Prospective Analysis of the Relation Between DSM-III Anxiety Disorders and Alcohol Use Disorders

Individuals suffering with an anxiety disorder frequently manifest a co-occurring alcohol use disorder (1–7). Although base rates vary for specific subtypes of both alcohol use disorders and anxiety disorders, studies in a variety of populations show that the risk for either is about three times greater given the presence of the other (4, 5, 8). This increase in risk defines alcohol disorders as “comorbid” with anxiety disorders.

Researchers have conceptualized this association as an epidemiological puzzle for which the potential causal relationships linking the comorbid conditions should be delineated and explored as alternative hypotheses (8–10). Possibilities include 1) having an anxiety disorder may promote the development or maintenance of an alcohol use disorder; 2) having an alcohol use disorder may promote the development or maintenance of an anxiety disorder; and 3) a third variable (e.g., familial/genetic factor) may promote the development or maintenance of both an alcohol use disorder and an anxiety disorder. Further, hybrid models involving several of these causal effects are possible.

Unfortunately, the study designs that typify research in this area (cross-sectional and descriptive) do not allow for strong causal inferences to be drawn (11). Highlighting the resulting ambiguity, two recent comprehensive reviews of this literature (8, 12) came to differing conclusions concerning the likely causes of high comorbidity rates. However, even these authors agree that the current level of evidence does not allow for strong conclusions concerning the nature of causal influences among comorbid conditions (13, 14).

Prospective studies are well suited to test the fit of competing causal models of comorbidity (9); however, as implied above, the use of this approach is rare. In fact, Schuckit and Hesselbrock (12) identified only four such studies in their recent review of this literature. Unfortunately, each of these studies had only a very limited capacity to elucidate comorbidity processes. For example, two of these studies assessed psychological symptoms (e.g., Minnesota Multiphasic Personality Inventory profiles) rather than anxiety disorders per se (15, 16). One study predicted substance use at age 16 from teacher-rated shyness of subjects in the first grade (17). Vaillant’s prospective work, although still providing useful information about the development of alcoholism (18), is limited in its quantitative treatment of questions related to comorbidity (19, 20).

We undertook the work described herein with the aim of further exploiting the power of prospective studies to elucidate the longitudinal relationship between anxiety disorders and alcohol use disorders. Notable advantages of this study over those few prospective studies referenced earlier include 1) modern diagnostic technology aimed at reliably identifying clinical syndromes of interest; 2) a sample size and analytic approach adequate for probing tests of putative causal models; 3) the use of a nonclinical (i.e., student) sample assessed early in the window of risk for onset of the disorders measured; and, 4) repeated measurements at 3-year intervals (i.e., measurement occasions spaced closely enough in time as to be likely to capture critical interactions between disorders but not so closely that stability of diagnosis might obscure effects of interest).

METHOD

Participants

Baseline sample. During the academic year 1987–1988, a mixed-sex sample of first-time freshmen (mean age=18.6 years, SD=1.0) was chosen for participation in a longitudinal study examining a range of variables related to alcohol use and abuse (21). This baseline sample (i.e., the group available for analysis at the year 1 assessment) consisted of 489 young adults screened from all incoming first-time college freshmen at a large Midwestern university. Screening instruments included two versions of the Short Michigan Alcoholism Screening Test (22) adapted for rating paternal and maternal drinking problems (23) and sections of the Family History Research Diagnostic Criteria (FH-RDC) interview (24). Participants were classified as family-history-positive for alcoholism if they scored a 4 or greater on the Short Michigan Alcoholism Screening Test (in some cases, 3 was used as a cutoff score) (see reference 21) and if their biological father was diagnosed with alcoholism on the basis of the FH-RDC. Family-history-negative for alcoholism was assigned to participants if, on the basis of FH-RDC, they had no biological first-degree relatives with alcoholism or drug use disorders or antisocial personality disorder and no second-degree relatives with alcoholism or drug use disorders. Participants were assessed on a range of psychosocial measures at baseline and on four subsequent occasions over the next 6 years. Written informed consent was obtained at each measurement occasion, following a description of the procedures.

Longitudinal sample. Analyses for the current study focused on those 454 participants (109 family-history-positive men, 109 family-history-negative men, 124 family-history-positive women, 112 family-history-negative women) who provided complete data on the National Institute of Mental Health Diagnostic Interview Schedule (DIS) interview (25) at years 1, 4, and 7. The current analyses focused on these three occasions because they occurred at roughly equal 3-year intervals corresponding to the freshman year of college, 3 years later when many participants were seniors, and again 3 years later when many participants had entered the workforce. The longitudinal sample across these three waves represents 92.8% of the baseline sample.

Psychiatric Diagnoses

Results presented here used past year diagnoses for a number of anxiety disorders (generalized anxiety disorder, social phobia, agoraphobia with panic attacks, panic attacks in the absence of agoraphobia) and alcohol abuse and dependence by using data obtained from the DIS. The DIS Version III-A (25) was administered at year 1, and the DIS Version III-R (26) was administered at years 4 and 7. To maintain continuity for the current report, all diagnoses were made according to criteria from DSM-III.

Four 12-month DIS/DSM diagnoses relating to anxiety problems were used: generalized anxiety disorder (yearly prevalence rates ranged from 6.2% to 14.3%), social phobia (yearly prevalence rates ranged from 1.3% to 2.4%), agoraphobia with panic attacks (yearly prevalence rates ranged from 0.4% to 0.7%), and panic attacks in the absence of agoraphobia (yearly prevalence rates ranged from 1.8% to 5.3%).

The DIS interview used in this study is keyed to DSM-III diagnostic criteria and, therefore, requires “worry for at least one month” to fulfill the duration requirement for a diagnosis of general anxiety disorder. However, the DSM-III-R increased the required duration of “worry” for a diagnosis of general anxiety disorder to 6 months. Therefore, some participants diagnosed with general anxiety disorder in this study would not have met the criteria for a diagnosis of general anxiety disorder based on the DSM-III-R criteria.

While DSM-III-R requires experiencing four panic attacks in a 4-week period or one panic attack and at least 1 month of worry about panic, DSM-III requires three panic attacks in a 3-week period. In order to identify nonagoraphobic individuals who have experienced repeated panic attacks at some point in their lives, we grouped those reporting either more than five panic attacks over their lifetime or at least three panic attacks having occurred over any 1-month period in the absence of agoraphobic avoidance. Because of the uncertain correspondence between this group and the DSM-III and DSM-III-R diagnosis of panic disorder, we labeled this group “DIS panic attacks.”

Agoraphobia without panic attacks is a highly controversial and not well understood DSM-III disorder. Therefore, we excluded from the “agoraphobia” group those for whom panic attacks were not identified by the interview. Consistent with DSM-III criteria for the diagnosis of agoraphobia with panic attacks, all participants in the “agoraphobia” group had experienced panic attacks in the course of their avoidance.

A new variable representing any DIS anxiety disorder was coded present or absent on the basis of whether an individual met DSM-III diagnostic criteria for the past 12 months for either generalized anxiety disorder, social phobia, agoraphobia with panic attacks, or panic attacks.

Two DIS/DSM-III alcohol diagnoses were used: alcohol dependence and alcohol abuse. Alcohol abuse was diagnosed if the individual showed a pattern of pathological alcohol use (e.g., need for daily use, binges) or impairment in social or occupational functioning (e.g., legal problems related to alcohol use, violence while intoxicated). