Introduction
Within the United States (U.S.), alcohol is the most misused psychoactive substance (Grant et al., Reference Grant, Goldstein, Saha, Chou, Jung, Zhang and Hasin2015; Kessler et al., Reference Kessler, Berglund, Demler, Jin, Merikangas and Walters2005). Most persons with an alcohol use disorder (AUD) history will experience at least one other lifetime psychiatric disorder, typically a non-alcohol substance use disorder, or DRUG (Grant et al., Reference Grant, Goldstein, Saha, Chou, Jung, Zhang and Hasin2015; Kessler et al., Reference Kessler, Crum, Warner, Nelson, Schulenberg and Anthony1997). During 2020, nearly 1 in 4 persons in the U.S. with a past-year AUD also had a DRUG diagnosis within the same period (Substance Abuse and Mental Health Services Administration, 2021). Despite the high prevalence of AUD/DRUG comorbidity, only a small number of long-term prospective studies have investigated factors that increase risk for the misuse of additional substances among those who misuse alcohol (Anthony, Barondess, Radovanovic, & Lopez-Quintero, Reference Anthony, Barondess, Radovanovic, Lopez-Quintero and Sher2016).
Accumulating evidence suggests that persons with lifetime AUD/DRUG comorbidities constitute a distinct and potentially important subgroup among those with AUD histories. Genetic modeling studies, for example, indicate substantial genetic overlap between AUD and DRUGs (Kendler, Prescott, Myers, & Neale, Reference Kendler, Prescott, Myers and Neale2003; Xian et al., Reference Xian, Scherrer, Grant, Eisen, True, Jacob and Bucholz2008). There is also evidence for unique environmental influences associated with individual substance use disorder (SUD) categories inclusive of AUD (Xian et al., Reference Xian, Scherrer, Grant, Eisen, True, Jacob and Bucholz2008), SUD-specific genetic risks (Kendler et al., Reference Kendler, Prescott, Myers and Neale2003), and separate licit and illicit drug genetic factors (Kendler, Myers, & Prescott, Reference Kendler, Myers and Prescott2007). Overall, a broad liability for problematic substance misuse appears to account for at least some of the heterogeneity among those with AUD histories.
Persons with comorbid AUD/DRUG, when compared to those with AUD as the only lifetime SUD, are also distinguished by greater AUD severity (Grant et al., Reference Grant, Goldstein, Saha, Chou, Jung, Zhang and Hasin2015; Saha et al., Reference Saha, Grant, Chou, Kerridge, Pickering and Ruan2018). The extent to which variations in AUD severity are explained by risk factors, AUD/DRUG comorbidity, or other processes that emerge during AUD or DRUG episodes, however, remains unclear. Similarly, psychosocial and health-related functioning are more impaired among persons with AUD/DRUG comorbidity v. AUD alone (Abé et al., Reference Abé, Mon, Durazzo, Pennington, Schmidt and Meyerhoff2013; Connor, Gullo, White, & Kelly, Reference Connor, Gullo, White and Kelly2014; McCabe, West, Jutkiewicz, & Boyd, Reference McCabe, West, Jutkiewicz and Boyd2017; Saha et al., Reference Saha, Grant, Chou, Kerridge, Pickering and Ruan2018). Differences in functioning, however, may have less to do with AUD/DRUG comorbidity or historical risk factors than overall SUD severity, a variable that is also associated with poorer psychosocial outcomes (Dawson, Saha, & Grant, Reference Dawson, Saha and Grant2010; Kirisci, Vanyukov, Dunn, & Tarter, Reference Kirisci, Vanyukov, Dunn and Tarter2002; Kirisci et al., Reference Kirisci, Tarter, Vanyukov, Martin, Mezzich and Brown2006).
The present study
Studies summarized above collectively suggest that histories of AUD alone v. AUD/DRUG comorbidity likely represent distinct behavioral phenotypes. Uncertainty exists, however, about whether the course of these conditions and associated psychosocial outcomes are primarily explained by processes or experiences that occurred early in development or emerged later. Early experiences rarely have enduring effects independent of circumstances that arise later in development (Rutter & Sroufe, Reference Rutter and Sroufe2000). Although early experiences or processes may have contributed to developmental trajectories that culminated in AUD alone v. AUD/DRUG comorbidity, for example, more proximal synergistic effects of drug comorbidity or the severity of substance involvement might ultimately have greater influence on subsequent outcomes. Two organizing concepts in developmental psychopathology are particularly pertinent in investigations of such questions: equifinality and multifinality (Cicchetti & Rogosch, Reference Cicchetti and Rogosch1996). Equifinality suggests that a common outcome (e.g. AUD) is often produced by diverse sets of processes or experiences. Multifinality, in comparison, suggests that diverse outcomes often result from the same event or condition. Although a group of persons may have AUD in common, for example, the subsequent outcomes of AUD may substantially differ among affected individuals.
Demonstrations that persons with comorbid AUD/DRUG histories display etiologically distinct psychosocial pathways compared to those with AUD alone (i.e. equifinality) would support these phenotype distinctions, as would findings that show significant differences in the course and outcomes of AUD among members of these two groups (i.e. multifinality). Correspondingly, we investigated putative risk factors, AUD and SUD course features, and age 30 psychosocial outcomes for four mutually exclusive groups defined by persons (1) without AUD or DRUG histories (or NEITHER); (2) with AUD but not DRUG histories (or AUD ONLY); (3) without AUD but with DRUG histories (or DRUG ONLY); and (4) with AUD and DRUG histories (or BOTH). Because of limited relevant research, our primary hypotheses are provisional:
• Risk-related pathways that culminate in AUD ONLY v. BOTH would be distinguished in magnitude and kind.
