Anxiety Disorders as Predictors of Substance Use: Social Anxiety, Alcohol Use Disorder, and Longitudinal Evidence

The relationship between anxiety disorders and substance use disorders (SUDs) represents one of the most clinically significant comorbidity patterns in psychiatric practice. Among all anxiety–substance pairings, the link between social anxiety disorder (SAD) and alcohol use disorder (AUD) has generated the most robust longitudinal evidence. The prevailing theoretical framework — the self-medication hypothesis, originally articulated by Khantzian (1985) — posits that individuals use psychoactive substances to alleviate distressing affective states. In the case of SAD and alcohol, this hypothesis has received strong but nuanced empirical support.

Epidemiological data consistently demonstrate that anxiety disorders, as a class, precede the onset of substance use disorders in the majority of comorbid cases. The National Comorbidity Survey Replication (NCS-R) found that among individuals with both an anxiety disorder and an SUD, the anxiety disorder had an earlier age of onset in approximately 75% of cases. Social anxiety disorder is particularly implicated: its typical onset in early adolescence (median age of onset: 13 years per DSM-5-TR) places it years before the normative window of alcohol initiation, creating a developmental period during which untreated social fears may drive alcohol-seeking behavior.

This article examines the specific mechanisms — neurobiological, psychological, and developmental — through which anxiety disorders, and SAD in particular, function as predictors of subsequent substance use. We review landmark longitudinal studies, dissect the neural circuitry involved, analyze treatment response data for comorbid populations, and identify prognostic factors that differentiate patients who develop SUDs from those who do not.

The co-occurrence of SAD and AUD is far more common than chance would predict. According to the National Epidemiological Survey on Alcohol and Related Conditions (NESARC) — one of the largest psychiatric epidemiological studies ever conducted (N > 43,000) — the lifetime prevalence of SAD in the general U.S. population is approximately 5.0%, while the lifetime prevalence of AUD is approximately 29.1% (using DSM-5 criteria applied retrospectively). Among individuals with SAD, however, the lifetime prevalence of AUD rises to approximately 48%, representing an odds ratio of roughly 2.2–2.8 depending on the analytic model and demographic adjustments.

Conversely, among individuals presenting for treatment of AUD, the prevalence of comorbid SAD ranges from 20% to 40% across clinical samples — substantially higher than the general-population base rate. The National Comorbidity Survey (NCS; Kessler et al., 1997) identified SAD as the anxiety disorder most strongly associated with subsequent alcohol dependence, with a prospective odds ratio of 2.4 (95% CI: 1.8–3.2) after adjusting for demographics and other psychiatric comorbidities.

Several demographic and clinical modifiers shape this comorbidity:

• Sex: Males with SAD are at higher risk for developing AUD than females with SAD, likely reflecting sex-differentiated sociocultural norms around alcohol use. However, the relative increase in risk conferred by SAD is actually larger in women.
• Severity: Generalized SAD (fear across most social situations, as distinguished from performance-only SAD in DSM-5-TR) carries a substantially higher risk for AUD than performance-only subtype.
• Age of onset: Earlier SAD onset (before age 11) is associated with greater AUD risk, likely due to longer exposure to untreated distress during critical developmental windows.
• Comorbidity cascades: SAD frequently co-occurs with major depressive disorder (MDD; approximately 40–50% lifetime comorbidity), and the triad of SAD + MDD + AUD is common in clinical settings, with each condition worsening the prognosis of the others.

The critical question in this literature is directionality: does anxiety cause substance use, does substance use cause anxiety, or are both driven by shared vulnerabilities? Longitudinal studies have been instrumental in disentangling these pathways.

The Oregon Adolescent Depression Project (OADP; Stein et al., 2001) followed a community sample from adolescence into early adulthood and found that baseline SAD in adolescence predicted new-onset alcohol dependence at follow-up, even after controlling for baseline alcohol use, depression, and other anxiety disorders. The hazard ratio was approximately 2.1 for the transition from no AUD to AUD over the follow-up period.

