Resilience profiles predict polysubstance use in adolescents with a history of childhood maltreatment

Abstract

Background.

Childhood maltreatment (CM) can be an impediment to normative development and consistently predicts increased risk for substance misuse and polysubstance use (polySU). Yet, a subset of individuals who experience CM exhibit successful adaptations across the lifespan. Although there is an expansive literature on socioemotional and cognitive protective factors that mitigate impacts of CM, less is known about other, intra-individual resilience-promoting factors (e.g., positive future orientation) known to assuage high-risk SU patterns during adolescence.

Method.

This study examined heterogeneity in individual-level resilience characteristics in maltreated youth as it related to CM characteristics and SU patterns during adolescence. Participants included maltreated youth from the longitudinal LONGSCAN sample (N=355; 181 females). Latent Profile Analysis was used to identify subgroups of CM-exposed individuals based on 5 resilience indicator variables (i.e., commitment to goals, engaging in demanding activities, self-reliance, positive future orientation, and externalizing behaviors). Tests for differences in SU patterns and CM characteristics between the resultant profiles were performed.

Results.

Data models revealed 3 latent profiles based on participants’ resilience traits (i.e., Low Resilience, Average Resilience, and High Resilience). There were no profile differences on the basis of CM characteristics. Those in the High Resilience profile were less likely to engage in polySU compared to the Average Resilience profile.

Implications.

These findings highlight the promise of individual-level resilience factors that are not necessarily dependent upon caregiver or environmental inputs as protective against polySU following CM. This work represents a promising avenue for future preventative intervention efforts targeting emergent SU behaviors in high-risk youth.

Introduction

Childhood maltreatment (CM) is pervasive in the US, with 1 in 7 children experiencing abuse or neglect in the past year (U.S. Department of Health & Human Services, Administration for Children and Families, 2020). CM represents a significant risk factor for poor health sequelae across the life span, including substance use (SU) and SU disorders (SUD; Buckingham & Daniolos, 2013). Adolescents who experience CM are more likely to initiate SU earlier, escalate SU more rapidly, eventually meet criteria for a SUD, and experience relapse following SU treatment compared to non-maltreated peers (Alvarez-Alonso et al., 2016; Mills et al., 2014; Shin et al., 2013). A burgeoning literature also suggests that youth who experience CM are at greater risk of polysubstance use (polySU; Rivera et al., 2018), which is associated with amplified risk for SUD (Moss et al., 2014). PolySU is typically defined as the use of more than one substance concurrently, or in developmental samples, the reported general use of more than one substance (e.g., (Moss et al., 2014). In addition, polySU represents a major risk factor for mental and physical health issues (Hakansson et al., 2011; Jones et al., 2017; Timko et al., 2018), making it a pertinent behavioral pattern to investigate in those at elevated risk for SU outcomes. Thus, early SU initiation, severity of use, and polySU each represent high-risk patterns of SU with potential to portend poor outcomes, particularly in youth who have experienced CM.

Despite the major mental health risks that CM poses, some individuals appear resilient to the impacts of CM and achieve an adaptive level of functioning (Meng et al., 2018). There are numerous resilience factors that mitigate SU outcomes following CM, including adaptive psychological functioning (Goldstein et al., 2013), cognitive factors (Burt & Paysnick, 2012), as well as social and support factors (Shin et al., 2019). However, less is known about individual-level variables as potential protective factors against maladaptive outcomes that may not be as dependent upon cognitive or environmental factors. Specifically, burgeoning work has highlighted individual-level traits such as perseverance (Yoon et al., 2020), self-reliance (Cicchetti & Rogosch, 2009), and positive future orientation (e.g., Cui et al., 2020), as individual-level resilience factors that promote healthy outcomes following CM. Taken together, existing literature on resilience-promoting factors in youth who have experienced CM highlights the highly multi-faceted nature of resilience that few studies have captured to date.

Person-Centered Approaches to Resilience

Critically, prior research has largely focused on comparing mean levels of resilience factors in youth with and without CM, or composite scores of resilience factors used to predict health outcomes. However, there is a growing literature that has used person-centered statistical approaches (e.g., latent profile/class analyses) with the goal of characterizing resilience following CM with a more process-oriented lens rather than as a singular trait (Green et al., 2021; Yoon et al., 2022). For example, using latent profile analysis (LPA) methods in a sample of 12-year old children with suspected CM, 5 distinct profiles varying in resilience were derived from multiple domains of functioning (i.e., child competence, behavioral problems, as well as family and neighborhood traits) (Martinez-Torteya et al., 2017), highlighting the highly heterogenous nature of adaptive functioning following CM. Similarly, in samples of older adolescents and emerging adults youth with CM histories, person-centered analytic approaches revealed 4 distinct profiles with a complex interplay of individual-level, as well as school/occupational competence, and interpersonal competence (Russotti et al., 2020; Yates & Grey, 2012). These person-centered studies underscore the utility of examining resilience across multiple domains to garner a more nuanced perspective of more or less adaptive patterns of functioning following CM.