• AUD-related course features would differ between AUD ONLY and BOTH groups, with the latter demonstrating a more severe course overall.
• Group differences in AUD course features would be partially explained by risk factors.
• BOTH would demonstrate poorer age 30 outcomes compared to AUD ONLY, with observed differences partially or fully explained by overall SUD severity, risk factors, and demographic variables.
Method
Participants
Probands
The Oregon Adolescent Depression Project (OADP) began as a two-panel study (T1, T2; ~ ages 16 and 17, respectively) of randomly selected high school students. The T1 sample (n = 1709) was demographically similar to corresponding census data for the region (Lewinsohn, Hops, Roberts, Seeley, & Andrews, Reference Lewinsohn, Hops, Roberts, Seeley and Andrews1993). At T3 (~ age 24), a stratified sampling procedure was implemented whereby recruitment involved all persons with a psychiatric disorder history by T2 (n = 644), all racial and ethnic minorities (to increase sample diversity), and a randomly selected subgroup without a psychiatric history by T2 (n = 457 of 863 persons). Around participants’ 30th birthday, a fourth panel (T4, n = 816) was initiated. From T1 to T2, T2 to T3, and T3 to T4, retention rates for eligible probands were 88, 85 and 87%, respectively.
Extensive analyses of participant attrition (Farmer, Kosty, Seeley, Olino, & Lewinsohn, Reference Farmer, Kosty, Seeley, Olino and Lewinsohn2013; Lewinsohn et al., Reference Lewinsohn, Hops, Roberts, Seeley and Andrews1993; Seeley, Farmer, Kosty, & Gau, Reference Seeley, Farmer, Kosty and Gau2019) revealed minimal sample biases related to study discontinuation, although higher rates of attrition were observed (a) for men compared to women, (b) by T2 among those with an externalizing behavior disorder or any SUD diagnosis undifferentiated by substance type, and (c) after T3 among those with histories of AUD or DRUG.
Family members
Around T3, adult first-degree family members of probands were interviewed about their psychiatric histories. Lifetime psychiatric data were available for 732 of 816 families (90%). Of these, 490 (67%) families produced data on all first-degree relatives and 242 (33%) provided data on some members (M = 62% of all first-degree relatives within families) but missing data on one or more members. As reported elsewhere (Farmer et al., Reference Farmer, Seeley, Gau, Klein, Merikangas, Kosty and Lewinsohn2018), no differences on proband AUD history status were observed as a function of missing family data although there were some demographic differences related to missingness in some comparisons. Overall, 2414 adult first-degree relatives (730 biological mothers, 719 biological fathers, 476 female siblings, 489 male siblings) contributed data to this research.
Diagnostic assessments
Probands
Probands participated in semi-structured diagnostic interviews at each assessment wave during which SUD and psychiatric disorder diagnostic criteria were assessed for the 12-months prior to interview, lifetime (at T1), and ever-since-last-interview (at T2, T3, T4), thus permitting continuous representation of proband diagnostic status from childhood through age 30. Diagnostic interviews involved the Schedule for Affective Disorders and Schizophrenia for School-Age Children (K-SADS, from T1 through T3; Chambers et al., Reference Chambers, Puig-Antich, Hirsch, Paez, Ambrosini, Tabrizi and Davies1985) and the Structured Clinical Interview for Axis I DSM-IV Disorders–Non-Patient Edition (SCID-NP, at T4; First, Spitzer, Gibbon, and Williams, Reference First, Spitzer, Gibbon and Williams1994). At T1, lifetime psychiatric disorders were also assessed with the K-SADS Epidemiologic interview (Orvaschel, Puig-Antich, Chambers, Tabrizi, & Johnson, Reference Orvaschel, Puig-Antich, Chambers, Tabrizi and Johnson1982). At T2, T3, and T4, the Longitudinal Interval Follow-Up Evaluation (LIFE; Keller et al., Reference Keller, Lavori, Friedman, Nielsen, Endicott and McDonald-Scott1987) was jointly administered with the K-SADS or SCID-NP to produce detailed information related to the presence and course of disorders since participation in the previous interview.
Although diagnostic categories were evaluated in accordance with DSM-III-R criteria at T1 and T2 and DSM-IV criteria at T3 and T4, sufficient additional symptom information was collected during the first two assessment waves to permit DSM-IV-based evaluations of individual SUD categories (see Rohde et al., Reference Rohde, Lewinsohn, Seeley, Klein, Andrews and Small2007). Consequently, all individual SUD diagnostic categories, including AUD, were based on DSM-IV criteria. Mean interrater reliability was satisfactory across the four assessment waves for AUD [range of kappa (κ): 0.69 to 0.89; median = 0.79] and individual DRUG categories (range of κ: 0.57 to 0.93).
Family members
Psychiatric histories of first-degree family members were ascertained from direct interviews with the SCID-NP and Personality Disorder Examination (Loranger, Reference Loranger1988). When direct interviews were not possible (38% of all first-degree relatives), we attempted to interview at least two first-degree relatives about the psychiatric history of the target relative based on the family history assessment method (Mannuzza & Fyer, Reference Mannuzza and Fyer1990). Final diagnostic decisions were based on the best-estimate method (Leckman, Sholomskas, Thompson, Balanger, & Weissman, Reference Leckman, Sholomskas, Thompson, Balanger and Weissman1982). Inter-diagnostician agreement, calculated from independently derived best-estimate diagnoses before resolution of any discrepancies, was satisfactory [all κs > 0.87 for individual SUD disorder categories; κ = 0.80 for antisocial personality disorder (ASPD)].
Risk factors assessment
Seven putative family, psychiatric, and psychosocial risk factors were evaluated as predictors of group membership.
Family densities of AUD, DRUG, and ASPD
Within the family risk-transmission literature, evidence exists for substance-specific influences (e.g. parental AUD to offspring AUD) as well as non-specific or generalized influences (e.g. parental cannabis use disorder to offspring AUD) in relation to offspring AUD risk (Bierut et al., Reference Bierut, Dinwiddie, Begleiter, Crowe, Hesselbrock, Nurnberger and Reich1998; Merikangas et al., Reference Merikangas, Stolar, Stevens, Goulet, Preisig, Fenton and Rounsaville1998; Needle et al., Reference Needle, McCubbin, Wilson, Reineck, Lazar and Mederer1986; Tsuang et al., Reference Tsuang, Lyons, Meyer, Doyle, Eisen, Goldberg and Eaves1998). We therefore investigated family densities of AUD and DRUG as putative risk factors. Family density of ASPD was also evaluated as a risk factor. Prior findings demonstrated that parental ASPD predicted offspring SUD, even after variance associated with parental SUD was controlled (Chassin, Pillow, Curran, Molina, & Barrera, Reference Chassin, Pillow, Curran, Molina and Barrera1993; Johnson, Cohen, Chen, Kasen, & Brook, Reference Johnson, Cohen, Chen, Kasen and Brook2006). Individuals with more severe forms of AUD are also more likely to have parental histories of ASPD (Chassin, Pitts, & Prost, Reference Chassin, Pitts and Prost2002), which is pertinent to this research given that the odds of AUD/DRUG comorbidity have been shown to increase as AUD severity becomes greater (Grant et al., Reference Grant, Goldstein, Saha, Chou, Jung, Zhang and Hasin2015; Saha et al., Reference Saha, Grant, Chou, Kerridge, Pickering and Ruan2018).
Family density scores were computed as the sum of all first-degree relatives who were positive for a lifetime diagnosis divided by the number of relatives within a family minus the proband (Milne et al., Reference Milne, Caspi, Harrington, Poulton, Rutter and Moffitt2009). The resulting values denote the proportion of first-degree relatives within families with positive disorder histories. For this study, when data were missing for a given family member, that member was excluded from density score computations. Because family density score distributions were discontinuous and multimodal, four ordered categories were derived for each disorder category and characterized as absent to mild (0% to 24.9%), mild to moderate (25% to 49.9%), moderate to severe (50% to 74.9%), or severe (75% to 100%).
Childhood adversities
The experience of severe adversities during childhood has been found to increase risk for substance use, AUD, and DRUG throughout the life course (Carliner et al., Reference Carliner, Keyes, McLaughlin, Meyers, Dunn and Martins2016; Fergusson, Boden, & Horwood, Reference Fergusson, Boden and Horwood2008; Green et al., Reference Green, McLaughlin, Berglund, Gruber, Sampson, Zaslavsky and Kessler2010; Kendler et al., Reference Kendler, Neale, Prescott, Kessler, Heath, Corey and Eaves1996; Sartor et al., Reference Sartor, Waldron, Duncan, Grant, McCutcheon, Nelson and Heath2013; Schlossarek, Kempkensteffen, Reimer, & Verthein, Reference Schlossarek, Kempkensteffen, Reimer and Verthein2016; Thompson, Lizardi, Keyes, & Hasin, Reference Thompson, Lizardi, Keyes and Hasin2008), including polysubstance misuse (Armour, Shorter, Elhai, Elklit, & Christoffersen, Reference Armour, Shorter, Elhai, Elklit and Christoffersen2014; Martinotti et al., Reference Martinotti, Carli, Tedeschi, Di Giannantonio, Roy, Janiri and Sarchiapone2009). Childhood adversities were retrospectively assessed at T3 with two self-report measures: the Sexual Abuse subscale (α = 0.96) of the Childhood Trauma Questionnaire (Bernstein & Fink, Reference Bernstein and Fink1998) and the Physical Punishment/Abuse scale (α = 0.71) from the Assessing Environments III questionnaire (Berger, Knutson, Mehm, & Perkins, Reference Berger, Knutson, Mehm and Perkins1988).
Childhood internalizing and externalizing disorders before age 15
Epidemiological research suggests that developmental pathways that culminate in AUD are established well before problematic alcohol use begins, and likely causally related to processes that increase vulnerabilities to externalizing and internalizing psychiatric disorders (Clark, Reference Clark2004; Hussong, Jones, Stein, Baucom, & Boeding, Reference Hussong, Jones, Stein, Baucom and Boeding2011; Vanyukov et al., Reference Vanyukov, Tarter, Kirillova, Kirisci, Reynolds, Kreek and Neale2012). Consistent with this view, prospective studies have documented that externalizing tendencies or disorders robustly predict the future onset of alcohol use problems or AUD when undifferentiated with respect to SUD comorbidity (Chassin, Flora, & King, Reference Chassin, Flora and King2004; Elkins, King, McGue, & Iacono, Reference Elkins, King, McGue and Iacono2006; Englund, Egeland, Oliva, & Collins, Reference Englund, Egeland, Oliva and Collins2008; Feingold, Capaldi, & Owen, Reference Feingold, Capaldi and Owen2015; Fergusson, Horwood, & Ridder, Reference Fergusson, Horwood and Ridder2007; Grekin, Sher, & Wood, Reference Grekin, Sher and Wood2006) as well as drug co-use or the joint occurrence of AUD and DRUG (Disney, Elkins, McGue, & Iacono, Reference Disney, Elkins, McGue and Iacono1999; Sibley et al., Reference Sibley, Pelham, Molina, Coxe, Kipp, Gnagy and Lahey2014; Young et al., Reference Young, Mikulich, Goodwin, Hardy, Martin, Zoccolillo and Crowley1995).
