Subjective Effects as Predictors of Substance Use Disorders in a Clinical Sample: A Longitudinal Study

Abstract

Background

The literature on the association between subjective effects (SEs; i.e., how an individual perceives their physiological and psychological reactions to a drug) and substance use disorders (SUDs) is largely limited to community samples. The present study addressed the following aims in a clinical sample: whether SEs predict general versus substance-specific SUD in adolescence and adulthood after controlling for conduct disorder symptoms (CDsymp); whether SEs predict SUDs across drug classes; whether SEs predict change in SUD from adolescence to adulthood; and whether there are racial/ethnic differences in associations.

Methods

Longitudinal analyses were conducted using data from a sample of 744 clinical probands recruited from residential and outpatient SUD treatment facilities in CO during adolescence (M_age = 16.26) and re-assessed twice in adulthood (M_ages = 22.56 and 28.96), approximately seven and twelve years after first assessment. SEs and CDsymp were assessed in adolescence. SUD severity was assessed at adolescence and twice during adulthood.

Results

SEs assessed in adolescence robustly predicted general SUD for legal and illegal substances in adolescence and adulthood, whereas CDsymp predicted SUD primarily in adolescence. Higher positive and negative SEs in adolescence were associated with greater SUD severity after controlling for CDsymp, with similar magnitudes. Results indicated cross-substance effects of SEs on SUD. We found no evidence for racial/ethnic differences in associations.

Conclusions

We investigated the progression of SUD in a high-risk sample with greater odds of sustained SUD. In contrast to CDsymp, both positive and negative SEs consistently predicted general SUD across substances in adolescence and adulthood.

1. INTRODUCTION

Substance use disorders (SUDs) are a pressing public health issue. In 2019, 20.4 million individuals aged 12 and older had a SUD diagnosis (SAMHSA, 2020). Typically, in community samples, substance use and polysubstance experimentation peak in emerging adulthood (Moss et al., 2014; Stone et al., 2012) and decline in adulthood (Johnston et al., 2009; Merrin & Leadbeater, 2018). However, trajectories are heterogenous, especially in clinical samples: SUDs may resolve, decrease in severity, or persist into adulthood (Brown et al., 2011; Trim et al., 2015). Adolescents with SUDs are highly likely to relapse shortly after receiving treatment (Brown et al., 2011). Also, adolescent SUD severity is associated with greater psychological distress and impairment (Ciraulo et al., 2003; Mirza & Mirza, 2008), later crime involvement (Crowley et al., 1998), higher later SUD severity (Ciraulo et al., 2003; Crowley et al., 1998), and early mortality (Clark et al., 2008; Cornelius et al., 2008). Understanding predictors of adolescent SUD and atypical trajectories strengthens our ability to predict and prevent SUDs and their associated consequences.

There are well-established predictors of general SUDs across development, including conduct disorder (CD; Palmer et al., 2013) and novelty seeking (Pardini et al., 2007). For example, CD explains co-occurring SUDs (Button et al., 2007; Iacono et al., 2008) and shares etiological influences with general SUDs (Grant et al., 2015; Krueger et al., 2009). Conversely, we know less about predictors of substance-specific SUDs, although general and substance-specific genetic and environmental influences contribute to SUDs (Palmer et al., 2013). The present study examines subjective effects (SEs; individual differences in physiological and psychological drug experiences) in adolescence as predictors of SUDs. We examined data from a clinical sample in treatment for SUDs as adolescents, then followed in two waves during adulthood. We addressed whether SEs explain general versus specific SUD from adolescence to adulthood after controlling for CD symptoms (CDsymp), a well-established risk factor for general SUDs (Button et al., 2007; Erskine et al., 2016; Palmer et al., 2013).¹

1.1. Associations Between SEs and SUDs

Studies have found evidence for two moderately correlated SE factors (Ríos-Bedoya et al., 2009; Zeiger et al., 2010, 2012): positive (e.g., euphoria, buzz, or relaxation) and negative (e.g., paranoia, sadness, or nausea; 9,14–18). Positive SEs at early use are consistently, positively associated with cannabis (Fergusson et al., 2003; Le Strat et al., 2009; Zeiger et al., 2010), alcohol (King et al., 2014; Morean & Corbin, 2008), tobacco (de Wit & Phillips, 2012; Hu et al., 2008; Pomerleau et al., 1998; Ríos-Bedoya et al., 2009), and opioid (Bieber et al., 2008; Haertzen et al., 1983) dependence. Research on other illegal substances is more limited and mixed (J. D. Grant et al., 2005; Karsinti et al., 2018; Lambert et al., 2006), potentially because research focuses on community samples with low rates of SUDs.

In contrast to results on positive SEs, associations between negative SEs and SUDs are less consistent and remain understudied. Researchers have found mixed results regarding negative SEs predicting cannabis (Fergusson et al., 2003; Le Strat et al., 2009; Zeiger et al., 2012), alcohol (King et al., 2014; Ríos-Bedoya et al., 2009; Zeiger et al., 2012), and tobacco SUDs (DiFranza et al., 2004; Pomerleau et al., 1998; Ríos-Bedoya et al., 2009) in community samples. One study found positive associations between cannabis SEs and cannabis SUD in a combined community and clinical sample (Zeiger et al., 2010). Literature on negative SEs and illegal substances is very limited (Karsinti et al., 2018).

1.2. Limitations of the Present Literature

To examine predictors for less commonly used substances (e.g., stimulants and opioids), additional studies including clinical samples are needed. Also, existing studies have mostly focused on associations between SEs and SUDs for one substance (typically alcohol, cannabis, or tobacco; e.g., Karsinti et al., 2018; King et al., 2014; Zeiger et al., 2010). Although existing results are inconsistent (Pomerleau et al., 2004; Zeiger et al., 2012), past research has identified significant cross-substance effects for alcohol, tobacco, and cannabis (Zeiger et al., 2012).