The Zurich Cohort Study — a prospective study following a cohort from age 20 to age 50 — demonstrated that social fears assessed at age 20 predicted the development of alcohol problems by age 30, with this association remaining significant after adjustment for personality factors and baseline substance use. Importantly, the reverse pathway (early alcohol problems predicting later social anxiety) was not significant in this sample, supporting the temporal primacy of anxiety.

The National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) Wave 1 to Wave 2 analysis (N ≈ 34,653; 3-year follow-up) found that baseline SAD significantly predicted the incidence of new-onset AUD (adjusted OR: 1.6–1.9), whereas baseline AUD did not reliably predict new-onset SAD. This asymmetry is critical: it suggests that the predominant causal arrow, at the population level, points from anxiety to substance use.

A meta-analysis by Wolitzky-Taylor et al. (2012) synthesizing longitudinal data from 12 prospective studies confirmed that anxiety disorders assessed at baseline predicted subsequent development of SUDs (pooled OR: 1.5–2.0), with stronger effects for SAD than for generalized anxiety disorder (GAD) or specific phobia. Panic disorder showed comparable predictive strength to SAD for alcohol-related outcomes.

However, the evidence is not entirely unidirectional. A subset of individuals develop anxiety secondary to chronic heavy alcohol use — particularly during withdrawal and early abstinence. The substance-induced anxiety disorder pathway is recognized in both DSM-5-TR and ICD-11, and distinguishing primary from substance-induced anxiety is a diagnostic challenge discussed further below.

The neurobiological overlap between anxiety processing and alcohol's pharmacological effects provides a mechanistic explanation for why individuals with SAD preferentially use alcohol over other substances. Several interacting neural systems are implicated.

GABAergic System

Alcohol's acute anxiolytic effect is mediated primarily through positive allosteric modulation of GABA_A receptors, particularly those containing the α2 and α3 subunits, which are concentrated in the amygdala and prefrontal cortex — the core nodes of the fear and social evaluation circuitry. In individuals with SAD, there is evidence of reduced baseline GABAergic tone. A positron emission tomography (PET) study by Pollack et al. using [¹¹C]flumazenil demonstrated reduced benzodiazepine binding potential in the amygdala, hippocampus, and prefrontal cortex in patients with SAD compared to controls. This GABAergic deficit may render alcohol's GABA-enhancing properties disproportionately reinforcing in this population.

Amygdala Hyperreactivity and Threat Processing

Functional neuroimaging studies consistently show that individuals with SAD exhibit exaggerated amygdala activation in response to social threat cues (e.g., angry or contemptuous faces). Acute alcohol administration reduces amygdala reactivity to threatening stimuli in both healthy volunteers and individuals with SAD, effectively normalizing the hyperactive threat-detection system. This pharmacological "correction" constitutes the neural basis of self-medication: alcohol temporarily resolves a circuit-level dysfunction.

Mesolimbic Dopamine and Reinforcement Learning

The mesolimbic dopamine pathway (ventral tegmental area → nucleus accumbens) governs reward prediction and reinforcement learning. In individuals with SAD, the relief from social distress produced by alcohol generates a strong negative reinforcement signal — the removal of an aversive state — which is encoded as a dopaminergic prediction error. Over repeated exposures, this creates a conditioned association between social contexts, distress, and alcohol-seeking behavior. Importantly, this reinforcement pathway is distinct from the positive reinforcement (euphoria) that characterizes early alcohol use in many individuals without anxiety, and it may explain why anxiety-motivated drinkers develop dependence through a different trajectory.

HPA Axis and Stress Reactivity

The hypothalamic-pituitary-adrenal (HPA) axis shows characteristic dysregulation in both anxiety disorders and AUD. Individuals with SAD demonstrate elevated cortisol reactivity to social evaluative stressors (e.g., the Trier Social Stress Test). Chronic alcohol use, paradoxically, further dysregulates the HPA axis — producing blunted cortisol responses during intoxication but exaggerated cortisol release during withdrawal. This creates a neurobiological vicious cycle: anxiety drives alcohol use, alcohol withdrawal amplifies anxiety, and the HPA axis becomes progressively more dysregulated.

