Premature mortality represents an important outcome for crime and violence prevention and related social interventions, but little is known about premature mortality in relation to these interventions. This article assesses the impact of interventions for at-risk and ...
Premature mortality represents an important outcome for crime and violence prevention and related social interventions, but little is known about premature mortality in relation to these interventions. This article assesses the impact of interventions for at-risk and criminally-involved youths and adults on premature mortality over the life-course. Systematic review methods are used, including comprehensive search strategies to identify, screen, and code eligible studies. Meta-analytic techniques are used to assess the impact of interventions on premature mortality, the influence of key moderators, and cause of mortality (natural vs. unnatural). A total of 11 studies met the inclusion criteria. Studies originated in five countries and were reported in the last two decades. Sample size can be considered large (> 500 participants) for all but two of the studies. Analysis of pooled effects showed a non-significant impact of interventions on premature mortality (OR = 0.89; 95% CI: 0.46, 1.74). Life-course stage, intervention type, and evaluation design did not moderate the mean effect. Of the nine studies that reported cause of mortality, interventions were associated with an increased likelihood of death by unnatural causes (OR = 1.78; 95% CI: 1.05, 2.39; p = .03). We conclude that greater attention needs to be paid to evaluating and understanding premature mortality over the life-course as part of the study of the effectiveness of crime and violence prevention interventions, and research opportunities exist to make an immediate contribution to this body of knowledge.
Keywords: crime and violence prevention; premature mortality; public health; systematic review
Welsh, B.C., Zane, S.N, and Reeves, J. Steven (2022). Impact of Interventions for At-Risk and Criminally Involved Youths and Adults on Premature Mortality over the Life-Course: A Systematic Review and Meta-analysis. Journal of Developmental and Life-Course Criminology, 8, 25-46
There is an extensive and growing body of research on the effectiveness of interventions for at-risk and criminally-involved youths and adults in preventing delinquency, crime, and violence over the life-course (e.g., Catalano et al., 2012; Mikton et al., 2016; Farrington et al., 2017). Oftentimes, these interventions also have the effect of leading to improvements in other life-course outcomes, including education, parenting, employment, substance abuse, mental health, and physical health (Farrington and Welsh, 2007; Catalano et al., 2012; Farrington et al., 2017). It has been hypothesized that this is related to the co-morbidity of risk factors for delinquency and later offending, as well as many interventions being multi-modal (Yoshikawa, 1994; Elliott and Fagan, 2017).
As part of a larger construct of physical health, premature mortality represents another important outcome of these interventions. The importance of this outcome extends over the life-course, from childhood through old age, and it is influenced by both distal and proximal risk factors. It is also the case that, whether in the context of research trials or large-scale programs, the avoidance of loss of life needs to be of paramount concern.
The definition of premature mortality can vary with the nature of the research. Long-term follow-up studies use a range of fixed or minimum ages to demarcate premature mortality, often considering the average age of study participants or the expected life span of the cohort. For example, McCord’s (1978) long-term follow-up of the Cambridge-Somerville Youth Study (mean age = 45 years) defined premature mortality as occurring before age 35. In studies with high-risk participants (e.g., gang members, incarcerated offenders), a cut-off age may be less important than their relative risk of mortality compared to a control group or the general population (e.g., Teplin et al., 2005). In other studies, sometimes with higher-risk participants, premature mortality may be examined ahead of schedule and owing to its “public health importance,” and this was the case in the 18-year follow-up of the Nurse-Family Partnership program in Memphis, Tennessee (mean ages: mothers = 39.4 years; children = 20.6 years; Olds et al., 2014, p. 801).
Drawing upon this group of studies, the main objective of this article is to conduct a systematic review and meta-analysis to assess the impact of interventions for at-risk and criminally-involved youths and adults on premature mortality over the life-course. Several key factors motivated this review’s focus on premature mortality in relation to interventions. In the first place, there is some indication of increased research activity in this area in the last decade, with studies reporting mixed effects on premature mortality (see e.g., Sherman and Harris, 2013; Olds et al., 2014; Welsh et al., 2019). Carrying out a systematic review will allow us to identify all of the relevant studies and quantify the overall impact of these interventions on premature mortality. To our knowledge the current study represents the first systematic review on this specific topic. It is important to note that studies on imprisonment and mortality are beyond the scope of this review (see Kinner et al., 2013).
A second factor is that there is an extensive and robust body of research that has investigated the association of criminal offending and premature mortality, with the first systematic review recently published (Skinner and Farrington, 2020). We summarize the key findings of this literature in the next section, but it is worth noting that the research shows that involvement in criminal offending is associated with increased risk of early death. By implication, interventions that are effective in preventing delinquency and later criminal offending could also have a desirable impact on premature mortality.
Another key factor that motivated this systematic review’s focus on premature mortality in relation to interventions is the need to better understand the impact of interventions on premature mortality over different stages of the life-course. It is not satisfactory to draw blanket conclusions about the impact of interventions on premature mortality without consideration for the life-course stage of participants.
Association of Criminal Offending and Premature Mortality
There is an extensive and robust literature on the relationship between criminal and violent offending and premature mortality over the life-course (see Skinner and Farrington, 2020). As documented by Zane et al. (2019), much of this literature is focused on offender-based samples, either in adolescence or adulthood (see, e.g., Chassin et al., 2013; Coffey et al., 200; Daigle and Naud, 2012; Lattimore et al., 1997; Lindqvist et al., 2007; Teplin et al., 2005), as well as on prisoner samples (see, e.g., Ferriter et al., 2016; Patterson, 2013; Spaulding et al., 2011). The consensus of these studies is that rates of premature mortality are higher among offenders compared to the general population, which includes findings of unnatural causes of mortality, such as homicide or suicide (Zane et al., 2019). It is also the case that a number of general population studies show a relationship between offending and premature mortality (e.g., Piquero et al., 2014; Elonheimo et al., 2017).
In the first systematic review to examine the relationship between offending and premature mortality, Skinner and Farrington (2020) drew upon prospective and retrospective longitudinal studies targeted on community offenders (i.e., not in prison or in psychiatric facilities) compared to community non-offenders or the general population. The authors found that community offenders (compared to community non-offender and general population samples) were significantly more likely to die prematurely and from both natural and unnatural causes.
Several studies have also investigated the relationship between criminal offending and mortality over the full life-course, from childhood to old age (i.e., beyond the age of 60). Laub and Vaillant (2000) carried out the first study on offending and mortality into old age, drawing on the Gluecks’ longitudinal survey of 500 delinquent and 500 matched non-delinquent boys from Boston (Glueck and Glueck, 1950). By the latest follow-up at age 65, 201 men (or 42.3%) from the delinquent group (n = 475) had died compared to 123 (or 27.0%) from the non-delinquent group (n = 456), a significant difference (p < .001). While Laub and Vaillant found that offending was strongly associated with mortality through this advanced age, key early life-course explanations of this association (e.g., dysfunctional upbringing, poor education, adolescent delinquency) appeared to be much less important than unhealthy behaviors in adulthood (e.g., alcohol abuse, poor self-care).
Van de Weijer and colleagues (2016) report on a long-term follow-up of a high-risk group of males and females (n = 501) who were born between 1918 and 1959 (mean = 1932) in the Netherlands. Offending was measured as total convictions over the life-course, and the authors controlled for parental age of death in addition to the participant’s gender, marital status, and offense type. By the latest follow-up in 2007, 43% of the sample had died, with an average age of mortality of 62.4 years. The authors found no significant difference in mortality hazard between offenders and non-offenders. When offending was divided into categories, driving under the influence was the only offense that was predictive of mortality.
In the most recent study, Zane et al. (2019) report on a 70-year follow-up of the Cambridge-Somerville Youth Study (CSYS), a longitudinal-experimental study of 506 disadvantaged boys that began in 1939. Following the analytic strategy of McCord (1984), participants were drawn from the study’s longitudinal arm (i.e., treatment group; n = 253). Data included court convictions of criminal offenses collected during middle age (mean = 47 years) and death records collected during old age (up to age 89). Through the end of 2016, a total of 216 participants (or 85.4%) were deceased. The authors found that mortality was related to offending over the life-course, but only when offending was measured using group-based trajectory modeling, and only from middle age into old age. Specifically, from middle age onward, life-course persistent offenders were more likely than adolescent-limited offenders and non-offenders to die earlier and from unnatural causes. Results also indicated that childhood risk factors for delinquency were not associated with mortality risk over the life-course.
Premature Mortality as a Key Outcome of Crime and Violence Prevention Interventions
Several factors point to the importance of evaluating the effects of crime and violence prevention and related social (e.g., foster care, residential care) interventions on premature mortality. One factor has to do with premature mortality being part of a larger construct of health problems (Shepherd et al., 2009), which are often a direct or indirect focus of developmental or other types of interventions designed to prevent crime or violence (Farrington et al., 2017; Farrington and Welsh, 2007). Closely related to this is the moderate to strong association of criminal offending and premature mortality over the life-course (Skinner and Farrington, 2020). Yet another factor has to do with a more comprehensive approach to crime prevention, perhaps best captured by the following:
If offending increases mortality risk, then crime may be framed as a public health problem as well: saving individuals from lives of crime would then also lead to healthier lives, fewer years of life lost, and possibly more productive, better socially integrated citizens. (van der Weijer et al., 2016, p. 92)
As we find in the present systematic review, only a small number of studies have investigated premature mortality as an outcome in evaluating crime and violence prevention and related social interventions. In recent years, some well-known intervention studies in criminology, criminal justice, and social welfare have been the subject of long-term follow-ups to investigate premature mortality. These studies have taken advantage of low participant attrition rates and improved availability of and access to official death records (e.g., the National Death Index in the U.S.).
One of these studies is the Milwaukee Domestic Violence Experiment carried out by researchers and the Milwaukee Police Department between 1987 and 1988 (Sherman et al., 1991). In a 23-year follow-up, Sherman and Harris (2013) investigated the effects of arrest on the mortality rates of suspects. Although marginally significant, the authors found that suspects who received the sanction of arrest, compared to a warning condition, were almost three times as likely to have died of homicide.
Another study is the CSYS. In a 72-year post-intervention follow-up of treatment and control participants (n = 506), Welsh et al. (2019) investigated whether the iatrogenic effects on mortality that were observed by McCord (1981) in middle age (mean = 47 years) persisted or changed in old age (up to age 90 years). Records were located for 96.4% of the participants (488 of 506), with a total of 446 participants (or 88.1%) confirmed deceased. Matched-pairs analyses showed no significant differences for all outcomes of interest: mortality at latest follow-up; premature mortality (younger than 40 years); and cause of mortality (natural vs. unnatural). In not being able to detect iatrogenic effects on mortality, the main implication is that the iatrogenic effects on mortality experienced in middle age did not persist in old age.
The systematic review was conducted in accordance with the guidelines of the Campbell Collaboration (2014).
Criteria for Inclusion of Studies
In selecting studies for inclusion in the review, the following criteria were used:
Participants: Interventions must be targeted on (a) children or adolescents (hereafter youths) who are at-risk for involvement in delinquency or (b) youths or adults who are offenders in the community or under the supervision of the justice system. At-risk for involvement in delinquency involves the presence of one or more individual-, family-, or community-level risk factors for delinquency (e.g., poor school achievement, family discord, growing up in a high-crime neighborhood). Some studies referred instead to high risk, which involves an elevated state of risk for more serious delinquency or criminal offending and includes involvement in antisocial behaviors.
