Substance use disorders in prisoners: an updated systematic review and meta‐regression analysis in recently incarcerated men and women

Abstract Aims The aims were to (1) estimate the prevalence of alcohol and drug use disorders in prisoners on reception to prison and (2) estimate and test sources of between study heterogeneity. Methods Studies reporting the 12‐month prevalence of alcohol and drug use disorders in prisoners on reception to prison from 1 January 1966 to 11 August 2015 were identified from seven bibliographic indexes. Primary studies involving clinical interviews or validated instruments leading to DSM or ICD diagnoses were included; self‐report surveys and investigations that assessed individuals more than 3 months after arrival to prison were not. Random‐effects meta‐analysis and subgroup and meta‐regression analyses were conducted. Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines were followed. Results In total, 24 studies with a total of 18 388 prisoners across 10 countries were identified. The random‐effects pooled prevalence estimate of alcohol use disorder was 24% [95% confidence interval (CI) = 21–27], with very high heterogeneity (I 2 = 94%). These ranged from 16 to 51% in male and 10–30% in female prisoners. For drug use disorders, there was evidence of heterogeneity by sex, and the pooled prevalence estimate in male prisoners was 30% (95% CI = 22–38; I 2 = 98%; 13 studies; range 10–61%) and, in female prisoners, was 51% (95% CI = 43–58; I 2 = 95%; 10 studies; range 30–69%). On meta‐regression, sources of heterogeneity included higher prevalence of drug use disorders in women, increasing rates of drug use disorders in recent decades, and participation rate. Conclusions Substance use disorders are highly prevalent in prisoners. Approximately a quarter of newly incarcerated prisoners of both sexes had an alcohol use disorder, and the prevalence of a drug use disorder was at least as high in men, and higher in women.


INTRODUCTION
Prisons around the world detain large numbers of individuals with substance use problems, which increase the risk of mortality after prison release [1][2][3] and repeat offending [4,5]. In addition, alcohol use disorders (AUD) are associated with suicide inside prison [6] and of perpetrating violence and being victimized inside custody [7,8].
The treatment gap for substance use disorders (SUD) inside prison has been reported in many studies [9,10]. Estimates of the prevalence of these disorders in prisoners can assist in planning service provision effectively, targeting scarce resources and developing and evaluating initiatives to reduce the gap between health needs and interventions. A previous systematic review reported ranges for drug abuse and dependence of 10-48% in men and 30-60% in women on reception or arrival to prison. For alcohol abuse and dependence, ranges of 18-30% for men and 10-24% for women were reported [11]. There were very high rates of heterogeneity between these included studies (with I 2 values of more than 80%), which were investigated in subgroup analyses. Lower prevalences were associated with studies where psychiatrists acted as interviewers and higher prevalences for drug use disorders in remand prisoners. However, this review is now dated, with its search for primary studies ending in 2004, and a number of relevant investigations have been published subsequently. In addition, subgroup analyses were the limited number of primary studies by sex, and an updated review will allow for further investigation of sources of between-study variation.
The aim of the current paper is to provide an update of prevalence estimates of alcohol and drug use disorders in prisoners and estimate sources of between-study heterogeneity. As part of this, we have used the term 'substance use disorder', which does not distinguish between 'abuse' and 'dependence'. In this update, we have also conducted meta-analyses to report pooled prevalence estimates and meta-regression to examine sources of variation between included studies.

Search strategy
We identified surveys of alcohol and drug use disorder in general prison populations (defined as remand/detainee and/or sentenced prisoners who are sampled from the whole population of a correctional institution) published between January 1966 and August 2015. For the period January 1966 and January 2004, methods have been described in a previous systematic review conducted by one of the authors (S.F.) [11]. For this update, we searched the following databases from 1 January 2004 to 11 August 2015: PsycINFO, MEDLINE, Global Health, PubMed, CINAHL, National Criminal Justice Reference Service and EMBASE. We used a combination of search terms relating to substance use disorder (i.e. substance*, alcohol, drug*, misuse, dependen*, abuse) and prisoners (i.e. inmate*, sentenced, remand, detainee*, felon*, prison*, incarcerat*), which are same search terms used in the previous review except for the addition of 'incarcerat*'. Additional targeted searches covered relevant reference lists, and non-English papers were translated. We corresponded with authors to clarify data when necessary. We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [12] (Supporting information, Appendix S1) and registered the protocol for this review with PROSPERO (registration code CRD42016036416) [13].

Study eligibility
Inclusion criteria were studies: (a) reporting diagnoses of substance use disorder (i.e. substance abuse and/or dependence) based on clinical examination or by interviews using validated diagnostic instruments [based on DSM (versions III to IV-R; codes 303.90, 304.00-90, 305.00-90, excluding nicotine-related disorders) and ICD versions 9 and 10 (ICD-9: 303-305; ICD-10 codes: F10-19.1-2 except F17)]; (b) with diagnoses based on the previous 12 months from the time when participants were interviewed/examined; and (c) that sampled the general prison population within 3 months of arrival to prison. We excluded studies that selected subgroups for interview (e.g. prisoners referred for treatment, specific categories of offenders), as the aim was to provide a prevalence estimate for the whole prison population [14][15][16]. After correspondence with authors, if studies reported combined prevalence for alcohol and drug [17,18] or combined male and female prevalence, these were excluded [19], as we aimed to report estimates separately by sex and by drug and alcohol use disorder. Studies that reported specific drugs [20,21], self-screening measures [22,23] or solely life-time prevalence were also excluded [24].