Finally, the literature on associations between SEs and SUDs is largely retrospective, with adults reporting on initial SEs many years after first use (de Wit & Phillips, 2012). Longitudinal studies beginning in adolescence are limited to nicotine (Audrain-McGovern et al., 2007; Chen et al., 2003; DiFranza et al., 2007; Kandel et al., 2007) and cannabis (Fergusson et al., 2003; Le Strat et al., 2009) and showed that positive SEs predict corresponding SUD into early adulthood. Prospective studies on alcohol are typically limited to participants over age 21, and suggest that positive SEs predict alcohol SUD in later adulthood (King et al., 2014). Additional longitudinal studies continuing past early adulthood are needed.

1.3. Present Study

We evaluated positive and negative SEs as predictors of SUD severity (referred to as “SUD”) in a racially and ethnically diverse clinical sample assessed in three waves from adolescence through adulthood by the Center for Antisocial Drug Dependence. We addressed whether SEs predict general or substance-specific SUD for alcohol, cannabis, tobacco, hallucinogens, cocaine, amphetamines, and opioids. All analyses were preregistered, unless otherwise noted: https://osf.io/mc7w8/?view_only=6daabac7e797496db01743426e5683d0.

1.4. Research Questions

1.4.1. Do SEs and CD symptoms predict general or specific SUD?

We hypothesized that positive SEs would predict specific rather than general SUD after controlling for CDsymp and that CDsymp would predict general SUD. Analyses on negative SEs were exploratory, given limited literature on this topic.

1.4.2. Do SEs predict SUD across substances?

Limited literature indicates that SEs may influence SUDs across substances (Pomerleau et al., 2004; Zeiger et al., 2012). Exploratory analyses addressed cross-substance effects of SEs on SUDs.

1.4.3. Do SEs predict change in SUD over time?

Analyses examining whether SEs predict change in SUD from adolescence through adulthood were exploratory, as research on this area is limited.

1.4.4. Associations between predictors and SUD different across race/ethnic groups or sex?

Minoritized racial and ethnic groups experience differential substance use trajectories (Chen & Jacobson, 2012) and experience unique stressors (e.g., bicultural stress via discrimination) predicting SUDs (Romero et al., 2007). There are ethnic differences in predictors of SUD across time (e.g., social conformity as a more robust predictor for Hispanic individuals; Galaif & Newcomb, 1999). No prior studies have examined racial/ethnic differences in associations between SEs and SUD, and our analyses were exploratory. We could not examine whether sex moderates associations between predictors and SUD as intended, given our sample was only 14% women. Sex was a covariate in all models.

2. METHODS

2.1. Participants

The present study included 744 clinical probands participating in a longitudinal study of genetic and developmental influences on adolescent substance use (Hartman et al., 2006; Hopfer et al., 2003). Beginning in 2003, adolescent participants (Wave 1: N = 744, M_age = 16.26, range = 12.95–19.71, SD = 1.15) were recruited from residential and outpatient SUD treatment facilities in Denver, CO. Each participant was contacted at least twice annually with opportunities for updating contact information. Individuals were assessed twice in adulthood (Wave 2: N = 418, M_age = 22.56, SD = 1.83; Wave 3: N = 352, M_age = 28.96, SD = 1.96), approximately seven (M = 7.57, SD = 2.24) and twelve (M = 12.22, SD = 1.89) years after initial assessment. Nearly 53% of individuals were lost to follow-up by Wave 3, but neither greater Wave 1 CDsymp nor SUD severity predicted attrition (Table S1). By Wave 3, approximately two percent of attrition was due to direct refusal and approximately five percent was due to participant death. As this is a high-risk sample, housing instability also likely contributed to attrition. For study inclusion, interview staff had to judge probands as not currently suicidal, homicidal, psychotic, or severely developmentally delayed, with no current intoxication or physical illness preventing participation in evaluation and treatment. The university’s institutional review board approved all procedures. Parents or guardians provided written consent and participants provided assent or consent. The sample was 86% male and 50% non-Hispanic White (Table S2).²

2.2. Measures

2.2.1. SEs

SEs were measured using the modified Lyons Battery for Subjective Effects (Lyons et al., 1997) at Wave 1. At initial assessment, participants reported SEs retrospectively for alcohol, cannabis, hallucinogens, cocaine, amphetamines, and opioids³ if participants reported using the substance more than five times (a standard CIDI-SAM skip threshold). Tobacco SEs were assessed in participants endorsing use of pipes, cigars, and snuff five or more times or tobacco almost every day for a month. Participants were asked, “In the period shortly after you used [substance], did it make you feel [subjective effect]?” (yes/no); questions did not specifically address SEs at initial or first use. Prior confirmatory factor analyses (CFA) and Mokken Scale analyses (Haberstick et al., 2010; Zeiger et al., 2010) indicated evidence for two scales with six items each: positive effects (relaxed, creative, energetic, on top of the world, increased sex drive, sociable) and negative effects (dizzy, nauseous, lazy, out of control, drowsy, unable to concentrate). Responses were averaged to create composite scores for each scale and transformed or collapsed into ordinal categories to address skewness and kurtosis (Derks et al., 2004; see Tables S3 and S4 for frequencies and descriptive statistics).⁴ CFA results were similar for positive and negative SEs; they indicated that SEs for alcohol, tobacco, and cannabis loaded onto one factor and SEs for hallucinogens, amphetamines, cocaine, and opioids loaded onto a second factor (Table S6, Figure 1), labeled the “legal” and “illegal” SE factors, respectively.⁵

Figure 1.

Results of Confirmatory Factor Analyses for Positive and Negative Subjective Effects

(See Figure 1 File).

Note: Standardized factor loadings are presented from models examining positive SEs/negative SEs. All factor loadings and correlations significant at p ≤ .001. SEs = Subjective Effects

2.2.2. CD Symptoms

CDsymp at Wave 1 (M = 5.70, SD = 2.95) were assessed using the Diagnostic Interview Schedule for Children (DISC; Robins et al., 1996). The DISC is a structured interview assessing DSM-IV CD criteria and is reliable and valid (e.g., Piacentini et al., 1993). 84.86% of participants met diagnostic criteria for CD. A CDsymp composite score was created by summing endorsed symptoms (see Table S7 for descriptive statistics).