Nature of intervention: Studies must be focused on crime or violence prevention or related social interventions. Crime and violence prevention interventions are those that adopt a developmental- or community-based approach to reducing the onset or continuation of criminal activity. Developmental interventions aim to “prevent the development of criminal potential in individuals, especially those targeting risk and protective factors discovered in studies of human development” (Tonry and Farrington, 1995, p. 2). Examples include social skills training for children and behavioral parent training. Community interventions aim to “change the social conditions that influence offending in residential communities” (Tonry and Farrington, 1995, p. 2). Examples include substance abuse treatment services and gang intervention programs. It is important to reiterate that studies on imprisonment and mortality are beyond the scope of the review. Related social interventions are concerned with the social welfare and protection of youths who are at-risk for involvement in delinquency. Some examples include foster care or residential school.
Outcome of interest: There is an outcome measure of premature mortality or early death. As noted above, the definition of premature mortality can vary with the nature of the research. For example, this can include long-term follow-ups of interventions targeted at at-risk participants or short-term follow-ups of interventions targeted at high-risk participants.
Form of intervention: Intervention studies can take the form of programs, practices, or policies. Drawing on Elliott and Fagan (2017, p. 95, table 3.7), programs are defined as a “set of specific activities with defined protocols, manuals, and methods of delivery.” Practices are a “general type of approach which can be used as part of many specific programs.” Policies are “regulations or laws intended to reduce crime across large populations.”
Evaluation design: The evaluation design is of high methodological quality. This includes randomized controlled experiments or quasi-experiments. In the case of quasi-experiments, the minimum design involves a moderate to high degree of comparability between the treatment and control groups. Comparability can be achieved by a range of techniques, including different forms of matching (e.g., pair-matching, propensity-score matching, cohort matching) and the use of control variables.
Study type: Both published and unpublished studies are considered. The search for and identification of unpublished studies is important in an effort to address concerns about publication bias (Wilson, 2009).
Geography, timeframe, and language: Studies are drawn from any geographical region and period of time and are not limited to the English language.1
The following search strategies were used to locate studies meeting the review’s inclusion criteria:
Electronic bibliographic databases: The following 13 databases were searched: Google Scholar; PsycINFO; Criminal Justice Abstracts; National Criminal Justice Reference Service Abstracts; Sociological Abstracts; Dissertation Abstracts; Medline; Catalog of U.S. Government Publications; Australian Criminology Database; Social Sciences Abstracts; Social Sciences Citation Index (through Web of Science); Center for Research Libraries; and the International Bibliography of the Social Sciences. These databases were selected because they had the most comprehensive coverage of criminology and criminal justice, social and behavioral sciences, and public health literatures. They include the top databases recommended by the Campbell Collaboration.
The databases were searched using the base terms of “crime prevention,” “violence prevention,” and “delinquency prevention,” combined with each of the following secondary terms: “premature mortality,” “early death,” “mortality,” and “death.” For example, for “violence prevention,” the following combinations were used: “violence prevention AND premature mortality,” “violence prevention AND early death,” “violence prevention AND mortality,” and “violence prevention AND death.”
Literature reviews on mortality and crime or crime prevention: We identified two relevant reviews: a meta-analysis of drug-related deaths of previously incarcerated offenders (Merrall et al., 2010), and a systematic review and meta-analysis of premature mortality in offenders (Skinner and Farrington, 2020).
References of studies: We examined the references or bibliographies of all of the studies that met the criteria for inclusion in the review, as well as all of the excluded reports that we determined to be evaluation studies.
Forward citations: We used the “cited by” function in Google Scholar to conduct forward citation searches of all of the studies that met our inclusion criteria.
Coding and Protocol
The following key features of the studies were coded: author and date; outlet (published or not); location; baseline sample (number of participants, age, and risk level); intervention (type, duration, and context); control group; evaluation design; and follow-up period.
A coding protocol was established by the research team. The first step involved the researchers meeting to develop the criteria for inclusion of studies and the measures to be coded (as noted above). Next, studies were collected by the research team (see below for full details) and coding was carried out by one of the researchers. The researchers met and communicated periodically to discuss the coding of all of the studies and resolve any questions.
Meta-analytic techniques were used to determine the size, direction, and statistical significance of the impact of interventions on premature mortality. Program effect sizes were weighted on the variance of the effect size and the study sample size (Lipsey and Wilson, 2001). The main measure of effect size was the odds ratio (OR). This was calculated from raw binary event data that was provided by most of the studies. Computations were performed using Biostat’s Comprehensive Meta Analysis version 3.3.2
The search strategies yielded an estimated 7,100 references, with an overwhelming majority from electronic bibliographic databases.3 Figure 1 summarizes the process of identifying, collecting, and screening studies that met the criteria for inclusion in the systematic review. Based on a review of titles, a total of 152 studies (64 from electronic databases and 88 from other search strategies) were identified as potentially relevant. These studies were retrieved and their abstracts reviewed. One-third of these studies (n = 51) were excluded, mostly because there was no intervention (n = 22) or there was no focus on premature mortality or morality in general (n = 15). The next step involved acquiring and screening the full-text articles or reports of the remaining 101 studies. As shown in the final box of Figure 1, a total of 11 studies—based on 11 independent samples—met the inclusion criteria. The other 90 studies were excluded for four reasons: (a) no intervention (n = 47); (b) no focus on premature mortality or mortality in general (n = 15); (c) not an impact evaluation (n = 7); and (d) no control or comparison group (n = 21).
Flowchart for selection of studies
Details of the Included Studies
Studies originated in five countries and three continents (Australia, Europe, and North America), with half of the studies taking place in the United States. Table 1 summarizes key characteristics and the results of the 11 included studies. Nine of the 11 studies (including the latest follow-up of the CSYS) were published in the last 10 years. There was a fair degree of variability in the age and type of study participants: at-risk for delinquency (children) or offending (mothers) (n = 4); high risk children and youths, which involved a mix of individuals in foster or residential care and juvenile corrections (n = 5); and criminally-involved (current and former) adults (n = 2). Not including the Nurse-Family Partnership study for mothers, males were overrepresented in nine studies. The baseline sample size can be considered large (> 500 participants) for all but two of the studies.
