Prediction of incidence and stability of alcohol use disorders by latent internalizing psychopathology risk profiles in adolescence and young adulthood

Accepted Manuscript Title: Prediction of incidence and stability of alcohol use disorders by latent internalizing psychopathology risk profiles in adolescence and young adulthood Authors: Silke Behrendt, Gerhard Buhringer, ¨ Michael H¨ofler, Roselind Lieb, Katja Beesdo-Baum PII: DOI: Reference:

S0376-8716(17)30330-7 http://dx.doi.org/doi:10.1016/j.drugalcdep.2017.06.006 DAD 6538

To appear in:

Drug and Alcohol Dependence

Received date: Revised date: Accepted date:

25-11-2016 2-6-2017 3-6-2017

Please cite this article as: Behrendt, Silke, Buhringer, ¨ Gerhard, H¨ofler, Michael, Lieb, Roselind, Beesdo-Baum, Katja, Prediction of incidence and stability of alcohol use disorders by latent internalizing psychopathology risk profiles in adolescence and young adulthood.Drug and Alcohol Dependence http://dx.doi.org/10.1016/j.drugalcdep.2017.06.006 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Prediction of incidence and stability of alcohol use disorders by latent internalizing psychopathology risk profiles in adolescence and young adulthood

Behrendt, Silke1; Bühringer, Gerhard1,2; Höfler, Michael1; Lieb, Roselind4; Beesdo-Baum, Katja1,3

1

Institute of Clinical Psychology and Psychotherapy, Technische Universität Dresden, Chemnitzer Str. 46, D-01187 Dresden, Germany;

2

IFT Institut für Therapieforschung, Parzivalstr. 25, D-80804 Munich, Germany;

3

Behavioral Epidemiology, Technische Universität Dresden, Chemnitzer Str. 46, D-01187 Dresden, Germany;

4

Department of Psychology, Division of Clinical Psychology and Epidemiology, University Basel, Missionsstr. 60-62, 4055 Basel, Switzerland

Correspondence: Silke Behrendt, PhD Institute of Clinical Psychology and Psychotherapy Technische Universität Dresden Chemnitzer Str. 46 D-01187 Dresden Phone: +49 351 463-39860 Fax: +49 351 463-39830 E-mail: [email protected]

1

Highlights  Different latent psychopathology profiles predicted DSM-IV alcohol dependence onset  Different latent psychopathology profiles predicted problematic alcohol use  A male-normative profile predicted problematic alcohol use and disorder onset

Abstract Background: Comorbid internalizing mental disorders in alcohol use disorders (AUD) can be understood as putative independent risk factors for AUD or as expressions of underlying shared psychopathology vulnerabilities. However, it remains unclear whether: 1) specific latent internalizing psychopathology risk-profiles predict AUD-incidence and 2) specific latent internalizing comorbidity-profiles in AUD predict AUD-stability. Aims: To investigate baseline latent internalizing psychopathology risk profiles as predictors of subsequent AUD-incidence and -stability in adolescents and young adults. Methods: Data from the prospective-longitudinal EDSP study (baseline age 14 – 24 years) were used. The study-design included up to three follow-up assessments in up to ten years. DSM-IV mental disorders were assessed with the DIA-X/M-CIDI. To investigate risk-profiles and their associations with AUD-outcomes, latent class analysis with auxiliary outcome variables was applied. Results: AUD-incidence: a 4-class model (N=1683) was identified (classes: normative-male [45.9%], normative-female [44.2%], internalizing [5.3%], nicotine dependence [4.5%]). Compared to the normative-female class, all other classes were associated with a higher risk of subsequent incident alcohol dependence (p<0.05). AUD-stability: a 3-class model (N=1940) was identified with only one class (11.6%) with high probabilities for baseline AUD. This class was further characterized by elevated substance use disorder (SUD) probabilities and predicted any subsequent AUD (OR 8.5, 95% CI 5.4 – 13.3).

2

Conclusions: An internalizing vulnerability may constitute a pathway to AUD incidence in adolescence and young adulthood. In contrast, no indication for a role of internalizing comorbidity profiles in AUD-stability was found, which may indicate a limited importance of such profiles – in contrast to SUD-related profiles – in AUD stability.

Keywords: alcohol dependence, latent class analysis, risk factor, mental disorders, community youth, prospective-longitudinal

1.

Introduction

Alcohol use disorders (DSM-IV alcohol abuse or dependence, AUD) are associated with substantial morbidity and mortality (Rehm et al., 2009; Rehm et al., 2013). Identifying risk factors for AUD onset and stability is crucial for tailoring preventive and intervention efforts (Perkonigg et al., 1998). Prospective-longitudinal epidemiological studies have identified a range of different mental disorders (MD) as risk factors for AUD onset, including several anxiety and substance use disorders (SUD), bipolar and conduct disorder (Behrendt et al., 2011; Brückl et al., 2007; Elkins et al., 2007; Lieb et al., 2016; Swendsen et al., 2010). However, existing evidence is somewhat more consistent for externalizing compared to internalizing MD in SUD prediction (Farmer et al., 2015). Thus, the importance of an ‘internalizing pathway’ to AUD onset warrants further research as it remains unclear, whether specific internalizing MD (Buckner et al., 2008), specific combinations thereof (Grant et al., 2015a) or a general underlying internalizing vulnerability (Elkins et al., 2006) might contribute to the inconsistent associations between internalizing MD and AUD. Many studies on MD as risk factors for AUD have carefully controlled for other putatively important MD to ensure an association is not merely an artefact of comorbidity. Undoubtedly, this is important for understanding the role of specific internalizing MD in AUD aetiology. However, the identification of a specific MD as a single risk factor beyond 3

the role of other MD may bear a conceptual disadvantage as it does not take into account the putative role of specific combinations of MD risk or underlying vulnerabilities. High MD comorbidity and heterotypic continuity frequently found in AUD (de Graaf et al., 2003; Hasin et al., 2007; Jacobi et al., 2015; Jacobi et al., 2014; Kessler et al., 2012; Lahey et al., 2014; Lai et al., 2015; Zimmermann et al., 2003) can be understood as expressions of underlying shared vulnerabilities (Krueger and Markon, 2006). Several epidemiological studies in adult and adolescent/young adult populations on the comorbidity structure of MD find separate externalizing and internalizing underlying factors (Beesdo-Baum et al., 2009; Kendler et al., 2003; Kessler et al., 2011a), however limited stability of the model in different age groups and upon inclusion of additional MD diagnoses has been noted (Wittchen et al., 2009). Moreover, underlying vulnerability factors can be modelled as mediators in the association between prior MD and other MD incidence (Kessler et al., 2011a). Given evidence for high comorbidity, underlying vulnerabilities, heterotypic continuity and the range of different MD identified as AUD risk factors, it is important for the clarification of the role of internalizing risk factors of AUD to go beyond single internalizing risk factor identification and retrospective identification of factors within AUD subtypes (Babor and Caetano, 2006; Cloninger et al., 1996) and determine internalizing MD risk profiles (i.e., clusters of risk for several internalizing MD with or without other MD) that predict AUD onset. Identifying such profiles can further the understanding of internalizing AUD risk factors that relies on single risk factor identification by identifying specific combinations of internalizing MD risk and their roles in AUD onset and create information for targeted prevention and intervention. To our best knowledge, so far, one cross-sectional study has investigated associations between MD profiles and a latent factor representing DSM-IV AUD-criteria (Harford et al., 2015). Therefore, we aim to investigate the prospective association between latent groups with different risk profiles of a range of baseline DSM-IV MD and the risk of subsequent 4

AUD incidence. As, with few exceptions (Lopez-Quintero et al., 2011), much less evidence is available for the role of internalizing MD in AUD stability compared to onset, and as correlates of stable vs. unstable AUD have been shown to differ (Hicks et al., 2010), we also aim to investigate in an explorative fashion latent AUD comorbidity profiles in a community sample and their prospective association with AUD stability. For the prediction of AUD incidence, we hypothesize to find one baseline latent risk profile with a high risk of SUD diagnoses other than AUD, one with a high risk of internalizing MD including somatoform MD (Kendler et al., 2011) and one with low MD probabilities. For the prediction of AUD stability we hypothesize to find the same overall profile structure with AUD-risk being elevated within one SUD-related and one internalizing risk profile. Given existing results on MD as AUD risk factors, we expect the SUD-related and the internalizing profiles to be associated with a higher risk of subsequent AUD incidence and stability, however with stronger associations for SUD-related profiles. 2.

