Predictors of symptom course in alcohol use disorder

Abstract

Background:

Symptoms often play an important role in the scientific inquiry of psychological disorders and have been theorized to play a functional role in the disorders themselves. However, little is known about the course of specific symptoms and individual differences in course. Understanding the course of specific symptoms and factors influencing symptom course can inform psychological theory and future research on course and treatment.

Methods:

The current study examined alcohol use disorder (AUD) criteria to explore how etiologically relevant covariates differentially affected the course of individual criteria. The study examined 34,653 participants from Wave 1 (2001–2002) and Wave 2 (2003–2004) of the National Epidemiological Survey on Alcohol and Related Conditions (NESARC), to analyze the extent to which AUD symptom course is predicted by alcohol consumption patterns, family history of alcoholism, the presence of internalizing and externalizing disorders, and race.

Results:

The course of all AUD criteria was significantly influenced by these predictors, with the magnitude of the influence varying across different criteria and different aspects of the course (i.e., onset, persistence, recurrence). The strength of the relationship is partially related to the theoretical proximity of a given covariate to AUD symptomatology, with heavy drinking being the strongest and family history of AUD being the weakest. The course of all criteria was strongly associated with the prevalence of the criterion in the overall sample.

Conclusions:

The course of AUD criteria is heterogeneous, appearing to be influenced by conceptually proximal predictors, the prevalence of the criterion, and perhaps an underlying common factor. Diagnostic accuracy may be improved by including a criterion related to alcohol consumption. Future work should include exploring the interchangeability of criteria and alternative operationalization of them.

INTRODUCTION

Current conceptualizations of alcohol use disorder (AUD) have been strongly influenced by the criteria used in diagnostic manuals such as the Diagnostic and Statistical Manual of Mental Disorders (DSM) and the International Classification of Diseases (ICD). These diagnostic categories make several assumptions about the underlying structure of the disorder. One significant assumption is that symptoms are observable phenomena indicative of a latent disorder. In a simple symptom-count model of diagnosis, symptoms are assumed to be polythetic (i.e., neither necessary nor sufficient for diagnosis) and interchangeable. These features have been suggested to create conceptually messy and heterogenous disorders (e.g., Lane & Sher, 2015), concerns about diagnostic validity (e.g., Martin et al., 2011), and a lack of consistent targets for treatment (Fleury et al., 2016).

In response to the concerns about how accurately diagnostic categories “carve nature at the joints,” researchers have considered alternative perspectives. New models have emerged (e.g., symptom network models, the Cambridge model), in which symptoms are treated as meaningful in their own right (Borsboom, 2017; Marková & Berrios, 2009; Wilshire et al., 2021) and not as mere manifestations of an underlying latent entity. These symptom-focused models have rapidly gained relevance in psychiatric nosology including AUD (Conlin et al., 2022).

Among the AUD criteria, there is evidence that some risk factors are more or less associated with different, specific criteria. For example, one study found that externalizing personality was more strongly associated with the criteria of Social/Interpersonal Problems and Role Interference, and least strongly associated with Tolerance (e.g., McDowell et al., 2019). Another recent study found four different patterns of how each AUD criterion was associated with internalizing and externalizing, with Hazardous Use being most associated with externalizing and Withdrawal being most associated with internalizing (Watts et al., 2022). In addition to personality/psychopathology correlates, there appear to be unique associations with consumption-related variables; Morgenstern et al. (2016) found that experiencing Craving and Drinking Larger/Longer than Intended early in one’s alcohol experimentation is predictive of later episodes of binge drinking.

Symptom course

The examination of the syndrome course is a valuable tool for understanding the nature of psychiatric disorders. Dating back to Kraepelin (1896), some psychopathologists have viewed prognosis as a factor to consider when establishing diagnosis. While countless studies have been conducted on the syndromal course of mental disorders, much can still be learned from the relatively unexplored course of symptoms. Studies have been conducted on the symptom course of various forms of psychopathology (e.g., Conradi et al., 2011; Larsson et al., 2004; Nivard et al., 2015), providing evidence that different symptoms have different longitudinal trajectories, with some of the variability in symptom course being accounted for by etiologically relevant covariates.

There are currently few studies on the course of individual AUD criteria. One study by O’Neill and Sher (2000), examined the course of DSM-IV AD criteria and found that Tolerance and Withdrawal had moderate persistence after 1 year and that both had prognostic significance for later AUD. One recent study by Vergés et al. (2021) found that the course of individual alcohol dependence (AD) criteria differed as a function of age. For example, in younger adults, Failed Attempts to Cut Down and Drinking Despite Physical/Psychological Problems were both less persistent and less predictive of AD course, relative to their effects in older adults. A prospective study exploring the onset of individual AUD criteria found that Cut Down and Social Problems had the highest probability of occurring during the onset of drinking (Buu et al., 2012), while another found that the rate progression from first drink to first AD criterion differed across the seven AD criteria (Behrendt et al., 2008). Two studies have also been conducted which have examined changes in the rates of endorsement for specific AUD criteria over time in persistent and recurrent syndromal diagnosis (Schuckit et al., 2023; Schuckit & Smith, 2021). The dearth of empirical research on the course of AUD criteria is surprising, as the heterogeneous nature of AUD criteria suggests that it is unlikely for all criteria to maintain a uniform course. Tolerance and Withdrawal are thought to be the result of neurobiological adaptations and suggest that the organism has undergone a physical change, therefore being relatively stable. Conversely, Social/Interpersonal Problems and Role Interference are inherently context-dependent criteria and thus may be less persistent. Examining the course of symptoms may help to evaluate which symptoms appear to be core, stable features of the syndrome.