Data extraction and analysis
Two researchers (I.Y. and A.H.) extracted independently information on year of publication, geographical location, total sample, sex, prisoner status (remand/sentenced), average age, method of sampling, sample size, participation rate, type of interviewer, diagnostic instrument, diagnostic criteria (ICD versus DSM) and number diagnosed with substance use disorders. If older studies reported dependence prevalence, this was prioritized over abuse, as we considered that these had higher diagnostic validity [32,33] (except when only combined prevalence for abuse and dependence was available). Eligible studies were assessed for quality using the JBI Critical Appraisal Checklist for Studies Reporting Prevalence Data, which uses nine criteria including sample size, sampling, sample description, appropriate statistical analysis and response rates (Supporting information, Appendix S2) [34].
We conducted a random-effects analysis, which assigns similar weights to all studies included in the meta-analysis regardless of sample size [35]. If there were high levels of overall heterogeneity (I 2 > 75%), we also reported estimate ranges as an alternative. Meta-regression analysis was performed to examine sources of between-study heterogeneity on a range of study pre-specified characteristics [i.e. sex, age, publication year, country (United States versus other countries), prisoner status (sentenced versus remand/detainee/unsentenced), participation rate, sample size, diagnostic criteria (ICD versus DSM) and psychiatric interviewer]. Univariable analysis was conducted for both dichotomous and continuous definitions of a variable (e.g. publication year: continuous versus before or after 2000). Multivariable analyses were not conducted due to the limited number of primary studies. If there were fewer than 10 studies that reported an explanatory variable, it was excluded from the meta-regression [36]. Selected continuous variables (study year and proportion sentenced) were converted to dichotomous variables for reporting of pooled prevalence estimates of subgroups. Accordingly, in the metaregression, studies that combined both remand and sentenced prisoners were excluded if: (1) prisoner type comprised more than 10% of the total study participants or was unspecified and (2) separate prevalence data were not provided for each type [37][38][39][40][41][42].
In addition, pooled prevalence estimates of the subgroups that did not have more than one study in each relevant category were not reported, even if they had significant results on meta-regression. Further, we conducted subgroup analyses stratified on pre-specified variables based on our previous review-sex, whether or not the country of origin was United States, remand/detainee versus sentenced prisoner status and whether or not the assessment was conducted by a psychiatrist. We added a new subgroup analysis based on the date of publication (2000, which was approximately the median date). To test for publication bias, funnel plot analysis and Egger's test were conducted on all studies stratified by disorder (i.e. AUD and SUD) and also by sex and disorder [43]. Thus, six separate Egger's tests were performed. Studies with juvenile prisoners or LMI countries were not included, as they were clinically heterogeneous and limited in number. The Egger test quantifies bias captured in the funnel plot analysis with linear regression using the value of effect sizes and their precision [standard error (SE)] and assumes that the quality of study conduct is independent of study size [35] All analyses were conducted in Stata (STATA-IC) version 14 using the following Substance use disorder in prisoners 1727 commands: metan (for random-effects meta-analysis), metareg (for meta-regression), metabias (for publication bias analysis) and heterogi (for calculation of confidence intervals for heterogeneity level).

Drug use disorder
There was evidence of heterogeneity by sex in univariable meta-regression, and prevalence estimates for drug use disorder are stratified accordingly.
Using subgroup analysis, we also investigated possible explanations for between-study variation ( Table 3). This found that there were higher estimates for drug use  disorders in women, and for both drug and alcohol use disorders since 2000, which were consistent with findings on meta-regression. In addition, in alcohol use disorders, there were higher prevalence estimates in sentenced (than remand) prisoners. However, these subgroup analyses had overlapping CIs, apart from a higher estimate for women with drug use disorder compared to men (Figure 3).

Publication bias
There was no evidence of publication bias overall and in subgroups stratified by sex apart from drug use disorder in male prisoners, where there was non-significant evidence of publication bias in the funnel plot analysis (Egger's test, t = 2.19, SE(t) = 4.27, P = 0.051) [37,38,[40][41][42]44,47,48,[50][51][52]57,60]. Visual analysis of the funnel plot suggested asymmetry, but appeared to be mainly attributable to one study [60] with a high prevalence and large standard error, which when removed did not suggest clear publication bias (Supporting information, Appendix S4).

DISCUSSION
This updated systematic review of the prevalence of substance use disorder in prisoners is based on 24 studies and 18 388 individuals in 10 countries. In addition, we identified five studies in juvenile prisoners and two investigations in LMI countries. The sample size in this update is more than double of that a previous systematic review [11], which identified relevant prevalence studies until 2004, and this updated synthesis allowed for an investigation of sources of heterogeneity between included studies. We report two main findings. The first is that alcohol use disorder was highly prevalent in prisoners, with a pooled estimate of 24% (95% CI = 21-27). In men, the lowest estimate suggests that one in six (16%) met the threshold for alcohol use disorder on arrival into prison, and in women it was one in 10. By way of comparison, in the United States in 2013 community rates of past year alcohol use disorder were estimated at 8.7% for men and 4.6% in women [62]. According to the Global Burden of Disease 2015 Study, the global prevalence of alcohol use disorder was 1.5% for males and 0.3% for females (0.9% for both sexes) [63]. The second major finding was that drug use disorder was as high as the alcohol estimates, and possibly higher in female prisoners, with a pooled estimate of 51% (95% CI = [43][44][45][46][47][48][49][50][51][52][53][54][55][56][57][58]. Importantly, the lowest prevalence study in women found that 30% had a drug use disorder. This can be contrasted with US community samples, where 3.4% of men and 1.9% women had such a disorder [62], and 0.8% in men and 0.4% in women (0.6% for both sexes) worldwide [63].
We investigated sources of heterogeneity more carefully than previous work, which led to a number of potentially important findings. First, using meta-regression, we found evidence of increasing drug use disorder in prison studies during the past three decades. This is in contrast with community trends in some high-income countries such as the United States, where drug use disorder had not increased (and alcohol reduced slightly) between 2000 and 2013 [64]. Secondly, two other study characteristics were associated with significant variations in prevalence. Having a higher participation rate was associated with lower rates in drug use disorder, and there were higher rates of drug use disorders in women prisoners. Being assessed by a psychiatrist was also linked with lower alcohol use disorder prevalence in subgroup analyses, although the confidence intervals overlapped. This should inform the interpretation of single studies, particularly if used for service planning and development. One possible explanation for heterogeneity that we did not investigate are the community baseline rates of substance use disorders, and future work could examine this using comparable measures of drug and alcohol use, such as the ongoing Global Burden of Disease [63]. In addition, the reported high prevalence range of 30-56% for substance use disorder in LMI countries needs further research, as it was based on only two investigations.
A number of implications arise from this updated metaanalysis. First, it highlights the opportunity that jails and prisons present to treat substance use disorders [65]. The high prevalences underscore the importance of evidencebased interventions being available to all individuals entering custody. Four areas should be considered to improve management of substance use disorders in prisoners. First, prison arrival centres need to have systems in place to identify individuals with high treatment needs, and treatments should be matched to individual needs [65]. Secondly, acute detoxification management should be available to all entrants to custody, which may include short-term prescription of benzodiazepines for alcohol withdrawal [66] and symptomatic treatment of withdrawal from other substances that may include opioid agonists (such as methadone or buprenorphine). Detoxification programmes may benefit from the use of clinical tools to document withdrawal symptoms [67]. Thirdly, combination pharmacological and psychosocial treatments should be available, considering the high prevalences and the subsequent effects on adverse outcomes, including mortality after release and violent re-offending [68,69]. Finally, considering the high relapse rates, programmes need to link prisoners with community services. Structured, simple and scalable tools to identify those at highest risk [70] and case management [71] may assist in this process. A second implication from the review is that prevalence research needs to consider some areas of improvement. These include separating prevalences by drug and alcohol use disorder, and also providing information stratified by sex and prisoner status (i.e. sentenced or not). Baseline information on sociodemographic and criminal history characteristics (such as those listed in Table 1, including the sample's age structure and index offence) should be provided in new studies, and supplemented with more clinically informative information, such as comorbidities with mental illness [72] and chronic pain, prevalence by individual drugs and most recent treatment. At the same time, as there are now at least 24 studies on prevalence on more than 18 000 prisoners, whether new research should prioritize how treatment can be delivered most effectively to prisoners and former prisoners needs to be considered by funding agencies, researchers and government agencies in criminal justice and public health. Table 3 Pooled prevalence estimates for drug and alcohol use disorders in newly incarcerated men and women by pre-specified subgroups. Some limitations to this review need to be considered. First, there was variation in the diagnostic tools and interviewers used to assess substance use disorders, and we found that psychiatrist interviewers were associated with lower prevalences for alcohol use disorder. To reflect this clinical and statistical heterogeneity, we also reported prevalence ranges. Secondly, as we focused upon substance use disorders on prison entry, these estimates may not reflect treatment needs later in prison or on prison release, where novel psychoactive substances are increasingly problematic and may require different treatment approaches [73]. In addition, the misuse of prescribed medication such as painkillers, anti-epileptics and anxiolytics inside custody needs to be considered, and may further increase treatment needs. Finally, some of the subgroup analyses were based on fewer than 10 studies, and should be interpreted with caution.
In summary, the high prevalence of alcohol and drug use disorders in prisoners remains a key challenge for prison health. Tackling this will probably require interventions at all stages of the criminal justice process -from identifying and treating withdrawal in police custody [74] and on arrival to prison, to opiate maintenance and other treatments during any period in prison [68], to community links being made and integrated treatment provided on release [75]. Comprehensive strategies to prevent relapse of substance dependence are likely to reduce premature mortality, recidivism and subsequent return to prison.