2.2.3. SUDs

At all waves, the Composite International Diagnostic Interview – Substance Abuse Module (CIDI-SAM; Cottler et al., 1989), a reliable measure of SUD severity for adolescents and adults in clinical and community settings (Crowley et al., 2001; Üstün et al., 1997), assessed diagnostic criteria for DSM-IV substance abuse and dependence criteria met during lifetime at Wave 1. At Waves 2 and 3, past year SUD criteria met was evaluated (see Tables S8 and S9 for frequencies). Following the present DSM-5, which does not distinguish between substance abuse and dependence, we indexed SUD severity for each substance: 0 = “no criteria endorsed”, 1 = “one criterion endorsed”, 2 = “mild”, 3 = “moderate”, 4 = “severe”, which was the outcome for all analyses. At Wave 1, 91.81% of participants met diagnostic criteria for two or more SUDs and the mean number of SUDs was 3.02 (SD = 1.91). Probands had used a mean of 4.12 substances five or more times by Wave 1 assessment (SD = 1.59). CFA provided evidence for a two-factor SUD model at each wave, with legal SUD loading on one factor and illegal SUD loading onto the second factor (Table S10, Figure 2).

Figure 2.

Results of Confirmatory Factor Analyses for Substance Use Disorder Factors at Waves 1, 2, and 3 (See Figure 2 File)

Note: Standardized factor loadings from models examining SUD criteria assessed in Wave1/Wave 2/Wave 3 are presented.

All factor loadings and correlations significant at p ≤ .001. SUD = substance use disorder.

2.3. Statistical Analyses

Structural equation models examining associations between SEs and SUD were estimated using MPlus Version 8.3 (Muthén & Muthén, 2017), including best-fitting models from CFAs. We used the weighted least squares mean and variance with pairwise deletion. Bentler’s Comparative Fit Indices (CFI; Bentler, 1990) > .95, root mean square error of approximation (RMSEA; Browne & Cudeck, 1992) < .06, and non-significant χ² statistics (interpreted with caution due to sensitivity to sample size) indicated good model fit. The False Discovery Rate (FDR), which controls for expected proportion of falsely rejected null hypotheses, addressed multiple testing (Benjamini & Hochberg, 1995; Radua & Albajes-Eizagirre, 2022).

First, we examined correlations between all measures. Further analyses controlled for CDsymp and sex. At each wave, we regressed SUD factors and residual variance of specific SUDs on SEs for each substance (question 1). Each SUD factor was regressed on each positive and negative SE factor (question 2). For question 3, latent growth curve modeling addressed whether SEs predict the slope of SUD, or change in SUD from wave 1, accounting for different ages at assessment (Figure 3; Muthén & Muthén, 2017). For statistically significant associations between predictors and SUD, we tested for racial/ethnic group differences (non-Hispanic White vs. all other categories) with a χ² difference test (question 4). Significant χ² difference tests between models with parameters freed vs. models equating parameters across groups would indicate group differences in associations.

Figure 3.

Example Growth Model Examining Associations between SEs, CD Symptoms, Initial SUD, and Change in SUD Over Time, After Controlling for Sex

(See Figure 3 File).

Note: For each model, the TYPE = RANDOM and TSCORES (time scores) options were used to allow time scores to vary across individuals (i.e., different ages of assessment at Waves 1 through 3), incorporated into the model using the AT command. Latent basis growth models with freed slope loadings are more efficient than separate models for linear, quadratic, and high polynomials while allowing estimation of nonlinear change without collinearity and are simpler to interpret (Ram & Grimm, 2007). The latent SUD intercept had unstandardized loadings of 1.0 for each timepoint (capturing individual stability of SUD across time). The latent slope factor reflected individual change in SUDs over time (Bollen & Curran, 2006). Results after controlling for sex are reported. SEs = subjective effects. SUD = substance use disorder.

3. RESULTS

3.1. Do SEs and CDsymp predict general or specific SUD?

All significant correlations between predictors (SEs and CDsymp) and SUD were positive (Tables S11 – S13). Positive and negative SEs for each substance were significantly correlated with corresponding Wave 1 SUDs. Fewer SEs were significantly correlated with corresponding Wave 2 and 3 SUDs (most significant correlations were between SEs for legal substances and corresponding SUDs). CDsymp were significantly correlated with all SUDs at Wave 1, only with Wave 2 hallucinogen SUD, and only with Wave 3 cannabis SUD.

Below, we address whether SEs and CDsymp independently influence SUDs, controlling for sex and after addressing multiple testing. All significant associations were positive.

3.1.1. Legal SUDs.

3.1.1.1. Legal SUD Factor

General legal SUD factors were predicted by positive and negative SEs for alcohol, tobacco, and cannabis at Wave 1 (β = .226 to .462), positive SEs for cannabis and negative SEs for alcohol, tobacco, and cannabis at Wave 2 (β = .210 to .292), and alcohol negative SEs at Wave 3 (β = .304; Table 1). CDsymp predicted general legal SUD only at Wave 1 (β = .382 to .514).

Table 1.

Legal SUD Factors and Residual Variances of Specific SUDs at Waves 1, 2, and 3 Regressed on Positive and Negative SEs for Each Substance and CD Symptoms, Controlling for Sex