Present study

The current study sought to explore the association of etiologically relevant covariates on individual criteria within AUD and to examine whether these covariates more strongly influence the course of criteria with closer conceptual proximity to the given predictor. As the empirical literature on predictors of individual symptom courses is limited, the current study is exploratory. However, we make some general hypotheses regarding the magnitude of effects and some specific hypotheses about predictors differentially affecting specific criteria. For example, alcohol consumption is expected to be more strongly related to Tolerance and Withdrawal due to allostasis. Similarly, externalizing is posited to be more strongly related to Hazardous Use and Social/Interpersonal Problems.

To examine the unique effects of multiple predictors on each criterion, we systematically evaluated each predictor’s effect on the onset (new endorsement in never symptomatic individuals), persistence (continued endorsement in past 12-month symptomatic individuals), and recurrence (endorsement in individuals who were not symptomatic in the past 12 months but reported being previously symptomatic in their lifetime) of each criterion. These operationalizations of onset, persistence, and recurrence, are henceforth referred to as “course.” The predictors selected for the current study were (a) heavy drinking, (b) externalizing (as proxied by Conduct Disorder diagnosis), (c) internalizing distress (as specified by Major Depressive Disorder, Persistent Depressive Disorder, and Generalized Anxiety Disorder; Kotov et al., 2017), (d) internalizing fear (as specified by Panic Disorder, with or without Agoraphobia, and Social Anxiety Disorder; Kotov et al., 2017), (e) having family history of AUD, and (f) race/ethnicity. Previous research has found a complex relationship between internalizing and AUD, including findings that the relationship between alcohol problems and anxiety differs across anxiety disorders (Kushner et al., 1990). Research on internalizing has uncovered multiple well-validated subfactors that have some degree of unique correlates (Krueger, 1999; Watson et al., 2022). Consistent with recent work examining AUD symptoms and internalizing (e.g., Watts et al., 2022), we chose to examine the sub-factors of internalizing distress and internalizing fear separately (henceforth referred to as “Distress” and “Fear”). Excluding Race, these predictors were expected to have robust effects and have extant research indicating differing degrees of relevance to different symptoms of AUD (e.g., McDowell et al., 2019; Morgenstern et al., 2016; Watts et al., 2022). Race and Family History of AUD were expected to have weaker effects than other predictors (although Family History likely overlaps with conduct disorder and heaviness of consumption pattern), with similar effect sizes across all criteria. While we do not have a priori hypotheses for differences in course based on Race, we are mindful that social and cultural factors have a significant effect on different racial groups in the United States, and best practice discourages treating individuals as part of a homogenous population (e.g., Bratter & Eschbach, 2005; Glass et al., 2020). Given the literature suggesting that both heavy and chronic consumption of alcohol provides the greatest risk for AUD (e.g., Hasin et al., 2017) and alcohol-related medical problems (e.g., Rehm et al., 2010), we also examined differences in AUD symptom course based on the chronicity of heavy consumption. Furthermore, treatment utilization was analyzed to determine whether the persistence of criteria was affected by the receipt of treatment for AUD.

We hypothesized that the strength of each predictor’s influence on symptom course would relate to the conceptual proximity to AUD, with more proximal predictors having a stronger influence on symptom course. As such, we expected the strength of the prediction to be ordered: Heavy Drinking, Conduct Disorder, Distress, Fear, Family History, Race. We also expected that the predictors would have differential effects on the criteria based on their nomological relevance to the given criterion. Based on past research (McDowell et al., 2019), we expected the course of Social/Interpersonal, Role Interference, and Hazardous Use to be more strongly related to Externalizing. Distress and Fear were expected to have similar effects, except for Hazardous Use, which was expected to have a significantly weaker association or even negative relationship with Fear given that fears over being stopped and/or arrested for a DUI would be more relevant for those high on general fearfulness. Criteria associated with neuroadaptations due to repeated alcohol exposure (Tolerance, Withdrawal) were expected to be most strongly associated with heavy drinking. Recurrence of criteria was expected to be most strongly associated with heavy consumption, as heavy alcohol use is a well-established predictor of AUD relapse (Tuithof et al., 2014).

METHOD

Sample

Data from the two waves of the National Epidemiological Study on Alcohol and Related Conditions (NESARC) study were utilized. NESARC is a nationally representative sample of 43,093 participants, 18 years and older, conducted by the National Institute on Alcohol Abuse and Alcoholism (NIAAA) over a three-year period, with baseline measurements conducted from 2001 to 2002 (Grant et al., 2003) and follow-up measurements from 2004 to 2005 (Grant et al., 2005). The response rate at Wave 2 was 86.7%, resulting in a final count of 34,653 participants. Participant characteristics and survey methodology for NESARC have been described in detail elsewhere (Grant et al., 2004).

Measures

Diagnosis and criteria were measured by the NIAAA Alcohol Use Disorder and Associated Disabilities Interview Schedule-DSM-IV Version (AUDADIS-IV). The AUDADIS-IV is a structured diagnostic interview developed to measure psychiatric problems in large-scale data, and there is extensive evidence supporting its reliability and validity (Grant et al., 2001, 2003).

Family History was defined as having at least one biological parent with an AUD (alcohol abuse or AD, per the AUDADIS-IV). Conduct Disorder was used as a proxy for externalizing. In addition, Conduct Disorder provides a behavioral operationalization of antisocial traits and captures a fixed period of life that all participants have already passed through, and is likely less contaminated with Substance Use Disorder than Antisocial Personality Disorder. Distress included a lifetime diagnosis of Major Depressive Disorder, Persistent Depressive Disorder, or Generalized Anxiety Disorder. Fear included Panic Disorder (with and without Agoraphobia) or Social Anxiety Disorder. Heavy drinking corresponds to the NIAAA (2019) guidelines defining risky drinking as “4 or more drinks on any day or 8 or more drinks per week for women and 5 or more drinks on any day or 15 or more drinks per week for men”. Individuals who exceeded either criterion were included in the Heavy Drinking category. The overall data was contrast-coded by racial identity, however, to ensure that the subsamples would have sufficient participants to produce reliable estimates, only participants in the White and Black categories were included. In the full sample, there were 18,256 White participants and 5346 Black participants. The race analyses did not include the Native (n = 502), Asian (n = 690), and Hispanic (n = 5199) categories. While the Hispanic sample may have been large enough to generate estimates, concerns have been raised about treating “Hispanic” as a single homogenous group, with a majority preferring to be categorized by their country of origin (e.g., Taylor et al., 2012; Umaña-Taylor & Fine, 2001). When determining the effect of treatment on symptom persistence, participants who received any form of treatment for AUD between Wave 1 and Wave 2 were compared with those who did not receive treatment. To examine chronic heavy drinking, we examined symptom course in individuals who exceeded NIAAA drinking guidelines at Wave 1, Wave 2, both waves, or neither wave.

Symptom course

We only used participants who provided data at both waves of NESARC (N = 34,653). Table 1 shows the method by which we defined persistence, onset, and recurrence. For persistence analysis, we conditioned the sample on participants who endorsed a given past-12-month criterion at Wave 1. For onset analyses (see Appendix 1), we conditioned the sample on participants who did not have a lifetime endorsement of the criterion. For symptom recurrence, we conditioned the sample on participants who had the criterion prior to the past year at Wave 1 but did not report the criterion during the past 12 months at Wave 1. For all the three analyses, the participants were only included if they met the stated condition and reported drinking at least 1 drink in their life prior to Wave 1 or at least 1 drink between Wave 1 and Wave 2. Endorsing a criterion at a given wave is defined as experiencing the criterion within the past 12 months. The criteria included in our analyses are 10 of the 11 DSM-IV AUD criteria (see Appendix 2); the criterion “legal consequences due to drinking” was dropped from the analyses, for reasons described by Hasin and colleagues (2013). The DSM-5 (American Psychiatric Association, 2013) criterion of Craving was not assessed at Wave 1 of NESARC and, therefore, could not be analyzed for the course.

Table 1.

Symptom Course as Defined by Endorsement Patterns

	Prior to Wave 1	Wave 1	Wave 2
Persistence	LTa	PY	PY
Onset	-	-	PY
Recurrence	LT	-	PY

Note. “LT” Represents the lifetime presence of the symptom, “PY” represents the presence of a symptom at the past year prior to the given timepoint. “-” represents the absence of a symptom at a given time.

^aFor persistence, lifetime endorsement prior to past year was allowed but not sufficient while PY at Wave 1 was necessary and sufficient.

Analytic approach

To account for the complex sampling design used in NESARC, the SAS procedure SURVEYLOGISTIC was used to determine odds ratios for symptom persistence, onset, and recurrence. This procedure accounted for the primary sampling unit, stratum, and sampling weights of the NESARC data using Wave 2 sampling weights. To assess the predictors of course, data were analyzed separately for each predictor, except for age and sex, which were included in all analyses. Separate sets of analyses were conducted for each “stage” of the course (i.e., persistence, onset, recurrence), along with each criterion and predictor, allowing us to examine the effect of each predictor on the course of each criterion. False discovery rate (FDR) was used to control for familywise error. After completing these analyses on the symptom course, additional analyses were conducted with all predictors included together (along with age and sex) in each model to test for unique prediction.

Odds ratios for relative changes (ORs) and risk differences for absolute changes (RDs) are reported. With the high variability in the base rate of different criteria, ORs are useful for ascertaining the strength of each predictor, while absolute risk difference can determine if this effect produces meaningful change at a population level (e.g., an OR of 2.0 can be a 1% change in risk or a 15% change in risk depending on the prevalence of the given phenomenon). This practice of reporting both absolute and relative measures is consistent with recommendations by the Consolidated Standards of Reporting Trials (CONSORT) and the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Holmberg & Andersen, 2020).

RESULTS

Due to the large number of models estimated, we will not discuss every finding in detail. Rather, we present the general trends found in the data and reference the overall course effects found in Figure 1 and Table 2. For specific results for each predictor on each symptom in each component of the course, see Table S1. The results and effect sizes reported below pertain to the single predictor, FDR-corrected models.

FIGURE 1.

Onset, persistence, and recurrence of AUD criteria. The axis is displayed using a logarithmic scale. CD, cut down; FF, role interference; GU, give up; HU, hazardous use; LL, larger/longer; PP, physical/psychological problems; SI, social/interpersonal problems; TL, tolerance; TS, time spent; WD, withdrawal.

Table 2.

