Abstract

Mental illness is firmly established as a risk factor for criminal legal system contact, particularly for women and Indigenous people. While patterns of criminal legal contact vary by gender and Indigenous status, we do not know how mental health contacts factor into these patterns. The aim of this research is to examine whether mental health characteristics and service contacts vary across patterns of criminal legal system contact defined by group-based trajectory modelling, and to explore whether any such variation is consistent across gender and Indigenous status. Using linked administrative data from a 1990 Australian birth cohort (to age 23/24 years, N = 45,141), we estimate trajectories of criminal legal system contact and assess variation across groups defined by gender and Indigenous status. We then examine whether types of mental illness diagnoses and mental health service contacts varied across trajectory groups, and whether this was consistent across gender and Indigenous status. Findings point to important differences in mental health system contact across offending trajectory groups. Differences are suggestive of variation in mental health system utilization at the intersection of gender and Indigenous statuses that are conditioned by patterns of criminal legal system contact. We conclude by outlining the implications of these patterns for life course theories of offending and for gender and culturally informed support and interventions directed towards system-involved individuals with mental health needs.

James M. Ogilvie^1,2†, Lisa Broidy^1,3, Carleen Thompson^1,2, Susan Dennison^1,2, Troy Allard^1,2, Aydan Kuluk^1,2, Belinda Crissman^1,2, Steve Kisely⁴ and Anna Stewart^1,2*

Postprint forthcoming in: Journal of Developmental and Life Course-Course Criminology

¹ Griffith Criminology Institute, Griffith University

² School of Criminology and Criminal Justice, Griffith University

³ Department of Sociology, University of New Mexico

⁴ School of Medicine, The University of Queensland

^† Corresponding author [email protected]. ORCID: 0000-0003-2294-5665. 176 Messines Ridge Road. Griffith Criminology Institute, Griffith University. Mount Gravatt, QLD 4122 AUSTRALIA

^* Deceased on 23 April 2021.

Introduction

Mental illness is firmly established as a risk factor for criminal legal system contact, particularly for women and Indigenous people (Ogilvie et al., 2021). While longitudinal trajectory patterns of criminal legal system contact vary by gender and Indigenous status (Broidy et al., 2015), we know less about how mental health outcomes factor into these patterns. The current research aims to address this gap by focusing on two key questions: 1) Do patterns of mental health system contact vary across longitudinal trajectories of offending; and 2) Does variation in mental health system contact across offending trajectories hold similarly by gender and Indigenous status? Given the overrepresentation of Aboriginal and Torres Strait Islander1 peoples within the Australian criminal legal system, this work has important implications for policy and practice addressing their mental health needs. Moreover, developmental life course theories identify mental illness as a prominent risk factor among female offenders, but rarely consider its racialized aspects or its role in the offending pathways of males. In detailing how mental health contacts and diagnoses link to offending trajectories and whether and how they vary across gender and race/ethnicity, our findings can inform developing theory and practice to move beyond the conflation of mental health with female offending.

Offending trajectories

There is a rich body of research utilizing group-based trajectory modelling (GBTM) to disaggregate longitudinal offending patterns (DeLisi & Piquero, 2011; Nagin & Odgers, 2010). Across diverse longitudinal datasets and studies, there is consistency in demonstrating that offending over the life course can be disaggregated into distinct patterns defined by timing and intensity (Nagin & Tremblay, 2005). The most consistently identified trajectories typically include early onset persistent/chronic; adolescent limited; and low offending patterns (DeLisi & Piquero, 2011), with more recent research also identifying late/adult onset offending patterns (Matsuda et al., 2022).

Research using GBTM has further demonstrated that offending trajectories can be differentiated by demographic characteristics, predictors/risk factors, and consequences, which has important implications for delivering targeted and early intervention to prevent the progression to poor outcomes (Nagin & Tremblay, 2005). Gender and racial/ethnic differences in the composition of different offending trajectories have been consistently demonstrated. Early onset and persistent/chronic offending trajectories are more prevalent among males compared to females, with females more likely to populate low/rare and adolescent-limited trajectories of offending (Broidy et al., 2015; Walker et al., 2019). North American (Steffensmeier et al., 2011) and Australian (Broidy et al., 2015; Ferrante, 2013) research highlights that persistent/chronic offending patterns are more prevalent among minorities compared to Whites, which is believed to reflect the disadvantages that accumulate among minority populations (Broidy et al., 2015).

Several studies have examined how gender and race/ethnicity intersect in the composition of offending trajectory patterns. For example, Piquero and Buka (2002) found that African American males and females were more likely than non-African American males and females to populate chronic offending trajectories. In the Australian context, Broidy et al. (2015) found that males and Indigenous Australians were more likely to be classified in serious and chronic offending trajectories. They further found that gender and race/ethnicity interact to influence offending patterns, with non-Indigenous females most likely to have low rates of offending, and Indigenous males having comparatively exaggerated rates of offending (Broidy et al., 2015). Similarly, Matthews et al. (2022) found that there was a higher proportion of non-Indigenous males and females in a no/low adult conviction trajectory group, while Indigenous males and females were more likely to populate a persistent trajectory. This clearly shows that the intersection of gender and race/ethnicity conditions patterns of criminal legal system contact. Although longitudinal patterns of offending and criminal legal system contact can be demarcated by demographic characteristics, less is known about how trajectories of offending may differ by other characteristics as well, including mental health.

Offending trajectories and mental health

Despite the well-established link between offending and mental illness, GBTM has been underutilized in exploring this relationship, resulting in only limited information about how mental health outcomes may vary across different longitudinal patterns of offending (Piquero et al., 2011). Odgers et al. (2007) made one of the first attempts to examine how mental health outcomes may differ across longitudinal patterns of conduct problems defined by GBTM using data from the Dunedin Mutlidisciplinary Health and Development Study. Odgers et al. (2007) found that by age 32 years, individuals classified by childhood onset and persistent conduct problems had the worst mental health outcomes (e.g., psychiatric disorders, receiving services and medication, hospitalizations, suicide attempts, homelessness) compared to other offending classes. Similarly, Testa and Semenza (2020) found that individuals classified in persistent high-rate offending trajectories were more likely to experience a higher level of depressive symptoms. These individual study findings are supported by the meta-analytic results of Reising et al. (2019), who examined the association between offending trajectories and mental health outcomes (i.e., depression and anxiety) from nine longitudinal studies. They found that the odds of mental health problems were nearly three-times higher for persistent offenders and almost twice as high for late-onset offenders compared to non-offenders (Reising et al., 2019). Overall, there is consistency across studies in demonstrating that more persistent/chronic offending patterns are associated with poorer mental health outcomes. Although available research demonstrates differences across offending trajectories in mental health outcomes, the outcomes studied have typically been limited to a restricted set of broad indicators (i.e., anxiety or depression symptoms), and without considering whether differences are consistent across gender and race/ethnicity.

Offending trajectories, mental health, and gender

Gender differences in the association between offending and mental health have been established, mainly based on cross-sectional studies examining the prevalence of mental illnesses in prison populations. For example, a higher prevalence of mental illnesses among female compared to male prisoners is consistently documented (Australian Institute of Health and Welfare, 2019; Browne et al., 2022). Females in the criminal legal system are more likely than males in the criminal legal system to report a range of mental health symptoms and they are also more likely to utilize mental health services (Albizu-Garcia et al., 2001; Cauffman et al., 2007). Female inmates are also significantly more likely than male inmates to seek mental health treatment during incarceration (Drapalski et al., 2009). Research also suggests that females with a lifetime history of arrest are more likely than females with no history of criminal legal system contact to access mental health services, which likely reflects the histories of victimization that characterize most women in the system (Nowotny et al., 2019). At the same time, research suggests that when mental illness and offending co-occur, males have more extensive criminal histories and are more likely to exhibit histories of violent and sexual offending than females (de Vogel & de Spa, 2019). There is also some evidence to suggest gender differences in the types of mental illnesses present among prisoners, with females more likely to be diagnosed with anxiety or personality disorders, and males more likely to be diagnosed with substance abuse or psychotic disorders (Stewart et al., 2021). However, cross-sectional findings do not provide insight into how gender may condition the association between mental health diagnoses, mental health system contacts, and criminal legal system contacts over the life course.

A small but growing body of literature has examined the association between mental health and offending trajectories for males and females using longitudinal data (Bergman & Andershed, 2009; Odgers et al., 2008; Walker et al., 2019). Two longitudinal studies found that life course persistent females were significantly more likely than life course persistent males to experience mental health problems including anxiety and major depression (Odgers et al., 2008; Walker et al., 2019). Odgers et al. (2008) also found that females in the low offending trajectory were more likely than low offending males to have internalizing mental health problems. Another longitudinal study found that psychiatric care was far more common across the female offending trajectories compared to those of males. Specifically, females following persistent (60%), adolescent-limited (15.8%), and adult-onset (21.2%) offending trajectories were significantly more likely than males following persistent (30.8%), adolescent-limited (7.1%), and adult-onset (9%) offending trajectories to have been registered for psychiatric care (Bergman & Andershed, 2009). At the same time, this work shows patterns that suggest more significant mental health needs among those with the most serious offending involvement across both females and males.