Summary of included studies
Study name and authors
Participants and sample size (baseline)
Intervention type, context, and duration
Evaluation design and follow-up (post-intervention)
Impact on premature mortality, by life-course stage
Prevalence of unnatural deaths
Pritchard and King (2000)
“English Southern county,” UK
High risk male youths (age = 11-15 years); N = 653 (T=438 youths in residential care; C=215 youths in exclusion units)
T = “Looked-After” or residential care (for youths with family breakdown); 1-5 years
C = “Exclusion Units” (for youths excluded from school but no services)
QExp (cohort matching); homicide (3-8 years); mortality by suicide (3-8 years)
Adolescence to early adulthood (age = 16-25 years): 0% (T=0) vs. 0.9% (C=2)
and Ribe (2001)
High risk children (age < 13 years); N = 23,768 (T = 13,100 children in foster care [52.8% male]; C = 10,668 children from adverse backgrounds [54.6% male])
Foster care services (45% > 5 years)
QExp (cohort matching); mortality (1-8 years)
Early adulthood (age = 19-26 years): 0.8% (T=103) vs. 1.4% (C=152)
Unnatural deaths: 66% (T=68/103) vs. 50.7% (C=77/152)
Bullock and Gaehl (2012)
United Kingdom (England and Wales)
High risk children (age = 5-17 years); N = 301 (T=152 [55.9% male]; C=149 [52.3% male])
State “in care” services
T = placement > 2 years
C = placement < 6 weeks
QExp (cohort matching); offending (25-30 years); mortality (25-30 years)
Early to middle adulthood (age = 30-47 years): 4.6% (T=7) vs. 2.7% (C=4)
Unnatural deaths: 57.1% (T=4/7) vs. 50% (C=2/4)
Milwaukee Domestic Violence Experiment; Sherman and Harris (2013)
Milwaukee, WI, USA
Adult suspects (mean age = 32 years; 90% male, 56% unemployed, 64% prior arrest); N = 1,128 cases (T=756; C=372)
Domestic violence police response
T = arrest at scene of violence and detain at police station for 3-12 hours
C = warning
RCT; crime (1 year); mortality (23 years)
Middle adulthood: 13.6% (T=103) vs. 12.6% (C=47)
Unnatural deaths: 31.1% (T=32/103) vs. 25.5% (C=12/47)
Nurse-Family Partnership (children); Olds et al. (2014)
Memphis, TN, USA
At-risk children (age = preterm to birth); N = 706 (T=217 [49.0% male]; C=489 [48.3% male])
Home visitation services by nurses; 2 years
RCT; crime (16 years); mortality (18 years)
Early adulthood (mean age = 20.6 years): 0.9% (T=2) vs. 2.9% (C=14)
Unnatural deaths: 0% (T=0/2) vs. 64.3% (C=9/14)
Nurse-Family Partnership (mothers); Olds et al. (2014)
Memphis, TN, USA
At-risk mothers (with at least 2 socio-demographic risk factors); N = 1,138 (T=458; C=680)
Home visitation services by nurses; 2 years
RCT; crime (10 years); mortality (19 years)
Middle adulthood (mean age = 39.4 years): 1.3% (T=6) vs. 4.0% (C=27)
Unnatural deaths: 16.7% (T=1/6) vs. 40.7% (C=11/27)
Gisev et al. (2015)
New South Wales, Australia
Ex-prisoners with past opioid dependence (median age = 30.6 years); N = 16,073 (T=7,957 [75.2% male]; C=8,116 [82.3% male])
Opioid substitution therapy administered in community within 7 days of release from prison
QExp (propensity score matching); mortality (6 months)
Early adulthood (approx. median age = 31 years): OR = 0.44 (95% CI: 0.26, 0.74)
Manninen et al. (2015)
High-risk youths (mean age = 15.2 years; 67.1% male); N = 5,201 (T=885; C=4,316)
Residential school with rehabilitation services (for youths with severe conduct disorder) plus voluntary 5-year after-care program; 3-8 years
QExp (matching by socio-demographics); mortality (1-22 years)
Early to middle adulthood (age = 19-41 years): 6.7% (T=59) vs. 1.0% (C=42)
Unnatural deaths: 89.8% (T=53/59) vs. 64.3% (C=27/42)
Genetics of Antisocial Drug Dependence Study; Border et al. (2018)
Denver, CO, and San Diego, CA, USA
High-risk youths (mean age = 16.8 years; 69.0% male; N = 3,766 (T=1,463; C1=1,399; C2=904)
C1=siblings of T
C2=matched to T
Substance abuse and delinquency treatment services (residential and outpatient); treatment and correctional facilities, community, and schools for troubled youths; n.a.
QExp (matching by socio-demographics); mortality (approx. 16 years)
Early adulthood (mean age = 32.7 years): 4.2% (T=62) vs. 2.4% (C1=34) vs. 0.9% (C2=8)
Unnatural deaths: 93.6% (T=58/62) vs. 70.6% (C=24/34)
Perry Preschool; Heckman and Karapakula (2019)
Ypsilanti, MI, USA
At-risk children (age 3-4 years; 58.5% male); N = 123 (T=58; C=65)
Preschool intellectual enrichment and parent education; preschool and home; 1-2 years
RCT; crime (45 years); mortality (45 years)
Middle adulthood (age = 50 years): 6.9% (T=4) vs. 11.8% (C=7)
Unnatural deaths: 50% (T=2/4) vs. 42.9% (C=3/7)
CSYS; Welsh et al. (2019)
Cambridge and Somerville, MA, USA
Under-privileged boys (median age = 10.5 years); N = 506 (T = 253; C = 253)
Individual counseling and mentoring; community and summer camps; mean = 5.5 years
Pair-matched RCT; crime (31 years); mortality (up to 25 years)
Middle adulthood (< 40 years): OR = 1.15 (95% CI: 0.51, 2.64)
Unnatural deaths: OR = 1.23 (95% CI: 0.56, 2.78)
Notes: All effects are reported as treatment (T) versus control (C); CSYS = Cambridge-Somerville Youth Study; RCT = randomized controlled trial; QExp = quasi-experiment; n.a. = not available.
