Beyond Risk Level: Timing and Change in Risk Factors as Predictors of Early Mortality Among Formerly Incarcerated Youth

Abstract

Objectives

The current study examined whether risk factors associated with recidivism also influence likelihood of early mortality. The baseline model and shelf-life models assess whether the level and timing of risk factor measurement is associated with differential mortality risk. The dynamic change model assesses whether within-individual change provides explanatory information beyond that obtained from single timepoint measures.

Methods

A cohort of 1,155 formerly incarcerated Canadian youth were followed into adulthood. Items from recidivism risk assessment tools (interpersonal support, antisocial attitudes, academic/vocational skills, substance use, and mental health instability) were used in a series of Cox proportional hazard models.

Results

Earlier baseline measurements of risk factors did not predict mortality, while those assessed closer to the time of death were significantly associated with increased mortality risk. Greater substance use increased mortality risk, while mental health instability unexpectedly decreased mortality risk. The dynamic change model found that within-individual increases in antisocial attitudes were associated with elevated mortality risk.

Conclusions

Timing and change in risk factors are more predictive of early mortality than static measures from adolescence. Future research should consider both community risk factors and correctional system monitoring to better understand the contribution of individual-level risk.

Keywords

early mortality risk assessment longitudinal criminological theory

Based on evidence that criminal legal system-involved populations face an elevated risk of early mortality, research has explored whether established risk factors for offending (e.g., onset of offending, substance use, gang involvement) are also risk factors for early mortality (Binswanger et al., 2020; Kullgren et al., 1998; Laub & Vaillant, 2000; Lattimore et al., 1997; Lidberg et al., 1989; Pritchard et al., 1997; Skinner et al., 2022; Teplin et al., 2005). Although this literature has advanced the idenitification of risk factors for early mortality, focusing on statistical prediction can oversimplify how risk factors operate (Wikström, 2020). More precise means of identifying individuals at heightened risk of early mortality is an important step in responding to calls to reduce early mortality among criminal legal system-involved populations (Butler et al., 2023). The forensic psychology literature and developmental life-course paradigm have provided useful roadmaps for conceptualizing risk, not just in terms of (a) the level of a risk factor (i.e., propensity) but also (b) its temporal proximity in relation to the outcome of interest (i.e., timing; Viljoen et al., 2017a), and (c) whether the level of the risk factor has changed over time (i.e., within-individual change; Le Blanc & Loeber, 1998; Nagin & Paternoster, 2000). The latter two components of risk have been neglected in research on early mortality.

The current study used data on 1,155 participants from the Incarcerated Serious and Violent Young Offender Study (ISVYOS) to examine whether risk factors traditionally associated with recidivism (e.g., substance use, antisocial attitude, lack of social support) also influence the likelihood of early mortality. The analytic sample included participants followed into adulthood who had received a structured intake interview during adolescence and at least two evaluations on the Community Risk Needs Assessment, a tool used by the British Columbia Corrections Service for case management and planning. Approximately 12% of participants died during the follow-up period. A series of Cox proportional hazard models were used to examine whether level, timing, and within-individual change in risk factors were associated with early mortality. We applied two approaches to understanding risk factors for mortality: the shelf-life hypothesis and the dynamic change hypothesis (Vincent & Grisso, 2005).

Shelf-Life Hypothesis

A basic principle of the risk factor paradigm is that higher levels of risk correspond with a greater probability of adverse outcomes (Farrington, 2000). This level-based perspective reflects the dominant approach within the literature on risk factors for mortality, where studies have found that higher levels of gang involvement (Lattimore et al., 1997; Teplin et al., 2014), substance use (Chassin et al., 2013; Kinner et al., 2015; Stattin & Romelsjo, 1995; Stenbacka et al., 2012), and alcohol dependency (Laub & Vaillant, 2000; Teplin et al., 2014) increased risk of mortailty. However, in some cases, these risk factors were measured up to five decades prior to mortality (e.g., Laub & Vaillant, 2000; Piquero et al., 2014). For a static risk factor (e.g., age of offending onset), which does not change, the timing of risk factor measurement may be considered less consequential (Farrington, 2000). From a population heterogeneity perspective, static risk factors are assumed to reflect relatively stable underlying propensities, such as low self-control (Gottfredson & Hirschi, 1990), such that the period of the life-course in which a risk factor was measured may be considered trivial. However, some commonly studied risk factors for mortality (e.g., substance use, gang involvement) may not be static nor stable, suggesting that level alone represents only one component of understanding mortality.

The shelf-life hypothesis addresses the importance of timing by positing that the association between a risk factor and an outcome weakens as time passes between measurement and outcome (Vincent & Grisso, 2005). Although the level of a risk factor is still important, a risk factor measured more proximally to the outcome should exhibit a stronger association than the same risk factor measured further from the outcome. In other words, a risk factor has a shelf life, and waiting too long to reassess someone's risk implies that their risk status may “expire.” This premise justifies correctional practices that emphasize updating risk assessments, on the assumption that more recent assessments better reflect individuals’ current circumstances and risk status (Viljoen et al., 2017a). The shelf-life hypothesis aligns with developmental criminology principles that suggest proximal risk factors better capture individuals’ current situations and therefore may have a stronger impact on the outcome than the same risk factor measured months or years earlier (Chung et al., 2002; Losel & Bender, 2003).

The shelf-life hypothesis can be tested by examining whether risk assessments for reoffending taken more proximally to the occurrence of recidivism improved predictive validity (Davies et al., 2022; Lloyd et al., 2020). The hypothesis is not supported if the timing of risk factor measurement fails to improve statistical prediction, and the association with mortality remains stable regardless of measurement timing (Skinner et al., 2022; Zane et al., 2019). Chassin et al. (2013), using data from the Pathways to Desistance Study, found that some risk factors (i.e., substance use severity, gang membership, gun carrying) measured approximately 18 months prior to death explained mortality above and beyond risk factors measured at baseline, when participants were between the ages of 14 and 18. Although Chassin et al.'s (2013) study supported the importance of accounting for risk factor timing, this was not a direct test of the shelf-life hypothesis because the study did not compare the impact of the same risk factor measured at different points in the life-course. Thus, it remains to be seen whether measurement of the same risk factor more proximal to mortality would improve statistical prediction.

Dynamic Change Hypothesis

The dynamic change hypothesis posits that changes in risk may themselves be consequential, independent of absolute level. For example, two individuals may exhibit similarly high risk at age 21, yet the person whose risk increased from age 20 to 21 may have a different likelihood of an adverse outcome, than the individual whose risk remained consistently high from age 20 to 21 (Olver et al., 2007). The forensic psychology literature has conceptually distinguished between static (e.g., age of offending onset) and dynamic factors (e.g., substance use), as well as explored whether changes in dynamic factors corresponds to a change in the probability of an adverse outcome (Viljoen et al., 2017b). This distinction parallels the risk-need-responsivity (RNR) model (Andrews et al., 1990), which differentiates between static risk factors (i.e., fixed historical characteristics such as criminal history) and dynamic criminogenic needs (e.g., antisocial attitude, substance use) that are modifiable. While the RNR model focuses on reoffending, conceptually similar distinctions may be useful for thinking about the impact of static and dynamic risk factors on mortality. Tremblay and Pare (2003) had the RNR model in mind when describing how transitioning from prosocial peer relationships to antisocial peer relationships was a risk factor for early mortality.

Although the examination of within-individual change is a fundamental principle of developmental and life-course criminology, this dimension has been neglected in research on early mortality. Chassin et al. (2013) specified a conceptual distinction between static (distal) and dynamic¹ risk factors, but they did not examine whether changes to risk factors impacted the probability of mortality. While static risk factors were measured as a combination of demographic (e.g., race/ethnicity, parental education) and historical factors (e.g., prior offenses, substance use, or mental health disorder diagnosis), dynamic risk factors included measures thought to be modifiable, such as peer antisocial behavior, family cohabitation, and exposure to violence. However, dynamic risk factors were measured at a single time point (also see Teplin et al., 2014).

To reiterate, the shelf-life hypothesis and dynamic change hypothesis are distinct. The shelf-life hypothesis focuses on recency of a measure but does not consider whether the level of a risk factor changed or remained stable. The dynamic change hypothesis focuses on the degree of change between assessments rather than overall level of risk (e.g., moving from a score of “2” to “4” is equivalent to moving from a score of “5” to “7”). A recent measure may be temporally close to the outcome of interest, without reflecting substantial change. Conversely, substantial change may occur even when assessments are separated by a relatively short time interval. Thus, recency and change capture different aspects of risk.

Role of Risk Assessments in Preventing Early Mortality

The risk-need-responsivity model, which guides criminal legal decision making in many jurisdictions, is premised on the notion that changes in risk factors should inform changes in treatment needs and treatment intensity (Andrews et al., 1990; Bonta & Andrews, 2007). Despite this straightforward premise, the risk assessment literature has spent decades grappling with how to make use of multiple assessments of the same risk factor instrument (e.g., use the first score, use the most recent, use the lowest score, use the highest score, use the average; see Babchishin et al., 2024).

If the criminal legal system is to contribute to calls to reduce early mortality rates in its population (Butler et al., 2023), then it is crucial to know how to best identify a person's risk for early mortality. Testing the shelf-life hypothesis and dynamic change hypothesis directly evaluates this question (also see Chassin et al., 2013). Doing so clarifies, for example, whether it is important for risk assessment tools to be updated to account for changes to the timing and level of a risk factor. Correctional system professionals may need to consider the process of risk reduction beyond just the level of a person's risk. This would entail communicating risk in terms of timing, change, whether change occurred as a result of treatment and is expected to continue (Davies et al., 2023). Two people may have the same level of risk but differ in treatment needs if one person has remained at that same level of risk for months, whereas the other person recently shifted from low risk to high risk. On the other hand, if neither hypothesis is supported, then the criminal legal system can avoid intensive and costly procedures of reassessing risk.