Absolute Risk Differences by Predictor

Onset
	LL	CUTd	HU	TL	WD	PP	TS	SI	FF	GU
	n=27661	n=27869	n=28008	n=28691	n=28512	n=29182	n=29432	n=29498	n=29781	n=29629
Overall	0.07	0.07	0.06	0.04	0.04	0.03	0.02	0.01	0.01	0.01
FH+	0.1	0.08	0.07	0.05	0.06	0.04	0.03	0.02	0.01	0.01
FH−	0.07	0.06	0.05	0.04	0.04	0.03	0.02	0.01	0.01	0.01
FHΔ	+0.03*	+0.02*	+0.02*	+0.01*	+0.02*	+0.01*	+0.01*	+0.01*	0*	0*
CD+	0.17	0.11	0.13	0.09	0.11	0.08	0.06	0.05	0.03	0.03
CD−	0.08	0.07	0.06	0.04	0.04	0.03	0.02	0.01	0.01	0.01
CDΔ	+0.09*	+0.04*	+0.07*	+0.05*	+0.07*	+0.05*	+0.04*	+0.04*	+0.02	+0.02*
INTD+	0.09	0.14	0.06	0.05	0.06	0.04	0.03	0.02	0.01	0.01
INTD−	0.08	0.07	0.06*	0.04	0.04	0.03	0.02	0.01	0.01	0
INTDΔ	+0.01*	+0.07*	0	+0.01*	+0.02*	+0.01*	+0.01*	+0.01*	0*	+0.01*
INTF+	0.09	0.08	0.05	0.06	0.06	0.04	0.03	0.02	0.01	0.01
INTF−	0.08	0.07	0.06	0.04	0.04	0.03	0.02	0.01	0.01	0.01
INTFΔ	+0.01*	+0.01*	−0.01	+0.02*	+0.02*	+0.01*	+0.01*	+0.01*	0*	0*
HD+	0.19	0.15	0.14	0.07	0.09	0.07	0.04	0.04	0.02	0.02
HD−	0.05	0.04	0.03	0.03	0.03	0.2	0.01	0.01	0	0
HDΔ	+0.14*	+0.11*	+0.11*	+0.04*	+0.06*	−0.13*	+0.03*	+0.03*	+0.02*	+0.02*
RACE+	0.08	0.06	0.06	0.03	0.06	0.03	0.02	0.01	0.01	0.01
RACE−	0.08	0.09	0.05	0.08	0.05	0.03	0.02	0.02	0.01	0.01
RACEΔ	0*	−0.03*	+0.01*	−0.05*	+0.01*	0*	0	−0.01	0*	0
Persistence
	LL n=2332	CUTd n=2124	HU n=1985	TL n=1302	WD n=1481	PP n=865	TS n=561	SI n=495	FF n=212	GU n=189
Overall	0.46	0.45	0.45	0.23	0.36	0.35	0.25	0.29	0.2	0.2
FH+	0.48	0.54	0.5	0.26	0.41	0.4	0.3	0.35	0.28	0.27
FH−	0.46	0.4	0.43	0.21	0.34	0.31	0.21	0.24	0.08	0.14
FHΔ	+0.02	+0.14*	+0.07*	+0.05*	+0.07*	+0.09*	+0.09*	+0.11*	+0.20*	+0.13*
CD+	0.48	0.53	0.51	0.35	0.47	0.47	0.34	0.38	0.26	0.26
CD−	0.46	0.44	0.45	0.21	0.35	0.32	0.23	0.27	0.18	0.18
CDΔ	+0.02	+0.09*	+0.06	+0.14*	+0.12*	+0.15*	+0.11*	+0.11*	+0.08	+0.08*
INTD+	0.5	0.49	0.42	0.26	0.42	0.41	0.3	0.33	0.23	0.24
INTD−	0.44	0.43	0.46	0.21	0.33	0.31	0.21	0.27	0.16	0.17
INTDΔ	+0.06*	+0.06*	−0.04	+0.05*	+0.09*	+0.10*	+0.09*	+0.06	+0.07*	+0.07*
INTF+	0.54	0.54	0.46	0.31	0.44	0.47	0.37	0.33	0.22	0.23
INTF−	0.45	0.43	0.45	0.21	0.34	0.31	0.22	0.28	0.19	0.19
INTFΔ	+0.09*	+0.11*	+0.01	+0.10*	+0.10*	+0.16*	+0.15*	+0.05*	+0.03	+0.04*
HD+	0.48	0.47	0.47	0.23	0.38	0.36	0.25	0.3	0.21	0.21
HD−	0.26	0.34	0.27	0.19	0.23	0.19	0.25	0.04	0.03	0.07
HDΔ	+0.22*	+0.13*	+0.20*	+0.04	+0.15*	+0.17*	0	+0.26*	+0.18*	+0.14*
RACE+	0.47	0.44	0.46	0.18	0.38	0.36	0.24	0.28	0.22	0.17
RACE−	0.42	0.53	0.43	0.4	0.3	0.28	0.29	0.26	0.22	0.3
RACEΔ	+0.05*	−0.09*	+0.03*	−0.22*	+0.08*	+0.08	−0.05*	+0.02	0	−0.13*
Recurrence
	LL	CUTd	HU	TL	WD	PP	TS	SI	FF	GU
	n=4261	n=3224	n=4448	n=4215	n=2415	n=1436	n=2146	n=1683	n=570	n=382
Overall	0.18	0.18	0.13	0.08	0.12	0.11	0.06	0.06	0.06	0.07
FH+	0.21	0.22	0.13	0.09	0.14	0.13	0.09	0.08	0.07	0.08
FH−	0.17	0.15	0.13	0.07	0.11	0.1	0.04	0.06	0.04	0.06
FHΔ	+0.04*	+0.07*	0	+0.02*	+0.03*	+0.03*	+0.05*	+0.02	+0.03*	+0.02
CD+	0.24	0.19	0.19	0.15	0.19	0.15	0.12	0.08	0.09	0.11
CD−	0.18	0.18	0.13	0.07	0.11	0.11	0.05	0.06	0.05	0.06
CDΔ	+0.06	+0.01	+0.06*	+0.08*	+0.08*	+0.04	+0.07*	+0.02	+0.04*	+0.05*
INTD+	0.19	0.19	0.12	0.08	0.14	0.12	0.09	0.08	0.08	0.1
INTD−	0.18	0.18	0.13	0.07	0.11	0.11	0.04	0.05	0.04	0.04
INTDΔ	+0.01*	+0.01*	−0.01	+0.01*	+0.03*	+0.01*	+0.05*	+0.03*	+0.04*	+0.06*
INTF+	0.18	0.21	0.12	0.1	0.15	0.16	0.09	0.08	0.08	0.14
INTF−	0.18	0.17	0.13	0.07	0.11	0.1	0.05	0.06	0.05	0.04
INTFΔ	0	+0.04*	−0.01	+0.03*	+0.04*	+0.06*	+0.04*	+0.02*	+0.03*	+0.10*
HD+	0.27	0.25	0.2	0.09	0.15	0.14	0.07	0.08	0.08	0.09
HD−	0.1	0.1	0.06	0.05	0.08	0.06	0.03	0.02	0.02	0.04
HDΔ	+0.17*	+0.15*	+0.14*	+0.04*	+0.07*	+0.08*	+0.04*	+0.06*	+0.06*	+0.05*
RACE+	0.18	0.17	0.13	0.07	0.12	0.11	0.05	0.06	0.05	0.07
RACE−	0.22	0.25	0.15	0.15	0.16	0.12	0.12	0.09	0.1	0.06
RACEΔ	−0.04*	−0.08*	−0.02	−0.08*	−0.04	−0.01	−0.07*	−0.03*	−0.05*	+0.01*