Thus far, existing studies have yet to employ a person-centered approach to resilience to examine mitigation against high-risk patterns of SU during adolescence, when risky SU is most likely to emerge and forecast subsequent SUDs. Moreover, to our knowledge, very few studies have sought to quantify the extent to which characteristics of maltreatment experiences relate to profiles of resilience (see Yates & Grey, 2012). This is a crucial gap to be filled, as CM characteristics such as maltreatment, type, timing, and severity have consistently been highlighted as potent factors impacting youth outcomes following CM, for better or for worse (Kwak et al., 2018; Manly et al., 2001). Therefore, building upon this prior literature, the current study sought to use an LPA approach to quantify individual-level resilience traits that have previously been highlighted as being linked to adaptive functioning, including lower SU rates, following CM. In what follows, we briefly review literature on resilience traits examined in the present investigation.

Individual-level Resilience Characteristics

Perseverance

Prior research has suggested that perseverance, or commitment to activities and goals, is protective against psychopathology following CM (Easterlin et al., 2019; Kwak et al., 2018). There is a longstanding literature demonstrating that, following adversity, youth who engage in extracurricular activities such as those measured in the current study (e.g., sports, arts, school organizations; Easterlin et al., 2019) exhibit reductions in delinquency as well as trauma and mood disorder symptoms (Kwak et al., 2018). Moreover, prior work in highly at-risk youth has suggested that engagement in specific activities (i.e., participation in student government and arts) is linked to lower levels of SU and reductions in SU over a 6 year period (Fauth et al., 2007). Thus, it stands to reason that measuring engagement and achievement may be a suitable measure of commitment to extracurricular activities known to mitigate poorer outcomes following adversity. Another aspect of perseverance examined here was challenging oneself (i.e., engaging in demanding activities), which predicts less SU following childhood adversity (Goldstein et al., 2013). That is, prior research in adults with CM histories has shown that greater levels of perseverance provide buffering effects against nicotine dependence (Goldstein et al., 2013).

Self-reliance

Previous work has suggested that self-reliance may be a critical individual-level factor for adaptive functioning following CM (Cicchetti & Rogosch, 2009). Although associations between self-reliance and SU following CM have not been previously examined, prior theoretical and empirical work suggest that it may be a positive coping strategy that fosters resiliency following CM (Flett et al., 2012). In particular, research examining self-reliance as one dimension of resilience following CM has shown that elevated self-reliance promotes more adaptive coping strategies (Flett et al., 2012), which tend to be related to lower rates of SU in youth (Cardoso, 2018; Lee-Winn et al., 2018; McConnell et al., 2014).

Positive future orientation

Future orientation is conceptualized as a multidimensional construct comprised of attitudes, judgments, and planning about one’s future (e.g., Schmidt et al., 1978). Positive future orientation is a potent predictor of adaptive functioning in multiple domains among maltreated adolescents (Cui et al., 2020; Zinn et al., 2020). Maltreated youth with a positive future orientation are less likely to have poor mental health, engage in SU and risky sexual behaviors, or have externalizing problems (Cabrera et al., 2009; Cui et al., 2020; Zinn et al., 2020). Critically, a previous longitudinal study demonstrated that future orientation is not a static trait; it has emergent and dynamic properties, suggesting malleability and, thus, potential as an intervention target (Oshri et al., 2018).

Externalizing behaviors

In concert with perseverance and positive future orientation, having fewer externalizing behaviors predicts adaptive functioning following CM (e.g., Cui et al., 2020; Oshri et al., 2011; Sasser et al., 2019). For example, multiple studies have underscored externalizing behaviors as a mediator, moderator, and outcome of associations between future orientation and mental health following CM (Cabrera et al., 2009; Cui et al., 2020). Namely, positive future orientation predicts fewer externalizing problems following CM. Moreover, externalizing problems have been established as one mechanism potentiating SU and SUDs following CM (for review, Kirsch et al., 2020). Thus, lower levels of externalizing behaviors may represent a key resilience trait, with the potential to mitigate SU in youth.