Study Aims

The current study contributes to research on the overlap between risk factors for mortality and risk factors for offending by examining whether the level, timing, and change in risk factors relate to mortality. We measure these different dimensions of risk using three complementary models. The baseline model uses the first available risk assessment to measure risk factors at a single time-point, earlier in the follow-up period. The shelf-life model uses the most recent available risk assessment to assess risk factors for mortality at a single time-point, later in the follow-up period. The dynamic change model uses all available risk assessments to capture within-individual change in risk factors over time. Risk factors examined in the current study include substance use, mental health instability, interpersonal support, antisocial attitudes and vocational/academic skills. Although this study is not designed to determine which modeling approach best explains mortality, applying the same set of risk factors across models allows us to test both the shelf-life hypothesis and the dynamic change hypothesis. The baseline model and shelf-life models assess whether the level and timing of risk factor measurement is associated with differential mortality risk. The dynamic change model further assesses whether within-individual change provides explanatory information beyond that obtained from single time-point measures. While the goal of this paper is not to adjudicate between within-person or between-person approaches, it directly evaluates whether incorporating change in risk factors is informative for understanding mortality.

Considerable attention has been given to identifying risk factors for offending that may also be risk factors for early mortality (e.g., Fazel et al., 2014; McCandless et al., 2015; Piquero et al., 2005). Identifying a common cause of offending and mortality allows intervention strategies to simultaneously reduce both offending and mortality. For example, substance use is a risk factor for both offending and early mortality, making substance use treatment programs beneficial for mitigating multiple adverse outcomes (e.g., Decker et al., 2017; Pitkänen et al., 2020). The current study contributes to the search for shared risk factors by examining whether the risk/need factors used in reoffending risk assessment tools also help explain early mortality.

Method

Sample and Procedures

The ISVYOS obtained permission from the Ministry of Children and Family Development to recruit boys and girls (n = 1,719) held in custody centers in British Columbia, Canada, between 1998 and 2011. Among 1,719 youth recruited between 1998 and 2011, 1,640 participated in an intake interview conducted by research assistants (RAs). RAs typically approached youth to participate in the study within the first week of their incarceration. Interviews occurred in a private room away from other youth and custody staff. To obtain informed assent, youth were both read and provided with an information sheet detailing the study's objectives, data collection procedures, the voluntary nature of participation, and assurances that responses were kept confidential except in instances where participants directly threatened themselves or someone else. Participants were also informed that involvement in the study would not impact their status in custody (e.g., security level) nor sentencing outcome. Approximately 5 percent of youth declined to participate.

Data were also collected through the Corrections Network (CORNET), a client management software program where criminal legal system professionals record information about their clients. CORNET includes youth and adult data for all persons involved in the criminal legal system in British Columbia. Excluding participants with sealed CORNET records or missing data resulted in a sample of 1,437, of whom 1,155 had completed a minimum of two risk assessments and were therefore included in the analysis. The analytic sample includes participants born between 1979 and 1997 and followed on average from age 15 to 32 (R = 12–42).

Measures

Descriptive statistics for all variables are presented in Table 1. The mi chained command in Stata (Statacorp, 2021) was used to impute missing values on all measures. Missing data were addressed with multiple imputation, generating 30 imputed datasets. Imputed variables included all risk assessment measures (i.e., substance use, mental health instability, antisocial attitude, academic/vocational skills, interpersonal support²), and proportion of missing data for each imputed variable ranged from five observations (0.04%) to 31 observations (2.68%). Imputations were predicted based on complete demographic information (e.g., age, gender, race), criminal history (e.g., conviction rate, incarceration rate), risk assessment details (e.g., number of total assessments, order of risk assessments), and deceased status. Parameter estimates were then pooled across imputations using Rubin's (1987) rules. Global model fit statistics are not reported because these are not directly pooled across imputed datasets.

Table 1.

Descriptive Statistics for Participants (n = 1,155).

	Mean (%)	SD	Min	Max
Baseline model (First risk assessment only)
Substance use	2.17	0.76	0	3
Antisocial attitude	2.02	0.83	0	3
Mental health instability	2.26	0.69	0	3
Interpersonal support	2.49	1.87	0	9
Academic/vocational skills	0.84	0.83	0	3
Age at first CRNA	17.25	3.27	13	38
Shelf-life model(Last risk assessment only)
Substance use	1.96	0.82	0	3
Antisocial attitude	1.69	0.93	0	3
Mental health instability	1.88	0.76	0	3
Interpersonal support	3.50	1.86	0	9
Academic/vocational skills	1.59	0.71	0	3
Age at last CRNA	25.86	5.62	15	41
Dynamic change model(All risk assessments)^a
Substance use	2.07	0.54	0	3
Antisocial attitude	1.81	0.59	0	3
Mental health instability	2.05	0.52	0	3
Interpersonal support	2.87	1.45	0	9
Academic/vocational skills	1.77	0.55	0	3
Sex
Male	82.9%	-	0	1
Female	17.1%	-	0	1
Ethnicity
White	56.2%	-	0	1
Indigenous	31.1%	-	0	1
Non-Indigenous Minority Group	12.7%	-	0	1
Criminal legal system involvement
Age at first conviction	15.05	2.26	12	29
Total number of CRNAs	5.98	3.62	2	21
Conviction rate (per year)	3.38	1.34	1	10
Incarceration rate (months/year)	3.78	2.28	0	11.20

All measures were averaged within-person.

Mortality. Mortality status and date of death were obtained primarily through CORNET. For participants with active criminal legal system contact (i.e., custodial or community supervision), their criminal histories are maintained by correctional staff on CORNET, which listed date of death and occasionally, cause of death. For participants no longer subject to criminal legal system monitoring, names were further queried through online search engines. Deaths reported in online obituaries and news media were verified by cross-referencing with the participant's name, demographic details (e.g., place of residence, sex), and date of birth. If one of these factors did not match, we did not code the participant as deceased. By the end of the follow-up period (October 31, 2022), 142 participants (12.3%) were deceased. The rate of mortality in the current study is as high or higher when compared to studies that sampled from youth custody facilities (e.g., Chassin et al., 2013; Teplin et al., 2005). The mean age at death was 28.27 (SD = 5.15; R = 19.17–42.52). Due to the inconsistency in reported causes of death (available for 61.3% of deaths), analyses were restricted to mortality status and date of death. However, of the known causes of death (n = 87), overdose was the most common (n = 41; 47.1%), followed by homicide (n = 25; 28.7%).³ These findings align with previous research indicating offending samples exhibit elevated mortality risk due to unnatural causes compared to nonoffending samples (Elonheimo et al., 2017; Skinner & Farrington, 2020; Teplin et al., 2005).

Risk Factors. The Community Risk Needs Assessment (CRNA) is a structured assessment tool used in the case management practices of criminal legal system professionals in British Columbia, based on the central eight risk factors of the RNR model (Bonta & Andrews, 2007). The intention was for the CRNA to be administered every 6 months, typically by probation officers, to ensure criminal legal system professionals remained aware of changes to the risk and needs factors of individuals under supervision (Gress, 2010). CRNA items include indicators of antisocial attitudes and personality traits, crime-involved peers, family and intimate partner issues, education and employment challenges, substance use severity, and mental health instability. These measures can be considered dynamic given their potential to change with each reassessment, which we include in the dynamic change model. For each item, individuals are rated on a four-point scale. Probation officers rely on several sources of information to rate items, including self-report information obtained from interviews with their client, other reports and assessments (e.g., presentence reports), and communication with friends and family. Higher scores indicate greater likelihood of challenges during reintegration.

To measure substance use, we used a single CRNA item that assessed substance dependency as a risk factor for reoffending, with higher values indicating greater substance use. Both alcohol and drug use are major risk factors for mortality (Coffey et al., 2003; van de Weijer et al., 2016). Formerly incarcerated persons face a heightened risk of fatal overdose following release into the community (Patterson, 2013), which may be due to reduced substance use during incarceration and subsequent decreased tolerance (Pratt et al., 2006). Substance use may also exacerbate chronic health conditions (e.g., cardiovascular disease), which tend to be more prevalent among justice-involved populations (Vaughan et al., 2017; Zane et al., 2019).

To measure mental health instability, we used a single CRNA item that assessed behavioral and emotional difficulties, with higher values reflecting more severe mental health issues. Research suggests a strong association between diagnoses of psychiatric disorders and elevated mortality in offending samples (Sailas et al., 2006; Stenbacka et al., 2012), particularly due to suicide (Kullgren et al., 1998; Marks et al., 2024). Furthermore, experiences of incarceration have been shown to exacerbate preexisting mental health conditions (Schnittker et al., 2012; Sugie & Turney, 2017) and in many cases, result in inadequate access to treatment (Reingle Gonzalez & Connell, 2014), thereby compounding the risk of mortality.

To measure antisocial attitudes, we used a single CRNA item that rated an individual's resistance to change and problematic impulses, with higher values denoting stronger antisocial dispositions. Antisocial attitudes may increase mortality risk due to more frequent and persistent offending (Elonheimo et al., 2017; Skinner et al., 2022), which may expose individuals to fatal victimization by engaging in conflict-prone activity (Piquero et al., 2005). Individuals with antisocial attitudes may be inclined toward sensation seeking and risky behaviors (e.g., reckless driving, firearm use), which may in turn increase likelihood of fatal accidents (Tremblay & Pare, 2003).

To measure academic/vocational skills, we used a single reverse-coded CRNA item that rated an individual's degree of academic or vocation skills, with higher values indicating more skills. Skill sufficiency captures an individual's employment potential and readiness. Being able to maintain legitimate employment may embed an individual in prosocial work institutions and discourage risky or criminal behavior, through exposure to potential formal (e.g., management) and informal (e.g., coworker) social control (Hollinger & Clark, 1982; Sampson & Laub, 1990), which may lead to decreased mortality risks. Additionally, an individual who has gained academic and vocational skills may have participated in programs or benefited from reintegration resources, suggesting lower likelihood of reoffending and therefore decreased mortality risk.

To measure interpersonal support, we summed scores from three CRNA items: supportive family relationships, prosocial intimate partners and associations, and living arrangement stability (polychoric ordinal α = .71). We used a summed scale rather than individual items given prior research has emphasized that social bonds and supports tend to be cumulative rather than based on a single item (Nguyen & Loughran, 2018; Savolainen, 2009). CRNA items were reverse-coded so that higher scores indicated stronger interpersonal support. Interpersonal supports could lower mortality risk by providing emotional assistance and instrumental assistance (e.g., financial support, housing), which could buffer stress, facilitate access to health and social services, and foster healthier coping behaviors (Meisenhelder, 1977; Mowen et al., 2019; Taylor, 2016). Additionally, close prosocial relationships may provide informal monitoring and control (Laub & Sampson, 2003), such as discouraging association with antisocial peers (Visher & Travis, 2003), which can further reduce exposure to violent or dangerous situations.

Control Variables. To adjust for past criminal legal system involvement, two indicators captured participants’ cumulative exposure: incarceration rate (number of months incarcerated divided by number of years with any incarceration) and the conviction rate (number of convictions divided by follow-up period).⁴ Standardizing exposure by individual observation time allows for comparison across participants of differing follow-up lengths. Periods in custody can reduce mortality in the short-term by providing stable shelter and legally mandated health care (Daigle & Naud, 2012). In contrast, a higher conviction rate indicates persistent offending, which may heighten risks of lethal conflicts (Binswanger et al., 2020).

To adjust for demographic disparities in mortality, we included self-identified race: White (55.4%), Indigenous (31.4%), and a non-Indigenous minority group (13.2%). Racialized individuals, particularly Indigenous peoples, face elevated mortality relative to White counterparts (Teplin et al., 2014; Zheng et al., 2023). Consistent with literature on incarcerated samples, Indigenous participants were overrepresented in the sample compared with their proportion in the Canadian population (e.g., Wiley et al., 2020). Additionally, we controlled for sex to account for the increased mortality risk that males experience compared to females in legal system-involved populations (Fitch et al., 2024; Kjelsberg & Laake, 2010). The sample was predominantly male (81.6%).

To account for the increased mortality risk that accompanies aging (Aalsma et al., 2016; Piquero et al., 2005), we included age at the time of CRNA completion, as well as a squared term to account for a potential curvilinear relationship (Kariminia et al., 2007). The mean age at first risk assessment was 17.86 (SD = 3.29, R = 13.03–38.56) and at last risk assessment was 26.32 (SD = 5.61, R = 15.14–41.92), with a mean interval between first and last assessments of 8.46 years (SD = 5.81, R = 0.17–24.92). Accounting for aging is critical when evaluating the dynamic perspective because a likely explanation for mixed results with respect to the importance of change in risk factors has to do with whether risk factor change is measured in short intervals in adolescence when change is more common (Viljoen et al., 2017a) or longer intervals in adulthood when stability is more common (Olver et al., 2007).

We also controlled for total number of risk assessments,⁵ given that individuals with greater offending may be more likely to receive repeated CRNAs. Participants were assessed an average of 5.98 times (SD = 3.62; R = 2–21) during the follow-up period, resulting in a total of 6,903 assessments across all participants.

Analytic Strategy

To estimate the relationship between risk factors and mortality, we used the Cox proportional hazard model (Cox, 1972), a semiparametric form of survival analysis designed to examine the time until an event occurs (Allison, 2014). The Cox model estimates the hazard function, which in the current study reflected the instantaneous risk of experiencing mortality at time t, given survival up to that point, as a function of covariates (Kleinbaum & Klein, 2012). The Cox model accommodates both time-invariant covariates (used in the baseline and shelf-life models) and time-varying covariates (used in the dynamic change model). An assumption of the Cox model is proportional hazards, meaning the ratio of hazards remains constant between subjects over the follow-up period (Cox, 1972). The assumption was tested⁶ using a combination of Schoenfeld residuals tests and visual inspections of partial residual plots (Grambsch & Therneau, 1994). Mental health instability and antisocial attitude failed the Schoenfeld residuals test (p < 0.05), but visual inspections indicated no discernable departures from the horizontal line in the residuals (see Figures S1–S3 in the Supplemental Materials).

Survival analysis is well-suited for studying early mortality because it addresses right-censoring situations where some participants have not experienced the event of interest by the end of their individual follow-up period. Survival analysis accounts for this partial information by incorporating participant follow-up times into the model, reflecting the fact that they remain at-risk. This is particularly important in studies of mortality, where participants will experience the event given sufficient time. Figure 1 shows the survival proportion of the sample at each age, without accounting for censoring. The survival proportion begins to decrease at age 19, with a steeper drop between age 23 (n = 35; 97.0%) and age 30 (n = 107; 90.7%). Subsequent declines are more gradual, with survival at 88.8% (n = 129) by age 35, and 87.7% (n = 142) by age 42.

Figure 1.

Survival proportion by age for the sample, not accounting for censoring.

Given the strong relationship between offending and mortality, participants were considered to be at-risk for mortality beginning at the age of first contact with the criminal legal system (i.e., earliest age at conviction or custody). The survival period ends at the age of death or the end of the participant's follow-up period (i.e., censoring). On average, participants’ first CRNA occurred 3.20 years (SD = 3.05; R = 0.03–22.89) into the survival period (i.e., after age at first criminal legal system contact) and 10.91 years (SD = 5.34; R = 1.73–25.09) before death. On average, participants’ last CRNA occurred 3.10 years prior to death (SD = 3.45; R = 0–13.71)⁷ or 7.11 years prior to end of follow-up for the nondeceased (SD = 4.87; R = 0–22.73). The Supplemental Materials provide a detailed explanation of the coding strategy as well as an example data structuring for the baseline, shelf-life, and dynamic change models (see Table S1 in the Supplemental Materials).

For the baseline model and shelf-life model, all risk factors were treated as fixed. These two models differ only in terms of whether risk factors were measured based on the first CRNA rating or last CRNA rating. When testing each of these perspectives, the hazard of death for participant i at time t was modeled as:

h_{i} (t) = h_{0} (t) \times \exp (β_{1} X_{i}_{1} + β_{2} X_{i}_{2} + . . . + β_{k} X_{i k}),

where h_o(t) is the baseline hazard function, and Xi1 represents the value of covariate k for individual i measured at the time of CRNA completion. The Cox model estimates the effects of covariates through partial likelihood, allowing an unspecified baseline hazard. No assumptions are made about the form of the hazard function (Cox, 1972; Kleinbaum & Klein, 2012). The model produces a hazard ratio (HR) which represents the relative risk of death as an exponentiated coefficient (Allison, 2014). The model treats covariates as fixed and assumes that this effect remains constant over the entire follow-up period. For the baseline model, this means that the HR represents the impact of the risk factor measured at the age of the first CRNA (M = 17.86; SD = 3.29, R = 13.03–38.56). For the shelf-life model, the HR represents the impact of the risk factor measured at the last CRNA (M = 26.32; SD = 5.60, R = 15.14–41.92).

For the dynamic change model, the hazard function allows for time-varying covariates:

h_{i} (t) = h_{0} (t) \times \exp (β_{1} X_{i}_{1} (t) + β_{2} X_{i}_{2} (t) + . . . + β_{k} X_{i k} (t)),

where h₀(t) represents the unspecified baseline hazard function, and Xi1(t) denotes the value of covariate k for individual i at time t. Whereas the previous Cox models assumed time-invariant covariates, this time-varying model allowed covariate values to be updated as new assessments emerged throughout the follow-up period. Although Cox regression is recommended for evaluating dynamic change in risk factors (Babchishin et al., 2024), Cox regression with time-varying covariates does not directly examine change; it merely recalculates the impact of a covariate across different measurement periods. Therefore, we calculated change scores for each time-varying covariate. Consistent with first-differencing approaches in fixed-effect regression analyses, a person's change score represented the difference between their raw score at time t and their raw score at time t–1. For example, interpersonal support ranged from 0 to 9. A person whose raw scores at time t and time t–1 were 2 and 5, respectively, would have the same change score as a person whose raw scores at time t and time t–1 were 5 and 8, respectively (i.e., −3 for both persons). This approach allows for the investigation of change while also controlling for time-invariant factors that impact the level of a person's risk (McCuish et al., 2025). Collinearity was assessed using variance inflation factors (VIF) and bivariate correlations (see Tables S2 and S3 in the Supplemental Materials), with no problematic multicollinearity detected across the three models (VIF range: 1.07–3.25, all below conventional threshold of 5.0).

Results

Table 2 presents pooled HRs and 95% confidence intervals from three Cox proportional hazard models examining predictors of mortality across the baseline, shelf-life, and dynamic change models. HRs for key risk factors across each of the three models are summarized in Figure 2.