Methods

2.1

Sample and Overall Design

The EDSP study is a 10-year prospective-longitudinal community study that has been described elsewhere (Beesdo-Baum et al., 2015; Lieb et al., 2000; Wittchen et al., 1998b). In short, the aim of the EDSP study is to investigate the course and risk-factors for substance use and SUD in a stratified sample of N= 3021 subjects aged 14–24 years at baseline. Focusing on early developmental stages of psychopathology, individuals aged 14–15 years were sampled at twice the probability of those aged 16–21 years. Individuals aged 22–24 years were sampled at half the probability of subjects aged 16–21 years. The baseline sample was drawn from metropolitan Munich (German government registries). Subjects were followed-up over a 10-year period with up to three follow-up examinations. The baseline survey took place in 1995 (T0, N= 3021). Follow-up examinations were conducted approximately after 1.6 years (T1, median interval since baseline, only for the younger cohort of N= 1228 subjects 5

aged 14–17 years at baseline), 3.5 years (T2) and 8.2 years (T3). The response rates (proportion of T0 sample) was 70.8% at T0 (N= 3021), 84.3% (N= 2548) at T2 and 73.2% (N= 2210) at T3. The T3 age range was 21–34 years. 2.2 Diagnostic Assessment At each study wave, assessments were conducted with computer-assisted baseline and follow-up versions of the Munich-Composite International Diagnostic Interview (DIA-X/MCIDI) (Wittchen et al., 1998a; Wittchen and Pfister, 1997). The M-CIDI is a fully standardized diagnostic interview for epidemiological research (Wittchen et al., 1998a), designed to assess symptoms, syndromes and diagnoses of 48 MD. The diagnoses presented here are based on computerized M-CIDI/DSM-IV algorithms. Test–retest reliability and validity of the DIA-X/M-CIDI diagnoses have been established (Reed et al., 1998; Wittchen, 1994; Wittchen et al., 1998a). DSM-IV AUD, alcohol use (AU), nicotine dependence (ND), and any illegal drug use disorder (DUD) were assessed with the respective DIA-X/M-CIDI-sections for nicotine, alcohol, and medication and illegal substance use that have been described elsewhere in detail (Behrendt et al., 2009). 2.3

Statistical Analysis

For descriptive analysis conducted with the Stata Software package 14.1 (StataCorp., 2015), data were weighted to account for different sampling probabilities at baseline, and response rates at baseline varying over age, sex, and geographic location. 2.3.1 Latent Class Analysis (LCA) with Auxiliary Outcome Variables. To empirically identify groups with different MD risk profiles at baseline and to investigate the associations between these groups and observed AUD outcomes we applied LCA (Skrondal and RabeHesketh, 2004) with auxiliary outcome variables (here: AUD-outcomes) using the DCATand BCH-method in MPlus version 7.31.

6

In line with our research goals to investigate AUD incidence and stability separately, two types of analysis with different subsamples were conducted: when AUD incidence was the outcome of interest (‘incidence model’) we excluded cases with baseline AUD (n=257; unweighted mean age of AUD onset: 16.8 years; SD: 2.2) to enable identification of baseline latent MD risk profiles that predict incident AUD at follow-up (n=352, unweighted mean age of AUD onset: 17.6 years, SD: 3.0). The respective weighted mean ages of onset were 17.0 years (SD: 2.3) before and 18.2 years (SD: 3.4) after baseline. When AUD stability was the outcome of interest (‘stability model’) we included baseline AUD to enable identification of baseline latent AUD comorbidity profiles that predict any subsequent AUD. In addition, for both analyses, subjects were excluded if they missed T2 or T3 or declined answering drug questions truthfully, leading to N=1683 for the incidence model and N=1940 for the stability model. LCA is a probabilistic classification method based on the assumption that the association structure among a set of variables can be reduced to at least two latent classes and that given class status, the variables in the set are independent from one another within the latent classes (“local independence”) (Muthén and Muthén, 1998-2006; Nylund et al., 2007). We fitted models with two to six (stability model) respectively two to five classes (incidence model). 6000 random sets of starting values were generated in the initial, 60 optimizations were carried out in the final stage1. Calculations were repeated to ensure that the solution was the global maximum of the likelihood function. The Bayesian information criterion (BIC), the sample size adjusted BIC (ABIC), and Akaike’s information criterion (AIC) were applied as model fit indices with lower values indicating better model fit. Model fit was additionally

1

Stability model: the number of LRT starts had to be increased to 120 40 400 120 in the 2-class and 60 20 200 60 in the 5-class model because the best loglikelihood value was not replicated in the bootstrap draws. The number of sets of starting values had to be increased to 32000 with 320 optimizations in the six-class model. Incidence model: the number of LRT starts had to be increased to 60 20 200 60 in the 3- and 4-class model because the best loglikelihood value was not replicated in the bootstrap draws.

7

assessed with the Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (LRT), the Lo-MendellRubin adjusted LRT, and the Parametric Bootstrapped LRT. When the number of classes was identified, the auxiliary outcome variable was added to the model using Lanza’s method (DCAT-method, for binary outcomes) and the BCHmethod (for continuous outcomes) to avoid class shifts when adding the auxiliary variable (Asparouhov and Muthén, 2014a, b; Lanza et al., 2013). The DCAT-method cannot be applied with weights. To ensure that this unweighted approach was appropriate, two class variables derived from respectively a weighted and unweighted application of the LCA-model estimation of the 3-Step approach described by Vermunt (2010) were transferred to Stata. Cross-tabulation and Cramer’s V were then used to assess the agreement in classification between the weighted and the unweighted approach. Hereby, maximum posterior probabilities were applied to estimate class membership. Unweighted analyses with the DCAT- and BCHmethod were only applied if agreement in classification was high. Although it has been reported that the BCH-method can be robust in case of non-normally distributed continuous distal outcomes (Asparouhov and Muthén, 2014b), we compared BCH-results with and without box-cox-transformation for all continuous outcomes. In case transformation led to different results, these are reported. 2.3.2 Variables in the LCA with Auxiliary Outcome Variables. For both models, binary indicator variables in the LCA were gender and DSM-IV baseline lifetime ND, DUD, panic disorder, generalized anxiety disorder, social phobia, specific phobia (for phobias, impairment was only requested for DSM-IV diagnosis after age 17 because these disorders occur especially early and because of the putatively limited reliability of impairment reports in young subjects (Wittchen et al., 1999a; Wittchen et al., 1999b)), any eating disorder, and any somatoform disorder or syndrome (somatoform disorder). The stability model additionally included baseline lifetime alcohol abuse and dependence (non-hierarchical). An unordered categorical indicator was used for affective disorders (0=none, 1=major depression or 8

dysthymia, 2= bipolar disorder). As some externalizing-spectrum diagnoses like conduct disorder were unavailable for the entire sample, SUD diagnosis were applied to enable identification of putative internalizing, SUD-related, and negative affect-SUD (Hussong et al., 2011) risk profiles. Given robust findings on externalizing vulnerability dimensions underlying SUD (Kendler et al., 2003; Kessler et al., 2011b), pure SUD-related profiles are seen as indicative of an externalizing vulnerability. Alcohol-related subsequent outcomes occurred between baseline and ten-year follow up (based on follow-up prevalence rates at T2 and T3). These outcomes were incident (incidence model) or any (stability model) DSM-IV alcohol abuse (non-hierarchical), alcohol dependence, and AUD as well as any hazardous alcohol use (HAU1: > 12 [females] or 24 [males] gram ethanol/day; HAU2: > 20 [females] or 40 [males] gram ethanol/day), binge drinking2, any SUD (without ND), and any anxiety or depressive disorder3. Additionally baseline lifetime characteristics (age at first AU4, average gram ethanol/day) were considered as outcomes. 3.