There was also variability in intervention type, context, and duration. Of the 11 included studies, four involved foster or residential care services, four were developmental prevention interventions, and the other three involved a mix of treatment services and police intervention delivered in the community. Five of the studies used randomized controlled experiments and the others used quasi-experiments with matching, including propensity score matching in one study. The post-intervention follow-up period was for the most part rather long, with eight studies in the range of five to 45 years;4 the other three studies reported either a range of years (based on the age of participants) or upon completion of the intervention.
Figure 2 summarizes the results of the 11 included studies in a forest-plot graph. This shows the odds ratio (OR) for premature mortality in each study plus its 95% confidence interval.5 The 11 studies are ordered according to the magnitudes of their effect size. Since the outcome of interest is premature mortality, an OR < 1.0 favors the treatment group. The majority of studies (n = 6) showed no significant effect of interventions on premature mortality, three studies showed a decreased likelihood of premature mortality for the treatment group, and two studies showed an increased likelihood of premature mortality for the treatment group. In pooling the data from the 11 studies, there was no significant impact of intervention on premature mortality, with a weighted mean OR = 0.89 (95% CI: 0.46, 1.74). We also conducted a sensitivity analysis to determine whether any of the 11 studies had a substantial influence on the weighted mean effect. Removal of individual studies did not alter the summary effect, with weighted means ranging from OR = 0.74 to 0.99 (compared to mean OR = 0.89), all of which remained non-significant.
Forest plot of the distribution of premature mortality effect sizes
We used the “trim-and-fill” procedure developed by Duval and Tweedie (2000) to estimate the effect of potential data censoring, such as publication bias, on the findings of the meta-analysis (see Figure 3). This procedure uses a diagnostic funnel plot to display the distribution of individual effect sizes around the mean effect size, where the y-axis represents sample size (standard error) and the x-axis represents effect size (logged OR). In the absence of publication bias, the plot should take the shape of a funnel evenly distributed around the mean (Wilson, 2010). As shown in Figure 3, no studies were imputed to account for any data censoring, indicating no evidence of publication bias.
Funnel plot of included studies with imputed value from trim-and-fill analysis
Given that the 11 studies are not necessarily replications of one another, it is especially important to examine potential heterogeneity among observed effects. The distribution of the 11 effect sizes was highly heterogeneous (Q = 144.13, I2 = 93.06, df = 10, p = .000). The I2 statistic is especially informative, indicating that the percentage of the dispersion in effects across studies is due to heterogeneity as opposed to chance (Altman et al., 2003). This suggests the presence of moderators of the effects, which can be either methodological or substantively related to the studies (Wilson, 2010). In other words, the effects of interventions on premature mortality may vary significantly across studies, and may vary according to study characteristics, rather than these being one uniform effect of such interventions on premature mortality.
Of particular importance is an understanding of whether effect sizes vary according to the life-course stage when premature mortality is measured. Two main categories of life-course stages could be discerned from the studies: early adulthood (approximate age range of 18 to mid-30s; n = 5) and middle adulthood (approximate age range of late 30s to 50s; n = 4). A third category of early to middle adulthood, representing an overlap of the other two life-course stages (approximate age range of 19 to early 40s), was created to account for the other two studies. We carried out a categorical moderator analysis to investigate the influence of life-course stage. Figure 4 summarizes the results in a forest-plot graph. For the five early adulthood studies, the summary effect was OR = 0.61 (95% CI: 0.30, 1.25, ns); for the two early to middle adulthood studies, the summary effect was OR = 4.05 (95% CI: 1.03, 15.99, p = 0.046); and for the four middle adulthood studies, the summary effect was OR = 0.78 (95% CI: 0.43, 1.39, ns). Heterogeneity among the three sub-groups was marginally significant (Q = 5.89, df = 2, p = .053), suggesting that life-course stage may be related to the observed effects. However, caution is needed in interpreting this finding. For one, only the early to middle adulthood studies reached a significant summary effect. Also, it appears that the study by Manninen et al. (2015), with a large positive effect size, represents an outlier across all 11 studies and influenced the overall summary effect for this life-course stage.
Forest plot of the distribution of premature mortality effect sizes by life-course stage
We also identified evaluation design as a categorical moderator of interest, and used it to explore variability in effects according to whether the intervention was evaluated using a randomized controlled experiment or quasi-experiment. Figure 5 summarizes the results in a forest-plot graph. For the five randomized controlled experiments, the summary effect was OR = 0.70 (95% CI: 0.40, 1.24), and for the six quasi-experiments, the summary effect was OR = 1.13 (95% CI: 0.38, 3.35)—both non-significant. Heterogeneity between the two subgroups was not significant, suggesting that evaluation design is not related to the observed effects.
Forest plot of the distribution of premature mortality effect sizes by evaluation design
Also important as a potential moderator of effect sizes was intervention type. Three broad categories of intervention could be discerned from the studies: developmental prevention (n = 4); foster/residential care (n = 4); and community treatment (n = 3). Figure 6 summarizes the results of the moderator analysis in a forest-plot graph. For the four developmental interventions, the summary effect was OR = 0.56 (95% CI: 0.28, 1.12); for the four foster/residential care programs, the summary effect was OR = 1.18 (95% CI: 0.19, 7.32); and for the three community treatment programs, the summary effect was OR = 0.96 (95% CI: 0.46, 1.99)—all non-significant. Heterogeneity among the three sub-groups was not significant (Q = 1.38, df = 2, p = .50).