Figure 2.

CRNA risk factor hazard ratios with 95% confidence intervals for baseline, shelf-life, and dynamic change models.

Table 2.

Cox Model Results (n = 1,155).

	Model 1: Baseline		Model 2: Shelf-Life		Model 3: Dynamic Change^b
	Hazard Ratio	95% Confidence Interval	Hazard Ratio	95% Confidence Interval	Hazard Ratio	95% Confidence Interval
CRNA risk factors
Substance use	1.24	[0.92, 1.67]	1.42**	[1.09, 1.85]	1.15	[0.88, 1.50]
Antisocial attitude	1.01	[0.77, 1.33]	1.21	[0.96, 1.52]	1.27*	[1.04, 1.55]
Mental health instability	1.12	[0.83, 1.52]	0.73*	[0.55, 0.98]	0.83	[0.61, 1.13]
Academic/vocational skills	1.13	[0.87, 1.46]	1.32	[0.97, 1.79]	0.98	[0.76, 1.25]
Interpersonal support	0.98	[0.87, 1.10]	0.98	[0.87, 1.10]	1.01	[0.91, 1.12]
Control variables
Age at CRNA	1.72	[0.88, 3.37]	1.00	[0.75, 1.32]	1.00	[0.63, 1.58]
Age at CRNA²	0.98	[0.96, 1.00]	1.00	[0.99, 1.00]	1.00	[0.99, 1.01]
Total number of CRNAs	0.91**	[0.87, 0.96]	0.98	[0.92, 1.04]	0.84***	[0.79, 0.89]
Female participant^a	-	-	-	-	-	-
Male participant	2.45**	[1.35, 4.44]	2.18*	[1.20, 3.95]	2.12*	[1.18, 3.80]
White^a	-	-	-	-	-	-
Indigenous	1.25	[0.86, 1.81]	1.15	[0.79, 1.68]	1.22	[0.83, 1.78]
Non-Indigenous minority group	1.29	[0.79, 2.12]	1.22	[0.73, 2.03]	1.35	[0.83, 2.20]
Incarceration rate (months/year)	0.97	[0.90, 1.04]	1.00	[0.93, 1.08]	0.97	[0.92, 1.03]
Conviction rate (per year)	1.00	[0.92, 1.09]	1.17*	[1.03, 1.33]	1.18***	[1.08, 1.28]
Model fit (Uno's C)	0.68 (95% CI [0.62, 0.74])		0.79 (95% CI [0.73, 0.84])
	Difference = 0.11, p < 0.01

Note: Conviction rate is standardized by length of follow-up period.

Reference category.

All CRNA risk factors are first-differenced (subtract CRNA value at t–1 from the value at t).

* p < 0.05, ** p < 0.01, *** p < 0.001.

The baseline model (Model 1) used only the earliest CRNA measurement. None of the CRNA-measured risk factors significantly impacted the likelihood of mortality. However, male participants exhibited significantly elevated mortality risk compared to female participants. Each additional risk assessment was associated with a 9% reduction in the hazard of death.

The shelf-life model (Model 2) used the most recent CRNA available.⁸ Each one-unit increase in substance use score was associated with a 42% increase in the hazard of mortality. Mental health instability was unexpectedly negatively related to mortality, with a one-unit increase associated with a 27% reduction in the hazard of death. As was the case in Model 1, male participants exhibited a higher risk of mortality. Each additional conviction per year observed was associated with a 17% increase in mortality risk.

To compare the performance of the baseline and shelf-life models, predictive discrimination was assessed using Uno's C-statistic, which represents a model's ability to correctly distinguish between participants who will experience the outcome (e.g., 0.5 refers to discriminative ability equal to a coin flip; 1.0 indicates perfect discrimination). Uno's C uses inverse probability of censoring weights to correct for unequal follow-up periods across participants (Uno et al., 2011). Uno's C indices were pooled across 30 multiple imputations (Rubin 1987). Uno's C requires the specification of truncation time (tau), beyond which comparisons between pairs are not made. This is necessary because as follow-up extends, fewer participants remain under observation, making pairwise comparisons increasingly unreliable. Tau was set to age 38, by which point 96% of all observed deaths in the sample had occurred, ensuring that the discrimination statistic reflects nearly the full mortality experience of the sample while remaining statistically stable. At age 38, the shelf-life model (C = 0.79; 95% CI [0.73, 0.84]) demonstrated significantly (p < 0.01) greater discriminative ability than the baseline model (C = 0.68; 95% CI [0.62, 0.74]). An increase of 0.11 indicates that the shelf-life model correctly stratifies risk for an additional 11% of all possible subject pairs compared to the baseline model. Sensitivity analyses at ages 28 and 30 produced consistent findings.⁹

The dynamic change model (Model 3) accounted for repeated assessments of dynamic factors by using first-differencing to capture wave-over-wave change in time-varying covariates. Antisocial attitude was the only significant CRNA risk factor. A one-unit increase in antisocial attitudes between time t–1 and time t was associated with a 27% increase in mortality risk between time t and time t + 1. Male participants and conviction rate remained significantly related to increased mortality risk, with each one-unit wave-over-wave increase in convictions subsequently increasing the wave-over-wave mortality hazard by 18%. Each additional risk assessment was associated with a 16% reduction in the hazard of death.

Discussion

Early mortality among individuals involved in the criminal legal system represents a persistent public health concern (Fischer et al., 2019; Jannetto, 2021). In response, researchers have explored the overlap between offending and early mortality by identifying shared risk factors. Although relying on criminological theory and the risk factor paradigm has contributed to identifying risk factors statistically associated with early mortality (Kullgren et al., 1998; Lidberg et al., 1989; Zheng et al., 2023), statistical prediction is not the same as explanation (Wikström, 2020). Using data on 1,155 incarcerated youth who were followed into adulthood, the goal of the current study was to move towards a better understanding of the association between risk factors and mortality. We addressed this goal by examining the impact of a given risk factor on mortality through the lens of baseline, shelf-life, and dynamic models. The baseline model proposes that the level of a risk factor determines the strength of its association with mortality, irrespective of the timing of measurement. The shelf-life model holds that the timing of the risk factor measurement relative to the mortality event is what matters most, not the level itself. The dynamic change model emphasizes the importance of directly examining whether change impacts the outcome of interest. Findings are interpreted with a focus on (a) the implications for the shelf-life hypothesis and the dynamic change hypothesis, and (b) the implications of risk factors identified as significantly associated with early mortality.

Evaluating Temporal Dynamics of Risk for Early Mortality

The current study found that risk factors measured at baseline, temporally distant from the mortality event, were generally not informative. However, when these same risk factors were measured closer to the timing of mortality (i.e., average 3.04 years from the event instead of 11.01 years from the event), several emerged as significantly associated with an increased risk of mortality. A comparison of the baseline model and shelf-life models using Harrell's C-indices also indicates an increase in performance of the shelf-life model.

Specifically, higher levels of substance use increased mortality risk (the unanticipated protective effect of mental health instability is discussed in the next section). The emergence of substance use as a significant predictor only when measured closer in time to the mortality event supports the shelf-life hypothesis (Vincent & Grisso, 2005). This hypothesis posits that risk factors lose their predictive validity over time. Therefore, it is important to update risk assessments as individuals may fall out of risk when earlier issues resolve or become less relevant, rendering historical profiles that are less informative.

Cox regression with time-varying covariates was used to evaluate the impact of dynamic change in risk factors on early mortality. This approach differed from the shelf-life model by accounting for all reassessments of dynamic factors as opposed to only the assessment in closest temporal proximity to the outcome of interest. Using this approach, higher levels of antisocial attitudes emerged as significantly associated with increased mortality risk. The emergence of antisocial attitudes as a theoretically consistent risk factor for mortality in the dynamic change model but not the shelf-life model implies that the relevance of a certain risk factor to early mortality may only be detected if examined through the appropriate lens. The dynamic change model, which uses repeated measurements, is suited especially for instances where acute risk factors (i.e., those that may change over shorter windows of time; Etzler & Rettenberger, 2019) increase the risk of early mortality. The measurement of antisocial attitudes in the current study may reflect the type of hostility considered to be an acute factor (see Hanson & Harris, 2007) that is subject to quicker changes (e.g., following incarceration, connection to new peer group, response to intervention) that need to be captured from a dynamic perspective.

Risk Factors Associated with Early Mortality

Substance use has been consistently recognized as a major risk factor for early mortality (Binswanger et al., 2020; Chen et al., 2010; Kinner et al., 2015; Kullgren et al., 1998; Teplin et al., 2014). Looking more closely at this relationship, we found that it might be too simplistic to claim that lifetime substance use matters for early mortality. Instead, public health policies aiming to disrupt the link between substance use and mortality should consider those with recent substance use issues as especially high-risk. These findings are particularly salient given the ongoing opioid epidemic in North America, which has contributed to substantial increases in overdose deaths and disproportionately affects criminal legal system-involved populations (Binswanger et al., 2007; Kariminia et al., 2007; Merrall et al., 2010).

The protective effect of mental health instability on mortality contrasts with existing research, which has typically found that individuals with psychiatric diagnoses are at elevated mortality risk (Sailas et al., 2006; Stenbacka et al., 2012). One explanation for this unanticipated finding is the visibility of higher needs in individuals with these risk factors. Individuals who display emotional dysregulation are more likely to trigger clinical concern, receive psychiatric referrals, or be placed under intensive supervision (Swartz et al., 1998), all of which may decrease mortality risk. Unfortunately, the current study lacked data on the level of resources each participant received from the criminal legal system.