The role of internalizing pathways in the development of AUD has been less researched (Hussong et al., Reference Hussong, Jones, Stein, Baucom and Boeding2011). Although internalizing symptoms or disorders are often concomitants with alcohol use problems or AUD in cross-sectional studies (Burns & Teesson, Reference Burns and Teesson2002; Hasin, Stinson, Ogburn, & Grant, Reference Hasin, Stinson, Ogburn and Grant2007; Kessler et al., Reference Kessler, Berglund, Demler, Jin, Merikangas and Walters2005), prospective studies on the risk posed by internalizing tendencies or disorders on future alcohol use problems or AUD have produced mixed findings (Boschloo et al., Reference Boschloo, Vogelzangs, van den Brink, Smit, Beekman and Penninx2012; Buckner & Turner, Reference Buckner and Turner2009; Crum & Pratt, Reference Crum and Pratt2001; Elkins et al., Reference Elkins, King, McGue and Iacono2006; Gilman & Abraham, Reference Gilman and Abraham2001; Grekin et al., Reference Grekin, Sher and Wood2006; Kushner, Sher, & Erickson, Reference Kushner, Sher and Erickson1999; Trautmann et al., Reference Trautmann, Schönfeld, Behrendt, Heinrich, Höfler, Siegel and Wittchen2015; Zimmermann et al., Reference Zimmermann, Wittchen, Höfler, Pfister, Kessler and Lieb2003).
In this research, externalizing disorders included conduct, oppositional defiant, and attention-deficit/hyperactivity. Internalizing disorders consisted of major depression, dysthymia, separation anxiety, overanxious/generalized anxiety, panic, obsessive-compulsive, post-traumatic stress, simple/specific phobia, and social phobia. Each variable was coded 0 (for absence) or 1 (for presence) of any domain-related disorder prior to age 15.
AUD and SUD course features and overall SUD severity
AUD and DRUG episode milestones
Symptom thresholds and timeline considerations specified in DSM-IV for AUD and individual DRUG categories were used to determine disorder onset and offset. Definitions of remission and recovery were informed by DSM-IV guidelines, LIFE interview naming conventions (Keller et al., Reference Keller, Lavori, Friedman, Nielsen, Endicott and McDonald-Scott1987), and earlier descriptions of these concepts (Chung & Maisto, Reference Chung and Maisto2006). With AUD as the example, remission referred to the offset of an AUD episode lasting at least 1 full month but less than 12 months during which the individual no longer met diagnostic criteria for the episode but may have continued alcohol use at subthreshold levels. The re-emergence of this episode during the remission period was regarded as a continuation of the initial episode (i.e. a relapse). The resolution of the episode, defined as a period of uninterrupted AUD remission lasting at least 12 months, was considered a recovery from that episode. Recovery, analogous to the DSM-IV concept of ‘sustained full remission’, was only achieved after a 12-month period following the offset of an AUD episode during which there was complete absence of AUD-related symptoms for a continuous 12-month period. A recurrence was defined as a new AUD episode after a period of recovery. These course milestone definitions were also applied to the determination of disorder offset and recovery for individual DRUG categories.
AUD and overall SUD severity
In the SUD literature, early-onset age, a dependence (v. abuse) diagnosis, episode persistence, disorder recurrence, and treatment seeking have been associated with a SUD course characterized by greater harms or impairments (Clark, Reference Clark2004; Delucchi & Kaskutas, Reference Delucchi and Kaskutas2010; Hanson, Medina, Padula, Tapert, & Brown, Reference Hanson, Medina, Padula, Tapert and Brown2011; Hicks, Iacono, & McGue, Reference Hicks, Iacono and McGue2010). In this research, course features that indexed AUD severity included: (a) onset age of first AUD episode (reverse coded, with early onset associated with greater severity); (b) an AUD dependence diagnosis (v. abuse); (c) AUD episode recurrence following first episode recovery; (d) treatment seeking during any AUD episode; and (e) the cumulative duration of all AUD episodes through age 30 (in months). For persons who remained within an AUD episode at age 30, cumulative duration was based on all AUD episodes until this age. For group contrasts, continuous AUD course variables (i.e. onset age, cumulative duration) were dichotomized at the sample median for probands with a lifetime AUD.
A similar set of variables was created that referenced SUD course features inclusive of AUD. Cumulative SUD duration computation was based on all non-overlapping AUD and DRUG episodes until age 30. Because all dichotomous SUD course indicators demonstrated moderate and positive intercorrelations (tetrachoric intercorrelation range: 0.35 to 0.69; median = 0.48), values for the five indicators were summed to produce an overall lifetime SUD severity score (range: 0 to 5). Earlier research has shown that unit counts of AUD severity indices performed as well as weighted models derived from statistical procedures (Dawson et al., Reference Dawson, Saha and Grant2010).