Present Study

Resilience can be conceptualized as a construct that is process-oriented and not singularly determined (Masten & Cicchetti, 2016), with growing evidence for the utility of person-centered approaches in capturing these processes in youth uniquely vulnerable to high-risk SU behaviors and SUDs (e.g., Green et al., 2021; Yates & Grey, 2012; Yoon et al., 2022). To this end, the current study sought to address existing gaps in the literature by examining individual-level resilience factors that have been both theoretically and empirically linked to mitigating SU/SUD in this population. Specifically, the present study used LPA to determine unique profiles based on heterogeneity in resilience characteristics in a cohort of maltreated youth from the Longitudinal Studies on Child Abuse and Neglect (LONGSCAN; Runyan & Dubowitz, 2014). In Figure 1, we provide a conceptual illustration of the LPA model (Fig 1a) as well as an overarching model of pathways toward or away from risky patterns of SU behavior following maltreatment (Fig 1b). Latent profiles of CM-exposed individuals were based on 5 resilience indicator variables (Fig 1a): 1) positive future orientation, 2) commitment to activities/leadership, 3) engaging in demanding activities, 4) self-reliance, and 5) externalizing behaviors. We examined whether profiles differed in high-risk patterns of SU (i.e., SU initiation prior to age 15, SU severity, and polySU (Hingson et al., 2003; Moss et al., 2014). Finally, we evaluated whether profiles differed in their CM characteristics, including type, timing, and severity, as these factors may be related to individuals’ resiliency; for instance, it may be the case that more severe maltreatment exposure results in lower levels of resilience, as others have demonstrated (see Leonard & Gudiño, 2021). It was generally hypothesized that the LPA would uncover 3–5 distinct resilience profiles, with individuals in profiles with the highest levels of resilience indicators exhibiting the lowest levels of high-risk SU behaviors and CM.

Figure 1.

Pathways toward and away from risky substance use patterns following child maltreatment. Fig 1a depicts the indicator variables (blue boxes) that comprise the latent resilience variable (green circle) in the latent profile analysis (LPA). Fig 1b provides an overarching conceptual model of pathways that individuals who have experienced child maltreatment (CM) may follow. The red arrows indicate a risk pathway toward high-risk patterns of substance use (SU; e.g., polySU, early SU, etc.), whereby experiences of CM predict risk of elevated internalizing and externalizing symptomology (Cicchetti & Handley, 2019), which in turn propels risk for substance misuse patterns. Conversely, the green arrow pathway indicates that resilience factors including those examined in Fig 1a are expected to buffer (grey arrow) against high-risk patterns of SU following CM.

Methods

Participants

The current subsample examined was drawn from the larger multisite LONGSCAN study (Runyan et al., 2014; dataset 170). The LONGSCAN study was completed from 1991 – 2012 through funding from the National Center on Child Abuse and Neglect. The central goal of the study was to identify the etiological processes and consequences of child maltreatment (CM). To achieve this, five data collection sites across the United States collected child, family, and community level factors relating to etiology and impacts of CM, with the coordinating center at the University of North Carolina Chapel Hill. Data were collected from children and families identified as at-risk for maltreatment (e.g., through screening at pediatric clinics and Child Protective Services allegations) at study sites located in the East, Midwest, South, Southwest, and Northwest regions of the United States. To date, the overall study has prospectively followed 1,354 children from ages 4 – 18, with caregiver and child interviews conducted every 2 years. Each LONGSCAN site obtained study approval from their respective institutional review boards and informed consent and/or assent from all child and caregiver participants. Complete details of study methods can be found in previous publications (Runyan et al., 2014).