Antisocial attitude emerged as a significant risk factor only within the dynamic perspective. Antisocial attitudes have been shown to be predictors for reoffending (Andrews et al., 1990; Mills et al., 2004), which in turn are linked to greater risk of violent deaths (Kariminia et al., 2007; Lattimore et al., 1997; Paanila et al., 1999). Antisocial attitudes may be context-dependent, such as being exacerbated by greater exposure to others with antisocial attitudes during incarceration, or from being frustrated by reintegration challenges (e.g., obtaining housing or employment). Conversely, new intimate relationships or obtaining legitimate employment may encourage more prosocial attitudes (Laub & Sampson, 2001; Maruna, 2001).

Consistent with existing research, conviction rate was also positively associated with mortality risk, further supporting a robust offending–mortality relationship evident in literature (Lidberg et al., 1989; Pritchard et al., 1997; Skinner & Farrington, 2020; Teplin et al., 2014). Conversely, the total number of risk assessments was associated with a reduced risk of mortality. Assessment frequency is expected to be positively related with offending, because risk assessments are routinely performed by probation officers as part of case planning and management, and therefore continued offending should result in more assessments. However, number of risk assessments and conviction counts were only moderately correlated (r = 0.56), suggesting that assessment frequency is not merely a proxy for offending. Instead, this inverse association may reflect other aspects of heightened system contact, which could attenuate mortality risk, such as closer supervision and monitoring, or more diligent and proactive probation officers. Increased contact likely facilitates higher levels of service engagement and identification of unmet needs, creating opportunities for target interventions (e.g., referrals to detox programs, housing assistance, medical care) that ultimately mitigate mortality risk.

Implications for Criminal Legal System Practices

Although the CRNA was not designed to identify a person's risk for mortality, its associations with offending in other studies (e.g., McCuish & Lussier, 2023) and mortality in this study reinforce the long-recognized overlap between offending and mortality. Results from the shelf-life and dynamic change models point to the importance of reassessment to maintain an accurate understanding of an individual's current risk profile and to inform relevant interventions. Acute dynamic risk factors are liable to change over short intervals. Without reassessment, outdated evaluations may obscure important needs.

Beyond simply capturing a new level of risk, reassessments should track whether a risk factor has increased or decreased relative to previous assessments. Findings from the first-differenced model suggest that it is not only the level of a risk factor but also its change over time (e.g., worsening antisocial attitudes) that indicate an increased risk of mortality. Integrating change-sensitive assessments into correctional practice would allow practitioners to intervene earlier, potentially preventing further deterioration from moderate to high risk, and reducing the likelihood of adverse outcomes.

Given the strong link between substance use and mortality (Kinner et al., 2020; Lidberg et al., 1989; Stenbacka et al., 2012), evidence-based substance use interventions should be prioritized within criminal legal systems. Because substance use is a risk factor for both offending and early mortality (Zane et al., 2019), targeted treatment programs can simultaneously address both public health and public safety concerns. Incorporating substance use treatment into community re-entry planning, probation supervision, and institutional programming could improve long-term outcomes and interrupt cycles of addiction, offending, and health deterioration.

Although incarcerated youth represent a relatively small fraction of the overall youth population, they are among the most vulnerable to early mortality, which warrants focused research and policy attention. Prior research has shown that criminal legal system-involved youth face significantly higher rates of mortality compared to their non-system-involved peers, particularly due to homicides and suicides (Aalsma et al., 2016; Lindberg et al., 2017; Zheng et al., 2023). These elevated risks persist into adulthood (Skardhamar & Skirbekk, 2013; Teplin et al., 2014), suggesting long-term consequences of early criminal legal system involvement. Avoiding research on this population risks overlooking a group for whom well-timed interventions could reduce preventable deaths and mitigate long-term harms. While no single agency is currently mandated to monitor or reduce mortality among criminal legal system-involved individuals, correctional systems regularly administer risk assessment instruments that identify the very factors shown here to predict early death. Correctional agencies internationally have implemented the risk-need-responsivity model, which targets dynamic criminogenic needs, such as substance use and antisocial attitudes, through cognitive-behavioral interventions (Andrews & Bonta, 2015; Andrews et al., 1990). Agencies that target criminogenic needs in the interest of reducing reoffending are therefore, by extension, already reducing mortality risk whether or not that is an explicit goal. Making mortality reduction an explicit correctional goal would not require systemic changes, but simply a recognition of what existing practice already has the capacity to achieve. For some youth agencies (e.g., Ministry of Children and Family Development in British Columbia), the goals of child well-being and youth corrections already coexist within a single mandate, demonstrating that public safety and health outcomes can be pursued together within the same organization.

Limitations and Future Research

As CRNAs are administered only during periods of supervision, individuals who desist from offending or otherwise refrain from having criminal legal system contact have fewer assessments. As a result, both the shelf-life model, which used the last available risk assessment, and the dynamic change model, which incorporated all available risk assessments, are more sensitive to risk patterns among individuals with prolonged system involvement. Therefore, mortality risk may be underestimated for individuals with shorter or less visible supervision histories, even if they continue to experience significant adversity. On average, participants’ last CRNA occurred 3.10 years prior to death or 7.11 years prior to the end of follow-up (for nondeceased), indicating that this limitation may have introduced greater bias for participants who did not die during the observation period.

The finding that the last assessment outperforms the first may not solely indicate that the person has changed over time, but may also reflect what could be termed a familiarity hypothesis: that repeated contact between a probation officer and the same individual produces more accurate assessments over time, as the officer develops a deeper understanding of the individual's background and circumstances. Therefore, improvement in predictive accuracy may reflect greater assessment reliability. These explanations are not mutually exclusive, but they carry different theoretical implications: one pointing to reassessment as a mechanism for capturing genuine risk change, and the other highlighting the role of assessor familiarity in producing accurate risk assessments. Future research could disentangle these mechanisms by examining whether predictive accuracy improves more rapidly when the same probation officer conducts repeated assessments compared to when different officers assess the same individual.

This imbalance in the CRNA data connects to a broader conceptual concern about the interpretation of individual-level risk factors. Items indexed in the CRNA reflected individual-level risk factors. However, these factors may instead serve as proxies for structural disadvantage, including systemic poverty, residential segregation, inadequate access to healthcare, and cumulative exposure to institutional neglect. The current study did not account for the broader social environment (e.g., social networks, the schools they attend, where they live, quality and access to healthcare in their jurisdiction). Accounting for such factors may change the interpretation of the contribution of individual-level risk factors to early mortality. Future research should address this limitation by integrating contextual variables, such as through multilevel models and data linkage projects. Also missing from the CRNA was direct measures of self-control and other individual-level risk factors more appropriate for testing questions about the impact of antisocial propensity on early mortality (Tremblay & Pare, 2003).

In the current study, mental health instability unexpectedly decreased the risk of early mortality, which is inconsistent with the majority of literature (Kullgren et al., 1998; Sailas et al., 2006; Siponen et al., 2023). This discrepancy may reflect differences in how mental health was measured. In this study, mental health was measured through a risk assessment item, while other studies typically use formal psychiatric diagnoses (e.g., Kullgren et al., 1998; Stenbacka et al., 2012). Measuring mental health in different ways could result in different correctional system responses to mental health issues. For instance, individuals with diagnosed psychiatric disorders are more often placed into solitary confinement (Clark, 2018; Stewart & Wilton, 2014), which may in turn exacerbate mortality risk (Wildeman & Andersen, 2020). Meanwhile, changes in mental health deterioration as indicated by repeated risk assessments may trigger responses and interventions that reduce mortality risk. Furthermore, formal diagnoses are made by psychiatrists, while probation officers complete CRNAs. Both groups differ in training and experiences with individuals who struggle with mental health, which may result in different perceptions of what mental health instability looks like. To explore the potential impact of correctional system staff on mortality, a next step is to analyze whether probation officers’ discretionary responses to perceived instability acts as a protective intervention. For example, future research could investigate if higher CRNA scores trigger escalation in supervision, such as increased home visits or program referrals.

Conclusion

This study followed a cohort of 1,155 formerly incarcerated Canadian youth into adulthood to examine the relationship between risk factors and early mortality. The same risk factor can be modeled in different ways, and the current study explored the ways in which the likelihood of mortality was impacted by a person's level of risk at baseline assessment, their level of risk closest to the outcome of interest (shelf-life hypothesis; Vincent & Grisso, 2005), and their change in level of risk. The results were most supportive of the shelf-life and dynamic change models. Consistent with the “shelf-life” hypothesis (Vincent & Grisso, 2005), earlier baseline measurements of risk factors did not predict mortality, while those assessed closer to the time of death were significantly associated with increased mortality risk. Specifically, greater substance use increased mortality risk, while mental health instability unexpectedly decreased mortality risk. The dynamic change model, which used time-varying covariates to evaluate the effects of changes in risk factors over time with first-differenced scores, found that within-individual increases in antisocial attitudes were associated with elevated mortality risk. These findings highlight the importance of repeated assessment and the value of modeling change over time when examining mortality among legal system-involved populations.

Supplemental Material

sj-docx-1-jrc-10.1177_00224278261459468 - Supplemental material for Beyond Risk Level: Timing and Change in Risk Factors as Predictors of Early Mortality Among Formerly Incarcerated Youth

Supplemental material, sj-docx-1-jrc-10.1177_00224278261459468 for Beyond Risk Level: Timing and Change in Risk Factors as Predictors of Early Mortality Among Formerly Incarcerated Youth by Jen-Li Shen and Evan C. McCuish in Journal of Research in Crime and Delinquency

Footnotes

ORCID iDs

Jen-Li Shen

Evan C. McCuish

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Social Sciences and Humanities Research Council of Canada (grant number 435-2020-0150).