Results

3.1

Incidence model

3.1.1 LCA Solution. The fit indices BIC and ABIC were smallest (i.e., indicated the best fit) for the 2-class solution, the fit indice AIC was smallest for the 4-class solution. For both solutions all three LRT-tests were significant, indicating improved model fit in comparison to a model with one class less. The 2-class solution had much lower classification quality (entropy: 0.68), so the 4-class solution (entropy: 0.89) was chosen (see Table 1). Cross-tabulation and Cramer’s V (0.94) indicated minimal differences between the latent class variables derived from a weighted and unweighted application of the 3-Step approach,

2

Binge drinking: >=five (males) or four (females) standard drinks à 9 gram ethanol Any DSM-IV alcohol or illegal drug use disorder respectively any DSM-IV major depression, dysthymia or anxiety disorder¸ added to investigate putative internalizing/externalizing pathways 4 Age at onset of alcohol use was available for n=1505 (incidence model) and n=1758 (stability model) 3

9

demonstrating appropriateness of the application of the unweighted DCAT- and BCHapproach (not shown in table). 3.1.2 Class Description. The respective proportions of the four classes were 45.9%5 (class 1), 4.5% (class 2), 5.3% (class 3), and 44.2% (class 4) (based on the estimated model). Class 1, named “Normative-male”, was characterized by a 0.0 probability of female gender and low disorder probabilities (shown in table 2, column 2). Class 2 (“Nicotine dependence”) had a 0.64 probability of female gender. Probabilities were high for ND (0.92), and elevated for somatoform disorder (0.35) and illegal SUD (0.23). Class 3 (“Internalizing”) had a high probability of female gender (0.87) and elevated probabilities for unipolar affective disorder (0.64), specific phobia (0.39), any somatoform disorder (0.31), social phobia (0.25), and ND (0.21). Class 4 (“Normative-female”) had a 1.0 probability of female gender and low disorder probabilities. 3.1.3 Results for Alcohol-Related Outcomes. As shown in table 3, at baseline, classes 1-3 did not differ from class 4 (reference class) by mean age at AU onset (p>=0.05). A significantly higher mean amount of ethanol per day was found for class 1 (mean: 5.5 gram) and class 2 (mean: 10.8 gram) compared to class 4 (mean: 2.7 gram, p<0.05). In prospective analyses, class 1 was associated with a higher risk of alcohol dependence onset (probability: 0.10) compared to class 4 (probability 0.02), a higher risk of alcohol abuse and any AUD (ORs 4.9 – 6.6). Compared to class 4, classes 2 and 3 were associated with a higher risk of incident alcohol dependence (ORs 18.2 and 5.7). The risk of subsequent HAU1 was elevated in classes 1-3, the risk of HAU2 in classes 1 and 2. Class 1 was associated with a higher risk of subsequent binge drinking. Classes 1-3 were associated with a higher risk of any subsequent SUD, classes 2 and 3 with any subsequent internalizing disorder. 3.2

Stability Model

5

Class counts are available upon request.

10

3.2.1 LCA Solution. BIC and ABIC were lowest (i.e., indicated best model fit) for the 3-class solution. The AIC decreased to the 5-class solution. The LRT indicated a significant (p<0.05) improvement in model fit until three and a borderline-significant improvement for five classes (in comparison to models with one class less). The parametric bootstrapped LRT indicated improvement for up to five classes. Based on the lowest BIC and ABIC, the more parsimonious 3-class solution (entropy: 0.77) was chosen (see Table 4). Cross-tabulation and a Cramer’s V of 0.96 indicated few minor differences between the two class variables derived from a weighted and unweighted application of the 3-Step approach, demonstrating appropriateness of the application of the unweighted DCAT- and BCH-approach (not in table). 3.2.2 Class Description. Proportions of the three classes were 11.6% (class 1), 10.7% (class 2), and 77.8% (class 3). Class 1, named “Substance use disorder”, was characterized by high probabilities for male gender (0.95) and alcohol abuse (0.64), and elevated probabilities for ND (0.40), alcohol dependence (0.32), and DUD (0.22) (see column 2, table 5). Class 2 (“Mixed-internalizing”) had a high probability of female gender (0.89). The highest disorder probabilities were 0.40 (ND), 0.40 (unipolar affective disorder), and 0.39 (somatoform disorder). Elevated probabilities showed for specific (0.25) and social phobia (0.18). Class 3 (“Normative class”) had an almost 0.5 probability of female gender and low disorder probabilities. 3.2.3 Results for Alcohol-Related Outcomes. At baseline, a significantly higher mean amount of ethanol per day was found for class 1 (mean: 36.8 gram) and class 2 (mean: 8.3 gram) compared to class 3 (mean: 3.1 gram, p<=0.05; see table 6). Mean age of alcohol use onset was lower in class 1 than in class 3 (box-cox-transformed, p<0.05). In prospective analyses, class 1 was associated with a higher risk of subsequent alcohol abuse (OR 5.1), 11

alcohol dependence (OR 22.7), HAU1, HAU2, and binge drinking, compared to class 3. Class 2 was associated with a higher risk of subsequent HAU1, HAU2, and internalizing disorder. 4.

Discussion

To our best knowledge, this is the first study that investigates the relationship between different latent MD risk profiles and subsequent AUD onset and stability in a community sample of adolescents and young adults. The main findings are: AUD incidence was predicted by different latent classes with respectively an internalizing MD profile, a nicotine dependence–profile, and a profile with high male gender and low disorder probabilities. As for the prediction of AUD-stability by latent AUD-comorbidity profiles, a class with elevated probabilities for AUD and other SUD was identified which predicted AUD at follow-up. No class with an internalizing MD-AUD comorbidity profile was found.

4.1 Limitations Data are representative for a relatively wealthy region in Germany. Different population characteristics may lead to different findings. Due to the focus on internalizing MD, a range of these were included as baseline indicators. The range of available externalizing MD diagnoses was more restricted. However, given the well-documented underlying vulnerability for externalizing disorders we expect the included different SUD diagnoses to be sufficiently informative for the latent profile structure, i.e., permitting identification of pure internalizing, pure SUD-related and mixed risk profiles. Drop-outs were excluded, but this is not expected to alter the observed associations between the latent risk profiles and AUD outcomes. Lifetime baseline diagnoses were used as indicators. This precludes insight into the temporal sequence of indicators and conclusions on specific developmental periods (Hussong et al., 2011) within adolescence and young adulthood in which risk profiles might pose the greatest risk for AUD outcomes, but enables covering the full disorder spectrum in the sample. A narrower age-rage might enable more specific 12

conclusions on selected developmental phases. In the stability model, entropy was modest but not poor (Asparouhov and Muthén, 2014b). To enable a strictly prospective approach, baseline data was used for risk profile identification. Given the sample’s age range, this data may have matched the incidence phases of AUD sub-optimally (Behrendt et al., 2009), leading to an attenuation of observed associations in the incidence model. Differences in associated correlates and consequences have been noted between adolescent-onset and early adulthood-onset AUD (Hicks et al., 2010). Here, some subjects with especially early AUD onset may have been excluded from the incidence model which might have led to limited consideration of such subtypes of AUD development. However, differences in age of onset distributions of those with AUD onset before and after baseline were small. Some CIs for the ORs for the associations with subsequent outcomes were wide, limiting the interpretability of these ORs.