Alcohol
	Legal SUD Factor		Residual Variance of Alcohol SUD			Legal SUD Factor		Residual Variance of Alcohol SUD
	β	p-value	β	p-value		β	p-value	β	p-value
Wave 1 (N = 642)
Alcohol Positive SEs	.227*** [.128, .326]	< .001	.214*** [.127, .300]	<.001	Alcohol Negative SEs	.462*** [.365, .558]	< .001	−.009 [−.132, .114]	.886
CD Symptoms	.382*** [.285, .480]	< .001	.150*** [.062, .238]	.001	CD Symptoms	.319*** [.224, .415]	< .001	.160*** [.066, .253]	.001
Wave 2 (N = 367)
Alcohol Positive SEs	.172* [.019, .325]	.027	.037 [−.093, .167]	.578	Alcohol Negative SEs	.237** [.090, .384]	.002	−.124 [−.255, .008]	.065
CD Symptoms	.060 [−.093, .214]	.443	.027 [−.093, .146]	.662	CD Symptoms	.052 [−.100, .204]	.501	.051 [−.071, .172]	.414
Wave 3 (N = 314)
Alcohol Positive SEs	.105 [−.070, .280]	.238	−.013 [−.159, .134]	.865	Alcohol Negative SEs	.304*** [.119, .489]	.001	−.093 [−.267, .082]	.298
CD Symptoms	.155 [−.031, .340]	.103	−.114 [−.266, .038]	.142	CD Symptoms	.147 [−.034, .328]	.111	−.113 [−.270, .044]	.159
Tobacco
	Legal SUD Factor		Residual Variance of Tobacco SUD			Legal SUD Factor		Residual Variance of Tobacco SUD
	β	p-value	β	p-value		β	p-value	β	p-value
Wave 1 (N = 608–610)
Tobacco Positive SEs	.275*** [.171, .378]	< .001	.018 [−.089, .124]	.747	Tobacco Negative SEs	.217*** [.109, .324]	< .001	−.049 [−.161, .063]	.393
CD Symptoms	.488*** [.388, .588]	< .001	−.109 [−.240, .022]	.103	CD Symptoms	.499*** [.397, .600]	< .001	−.125 [−.261, .011]	.072
Wave 2 (N = 353–355)
Tobacco Positive SEs	.108 [−.050, .267]	.180	.102 [−.003, .207]	.056	Tobacco Negative SEs	.210** [.057, .363]	.007	−.083 [−.216, .050]	.220
CD Symptoms	.086 [−.071, .243]	.283	−.045 [−.168, .077]	.469	CD Symptoms	.079 [−.076, .234]	.318	−.040 [−.164, .085]	.531
Wave 3 (N = 306–308)
Tobacco Positive SEs	.170 [−.007, .347]	.060	.038 [−.097, .174]	.581	Tobacco Negative SEs	.179* [.008, .349]	.040	−.030 [−.165, .104]	.661
CD Symptoms	.099 [−.087, .285]	.297	.018 [−.124, .159]	.805	CD Symptoms	.133 [−.051, .317]	.157	.003 [−.142, .148]	.967
Cannabis
	Legal SUD Factor		Residual Variance of Cannabis SUD			Legal SUD Factor		Residual Variance of Cannabis SUD
	β	p-value	β	p-value		β	p-value	β	p-value
Wave 1 (N = 665)
Cannabis Positive SEs	.226*** [.135, .317]	< .001	.110* [.023, .198]	.014	Cannabis Negative SEs	.368*** [.276, .459]	< .001	.163** [.056, .269]	.003
CD Symptoms	.514*** [.427, .600]	< .001	−.067 [−.181, .046]	.245	CD Symptoms	.452*** [.364, .540]	< .001	−.071 [−.176, .033]	.182
Wave 2 (N = 378)
Cannabis Positive SEs	.292*** [.154, .431]	< .001	−.106 [−.234, .022]	.106	Cannabis Negative SEs	.273*** [.120, .426]	< .001	.018 [−.097, .134]	.755
CD Symptoms	.093 [−.051, .237]	.208	.003 [−.116, .122]	.963	CD Symptoms	.059 [−.090, .209]	.436	−.002 [−.124, .119]	.969
Wave 3 (N = 326)
Cannabis Positive SEs	.003 [−.165, .170]	.977	.188** [.069, .308]	.002	Cannabis Negative SEs	.207* [.037, .378]	.017	−.046 [−.179, .087]	.500
CD Symptoms	.082 [−.095, .259]	.363	.067 [−.063, .197]	.315	CD Symptoms	.057 [−.120, .235]	.527	.074 [−.059, .207]	.276

^*p < .05

^**p < .01

^***p ≤ .001

Bolded estimate: FDR p < .05

Note: Reported Ns include individuals with data for both SEs and SUDs for that specific substance.

3.1.1.2. Residual Variance of Specific SUDs.

Substance-specific residual variances were predicted by positive SEs for alcohol and cannabis at Wave 1 (β = .110 to .214), no SEs at Wave 2, and positive SEs for cannabis at Wave 3 (β = .188; Table 1). CDsymp significantly predicted only Wave 1 alcohol SUD residual variance (β = .150 to .160).

3.1.2. Illegal SUDs

3.1.2.1. Illegal SUD Factor.

General illegal SUD factors were predicted by positive SEs for hallucinogens and cocaine and negative SEs for hallucinogens, amphetamines, and opioids at Wave 1 (β = .197 to .382), no SEs at Wave 2, and negative SEs for hallucinogens at Wave 3 (β = .273; Table 2). CDsymp predicted Wave 1 general SUD (β = .241 to .296), but not Wave 2 or 3.

Table 2.

Illegal SUD Factors and Residual Variances of Specific SUDs at Waves 1, 2, and 3 Regressed on Positive and Negative SEs for Each Substance and CD Symptoms, Controlling for Sex

Hallucinogens
	Illegal SUD Factor		Residual Variance of Hallucinogen SUD			Illegal SUD Factor		Residual Variance of Hallucinogen SUD
	β	p-value	β	p-value		β	p-value	β	p-value
Wave 1 (N =275)
Hallucinogen Positive SEs	.197** [.052, .342]	.008	.135* [.010, .261]	.035	Hallucinogen Negative SEs	.382*** [.257, .507]	< .001	.086 [−.050, .222]	.214
CD Symptoms	.292*** [.144, .440]	< .001	−.023 [−.159, .113]	.742	CD Symptoms	.275*** [.133, .416]	< .001	−.018 [−.151, .116]	.795
Wave 2 (N = 178)
Hallucinogen Positive SEs	.020 [−.205, .245]	.860	.075 [−.101, .251]	.402	Hallucinogen Negative SEs	.221* [.003, .439]	.047	.064 [−.129, .257]	.517
CD Symptoms	.071 [−.163, .304]	.552	.002 [−.178, .182]	.980	CD Symptoms	.075 [−.171, .320]	.552	.005 [−.172, .183]	.954
Wave 3 (N = 160)
Hallucinogen Positive SEs	−.133 [−.381, .115]	.292	.169 [−.045, .384]	.122	Hallucinogen Negative SEs	.273* [.056, .489]	.014	.100 [−.094, .294]	.313
CD Symptoms	−.068 [−.324, .188]	.603	−.110 [−.318, .097]	.297	CD Symptoms	−.142 [−.395, .111]	.271	−.086 [−.299, .127]	.428
Amphetamines
	Illegal SUD Factor		Residual Variance of Amphetamine SUD			Illegal SUD Factor		Residual Variance of Amphetamine SUD
	β	p-value	β	p-value		β	p-value	β	p-value
Wave 1 (N = 220)
Amphetamine Positive SEs	−.001 [−.179, .176]	.988	.378*** [.249, .507]	< .001	Amphetamine Negative SEs	.220** [.062, .378]	.006	.169* [.040, .299]	.011
CD Symptoms	.249** [.073, .425]	.006	.019 [−.127, .164]	.802	CD Symptoms	.241** [.066, .415]	007	.043 [−.111, .197]	.586
Wave 2 (N = 130)
Amphetamine Positive SEs	−.008 [−.251, .236]	.950	.031 [−.168, .230]	.762	Amphetamine Negative SEs	.084 [−.138, .307]	.458	−.064 [−.262, .135]	.530
CD Symptoms	−.001 [−.254, .252]	.994	−.183 [−.408, .042]	.110	CD Symptoms	−.007 [−.250, .235]	.952	−.184 [−.408, .040]	.107
Wave 3 (N = 113)
Amphetamine Positive SEs	.113 [−.308, .533]	.599	−.132 [−.435, .172]	.394	Amphetamine Negative SEs	.161 [−.284, .606]	.478	−.188 [−.544, .167]	.299
CD Symptoms	−.044 [−.447, .358]	.829	−.181 [−.465, .103]	.212	CD Symptoms	−.036 [−.466, .393]	.868	−.180 [−.491, .131]	.258
Cocaine
	Illegal SUD Factor		Residual Variance of Cocaine SUD			Illegal SUD Factor		Residual Variance of Cocaine SUD
	β	p-value	β	p-value		β	p-value	β	p-value
Wave 1 (N = 231)
Cocaine Positive SEs	.241** [.082, .401]	.003	.102 [−.044, .247]	.170	Cocaine Negative SEs	.184* [.023, .344]	.025	.111 [−.021, .243]	.098
CD Symptoms	.278*** [.119, .437]	.001	−.019 [−.164, .126]	.797	CD Symptoms	.268*** [.109, .427]	.001	−.032 [−.178, .114]	.665
Wave 2 (N = 133)
Cocaine Positive SEs	N/A	N/A	.048 [−.168, .264]	.666	Cocaine Negative SEs	N/A	N/A	.089 [−.115, .293]	.395
CD Symptoms	N/A+	N/A	−.081 [−.302, .139]	.471	CD Symptoms	N/A	N/A	−.092 [−.313, .129]	.415
Wave 3 (N = 127)
Cocaine Positive SEs	−.055 [−.333, .223]	.697	−.030 [−.224, .165]	.765	Cocaine Negative SEs	.097 [−.159, .353]	.458	−.109 [−.326, .108]	.325
CD Symptoms	−.213 [−.490, .064]	.131	.122 [−.097, .341]	.275	CD Symptoms	−.217 [−.491, .057]	.121	.128 [−.091, .348]	.252
Opioids
	Illegal SUD Factor		Residual Variance of Opioid SUD			Illegal SUD Factor		Residual Variance of Opioid SUD
	β	p-value	β	p-value		β	p-value	β	p-value
Wave 1 (N = 178)
Opioid Positive SEs	.021 [−.184, .227]	.838	.248** [.080, .416]	.004	Opioid Negative SEs	.258** [.076, .441]	.006	.268** [.103, .434]	.002
CD Symptoms	.268** [.098, .439]	.002	−.102 [−.287, .083]	.280	CD Symptoms	.296*** [.129, .464]	.001	−.073 [−.255, .109]	.433
Wave 2 (N = 110)
Opioid Positive SEs	.054 [−.180, .287]	.652	.224 [−.092, .540]	.164	Opioid Negative SEs	−.159 [−.475, .156]	.322	.425** [.115, .736]	.007
CD Symptoms	−.124 [−.410, .162]	.395	.044 [−.312, .400]	.809	CD Symptoms	−.229 [−.584, .126]	.206	.217 [−.248, .682]	.361
Wave 3 (N = 106)
Opioid Positive SEs	−.027 [−.325, .271]	.179	.295* [.036, .554].	.026	Opioid Negative SEs	.095 [−.215, .406]	.547	.148 [−.104, .400]	.249
CD Symptoms	−.318 [−.688, .051]	.091	.026 [−.422, .474]	.910	CD Symptoms	−.312 [−.699, .075]	.114	.035 [−.440, .510]	.885

^*p < .05

^**p < .01

^***p ≤ .001

Bolded estimate: FDR p < .05

⁺At Wave 2, we were unable to regress cocaine positive SEs on the general SUD factor, due to small cell sizes

Note: Reported Ns include individuals with data for both SEs and SUDs for that specific substance.

3.1.2.2. Residual Variance of Specific SUDs.

Substance-specific residual variances were predicted by positive and negative SEs for amphetamines and opioids at Wave 1 (β = .169 to .268), negative SEs for opioids at Wave 2 (β = .425), and no SEs at Wave 3 (Table 2). CDsymp did not predict specific illegal SUD residual variance.

3.1.3. Exploratory Analyses.

Exploratory analyses addressed whether individual negative SEs negatively predicted SUD, since positive correlations between negative SEs and SUDs appeared counterintuitive. We examined correlations between each individual negative SE item for legal substances and legal SUD factors at all waves (given robust associations between negative SE factors and legal SUD factors). There were no significant negative correlations (Table S14). We also tested whether substantial correlations between positive and negative SEs drove associations between negative SE factors and SUDs. After controlling for positive SE factors, negative SE factors predicted legal and illegal SUD independently at Wave 1 and illegal SUD at Wave 3 (Table S15).