Genetic Overlap

Twin studies estimate that the genetic correlation between anxiety disorders and AUD is approximately r = 0.30–0.40, indicating moderate shared genetic liability. Genome-wide association studies (GWAS) have identified shared risk loci, particularly in genes regulating GABAergic and serotonergic neurotransmission. The SLC6A4 gene (serotonin transporter) and GABRA2 gene (GABA_A receptor α2 subunit) have been implicated in both SAD and AUD in candidate gene studies, though GWAS replications have yielded mixed results. More recently, polygenic risk score analyses suggest that the shared genetic architecture is primarily driven by general internalizing liability (the "p" factor or internalizing spectrum) rather than disorder-specific genetic pathways.

Accurate diagnosis of comorbid SAD and AUD is complicated by significant symptom overlap, substance-induced confounds, and assessment timing issues. Several pitfalls deserve clinical attention.

Primary vs. Substance-Induced Social Anxiety

DSM-5-TR distinguishes between social anxiety disorder (300.23; F40.10) and substance/medication-induced anxiety disorder (F16.180, etc.). The key differentiating factors include: (a) whether the anxiety symptoms preceded the onset of substance use, (b) whether they persist during sustained abstinence (typically assessed after 4 or more weeks of sobriety), and (c) whether they are temporally linked to intoxication or withdrawal episodes. In clinical practice, this distinction requires detailed timeline reconstruction — a task complicated by recall bias and the insidious onset of both conditions.

Research from the Collaborative Study on the Genetics of Alcoholism (COGA) suggests that approximately 15–20% of individuals with co-occurring SAD and AUD have substance-induced anxiety that remits fully with sustained abstinence. The remaining 80–85% have independent (primary) SAD that persists and requires its own treatment.

Avoidance vs. Drinking Contexts

A diagnostic paradox exists: the DSM-5-TR criterion for SAD includes avoidance of social situations, yet many individuals with SAD do not avoid social situations because they use alcohol to endure them. Standard diagnostic interviews (e.g., SCID-5, MINI) may under-detect SAD in active drinkers because the avoidance criterion appears unmet. Clinicians should assess endurance with significant distress or substance use as an avoidance substitute, which the DSM-5-TR text acknowledges but which structured interviews do not always capture.

Disentangling Social Anxiety from Alcohol Withdrawal

Alcohol withdrawal produces autonomic hyperarousal, tremulousness, and acute anxiety that can mimic SAD symptoms. Early-sobriety assessments of social anxiety are unreliable. Best practice involves provisional diagnosis at intake, with diagnostic confirmation after at least 2–4 weeks of abstinence. The Liebowitz Social Anxiety Scale (LSAS) and the Social Phobia Inventory (SPIN) are validated self-report measures that can be administered serially to track symptom trajectories during early sobriety.

Comorbid Depression as a Confound

Given that SAD, AUD, and MDD frequently co-occur, clinicians must differentiate social withdrawal due to depressive anhedonia from social withdrawal due to social fear. The phenomenological distinction — social avoidance due to fear of negative evaluation (SAD) vs. social avoidance due to loss of interest or energy (MDD) — has different treatment implications. Both may be present simultaneously, requiring layered case formulation.

Treatment of comorbid SAD and AUD is an area where the evidence base has expanded substantially over the past two decades, though it remains insufficient relative to clinical need. The central question is whether to treat the disorders sequentially, in parallel, or through integrated protocols.

Sequential vs. Integrated Treatment

Historically, addiction treatment programs advocated for achieving sobriety before addressing anxiety, on the assumption that anxiety was primarily substance-induced. This approach has been challenged by evidence showing that untreated primary anxiety is a major predictor of relapse. A randomized controlled trial by Randall et al. (2001) compared CBT for alcohol dependence alone vs. CBT for alcohol dependence plus CBT for social phobia in comorbid individuals. Surprisingly, the combined group did not show superior alcohol outcomes — but this study had significant methodological limitations including small sample size and ceiling effects from intensive alcohol-focused treatment in both groups.