To be included in the present subsample, participants had to satisfy multiple criteria. First, participants had to have at least 1 Child Protective Services (CPS) report of CM and a maltreatment subtype coded (N= 934). Second, to ensure that participants were meaningfully characterized in the LPA analysis, they had to have at least 4 of the 5 resilience indicator variables completed (final sample N = 355; ~51% females; Table 2), which were subsequently included in the latent profile analysis (Table 1). Importantly, participants from the larger sample (N=934) did not systematically differ from participants in the subsample (N=355) on the basis of gender (χ²_(1,N=934)= 0.11, p = .75), race (χ²_(1,N=934)= 5.88, p = .44), or number of CPS reports (t(932) = −1.28, p = .20). Participants in the larger sample were a few months older at the time of study entry compared to subsample examined here (t(932) = 8.58, p < .001; full sample M_age = 4.64, SE = 0.03; subsample M_age = 4.29, SE = 0.02). Participants from the larger sample and the final subsample did not differ on any of the 5 LPA indicator variables; future expectations: t(548) = −0.38, p = .70; leadership and honors: t(612) = 1.84, p = .07; self-reliance: t(448) = 0.78, p = .43; engaging in demanding activities: t(448) = 0.02, p = .98; externalizing behaviors: t(585) = 0.82, p = .24. In addition, on all but one of the SU measures (i.e., polySU ever), participants from the larger sample and the subsample were matched; early SU: χ²_(1,N=934) = 0.84, p = .36; SU severity: t(932) = −0.75, p = .46; persistent polySU: χ²_(1,N=934) = 2.65, p = .10. Participants from the larger sample had a smaller proportion of polysubstance users (37%) compared to the current subsample (49%), χ²_(1,N=543) = 12.03, p = .001. Approximately 80% of participants had 1+ substantiated report of maltreatment (range=1–32;M=4.02,SE=0.23). Although the sample included children with unsubstantiated reports of CM, there is long-standing evidence that physical and psychological consequences of CM do not differ for substantiated and unsubstantiated reports (Kohl et al., 2009; Kugler et al., 2019).

Table 2.

Participant Demographics

Variable	Mean(SD); Range
Age at first CPS referral: M(SE); range	1.62(0.12); 0 – 15.87
T1 Age: M(SE)	4.32 (0.02)
T2 Age: M(SE)	6.23 (0.02)
T3 Age: M(SE)	8.25 (0.02)
T4 Age: M(SE)	12.31 (0.02)
T5 Age: M(SE)	14.37 (0.03)
T6 Age: M(SE)	16.37 (0.02)
T7 Age: M(SE)	18.47 (0.04)
Sex F:M	181:174
Race: W:B:L:NA:A:M:O	120:134:32:3:3:61:2

Note. W = White, B = Black, L = Latino/a, NA = Native American, A = Asian, M = Mixed Race, O = Other

Table 1.

Descriptive Statistics for LPA Indicator Variables

Measure	N	Age Assessed	Mean(SE)	Sample Range
Future Expectations	350	16	30.28 (.25)	15 – 40
Leadership and Honors	353	18	3.65 (.18)	0 – 16
Self-Reliance	355	18	19.56 (.25)	6 – 30
Demanding Activities	355	18	10.92 (.18)	4 – 20
Externalizing Behaviors	332	18	9.45 (.39)	0 – 35

Measures

Childhood maltreatment

CPS maltreatment narrative records were coded by LONGSCAN researchers using the Modified Maltreatment Classification System (MMCS) every 2 years from 0 to 18 years of age (English et al., 1997). The MMCS quantifies CM characteristics, such as type (i.e., physical abuse (55% reported; M severity=1.22,SE=.07), sexual abuse (32%;M severity=1.07,SE=.08), psychological/emotional abuse (58%; M severity=2.13, SE=.11), and neglect (90%;M severity=3.40,SE=.08)), timing, and severity (i.e., ranging from 1=less severe to 6= most severe abuse or neglect).

We considered several dimensions of CM, including type, maximum severity/type, and the total number of CPS records (M=5.43,SE=0.24;range =1–21) across all time points (ages 0–16)¹. To assess the effects of timing of CM, we examined whether participants had reports for each CM type earlier (ages 0–8) or later (9–16) in development².

Resilience variables

Table 1 includes descriptive information for each resilience indicator variable and Figure 1a illustrates a conceptual model of the resilience variables. More detail regarding indicator variables can be found in the Supplement.

Future expectations.

The Future Events Questionnaire (FEQ; (Knight et al., 2008; McCabe & Barnett, 2000) assessed adolescents’ future planning and expectations about employment, education, and family. Adolescents answered 10 questions regarding the likelihood that each outcome would occur in their future (1=very unlikely to 5=very likely). Only the education/career and financial stability items were used to assess future expectations since the family-related items yielded low variability (i.e., 69% endorsed wanting to get married and/or have children; (range α=.65–.72) and were summed to obtain an overall future expectations measure.

Leadership and honors.

A well-validated measure of perseverance is not yet available. Therefore, we considered multiple indicators to provide an approximation of this construct. Since specific achievements may be one tangible way to measure commitment and, therefore, perseverance, we used these measures to discern individuals’ commitment to activities over time. This included measures of how many leadership roles or how many honors/awards participants received (e.g., for sports, schoolwork, community engagement, work or extracurricular activities). To assess commitment to school and extracurricular activities, we used the Community Connectedness measure to examine if they had received recognition through leadership positions or honors in the prior year (Knight et al., 2009). A sum of the quantity of leadership activities and honors was examined.