3.1.4. Summary

SEs were more robust predictors of general SUD than substance-specific SUD. Legal SEs predicted general legal SUD into adulthood, whereas illegal SEs primarily predicted general illegal SUD in adolescence. Effect sizes were similar across legal and illegal SUD, although confidence intervals were broader for illegal SUD. Results were robust for both positive and negative SEs. CDsymp consistently predicted general, rather than specific, SUD, and only in adolescence.

3.2. Do SEs predict SUD across substances?

Positive and negative SEs were significantly correlated across substances (Table S16). There was little evidence that correlations between SEs and SUDs for corresponding substances were larger in magnitude than those across substances (Tables S11, S12).

We also examined correlations between SEs and SUD latent factors (Table 3). Positive and negative legal SEs factors and legal SUD factors were significantly correlated at all waves. Illegal SE factors and illegal SUD factors were significantly correlated at Wave 1 (positive and negative SEs) and Wave 3 (negative SEs). Legal SEs factors and illegal SUD factors were significantly correlated at Wave 1 (positive and negative) and Waves 2 and 3 (negative). Illegal SEs factors (positive and negative) and legal SUD factors were significantly correlated at Waves 1 and 2.

Table 3.

Correlations Between SUD Latent Factors, Positive and Negative SEs Latent Factors, and CD Symptom Count

	Legal SUD W1	Legal SUD W2	Legal SUD W3	CDsymp	Pos Legal SE	Neg Legal SE	Illegal SUD W1	Illegal SUD W2	Illegal SUD W3	Pos Illegal SE
Legal SUD W1	1
Legal SUD W2	.494*** [.349, .639]	1
Legal SUD W3	.530*** [.384, .676]	.945*** [.780, 1.111]	1
CDsymp	.503*** [.424, .581]	.109 [−.019, .238]	.139 [−.012, .289]	1
Pos Legal SE	.593*** [.492, .693]	.354*** [.192, .515]	.255* [.058, .451]	.189*** [.095, .284]	1
Neg Legal SE	.773*** [.670, .875]	.383*** [.216, .550]	.338*** [.158, .518]	.284*** [.190, .379]	.640*** [.542, .739]	1
Illegal SUD W1	.796*** [.711, .882]	.209* [−.059, .359]	.358*** [.201, .515]	.452*** [.367, .537]	.163* [.037, .289]	.556*** [.448, .665]	1
Illegal SUD W2	.327*** [.144, .509]	.549*** [.382, .717]	.265* [.010, .521]	.232** [.078, .386]	.010 [−.177, .197]	.211* [.026, .407]	.771*** [.608, .933]	1
Illegal SUD W3	.197* [.002, .391]	.324** [.123, .524]	.556*** [.371, .741]	.098 [−.069, .265]	.086 [−.129, .302]	.353*** [.141, .632]	.554*** [.369, .738]	.889*** [.671, 1.108]	1
Pos Illegal SE	.403*** [.265, .540]	.368*** [.167, .569]	.188 [−.020, .395]	.040 [−.073, .152]	.896*** [.786, 1.005]	.473*** [.341, .605]	.232*** [.141, .324]	.114 [−.066, .294]	.012 [−.223, .247]	1
Neg Illegal SE	.575*** [.421, .728]	.241* [.013, .470]	.212 [−.037, .462]	.044 [−.084, .173]	.592*** [.442, .742]	.877*** [.759, .994]	.343*** [.239, .448]	.187 [−.038, .412]	.272* [.040, .503]	.504*** [.365, .644]

Note: Legal SUD W1, W2, W3: Legal SUD Factors for Waves 1, 2, and 3. Illegal SUD W1, W2, W3: Illegal SUD Factors for Waves 1, 2, and 3. Pos Legal SE: Positive SEs Factor for Legal Substance. Neg Legal SE: Negative SEs Factor for Legal Substances. Pos Illegal SE: Positive SEs Factor for Illegal Substance. Neg Illegal SE: Negative SEs Factor for Illegal Substances. CDsymp: CD Symptom Count.

^***p ≤ .001

^**p = .003

^*p < .05

Below are results from models examining whether each latent SEs factor and CDsymp independently influenced latent SUD factors, after controlling for sex and addressing multiple testing. All significant associations were positive.

3.2.1. Legal SUDs

The legal positive SEs factor predicted general legal SUD at Waves 1 and 2 (β = .328 to .497), and the legal negative SEs factor predicted general legal SUD at all waves (β = .327 to .631; Table 4). Positive and negative illegal SEs factors predicted general legal SUD at Wave 1 (β = .371 to .550; Table 4). CDsymp only predicted general legal SUD at Wave 1 (β = .499 to .506).

Table 4.

Legal and Illegal SUD Factors at Waves 1, 2, and 3 Regressed on Legal and Illegal Positive and Negative SE Factors and CD Symptoms, Controlling for Sex