Age 30 outcome measures
Twelve indicators of proband psychosocial functioning were assessed at T4 that collectively evaluated three domains of functioning: socioeconomic, interpersonal, and psychosocial.
Socioeconomic functioning
Three variables indexed socioeconomic functioning: (a) years of education, (b) weeks of unemployment during the prior 12 months, and (c) past year annual household income.
Interpersonal functioning
Six variables indexed various aspects of interpersonal functioning: (a) marriage history; (b) divorce or separation history; (c) biological parentage of a child; (d) relationship quality with family, assessed with the Perceived Social Support from Family scale (α = 0.90; Procidano and Heller, Reference Procidano and Heller1983); (e) poor social adjustment, assessed by the Social Adjustment Scale (α = 0.70; Weissman and Bothwell, Reference Weissman and Bothwell1976); and (f) engagement in high-risk sexual behavior, represented by the whole sample-derived mean of Z-score transformations of two variables, the number of past year concurrent sexual partners and total number of partners.
Psychosocial functioning
Three variables indexed various aspects of psychosocial functioning: (a) life dissatisfaction, assessed by 15 items (α = 0.77) adapted from earlier questionnaires (Andrews & Withey, Reference Andrews and Withey1976; Campbell, Converse, & Rodgers, Reference Campbell, Converse and Rodgers1976); (b) stressful major life events during the past year, based on endorsements of 33 events identified in earlier measures (Dohrenwend, Levav, & Shrout, Reference Dohrenwend, Levav, Shrout, Weissman, Myers and Ross1986; Hammen et al., Reference Hammen, Gordon, Burge, Adrian, Jaenicke and Hiroto1987; Holmes & Rahe, Reference Holmes and Rahe1967); and (c) number of lifetime psychiatric disorders, computed as the sum of distinct diagnostic categories for which probands met DSM-defined criteria by age 30. For this latter variable, individual Axis I psychiatric disorders excluding SUDs were considered regardless of the presence of other lifetime disorders from the same broad diagnostic domain (e.g. mood disorders; anxiety disorders). Recurring episodes of any disorder were treated as a single occurrence given our focus on disorders rather than episodes.
Data analysis
Missing data
We used multiple imputation methods to impute missing values on risk and outcome variables (5% to 11% missing) to the proband sample with family data (n = 732). This process involved the generation of 20 complete data sets following best-practice recommendations (Abayomi, Gelman, & Levy, Reference Abayomi, Gelman and Levy2008).
Statistical modeling procedures
Multinomial logistic regression methods were used to evaluate whether risk factors predicted group membership. Linear and binary logistic regression models were used to evaluate whether group membership predicted AUD and SUD course features and age 30 outcomes.
Unadjusted and adjusted analyses
Unadjusted results for risk factors, AUD and SUD course, and age 30 outcomes were based on separate models for each predictor. In analyses of risk factors, adjusted results controlled for all other risk factors. In analyses of group differences in AUD and SUD course features and overall SUD severity scores, adjusted analyses controlled for risk factors (Adjusted 1) and risk factors and demographics (Adjusted 2). In analyses of age 30 outcomes, adjusted analyses controlled for overall SUD severity (Adjusted 1) and overall SUD severity, risk factors, and demographics (Adjusted 2).
Results
Sample characteristics
Table 1 summarizes proband characteristics by group membership. Notably, men were disproportionately underrepresented within the NEITHER and DRUG ONLY groups and overrepresented in the AUD ONLY and BOTH groups. When referenced to NEITHER, DRUG ONLY and BOTH were more likely to be raised in a single-parent household during adolescence, with DRUG ONLY also more likely to be raised by a single parent compared to AUD ONLY.
Table 1. Proband characteristics by group membership
Note: M, mean; s.d., standard deviation; AUD, alcohol use disorder; DRUG, non-alcohol substance use disorder; NA, not applicable. Values that share subscripts within rows are not significantly different (α = 0.05).
Risk factors
Six sets of pairwise contrasts were performed based on the four-group configuration. Analyses presented in the first three columns in Table 2 compared AUD ONLY, DRUG ONLY, and BOTH, respectively, with NEITHER. When results were evaluated with Benjamini-Hochberg adjusted p values (Benjamini & Hochberg, Reference Benjamini and Hochberg1995), no significant differences emerged between AUD ONLY and NEITHER in unadjusted or adjusted analyses (column 1).
Table 2. Results of regressing group membership on risk factors
Note: *Benjamini-Hochberg adjusted p < 0.05. Confidence intervals do not account for Benjamini-Hochberg adjustment. M, mean; s.d., standard deviation; AUD, alcohol use disorder; DRUG, non-alcohol substance use disorder; ASPD, antisocial personality disorder; CI95, 95% confidence interval. Unadjusted results are based on separate models for each predictor. Adjusted results are based on a single model with all predictors entered simultaneously. The second group of each contrast is the reference category.
Significant differences emerged in DRUG ONLY v. NEITHER unadjusted contrasts (column 2). Findings indicated that DRUG ONLY, when contrasted with NEITHER, came from comparatively high DRUG-density families and experienced disproportionately higher levels of childhood sexual and physical abuse. These unadjusted effects, however, were not maintained in adjusted analyses that controlled for all other risk factors.