Self-reliance and engaging in demanding activities.

The Adolescent Coping Orientation for Problem Experiences questionnaire (ACOPE; McCubbin et al., 2012) evaluated coping behaviors. The ACOPE includes 12 subscales of strategies for managing stress and difficult situations. Adolescents were asked how frequently (1=never to 5=most of the time) they engage in each behaviors. Here, the self-reliance and engaging in demanding activities subscales were used (α=.67, .74, respectively). Subscales were entered as separate indicators into the LPA.

Externalizing behaviors.

Externalizing behaviors were assessed with the Youth Self-Report measure (YSR; Achenbach et al., 2001). We included raw self-report sum scores for externalizing behaviors (α=.88).

Substance use

Self-report SU measures were collected at ages 12, 14, 16, and 18. A version of the Tobacco Alcohol and Drugs questionnaire (TAD; Knight et al., 2009) was administered at all ages except 14. The TAD assesses use across a wide range of commonly used substances (e.g., cigarettes, alcohol, methamphetamine, cannabis, cocaine, hallucinogens, heroin, prescription drugs not prescribed by a doctor; (Brener & Kinchen, 2002). We included items that captured whether participants reported past year SU and how many days they used each substance in that year. The TAD was not available at age 14, instead the Diagnostic Interview Schedule for Children (DISC-IV; Shaffer et al., 2000) was administered and captured how often adolescents used substances in the past year. The percentage reporting SU at each age are reported in supplementary figure S1.

We computed multiple indices to capture SU including: 1) SU severity (i.e., frequency of drug use/past year at all ages/time points assessed (M=9.19, SE=0.66; range=0–63) 2) early SU (i.e., SU at 12 or 14 (0=no SU or 1=SU; 38% early use); 3) polySU (i.e., SU for >1 drug/time point (0=use of ≤1 or 1=use of >1 drugs; 51% polySU); 4) persistent polySU (i.e., use of more than 1 drug at 2+ time points (0= polySU ≤ 1 time points or 1=poly SU ≥2 time points; 21% polySU).

Analytic Plan

Missing data ranged from 0% to 6.5% for all variables. Individuals were included only if they were missing 1 or less indicators. We conducted the LPA with and without missing data estimation using full-information maximum likelihood (FIML) and the same conclusions were reached. The LPA was conducted using Mplus v.7.4 (Muthén & Muthén, 2015). We examined several established criteria to identify the optimal solution. To compare model fit, we used Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample size-adjusted BIC with lower values indicating improved model fit (Nylund et al., 2007). The Lo-Mendell-Rubin (LMR) likelihood ratio test compared whether the current number of profiles (k) is a superior fitting model than the model with one fewer profile (k – 1; (Wang & Wang, 2012). Relative entropy for each model was used to determine how accurately the model delineated separate profiles, with higher values indicating better accuracy (Celeux & Soromenho, 1996). Moreover, to ensure that the level of each indicator meaningfully differed across profiles, we used one-way ANOVAs with post-hoc testing to examine whether there were between-profile differences.

Final profiles were compared on patterns of SU and CM, respectively, using the Auxiliary (BCH or DCAT for continuous and categorical outcome variables, respectively) function in MPlus (Asparouhov & Muthén, 2014; Bolck et al., 2004). Both methods estimate an omnibus result based upon profile-specific differences and conduct pairwise comparisons.

Results

Resilience Latent Profiles

The fit statistics for models containing 1 to 5 profile solutions are shown in Table 3. Although the AIC, BIC, and adjusted BIC all decreased with a greater number of profiles, the LMR p-value indicated the 3-profile solution was optimal. In addition, the relative entropy indicated that the 3-profile solution provided the most accurate classification. The estimated means for this solution for each indicator variable contributing to the latent resilience construct are shown in Figure 2. Profile 1 (10% of the sample) was labeled the Low Resilience group as it was characterized by the highest level of externalizing behaviors (M=10.49, SE=1.38) and the lowest levels of expectations about their future (M=27.39, SE=1.12), leadership and honors (M=2.83, SE=0.67), self-reliance (M=11.48, SE=1.05), engaging in demanding activities (M=6.81, SE=6.81). Profile 2 (69%), labeled as the Average Resilience group, was characterized by lower levels of externalizing behaviors (M=9.62, SE=0.54) than the Low Resilience profile and scores that were in between the Low and High Resilience profiles for expectations about their future (M=30.19, SE=0.40), leadership and honors (M=3.05, SE=0.27), self-reliance (M=19.20, SE=0.37), and engaging in demanding activities (M=10.31, SE=0.24). Profile 3 (20%), the High Resilience group, was characterized by the lowest levels of externalizing behaviors (M=8.35, SE=0.84) and the highest levels of positive behaviors; future expectations (M=32.00, SE=0.56), leadership and honors (M=5.92, SE=0.56), self-reliance (M=24.58, SE=0.60), and engaging in demanding activities (M=14.83, SE=0.74).