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biographies

Jen-Li Shen is a PhD Candidate in Criminology. Her areas of research are environmental criminology and life-course criminology.

Evan C. McCuish is an Associate Professor in Criminology. His area of research is developmental and life-course criminology with a focus on psychopathy.

References

Aalsma

Matthew C.

Katherine

S. L. Lau

Perkins

Anthony J.

Schwartz

Katherine

Wanzhu

Wiehe

Sarah E.

Monahan

Patrick

Rosenman

Marc B.

. 2016. “Mortality of Youth Offenders Along a Continuum of Justice System Involvement.” American Journal of Preventive Medicine 50(3):303–310. doi:10.1016/j.amepre.2015.08.030

Allison

Paul.

2014. Event History and Survival Analysis. California, US: Sage Publications, Inc.

Andrews

D. A.

Bonta

James

. 2015. The Psychology of Criminal Conduct. 5th ed. New York: Routledge.

Andrews

D. A.

Bonta

James

Hoge

R. D.

. 1990. “Classification for Effective Rehabilitation: Rediscovering Psychology.” Criminal Justice and Behavior 17(1):19–52. doi:https://doi.org/10.1177/0093854890017001004

Babchishin

Kelly M.

Hanson

R. Karl

Lee

Seung C.

. 2024. “Risk to Reoffend Changes Over Time: Improving Correctional Programming Through Progress Monitoring.” Psychological Assessment 36(10):595–605. doi:10.1037/pas0001335

Binswanger

Ingrid A.

Marc

F. Stern

Deyo

Richard A.

Heagerty

Patrick J.

Cheadle

Allen

Elmore

Joann G.

Koepsell

Thomas D.

. 2007. “Release from Prison — A High Risk of Death for Former Inmates.” New England Journal of Medicine 356(2):157–65. doi:10.1056/NEJMsa064115

Binswanger

Ingrid A.

Nguyen

Anh P.

Morenoff

Jeffrey D.

Stanley

Harding

David J.

. 2020. “The Association of Criminal Justice Supervision Setting with Overdose Mortality: A Longitudinal Cohort Study.” Addiction 115(12):2329–38. doi:10.1111/add.15077

Bonta

James

Andrews

Donald A.

. 2007. Risk-Need-Responsivity Model for Offender Assessment and Rehabilitation. Ottawa: Public Safety Canada.

Butler

Amanda

Croxford

Ruth

Bodkin

Claire

Akbari

Hanaya

Bayoumi

Ahmed M.

Bondy

Susan J.

Guenter

Dale

McLeod

Katherine E.

Gomes

Tara

Kanagalingam

Tharsan

Kiefer

Lori A.

Orkin

Aaron Michael

Owusu-Bempah

Akwasi

Regenstreif

Leonora

Kouyoumdjian

Fiona

. 2023. Burden of Opioid Toxicity Death in the Fentanyl-Dominant Era for People Who Experience Incarceration in Ontario 13(5):E071867, Canada, 2015–2020: A Whole Population Retrospective Cohort Study 13:e071867. doi:10.1136/bmjopen-2023-071867

10.

Chassin

Laurie

Piquero

Alex R.

Losoya

Sandra H.

Mansion

Andre D.

Schubert

Carol A.

. 2013. “Joint Consideration of Distal and Proximal Predictors of Premature Mortality Among Serious Juvenile Offenders.” Journal of Adolescent Health 52(6):689–96. doi:10.1016/j.jadohealth.2012.11.018

11.

Chen

Chuan-Yu

Pei-Ning

Lien-Wen

Chou

Yiing-Jenq

Lin

Keh-Ming

. 2010. “Three-Year Mortality and Predictors After Release: A Longitudinal Study of the First-Time Drug Offenders in Taiwan.” Addiction 105(5):920–27. doi:10.1111/j.1360-0443.2009.02894.x

12.

Chung

Ick-Joong

Hill

Karl G.

Hawkins

J. David

Gilchrist

Lewayne D.

Nagin

Daniel S.

. 2002. “Childhood Predictors of Offense Trajectories.” Journal of Research in Crime and Delinquency 39(1):60–90. doi:10.1177/002242780203900103

13.

Clark

Kyleigh.

2018. “The Effect of Mental Illness on Segregation Following Institutional Misconduct.” Criminal Justice and Behavior 45(9):1363–82. doi:10.1177/0093854818766974

14.

Coffey

Carolyn

Veit

Friederike

Wolfe

Rory

Cini

Eileen

Patton

George C.

. 2003. “Mortality in Young Offenders: Retrospective Cohort Study.” BMJ 326:4.

15.

Cox

D. R.

1972. “Regression Models and Life-Tables.” Journal of the Royal Statistical Society Series B: Statistical Methodology 34(2):187–202. doi:10.1111/j.2517-6161.1972.tb00899.x

16.

Daigle

Marc S.

Naud

Hélène

. 2012. “Risk of Dying by Suicide Inside or Outside Prison: The Shortened Lives of Male Offenders.” Canadian Journal of Criminology and Criminal Justice 54(4):511–28. doi:10.3138/cjccj.2011.E.54

17.

Davies

Simon T.

Lloyd

Caleb D.

Polaschek

Devon L. L.

. 2022. “Does Reassessment Enhance the Prediction of Imminent Criminal Recidivism? Replicating Lloyd et al. (2020) With High-Risk Parolees.” Assessment 29(5):962–80. doi:10.1177/1073191121993216

18.

Davies

Simon T.

Lloyd

Caleb D.

Polaschek

Devon L. L.

. 2023. “Does Risk Change Simply Indicate Current Risk for Recidivism or Something More? A Review of the Predictive Value of Intraindividual Change.” Criminal Justice and Behavior 50(8):1102–19. doi:10.1177/00938548231174905

19.

Decker

Kathleen P.

Peglow

Stephanie L.

Samples

Carl R.

Cunningham

Tina D

. 2017. “Long-Term Outcomes After Residential Substance Use Treatment: Relapse, Morbidity, and Mortality.” Military Medicine 182(1):e1589–e1595. doi:10.7205/MILMED-D-15-00560

20.

Elonheimo

Henrik

Sillanmäki

Lauri

Sourander

André

. 2017. “Crime and Mortality in a Population-Based Nationwide 1981 Birth Cohort: Results from the FinnCrime Study: Crime and Mortality.” Criminal Behaviour and Mental Health 27(1):15–26. doi:10.1002/cbm.1973

21.

Etzler

Sonja

Rettenberger

Martin

. 2019 “Stable-2007 and Acute-2007.” Pp. 1434–37 In The SAGE Encyclopedia of Criminal Psychology, edited by Morgan

R. D.

. Los Angeles, London, New Delhi, Singapore, Washington DC, Melbourne: Sage.

22.

Farrington

D. P.

2000. “Explaining and Preventing Crime: The Globalization of Knowledge—the American Society of Criminology 1999 Presidential Address.” Criminology; An Interdisciplinary Journal 38(1):1–24. doi:10.1111/j.1745-9125.2000.tb00881.x

23.

Fazel

Seena

Wolf

Achim

Palm

Camilla

Lichtenstein

Paul

. 2014. “Violent Crime, Suicide, and Premature Mortality in Patients with Schizophrenia and Related Disorders: A 38-Year Total Population Study in Sweden.” The Lancet Psychiatry 1(1):44–54. doi:10.1016/S2215-0366(14)70223-8

24.

Fischer

Benedikt

Pang

Michelle

Tyndall

Mark

. 2019. “The Opioid Death Crisis in Canada: Crucial Lessons for Public Health.” The Lancet Public Health 4(2):e81–82. doi:10.1016/S2468-2667(18)30232-9

25.

Fitch

Kate Vinita

Pence

Brian W.

Rosen

David L.

Miller

Vanessa E.

Gaynes

Bradley N.

Swilley-Martinez

Monica E.

Kavee

Andrew L.

Carey

Timothy S.

Proescholdbell

Scott K.

Ranapurwala

Shabbar I.

. 2024. “Suicide Mortality Among Formerly Incarcerated People Compared With the General Population in North Carolina, 2000–2020.” American Journal of Epidemiology 193(3):489–99. doi:10.1093/aje/kwad214

26.

Gonzalez

Reingle

Jennifer

Connell

Nadine M.

. 2014. “Mental Health of Prisoners: Identifying Barriers to Mental Health Treatment and Medication Continuity.” American Journal of Public Health 104(12):2328–33. doi:10.2105/AJPH.2014.302043

27.

Gottfredson

Michael R.

Hirschi

Travis

. 1990. A General Theory of Crime. Stanford, CA: Stanford University Press.

28.

Grambsch

Patricia M.

Therneau

Terry M.

. 1994. “Proportional Hazards Tests and Diagnostics Based on Weighted Residuals.” Biometrika 81(3):515–26. doi:10.1093/biomet/81.3.515

29.

Gress

Carmen.

2010. Revealing Research & Evaluation (RREv). Community Corrections and Corporate Programs Division, May.

30.

Hanson

R. Karl

Harris

Andrew J. R.

. 2007. Acute-2007 Scoring Guide. Public Safety Canada.

31.

Hollinger

Richard C.

Clark

John P.

. 1982. “Formal and Informal Social Controls of Employee Deviance.” The Sociological Quarterly 23(3):333–43. doi:10.1111/j.1533-8525.1982.tb01016.x

32.

Jannetto

Paul J.

2021. “The North American Opioid Epidemic.” Therapeutic Drug Monitoring 43(1):1. doi:10.1097/FTD.0000000000000817

33.

Kariminia

Azar

Law

Matthew G

Butler

Tony G.

Corben

Simon P.

Levy

Michael H.

Kaldor

John M.

Grant

Luke

. 2007. “Factors Associated with Mortality in a Cohort of Australian Prisoners.” European Journal of Epidemiology 22(7):417–28. doi:10.1007/s10654-007-9134-1

34.

Kinner

Stuart A.

Degenhardt

Louisa

Coffey

Carolyn

Hearps

Stephen

Spittal

Matthew

Sawyer

Susan M.

Patton

George C.

. 2015. “Substance Use and Risk of Death in Young Offenders: A Prospective Data Linkage Study.” Drug and Alcohol Review 34(1):46–50. doi:10.1111/dar.12179

35.

Kinner

Stuart A.

Gan

Wenqi

Slaunwhite

Amanda

. 2020. “High Rate of Fatal Overdose After Release from Prison In BC, Canada: A Data Linkage Study.” International Journal of Population Data Science 5(5). doi:https://doi.org/10.23889/ijpds.v5i5.1606

36.

Kjelsberg

Ellen

Laake

Petter

. 2010. “Is the High Mortality Risk in Sentenced Offenders Independent of Previous Imprisonment?” European Journal of Epidemiology 25(4):237–43. doi:10.1007/s10654-010-9436-6

37.

Kleinbaum

David G.

Klein

Mitchel

. 2012. Survival Analysis: A Self-Learning Text. Statistics for Biology and Health. 3rd ed. New York: Springer.

38.

Kullgren

Tengström

Grann

. 1998. “Suicide among Personality-Disordered Offenders: A Follow-up Study of 1943 Male Criminal Offenders.” Social Psychiatry and Psychiatric Epidemiology 33(0):S102–6. doi:10.1007/s001270050217

39.

Lattimore

Pamela

Linster

Richard

MacDonald

John

. 1997. “Risk of Death among Serious Young Offenders.” Journal of Research in Crime and Delinquency 34(2):187–209. doi:10.1177/0022427897034002002

40.

Laub

John H.

Sampson

Robert J.

. 2001. “Understanding Desistance from Crime.” Crime and Justice: A Review of Research 28:1–70.

41.

Laub

John H.

Sampson

Robert J.

. 2003. Shared Beginnings, Divergent Lives: Delinquent Boys to Age 70. Cambridge, MA: Harvard University Press.

42.

Laub

John H.

Vaillant

George E.

. 2000. “Delinquency and Mortality: A 50-Year Follow-Up Study of 1,000 Delinquent and Nondelinquent Boys.” American Journal of Psychiatry 157(1):96–102. doi:10.1176/ajp.157.1.96

43.

Le Blanc

Marc

Loeber

Rolf

. 1998. “Developmental Criminology Updated.” Crime and Justice 23:115–98.

44.

Lidberg

Lars

Wiklund

Nils

Jakobsson

Sten W.

. 1989. “Mortality among Criminals with Suspected Mental Disturbance.” Scandinavian Journal of Social Medicine 17(1):59–65. doi:10.1177/140349488901700110

45.

Lindberg

Nina

Miettunen

Jouko

Heiskala

Anni

Kaltiala-Heino

Riittakerttu

. 2017. “Mortality of Young Offenders: A National Register-Based Follow-up Study of 15- to 19-Year-Old Finnish Delinquents Referred for Forensic Psychiatric Examination Between 1980 and 2010.” Child and Adolescent Psychiatry and Mental Health 11(1):37. doi:10.1186/s13034-017-0174-3

46.

Lloyd

Caleb D.

Karl Hanson

Richards

Dylan K.

Serin

Ralph C.

. 2020. “Reassessment Improves Prediction of Criminal Recidivism: A Prospective Study of 3,421 Individuals in New Zealand.” Psychological Assessment 32(6):568–81. doi:10.1037/pas0000813

47.

Lösel

Friedrich

Bender

Doris

. 2003 “Protective Factors and Resilience.” Pp. 130–204 in Early Prevention of Adult Antisocial Behaviour, edited by Farrington

D. P.

Coid

Jeremy W.

. Cambridge: Cambridge University Press.

48.

Marks

Rocky B.

Moreira

Nicole

O’Connell

Katherine L.

Hearne

Alera

Law

Keyne C.

. 2024. “Suicide While Locked Up in Texas: Risk Factors for Death by Suicide in Custody”. Journal of Interpersonal Violence 39(23-24):4896–4923. doi:10.1177/08862605241243366

49.

Maruna

Shadd.

2001. Making Good: How Ex-Convicts Reform and Rebuild Their Lives. 1st ed. Washington, D.C: American Psychological Association.

50.

McCandless

Lawrence C.

Lauren

C. Stewart

Rempel

Emily S.

Venners

Scott A.

Somers

Julian M.

. 2015. “Criminal Justice System Contact and Mortality among Offenders with Mental Illness in British Columbia: An Assessment of Mediation.” Journal of Epidemiology and Community Health 69(5):460–66. doi:10.1136/jech-2013-203705

51.

McCuish

Evan

Lussier

Patrick

. 2023. “Twenty Years in the Making: Revisiting Laub and Sampson’s Version of Life-Course Criminology.” Journal of Criminal Justice 88:102117. doi:10.1016/j.jcrimjus.2023.102117

52.

McCuish

Evan C.

Bushway

Shawn

Lussier

Patrick

Gushue

Kelsey

. 2025. “The Impact of Incarceration on Reoffending: A Period-to-Period Analysis of Canadian Youth Followed into Adulthood.” Journal of Criminal Justice 96:102335. doi:10.1016/j.jcrimjus.2024.102335

53.

Meisenhelder

Thomas.

1977. “An Exploratory Study of Exiting from Criminal Careers.” Criminology; An Interdisciplinary Journal 15(3):319–34. doi:10.1111/j.1745-9125.1977.tb00069.x

54.

Merrall

Elizabeth L. C.

Azar Kariminia

Ingrid A. Binswanger

Hobbs

Michael S.

Farrell

Michael

Marsden

John

Hutchinson

Sharon J.

Bird

Sheila M.

. 2010. “Meta-Analysis of Drug-Related Deaths Soon After Release from Prison.” Addiction 105(9):1545–54. doi:10.1111/j.1360-0443.2010.02990.x

55.

Mills

Jeremy F.

Daryl

G. Kroner

Hemmati

Toni

. 2004. “The Measures of Criminal Attitudes and Associates (MCAA): The Prediction of General and Violent Recidivism.” Criminal Justice and Behavior 31(6):717–33. doi:10.1177/0093854804268755

56.

Mowen

Thomas J.

Stansfield

Richard

Boman

John H.

. 2019. “Family Matters: Moving Beyond ‘If’ Family Support Matters to ‘Why’ Family Support Matters During Reentry from Prison.” Journal of Research in Crime and Delinquency 56(4):483–523. doi:10.1177/0022427818820902

57.

Nagin

Daniel

Paternoster

Raymond

. 2000. Population Heterogeneity and State Dependence: State of the Evidence and Directions for Future Research. 28.

58.

Nguyen

Loughran

T. A.

. 2018. “On the Measurement and Identification of Turning Points in Criminology.” Annual Review of Criminology 1(1):335–58. 10.1146/annurev-criminol-032317-091949

59.

Olver

Mark E.

Stephen

C. P.

Wong

Terry

Nicholaichuk

Audrey Gordon.

. 2007. “The Validity and Reliability of the Violence Risk Scale-Sexual Offender Version: Assessing Sex Offender Risk and Evaluating Therapeutic Change.” Psychological Assessment 19(3):318–29. doi:10.1037/1040-3590.19.3.318

60.

Paanila

Jarmo

Hakola

Panu

Tiihonen

Jari

. 1999. “Mortality among Habitually Violent Offenders.” Forensic Science International 100(3):187–91. doi:10.1016/S0379-0738(98)00209-6

61.

Patterson

Evelyn J.

2013. “The Dose–Response of Time Served in Prison on Mortality: New York State, 1989–2003.” American Journal of Public Health 103(3):523–28. doi:10.2105/AJPH.2012.301148

62.

Piquero

Alex R.

David

Farrington

Jonathan P.

Shepherd

Katherine Auty.

. 2014. “Offending and Early Death in the Cambridge Study in Delinquent Development.” Justice Quarterly 31(3):445–72. doi:10.1080/07418825.2011.641027

63.

Piquero

Alex R.

MacDonald

John

Dobrin

Adam

Daigle

Leah E.

Cullen

Francis T.

. 2005. “Self-Control, Violent Offending, and Homicide Victimization: Assessing the General Theory of Crime.” Journal of Quantitative Criminology 21(1):55–71. doi:10.1007/s10940-004-1787-2

64.

Pitkänen

Tuuli

Kaskela

Teemu

Levola

Jonna

. 2020. “Mortality of Treatment-Seeking Men and Women with Alcohol, Opioid or Other Substance Use Disorders – A Register-Based Follow-up Study.” Addictive Behaviors 105:106330. doi:10.1016/j.addbeh.2020.106330

65.

Pratt

Daniel

Piper

Mary

Appleby

Louis

Webb

Roger

Shaw

Jenny

. 2006. “Suicide in Recently Released Prisoners: A Population-Based Cohort Study.” The Lancet 368(9530):119–23. doi:10.1016/S0140-6736(06)69002-8

66.