Several significant differences were observed between BOTH and NEITHER (column 3). In unadjusted analyses, significantly greater odds of BOTH group membership were observed for increasing family densities of AUD, DRUG, and ASPD; histories of childhood sexual abuse and physical abuse; and histories of an externalizing disorder prior to age 15. Significant unadjusted effects were maintained for all risk factors in adjusted analyses except for family AUD and ASPD densities.
Column 4 evaluated differences in risk factors between DRUG ONLY and AUD ONLY groups. Only one significant unadjusted contrast emerged, for childhood sexual abuse, which was not maintained in adjusted analyses. Overall, these groups appeared more similar than different with respect to risk factors, with the pattern of findings similar to, albeit smaller in magnitude than, those observed in the second contrast column (DRUG ONLY v. NEITHER).
Analyses presented in columns 5 and 6 compared AUD ONLY and DRUG ONLY, respectively, with BOTH. No statistically significant unadjusted or adjusted contrasts emerged between DRUG ONLY and BOTH (column 6). Several significant unadjusted contrasts were evident between the AUD ONLY and BOTH (column 5). In unadjusted analyses, the odds of BOTH group membership were comparatively greater given higher family densities of AUD and DRUG, childhood sexual and physical abuse, and an externalizing disorder prior to age 15. Except for AUD family densities, the significance of these effects was maintained in adjusted analyses.
AUD and SUD course features
Table 3 presents group contrasts for AUD-specific course-related features for the two groups with AUD histories. All five AUD course features differed between BOTH and AUD ONLY in separate regression analyses, with BOTH demonstrating greater severity on each indicator. These effects were maintained in adjusted analyses that controlled for risk factors and demographic variables.
Table 3. Descriptive statistics and results of regressing AUD-specific course features on group membership
Note: *Benjamini-Hochberg adjusted p < 0.05. Confidence intervals do not account for Benjamini-Hochberg adjustment. M, mean; s.d., standard deviation; CI95, 95% confidence interval; AUD, alcohol use disorder. Adjusted analysis 1 included the risk factors in Table 2 as covariates. Adjusted analysis 2 included risk factors in Table 2 and proband demographic characteristics in Table 1 as covariates. Early-onset and cumulative duration indicators were calculated using median age of first AUD onset (19.7 years) and median cumulative AUD duration across episodes (27.4 months), respectively, as cutoffs. The age of onset binary variable was recoded to indicate greater severity associated with a younger age of onset. The AUD ONLY group is the reference category.
Table 4 presents analyses for the same course indicators but broadened to include each SUD-defined group and combined AUD and DRUG course features. Unadjusted group contrasts revealed several significant differences. The most robust differences were observed between BOTH with AUD ONLY and DRUG ONLY (columns 2 and 3, respectively), with BOTH displaying greater severity on each indicator. Unadjusted contrasts also demonstrated two significant contrasts between the DRUG ONLY and AUD ONLY groups (early onset and treatment seeking; column 1), with DRUG ONLY demonstrating a more severe course on these indicators. The significance of contrasts involving individual course indicators was mostly maintained after adjustment for risk factors (Adjusted 1 in table) and risk factors and demographic variables (Adjusted 2).
Table 4. Descriptive statistics and results of regressing combined AUD and DRUG course features on group membership
Note: *Benjamini-Hochberg adjusted p < 0.05. Confidence intervals do not account for Benjamini-Hochberg adjustment. M, mean; s.d., standard deviation; CI95, 95% confidence interval; AUD, alcohol use disorder. DRUG, non-alcohol substance use disorder. Table entries for the group contrasts show Cohen's d effect sizes for the overall severity score and odds ratios for dichotomous indicators. Adjusted analysis 1 included the risk factors in Table 2 as covariates. Adjusted analysis 2 included risk factors and proband demographic features in Table 1. The overall severity score represents the number of severity indicators present across AUD and DRUG episodes (range = 0 to 5). Early-onset and cumulative duration indicators were calculated using median age of first onset (18.6 years) and median cumulative duration (37.0 months), respectively, as cut-offs for all non-overlapping substance use disorder episodes. The age of onset binary variable was recoded to indicate greater severity associated with a younger age of onset. The second group of each contrast is the reference category.
As with the individual indicators, BOTH had a higher overall SUD severity score than AUD ONLY and DRUG ONLY, and these differences were maintained after adjustments for risk factors and demographics. DRUG ONLY also demonstrated a greater overall SUD severity score compared to AUD ONLY in unadjusted and adjusted analyses.
Age 30 outcomes
Analyses presented in the first three columns of Table 5 contrasted AUD ONLY, DRUG ONLY, and BOTH, respectively, with NEITHER. Among the AUD ONLY contrasts with NEITHER (column 1), two significant unadjusted effects emerged: for life dissatisfaction and number of lifetime psychiatric disorders. When DRUG ONLY was contrasted with NEITHER (column 2), several differences were apparent in unadjusted analyses. These were observed for years of education, weeks of unemployment, annual household income, stressful life events, and number of lifetime psychiatric disorders. The third column contrasted outcomes between BOTH with NEITHER. Significant unadjusted effects were noted for marital history, years of education, weeks unemployed during the past year, annual household income, relationship quality, poor social adjustment, high-risk sexual behavior, life dissatisfaction, stressful life events, and number of lifetime psychiatric disorders. Because covariance analyses involved control of overall SUD course severity, adjusted analyses were not performed for contrasts involving NEITHER.
