exceeds the less than 1% of adults in the general population who self-harm each year.
Prisoners who self-harm are at a six to eight times increased risk of suicide while incarcerated and remain so after release into the community. Half of people who die by suicide in prison have a recorded history of self-harm, with many deaths occurring within a month of self-harm.
In addition, the impact of self-harm extends to other prisoners and to prison staff, and can lead to substantial costs for the prison system, especially if associated with suicide. Transfer of prisoners to local health-care services for the more severe incidents can further increase costs.
Understanding the risk factors for self-harm can help to improve prevention efforts in this population at high risk, particularly if there is evidence of modifiable risk factors.
Previous research has examined a range of individual and environmental correlates of self-harm in prisoners, although findings are inconsistent across primary studies.
Reviews have been limited by being narrative syntheses of the literature that do not use quantitative methods to evaluate the strength, quality, and consistency of the available evidence. Therefore, we have done a systematic review and meta-analysis of risk factors associated with self-harm inside prison.
Our findings could identify appropriate targets for interventions and future treatment trials, and assist decision makers in allocating scarce prison resources
Evidence before this study
We searched four databases (PubMed, Embase, Web of Science, and PsycINFO) for systematic reviews of self-harm risk factors in adult prisoners published from database inception to Oct 31, 2019. The same keywords were used for each database search for self-harm, “(self-harm OR suicide OR attempt OR NSSI OR self-injury OR self-mutilate OR self-destruct OR poison OR overdose)”, and prison “(inmate OR penal OR correction OR sentence OR remand OR detainee OR felon OR prison OR incarcerated)”.
No language restrictions were set. We identified three systematic reviews with narrative summaries on risk factors for self-harm, non-suicidal self-injury, and near-lethal suicide attempt.
These reviews reported that risk factors span many different individual and environmental domains, although there were many inconsistencies in the magnitude and direction of the effects.
We did not identify any reviews that meta-analysed findings or evaluated the strength and consistency of risk factors for self-harm inside prison.
We found one meta-analysis published in 2020 that examined the association between childhood maltreatment and suicide attempt in a population who has been in contact with the criminal justice system, but this meta-analysis included non-prisoners, juvenile offenders, and outcomes in the community.
This meta-analysis synthesised data from almost 50 years of research examining risk factors for self-harm in over half a million prisoners. Although we identified many risk factors for self-harm across sociodemographic, criminological, custodial, clinical, and historical domains.
Implications of all the available evidence
Our results show that risk factors associated with self-harm in prisoners include a range of potentially modifiable clinical, psychosocial, and environmental factors. These data emphasise the need for a whole-prison approach and multiagency collaboration in the prevention of self-harm
Data were independently extracted by two researchers (LF and Isabel Yoon). A standardised form was used to extract data and included information on study characteristics (ie, publication year, country, design, and number of prisoners included), sample details (ie, age and sex), outcomes (ie, definition and assessment), and risk factors.
Extraction sheets for each study were crosschecked for consistency and any discordance was resolved by discussion between study authors. When the study characteristics were unclear, the corresponding authors of included papers were contacted.
When multiple publications from the same population were available, information on risk factors was extracted from the investigation with the largest sample size. Data were only extracted from overlapping publications when a new risk factor was reported.
As the reporting of effect sizes varied between studies, data were converted to a comparable measure for meta-analysis. Odds ratios (ORs) and their 95% CIs were extracted when reported or calculated from available data in the paper (eg, converted from standardised effect sizes) by use of standard formulas.
Most studies did not provide adjusted effect sizes and, for many, we had to calculate the ORs on the basis of raw prevalence data. In addition, different studies used contrasting approaches to adjustment (from basic demographics to clinical and custodial factors), which would make adjusted estimates difficult to compare. Therefore, to obtain a consistent measure across studies, data were extracted from the most parsimonious model (ie, the least adjusted model).
LF assessed all studies for risk of bias using the Newcastle-Ottawa Scale for cohort and case-control studies, with 9 points indicating high quality and low risk of bias. A modification of the Newcastle-Ottawa Scale was adopted for the assessment of cross-sectional studies, which has been used in suicide research before and is out of 8 points. This scale assesses quality in terms of sample representativeness and size, comparability between respondents and non-respondents, ascertainment of self-harm, and statistical quality. On the basis of these scores, we calculated a summary score (the sum of items divided by the total possible sum) ranging from 0 to 100 and each study was then categorised as low (≤49), moderate (50–74), or high (≥75) quality. Uncertainties were resolved by discussion among study authors.
We grouped risk factors into five categories: sociodemographic, criminological, clinical, custodial, and historical. Three separate outcomes were identified: self-harm, suicide attempt, and non-suicidal self-injury. We have taken a broad definition of self-harm as any act of intentional self-poisoning or self-injury irrespective of the degree of suicidal intent or underlying motive, which includes both suicide attempt (self-injurious behaviour with inferred or actual intent to die) and non-suicidal self-injury (self-injurious behaviour without any intent to die). The difficulty in establishing suicidal intent and the high co-occurrence of both behaviours and their overlapping risk factors explains our approach of combining non-suicidal self-injury and suicide attempt into a single self-harm outcome. This method is consistent with policy and reporting in many prison jurisdictions, including in England and Wales, which has the largest prison population in western Europe
In three instances, both suicide attempt and non-suicidal self-injury were investigated in the same study sample. To avoid double counting of participants, we contacted the authors from these three studies for data on an aggregated outcome measure of any self-harm: suicide attempt, non-suicidal self-injury, or both.
In addition, four studies reported in seven articles specifically focused on near-lethal suicide attempt, defined as acts that could have been fatal had it not been for intervention or chance, involved methods that are associated with a reasonably high chance of death, or both. We included this outcome as there were no material differences in the effects of risk factors for this outcome compared with other self-harm outcomes. Furthermore, other studies did not differentiate according to the severity or lethality of outcomes and might thus also have included near-lethal self-harm.
Where possible, we examined risk factors for men and women separately. We did the meta-analysis in Stata IC (version 13) using the metan command. For all analyses, we generated random effects models that accounted for the anticipated high heterogeneity between studies resulting from differences in samples, measures, and design. Heterogeneity was estimated by use of the I2 statistic, which quantifies the percentage of variance across studies that can be attributed to true variation in effect sizes rather than sampling error as low (0–40%), moderate (30–60%), substantial (50–90%), and considerable (75–100%).
The extent to which methodological variations across studies affected the association between risk factors and self-harm was examined by applying meta-regression models (by use of the metareg command).
Specifically, univariate meta-regression analyses were done to explore sample size (n<median=0 and n≥median=1) and outcome definition (self-harm=0, suicide attempt=1, and non-suicidal self-injury=2) as possible sources of between-study heterogeneity for all risk factors.
The presence of potential publication bias was assessed by examination of asymmetry in funnel plots and by applying Egger’s test for the top three risk factors that had the most information
Included studies were done across 20 countries (11 [31%] in England and Wales) and published from 1972 to 2020. Median sample size was 785 (IQR 142–2119), ranging from 60 to 263 794 prisoners. 20 (57%) of the 35 studies focused solely on either men (k=15) or women (k=5). In 15 studies, the sample included both male and female prisoners, with the mean proportion of women equalling 12·3% (SD 8·7), but only two studies provided data disaggregated by sex
The most frequent designs were case-control studies (k=17; 49%); 12 studies were cross-sectional and six were cohorts. We identified only two prospective studies.
The three largest studies were retrospective analyses of routinely collected data, accounting for 609 366 (91·8%) of the people in the pooled sample. The most common outcome investigated was self-harm (k=15), followed by (near-lethal) suicide attempt (k=12) and non-suicidal self-injury (k=8). Of all 663 735 prisoners included, 24 978 (3·8%) had self-harmed in prison.
In terms of study quality measured by the Newcastle-Ottawa Scale, of 9 possible points, the median score for the cohort studies was 8·5 (IQR 7–9) and the median score for the case-control studies was 6 (5–8).
Of 8 possible points available in the modified Newcastle-Ottawa Scale, the median score for the cross-sectional studies was 6 (5–6). Overall, 18 (51%) of the 35 studies included were judged to be of high quality and four (11%) were categorised as being of low methodological quality
There were large variations in the sample sizes contributing to risk estimates. The largest samples were for sex (n=644 812) and violent offending (n=520 581). Only two risk factors (substance use disorder [n=766] and family history of suicide [n=382]) were calculated on the basis of a pooled sample of less than 1000 prisoners. Various sociodemographic factors were associated with self-harm in prison, with pooled ORs ranging from 1·5 to 2·5 . The three strongest risk factors within this domain were homelessness (OR 2·5, 95% CI 1·8–3·3), unemployment before incarceration (1·6, 1·3–2·1), and being younger than 30 years (2·0, 1·4–2·9). Female sex showed a small increase in risk but a non-significant association with self-harm (1·3, 0·7–2·2). Criminological variables, including violent offences (1·8, 1·3–2·4) and having a previous incarceration (2·0, 1·3–3·1), were risk factors for self-harm in prison . Being sentenced for more than 5 years (2·3, 1·9–2·7) or serving a life sentence (2·0, 1·2–3·3) doubled the odds of self-harm.