Note. Values listed are the proportion of individuals who experienced persistence, onset, and recurrence of a symptom given the presence and absence of each predictor. “Overall” reflects the symptom course when no predictors are applied.

^*indicates a statistically (p < .05) significant effect.

CUTd=Cut Down, FF=Role Interference, GU=Give Up, HU=Hazardous Use, LL=Larger/Longer, PP=Physical/Psychological Problems, SI=Social/Interpersonal Problems, TL=Tolerance, TS=Time Spent, WD=Withdrawal. FH=Family History, CD=Conduct Disorder, INTD = Distress, INTF = Fear, HD = Heavy Drinking, RACE = Race/ethnicity.

Correlations among the predictors in the overall sample were modest, with absolute correlations between 0 and 0.15, except for Fear and Distress (r = 0.31). Significant results were found for 85% (153/180) of the total number of logistic regression analyses conducted. For symptom persistence, significant results were found in 78% (47/60) of the models. For symptom onset, significant results were found for 93% (56/60) of the models. For symptom recurrence, significant results were found for 80% (48/60) of the models. Across all the analyses (persistence, onset, and recurrence), significant results were found for Family History, Conduct Disorder, Distress, Fear, Heavy Drinking, and Race/Ethnicity in 87% (26/30), 76% (23/30), 93% (28/30), 83% (25/30), 93% (28/30), and 70% (21/30) models, respectively.

Magnitude of effects

Overall

Figure 1 displays the odds ratios for each set of analyses. When examining trends across all sets of analyses, the results were as hypothesized, with the strength of the predictor’s influence appearing to be related to the conceptual proximity to AUD. Heavy Drinking had the overall strongest effects (RD range = 13%–26%, median RD = 7.5%; OR range = 1.02–11.37, median OR = 2.99) followed by Conduct Disorder (RD range = 1%–15%, median RD = 6.0%; OR range = 1.01–3.51, median OR = 1.63). Distress (RD range = 4%–10%, median RD = 2.5%; OR range = 0.95–3.1, median OR = 1.53) and Fear (RD range = 1%–16%, median RD = 3.0%; OR range = 0.99–4.31, median OR = 1.59) had very similar risk profiles. Family History (RD range = 0%–20% median RD = 3.0%; OR range = 1.06–5.58, median OR = 1.48) had effect sizes only slightly lower than Fear and Distress, and Race did not produce consistent directional effects (RD range = −22% to 8% median RD = −1%; OR range = 0.62–3.0, median OR = 1.15).

Persistence

Symptom persistence was generally low, with the greatest chronicity found in Larger/Longer (persistence = 46.3%) and the lowest chronicity in Give Up (persistence = 20.2%). The strength of prediction was modest, with effect sizes similar to the recurrence analyses but weaker than the onset analyses (see below). As with onset and recurrence, Heavy Drinking was the strongest overall predictor of symptom persistence, although the effect was less pronounced and was not consistent across all criteria. There was mixed support for our hypothesis, with Heavy Drinking being the strongest predictor of Withdrawal (OR = 2.4, CI = 1.8–2.4), but being a non-significant predictor of Tolerance (OR = 1.2, CI = 1.0–1.5). As hypothesized, Family History had similar effects across criteria. Role Interference was an exception to this trend, producing an unstable estimate with wide confidence intervals (OR = 5.6, CI = 2.4–12.9). Notwithstanding Role Interference, Family History had a median OR of 1.5, and an OR range of 1.1–2.3. Contrary to our hypothesis, Conduct Disorder had relatively little variability across criteria (Median OR = 1.61, Range = 1.0–2.0) and was not a significant predictor of persistence of Hazardous Use (OR = 1.1, CI = 0.9–1.3). Fear and Distress had similar effects on symptom persistence, with no discernable trends. Race had strong effects on symptom persistence, however, these effects were divided between positive, negative, and nonsignificant effects (see Figure 1).