Table 5. Results of regressing age 30 outcomes on group membership
Note: *Benjamini-Hochberg adjusted p < 0.05. Confidence intervals do not account for Benjamini-Hochberg adjustment. M, mean; s.d., standard deviation; CI95, 95% confidence interval; AUD, alcohol use disorder; DRUG, non-alcohol substance use disorder. Table entries for group contrasts show Cohen's d effect sizes for scales and odds ratios for dichotomous outcomes. Adjusted analysis 1 included the overall severity score as a covariate. Adjusted analysis 2 included the overall severity score, risk factors, and proband demographic characteristics as covariates. NA = the overall severity score does not apply to the NEITHER group; consequently, adjusted contrasts exclude that group. The second group of each contrast is the reference category.
Column 4 evaluated differences in outcomes between DRUG ONLY and AUD ONLY. Three significant unadjusted effects were noted: years of education, weeks unemployed, and household income. The effects for years of education and household income were maintained following the control of overall SUD severity (Adjusted 1) but were not maintained when confounder control included SUD severity, demographic variables, and risk factors (Adjusted 2).
The fifth and sixth columns contrasted the AUD ONLY and DRUG ONLY groups, respectively, with BOTH. Similar to the pattern of findings for risk factors, no statistically significant unadjusted or adjusted contrasts emerged in outcomes between DRUG ONLY and BOTH (column 6). Several significant unadjusted contrasts were observed between AUD ONLY and BOTH (column 5). These were evident for years of education, weeks unemployed, household income, high-risk sexual behavior, stressful life events, and number of lifetime psychiatric disorders. None of these significant contrasts were maintained following control of overall SUD severity (Adjusted 1), thus implying that unadjusted effects were influenced similarly by overall SUD severity and group membership.
Discussion
Study findings suggest that important manifestations of heterogeneity among persons with AUD are evident between individuals whose histories of substance use problems include alcohol along with other drugs (BOTH) and those with histories where alcohol is the only misused substance (AUD ONLY). Evidence in support of this distinction was broadly established with divergent findings related to risk factors, AUD course features, and in unadjusted analyses of psychosocial outcomes assessed at age 30. Contrary to predictions, course features associated with AUD episodes were not accounted for by risk factors or demographic variables and instead likely resulted from emergent processes within SUD episodes. Differences in age 30 outcomes between BOTH and AUD ONLY were explained by overlapping characteristics associated with group membership and overall SUD severity inclusive of AUD severity. Together, these findings suggest that divergent developmental trajectories culminating in AUD v. BOTH can be predicted by risk factors and demographic variables. Following an AUD diagnosis, however, the severity of SUD episodes inclusive of AUD also has a significant role in charting the course for subsequent socioeconomic, interpersonal, and psychosocial functioning.
Specific study findings on the risk, course, and outcomes of AUD as a function of group membership included the following. In unadjusted and adjusted analyses of risk factors (Table 2), AUD ONLY was highly similar in putative risk factor endorsements to NEITHER, whereas BOTH displayed significantly greater levels of most risk factors compared to NEITHER or AUD ONLY. In unadjusted and adjusted analyses of AUD course (Table 3), BOTH compared to AUD ONLY experienced earlier, longer, recurring, and more severe AUD episodes and a greater likelihood of treatment seeking, findings that are consistent with those reported in earlier cross-sectional studies (Grant et al., Reference Grant, Goldstein, Saha, Chou, Jung, Zhang and Hasin2015; Saha et al., Reference Saha, Grant, Chou, Kerridge, Pickering and Ruan2018). When analyses were broadened to include the DRUG ONLY group and course features for all SUD episodes combined, unadjusted and adjusted analyses revealed that BOTH demonstrated greater severity across most indicators when separately contrasted with AUD ONLY and DRUG ONLY (Table 4). Adjusted analyses further indicated that between group differences in AUD (Table 3) and SUD course severity (Table 4) were mostly unexplained by demographic features or risk factors despite the predictive relevance of these variables for group membership (Tables 1 and 2, respectively). In analyses of age 30 outcomes, the overall pattern of findings (Table 5) resembled those reported for risk factors. In unadjusted analyses, AUD ONLY was mostly similar to NEITHER, whereas BOTH displayed significantly poorer outcomes in multiple domains compared to NEITHER or AUD ONLY. Whereas unadjusted analyses implied that persons with BOTH are distinct in age 30 outcomes from those with AUD ONLY histories, adjusted analyses indicated that significant differences observed between these groups were accounted for by the severity of lifetime SUD episodes inclusive of AUD.
This overall pattern of study findings suggests that important sources of heterogeneity of AUD shift between childhood and early adulthood. Whereas several risk factors significantly predicted membership in either AUD ONLY or BOTH, group membership and the severity of lifetime SUD episodes each accounted for observed differences in age 30 outcomes. Together, these findings imply that addiction-specific processes alter life trajectories among those with AUD ONLY or BOTH, and that the combination of AUD and DRUG produces an especially malignant synergy that culminates in a more severe course and greater psychosocial impairments.
Implications
There are four primary implications associated with this research. First, there appear to be at least two pathways that culminate in AUD, an example of the equifinality concept. Risk factor findings imply that AUD ONLY and BOTH groups are likely drawn from dissimilar populations and may represent distinct and important behavioral phenotypes that account for a significant portion of the heterogeneity observed within the AUD population.