	β	p-value		β	p-value
Legal SUD Factor
Wave 1 (N = 723)
General Legal Positive SEs	.497*** [.404, .591]	< .001	General Legal Negative SEs	.631*** [.532, .730]	<.001
CD Symptoms	.506*** [.428, .583]	<.001	CD Symptoms	.499*** [.421, .578]	<.001
Wave 2 (N = 696)
General Legal Positive SEs	.328*** [.173,.484]	<.001	General Legal Negative SEs	.362*** [.195, .529]	<.001
CD Symptoms	.090 [ −.043, .223]	.185	CD Symptoms	.092 [−.041, .224]	.176
Wave 3 (N = 693)
General Legal Positive SEs	.216* [.025, .408]	.027	General Legal Negative SEs	.327*** [.148, .507]	<.001
CD Symptoms	.125 [−.033, .283]	.122	CD Symptoms	.121 [−.038, .279]	.136
Legal SUD Factor
Wave 1 (N = 723)
General Illegal Positive SEs	.371*** [.246, .495]	< .001	General Illegal Negative SEs	.550*** [.408, .692]	< .001
CD Symptoms	.504*** [.426, .581]	< .001	CD Symptoms	.503*** [.426, .581]	< .001
Wave 2 (N = 572)
General Illegal Positive SEs	.360*** [.159, 562]	< .001	General Illegal Negative SEs	.249* [.016, .482]	.036
CD Symptoms	.089 [−.042, .220]	.183	CD Symptoms	.090 [−.041, .221]	.180
Wave 3 (N = 547)
General Illegal Positive SEs	.181 [−.030, .393]	.093	General Illegal Negative SEs	.238 [−.018, .493]	.068
CD Symptoms	.115 [−.041, .271]	.149	CD Symptoms	.116 [−.039, .272]	.143
Illegal SUD Factor
Wave 1 (N = 723)
General Illegal Positive SEs	.221*** [.136, .307]	< .001	General Illegal Negative SEs	.340*** [.239, .441]	< .001
CD Symptoms	.457*** [.371, .542]	< .001	CD Symptoms	.457*** [.371, .542]	< .001
Wave 2 (N = 572)
General Illegal Positive SEs	.100 [−.072, .272]	.253	General Illegal Negative SEs	.185 [−.044, .415]	.114
CD Symptoms	.202* [.046, .358]	.011	CD Symptoms	.201* [.046, .357]	.011
Wave 3 (N = 547)
General Illegal Positive SEs	−.010 [−.238, .217]	.930	General Illegal Negative SEs	.259* [.027, .492]	.029
CD Symptoms	.077 [−.089, .243]	.365	CD Symptoms	.074 [−.098, .247]	.399
Illegal SUD Factor
Wave 1 (N = 723)
General Legal Positive SEs	.078 [−.034, .189]	.172	General Legal Negative SEs	.451*** [.349, .552]	< .001
CD Symptoms	.456*** [.371, .542]	< .001	CD Symptoms	.456*** [.371, .542]	< .001
Wave 2 (N = 696)
General Legal Positive SEs	−.030 [−.201, .140]	.728	General Legal Negative SEs	.157 [ −.028, .341]	.095
CD Symptoms	.202* [.045, .358]	.011	CD Symptoms	.201* [.045, .357]	.011
Wave 3 (N = 693)
General Legal Positive SEs	−.013 [−.209, .182]	.892	General Legal Negative SEs	.370*** [.169, .570]	< .001
CD Symptoms	.078 [−.091, .247]	.365	CD Symptoms	.070 [−.109, .248]	.443

^*p < .05

^***p ≤ .001

Bolded estimate: FDR p < .05

3.2.2. Illegal SUDs

Illegal positive and negative SEs factors predicted general illegal SUD at Wave 1 (β = .221 to .340; Table 4). The legal negative SE factor predicted general illegal SUD (β = .370 to .451; Waves 1 and 3; Table 4). CDsymp predicted general illegal SUD at Waves 1 and 2 (β = .201 to .457).

3.2.3. Exploratory Analyses

Exploratory analyses, not preregistered, indicated some significant residual correlations between SEs for each substance and corresponding SUDs at all waves. These results suggest that SEs also explain substance-specific SUD (Tables S17 – S20).

3.2.4. Summary

Overall, results provided strong evidence for cross-substance effects. Correlations between SEs and SUD across substances were significant, particularly at Wave 1. Legal and illegal SE factors predicted legal and illegal SUD in adolescence and adulthood.

3.3. Do SEs predict change in SUD over time?

Overall, mean SUD (lifetime at Wave 1; past year at Waves 2 and 3) decreased significantly from Wave 1 to 3 for most substances, except for cocaine and amphetamines (Table S20). Alcohol, cannabis, and cocaine slope variances were not significant, suggesting lack of variability to examine differential predictors of change in SUD over time. Associations between SEs and change in SUD from adolescence to adulthood for models with significant slope variance (tobacco, hallucinogens, amphetamines, and opioids; Table S21) indicated only positive amphetamine SEs significantly predicted greater decrease in SUD (β = −.280; Table 5). Although this negative association appears contradictory to positive associations between SEs and SUD at each wave, amphetamine SEs predicted amphetamine SUD only at Wave 1. The negative association indicates a floor effect; individuals with lower initial SEs and CDsymp decreased less in SUD over time.

Table 5.

Change in SUD Criteria Over Time for Individual Substances Regressed on SEs and CD Symptoms, Controlling for Sex

	SUD Slope of Individual Substance			SUD Slope of Individual Substance
	β	p-value		β	p-value
Tobacco SUD (N = 608–610)
Tobacco Positive SEs	−.016 [−.039, .007]	.161	Tobacco Negative SEs	−.008 [−.031, .015]	.489
CD Symptoms	−.009 [−.018, .001]	.068	CD Symptoms	−.010* [−.019, −.001]	.032
Hallucinogen SUD (N = 275)
Hallucinogen Positive SEs	.021 [−.158, .201]	.815	Hallucinogen Negative SEs	.004 [−.187, .195]	.969
CD Symptoms	−.021* [−.039, −.003]	.023	CD Symptoms	−.020* [−.038, −.002]	.025
Amphetamine SUD (N = 220)
Amphetamine Positive SEs	−.280** [−.487, −.072]	.008	Amphetamine Negative SEs	−.216* [−.407, −.026]	.026
CD Symptoms	−.021* [−.039, −.003]	.023	CD Symptoms	−.019* [−.036, −.003]	.022
Opioid SUD (N = 178)
Opioid Positive SEs	.239 [−.053, .531]	.109	Opioid Negative SEs	.087 [−.265, .439]	.630
CD Symptoms	−.017 [−.048, .015]	.304	CD Symptoms	−.018 [−.052, .015]	.284

^*p < .05

^**p ≤ .01

Bolded estimate: FDR p < .05

Note: Reported Ns include individuals with data for both SEs and SUDs for that specific substance.

3.4. Are there sex or racial/ethnic differences in associations between predictors and SUD?

We found no evidence for racial/ethnic group differences in associations (Tables S22, S23). Limited number of girls/women in our sample prevented direct testing of sex differences in associations between SEs and SUDs. We instead regressed sex on SUD of each substance (Tables S24, S25).

4. DISCUSSION

The present study aimed to increase understanding of the predictors of SUDs across development in a high-risk sample with greater odds of sustained SUD. Extant literature primarily focuses on retrospectively reported predictors for commonly used substances in community samples. Our study extends prior research by addressing whether CD symptoms and SEs for legal and illegal substances assessed during adolescence predict general versus specific SUD from adolescence to adulthood in a longitudinal clinical sample.