Onset

Because our selection was based on risk factors for the prevalence of AUD, it is not surprising that the predictors’ effects were generally greatest in the onset analyses, relative to persistence and recurrence. Heavy drinking had the largest estimated effects in nine of the 10 criteria (Median OR = 3.4, Range = 1.9–4.6), with Tolerance being the exception (OR = 1.89). Consistent with our hypotheses, Family History appeared to have a similar pattern of effects as Conduct Disorder, but with smaller magnitudes. Conduct Disorder was a significant predictor of onset for all criteria, and as hypothesized, Conduct Disorder appeared to be an especially strong predictor of Hazardous Use (OR = 1.66, CI = 1.4–1.9) and Social/Interpersonal Problems (OR – 2.6, CI = 2.2–3.2). Fear and Distress had relatively weak associations with Hazardous Use and were significant (but weaker than externalizing) predictors of all other criteria. Race had a pattern similar to that found in persistence, with four significant negative effects and three significant positive effects.

Recurrence

Heavy Drinking, relative to other predictors, had the greatest average effect on symptom recurrence (Median OR = 2.7, Range = 1.60–4.0). This difference was evident in high-frequency criteria (Larger/Longer, Cut Down, Hazardous Use), but less clear in other criteria. Contrary to our hypothesis, Tolerance and Withdrawal did not appear to have uniquely strong associations with Heavy Drinking (See Figure 1). As expected, Family History had weaker overall effects than other predictors (median OR = 1.4, Range = 1.1–2.6), with the pattern of effects somewhat similar to Conduct Disorder. While Conduct Disorder had weaker effects than hypothesized, there was some support for the conceptual similarity hypothesis, as Conduct Disorder was the strongest predictor of Hazardous Use (OR = 1.7, CI −1.4 to 1.9). Distress produced slightly greater odds ratios than Fear, (median ORs = 1.6 and 1.4, respectively) but was a weaker predictor of absolute risk (Median RDs = 2% and 3.5%, respectively). Neither Fear nor Distress was associated with the recurrence of Hazardous Use. There was mixed support for the hypothesis that Heavy Drinking would be an especially strong predictor of symptom recurrence. The effects produced by heaver drinking were stronger (relative to other predictors) than those found in persistence, but not onset.

Predictors contained a modest amount of overlapping variance. After including all predictors and using FDR to correct for multiple tests, the persistence analyses had significant effects in 41/60 (68%) predictions. However, Give Up and Role Interference had insufficient sample sizes to produce persistence estimates with all predictors included. In the onset analyses, 51/60 (85%) were significant, as opposed to 56/60 in the single predictor analyses. In the recurrence analyses, 40/60 (67%) were significant, as opposed to 48/60 in the single predictor analyses. The binomial probability of observing such effects by chance is equal to 0.003, <0.0001, and 0.003 for persistence, onset, and recurrence, respectively. This indicates that there is, overall, consistent specificity of prediction. Most differences were due to Distress and Fear, which appeared to be accounting for a fair amount of overlapping variance in the symptom course models. Other results that lost significance were in predictors that already had marginally significant effects in the single predictor models. Details on which results were significant in the uncorrected, corrected, and multiple predictor models are found in Table S3.

Ancillary analyses

Base rate

Additional analyses found that the overall prevalence of a given criterion at baseline (Wave 1) was significantly associated with the persistence (r = 0.90, r_s = 0.87), onset (r = 0.99, r_s = 1.00), and recurrence (r = 0.91, r_s = 0.89) of the criterion. These analyses revealed that, although criteria are affected differently by the six predictors, the overall pattern of symptom course is inextricably tied to its base rate.

Treatment

Receiving treatment for AUD was associated with increased persistence of the criterion, relative to those who did not receive treatment. This effect may be related to the severity of AUD, given that individuals seeking treatment for AUD were likely more severe, and therefore had criteria that were more likely to persist. Further analysis examining the proportion of individuals in treatment by the level of AUD diagnosis found a strong association (Rao-Scott χ²[3, N = 29,993] = 2216.68, p < 0.0001). Of those people who had a severe diagnosis, 23% received treatment. Of those people who had a moderate diagnosis (4–5 symptoms), 8% received treatment, while only 4% of people with a mild diagnosis (2–3 symptoms) received treatment. Only 1% of those who did not qualify for an AUD diagnosis received treatment. Interestingly, even those who received treatment with mild and moderate AUD diagnoses tended to have greater persistence of criteria (mean = 48%, median = 51% than those who did not receive treatment (mean = 27%, median = 23%). Criteria with lower prevalence characterized by larger degrees of impairment (e.g., Role Interference, Give Up) had the largest differences in persistence between those who received treatment and those who did not. This may be in part due to them also having the lowest persistence in general, or due to false positives in the nontreatment seeking groups. Indeed, these low-frequency symptoms had extremely low rates of persistence in the non-treatment-seeking groups and in individuals who did not endorse severe AUD (Table S4).

Heavy drinking over time

Among heavy drinkers, individuals who exceeded daily or weekly drinking limits only at Wave 1 had the lowest rates of onset (mean = 3.3%, range = 1.8%–3.8%) persistence (mean = 8.1%, range = 4.4%–16.3%), and recurrence (mean = 3.8%, range = 2.5%–6.1%). Individuals who exceeded daily or weekly drinking limits at both waves had similar rates of onset (mean = 9.9%, range = 9.0%–10.8%) compared to those who only exceeded at Wave 2 (mean = 9.6%, range = 3.8%–10.5%) Unexpectedly, Heavy Drinking at both waves resulted in significantly lower persistence (mean = 23.5%, range = 14.1%–34.3%) than the persistence found in those who only exceeded drinking limits at Wave 2 (mean = 42.3%, range = 19.6%–96.0%). For recurrence, there was evidence that Heavy Drinking at both waves presented greater risk (mean = 12.5%, range = 8.5%–23.6%) than only exceeding at Wave 2 (mean = 10.2%, range = 7.1%–15.9%). Full results are found in Table S2.

We observed very low criterion stability in non-heavy drinkers. In these light-moderate drinking individuals, symptom persistence ranged from 13.9% (SE = N/A; n = 1) for Time Spent, to 0% (SE = N/A; n = 0) for Role Interference and Give Up. Perhaps less surprisingly, non-heavy drinkers also had very low rates of symptom onset (mean = 1.8%, range = 1.6%–1.8%), and recurrence (mean = 1.9%, range = 1.3%–2.5%).

DISCUSSION

The current study examined the symptom-level course of AUD and explored potential predictors that may relate to their trajectories over 3 years. The results indicate that factors associated with the course of syndromal AUD are also associated with each criterion but vary considerably by criterion and type of course (onset, persistence, or recurrence). Our hypothesis that more nomologically proximal predictors would have greater effects on symptom course was well-supported, and our hypothesized order of the magnitude of effects (Heavy Drinking, Conduct Disorder, Distress, Fear, Family History, Race) was supported when effects were examined across all sets of analyses. The hypothesis that Family History would have a similar, albeit weaker, pattern of effects as Conduct Disorder had some support. The expectation that Heavy Drinking would be the strongest predictor of symptom recurrence was supported, however, Heavy Drinking also tended to be the strongest predictor of symptom onset and persistence.

The results provided only modest support for the hypothesis that predictors conceptually related to specific criteria would have stronger effects on these criteria. Little evidence supported the prediction that heavy drinking would be an especially strong predictor of Tolerance and Withdrawal. While heavy drinking was often the strongest predictor of the course of these criteria, relative to other criteria, the effects on Tolerance and Withdrawal did not appear to be uniquely strong. Counterintuitively, heavy drinking was a weaker predictor of Tolerance than other criteria. This lack of effect may be due to the nature of the sample; being an epidemiological sample rather than a clinical sample, there are proportionately fewer individuals who endorse clinically severe symptoms of physiological dependence on alcohol.

Regarding our hypothesis that Conduct Disorder more strongly influences course in criteria related to externalizing behavior (Hazardous Use, Social/Interpersonal Problems Role Interference), results were mixed. Again, Conduct Disorder was a significant predictor of course for these criteria, but it was unclear if these effects were uniquely strong. The onset analyses yielded the most support for this hypothesis, but the persistence and recurrence analyses were less clear. The hypothesis that Fear would have weak associations with Hazardous Use was supported. Interestingly, Distress also shared this pattern of weak association with Hazardous Use. Many of the overall patterns remained similar across criteria, indicating that there may be an underlying common factor influencing course.

The finding that symptom persistence is greater in individuals who received treatment for AUD has several potential explanations. Some of this effect may result from individuals in treatment having more severe AUD, and thus being more likely to experience chronic symptoms. The data supported this idea, with proportionally more individuals who received treatment having severe AUD. In addition, these findings may be due to comorbidity. Individuals who enter treatment are likely those who are experiencing significant distress or impairment, which presages a more deleterious course (e.g., Berkson’s bias). As such, individuals in the treatment subsample may have greater barriers to recovery and a higher likelihood of continuing to experience AUD symptoms.

Heavy drinking at Wave 1 and Wave 2 resulted in greater rates of onset, persistence, and recurrence than heavy drinking at Wave 1 only and non-heavy drinking at both waves, with these effects most clear in the recurrence analyses. However, those who engaged in heavy drinking at Wave 1 and Wave 2 had comparable rates of onset and persistence to those who engaged in heavy drinking at Wave 2 only. Regarding persistence, one explanation for this finding is that individuals who only engaged in heavy drinking at Wave 2 necessarily increased their drinking between Wave 1 and Wave 2. If they had already endorsed the criterion at Wave 1, it would make sense that they would continue to endorse it at Wave 2, given that their drinking had increased. For onset, it is possible that heavy drinkers had already endorsed some symptoms at Wave 1, meaning that those who had begun drinking heavily between Wave 1 and Wave 2 would be more likely to experience the onset of new symptoms at Wave 2. In addition, it is possible that consumption at Wave 2 is particularly strongly associated with AUD symptoms at Wave 2 due to a time of measurement confound.

One important finding was that, regardless of the unique effect of each predictor on the criterion, the prevalence of any given criterion in the population was the strongest predictor of course. Although it makes statistical sense that higher frequency events are more likely to be observed at any given time point, the magnitude of this effect was striking. As such, it appears that the course of a given symptom cannot be determined independently of the symptom’s prevalence.

Implications

Our findings regarding the lack of symptom persistence highlight the importance of monitoring individual symptoms of AUD. While the evidence of this study is not sufficient to propose a hierarchical model of AUD symptoms, varying stability found in AUD symptoms raises concerns about symptom count approaches to diagnosis (see also Lane & Sher, 2015). In contrast, the relatively consistent overall patterns among criteria and among persistence, onset, and recurrence indicate that a mixed approach incorporating both an underlying syndrome and individual symptom effects may provide a compelling avenue for future research.

Given our findings on the low level of symptom persistence in this epidemiological sample in non-heavy drinkers, we believe it may be important to consider including a minimum consumption requirement in future revisions of the diagnostic criteria. The notion that measures of heavy consumption can supplement and improve the validity of diagnosis has been previously suggested in the AUD literature (e.g., Hoeppner et al., 2011; Saha et al., 2007). Our findings suggest that many individuals who endorse a criterion in the absence of a heavy drinking pattern may be providing false positive responses. If these responses do actually represent true positives, the lack of temporal stability would indicate that these “symptoms” in moderate alcohol consumers do not represent a chronic relapsing disorder, but rather a transient consequence more likely related to circumstances. However, we acknowledge that certain populations may experience serious consequences from relatively low amounts of alcohol consumption (e.g., pregnant women, individuals with medical comorbidities), and the incorporation of a “consumption cutoff” would require further research.