4.2 Risk Profiles and Prediction of AUD in the Incidence Model The latent class solution that fit the data best in the incidence model included, apart from a normative- male and a normative-female class two classes with elevated nicotine dependence respectively internalizing disorder probabilities. The two latter classes and the normative-male class were associated with a higher risk of incident alcohol dependence but differed in their associations with incident alcohol abuse and drinking behaviors. Overall, these findings may suggest the existence of different psychopathology pathways to AUD in adolescence and young adulthood. The association between the nicotine dependence-class and AUD-outcomes could possibly be understood as an expression of an underlying vulnerability for externalizing disorders and homotypic continuity (Beesdo-Baum et al., 2012; Hicks et al., 2004; Kessler et al., 2011a). The internalizing class showed higher probabilities for all internalizing disorders compared to the other classes, including disorders as specific phobia, a risk factor for alcohol dependence 13

(Behrendt et al., 2011). In contrast, no indication of latent classes with specific combinations of internalizing disorders was found. Thus, the finding may suggest that not specific internalizing disorder combinations but an overall internalizing vulnerability is predictive of a heterotypic continuity to AUD, possibly especially among women (Foster et al., 2015). The internalizing class was not, however, associated with alcohol abuse or most forms of problematic drinking under consideration. This largely agrees with the observation that associations between single internalizing predictors and SUD outcomes (Behrendt et al., 2011; Buckner et al., 2008; Swendsen et al., 2010) are found less consistently than associations for externalizing predictors (Farmer et al., 2015; Lahey et al., 2014; Lieb et al., 2016). The lacking association with alcohol abuse might be due to the disruption of social norms in alcohol abuse being more likely in subjects with externalizing MD. Alternatively, the degree of disability or of functionality in terms of self-medication (Thomas et al., 2003), subjective suffering or other disorder features including early onset, stability or occurrence in specific vulnerability periods (Asselmann et al., 2014; Behrendt et al., 2011; Hussong et al., 2011) in internalizing disorders could be more important for associations with AUD than the dichotomous diagnoses investigated here. Of importance for preventive efforts, the finding of an elevated risk of subsequent HAU in the internalizing class indicates that subjects with this risk profile, apart from their risk for incident alcohol dependence, are at risk for problematic drinking behavior associated with an elevated morbidity risk (Rehm et al., 2010). The finding of the consistent associations between the normative-male class with low disorder probabilities and all AUD- and problematic drinking outcomes is of interest for preventive efforts as it may indicate the existence of a large subgroup of predominantly male subjects at risk for AUD and problematic drinking in spite of a low psychopathology burden. As the elevated risk of this group can likely not be attributed to higher psychopathology liabilities, other factors than threshold MD might be influential in this group. Overall, the finding of the normative-male class within the latent profile structure of the incidence model 14

is in line with the conclusion that internalizing and externalizing liabilities may constitute “an incomplete indication of overall risk” (Farmer et al., 2015) of SUD as they are only prevalent in a minority (Dawson et al., 2010). 4.3 Risk Profiles and Prediction of AUD in the Stability Model Against our expectation we found no comorbidity profile characterized by high AUD and internalizing disorder probabilities. Instead, AUD probability was only elevated within a class with elevated probabilities for other SUD, giving no indication of MD-related AUDsubtypes (Babor and Caetano, 2006) in this sample. Internalizing MD probabilities were elevated in a class with negligible AUD probabilities. While differences in included variables, statistical methods and study designs limit comparability, the two classes described above do roughly agree with the externalizing-internalizing structure underlying comorbidity risk that has been identified in epidemiological studies (Beesdo-Baum et al., 2009; Harford et al., 2015; Kendler et al., 2011; Kendler et al., 2003; Kessler et al., 2011a; Kessler et al., 2011b), higher prevalence rates of unipolar mood and anxiety disorders in females and higher rates of AUD and DUD in males (Hasin and Grant, 2015; Jacobi et al., 2015; Kessler et al., 2012). It may be the case that internalizing comorbidity profiles are rare as distinct AUD comorbidity patterns and thus not relevant for the prediction of AUD stability in adolescence and young adulthood. However, the underlying vulnerability structure has proven unstable upon inclusion of further factors (Wittchen et al., 2009) and in this context two indicators investigated here warrant further discussion: ND and somatoform disorder. Both classes had a comparable likelihood for ND, which can be regarded as an externalizing disorder (Marmorstein et al., 2009) but is often not included in studies on underlying vulnerability factors (Beesdo-Baum et al., 2009; Harford et al., 2015; Hicks et al., 2004; Kendler et al., 2003; Kessler et al., 2011a). ND may occur as a result of an underlying vulnerability for externalizing disorders, however, it is also predictive of and predicted by internalizing disorders (Behrendt et al., 2011; Grant et al., 2015b; Lahey et al., 2014; Marmorstein et al., 15

2009; Swendsen et al., 2010). Our results might indicate that ND may occur in the context of an externalizing as well as an internalizing profile. In the latter, factors other than an externalizing vulnerability as self-medication or reward learning (Diehl and Scherbaum, 2008) may play a core role in ND development. Many studies on shared MD liabilities do not include somatoform disorder (BeesdoBaum et al., 2009; Harford et al., 2015; Kessler et al., 2011a; Lahey et al., 2014). Here, somatoform disorder was among the indicators with the highest probability in the mixedinternalizing class. Probabilities in the SUD- and normative class were low. This agrees with a study on underlying comorbidity structures, in which one factor that loaded on (sub)threshold somatoform and on anxiety and eating disorders and major depression (Kendler et al., 2011) was found. The SUD-class in the stability-model predicted subsequent DSM-IV AUD. The observation of this association is in line with previous consistent findings of cross-sectional and prospective associations between SUD (Dawson et al., 2010; Harford et al., 2015; Swendsen et al., 2010) and the prediction of AUD by an underlying externalizing vulnerability (Kessler et al., 2011a). However, as the highest probability for an AUD in this class was 0.6, the results for this class might be best understood as an expression of continued homotypic vulnerability for SUD instead of definite AUD stability. 4.4 Implications The identification of latent risk profiles could help evaluating the comorbidity profiles of subjects for the adaptive planning of preventive and treatment measures for AUD. While our findings warrant replication, it may be important for targeted preventive efforts that different

risk

profiles,

namely

SUD-related,

internalizing,

and

male-gender/low

psychopathology profiles predict incident alcohol dependence in adolescents and young adults. The large male-normative group possibly warrants low-threshold preventive efforts. A comorbidity profile characterized by SUD-related but not internalizing psychopathology 16

appeared especially relevant for targeting AUD stability. Of further importance for prevention, internalizing profiles were associated with subsequent problematic drinking. Future research should broaden the range of indicators to social and psychological factors to identify factors related to risk among males with low psychopathology burden, to externalizing disorders as conduct disorder, and to MD features as severity and impairment. In the absence of specific internalizing MD combinations that predict AUD, the underlying mechanisms in the heterotypic continuity from internalizing vulnerability to AUD needs to be investigated. In summary, investigating latent MD profiles and their associations with subsequent AUD incidence we found that different latent profiles (internalizing, nicotine dependencerelated, normative-male) are associated with an elevated risk of subsequent incident alcohol dependence, putatively indicating different pathways to incident AUD and different prevention needs. Investigating latent AUD comorbidity profiles and their association with AUD stability we found no indication of a profile characterized by high AUD and internalizing MD probabilities being relevant for AUD continuity, instead results suggested a homotypic continuity of SUD.