Offending trajectories, mental health, and race/ethnicity

Available research shows racial/ethnic disparities in mental health needs and utilization among justice involved populations. For example, two US studies found, compared to whites, Black youth in detention/probation presented with more mental health needs, yet received lower rates of mental health care. Furthermore, white youth were twice as likely as Black youth to be detected as needing mental health treatment (Rawal et al., 2004; Teplin et al., 2005). Australian studies have found that diagnosed mental illnesses are more prevalent among Indigenous compared to non-Indigenous justice-involved populations, including amongst prisoners (Butler et al., 2007; Stewart et al., 2021) and individuals with proven offenses from court appearances (Ogilvie et al., 2021). However, research examining the influence of race/ethnicity on the association between longitudinal offending patterns and mental health is rare. An exception is the Australian study conducted by Valuri et al. (2021), who used population-based data to compare the offending trajectories among individuals with psychotic disorders, other mental disorders, and no mental disorders. Among individuals with psychotic disorders and other mental disorders, Indigenous people were more likely to be classified in any of the offender groups compared to non-offender groups.

Offending trajectories and mental health at the intersection of gender and race/ethnicity

Based on available evidence, criminal legal and mental health system contacts appear to be simultaneously impacted by gender and race/ethnicity in complicated ways. However, studies tend to examine these demographic characteristics in isolation, masking differences across intersecting demographic subgroups that might help explain variation in mental health and criminal legal system contacts. Ogilvie et al. (2021) examined how prevalence rates of mental disorders among individuals with criminal legal system contact differed across both gender and Indigenous status. They found that Indigenous females (24.1%) and Indigenous males (20.8%) with a proven offense were more likely than non-Indigenous females (16.3%) and non-Indigenous males (11%) with a proven offense to be diagnosed with any mental disorder. This demonstrated that Indigenous status moderated the effect of gender on prevalence rates of mental disorder, with Indigenous males having higher rates compared to non-Indigenous females, for example. These findings reinforce the importance of considering intersectionality to improve our understanding of the influence of gender and race/ethnicity on the nexus between mental health and criminal legal system contacts. To date, we are not aware of any studies that have sought to examine the connection between mental health system contacts and diagnoses and longitudinal trajectories of criminal legal system contact at the intersection of gender and race/ethnicity, with this being the focus of the current study.

Current study and research questions

Research utilizing GBTM has demonstrated that mental health outcomes vary across longitudinal patterns of offending and criminal legal system contact, highlighting the capacity of the methodology to provide deeper insights into the heterogeneity of offending and system contact across the life course. At present there is limited knowledge about how mental health system contacts and diagnoses overlap with trajectories of criminal legal system contact and are impacted by both gender and race/ethnicity. Our study seeks to address this gap in knowledge by firstly identifying longitudinal trajectory patterns of criminal legal system contact in an Australian birth cohort, that were then used to examine two research questions:

1. Do patterns of mental health system contact vary across longitudinal trajectories of offending; and

2. Does variation in mental health system contact across offending trajectories hold similarly by gender and Indigenous status?

Regarding the first research question, consistent with available evidence (Reising et al., 2019; Testa & Semenza, 2020; Walker et al., 2019), we expect that patterns of criminal legal system contact reflecting more serious and persistent/chronic offending trajectories will be characterized by a higher burden of mental illness and greater mental health service contacts compared to low offending trajectories. There is insufficient information from previous research to make specific predictions about variation in the types of mental illnesses that will be present across different groups, apart from depression and anxiety being likely in persistent/chronic groups (Reising et al., 2019).

For the second research question, previous studies (Broidy et al., 2015; Odgers et al., 2008) would suggest that we will find both gender and Indigenous status differences in the connections between the offending patterns reflected in trajectories of criminal legal system contact and mental health outcomes and service use. Specifically, males and Indigenous individuals will be more likely to populate persistent/chronic offending trajectories, with females more concentrated in low and adolescent-limited offending trajectories. Further, we expect Indigenous males and females to have higher rates of diagnosed mental disorders and mental health service contacts compared to non-Indigenous males and females across trajectories of offending. Given the absence of evidence, we have no expectations for how the intersection of gender and Indigenous status might affect the links between offending trajectories and mental health outcomes, with our analyses being an important first step in identifying these linkages.

Addressing our research questions requires a longitudinal dataset with large samples of both females and racial/ethnic minorities to be able to disaggregate by gender and race/ethnicity within various life course offending trajectories. Here we utilize a population-based longitudinal administrative dataset of an entire cohort of individuals born in Queensland in 1990. This large dataset allows us to examine the differences in offending reflected in trajectories of criminal legal system contact at the intersection of gender and Indigenous status and to assess how mental health outcomes and service use presents within and across these life course offending trajectories by gender and Indigenous status. This has important implications for both theory and practice in relation to meeting overlapping needs of individuals reflected in criminal legal and mental health system contact.

Methods

Data sources and sample

This study uses population-based linked administrative data from the Queensland Cross-sector Research Collaboration (QCRC) repository (Stewart et al., 2021) for all individuals born in Queensland in 1990. We utilize longitudinal administrative records across the following government agencies/systems: Queensland Registry of Births, Deaths and Marriages (births and deaths records); Queensland Police (youth cautions and conferences); youth justice (youth courts)2; Queensland Department of Justice and Attorney General (adult courts); and Queensland Health (mental health service records). Data are stored in the Social Analytics Lab, which is a secure purpose-built facility for storing sensitive data. The study was approved by the Griffith University Human Research Ethics Committee (HREC 2010/479).

The full cohort comprises 45,422 individuals born in Queensland, Australia in 1990 (demographic details provided in Table 1).

Table 1. Offending and mental health service contacts up to age 24 years by gender and Indigenous status for the total 1990 cohort.

Note: Mental health service contact data are to age 23 potentially resulting in minor right censoring of this variable.

Among the cohort, 12,690 individuals (27.9%) had at least one proven offense from a court finalization or youth police diversion by age 24 years, 5,359 (11.8%) had at least one mental health service contact by age 23 years, and 2,828 (6.2%) had experienced both. The sample for this study was limited to cohort members with at least one proven offense (n = 12,690; 30.3% female; 16.2% Indigenous). In the sample of individuals with at least one proven offense, 22.3% (n = 2,828) had at least one mental health service contact. Male, χ²(1) = 2351.84, p<0.001, Cramer’s V [ϕ_c] = .23, and Indigenous, χ² (1) = 2887.32, p<0.001, ϕ_c =.25, individuals were significantly overrepresented among those with a proven offense compared to female and non-Indigenous individuals, respectively.

Criminal legal system contact

We extracted data on criminal legal system contacts from finalized court outcomes as well as police diversions. Court information was available from the age of criminal responsibility (i.e., 10 years in Queensland) up to age 24 years. We aimed to only include those contacts associated with an offending incident. As such, we only include offenses where an individual is found guilty or pleads guilty to an offense(s) and excluded offenses with not guilty outcomes. We also included Queensland Police Service data on diversions up to age 24 years (i.e., cautions and conferences)3, as these diversions required an admission of guilt by the accused. Multiple offenses and offense occurrences (i.e., offenses occurring on separate dates) can be finalized on a single date. For each individual, we summed the number of offense related contacts for each calendar year (i.e., based on court and/or diversion finalization date) in the observation period. Offense counts were significantly positively skewed and zero-inflated across the observation period (i.e., most individuals offended for short periods). To minimize skew and the influence of outlier cases and to ensure model convergence, we capped offense counts (i.e., top-coded) at a maximum of 75 offenses 4 per year for each individual. For covariate analyses, we classified offenses into the three broad categories of violent, nonviolent and other minor offenses based on the Australian Standard Offence Classification, Queensland Extension (see Supplementary Table S1 for detailed divisions, subdivisions and corresponding QASOC codes for offenses within these categories). We also coded a binary violent offense history indicator (present/not present) to reflect whether an individual had ever had a proven violent offense finalized in court between 10-24 years (see Supplementary Table S1 for a list of violent offenses as defined by the QASOC).