Second, the severity of SUD episodes inclusive of AUD accounted for all significant unadjusted group contrasts on age 30 outcomes between AUD ONLY and BOTH. Although differences in risk factor exposure may have influenced the life course that culminated in membership in either of these groups, AUD course features and overall SUD severity were largely uninfluenced by risk factors and were instead likely the product of emergent processes within SUD episodes. Additionally, even though AUD ONLY and BOTH groups had AUD histories in common, differences in age 30 outcomes were observed between these groups. Differences between AUD ONLY and BOTH in AUD course indicators, overall SUD severity, and age 30 outcomes exemplify the multifinality of AUD.
Third, AUD ONLY was similar to and often indistinguishable from NEITHER in risk factors and age 30 outcomes. These findings raise questions about meaningful etiologic distinctions between these groups in community samples. The absence of clear group distinctions might be related to several factors, including the commonality of AUD, symptom thresholds required for an AUD diagnosis in DSM-IV, cultural factors, and distinctions between excessive alcohol consumption and alcohol-related problems. The commonality of lifetime AUD in the U.S. general population is high, estimated to be 29% (Grant et al., Reference Grant, Goldstein, Saha, Chou, Jung, Zhang and Hasin2015). This high rate is a possible artifact of relatively low symptom thresholds required for an AUD diagnosis. Over half of all past-year AUD positive cases in the U.S. are categorized as mild (Grant et al., Reference Grant, Goldstein, Saha, Chou, Jung, Zhang and Hasin2015). The commonality of mild AUD cases in the general population may, in part, reflect the impact of cultural factors. Drinking patterns and per capita alcohol consumption within the U.S. have varied substantially over successive generations and within subcultures during fixed time periods (Castro, Barrera, Mena, & Aguirre, Reference Castro, Barrera, Mena and Aguirre2014; Martin, Chung, & Langenbucher, Reference Martin, Chung, Langenbucher and Sher2016). AUD among some community members may reflect the influence of cultural socialization processes with respect to alcohol, which include conceptualizations of normative, excessive, or harmful use.
Emerging research further suggests etiological distinctiveness between excessive alcohol consumption and moderate-to-severe alcohol problems in the general population. Recent findings from genome-wide association studies, for example, indicate that the genetic risk for alcohol consumption only partially overlaps with alcohol-related problems or AUD (Kranzler et al., Reference Kranzler, Zhou, Kember, Smith, Justice, Damrauer and Gelernter2019; Sanchez-Roige et al., Reference Sanchez-Roige, Palmer, Fontanillas, Elson and Clarke2019; Walters et al., Reference Walters, Polimanti, Johnson, McClintick, Adams, Adkins and Agrawal2018). Additionally, when compared with the genetic risk for alcohol consumption, alcohol-related problems and AUD demonstrate larger genetic correlations with other substance use phenotypes such as tobacco and cannabis (Kranzler et al., Reference Kranzler, Zhou, Kember, Smith, Justice, Damrauer and Gelernter2019; Sanchez-Roige et al., Reference Sanchez-Roige, Palmer, Fontanillas, Elson and Clarke2019; Walters et al., Reference Walters, Polimanti, Johnson, McClintick, Adams, Adkins and Agrawal2018). These findings collectively imply that the genetic risks for alcohol consumption and alcohol-related problems are associated but distinct phenotypes, and that the genetic risk associated with alcohol-related problems extends to the misuse of a broader range of substances. The AUD ONLY group in this research may not have been well-distinguished from those with excessive alcohol consumption given the threshold for alcohol-related problems required for an AUD diagnosis, particularly alcohol abuse, in DSM-IV.
Fourth, findings from this research suggest that prevention and intervention programs should emphasize different processes depending on one's location along the developmental continuum. Prior to the onset of AUD, prevention efforts might emphasize youth whose life experiences place them at high risk. For persons who subsequently develop an AUD, interventions might emphasize substance misuse itself, including the contexts that support such behavior.
Limitations
This study has noteworthy limitations. First, this research only extended to proband age 30. Differences have been reported in genetic risk factors associated with AUD first emergence during adolescence through the early 20s when compared to mid-20s and beyond, whereby the former are most strongly linked to non-specific externalizing genetic risk factors and the latter to specific alcohol-related genetic risk factors (Kendler, Gardner, & Dick, Reference Kendler, Gardner and Dick2011). Consequently, first-episode AUD onsets after age 30 may be etiologically distinct from those that emerge during adolescence or the early 20s. The generalization of the present findings to cases with AUD onsets after age 30 is therefore uncertain. Second, at some assessment waves, attrition since the previous wave was related to histories of AUD, SUD, externalizing behavior disorders, or sex. The effects of participant attrition over the 15-year study period on observed findings is unknown. Third, the diversity of the OADP sample, although representative of the racial and ethnic distribution of western Oregon, is limited.
Conclusions
Study findings demonstrate the equifinality and multifinality of AUD and the importance of distinguishing persons with AUD ONLY from BOTH, whether this be done through subgroup analyses or statistical control of DRUG histories. This research also highlights the importance of overall SUD severity inclusive of AUD severity when accounting for the course and outcomes of AUD. Factors that account for the high degree of similarity between AUD ONLY and NEITHER among community members, such as cultural influences and AUD diagnosis symptom thresholds, warrant additional research attention.
Acknowledgements
The authors wish to thank Peter Lewinsohn for making OADP data available for this investigation.
Financial support
National Institutes of Health grants R01AA027543 to Richard Farmer and John Seeley and MH40501, MH50522, and DA12951 to Peter Lewinsohn supported this research. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflict of interest
None.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