Any current psychiatric diagnosis was significantly associated with self-harm, particularly major depression and borderline personality disorder . By diagnosis, the odds of self-harm were increased for major depression, borderline personality disorder, psychotic disorder, anxiety disorder, and substance use disorder (OR range 2·3–9·3; Proxies for psychiatric disorders were also associated with increased odds of self-harm, particularly psychiatric treatment in prison.
Sample size was only significantly associated with heterogeneity in meta-regression for nationality (B=1·2; p=0·039) in that studies with larger samples (n≥785) found a larger effect for nationality relative to studies with smaller samples (n<785). Outcome definition was a moderator only for the relationship between major depression and self-harm (B=–2·1; p=0·031), with a stronger effect observed for major depression in studies examining suicide attempt compared with non-suicidal self-injury.
Overall, the results suggest that neither sample size, nor outcome definition, explained the heterogeneity in the association between most risk factors and self-harm.
In meta-regression analyses, we examined sample size and outcome definition as possible sources of between-study heterogeneity in risk estimates
Additionally, we did post-hoc analyses on the leading risk factors from each domain (ie, homelessness, sentenced for 5 years or more, current suicidal ideation, solitary confinement, and childhood sexual abuse) and again found no evidence for publication bias (all p≥0·18).
Sensitivity analyses showed that risk factor estimates did not materially change (both in terms of strength and significance of effects) when low quality studies were excluded from the analyses
The present meta-analysis synthesised data from nearly 50 years of research examining risk factors for self-harm in more than half a million prisoners. Across 40 risk factors investigated, the strongest associations with self-harm were past and current suicidality and markers of psychiatric morbidity.
Overall, we found strong effects for modifiable clinical and custodial variables, moderate effects for historical variables, and smaller effects for sociodemographic and criminological variables.
Many of the identified risk factors are similar to those found for self-harm in the general population.
Meta-analyses of longitudinal studies have highlighted suicidal ideation, previous self-harm, and psychiatric disorders as replicated risk factors, although the strength of associations was typically stronger in our meta-analysis.
However, there was one notable difference. Antisocial personality disorder, despite being strongly associated with self-harm in the community, was not linked with self-harm in prison.
This difference might reflect the high prevalence of antisocial personality disorder in prisoners, for which diagnostic criteria overlap with the reasons for entering prison.
There was an increased risk of self-harm in female prisoners, although this was not statistically significant. This non-significance contrasts with findings in the general population, in which female sex as a risk factor for self-harm is stronger, and one high-quality population study of UK prisoners that reported that the odds of self-harm was four times higher in women than in men
In addition, we found that environmental factors specific to prison, including solitary confinement, disciplinary infractions, victimisation during imprisonment, and poor social support, were clearly associated with self-harm
Walmsley, R
World prison population list, twelfth edition
https://www.prisonstudies.org/sites/default/files/resources/downloads/wppl_12.pdf
Date accessed: March 30, 2020
Fazel, S ∙ Hayes, AJ ∙ Bartellas, K ∙ et al.
Mental health of prisoners: prevalence, adverse outcomes, and interventions
Lancet Psychiatry. 2016; 3:871-881
Hawton, K ∙ Linsell, L ∙ Adeniji, T ∙ et al.
Self-harm in prisons in England and Wales: an epidemiological study of prevalence, risk factors, clustering, and subsequent suicide
Borges, G ∙ Nock, MK ∙ Haro Abad, JM ∙ et al.
Twelve-month prevalence of and risk factors for suicide attempts in the World Health Organization World Mental Health Surveys
J Clin Psychiatry. 2010; 71:1617-1628
Klonsky, ED
Non-suicidal self-injury in United States adults: prevalence, sociodemographics, topography and functions
Psychol Med. 2011; 41:1981-1986
Fazel, S ∙ Cartwright, J ∙ Norman-Nott, A ∙ et al.
Suicide in prisoners: a systematic review of risk factors
J Clin Psychiatry. 2008; 69:1721-1731
Pratt, D ∙ Appleby, L ∙ Piper, M ∙ et al.
Suicide in recently released prisoners: a case-control study
Psychol Med. 2010; 40:827-835