Subsequent research has generally supported integrated treatment. A study by Terra et al. (2006) demonstrated that treating SAD with paroxetine during inpatient alcohol rehabilitation led to greater SAD symptom reduction and lower relapse rates at 6-month follow-up compared to placebo.

Cognitive-Behavioral Therapy (CBT)

CBT for SAD has robust evidence as a standalone treatment, with response rates of 50–65% and a number needed to treat (NNT) of approximately 3–4 vs. waitlist controls. In comorbid SAD–AUD populations, CBT for SAD produces similar anxiety reductions but with smaller and more inconsistent effects on alcohol consumption. The Clark and Wells cognitive model — targeting self-focused attention and post-event rumination — has shown particular promise in comorbid samples because it directly addresses the cognitive processes that drive drinking in social contexts.

Pharmacotherapy

SSRIs are first-line pharmacotherapy for SAD, with paroxetine, sertraline, and fluvoxamine holding the strongest evidence bases. In SAD monotherapy trials, SSRI response rates are approximately 50–60% vs. 30–35% for placebo (NNT ≈ 4–6). For comorbid SAD–AUD, the evidence is more nuanced:

• Paroxetine: The Terra et al. study showed benefit in comorbid patients. A larger study by Book et al. (2008) found that paroxetine reduced social anxiety but not drinking outcomes in comorbid individuals. Notably, paroxetine reduced drinking only in the subgroup with lower baseline alcohol severity.
• Sertraline: The COMBINE-related analysis and other trials suggest sertraline has modest effects on both anxiety and alcohol consumption, but results have been inconsistent across studies.
• Gabapentin: Emerging evidence suggests gabapentin may uniquely address both anxiety and alcohol craving through GABAergic modulation. A randomized trial by Furieri and Nakamura-Palacios (2007) showed gabapentin reduced both social anxiety and alcohol consumption in comorbid patients, though replication with larger samples is needed.
• Naltrexone: The opioid antagonist naltrexone is FDA-approved for AUD and has been studied as adjunctive treatment in comorbid populations. While naltrexone does not directly treat anxiety, it may reduce the reinforcing properties of alcohol's anxiolytic effect, potentially disrupting the self-medication cycle. No large trial has tested this hypothesis specifically in SAD–AUD comorbidity.

Motivational Enhancement Therapy (MET)

MET and motivational interviewing approaches have demonstrated efficacy for AUD (Project MATCH showed MET produced comparable outcomes to CBT and twelve-step facilitation). In comorbid populations, MET may be particularly useful for addressing ambivalence about reducing alcohol use in individuals who rely on it as an anxiety management strategy. However, MET does not directly target anxiety symptoms and is insufficient as a standalone treatment for SAD.

Several factors have been identified that predict treatment response and long-term outcome in individuals with comorbid anxiety and substance use disorders.

Predictors of Poor Outcome

• Earlier onset of SAD: Onset before age 10 is associated with more severe alcohol pathology and poorer treatment response, likely reflecting a longer period of reinforced self-medication learning.
• Generalized subtype of SAD: Compared to performance-only SAD, generalized SAD is associated with greater functional impairment, more pervasive avoidance, and higher relapse rates after AUD treatment.
• Comorbid depression: The SAD–MDD–AUD triad is associated with higher dropout rates from treatment (estimated at 35–50% vs. 20–30% for SAD–AUD without depression), more suicide attempts, and poorer social functioning at follow-up.
• Drinking to cope motives: Assessed via the Drinking Motives Questionnaire (DMQ), coping-motivated drinking (as opposed to social or enhancement motives) is the strongest motivational predictor of poor AUD outcome in anxious drinkers.
• Severity of alcohol dependence at treatment entry: Higher scores on the Alcohol Use Disorders Identification Test (AUDIT) and more prior detoxification episodes predict lower likelihood of sustained remission.
• Avoidant personality disorder: DSM-5-TR recognizes substantial overlap between generalized SAD and avoidant personality disorder (AvPD). Comorbid AvPD is associated with slower treatment response and lower remission rates in both anxiety and substance use domains.