Failing to account for time in custody can both underestimate the level of offending (and potential number of criminal legal system contacts) and distort the shape of estimated trajectories (Piquero et al., 2001). Time in custody reduces exposure time in the community, thereby limiting opportunities to both offend and get caught. We used court sentencing records to estimate time at risk (i.e., time not spent in custody). We calculated time at risk for each individual by summing the number of days sentenced to custody at each distinct finalized court appearance for each year of observation. If this value was greater than 365 days within the year, the excess value was carried over into the next calendar year. We then converted days in custody per year to a proportion between 0-1, where 0 represented no time spent at risk in community and 1 represented an individual being at risk for the full year. We used this time at risk measure to adjust the offending data using the divide and round method (see Nielsen et al., 2014), which involved adjusting each yearly offense count by dividing it by the corresponding yearly time at risk and rounding the result to the nearest integer. Court sentencing information may not provide the most accurate estimation of time at risk, given it does not measure actual time spent in custody and does not account for possible days spent on remand during the legal process. However, days sentenced to custody is likely to have strong correlation with actual time in custody, and therefore an adequate indicator for exposure time. Top-coding/capping of offense counts was performed after the exposure adjustment.

Mental health service contacts and psychiatric diagnoses

Mental health service contacts. We secured details about mental health service usage from the Queensland Health Consumer Integrated Mental Health Application (CIMHA) dataset. CIMHA covers information on contacts with Community Mental Health Services. Contacts recorded in CIMHA relate to service events delivered by healthcare providers, which could include in-person contacts (e.g., individual and group treatment sessions), telephone contacts (both with consumer and other service providers), emails, consultation, and assessment activities, for example. CIMHA recorded contacts relate to healthcare provision activities and therefore act as a proxy for the intensity of service utilization for consumers. We extracted the total number of service contacts for everyone in the sample. These data cover the period from September 2000 to December 2013, providing utilization data for sample members from age 10 through to age 23.

Psychiatric diagnoses. We extracted information about psychiatric diagnoses from the Queensland Hospital Admitted Patient Data Collection (QHAPDC), which contains information for public and private hospital admissions in Queensland. QHAPDC data were available from 1 July 1995, with hospital admissions therefore left-censored until age four/five years for the cohort. QHAPDC separations to June 2014 were extracted and were therefore available from age four/five to age 23/24 for the cohort. Psychiatric diagnoses were coded according to the ICD-9CM (International Statistical Classification of Diseases – clinical modification) or ICD-10AM (Australian Modification; World Health Organization, 2004). ICD-9 codes were converted to corresponding ICD-10 codes. We included all primary and secondary psychiatric diagnoses across all admissions, therefore representing the lifetime prevalence of diagnosed psychiatric illness from hospital admissions between the ages 4/5 to 23/24 years. For the current study, we coded the presence of any psychiatric disorder (ICD-10 Mental and behavioral disorders [codes F00 to F99], as well as suicidal ideation [code R45.8] and self-harm [codes X60 through X84]) as either present or absent. Within this overarching classification, we coded individuals for the presence or absence of three specific diagnostic classes: psychotic disorders (schizophrenia, schizotypal and delusional disorders [F20-F29], severe or psychotic affective disorders [F30, F31, F32.2, F32.3], and psychotic disorders related to substance use [F10.5, F11.5, F12.5, F13.5, F14.5, F15.5, F15.50, F15.51, F15.59, F15.70, F16.5, F17.5, F18.5, F19.5, F19.7]); mood and anxiety disorders [F32.0, F32.1, F32.8, F32.9, F33-F48]; and substance use disorders excluding those involving psychosis [F10-F19].

Covariates

Our analyses focus on variation in the overlaps across criminal legal and mental health system contacts by gender and Indigenous status. Consistent with best-practice guidelines for linked data (Australian Institute of Health and Welfare, 2012), Indigenous status was assigned if an individual had ever self-identified or been labelled within a system as Indigenous (Aboriginal and/or Torres Strait Islander) in any of the QCRC databases. We recognise the problems inherent in using the overarching term of Indigenous status, since it obscures the diversity of the more than 500 Aboriginal and Torres Strait Islander nations in Australia (Walter et al., 2020). However, the use of this overarching term is unavoidable in our analyses, as the QCRC datasets do not provide further detail on Indigenous identity. We classified individuals as either female or male based on what was most commonly recorded across all QCRC databases. While not a comprehensive measure of sex or gender, we believe that any difference across females and males reflect a range of bio-psycho-social dynamics and use the term gender (rather than sex) to describe this demographic categorization. We also examine age of onset for both criminal legal and mental health system contacts. Age at first contact with the criminal legal system represents the date of finalization in court for each individual’s first court appearance or the date at caution/conference for diversions (if these precede the first court finalization). Age at first mental health service use represents the date of first recorded CIMHA contact.

Analytical strategy

To examine offending trajectories and mental health outcomes, and how these vary by gender and Indigenous status, we employed a two-stage analytical approach.

Stage one: Group-based trajectory modelling of offending.

Group-based trajectory modelling (GBTM) is a specialized application of finite mixture modelling used to identify clusters of individuals following similar progressions of behaviors or outcomes longitudinally (GBTM; Muthén & Muthén, 2000; Nagin & Tremblay, 2001; Nagin et al., 2016). GBTM has been extensively applied to model heterogeneity in longitudinal patterns of both self-reported offending and administrative records of criminal legal system contact (Nagin & Odgers, 2010). The current trajectory analyses are reported in line with the Guidelines for Reporting on Latent Trajectory Studies checklist (van de Schoot et al., 2016).

We used the FLXMRglm driver in the flexmix package (version 2.3-17; Grün & Leisch, 2008) for R (version 4.2.0; R Core Team, 2022) to fit trajectory models to the data. To estimate offending trajectories reflected in patterns of criminal legal system contacts, we used Poisson models with quadratic functions of age/time. Poisson models are appropriate to model count variables characterized by a minimum value of zero and an unbound maximum. We estimated models with one to six classes, settling on six as the upper limit to preserve parsimony and interpretability, and ensure that class sizes were large enough to provide adequate statistical power for subsequent analyses. To guide model selection we relied on multiple goodness-of-fit and classification accuracy statistics, including log likelihood, Akaike information criterion (AIC), Bayesian information criterion (BIC), entropy (Celeux & Soromenho, 1996), the average posterior probability (AvePP) of group classification for most likely group membership, and odds of correct classification (OCC; Nagin, 2005). The best fitting models are typically identified by low log likelihood, AIC and BIC values (Nylund et al., 2007), and entropy and AvePP values closest to one. AvePPs above 0.70 and OCCs greater than 5 for all groups are considered indicative of adequate fit (Nagin, 2005). Following convention, we also considered the interpretability of the estimated models in reference to existing theory and research, and the principle of parsimony, referring to the preference for selecting the model with the fewest number of classes that is statistically significant and substantively adequate (Fergusson et al., 2000; Masyn, 2013). We assigned individuals to trajectory groups based on maximum posterior group probabilities, with group membership used as dependent variables for subsequent analyses. This approach was appropriate given the high entropy values (>.80) of the fitted models, which indicate high separation between the modelled groups (Ram & Grimm, 2009). At the initial stage, we examined potential correlates of group membership descriptively using chi-square tests for categorical variables (i.e., gender, Indigenous status, violent offense history) and ANOVA for continuous variables (i.e., age at first offense and mean offenses).

Stage two: Comparison of mental health outcomes

To evaluate variation in mental health outcomes across trajectory groups, generally and in relation to gender and ethnicity, we used chi-square tests for categorical variables (i.e., gender, Indigenous status, mental health service contact, psychiatric diagnosis) and ANOVA for continuous variables (i.e., age at first mental health service contact and mean number of service contacts from the CIMHA dataset). We then conducted multinomial logistic regressions to examine whether the offending trajectory groups could be distinguished by the combined effects of demographic (gender and Indigenous status), offense (ever committed a violent offense), and mental health variables (ever had contact with a mental health service, psychiatric diagnosis). We estimated two models, the first including a binary indicator of the presence of any psychiatric disorder, and the second including three binary indicators of the presence of specific categories of psychiatric disorders (i.e., psychotic disorders, substance use disorder, mood/anxiety disorders).

Results

Descriptive information

We begin by describing patterns of criminal legal system contacts as well as mental health system contacts and comparing contacts across gender and Indigenous status for the full observation period (Table 2). The sample accrued 122,870 offenses and 82,309 mental health service contacts over the observation period. Gendered patterns showed variation across criminal legal contacts and mental health system contacts with males exhibiting a significantly higher mean number of offenses compared to females (t = 15.54, p <.001) while females had a significantly higher mean number of mental health service contacts (t = 5.11, p <.001) relative to males. Further patterns were evident when examining differences by Indigenous status. Compared to their non-Indigenous counterparts, Indigenous individuals exhibited significantly higher offense counts (t = 18.46, p <.001) and mental health service contacts (t = 4.56, p <.001). Notably, Indigenous females have higher offense counts than non-Indigenous males, and Indigenous males have more mental health service contacts than non-Indigenous females.

Table 2. Descriptive information for offending and mental health service contacts by gender and Indigenous status (n = 12,690).

* p < .05, ** p <.01, *** p < .001.

Descriptive statistics for offending counts across age (see Supplementary Table S2) demonstrated that within each year offending counts were significantly positively skewed and consistent with a Poisson distribution. Mean counts of mental health service contacts and offenses were plotted by age and stratified by gender and Indigenous status (Figure 1). For mental health service contacts, there were two prominent peaks at 17 and 21 years, with a dip in contacts between these years. It is possible this dip reflected transitional effects between youth and adult mental health services. Indigenous and non-Indigenous females generally exhibited the highest means of service contacts across the entire observation period.

The pattern of mean offenses by age was largely consistent with the typical age-crime curve (i.e., mean offenses increasing markedly throughout adolescence to a peak at 18/19, declining through the early 20s), with the most prominent peak for Indigenous males at 18-years-old. The exception was a persistent gradual increase in mean offending for non-Indigenous females across the entire observation period. Given differences across gender and Indigenous status in both criminal legal and mental health system contacts, we next model group-based offending trajectories and then examine variation in mental health contact patterns across offender groups. We also explore differences in group membership across gender and Indigenous status and links to patterns of mental health system utilization.

Figure 1. Mean count of mental health service contacts and offenses by age, stratified by gender and Indigenous status.

Offending trajectories

Goodness-of-fit and classification accuracy statistics for the model selection process are provided in Supplementary Table S3. AIC and BIC improved iteratively with the addition of groups with no worsening of fit up to six groups, and entropy, AvePP and OCC values remained adequate across all models. We rejected the six-group model because it lacked parsimony, with three groups exhibiting low trajectories of offending. The five-group model contained two groups comprising < 2.0% of the sample but had trajectory group coverage for low, moderate and high levels of offending, which the four-group model was missing. Therefore, we selected the five-group trajectory model as the best model for offending with interpretable trajectory shapes (model coefficients and classification statistics are provided in Supplementary Table S4, and a rootogram of posterior probabilities for group assignment is displayed in Supplementary Figure S1). Table 3 details descriptive information for the five-group trajectory model.

Table 3. Five-group offending trajectory model coefficients, classification diagnostic statistics and descriptive information (n = 12,690).

^a Cases with no mental health service contacts excluded.

As illustrated in Figure 2, the five-group trajectory model includes the following offending groups: high-rate late adolescent peak (n = 212, 1.7%), low (n = 9,573, 75.4%), low early adult escalating (n = 1,688, 13.3%), adolescent limited (n = 987, 7.8%) and high-rate escalating (n = 230, 1.8%). Observed individual trajectory profiles by group classification are displayed in Supplementary Figure S2 and demonstrated significant variability of individual trajectories within each group. However, collectively the individual trajectories approximated the shapes of the fitted trajectories.

Figure 2. Fitted mean offense counts by age for the five-group trajectory model.

Tests of group differences in mean offense counts and mean age at first offense were conducted to confirm separation of the groups generated by the GBTM process. There was a significant difference across groups in the mean number of offenses per individual, F(4, 12,685) = 11,545.05, p <.001, η² = .79, with all post-hoc trajectory group comparisons significant using Tukey multiple comparisons of means adjustment. The high-rate late adolescent peak offending group had the highest mean number of offenses per individual and the low offenses group had the lowest mean number of offenses per individual. There was a significant difference across groups in the age at first offense, F(4, 12,685) = 572.85, p <.001, η² = .15, with all group comparisons significant using Tukey multiple comparisons of means adjustment. The high-rate late adolescent peak offending group were youngest on average at their first offense, and the low offending group was oldest on average at their first offense.

The high-rate late adolescent peak and high-rate escalating groups contained the fewest individuals (1.7% and 1.8% of the sample, respectively) but were responsible for disproportionately high counts of offenses (19.4% and 17.9% of all offenses, respectively). The low offending group contained most individuals (75.4%) but only accounted for 22.1% of all offenses. These findings demonstrate the concentration of offending within small groups of individuals. The proportion of individuals with a violent offense history differed significantly across groups, χ² (4, N = 12,690) = 2,382.30, p<0.001, ϕ_c = .43), with the high-rate late adolescent peak (74.1%) and high-rate escalating (69.6%) groups having the highest proportions of these individuals. In contrast, only 8.3% of individuals classified in the low offending group had a violent offense history.

The proportion of individuals assigned to trajectory groups differed significantly by gender, χ² (4, N = 12,690) = 265.04, p<0.001, ϕ_c = .14, and Indigenous status, χ² (4, N = 12,690) = 1,123.98, p<0.001, ϕ_c = .30). The composition of trajectory groups by gender and Indigenous status is displayed in Figure 3a. Indigenous males comprised the majority (50.9%) of the high-rate late adolescent peak group and comprised the second highest proportion of individuals in the high-rate escalating (37.0%), adolescent limited (29.2%) and low early adult escalating (17.8%) groups behind non-Indigenous males. The low group contained the highest proportion of non-Indigenous females (28.8%). Indigenous females were most concentrated in the high-rate escalating (9.6%) and adolescent limited (9.6%) groups.

Considering the intersection of gender and race/ethnicity (Figure 3b) we see notable differences in the distribution of individuals across trajectory groups. The greatest proportion of individuals in each gender by race/ethnicity subgroup populate the low offending group. However, for Indigenous males, this constitutes only 39.9% of the subsample, while it is 89.3% for non-Indigenous females. Indigenous females and non-Indigenous males fall between these two groups, but both have more than 65% in the low group. Similarly, although there are notably few non-Indigenous males and females in the two highest rate offending groups (high-rate escalating and high-rate adolescent peak), 14.8% of Indigenous males populate these two groups (combined), compared to 4.9% of Indigenous females.

Figure 3. Proportions of a) individuals assigned to trajectory groups by gender and Indigenous status; and b) individuals in each demographic group by trajectory group assignment.

Mental health outcomes across offending trajectories

There were significant differences across offending trajectory groups in the proportion of individuals who had experienced mental health service contact, χ² (4, N = 12,690) = 986.20, p<0.001, ϕ_c = .28 (Table 3). Patterns indicated a significant overlap between offending and mental health system contact. The high-rate late adolescent peak (68.4%) and high-rate escalating (65.7%) offending groups had the highest proportions of individuals with mental health service contact, and the low offending group had the lowest proportion of individuals with a mental health service contact (16.8%). There was a significant difference across trajectory groups in the mean number of mental health service contacts, F(4, 12,685) = 62.19, p <.001, η² = .02, with all post-hoc trajectory group comparisons significant using the Tukey multiple comparisons of means adjustment, except for between the adolescent limited and high-rate escalating groups. The high-rate late adolescent peak group had the highest mean number of contacts per individual, and the low group having the lowest mean number of mental health service contacts. There was also a significant difference across trajectory groups in the mean age at first mental health service contact, F(4, 2,823) = 16.54, p <.001, η² = .02, with the high-rate late adolescent peak group on average being younger at first contact when compared to the low, low early adult escalating and high-rate escalating trajectory groups. The proportion of individuals with any mental health disorder from a hospital admission differed significantly across offending trajectories, χ² (4, N = 12,690) = 514.09, p<0.001, ϕ_c = .20, with the highest rate in the high-rate escalating group (41.3%) and the lowest rate in the low offending group (10.2%). There were significant differences in the proportion of individuals with psychotic disorder, χ² (4, N = 12,690) = 72.50, p<0.001, ϕ_c = .08, mood or anxiety disorder, χ² (4, N = 12,690) = 117.45, p<0.001, ϕ_c = .10, and substance use disorder, χ² (4, N = 12,690) = 548.40, p<0.001, ϕ_c = .21, across trajectory groups. Across all trajectories, substance use disorders were the most prevalent disorders, with the highest rates observed in the high-rate escalating (34.4%) and high-rate late adolescent peak (25.0%) groups. Psychotic illnesses were the least prevalent disorders, with the high-rate escalating (7.0%) and high-rate late adolescent peak (6.6%) groups again exhibiting the highest rates. The high-rate escalating group had the highest prevalence of mood or anxiety disorders (14.8%) followed by the adolescent limited group (9.7%).

Mental health outcomes across offending trajectory groups were further disaggregated by gender and Indigenous status (Table 4) to reveal some notable intersectional differences. Compared to all other demographic groups, Indigenous females had the highest rates of mental health service contact and diagnosed mental illness. Specifically, for Indigenous females classified in the high-rate escalating offending group, over 90% had contact with mental health services and over 54% had been diagnosed with any mental illness, which were the highest rates for all demographic groups. For Indigenous and non-Indigenous females, individuals classified in the adolescent limited offending group had the highest average mental health service contacts, but with non-Indigenous females having much higher rates of contact on average. In contrast, for Indigenous and non-Indigenous males, the highest average mental health service contacts were observed for the high-rate late adolescent peak offending group. This highlights differences in mental health service use across offending trajectories that are conditional on gender. Regardless of Indigenous status, females in all offending trajectories exhibited higher rates of mood or anxiety disorders compared to their male counterparts.