Forest plot of the distribution of premature mortality effect sizes by intervention type
Lastly, nine studies reported cause of mortality among participants who died, and we performed a supplemental meta-analysis for whether premature mortality was unnatural (i.e., not due to natural causes such as medical conditions).6 Figure 7 summarizes the results in a forest-plot graph, ordered according to the magnitudes of their effect size. Six studies showed no significant effect of intervention on cause of mortality, while the other three showed an increased likelihood of unnatural mortality for the treatment group. In pooling effects of the nine studies, interventions were associated with an increased likelihood of participants dying of unnatural causes, with a weighted mean OR = 1.78 (95% CI: 1.05, 2.39; p = .03). A test of heterogeneity showed that the distribution of effect sizes was not significantly heterogeneous (Q = 13.95, I2 = 42.66, df = 8, p = .08).
Forest plot of the distribution of cause of mortality effect sizes
Discussion and Conclusions
Some of the studies included in the present review reported lower rates of premature mortality at different stages of the life-course. While it is long-established that many types of crime and violence prevention and related social interventions can lead to improved health outcomes over the life-course (Catalano et al., 2012; Mikton et al., 2016), the weight of the evidence at this point in time suggests that these interventions are not effective in mitigating the risk of premature mortality. Using life-course stage (i.e., early adulthood, early to middle adulthood, and middle adulthood) as a moderator variable did not substantially alter this main finding. Also, of the nine studies that reported cause of mortality, interventions were associated with an increased likelihood of participants dying of unnatural causes.7
One of the key limitations has to do with the small number of studies that met the criteria for inclusion in the review. This can be a limiting factor for drawing conclusions, as well as being able to generalize results. On the one hand, the small number of studies might be expected because not many studies have carried out long-term follow-ups that would allow for a robust assessment of premature mortality (Farrington and MacKenzie, 2013). On the other hand, it is somewhat surprising because studies of criminally-involved participants with even modest post-intervention follow-ups (e.g., 4 or 5 years)—of which there are many more—could conceivably allow for an equally robust assessment of premature mortality. The latter draws attention to an important priority for future research, and we return to this point below.
Nevertheless, in the absence of a large number of studies on a topic—something that confronts many systematic reviews—there can be good justification for conducting a systematic review. One of the central arguments has been that not doing so does a disservice to policymakers, practitioners, and researchers alike; specifically, these groups need to have access to the best available information.
Another key limitation of the review is that we defined in broad terms the nature of the intervention. This meant considering the full range of crime and violence prevention interventions and related social interventions concerned with the social welfare and protection of youths who are at-risk for involvement in delinquency. As a result, it is more difficult to make recommendations for specific groups or types of interventions. This may be particularly relevant for foster/residential care programs, which had the largest impact (albeit non-significant) on premature mortality among the three intervention types. Ideally, future updates of this review or other similar reviews in the years ahead will have more studies to draw upon and, thus be in a position to examine a more narrowly defined group of interventions.
Directions for Policy and Research
The findings of the present review are somewhat at odds with the state of evidence on the effectiveness of crime and violence prevention interventions in general (see Weisburd et al., 2016). To take developmental crime prevention as one example, a review of systematic reviews by Farrington et al. (2017) found that a wide range of intervention modalities in different domains (e.g., family, school, community) were highly effective in preventing delinquency and later criminal offending. The authors also found that many of the interventions led to improvements in other life-course outcomes, including education, employment, and substance abuse. This difference from the findings of our review may be due to a difference in focus; namely, that our review assesses long-term impacts of interventions whereas most studies on the effectiveness of interventions focus on shorter-term impacts.
Even with a small number of developmental crime prevention studies in this review (n = 4), it is noteworthy that not one of them reported a desirable impact on premature mortality. As an outcome that is a low base rate event, at least during the life-course stages of early through middle adulthood, it seems reasonable for policymakers to consider what is known about premature mortality effects alongside the body of knowledge about other life-course outcomes. Giving greater weight to premature mortality effects as more studies become available seems prudent.
Notwithstanding the top priority of preventing crime and violence, knowledge about premature mortality, including other health outcomes, should be a key priority of intervention research. We see two approaches as having the potential to make an immediate contribution to the knowledge base in this area. The first involves prioritizing early developmental crime prevention interventions that have been the subject of long-term follow-ups and whose participants are approaching or are in middle adulthood. For example, the Abecedarian Project, carried out in North Carolina, provided enriched preschool to children born to low-income, multi-risk families. Participants were in their early thirties at the last follow-up more than 10 years ago (Campbell et al., 2012). Another example is the first trial of the NFP program, carried out in Elmira, New York (Eckenrode et al., 2010), with participant mothers now in their late fifties. Other candidate studies that are eligible now or will be in the next few years to evaluate the impact on premature mortality include the Chicago Child-Parent Center program (Reynolds et al., 2018; participants now in early 40s), the Montreal-Longitudinal Experimental Study (Vitaro et al., 2013; participants now in early 40s), and the Seattle Social Developmental Project (Kosterman et al., 2019; participants now in mid-40s).
The second approach involves giving equal priority to studies of interventions for criminally-involved youths or adults that have carried out post-intervention follow-ups in the range of four to five years or beyond. Participants in some of these studies are at the highest levels of risk for serious and violent reoffending and associated physical injuries (morbidity) and premature mortality (Elliott et al., 2020). Even among a relatively younger age group (e.g., mid- to late 20s), the nature of criminal involvement and risk exposure could warrant a robust examination of premature mortality. Candidate studies could include a range of brand-name evidence-based programs, including multisystemic therapy (MST), functional family therapy, multidimensional treatment foster care, and aggression replacement therapy. As one example, Sawyer and Borduin (2011) conducted a 22-year follow-up of a MST trial, known as the Missouri Delinquency Project, involving 176 serious juvenile offenders; participants would now be in their late forties.
As part of these priorities for future research, it will also be important that future studies differentiate the causes of premature mortality. Did the deceased die of natural or unnatural causes? In the case of unnatural causes, such as homicide or suicide, were these associated with distal or proximal risk factors? Investigating these and other questions will help better understand this important outcome in the study of the effectiveness of crime and violence prevention interventions over the life-course.
References marked with an asterisk (*) indicate studies included in the systematic review.