Author Disclosures

Role of the Funding Source This work is part of the Early Developmental Stages of Psychopathology (EDSP) Study and is funded by the German Federal Ministry of Education and Research (BMBF), project numbers 01EB9405/6, 01EB 9901/6, EB01016200, 01EB0140 and 01EB0440. Some of the fieldwork and analyses was also supported by grants from the Deutsche Forschungsgemeinschaft (DFG) LA1148/1-1, WI2246/1-1, WI 709/7-1 and WI 709/8-1. The principal investigators are Dr. H.-U. Wittchen and Dr. R. Lieb. Core staff members of the EDSP group are: Dr. K. von

17

Sydow, Dr. G. Lachner, Dr. A. Perkonigg, Dr. P. Schuster, Dr. M. Höfler, H. Sonntag, Dr. T. Brückl, E. Garczynski, Dr. B. Isensee, Dr. A. Nocon, Dr. C. Nelson, H. Pfister, Dr. V. Reed, B. Spiegel, Dr. A. Schreier, Dr. U. Wunderlich, Dr. P. Zimmermann, Dr. K. Beesdo-Baum, Dr. A. Bittner, Dr. S. Behrendt and Dr. S. Knappe. Scientific advisers are Dr. J. Angst (Zurich), Dr. J. Margraf (Basel), Dr. G. Esser (Potsdam), Dr. K. Merikangas (NIMH, Bethesda), Dr. R. Kessler (Harvard, Boston) and Dr. J. van Os (Maastricht). The EDSP project and its family genetic supplement have been approved by the Ethics Committee of the Medical Faculty of the Technische Universitaet Dresden (no: EK-13811). All participants provided informed consent.

Contributors SB conceptualized and wrote the manuscript and conducted the analysis. MH contributed to the analysis and the writing of the manuscript. All other authors contributed to the writing of the manuscript. All authors have approved the final version of the manuscript.

Conflict of interest Gerhard Bühringer has received unrestricted gambling research grants from the Bavarian State Ministry of Finance (regulatory authority for and operator of the State Gambling Monopoly) via the Bavarian State Ministry of Public Health and Care Services, from the German Federal Ministry of Economics and Technology (regulatory authority for parts of the commercial gambling industry), and from public and commercial gambling providers. The other authors state that they do not have a conflict of interest.

18

References Asparouhov, T., Muthén, B., 2014a. Auxiliary Variables in Mixture Modeling: 3-Step Approaches Using Mplus. Mplus Web Notes: No. 15. Version 8, August 5, 2014. Asparouhov, T., Muthén, B., 2014b. Auxiliary Variables in Mixture Modeling: Using the BCH Method in Mplus to Estimate a Distal Outcome Model and an Arbitrary Secondary Model Mplus Web Notes: No. 21, Version 2. Asselmann, E., Wittchen, H.-U., Lieb, R., Hoefler, M., Beesdo-Baum, K., 2014. Associations of fearful spells and panic attacks with incident anxiety, depressive, and substance use disorders: A 10-year prospective-longitudinal community study of adolescents and young adults. J. Psychiatr. Res. 55, 8-14. Babor, T.F., Caetano, R., 2006. Subtypes of substance dependence and abuse: Implications for diagnostic classification and empirical research. Addiction 101, 104-110. Beesdo-Baum, K., Hoefler, M., Gloster, A.T., Klotsche, J., Lieb, R., Beauducel, A., Buehner, M., Kessler, R.C., Wittchen, H.-U., 2009. The structure of common mental disorders: A replication study in a community sample of adolescents and young adults. Int. J. Methods Psychiatr. Res. 18, 204-220. Beesdo-Baum, K., Knappe, S., Asselmann, E., Zimmermann, P., Brückl, T., Höfler, M., Behrendt, S., Lieb, R., Wittchen, H.-U., 2015. The ‘Early Developmental Stages of Psychopathology (EDSP) study’: A 20 years review of methods and findings. Soc. Psychiatry Psychiatr. Epidemiol. 50, 851-866. Beesdo-Baum, K., Low, N.C.P., Knappe, S., Behrendt, S., Hofler, M., Lieb, R., Wittchen, H.U., 2012. Age-specific cumulative incidence risk for mental disorders as a function of familial liability. Eur. Neuropsychopharmacol. 22, S423-S424. Behrendt, S., Beesdo-Baum, K., Zimmermann, P., Höfler, M., Perkonigg, A., Bühringer, G., Lieb, R., Wittchen, H.U., 2011. The role of mental disorders in the risk and speed of

19

transition to alcohol use disorders among community youth. Psychol. Med. 41, 1073– 1085. Behrendt, S., Wittchen, H.-U., Höfler, M., Lieb, R., Beesdo, K., 2009. Transitions from first substance use to substance use disorders in adolescence: Is early onset associated with a rapid escalation? Drug Alcohol Depend. 99 68–78. Brückl, T.M., Wittchen, H.-U., Höfler, M., Pfister, H., Schneider, S., Lieb, R., 2007. Childhood separation anxiety and the risk of subsequent psychopathology: Results from a community study. Psychother. Psychosom. 76, 47-56. Buckner, J.D., Schmidt, N.B., Lang, A.R., Small, J.W., Schlauch, R.C., Lewinsohn, P.M., 2008. Specificity of social anxiety disorder as a risk factor for alcohol and cannabis dependence. J. Psychiatr. Res. 42, 230-239. Cloninger, C.R., Sigvardsson, S., Bohman, M., 1996. Type I and type II alcoholism: An update. Alcohol Health Res. World 20, 18-23. Dawson, D.A., Goldstein, R.B., Moss, H.B., Li, T.-K., Grant, B.F., 2010. Gender differences in the relationship of internalizing and externalizing psychopathology to alcohol dependence: Likelihood, expression and course. Drug Alcohol Depend. 112, 9-17. de Graaf, R., Bijl, R.V., Spijker, J., Beekman, A.T.F., Vollebergh, W.A.M., 2003. Temporal sequencing of lifetime mood disorders in relation to comorbid anxiety and substance use disorders. Findings from the Netherlands Mental Health Survey and Incidence Study. Soc. Psychiatry Psychiatr. Epidemiol. 38, 1-11. Diehl, A., Scherbaum, N., 2008. Nicotine dependence as comorbidity of alcohol dependence Epidemiology, etiology and therapy. Fortschritte der Neurologie-Psychiatrie 76, 1420. Elkins, I.J., King, S.M., McGue, M., Iacono, W.G., 2006. Personality traits and the development of Nicotine, Alcohol and Illicit Drug Disorders: Prospective Links from Adolescence to young Adulthood. J. Abnorm. Psychol. 115, 26-39. 20

Elkins, I.J., McGue, M., Iacono, W.G., 2007. Prospective effects of attentiondeficit/hyperactivity disorder, conduct disorder, and sex on adolescent substance use and abuse. Arch. Gen. Psychiatry 64, 1145-1152. Farmer, R.F., Seeley, J.R., Kosty, D.B., Gau, J.M., Duncan, S.C., Lynskey, M.T., Lewinsohn, P.M., 2015. Internalizing and externalizing psychopathology as predictors of cannabis use disorder onset during adolescence and early adulthood. Psychol. Addict. Behav. 29, 541-551. Foster, K.T., Hicks, B.M., Iacono, W.G., McGue, M., 2015. Gender differences in the structure of risk for alcohol use disorder in adolescence and young adulthood. Psychol. Med. 45, 3047-3058. Grant, B.F., Goldstein, R.B., Saha, T.D., Chou, S.P., Jung, J., Zhang, H.T., Pickering, R.P., Ruan, W.J., Smith, S.M., Huang, B.J., Hasin, D.S., 2015a. Epidemiology of DSM-5 alcohol use disorder results from the national epidemiologic survey on alcohol and related conditions III. JAMA Psychiatry 72, 757-766. Grant, J.D., Lynskey, M.T., Madden, P.A.F., Nelson, E.C., Few, L.R., Bucholz, K.K., Statham, D.J., Martin, N.G., Heath, A.C., Agrawal, A., 2015b. The role of conduct disorder in the relationship between alcohol, nicotine and cannabis use disorders. Psychol. Med. 45, 3505-3515. Harford, T.C., Yi, H.-y., Chen, C.M., Grant, B.F., 2015. Psychiatric symptom clusters as risk factors for alcohol use disorders in adolescence: A national study. Alcohol. Clin. Exp. Res. 39, 1174-1185. Hasin, D.S., Grant, B.F., 2015. The National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) Waves 1 and 2: Review and summary of findings. Soc. Psychiatry Psychiatr. Epidemiol. 50, 1609-1640. Hasin, D.S., Stinson, F.S., Ogburn, E., Grant, B.F., 2007. Prevalence, correlates, disability, and comorbidity of DSM-IV alcohol abuse and dependence in the United States. 21

Results from the national epidemiologic survey on alcohol and related conditions. Arch. Gen. Psychiatry 64, 830-842. Hicks, B.M., Iacono, W.G., McGue, M., 2010. Consequences of an Adolescent onset and persistent course of alcohol dependence in men: Adolescent risk factors and adult outcomes. Alcohol. Clin. Exp. Res. 34, 819-833. Hicks, B.M., Krueger, R.F., Iacono, W.G., McGue, M., Patrick, C.J., 2004. Family transmission and heritability of externalizing disorders - A twin-family study. Arch. Gen. Psychiatry 61, 922-928. Hussong, A.M., Jones, D.J., Stein, G.L., Baucom, D.H., Boeding, S., 2011. An internalizing pathway to alcohol use and disorder. Psychol. Addict. Behav. 25, 390-404. Jacobi, F., Hoefler, M., Strehle, J., Mack, S., Gerschler, A., Scholl, L., Busch, M.A., Hapke, U., Maske, U., Seiffert, I., Gaebel, W., Maier, W., Wagner, M., Zielasek, J., Wittchen, H.-U., 2015. Twelve-months prevalence of mental disorders in the German Health Interview and Examination Survey for Adults - Mental Health Module (DEGS1-MH): a methodological addendum and correction. Int. J. Methods Psychiatr. Res. 24, 305313. Jacobi, F., Hofler, M., Siegert, J., Mack, S., Gerschler, A., Scholl, L., Busch, M.A., Hapke, U., Maske, U., Seiffert, I., Gaebel, W., Maier, W., Wagner, M., Zielasek, J., Wittchen, H.U., 2014. Twelve-month prevalence, comorbidity and correlates of mental disorders in Germany: The Mental Health Module of the German Health Interview and Examination Survey for Adults (DEGS1-MH). Int. J. Methods Psychiatr. Res. 23, 304-319. Kendler, K.S., Aggen, S.H., Knudsen, G.P., Roysamb, E., Neale, M.C., Reichborn-Kjennerud, T., 2011. The structure of genetic and environmental risk factors for syndromal and subsyndromal common DSM-IV axis I and all axis II disorders. Am. J. Psychiatry 168, 29-39. 22

Kendler, K.S., Prescott, C.A., Myers, J., Neale, M.C., 2003. The structure of genetic and environmental risk factors for common psychiatric and substance use disorders in men and women. Arch. Gen. Psychiatry 60, 929-937. Kessler, R.C., Avenevoli, S., Costello, E.J., Georgiades, K., Green, J.G., Gruber, M.J., He, J.p., Koretz, D., McLaughlin, K.A., Petukhova, M., Sampson, N.A., Zaslavsky, A.M., Merikangas, K.R., 2012. Prevalence, persistence, and sociodemographic correlates of DSM-IV disorders in the national comorbidity survey replication adolescent supplement. Arch. Gen. Psychiatry 69, 372-380. Kessler, R.C., Ormel, J., Petukhova, M., McLaughlin, K.A., Green, J.G., Russo, L.J., Stein, D.J., Zaslavsky, A.M., Aguilar-Gaxiola, S., Alonso, J., Andrade, L., Benjet, C., de Girolamo, G., de Graaf, R., Demyttenaere, K., Fayyad, J., Haro, J.M., Hu, C.Y., Karam, A., Lee, S., Lepine, J.-P., Matchsinger, H., Mihaescu-Pintia, C., Posada-Villa, J., Sagar, R., Uestuen, T.B., 2011a. Development of lifetime comorbidity in the world health organization world mental health surveys. Arch. Gen. Psychiatry 68, 90-100. Kessler, R.C., Petukhova, M., Zaslavsky, A.M., 2011b. The role of latent internalizing and externalizing predispositions in accounting for the development of comorbidity among common mental disorders. Curr. Opin. Psychiatry 24, 307-312. Krueger, R.E., Markon, K.E., 2006. Reinterpreting comorbidity: A model-based approach to understanding and classifying psychopathology. Ann. Rev. Clin. Psychol. 2, 111-133. Lahey, B.B., Zald, D.H., Hakes, J.K., Krueger, R.F., Rathouz, P.J., 2014. Patterns of heterotypic continuity associated with the cross-sectional correlational structure of prevalent mental disorders in adults. JAMA Psychiatry 71, 989-996. Lai, H.M.X., Cleary, M., Sitharthan, T., Hunt, G.E., 2015. Prevalence of comorbid substance use, anxiety and mood disorders in epidemiological surveys, 1990-2014: A systematic review and meta-analysis. Drug Alcohol Depend. 154, 1-13.

23

Lanza, S.T., Tan, X.M., Bray, B.C., 2013. Latent class analysis with distal outcomes: A flexible model-based approach. Struct. Equat. Model. 20, 1-26. Lieb, R., Isensee, B., von Sydow, K., Wittchen, H.-U., 2000. The Early Developmental Stages of Psychopathology Study (EDSP): A methodological update. Eur. Addict. Res. 6, 170-182. Lieb, R., Miche, M., Gloster, A.T., Beesdo-Baum, K., Meyer, A.H., Wittchen, H.-U., 2016. Impact of specific phobia on the risk of onset of mental disorders: A 10-year prospective-longitudinal community study of adolescents and young adults. Depress. Anxiety 0, 1-9. Lopez-Quintero, C., Perez de los Cobos, J., Hasin, D.S., Okuda, M., Wang, S., Grant, B.F., Blanco, C., 2011. Probability and predictors of transition from first use to dependence on nicotine, alcohol, cannabis, and cocaine: Results of the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). Drug Alcohol Depend. 115, 120-130. Marmorstein, N.R., Iacono, W.G., McGue, M., 2009. Alcohol and illicit drug dependence among parents: Associations with offspring externalizing disorders. Psychol. Med. 39, 149-155. Muthén, L., Muthén, B. (Eds.), 1998-2006. Mplus User’s Guide Muthén and Muthén., Los Angeles, CA. Nylund, K.L., Asparoutiov, T., Muthen, B.O., 2007. Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Struct. Equat. Model. 14, 535-569. Perkonigg, A., Lieb, R., Wittchen, H.-U., 1998. Substance use, abuse and dependence in Germany. A review of selected epidemiological data. Eur. Addict. Res. 4, 8-17. Reed, V., Gander, F., Pfister, H., Steiger, A., Sonntag, H., Trenkwalder, C., Sonntag, A., Hundt, W., Wittchen, H.-U., 1998. To what degree does the Composite International 24