Table 4. Mental health outcomes by trajectory group and disaggregated by gender and Indigenous status (n = 12,690).

We ran two multinomial logistic regression models to estimate the combined effects of demographic (i.e., gender and Indigenous status), offense (i.e., violent offense history), and mental health (i.e., mental health service contact, mental disorder diagnosis) characteristics on membership in the five offending trajectory groups, with the low offending trajectory as the reference group (see Table 5).

Table 5. Multinomial regression models of offending trajectory membership (n = 12,690).

Note: AIC = Akaike information criterion; RRR = relative risk ratios; SUD = substance use disorder; 95%CI = 95% confidence intervals.

^a Reference group is low offending trajectory.

^b Reference group is Indigenous females.

* p < .05, ** p <.01, *** p < .001.

For Model 1, the presence of any psychiatric disorder was included as a covariate, and for Model 2, the presence of specific psychiatric disorders was included. Consistent across Models 1 and 2, those with violent offense histories and who experienced mental health service contact were more likely to be in all higher rate offending trajectories compared to the low offending trajectory. The highest relative risk ratios for these variables were observed for high-rate late adolescent peak offending trajectory membership. Violent offense histories were associated with the highest relative risk ratios for being classified in the high-rate adolescent peak and high-rate escalating offending trajectory groups. Regarding demographic group with Indigenous females as the reference category, Indigenous males had the highest relative risk ratios for being classified in all higher rate offending trajectories compared to the low trajectory. Specifically, Indigenous males exhibited the highest relative risk of being classified in the high-rate late adolescent peak trajectory. Non-Indigenous females were significantly less likely to be classified in each of the non-low trajectories, with Indigenous females as the comparison. There was no significant difference between Indigenous females and non-Indigenous males in the likelihood of being classified in each of the non-low offending trajectories.

In Model 1, having any psychiatric disorder was significantly associated with being classified in all the higher rate offending trajectories compared to the low offending trajectory. The highest relative risk ratio for any psychiatric disorder was observed for the high-rate escalating offending trajectory. The results of Model 2 demonstrated that not all psychiatric disorder categories significantly differentiated offending trajectory groups, controlling for all other predictors. Psychotic disorders were not significantly associated with any of the offending trajectory groups. Mood/anxiety disorders were only significantly associated with a lower relative risk of being classified in the high-rate late adolescent peak trajectory compared to the low trajectory. In contrast, substance use disorders were significantly associated with higher relative risks of being classified in each of the higher rate offending trajectories compared to the low trajectory, particularly the high-rate escalating trajectory.

Discussion

In this study using linked longitudinal data for a birth cohort followed up to age 23/24 years, we sought to address the research questions: 1) Do patterns of mental health system contact vary across longitudinal trajectories of offending; and 2) Does variation in mental health system contact across offending trajectories hold similarly by gender and Indigenous status? The findings demonstrate the concentration of mental health service use and mental illnesses within specific groups of individuals who encounter the criminal legal system and provide further insight into how longitudinal patterns of offending vary across the intersection of gender and Indigenous status.

Prior to addressing the research questions, it is of direct policy relevance to note that our findings show significant differences in life course offending patterns reflected in criminal legal system contact at the intersection of gender and Indigenous status that are consistent with existing work in this area (Broidy et al., 2015; Matthews et al., 2022). The findings confirm that examining gender differences without considering race/ethnicity (or vice-versa) masks important within group differences in offending patterns. For instance, while males have more extensive criminal histories than females and are more likely than females to populate offending trajectories that reflect more persistent offending patterns, these patterns do not hold when we disaggregate by Indigenous status. In fact, differences in overall offending patterns and trajectory group membership vary significantly across Indigenous and non-Indigenous females. And, although Indigenous females exhibit significantly less offending than Indigenous males, who are the group most likely to populate the highest rate offending trajectories, the offending rates and life course patterns of Indigenous females are very similar to those of non-Indigenous males. In contrast to what is known about non-Indigenous male offending, relatively little is known about Indigenous female offending patterns, and despite their broad similarities, the associated risk factors likely diverge in important ways. Available evidence highlights distinct features of Indigenous female offending patterns, including earlier ages of contact with all levels of the criminal legal system, significant overlap with victimization for violent offenses, and higher proportions of minor offenses, such as public order offenses (Bartels, 2010; Matthews et al., 2022). Further research is needed to better understand these offending patterns, including how they are shaped by systemic biases (e.g., over-policing), historical trauma, victimization, and cumulative disadvantage stemming from colonization.

The overrepresentation of Indigenous males in the high-rate offending trajectories is stark. Indigenous male criminal legal system overrepresentation is well documented and has been linked to the concentration of several established risk factors, including drug/alcohol problems, low educational attainment, adverse childhood experiences, violence exposure, unemployment, family and neighborhood dysfunction, poverty, negative peer group influences and mental disorder (Heffernan et al., 2012; Shepherd et al., 2020; Snowball & Weatherburn, 2006; Weatherburn et al., 2008). These risk factors are universal to offending and not unique to Indigenous males but highlight the cumulative nature of difficulties and disadvantage experienced by this group that ultimately stem from the ongoing effects of colonization, intergenerational trauma and marginalization and systemic racism (Behrendt et al., 2019). Shepherd et al. (2020) demonstrated that the most prominent risk factors evident for Indigenous males who experienced incarceration (compared to those who did not) were poor educational attainment and labor force participation, and drug/alcohol problems. Although there is a need for a whole of system approach to address and prevent Indigenous over-representation in the criminal legal system, there may be a specific need for place-based, local strategies that address the local drivers of educational disengagement and substance use while creating employment opportunities and cultural connectedness. Such strategies should respect calls for self-determination by Indigenous peoples, and be driven by Indigenous leaders, community members and services, building on the strengths of that community. Previous and current government approaches have largely failed in reducing Indigenous over-representation in the criminal legal system and most starkly for Indigenous males, whose offending is most likely to escalate into adulthood.

Consistent with expected results relating to the first research question and Moffitt’s (2006) hypothesis that different offending pathways may bear differential risks for mental health, we found significant variation in mental health service use and mental illness diagnoses across trajectory groups. The two highest rate offending trajectory groups exhibited the highest rates of mental health contacts (i.e., greater than 60%) and the highest rates of psychiatric diagnoses (i.e., approximately one third of individuals). These were also the trajectories with the most notable histories of violent offending, reinforcing links between mental illness and serious offending. Further, these trajectories are overpopulated by Indigenous males, highlighting the too often overlooked role of mental health needs and challenges for males, particularly marginalized males in contact with the criminal legal system.

Our results are consistent with an expanding body of evidence demonstrating that individuals characterized by persistent and serious offending patterns experience the highest rates of mental and physical health problems as they age (Odgers et al., 2007; Piquero et al., 2011; Reising et al., 2019; Skinner et al., 2020; Testa & Semenza, 2020; Walker et al., 2019). Early onset and serious offending pathways have been hypothesized to interfere with the development of core psychosocial competencies, including limiting opportunities for education and employment, that in turn make this group more vulnerable to the development of mental health and substance use problems (Wiesner et al., 2005). The behaviors and patterns of system contact and surveillance associated with persistent and serious offending carry long-term negative consequences for psychological wellbeing, highlighting the importance of childhood and adolescent interventions and supports as a key public health strategy to minimise health problems in a vulnerable segment of the population and reduce burden on the health, welfare, and criminal legal systems.

Our multivariate results suggest that the role of mental illness in increasing the risk of more persistent and serious offending pathways is largely driven by substance use disorders. This finding is consistent with the well-established links between substance use, mental illness, offending and criminal legal system contact. Substance use is likely to contribute to persistence in criminal behavior (Wiesner et al., 2005), which is consistent with our finding of substance use disorders imparting the greatest risk of being classified in the high-rate escalating offending trajectory. For these individuals, substance use may disrupt the maturing out of criminal behavior in late adolescence and early adulthood (Welte et al., 2009) and increase vulnerability for comorbid mental health problems. Additionally, the high rates of substance use and mood disorders observed in the high-rate escalating group may reflect the negative collateral impacts of criminal legal system contacts on psychological wellbeing. The relationships among substance use, offending and mental health appear best characterized as reciprocal and dynamic, where further research is required to understand how these relationships co-evolve over time.