Another important finding for symptom-focused research is the strength of the effect that base rate has on the course of a symptom. Additional research is needed to determine the extent to which individual symptom course varies beyond base rate effects. It has been suggested that part of this variability in prevalence may be due to different instruments/questions capturing different criterion thresholds, which has significant implications for the base rate of a criterion (Boness et al., 2019; Hoffman et al., 2018; Lane et al., 2016). Future work should examine multiple operationalizations of each symptom, with varying “severity” thresholds, examining symptom course with criterion definitions that yield more similar base rates. Until such effects can be isolated, the most prevalent (and presumably least severe) symptoms will continually rise to the forefront of symptom-focused analyses simply on the basis of prevalence, obfuscating which symptoms are truly the most fundamental to a given disorder.

Limitations

Self-report of lifetime criteria (such as those used for recurrence in this study) tends to have questionable accuracy, with respect to AUD symptomatology (Haeny et al., 2016) and psychiatric symptomatology more generally (e.g., Moffitt et al., 2010; Vandiver & Sher, 1991). Self-reported family history of AUD has also been shown to have poor sensitivity (Andreasen et al., 1986; Roy et al., 1994; Schuckit et al., 2020). Consequently, there may have been more Family History than reported, and the true effects of Family History may be stronger than those found by our study. In addition, NESARC is a population based, epidemiological sample rather than a clinical sample, limiting the extent to which these findings can be generalized to clinical populations. It is possible that Family History and Conduct Disorder would have greater predictive effects in a clinical sample, as they would be less influenced by the relatively low rates of endorsement of both the predictors and AUD criteria themselves. Another limitation was the usage of only 10 of the 11 AUD criteria. As the first wave of NESARC only assessed DSM-IV criteria, we were unable to analyze the DSM-5 diagnostic criterion Craving. A broader limitation is the survey nature of data collection, relying exclusively on self-report for measuring symptoms, drinking, and the various predictors used in the analyses. Issues with insight, reporting bias, poor recollection, cognitive impairment, misinterpretation, and a host of other concerns have been discussed in the literature (e.g., Schwarz, 1999). Along with improvements in the diagnostic system and symptom measurement, future research should consider including objective measures of consumption over time using advanced biosensors (Tehrani et al., 2022).

Some analyses required individuals to simultaneously endorse two different low base rate phenomena (e.g., Role Interference and Conduct Disorder). Despite the large initial sample, some subsamples contained few individuals (e.g., Family History predicting the persistence of Give Up, n = 36). Replications will be needed to ensure that the effects discovered in these sparse subpopulations are robust enough for valid interpretation.

CONCLUSIONS

The criteria that constitute AUD vary considerably in prevalence, course, and the extent to which etiologically relevant covariances influence their course. In part, the strength of this influence appears to relate to how proximal a given risk factor is to the behavioral outcome of AUD. In some instances, covariates that are conceptually related to certain symptoms appear to have a significant influence on their course, leading to the differential prediction of symptoms that goes beyond the overall pattern associated with the behavioral proximity of the predictor. However, the strength of the findings was mixed, with a large degree of heterogeneity and a strong effect seemingly driven by the overall prevalence of the symptom. Given that the prevalence of a symptom is, by definition, strongly related to its severity and there is great heterogeneity in the relative prevalence of symptoms due to alternative operationalizations (Lane et al., 2016), it seems likely that the course of symptoms ascertained via different assessment instruments is likely to vary accordingly. We maintain that research on symptom “behavior” and diagnosis more generally needs to address this fundamental problem if we hope to legitimately put “precision” in our “medicine.”

APPENDIX 1

It would be possible to measure onset at Wave 1, by using individuals who did not endorse a symptom prior to Wave 1 and did endorse the symptom in the past 12 months at Wave 1. This was not used for several reasons. First, previous research (Vergés et al., 2014) found that using a contemporaneous outcome (having the predictor and outcome at the same measurement occasion) produces considerably stronger effects than those found in offset measurements. In supplemental analyses with the current data, we observed the same phenomenon. Second, we aimed to minimize the amount to which the results rely on retrospection. Thus, to avoid introducing a time-of-measurement confound and unnecessary retrospection into the results, only Wave 2 onsets were used in the study.

APPENDIX 2

Tolerance = “Tolerance, as defined by either of the following: (a) A need for markedly increased amounts of alcohol to achieve intoxication or desired effect (b) A markedly diminished effect with continued use of the same amount of alcohol.” Cut Down = “There is a persistent desire or unsuccessful efforts to cut down or control alcohol use;” Larger/Longer = “Alcohol is taken in larger amounts or over longer periods than was intended;” Time Spent = “A great deal of time is spent in activities necessary to obtain alcohol, use alcohol, or recover from its effects;” Give Up = “Important social, occupational, or recreational activities given up or reduced because of alcohol use;” Physical/Psychological = “Alcohol use is continued despite knowledge of having a persistent or recurrent physical or psychological problem that is likely to have been caused or exacerbated by alcohol.” Withdrawal = “Withdrawal, as manifested by either of the following: (a) The characteristic withdrawal syndrome for alcohol, (b) Alcohol (or a closely related substance, such as benzodiazepine) is taken to relieve or avoid withdrawal symptoms.” Hazardous Use = “Recurrent alcohol use in situations in which it is physically hazardous.” Social/Interpersonal = “Continued alcohol use despite having persistent or recurrent social or interpersonal problems caused or exacerbated by the effects of alcohol.” Role Interference = “Recurrent alcohol use resulting in a failure to fulfill major role obligations at work, school, or home”.