Diagnostic Interview (CIDI) correctly identify DSM-IV Disorders? Testing validity issues in a clinical sample. Int. J. Methods Psychiatr. Res. 7, 142-155. Rehm, J., Baliunas, D., Borges, G.L.G., Graham, K., Irving, H., Kehoe, T., Parry, C.D., Patra, J., Popova, S., Poznyak, V., Roerecke, M., Room, R., Samokhvalov, A.V., Taylor, B., 2010. The relation between different dimensions of alcohol consumption and burden of disease: an overview. Addiction 105, 817-843. Rehm, J., Mathers, C., Popova, S., Thavorncharoensap, M., Teerawattananon, Y., Patra, J., 2009. Alcohol and Global Health 1 Global burden of disease and injury and economic cost attributable to alcohol use and alcohol-use disorders. Lancet 373, 2223-2233. Rehm, J., Shield, K.D., Gmel, G., Rehm, M.X., Frick, U., 2013. Modeling the impact of alcohol dependence on mortality burden and the effect of available treatment interventions in the European Union. Eur. Neuropsychopharmacol. 23, 89-97. Skrondal, A., Rabe-Hesketh, S. (Eds.), 2004. Generalized latent variable modeling. Chapman and Hall/CRC., Washington D.C. StataCorp. (Ed.), 2015. Stata Statistical Software: Release 14.1. StataCorp LP. College Station, TX. Swendsen, J., Conway, K.P., Degenhardt, L., Glantz, M., Jin, R., Merikangas, K.R., Sampson, N., Kessler, R.C., 2010. Mental disorders as risk factors for substance use, abuse and dependence: Results from the 10-year follow-up of the National Comorbidity Survey. Addiction 105, 1117-1128. Thomas, S.E., Randall, C.L., Carrigan, M.H., 2003. Drinking to cope in socially anxious individuals: A controlled study. Alcohol. Clin. Exp. Res. 27, 1937-1943. Vermunt, J.K., 2010. Latent class modeling with covariates: Two improved three-step approaches. Politic. Anal. 18, 450-469. Wittchen, H.-U., 1994. Reliability and validity studies of the WHO-Composite International Diagnostic Interview (CIDI): A critical review. J. Psychiatr. Res. 28, 57-84. 25

Wittchen, H.-U., Beesdo-Baum, K., Gloster, A.T., Hoefler, M., Klotsche, J., Lieb, R., Beauducel, A., Buehner, M., Kessler, R.C., 2009. The structure of mental disorders reexamined: Is it developmentally stable and robust against additions? Int. J. Methods Psychiatr. Res. 18, 189-203. Wittchen, H.-U., Lachner, G., Wunderlich, U., Pfister, H., 1998a. Test-retest reliability of the computerized DSM-IV version of the Munich-Composite International Diagnostic Interview (M-CIDI). Soc. Psychiatry Psychiatr. Epidemiol. 33, 568-578. Wittchen, H.-U., Lieb, R., Schuster, P., Oldehinkel, A.J., 1999a. When is onset? Investigations into early developmental stages of anxiety and depressive disorders. In: Rapoport, J.L. (Ed.), Childhood onset of "adult" psychopathology, clinical and research advances. American Psychiatric Press, Washington. pp. 259-302. Wittchen, H.-U., Perkonigg, A., Lachner, G., Nelson, C.B., 1998b. Early Developmental Stages Of Psychopathology Study (EDSP): Objectives and design. Eur. Addict. Res. 4, 18-27. Wittchen, H.-U., Pfister, H., 1997. DIA-X Interview. Instruktionsmanual zur Durchführung von DIA-X-Interviews (Instruction manual for the DIA-X-Interview) Swets and Zeitlinger, Frankfurt. Wittchen, H.U., Stein, M.B., Kessler, R.C., 1999b. Social fears and social phobia in a community sample of adolescents and young adults: Prevalence, risk factors and comorbidity. Psychol. Med. 29, 309-323. Zimmermann, P., Wittchen, H.-U., Pfister, H., Kessler, R.C., Lieb, R., 2003. Primary anxiety disorders and the development of subsequent alcohol use disorder: A 4-year community study of adolescents and young adults. Psychol. Med. 33, 1211-1222.

26

Figure Legends

Figure 1: Item probabilities for gender and baseline lifetime mental disorders (incidence model)

Figure 2: Item probabilities for gender and baseline lifetime mental disorders (stability model)

27

28

29

Table 1: Model-fit-information for the different latent class solutions in the incidence model (N=1683) 2-class solution 3-class solution

4-class solution

5-class solution

Loglikelihood

-4183.410

-4162.376

-4142.454

-4130.588

Number of free parameters

23

35

47

59

Akaike's Information Criterion (AIC)1 Bayesian Information Criterion (BIC) Sample-size adjusted Bayesian Information Criterion (ABIC)

8412.820 8537.671

8394.751 8584.743

8378.908 8634.039

8379.175 8699.447

8464.604

8473.553

8484.726

8512.012

Vuong-Lo-Mendell-Rubin Likelihod Ratio Test H0 Loglikelihood Value P-value

-4315.915 <0.0001

-4183.410 0.1195

-4162.376 0.0113

-4142.454 0.1018

Lo-Mendell-Rubin Adjusted Likelihood Ratio Test Value P-value

261.807 <0.0001

41.510 0.1230

39.314 0.0121

23.417 0.1046

Parametric bootstrapped Likelihood Ratio Test Value Approximate p-Value

-4315.915 <0.0001

-4183.410 <0.0001

-4162.376 <0.0001

-4142.454 0.2500

Entropy2 0.682 0.788 0.890 1AIC, BIC, ABIC: lower values indicate better model fit 2Quality of classification Summary: BIC and ABIC were smallest for the 2-class solution, the AIC for the 4-class solution. For both solutions, all three LRT-tests were significant. The 2-class solution had lower entropy (0.68), so the 4-class solution (entropy: 0.89) was chosen.

0.927

30

Table 2: Class proportions (%) and estimated indicator probabilities for the 4-class solution in the incidence model Reference: total sample Class 1 (45.94 Class 2 (4.47 Class 3 Baseline indicators (N=1683) %)12 %)12 %)12 4 3 Gender (female) 54.82 0.00 0.64 0.87 5 Nicotine dependence 12.92 0.06 0.92 0.21 Any illegal drug use disorder5 1.89 0.01 0.23 0.04 5 Major depression or dysthymia 12.28 0.05 0.16 0.64 Bipolar disorder I or II5 1.67 0.00 0.08 0.14 5 Panic disorder 1.10 0.00 0.00 0.12 Generalized anxiety disorder5 1.92 0.00 0.08 0.19 5 Social phobia 3.87 0.02 0.08 0.25 5 Specific phobia 6.42 0.05 0.07 0.39 Any somatoform disorder or syndrom5 12.39 0.06 0.35 0.31 5 Any eating disorder 2.50 0.00 0.05 0.10 1In brackets: class proportions (%) based on the estimated model 2Class 1 "Normative-male"; class 2 "Nicotine dependence"; class 3 "Internalizing"; class 4 "Normative-female" 3Estimated probabilities for the baseline lifetime indicators 4Weighted proportions (%) in the entire sample; unweighted information available upon request 5Mental disorders and syndromes assessed according to DSM-IV

(5.34

Class %)12 1.00 0.07 0.00 0.08 0.00 0.01 0.00 0.03 0.08

4

(44.23

0.13 0.02

31

Table 3: Equality test of probabilities and means across classes for auxiliary outcome variables in the incidence model Auxiliary variable: baseline characteristics Mean

S.E.