Predictors of Good Outcome

• Later onset of AUD: When AUD develops later (e.g., in the 20s rather than mid-teens), it may be less entrenched and more responsive to treatment of the underlying anxiety.
• Treatment engagement with both conditions: Individuals who receive concurrent anxiety and substance use treatment show better outcomes across both domains than those treated for one condition alone.
• Social support and non-drinking social network: Access to alcohol-free social contexts where exposure to feared social situations can occur without alcohol is a significant positive prognostic factor.
• Higher baseline motivation for change: Measured by readiness rulers or the URICA (University of Rhode Island Change Assessment), higher motivation at intake predicts better adherence and outcomes in integrated treatment.

While SAD–AUD is the most extensively studied anxiety–substance comorbidity, other pairings carry significant clinical importance.

Panic Disorder and Alcohol/Benzodiazepines

Panic disorder (PD) is associated with elevated rates of AUD (lifetime comorbidity approximately 25–30%) and problematic benzodiazepine use. The acute anxiolytic effect of both substances on panic symptoms creates strong negative reinforcement. Notably, benzodiazepine dependence in PD patients is iatrogenic in a substantial minority of cases, highlighting the importance of prescribing guidelines that limit duration and dosage.

PTSD and Polysubstance Use

Post-traumatic stress disorder shows the highest rates of SUD comorbidity among the anxiety-related disorders, with lifetime SUD prevalence of approximately 40–50% in individuals with PTSD. Alcohol, cannabis, and opioids are all commonly used. The longitudinal evidence supports the self-medication model: PTSD typically precedes SUD onset, and the relationship is mediated by hyperarousal and re-experiencing symptoms more than by avoidance symptoms.

Generalized Anxiety Disorder and Alcohol

GAD–AUD comorbidity is common (lifetime AUD in GAD patients: approximately 30–35%), but the longitudinal evidence for directionality is less consistent than for SAD. Some studies suggest a bidirectional relationship, where chronic worry drives drinking and chronic drinking exacerbates worry through neuroadaptive changes in stress systems.

A meta-analysis by Lai et al. (2015) examining anxiety disorders broadly as predictors of SUD found that the strongest predictive associations were for SAD and PD, with weaker associations for specific phobia and GAD. This hierarchy likely reflects the degree to which each anxiety disorder creates distress in contexts where alcohol is readily available (i.e., social settings for SAD; any setting for PD).

The typical developmental sequence — SAD onset in early adolescence followed by alcohol initiation in mid-to-late adolescence — creates a critical window for prevention. If SAD is identified and treated before alcohol experimentation begins, the self-medication pathway may be interrupted.

School-based anxiety prevention programs, such as FRIENDS for Life and Cool Kids, have demonstrated efficacy in reducing anxiety symptoms in children and adolescents, with effect sizes (Cohen's d) of approximately 0.30–0.50 for anxiety reduction. However, no study has yet demonstrated that early anxiety treatment reduces subsequent SUD incidence — this is a critical gap in the prevention literature that requires long-duration longitudinal trials.

The Australian National Survey of Mental Health and Wellbeing data suggest that only approximately 35% of individuals with SAD ever receive any mental health treatment, and the median delay from onset to first treatment contact is 15–20 years. This treatment gap means that most individuals with SAD pass through the highest-risk period for alcohol misuse initiation entirely untreated.

Interventions targeting adolescents with elevated social anxiety who are beginning to experiment with alcohol represent a theoretically optimal prevention strategy. Brief interventions combining social anxiety psychoeducation with motivational interviewing about alcohol use have shown preliminary promise in college-age samples (e.g., Clerkin et al., 2016), though large-scale trials are lacking.

Despite substantial progress, the field faces several limitations and active areas of investigation.