Moving beyond the consistent link between persistent offending trajectories and worse mental health outcomes, we found that those who had the earliest contact with the mental health system were concentrated in trajectory groups characterized by an adolescent peak in offending. Those in both the high-rate late adolescent peak and the adolescent limited trajectory groups on average had their first mental health system contact before age 15 years. It is possible that this overlaps with their offending onset, such that their adolescent offending triggered mental health system intervention. Moreover, these groups exhibit repeat mental health system contact with the high-rate adolescent peak group having over 30 mental health contacts per individual on average and the adolescent limited group having over 15 contacts per person on average. Comparatively, the two low-rate offending groups average well under 10 visits per person on average. These results highlight that the overlap between offending and mental health is evident early in developmental trajectories, which likely reflects shared risk factors (e.g., childhood maltreatment and other adverse events) experienced by a vulnerable segment of the population.

Turning to the second research question, we found that the extent of the link between mental health outcomes and offending trajectories varied across the intersection of gender and Indigenous status. Compared to all other demographic groups, Indigenous females had the highest rates of mental health service contact and diagnosed mental illness. Specifically, for Indigenous females classified in the high-rate escalating offending group, over 90% had contact with mental health services and over 54% had been diagnosed with a mental illness, which were the highest rates for all demographic groups. For Indigenous and non-Indigenous females, individuals classified in the adolescent limited offending group had the highest average mental health service contacts, but with non-Indigenous females having much higher rates of contact on average. In contrast, for Indigenous and non-Indigenous males, the highest average mental health service contacts were observed for the high-rate late adolescent peak offending group. This highlights differences in mental health service use across offending trajectories that are conditional on gender.

Our findings reveal further variation in the patterns linking offending and mental health system contact across gender and Indigenous status. Non-Indigenous females were over-represented in the low-rate trajectory group, which is least likely to have mental health service contacts or psychiatric diagnoses. By contrast, Indigenous women had offending patterns that mirrored those of non-Indigenous men, but with notably higher rates of mental health system contacts on average. For women, then, the overlap between mental health contacts and offending trajectories is significantly more notable among Indigenous populations. Comparatively, Indigenous males had fewer mental health contacts than their Indigenous female counterparts but were significantly more likely to populate the high-rate offending trajectories. These variations across gender and Indigenous status highlight that the overlaps between mental health system contacts and offending patterns are conditional. Findings relating to mental illness diagnoses, mental health service utilization, and criminal legal system contacts in the current study need to be understood within the context of a history of government policies which have impacted both health and criminal legal system involvement of Indigenous people in Australia. Thus, high rates of mental health service contact may be reflecting policy approaches of the period or may be reflecting a true indication of higher rates of mental illness. Moreover, mental illness diagnoses differ substantially from an Indigenous conceptualization of social and emotional and well-being (Calma et al., 2017).

Our findings highlight that further research is needed to better understand the contexts that both exaggerate and minimize the links between offending and mental health. For instance, subsequent work can focus on those with mental health contacts to better understand when these contacts accompany criminal legal system involvement and when they might reduce the likelihood of system contact. This is important because policy and practice often calls for increased mental health services for at-risk populations as a strategy for addressing risks and needs associated with criminal legal system involvement. However, very little is known about the effectiveness and/or appropriateness of mental health interventions for system-involved young people, particularly Indigenous youth. In the Australian context, it will be important to examine the availability and effectiveness of culturally led and operated mental health services in communities that are home to a disproportionate number of young people on a trajectory toward persistent offending and repeated contact with the criminal legal system. Early access to, and engagement in, mental health and wellbeing services, as well as localized support to enhance the wellbeing of the entire community, may be critical in reducing pathways into serious offending. So too is serious consideration of whether, when, and how criminal legal system interventions should be utilized to address risk behaviors that, at least partly, reflect both mental health and broader community resource needs.

The primary strength of our study was the use of a longitudinal population-based cohort with a long follow-up period. Although specific cohort effects may be present, our findings were largely consistent with other studies using GBTM with cohort data (e.g., Ferrante, 2013; Odgers et al., 2008; Valuri et al., 2021). Despite this strength, our study has some limitations. First, we acknowledge the limitations of the GBTM method, where individuals do not actually belong to any group identified by the process. Instead, the groups reflect unobserved patterns in trajectories of offending. A further criticism of GBTM is that it can generate spurious findings (e.g., identifying patterns where none exist; Skardhamar, 2010), necessitating the use several model accuracy evaluation criteria (Mésidor et al., 2022), which was the approach adopted for the current study. Second, the data sources for community mental health service use and mental health disorders (i.e., hospital admissions) do not capture the full extent of service use and mental health disorders. These sources are biased toward more severe and/or acute presentations and do not capture private mental health service use or the bulk of mental health issues that do not result in hospitalization. This bias may partly explain the prominence of substance use disorders as a risk factor for offending in our results, since these presentations are highly prevalent in hospital inpatient settings (Brown et al., 1998; Peterson et al., 2021). Therefore, the extent of mental health service use and disorders should be viewed as conservative. Similarly, the use of court records and police diversions does not capture the full extent of offending, since not all offenses come to the attention of police and/or progress through the court system. Moreover, we know that racialized patterns of state surveillance create biases in criminal legal system contacts that result in disproportionately higher criminal legal system contacts for Indigenous peoples (Cunneen & Tauri, 2019). This means that some of the overrepresentation of Indigenous males and females in our offending data may be more reflective of race/ethnicity-based differences in the likelihood of system contact than in actual differences in offending. We tried to be cognizant of this is our language throughout as well as our interpretation of the patterns we identified. Given the QCRC repository only contains data from Queensland agencies, it was not possible to identify outward migration of the cohort. As a result, the data is likely to under-estimate justice and health system contacts. Finally, we did not examine temporal ordering or sequencing of offending and mental health problems, limiting insights about causality, with these analyses planned for future studies with the cohort.

In conclusion, the novel contribution of this paper is the linkage of mental health service utilization and diagnosis data with data on criminal legal system contacts. Broadly, our findings reinforce that mental health needs are most pronounced among those with the most extensive, serious, and enduring criminal legal system contacts. Moving beyond this, our findings also suggest that these links are most acute for Indigenous women and men, who present with the most mental health system contacts and highest rates of mental illness diagnoses. While non-Indigenous women also have relatively frequent mental health system contacts (like Indigenous men), they are significantly less likely to exhibit the more serious and enduring patterns of criminal legal system contact. We call for more research attention to the ways in which mental health patterns across the life course relate to life course patterns of criminal legal system contact for different demographic groups. Mental health needs and impacts differ across the intersection of gender and race/ethnicity in ways that should inform treatment, and more research is critical to detailing these differences and their related policy and practice implications.

Acknowledgements: The industry partners on the grant supporting this research were Queensland Health; Department of Premier and Cabinet, Office of Economic and Statistical Research (Queensland Treasury, now called the Queensland Government Statistician’s Office [QGSO]); Department of Children, Youth Justice and Multicultural Affairs; Queensland Police; Queensland Department of Justice and Attorney General; and Queensland Registry of Births, Deaths and Marriages. We thank our government partners for helpful comments on a previous version of this paper. The authors gratefully acknowledge use of the services and facilities of the Griffith Criminology Institute's Social Analytics Lab at Griffith University. The views expressed are not necessarily those of the departments or agencies, and any errors of omission or commission are the responsibility of the authors.

Ethics approval and consent to participate: The requirement to obtain informed individual participant consent was waived given the use of historical de-identified administrative data, which was approved by the Griffith University Human Research Ethics Committee. The study was approved by the Griffith University Human Research Ethics Committee (HREC 2010/479). The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation, and with the Helsinki Declaration of 1975, as revised in 2008.

Availability of data and materials: The data for the study are held in Social Analytics Lab (SAL) at Griffith University and used with permission from the relevant data custodians. The linked administrative data used in this study is owned by the respective Queensland Government agencies and access is managed by the Queensland Government Statistician’s Office and cannot be made available to third parties by the authors. The datasets analyzed during the current study are not publicly available due to restrictions placed on the datasets by the data custodians but can be made available upon reasonable request and with permission of the relevant data custodians and the Queensland Government Statistician’s Office. Any researcher interested in accessing the data can submit an application to the SAL management committee ([email protected]) with the relevant support and approvals.

Competing interests: The authors declare that they have no competing interests.

Funding: This research was funded by the Australian Research Council grant number LP100200469. The funder had no role in study design, the collection, analysis and interpretation of data, the writing of the report or the decision to submit the article for publication. JO was supported by a Griffith University Postdoctoral Research Fellowship.

Author contributions: LB, AS, CT, and SD conceived the study, with JO, LB, SD, CT and TA contributing to the study design. Data preparation and analysis were performed by JO. The first draft of the method and results were written by JO. AK and LB wrote the first draft of the introduction. LB and JO wrote the first draft of the discussion. All authors contributed to subsequent versions of the manuscript. All authors read and approved the final manuscript.