Altman, D. A., Deeks, J. A., Higgins, J. P. T., & Thompson, S. G. (2003). Measuring inconsistency in meta-analysis. British Medical Journal, 327, 557-559.
*Border, R., Corley, R. P., Brown, S. A., Hewitt, J. K., Hopfer, C. J., McWilliams, S. K., et al. (2018). Independent predictors of mortality in adolescents ascertained for conduct disorder and substance use problems, their siblings and community controls. Addiction, 113, 2107-2115.
*Bullock, R., & Gaehl, E. (2012). Children in care: A long-term follow up of criminality and mortality. Children and Youth Services Review, 34, 1947-1955.
Campbell Collaboration (2014). Campbell systematic reviews: Policies and guidelines. Campbell Policies and Guidelines Series, No. 1. https://doi.org/10.4073/cpg.2016.1.
Campbell, F. A., Pungello, E. P., Burchinal, M., Kainz, K., Pan, Y., Wasik, B. H., Barbarin, O. A., Sparling, J. J., & Ramey, C. T. (2012). Adult outcomes as a function of an early childhood educational program: An Abecedarian Project follow-up. Developmental Psychology, 48, 1033-1043.
Catalano, R. F., Fagan, A. A., Gavin, L. E., Greenberg, M. T., Irwin, C. E., Ross, D. A., & Shek, D. T. (2012). Worldwide application of prevention science in adolescent health. Lancet, 379, 1653-1664.
Chassin, L., Piquero, A. R., Losoya, S. H., Mansion, A. D., & Schubert, C. A. (2013). Joint consideration of distal and proximal predictors of premature mortality among serious juvenile offenders. Journal of Adolescent Health, 52, 689-696.
Coffey, C., Veit, F., Wolfe, R., Cini, E., & Patton, G. C. (2003). Mortality in young offenders: Retrospective cohort study. British Medical Journal, 326, 1064-1066.
Daigle, M. S., & Naud, H. (2012). Risk of dying by suicide inside or outside prison: The shortened lives of male offenders. Canadian Journal of Criminology and Criminal Justice, 54, 511-528.
Duval, S., & Tweedie, R. (2000). A nonparametric ‘trim and fill’ method of accounting for publication bias in meta-analysis. Journal of the American Statistical Association, 95, 89-98.
Eckenrode, J., Campa, M., Luckey, D. W., Henderson, C. R., Cole, R. E., Kitzman, H. J., Anson, E. A., Sidora-Arcoleo, K., Powers, J., & Olds, D. L. (2010). Long-term effects of prenatal and infancy nurse home visitation on the life course of youths: 19-year follow-up of a randomized trial. Archives of Pediatrics and Adolescent Medicine, 164, 9-15.
Elliott, D. S., Buckley, P. R., Gottfredson, D. C., Hawkins, J. D., & Tolan, P. H. (2020). Evidence-based juvenile justice programs and practices: A critical review. Criminology & Public Policy, 19, 1305-1328.
Elliott, D. S., & Fagan, A. A. (2017). The prevention of crime. Malden, MA: Wiley.
Elonheimo, H., Sillanmäki, L., & Sourander, A. (2017). Crime and mortality in a population-based nationwide 1981 birth cohort: Results from the FinnCrime study. Criminal Behaviour and Mental Health, 27, 15-26.
Farrington, D. P., Gaffney, H., Lösel, F., & Ttofi, M. M. (2017). Systematic reviews of the effectiveness of developmental prevention programs in reducing delinquency, aggression, and bullying. Aggression and Violent Behavior, 33, 91-106.
Farrington, D. P., & MacKenzie, D. L. (2013). Long-term follow-ups of experimental interventions. Journal of Experimental Criminology, 9, 385-388.
Farrington, D. P., & Welsh, B. C. (2007). Saving children from a life of crime: Early risk factors and effective interventions. New York: Oxford University Press.
Ferriter, M., Gedeon, T., Buchan, S., Findlay, S., Mbulawa, D., Powney, M., & Cormac, I. (2016). Eight decades of mortality in an English high-security hospital. Criminal Behaviour and Mental Health, 26, 403-416.
*Gisev, N., Shannahan, M., Weatherburn, D. J., Mattick, R. P., Larney, S., Burns, L., & Degenhardt, L. (2015). A cost-effectiveness analysis of opioid substitution therapy upon prison release in reducing mortality among people with a history of opioid dependence. Addiction, 110, 1975-1984.
Glueck, S., & Glueck, E. T. (1950). Unraveling juvenile delinquency. New York: Commonwealth Fund.
*Heckman, J. J., & Karapakula, G. (2019). The Perry preschoolers at late midlife: A study in design-specific inference. Working Paper No. 25888. Cambridge, MA: National Bureau of Economic Research.
Kinner, S. A., Forsyth, S., & Williams, G. (2013). Systematic review of record linkage studies of mortality in ex-prisoners: Why (good) methods matter. Addiction, 108, 38-49.
Kosterman, R., Hawkins, J. D., Hill, K. G., Bailey, J. A., Catalano, R. F., & Abbott, R. D. (2019). Effects of social development intervention in childhood on adult life at ages 30 to 39. Prevention Science, 20, 986-995.
Lattimore, P. K., Linster, R. L., & MacDonald, J. M. (1997). Risk of death among serious young offenders. Journal of Research in Crime and Delinquency, 34, 187-209.
Laub, J. H., & Vaillant, G. E. (2000). Delinquency and mortality: A 50-year follow-up study of 1,000 delinquent and nondelinquent boys. American Journal of Psychiatry, 157, 96-102.
Lindqvist, P., Leifman, A., & Eriksson, A. (2007). Mortality among homicide offenders: A retrospective population-based long-term follow-up. Criminal Behaviour and Mental Health, 17, 107-112.
Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Thousand Oaks, CA: Sage.
*Manninen, M., Pankakoski, M., Gissler, M., & Suvisaari, J. (2015). Adolescents in a residential school for behavior disorders have an elevated mortality risk in young adulthood. Child and Adolescent Psychiatry and Mental Health, 9(46), 1-7.