Pritchard

Colin

Cox

Malcolm

Dawson

Adrian

. 1997. “Suicide and ‘Violent’ Death in a Six-Year Cohort of Male Probationers Compared with Pattern of Mortality in the General Population: Evidence of Accumulative Socio-Psychiatric Vulnerability.” Journal of the Royal Society of Health 117(3):180–85. doi:10.1177/146642409711700310

67.

Rubin, Donald B. 2001. Multiple Imputation for Nonresponse in Surveys. New York: John Wiley & Sons.

68.

Sailas

Eila S.

Feodoroff

Benjamin

Lindberg

Nina C.

Virkkunen

Matti E.

Sund

Reijo

Wahlbeck

Kristian

. 2006. “The Mortality of Young Offenders Sentenced to Prison and Its Association with Psychiatric Disorders: A Register Study.” European Journal of Public Health 16(2):193–97. doi:10.1093/eurpub/cki169

69.

Sampson

Robert J.

Laub

John H.

. 1990. “Crime and Deviance Over the Life Course: The Salience of Adult Social Bonds.” American Sociological Review 55(5):609. doi:10.2307/2095859

70.

Savolainen

2009. “Work, Family and Criminal Desistance: Adult Social Bonds in a Nordic Welfare State.” British Journal of Criminology 49(3):285–304. doi:10.1093/bjc/azn084

71.

Schnittker

Jason

Massoglia

Michael

Uggen

Christopher

. 2012. “Out and Down: Incarceration and Psychiatric Disorders.” Journal of Health and Social Behavior 53(4):448–64. doi:10.1177/0022146512453928

72.

Siponen

Rebecca

Andersson

Anneli

Oskarsson

Sofi

Garcia-Argibay

Miguel

Beckley

Amber L.

Långström

Niklas

Fazel

Seena

Chang

Zheng

Larsson

Henrik

Evans

Brittany

Tuvblad

Catherine

. 2023. “A Population-Based Study of Unintentional Injury and Premature Death among Non-Imprisoned and Imprisoned Youth Offenders.” Journal of Criminal Justice 84:102009. doi:10.1016/j.jcrimjus.2022.102009

73.

Skardhamar

Torbjørn

Skirbekk

Vegard

. 2013. “Relative Mortality among Criminals in Norway and the Relation to Drug and Alcohol Related Offenses.” PLOS ONE 8(11):e78893. doi:https://doi.org/10.1371/journal.pone.0078893

74.

Skinner

Guy C. M.

David

P. Farrington

Darrick

Jolliffe.

. 2022. “Criminal Careers and Early Death: Relationships In the Cambridge Study in Delinquent Development.” The British Journal of Criminology 62(4):840–56. doi:10.1093/bjc/azab092

75.

Skinner

Guy C. M.

Farrington

David P.

. 2020. “A Systematic Review and Meta-Analysis of Premature Mortality in Offenders.” Aggression and Violent Behavior 53:101431. doi:10.1016/j.avb.2020.101431

76.

Statacorp. 2021. Stata Statistical Software: Release 17. College Station, Texas: StataCorp LLC.

77.

Stattin

Hakan

Romelsjo

Anders

. 1995. “Adult Mortality in the Light of Criminality, Substance Abuse, and Behavioural and Family-Risk Factors in Adolescence.” Criminal Behaviour and Mental Health 5(4):279–311. doi:10.1002/cbm.1995.5.4.279

78.

Stenbacka

Marlene

Moberg

Tomas

Romelsjö

Anders

Jokinen

Jussi

. 2012. “Mortality and Causes of Death among Violent Offenders and Victims—A Swedish Population Based Longitudinal Study.” BMC Public Health 12(1):38. doi:10.1186/1471-2458-12-38

79.

Stewart

Lynn A.

Wilton

Geoff

. 2014. “Correctional Outcomes of Offenders with Mental Disorders.” Criminal Justice Studies 27(1):63–81. doi:10.1080/1478601X.2013.873205

80.

Sugie

Naomi F.

Turney

Kristin

. 2017. “Beyond Incarceration: Criminal Justice Contact and Mental Health.” American Sociological Review 82(4):719–43. doi:10.1177/0003122417713188

81.

Swartz

Marvin S.

Jeffrey

W. Swanson

Hiday

Virginia A.

Borum

Randy

Wagner

H. Ryan

Burns

Barbara J.

. 1998. “Violence and Severe Mental Illness: The Effects of Substance Abuse and Nonadherence to Medication.” American Journal of Psychiatry 155(2):226–31. doi:10.1176/ajp.155.2.226

82.

Taylor

Caitlin J.

2016. “The Family’s Role in the Reintegration of Formerly Incarcerated Individuals: The Direct Effects of Emotional Support.” The Prison Journal 96(3):331–54. doi:10.1177/0032885516635085

83.

Teplin

Linda A.

Gary

M. McClelland

Abram

Karen M.

Mileusnic

Darinka

. 2005. “Early Violent Death Among Delinquent Youth: A Prospective Longitudinal Study.” Pediatrics 115(6):1586–93. doi:10.1542/peds.2004-1459

84.

Teplin

Linda A.

Jakubowski

Jessica A.

Abram

Karen M.

Olson

Nichole D.

Stokes

Marquita L.

Welty

Leah J.

. 2014. “Firearm Homicide and Other Causes of Death in Delinquents: A 16-Year Prospective Study.” Pediatrics 134(1):63–73. doi:10.1542/peds.2013-3966

85.

Tremblay

Pierre

Pare

Paul-Philippe

. 2003. “Crime and Destiny: Patterns in Serious Offenders’ Mortality Rates.” Canadian Journal of Criminology and Criminal Justice 45(3):29.

86.

Uno

Hajime

Cai

Tianxi

Pencina

Michael J.

D’Agostino

Ralph B.

Wei

L. J.

. 2011. “On the C-Statistics for Evaluating Overall Adequacy of Risk Prediction Procedures with Censored Survival Data.” Statistics in Medicine 30(10):1105–17. doi:10.1002/sim.4154

87.

van de Weijer

Steve

Bijleveld

Catrien

Huschek

Doreen

. 2016. “Offending and Mortality.” Advances in Life Course Research 28:91–9. doi:10.1016/j.alcr.2015.11.004

88.

Vaughan

Adam D.

Denise

M. Zabkiewicz

Verdun-Jones

Simon N.

. 2017. “In Custody Deaths of Men Related to Mental Illness and Substance Use: A Cross-Sectional Analysis of Administrative Records in Ontario, Canada.” Journal of Forensic and Legal Medicine 48:1–8. doi:10.1016/j.jflm.2017.03.002

89.

Viljoen

Jodi L.

Andrew

L. Gray

Shaffer

Catherine

Bhanwer

Aisha

Tafreshi

Donna

Douglas

Kevin S.

. 2017a. “Does Reassessment of Risk Improve Predictions? A Framework and Examination of the SAVRY and YLS/CMI.” Psychological Assessment 29(9):1096–110. doi:10.1037/pas0000402

90.

Viljoen

Jodi L.

Andrew

L. Gray

Shaffer

Catherine

Latzman

Natasha E.

Scalora

Mario J.

Ullman

Daniel

. 2017b. “Changes in J-SOAP-II and SAVRY Scores Over the Course of Residential, Cognitive-Behavioral Treatment for Adolescent Sexual Offending.” Sexual Abuse 29(4):342–74. doi:10.1177/1079063215595404

91.

Vincent

Gina

Grisso

Thomas

. 2005. “A Developmental Perspective on Adolescent Personality, Psychopathology, and Delinquency.” Pp. 22–43 in Mental Health Screening and Assessment in Juvenile Justice, edited by Grisso

Thomas

Vincent

Gina

Seagrave

Daniel

. New York: Guilford Press.

92.

Visher

Christy A.

Travis

Jeremy

. 2003. “Transitions from Prison to Community: Understanding Individual Pathways.” Annual Review of Sociology 29(Volume 29, 2003):89–113. doi:10.1146/annurev.soc.29.010202.095931

93.

Wikström, Per-Olof

2020. “Explaining Crime and Criminal Careers: The DEA Model of Situational Action Theory.” Journal of Developmental and Life-Course Criminology 6(2):188–203. doi:10.1007/s40865-019-00116-5

94.

Wildeman

Christopher

Andersen

Lars H.

. 2020. “Solitary Confinement Placement and Post-Release Mortality Risk among Formerly Incarcerated Individuals: A Population-Based Study.” The Lancet Public Health 5(2):e107–13. doi:10.1016/S2468-2667(19)30271-3

95.

Wiley

Stephanie A.

Love

Helene

Emmett

Kelin A.

. 2020. “Indigenous Over-Representation in Canada’s Youth Correctional System: An Assessment of Regional Variability.” Canadian Journal of Criminology and Criminal Justice 62(2):22–43. doi:10.3138/cjccj.2019-0049

96.

Zane

Steven N.

Brandon

C. Welsh

Zimmerman

Gregory 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(4):691–713. doi:10.1007/s10940-018-9399-4

97.

Zheng

Nanzi

Abram

Karen M.

Welty

Leah J.

Aaby

David A.

Meyerson

Nicholas S.

Teplin

Linda A.

. 2023. “Nonfatal Firearm Injury and Firearm Mortality in High-Risk Youths and Young Adults 25 Years After Detention.” JAMA Network Open 6(4):e238902. doi:10.1001/jamanetworkopen.2023.8902

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

1.46 MB

0.00 MB