References

Albizu-Garcia, C. E., Alegrı́a, M., Freeman, D., & Vera, M. (2001). Gender and health services use for a mental health problem. Social Science & Medicine, 53(7), 865-878. https://doi.org/10.1016/S0277-9536(00)00380-4

Australian Institute of Health and Welfare. (2012). National best practice guidelines for data linkage activities relating to Aboriginal and Torres Strait Islander people. https://www.aihw.gov.au/reports/indigenous-australians/national-best-practice-guidelines-for-data-linkage/contents/table-of-contents

Australian Institute of Health and Welfare. (2019). The health of Australia’s prisoners, 2018. https://www.aihw.gov.au/reports/prisoners/health-australia-prisoners-2018/contents/table-of-contents

Bartels, L. (2010). Indigenous women’s offending patterns: A literature review (Research and Public Policy Series, Issue. Australian Institute of Criminology.

Behrendt, L., Cunneen, C., Libesman, T., & Watson, N. (2019). Aboriginal and Torres Strait Islander Legal Relations. Oxford University Press.

Bergman, L. R., & Andershed, A.-K. (2009). Predictors and outcomes of persistent or age-limited registered criminal behavior: a 30-year longitudinal study of a Swedish urban population. Aggressive behavior, 35(2), 164-178. https://doi.org/10.1002/ab.20298

Broidy, L. M., Stewart, A. L., Thompson, C. M., Chrzanowski, A., Allard, T., & Dennison, S. M. (2015). Life Course Offending Pathways Across Gender and Race/Ethnicity. Journal of Developmental and Life-Course Criminology, 1(2), 118-149. https://doi.org/10.1007/s40865-015-0008-z

Brown, R. L., Leonard, T., Saunders, L. A., & Papasouliotis, O. (1998). The Prevalence and Detection of Substance Use Disorders among Inpatients Ages 18 to 49: An Opportunity for Prevention. Preventive Medicine, 27(1), 101-110. https://doi.org/https://doi.org/10.1006/pmed.1997.0250

Browne, C. C., Korobanova, D., Yee, N., Spencer, S.-J., Ma, T., Butler, T., & Dean, K. (2022). The prevalence of self-reported mental illness among those imprisoned in New South Wales across three health surveys, from 2001 to 2015. Australian & New Zealand Journal of Psychiatry, 0(0), 00048674221104411. https://doi.org/10.1177/00048674221104411

Butler, T., Allnutt, S., Kariminia, A., & Cain, D. (2007). Mental health status of Aboriginal and non-Aboriginal Australian prisoners. Australian and New Zealand Journal of Psychiatry, 41(5), 429-435. https://doi.org/10.1080/00048670701261210

Calma, T., Dudgeon, P., & Bray, A. (2017). Aboriginal and Torres Strait Islander Social and Emotional Wellbeing and Mental Health. Australian Psychologist, 52(4), 255-260. https://doi.org/10.1111/ap.12299

Cauffman, E., Lexcen, F., Goldweber, A., Shulman, E., & Grisso, T. (2007). Gender Differences in Mental Health Symptoms Among Delinquent and Community Youth. Youth violence and juvenile justice, 5(3), 287-307.

Celeux, G., & Soromenho, G. (1996). An entropy criterion for assessing the number of clusters in a mixture model. Journal of Classification, 13(2), 195-212. https://doi.org/10.1007/BF01246098

Cunneen, C., & Tauri, J. M. (2019). Indigenous Peoples, Criminology, and Criminal Justice. Annual Review of Criminology, 2(1), 359-381. https://doi.org/10.1146/annurev-criminol-011518-024630

de Vogel, V., & de Spa, E. (2019). Gender differences in violent offending: results from a multicentre comparison study in Dutch forensic psychiatry. Psychology, Crime & Law, 25(7), 739-751. https://doi.org/10.1080/1068316X.2018.1556267

DeLisi, M., & Piquero, A. R. (2011). New frontiers in criminal careers research, 2000-2011: A state-of-the-art review. Journal of Criminal Justice, 39(4), 289-301. https://doi.org/10.1016/j.jcrimjus.2011.05.001

Drapalski, A. L., Youman, K., Stuewig, J., & Tangney, J. (2009). Gender differences in jail inmates' symptoms of mental illness, treatment history and treatment seeking. Criminal Behaviour and Mental Health : CBMH, 19(3), 193-206. https://doi.org/10.1002/cbm.733

Ferrante, A. M. (2013). Assessing gender and ethnic differences in developmental trajectories of offending. Australian & New Zealand Journal of Criminology, 46(3), 379-402. https://doi.org/10.1177/0004865813490948

Grün, B., & Leisch, F. (2008). FlexMix Version 2: Finite Mixtures with Concomitant Variables and Varying and Constant Parameters. Journal of Statistical Software, 28(4), 1 - 35. https://doi.org/10.18637/jss.v028.i04

Heffernan, E. B., Andersen, K. C., Dev, A., & Kinner, S. (2012). Prevalence of mental illness among Aboriginal and Torres Strait Islander people in Queensland prisons. The Medical Journal of Australia, 197(1), 37-41. https://doi.org/10.5694/mja11.11352

Matsuda, M., Thornberry, T. P., Loughran, T. A., & Krohn, M. D. (2022). Are Late Bloomers Real? Identification and Comparison of Late-Onset Offending Patterns from Ages 14-40. J Dev Life Course Criminol, 8(1), 124-150. https://doi.org/10.1007/s40865-022-00189-9

Matthews, B., McVie, S., Thompson, C., & Stewart, A. (2022). From Childhood System Contact to Adult Criminal Conviction: Investigating Intersectional Inequalities using Queensland Administrative Data. Journal of Developmental and Life-Course Criminology. https://doi.org/10.1007/s40865-022-00204-z

Mésidor, M., Rousseau, M. C., O'Loughlin, J., & Sylvestre, M. P. (2022). Does group-based trajectory modeling estimate spurious trajectories? BMC Med Res Methodol, 22(1), 194. https://doi.org/10.1186/s12874-022-01622-9

Moffitt, T. E. (2006). Life-course-persistent versus adolescence-limited antisocial behavior. In D. Cicchetti & D. J. Cohen (Eds.), Developmental Psychopathology (2nd ed., Vol. 3, pp. 570-598). John Wiley & Sons, Inc.

Muthén, B., & Muthén, L. K. (2000). Integrating Person-Centered and Variable-Centered Analyses: Growth Mixture Modeling With Latent Trajectory Classes. Alcoholism: Clinical and Experimental Research, 24(6), 882-891. https://doi.org/10.1111/j.1530-0277.2000.tb02070.x

Nagin, D., & Tremblay, R. E. (2001). Analysing Developmental Trajectories of Distinct but Related Behaviours: A Group-Based Method. Psychological Methods, 6(1), 18-34. https://doi.org/10.1037/1082-989x.6.1.18

Nagin, D., & Tremblay, R. E. (2005). Developmental Trajectory Groups: Fact or a Useful Statistical Fiction? Criminology, 43(4), 873-904.

Nagin, D. S. (2005). Group-Based Modeling of Development. Harvard University Press.

Nagin, D. S., Jones, B. L., Passos, V. L., & Tremblay, R. E. (2016). Group-based multi-trajectory modeling. Statistical Methods in Medical Research, 27(7), 2015-2023. https://doi.org/10.1177/0962280216673085

Nagin, D. S., & Odgers, C. L. (2010). Group-based trajectory modeling in clinical research. Annu Rev Clin Psychol, 6, 109-138. https://doi.org/10.1146/annurev.clinpsy.121208.131413

Nielsen, J. D., Rosenthal, J. S., Sun, Y., Day, D. M., Bevc, I., & Duchesne, T. (2014). Group-based Criminal Trajectory Analysis Using Cross-validation Criteria. Communications in Statistics - Theory and Methods, 43(20), 4337-4356. https://doi.org/10.1080/03610926.2012.719986

Nowotny, K. M., Kuptsevych-Timmer, A., & Oser, C. (2019). Criminal Justice Contact and Health Service Utilization among Women across Health Care Settings: Analyzing the Role of Arrest. Women's health issues : official publication of the Jacobs Institute of Women's Health, 29(2), 125-134. https://doi.org/10.1016/j.whi.2018.12.005

Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study. Structural Equation Modeling: A Multidisciplinary Journal, 14(4), 535-569. https://doi.org/10.1080/10705510701575396

Odgers, C. L., Caspi, A., Broadbent, J. M., Dickson, N., Hancox, R. J., Harrington, H. L., . . . Moffitt, T. E. (2007). Prediction of differential adult health burden by conduct problem subtypes in males. Archives of General Psychiatry, 64(4), 476-484. https://doi.org/10.1001/archpsyc.64.4.476