McCord, J. (1978). A thirty-year follow-up of treatment effects. American Psychologist, 33, 284-289.
McCord, J. (1981). Consideration of some effects of a counseling program. In S. E. Martin, L. B. Sechrest, & R. Redner (Eds.), New directions in the rehabilitation of criminal offenders (pp. 394-405). Washington, DC: National Academy Press.
McCord, J. (1984). A longitudinal study of personality development. In S. A. Mednick, M. Harway, & K. M. Finello (Eds.), Handbook of longitudinal research (vol. 2, pp. 522-531). New York: Praeger.
Merrall, E. L. C., Kariminia, A., Binswanger, I. A., Hobbs, M. S., Farrell, M., Marsden, J., et al. (2010). Meta-analysis of drug-related deaths soon after release from prison. Addiction, 105, 1545-1554.
Mikton, C. R., Butchart, A., Dahlberg, L. L., & Krug, E. G. (2016). Global status report on violence prevention. American Journal of Preventive Medicine, 50, 652-659.
*Olds, D. L., Kitzman, H., Knudtson, M. D., Anson, E., Smith, J. A., & Cole, R. (2014). Effect of home visiting by nurses on maternal and child mortality: Results of a 2-decade follow-up of a randomized clinical trial. JAMA Pediatrics, 168, 800-806.
Patterson, E. (2013) The dose–response of time served in prison on mortality: New York state, 1989-2003. American Journal of Public Health, 103, 523-528.
Piquero, A. R., Farrington, D. P., Shepherd, J. P., & Auty, K. (2014). Offending and early death in the Cambridge Study in Delinquent Development. Justice Quarterly, 31, 445-472.
*Pritchard, C., & King, E. (2000). ‘High risk’ adolescents five years on: The subsequent suicide and homicide of former ‘looked-after’ and ‘exclussion unit’ adolescents. International Journal of Adolescence and Youth, 8, 139-148.
Reynolds, A. J., Ou, S.-R., & Temple, J. A. (2018). A multicomponent, preschool to third grade preventive intervention and educational attainment at 35 years of age. JAMA Pediatrics, 172, 247-256.
Sawyer, A. M., & Borduin, C. M. (2011). Effects of Multisystemic Therapy through midlife: A follow-up to a randomized clinical trial with serious juvenile offenders. Journal of Consulting and Clinical Psychology, 79, 643-652.
Shepherd, J. P., Shepherd, I., Newcombe, R. G., & Farrington, D. P. (2009). Impact of antisocial lifestyle on health: Chronic disability and death by middle age. Journal of Public Health, 31, 506-511.
*Sherman, L. W., & Harris, H. M. (2013). Increased homicide victimization of suspects arrested for domestic assault: A 23-year follow-up of the Milwaukee Domestic Violence Experiment (MilDVE). Journal of Experimental Criminology, 9, 491-514.
Sherman, L. W., Schmidt, J. D., Rogan, D. P., Gartin, P. R., Cohn, E. G., Collins, D. J., et al. (1991). From initial deterrence to long-term escalation: The effects of short-custody arrest for poverty ghetto domestic violence. Criminology, 29, 821-850.
Skinner, G. C. M., & Farrington, D. P. (2020). A systematic review and meta-analysis of premature mortality in offenders. Aggression and Violent Behavior, 53, 1-12 (e101431).
Spaulding, A. C., Seals, R. M., McCallum, V. A., Perez, S. D., Brzozowski, A. K., & Steenland, N. K. (2011). Prisoner survival inside and outside of the institution: Implications for health-care planning. American Journal of Epidemiology, 173, 479-487.
Teplin, L. A., McClelland, G. M., Abram, K. M., & Mileusnic, D. (2005). Early violent death among delinquent youth: A prospective longitudinal study. Pediatrics, 115, 1586-1593.
Tonry, M., & Farrington, D. P. (1995). Strategic approaches to crime prevention. In M. Tonry & D. P. Farrington (Eds.), Building a safer society: Strategic approaches to crime prevention (pp. 1-20). Chicago: University of Chicago Press.
van de Weijer, S., Bijleveld, C., & Huschek, M. (2016). Offending and mortality. Advances in Life Course Research, 28, 91-96.
*Vinnerljung, B., & Ribe, M. (2001). Mortality after care among young adult foster children in Sweden. International Journal of Social Welfare, 10, 164-173.
Vitaro, F., Brendgen, M., Giguère, C.-É., & Tremblay, R. E. (2013). Early prevention of life-course personal and property violence: A 19-year follow-up of the Montreal Longitudinal-Experimental Study (MLES).” Journal of Experimental Criminology, 9, 411-427.
Weisburd, D., Farrington, D. P., & Gill, C. E. (Eds.) (2016). What works in crime prevention and rehabilitation: Lessons from systematic reviews. New York: Springer.
*Welsh, B. C., Zane, S. N., Zimmerman, G. M., & Yohros, A. (2019). Association of a crime prevention program for boys with mortality 72 years after the intervention: Follow-up of a randomized clinical trial. JAMA Network Open, 2(3), 1-11 (e190782).
Wilson, D. B. (2009). Missing a critical piece of the pie: Simple document search strategies inadequate for systematic reviews. Journal of Experimental Criminology, 5, 429-440.
Wilson, D. B. (2010). Meta-analysis. In A. R. Piquero & D. Weisburd (Eds.), Handbook of quantitative criminology (pp. 181-208). New York: Springer.
Yoshikawa, H. (1994). Prevention as cumulative protection: Effects of early family support and education on chronic delinquency and its risks. Psychological Bulletin, 115, 28-54.
Zane, S. N., Welsh, B. C., & Zimmerman, G. M. (2019). Criminal offending and mortality over the full life-course: A 70-year follow-up of the Cambridge-Somerville Youth Study. Journal of Quantitative Criminology, 35, 691-713.