Table 3.

Summary of model fit for latent profile models

Profiles:	AIC	BIC	Adjusted BIC	Relative Entropy	LMR p	Proportion of smallest group
1	9766.678	9805.400	9773.675	--	--	--
2	9647.692	9709.646	9658.887	0.598	0.0110	.366
3	9595.832	9681.019	9611.225	0.781	0.0111	.101
4	9555.430	9663.849	9575.021	0.736	0.1290	.099
5	9541.723	9673.375	9565.513	0.755	0.8732	.096

Note. AIC = Akaike Information Criterion, BIC = Bayesian Information Criterion, and LMR = Lo-Mendell-Rubin likelihood ratio test. 3-profile solution is highlighted in grey.

Figure 2.

Estimated Means of Indicator Variables for 3-profile Latent Profile Model. Estimated means for the Low, Average, and High Resilience profiles from the 3-profile solution. FE = Future Expectations, LH = Leadership and Honors, EXT = externalizing behaviors, SR = Self-reliance, DA = Demanding Activities. Significant (i.e., p < .016) group differences from post-hoc analyses are indicated above each bar with the following letters: L = different from Low Resilience, A = different from Average Resilience, and H = different from High Resilience.

Note to journal: Intended to be reproduced in color on Web

To determine whether profiles were statistically distinguishable on each indicator, we conducted a one-way ANOVA with post-hoc analyses. Findings revealed significant differences between groups for 4 of 5 indicators: future expectations (F_(2,348)=14.10,p<.001), leadership/honors (F_(2,351)=26.76,p<.001), self-reliance (F_(2,353)=293.45,p<.001), and demanding activities (F_(2,353)=200.90,p<.001). There were no significant differences between any of the groups for the externalizing behaviors indicator (F_(2,330)=1.15,p=.32). Removing the externalizing behaviors indicator did not improve model fit and no solution remained optimal or accurate according to the LMR and relative entropy metrics.

Post-hoc analyses were conducted using Bonferroni corrected adjusted alpha levels (i.e., p<.016). For future expectations, Low Resilience had lower future expectations compared to Average Resilience (p<.001) and High Resilience (p<.001). Similarly, Average Resilience had lower future expectations than High Resilience (p=.01). For leadership and honors, the Average and Low Resilience profiles did not differ. However, the High Resilience profile engaged in higher levels of leadership/honors than the other profiles (p<.001). Self-reliance differed among all profiles. Low Resilience had less self-reliance compared to the Average and High Resilience profiles (p<.001). The High Resilience profile also had more self-reliance than the Average Resilience profile (p<.001). Similarly, for demanding activities the Low Resilience profile had the lower levels compared to the Average and High Resilience profiles (p<.001). The High Resilience profile engaged in more demanding activities than the Average Resilience profile as well (p<.001).

Resilience Profile Comparisons of SU and Maltreatment

Consistent with study hypotheses, significant differences were observed between the profiles for two SU outcome variables (see Figure 3a). Regarding lifetime polySU (χ²_{(2, N =355)}=5.77, p=0.05), participants in the High Resilience profile were less likely to engage in polySU than those in the Average Resilience (χ²_(1,N=319)=5.01, p=0.02) but not the Low Resilience (χ²_{(2,N =108)}=3.51, p=0.06) profiles. The Average Resilience and Low Resilience profiles did not differ on polySU (χ²_(2,N=283)=0.14, p=0.71). Persistent polySU was also significantly different between profiles (χ²_(2,N=355)=7.34, p=0.03; see Figure 3b). Specifically, the High Resilience profile was less likely to engage in persistent polySU compared to the Average Resilience profile (χ²_(1,N=319)=7.62, p=0.02). The Low Resilience group was not different in persistent polySU compared to other classes. Resilience profiles did not differ on SU severity, early SU, number of CPS records, maltreatment type, timing, or severity (see Table 4), which was not consistent with study hypotheses that more severe SU and histories of CM would relate to lower resilience.

Figure 3.

Between-Profile Differences in Lifetime and Persistent PolySU. Fig3a. Polysubstance Use Ever. Fig 3b. Persistent Polysubstance Use. Note. * = p < .05.

Table 4.

Between-profile comparisons for maltreatment type, severity, and timing

Measure	χ2	df	p
Number of CPS records	2.13	2	0.34
Type
Physical Abuse	0.01	2	0.99
Sexual Abuse	1.63	2	0.44
Emotional Abuse	0.60	2	0.74
Neglect	0.54	2	0.76
Severity
Physical Abuse Max Severity	0.66	2	0.72
Sexual Abuse Max Severity	2.74	2	0.25
Emotional Abuse Max Severity	0.20	2	0.91
Neglect Max Severity	1.41	2	0.49
Timing
Early Physical Abuse	0.06	2	0.97
Early Sexual Abuse	3.18	2	0.20
Early Emotional Abuse	0.32	2	0.85
Early Neglect	0.65	2	0.72

Discussion

The present study examined the extent of variability in individual-level resilience characteristics in a sample of adolescents with a history of CM. We examined the heterogeneity in resilience behaviors and how patterns of resilience were was associated with higher-risk SU behavior. We identified three resilience profiles: Low Resilience, Average Resilience, and High Resilience. Profiles differed on all resilience indicators with the exception of externalizing behaviors. The comparable levels of externalizing behaviors across groups is consistent with literature demonstrating that disruptions in self-regulation and the development of externalizing behaviors are lasting adverse impacts of CM (Herrenkohl & Herrenkohl, 2007; Mills et al., 2013). Despite similar levels of externalizing across profiles, in our iterative testing of different combinations of indicators to find the best-fitting LPA models, including externalizing behaviors was necessary to uncover an optimal solution. This suggests that, despite being comparable across profiles, the way in which externalizing behaviors interacted with other indicators in each profile may be critical for delineating those processes that drive resilience in this population. It is also plausible that despite disruptions in self-regulation, CM survivors who evince more resiliency are able to compensate with other competencies.

Our second aim was to examine whether resilience profiles were associated with high-risk SU behaviors. High Resilience adolescents were less likely to engage in polySU compared to Average Resilience adolescents. These SU outcomes are consistent with prior studies in CM samples demonstrating that higher levels of resilience factors examined predicts less SU in adolescents (Bethell et al., 2014; Cabrera et al., 2009; Cui et al., 2020; Goldstein et al., 2013; Zinn et al., 2020). Thus, using a more comprehensive and person-centered approach, the present study suggests that specific individual-level factors previously studied separately work synergistically to protect against polySU. These resilience findings are especially encouraging given that the current sample reported particularly high levels of lifetime polySU (51%) in comparison to what has been reported in prior work. That is, previous literature has reported polySU rates of anywhere from 2% to 51% of youth and young adults with histories of CM (Rivera et al., 2018; Sadeh et al., 2021; S. Shin et al., 2010; Snyder & Smith, 2015). This wide range in polySU rates is likely due to methodological differences, including how polySU is conceptualized (e.g., same session polySU or “co-use” versus lifetime reports of using multiple substances).

Interestingly, the Low Resilience adolescents did not differ from the Average and High Resilience profiles in overall SU. However, it is notable that the pattern of polySU was in the same direction and was statistically comparable to the Average Resilience profile (higher levels of polySU than no polySU). It is plausible that the characteristics of individual-level resilience that we examined play a more protective role against polySU at greater levels, as in the High Resilience profile. In contrast, among those with the lowest levels of resilience traits, the constructs measured here may be of less consequence for patterns of polySU and/or there may be other factors more pertinent to patterns of polySU for the Low Resilience profile, in particular.

Despite CM being a potent predictor of SU and SUDs (Alvarez-Alonso et al., 2016; Mills et al., 2014), our findings provide insight into potentially protective effects of having higher levels of resilience traits like self-reliance, perseverance, and future-oriented thinking. These results are particularly interesting given that polySU predicts multiple deleterious outcomes above and beyond developing SUDs, including decrements in neurocognitive functioning and psychopathology (Jones et al., 2017; Moody et al., 2016). For example, compared to individuals who use no or few substances, polysubstance users are at elevated risk of long-term physical and mental health issues (Connor et al., 2013; Hakansson et al., 2011; Jones et al., 2017; Timko et al., 2018; Trenz et al., 2013). Uncovering individual-level characteristics that protect against polySU could point to potential targets for future preventative interventions. Our results represent an important step toward identifying potentially malleable individual factors that may protect against common long-term sequelae of CM.

While we often consider ‘prevention’ as the effort to ameliorate risk, some of the most successful preventive interventions for high-risk behaviors in youth also focus on promoting those resilience factors that will impact proximal and more distal outcomes (e.g., Life Skills Training/problem solving and coping skills; PATHS/socioemotional learning skills). Despite concerted effort to identify key factors that drive adolescent SU and that could be targeted in SU/SUD prevention efforts, many gaps still remain. By revealing mechanistic pathways by which CM impacts on SU/SUD risk may be tempered (Fig. 1), our outcomes help to address a persistent gap in our knowledge regarding modifiable factors that may reduce adolescent SU-liability and, as such, should be the primary focus of the development or adaptation of preventive interventions for SU/SUD in those at heightened risk. Indeed, these pathways may serve as novel targets for more effective preventive intervention strategies for SU/SUDs among CM-exposed youth. Moreover, in addition to informing the types of capabilities or skills that programs should target, the determination of subgroups of those within a more globally “at-risk” cohort (i.e., CM-exposed youth) who have differential susceptibility to SU behaviors that portend future SUD (i.e., poly-SU) has the potential to inform which individuals should receive more selective (vs. universal) preventive interventions.

We also examined whether characteristics of CM were associated with individual-level resilience traits, with the prediction that greater CM severity would be related to lower levels of resilience as has been demonstrated in studies using more traditional analytic approaches (Kwak et al., 2018). There were no differences among the profiles in CM characteristics such as type, timing, and severity. Although contrary to our hypotheses, this finding is largely consistent with prior studies using person-centered approaches, which report comparability across resilience profiles in terms of features of maltreatment histories (Yates & Grey, 2012; Yoon et al., 2022). These results suggest that differences in resilience characteristics may be driven by other, non-CM related factors. That is, other aspects of the developmental context might be critical to the development of these characteristics. This finding may be particularly important as it suggests that independent of the specific types of traumatic exposure, there is a potential equivalency in characteristics that are protective against maladaptive outcomes. Delineating precisely what those influences are and how they mediate or moderate the impact of CM on subsequent risk or resilience has the potential to inform the development of more effective evidence-based interventions to prevent CM’s negative impacts.

Limitations

Despite LONGSCAN’s large sample of CM-exposed children, our analytic sample is relatively small. Although simulation studies indicate that the current sample is adequately powered for a 3-profile solution (Tein et al., 2013), future work should confirm findings in larger samples. Second, although our analysis included standard measures of CM, profiles could differ in other characteristics such as pubertal timing. Moreover, it is noteworthy that this subsample had very high rates of neglect (~90%). Thus, in a sample with a different distribution of maltreatment subtypes the outcomes regarding the specificity of CM impacts may be quite different.

Third, the future expectations measure was only administered at 16 years of age. While it is ideal measure all indicators at the same time point, this temporal difference is unlikely to meaningfully impact the associations with SU during adolescence. Moreover, since important transitions and emerging adulthood milestones typically occur before age 16 (education attainment, employment) that would likely already be occurring (or not) at age 18, it may have been more informative to collect adolescents’ future attitudes at this earlier age.

Fourth, a specific measure of perseverance was not available in the current study. We examined adolescents’ participation and number of accolades received in extracurricular activities and work. Although prior work has shown links between engaging in extracurricular activities and resilience following maltreatment (e.g., Easterlin et al., 2019; Kwak et al., 2018), adolescents could evince perseverance without being a high-achiever on the measures used here. In future studies, a more comprehensive and direct measure of perseverance will be helpful in delineating the specific types of activities and behaviors.

Finally, although the resilience characteristics in the current study were expressly chosen for their lack of reliance on cognitive, social, and environmental factors, structural and systemic barriers to success are pervasive and insidious for at-risk youth. Thus, we cannot rule out whether structural barriers were integral in determining access to resources that promote the resilience factors examined here (e.g., future expectations about attending college may be related to perceived environmental resources). Though it is not within the scope of this work to speak to these inequities, it is imperative that the issue be acknowledged and that future work focuses on how individual-level resilience factors can or cannot be realized in environments that place excessive impediments on youths’ adaptive functioning.

Conclusions

The current study advances our understanding of pathways toward adolescent SU following maltreatment. We uncovered profiles of resilience that were related to polySU patterns such that, in a cohort already at heightened risk due to CM, those who were most likely to evince competence in these individual-level resilience constructs during adolescence were least likely to engage in polySU. Future research should examine the extent to which these profiles change over time and are potentially predictive of subsequent SU and SUDs. Identifying traits that are less dependent upon context or cognitive ability, or qualities of the environment in those who are at elevated risk of developing SUDs represents a promising avenue for future prevention and intervention efforts.