Odgers, C. L., Moffitt, T. E., Broadbent, J. M., Dickson, N., Hancox, R. J., Harrington, H., . . . Caspi, A. (2008). Female and male antisocial trajectories: From childhood origins to adult outcomes. Development and Psychopathology, 20(2), 673-716. https://doi.org/http://dx.doi.org/10.1017/S0954579408000333

Ogilvie, J. M., Allard, T., Thompson, C., Dennison, S., Little, S. B., Lockwood, K., . . . Stewart, A. (2021). Psychiatric disorders and offending in an Australian birth cohort: Overrepresentation in the health and criminal justice systems for Indigenous Australians. Australian & New Zealand Journal of Psychiatry, 00048674211063814. https://doi.org/10.1177/00048674211063814

Peterson, C., Li, M., Xu, L., Mikosz, C. A., & Luo, F. (2021). Assessment of Annual Cost of Substance Use Disorder in US Hospitals. JAMA Netw Open, 4(3), e210242. https://doi.org/10.1001/jamanetworkopen.2021.0242

Piquero, A. R., Blumstein, A., Brame, R., Haapanen, R., Mulvey, E. P., & Nagin, D. S. (2001). Assessing the Impact of Exposure Time and Incapacitation on Longitudinal Trajectories of Criminal Offending. Journal of Adolescent Research, 16(1), 54-74. https://doi.org/10.1177/0743558401161005

Piquero, A. R., & Buka, S. L. (2002). Linking juvenile and adult patterns of criminal activity in the Providence cohort of the National Collaborative Perinatal Project. Journal of Criminal Justice, 30(4), 259-272. https://doi.org/10.1016/S0047-2352(02)00128-9

Piquero, A. R., Shepherd, I., Shepherd, J. P., & Farrington, D. P. (2011). Impact of offending trajectories on health: disability, hospitalisation and death in middle-aged men in the Cambridge Study in Delinquent Development. Crim Behav Ment Health, 21(3), 189-201. https://doi.org/10.1002/cbm.810

R Core Team. (2022). R: A language and environment for statistical computing. In (Version 4.2.0) R Foundation for Statistical Computing. https://www.R-project.org/

Ram, N., & Grimm, K. J. (2009). Growth Mixture Modeling: A Method for Identifying Differences in Longitudinal Change Among Unobserved Groups. Int J Behav Dev, 33(6), 565-576. https://doi.org/10.1177/0165025409343765

Rawal, P., Romansky, J., Jenuwine, M., & Lyons, J. S. (2004). Racial Differences in the Mental Health Needs and Service Utilization of Youth in the Juvenile Justice System. The Journal of Behavioral Health Services & Research, 31(3), 242-254. https://doi.org/10.1007/BF02287288

Reising, K., Ttofi, M. M., Farrington, D. P., & Piquero, A. R. (2019). Depression and anxiety outcomes of offending trajectories: A systematic review of prospective longitudinal studies. Journal of Criminal Justice, 62, 3-15. https://doi.org/10.1016/j.jcrimjus.2018.05.002

Shepherd, S. M., Spivak, B., Ashford, L. J., Williams, I., Trounson, J., & Paradies, Y. (2020). Closing the (incarceration) gap: assessing the socio-economic and clinical indicators of indigenous males by lifetime incarceration status. BMC Public Health, 20(1), 710. https://doi.org/10.1186/s12889-020-08794-3

Skardhamar, T. (2010). Distinguishing facts and artifacts in group-based modeling. Criminology, 48(1), 295-320. https://doi.org/10.1111/j.1745-9125.2010.00185.x

Skinner, G. C., Farrington, D. P., & Shepherd, J. P. (2020). Offender trajectories, health and hospital admissions: relationships and risk factors in the longitudinal Cambridge Study in Delinquent Development. J R Soc Med, 113(3), 110-118. https://doi.org/10.1177/0141076820905319

Snowball, L., & Weatherburn, D. (2006). Indigenous over-representation in prison: The role of offender characteristics. BOCSAR NSW Crime and Justice Bulletins, 20.

Steffensmeier, D., Feldmeyer, B., Harris, C. T., & Ulmer, J. T. (2011). Reassessing trends in Black violent crime, 1980–2008: Sorting out the “Hispanic effect” in uniform crime reports arrests, national crime victimization survey offender estimates, and U.S. prisoner counts. Criminology: An Interdisciplinary Journal, 49(1), 197-251. https://doi.org/10.1111/j.1745-9125.2010.00222.x

Stewart, A., Ogilvie, J. M., Thompson, C., Dennison, S., Allard, T., Kisely, S., & Broidy, L. (2021). Lifetime prevalence of mental illness and incarceration: An analysis by gender and Indigenous status. Australian Journal of Social Issues, 56(2), 244-268. https://doi.org/https://doi.org/10.1002/ajs4.146

Teplin, L. A., Abram, K. M., McClelland, G. M., Washburn, J. J., & Pikus, A. K. (2005). Detecting Mental Disorder in Juvenile Detainees: Who Receives Services. American Journal of Public Health, 95(10), 1773-1780.

Testa, A., & Semenza, D. (2020). Criminal offending and health over the life-course: A dual-trajectory approach. Journal of Criminal Justice, 68. https://doi.org/10.1016/j.jcrimjus.2020.101691

Valuri, G. M., Morgan, F., Ferrante, A., Jablensky, A., & Morgan, V. A. (2021). A comparison of trajectories of offending among people with psychotic disorders, other mental disorders and no mental disorders: Evidence from a whole-of-population birth cohort study. Crim Behav Ment Health. https://doi.org/10.1002/cbm.2204

van de Schoot, R., Sijbrandij, M., Winter, S. D., Depaoli, S., & Vermunt, J. K. (2016). The GRoLTS-Checklist: Guidelines for Reporting on Latent Trajectory Studies. Structural Equation Modeling: A Multidisciplinary Journal, 24(3), 451-467. https://doi.org/10.1080/10705511.2016.1247646

Walker, G. H., Boden, J. M., Fergusson, D. M., & Horwood, L. J. (2019). Examining the associations between offending trajectories in adolescence/young adulthood and subsequent mental health disorders. Journal of Criminal Justice, 62, 94-100. https://doi.org/10.1016/j.jcrimjus.2018.09.008

Walter, M., Lovett, R., Maher, B., Williamson, B., Prehn, J., Bodkin‐Andrews, G., & Lee, V. (2020). Indigenous data sovereignty in the era of big data and open data. Australian Journal of Social Issues, 56(2), 143-156. https://doi.org/10.1002/ajs4.141

Weatherburn, D., Snowball, L., & Hunter, B. (2008). Predictors of Indigenous Arrest: An Exploratory Study. Australian & New Zealand Journal of Criminology, 41(2), 307-322. https://doi.org/10.1375/acri.41.2.307

Welte, J. W., Barnes, G. M., Hoffman, J. H., Wieczorek, W. F., & Zhang, L. (2009). Substance Involvement and the Trajectory of Criminal Offending in Young Males. The American Journal of Drug and Alcohol Abuse, 31(2), 267-284. https://doi.org/10.1081/ada-47934

Wiesner, M., Kim, H. K., & Capaldi, D. M. (2005). Developmental Trajectories of Offending: Validation and Prediction to Young Adult Alcohol Use, Drug Use, and Depressive Symptoms Development and Psychopathology, 17, 251-270.

World Health Organization. (2004). ICD-10: International statistical classification of diseases and related health problems: tenth revision (2nd ed.). World Health Organization.

Supplementary information for “Trajectories of offending and mental health service use: Similarities and differences by gender and Indigenous status in an Australian birth cohort”

Supplementary Table S1. Offence categories by the Australian Standard Offence Classification; Queensland Extension.

Notes: nec = not elsewhere classified.

Supplementary Table S2. Descriptive information for offending counts by age for individuals with at least one offence in the 1990 cohort, aged 10-24 (n = 12,690).

Supplementary Table S3. Model selection goodness-of-fit and classification accuracy statistics for one to six group models of offending.

AvePP = Average posterior probability of group classification for most likely group membership; OCC = Odds of correct classification. ‘-’ = not applicable due to single group. Selected model in bold.

Table S4. Five-group offending trajectory model coefficients, classification diagnostic statistics and descriptive information (n = 12,690).

Figure S1. Rootogram of posterior probabilities for five group trajectory model of offending.

Figure S2. Individual trajectory profiles for offences separated by group membership.

Supplementary Table S5. Correlations (Pearson’s r) among variables included in the multinomial regression analyses.

Note: MD = mental disorder; MH = mental health; SMI = serious mental illness; SUD = substance use disorder.

* p <.05, ** p <.01, *** p <.001