Contrary to our hypothesis that SEs would predict specific SUD, results indicated that positive and negative SEs were robust predictors of general SUD. SEs also predicted substance-specific SUD, although less reliably. SEs for legal substances predicted general legal SUD from adolescence to adulthood, whereas SEs for illegal substances predicted general illegal SUD mostly in adolescence. Notably, effect sizes for associations between SEs and legal and illegal SUD were similar, but sample sizes were smaller and confidence intervals were wider in analyses addressing illegal SUDs. Overall, these results indicate that SEs may be relevant predictors for individuals engaging in both legal and illegal polysubstance use.

Our results suggest that intensity and quantity of sensation experienced may be more important predictors than valence, as associations between SEs and SUD were similar in magnitude across positive and negative SEs. Although the result that unpleasant experiences did not deter use seems counterintuitive, they align with prior research on substance use or SUD in a combined community/clinical sample (Zeiger et al., 2010) and community youth samples (Ríos-Bedoya et al., 2009; Zeiger et al., 2012). Prior research suggests that some individuals need higher doses of substances (specifically alcohol) to feel both sedative and stimulant effects, indicating individuals using more of a substance might report more SEs in general (Chung & Martin, 2009). Some individuals at high SUD risk may be “high responders” who are sensitive to both positive and negative substance effects (Chung & Martin, 2009; Zeiger et al., 2012). Lastly, significant associations between negative SEs and SUDs have been found in adolescent or very young adult populations (Ríos-Bedoya et al., 2009; Zeiger et al., 2010, 2012). Adolescents are less deterred by negative consequences and more motivated by rewards, specifically in the context of substance use (Bjork et al., 2007; Colder et al., 2013; Van Leijenhorst et al., 2010). Adolescents experiencing positive and negative SEs may be less deterred by negative physiological drug experiences and more motivated by positive experiences, in congruence with prior research (Colder et al., 2013).

Cross- and within-substance associations between SEs and SUD were equally robust, for both negative and positive SEs. Our results align with prior research finding associations between positive and negative SEs across cannabis, tobacco, and alcohol at one timepoint in a mixed clinical and community sample (Zeiger et al., 2012). Cross-substance effects in our results were general: legal and illegal SE factors predicted general legal and illegal SUD factors in adolescence and adulthood, providing clear evidence for cross-substance effects of SEs on polysubstance SUD for commonly and less commonly used substances.

CDsymp primarily predicted general SUD in adolescence, in contrast to evidence suggesting that CD predicts general SUD over time (Erskine et al., 2016; Palmer et al., 2013). We examined individuals ascertained for SUDs with high CD symptom rates (M = 5.70, SD = 2.95) and CD symptom range restriction may have limited the associations between CDsymp and adult SUDs. However, CDsymp—but not SUD—predicted later antisocial behavior (Border et al., 2018a) and mortality (Border et al., 2018b) in an overlapping sample.

Our clinical sample had high levels of SUDs, allowing us to investigate predictors of significant change in SUD over time. Overall, SUD severity decreased for most substances. However, neither SEs nor CDsymp consistently predicted change in SUD. Very few studies have examined change in SUD severity over time in clinical subjects, and additional research is needed to replicate these results. We also found no evidence for racial/ethnic differences in associations between SEs and SUD.

Substance-specific SEs, both positive and negative, were significantly correlated across substances. Many individuals in our sample met diagnostic criteria for SUD of multiple substances at Wave 1 and may have been using different substances simultaneously, and polysubstance use may have led to difficulties in discerning substance-specific SEs. Lastly, substances may act on common biological mechanisms, although different substances may act on the same neurotransmitters to varying degrees (Wise & Robble, 2020).

4.1. Limitations

First, we could not evaluate whether sex moderated associations; due to small numbers of women, multigroup analyses would not converge. Second, we replicated prior evidence that two-factor models (positive and negative SEs) fit significantly better than one-factor models for most substances but model nonconvergence prevented model fit comparisons for opioids. Third, results were most robust at Wave 1 than Waves 2 and 3. Analyses for the later waves likely had less power due to subject attrition—nearly 53% of individuals were lost to follow-up by Wave 3. Notably, greater Wave 1 CDsymp severity or SUD did not predict attrition (Table S1). Fourth, we collapsed all individuals who were not non-Hispanic White into one group due to small sample size of other racial/ethnic groups. Lastly, participants reported on SEs experienced “shortly after using” and may have referred to general experiences with a substance, recent use, or initial use (Zeiger et al., 2012). Regular or heavy users may have developed tolerance and reported SEs may have differed than those initially experienced.

4.2. Strengths

The present study’s longitudinal design is less vulnerable to recall bias than retrospective studies (Hardt & Rutter, 2004) because our assessment of SEs was more proximal to first use compared to existing retrospective research, as substance experimentation typically begins in adolescence. Mean age of first substance use in this sample ranged from 11.57–14.35 years (Table S26). We expanded the existing literature by examining a clinical, rather than community, sample ascertained for SUDs and CDsymp. CFA results replicated previous findings of a two-factor model of positive and negative SEs and provided evidence of a two-factor model of legal and illegal SUDs in a clinical sample. Finally, our sample’s racial and ethnic diversity facilitated examination of racial/ethnic differences in associations.

4.3. Conclusion

In a clinical sample, SEs predicted polysubstance SUD from adolescence through adulthood, whereas CDsymp predicted SUD primarily in adolescence only. SEs predicted general and substance-specific SUD, but prediction of general SUD was more robust. Associations between negative SEs and SUD were similar or stronger in magnitude compared to those between positive SEs and SUD. We demonstrated significant cross-substance associations between SEs and SUD of multiple substances. Finally, associations between SEs and SUD were consistent across race/ethnic groups. Overall, our results suggest that individuals at elevated risk for SUDs who experience high intensity and quantity of SEs likely have greater SUD severity, across multiple substances and into adulthood. In contrast, CDsymp were not strong longitudinal predictors of SUD in this clinical sample. Assessing SEs early and prospectively may help clinicians identify adolescents who are at highest risk for sustained SUDs.