Chi2

p-value

Gram ethanol/day Class 11

5.50

0.45

25.19

<0.001

Class 2

10.78

2.86

7.49

0.006

Class 3

4.58

1.42

1.55

0.212

Class 4

2.69

0.33

REF

Age at first Alcohol Use4 (years) Class 1

13.47

0.08

2.63

0.105

Class 2

13.14

0.40

1.58

0.209

Class 3

13.74

0.36

0.02

0.864

Class 4

13.67

0.09

REF

Auxiliary variable: outcome at follow-up Probability3

OR

95% CI

DSM-IV Alcohol Abuse Class 1

0.28

4.93

3.5 - 6.8

Class 2

0.26

4.42

1.03 - 18.9

Class 3

0.13

1.90

0.7 - 4.6

Class 4

0.07

REF

Class 1

0.10

6.60

3.3 - 13.0

Class 2

0.24

18.17

5.9 - 55.4

Class 3

0.09

5.73

1.8 - 18.3

Class 4

0.02

REF

DSM-IV Alcohol Dependence

DSM-IV Alcohol Use Disorder Class 1

0.32

4.94

3.6 - 6.7

Class 2

0.31

4.53

1.3 - 15.3

Class 3

0.17

2.17

0.9 - 4.8

Class 4

0.09

REF

Hazardous Alcohol Use (HAU1; >12/24 gram/day)2 Class 1

0.33

1.52

1.1 - 2.0

Class 2

0.53

3.59

1.8 - 7.1

32

Class 3

0.41

2.19

Class 4

0.24

REF

1.1 - 4.2

Hazardous Alcohol Use (HAU2; >20/40 gram/day)2 Class 1

0.16

1.66

1.1 - 2.3

Class 2

0.44

6.76

2.9 - 15.6

Class 3

0.19

1.99

0.9 - 4.3

Class 4

0.11

REF

Class 1

0.77

2.15

1.7 - 2.7

Class 2

0.76

2.06

0.3 - 12.6

Class 3

0.61

1.02

0.4 - 2.3

Class 4

0.61

REF

Class 1

0.37

4.68

3.5 - 6.2

Class 2

0.60

11.72

2.2 - 60.2

Class 3

0.25

2.59

1.2 - 5.5

Class 4

0.11

REF

Binge Drinking

Any Substance Use Disorder

Any Anxiety or Depressive Disorder Class 1

0.20

0.55

0.4 - 0.7

Class 2

0.61

3.48

1.4 - 8.4

Class 3

0.83

10.69

4.7 - 23.9

Class 4

0.31

REF

1

Class 1 "Normative-male"; class 2 "Nicotine dependence"; class 3 "Internalizing"; class 4 "Normative-female" More than 12 (females) or 24 (males) gram ethanol per day; more than 20 (females) or 40 (males) gram ethanol per day 2

3

Probability of the outcome of interest in the respective class n=1505 (subjects with available age of onset information)

4

Table 4: Model-fit-information for the different latent class solutions in the stability model (N=1940)3 2-class solution

3-class solution

4-class solution

5-class solution

6-class solution

Loglikelihood

-6164.164

-6034.550

-6003.870

-5986.286

-5973.203

Number of free parameters

27

41

55

69

83

Akaike's Information Criterion (AIC)1

12382.328

12151.099

12117.741

12110.573

12112.406

33

Bayesian Information Criterion (BIC)

12532.730

12379.487

12424.115

12494.933

12574.753

Sample-size adjusted Bayesian Information Criterion (ABIC)

12446.950

12249.230

12249.379

12275.719

12311.060

H0 Loglikelihood Value

-6421.258

-6164.164

-6034.550

-6003.870

-5986.286

P-value

<0.0001

<0.0001

0.3525

0.0471

1.0000

Value

509.383

256.406

60.690

34.785

25.882

P-value

<0.0001

<0.0001

0.3561

0.0485

1.0000

H0 Loglikelihood Value

-6421.258

-6164.164

-6034.550

-6003.870

-5986.286

Approximate p-Value

<0.0001

<0.0001

<0.0001

0.0400

0.6667

Entropy2

0.738

0.770

0.725

0.859

0.864

Vuong-Lo-Mendell-Rubin Likelihod Ratio Test

Lo-Mendell-Rubin Adjusted Likelihood Ratio Test

Parametric bootstrapped Likelihood Ratio Test

1AIC,

BIC, ABIC: lower values indicate better model

fit 2Quality of classification 3Summary: BIC and ABIC were lowest for the 3-class solution. The AIC decreased to the 5-class solution. The LRT indicated a significant (p<0.05) improvement in model fit for two and three and a borderline-significant improvement for five classes. The parametric bootstrapped LRT indicated improvement for up to five classes. Based on the lowest BIC and ABIC, the more parsimonious 3-class solution was chosen.

34

Table 5: Class proportions (%) and estimated indicator probabilities for the 3-class solution in the stability model Reference: total sample Class 1 (11.55 Class 2 (10.67 Class Baseline indicators (N=1940) %)12 %)12 %)12 4 3 Gender (female) 49.80 0.05 0.89 0.49 5 Alcohol abuse 14.04 0.64 0.15 0.03 Alcohol dependence5 5.88 0.32 0.08 0.00 5 Nicotine dependence 16.76 0.40 0.40 0.07 Any illegal drug use disorder5 4.18 0.22 0.11 0.00 5 Major depression or dysthymia 12.52 0.13 0.40 0.06 Bipolar disorder I or II5 2.13 0.04 0.11 0.00 5 Panic disorder 1.62 0.00 0.12 0.00 5 Generalized anxiety disorder 2.09 0.01 0.14 0.00 Social phobia5 4.07 0.02 0.18 0.03 5 Specific phobia 6.00 0.04 0.25 0.06 Any somatoform disorder or syndrom 5 12.38 0.08 0.39 0.09 5 Any eating disorder 2.67 0.01 0.11 0.01 1In brackets: class proportions (%) based on the estimated model 2Class 1 "Substance use disorders"; class 2 "Mixed-internalizing"; class 3 "Normative class" 3Estimated probabilities for the baseline lifetime indicators 4Weighted proportions (%) in the entire sample; unweighted information available upon request 5Mental disorders and syndromes assessed according to DSMIV

3

(77.77

35

Table 6: Equality test of probabilities and means across classes for auxiliary outcome variables in the stability model

Auxiliary variable: baseline characteristics Mean

S.E.

Chi2

p-value

Gram ethanol/day Class 11

36.81

2.99

119.35

<0.001

Class 2

8.29

1.54

9.90

0.002

Class 3

3.12

0.31

REF

Age at first Alcohol Use56 (years) Class 1

100.29

2.54

3.96

0.047

Class 2

105.08

3.10

0.06

0.799

Class 3

105.95

0.97

REF

Auxiliary variable: outcome at follow-up Probability3

OR

95% CI

DSM-IV Alcohol Abuse Class 1

0.53

5.10

3.3 - 7.8

Class 2

0.08

0.37

0.1 - 0.9

Class 3

0.18

REF

DSM-IV Alcohol Dependence Class 1

0.55

22.73

13.5 - 38.3

Class 2

0.07

1.40

0.4 - 4.6

Class 3

0.05

REF

DSM-IV Alcohol Use Disorder Class 1

0.69

8.50

5.4 - 13.3

Class 2

0.09

0.38

0.1 - 0.9

Class 3

0.21

REF

Hazardous Alcohol Use (HAU1; >12/24 gram/day)2

36

Class 1

0.68

5.35

3.5 - 8.0

Class 2

0.44

2.01

1.3 - 3.1

Class 3

0.28

REF

Hazardous Alcohol Use (HAU2; >20/40 gram/day)2 Class 1

0.51

6.92

4.6 - 10.4

Class 2

0.26

2.34

1.3 - 4.0

Class 3

0.13

REF

Class 1

0.94

6.64

3.0 - 14.4

Class 2

0.60

0.66

0.4 - 1.0

Class 3

0.69

REF

Class 1

0.91

na4

na

Class 2

0.18

0.74

0.3 - 1.7

Class 3

0.22

REF

Binge Drinking

Any Substance Use Disorder

Any Anxiety or Depressive Disorder Class 1

0.25

1.04

0.6 - 1.7

Class 2

0.78

11.23

7.1 - 17.7

Class 3

0.24

REF

1

Class 1 "Substance use disorders"; class 2 "Mixed-internalizing"; class 3 "Normative class" More than 12 (females) or 24 (males) gram ethanol per day; more than 20 (females) or 40 (males) gram ethanol per day 2

3

Probability of the outcome of interest in the respective class

4

Not applicable because of low proportion of subjects without the outcome of interest

5

n=1758 (subjects with available age of onset information)

6

Continuous distal outcome age at onset was box-cox-transformed

37