Limitations of Existing Evidence

• Measurement heterogeneity: Studies vary widely in how they assess social anxiety (diagnostic interviews vs. self-report screening tools) and alcohol problems (diagnostic criteria vs. consumption measures), complicating meta-analytic synthesis.
• Insufficient treatment trials: Fewer than 10 randomized controlled trials have specifically studied pharmacological or psychological treatments in well-characterized SAD–AUD comorbid samples. Most evidence is derived from post-hoc subgroup analyses of trials designed for one condition.
• Underrepresentation of diverse populations: Most longitudinal and treatment studies have been conducted in majority-White, Western samples. The cultural context of social anxiety (e.g., taijin kyofusho in Japanese culture) and alcohol use norms vary dramatically across populations.
• Reliance on self-report for substance use: Objective biomarkers (e.g., phosphatidylethanol, ethyl glucuronide) are underutilized in comorbidity research, potentially underestimating alcohol consumption.

Emerging Research Directions

• Computational psychiatry approaches: Reinforcement learning models are being applied to understand how anxious individuals differentially weight negative reinforcement signals from alcohol. Preliminary work by Gillan and colleagues suggests that anxiety disorders are associated with model-free (habitual) rather than model-based (goal-directed) decision-making, which may explain the compulsive quality of anxiety-driven drinking.
• Neuromodulation: Transcranial magnetic stimulation (TMS) targeting the dorsolateral prefrontal cortex (dlPFC) has shown promise for both anxiety disorders and AUD independently. A circuit-based approach targeting shared prefrontal–amygdala dysconnectivity is under investigation for comorbid populations.
• Psychedelic-assisted therapy: Psilocybin-assisted therapy has shown preliminary efficacy for both alcohol dependence (Bogenschutz et al., 2022) and treatment-resistant anxiety. Whether it addresses the shared pathology in comorbid populations is a question of active investigation.
• Digital phenotyping: Smartphone-based ecological momentary assessment (EMA) studies are capturing real-time associations between anxiety spikes and alcohol use in naturalistic settings, providing unprecedented granularity on the self-medication cycle. Early data confirm that within-person increases in social anxiety predict within-day increases in alcohol consumption, particularly on days with social events.

The evidence is clear that social anxiety disorder is a robust predictor of subsequent alcohol use disorder, with the predominant causal direction running from anxiety to substance use in the majority of comorbid cases. This has several direct clinical implications:

• Screen for anxiety in substance use treatment settings: All patients presenting for AUD treatment should be screened for SAD (and other anxiety disorders) using validated instruments such as the LSAS, SPIN, or the Social Phobia section of the MINI. Screening should be repeated after 2–4 weeks of abstinence to confirm diagnosis.
• Screen for substance use in anxiety treatment settings: Individuals with SAD should be routinely assessed for problematic alcohol use, including quantity/frequency measures, AUDIT scores, and assessment of drinking motives.
• Integrate treatment: The evidence favors concurrent, integrated treatment of both conditions over sequential approaches. This may involve combined CBT protocols addressing both social anxiety cognitions and alcohol-related coping, along with pharmacotherapy targeting both conditions.
• Prioritize early intervention for social anxiety: The developmental window between SAD onset (typically age 8–15) and alcohol initiation (typically age 14–18) represents a prevention opportunity. Effective treatment of childhood and adolescent SAD may reduce lifetime AUD risk.
• Assess drinking motives: Understanding why a patient drinks (coping vs. social vs. enhancement motives) is essential for tailoring treatment. Coping-motivated drinking is the strongest marker of anxiety-driven alcohol use and predicts the poorest outcomes without anxiety-focused intervention.
• Monitor the HPA axis vicious cycle: Clinicians should educate patients about how withdrawal-induced anxiety amplifies social fears, creating a neurobiological trap that reinforces continued drinking. This psychoeducation, grounded in neuroscience, can enhance treatment motivation and normalize the early-sobriety anxiety surge.

The convergence of epidemiological, longitudinal, neurobiological, and clinical trial evidence establishes the anxiety-to-substance-use pathway as one of the best-supported causal sequences in psychiatric comorbidity. Translating this knowledge into earlier identification, integrated treatment protocols, and targeted prevention strategies remains the central challenge for the field.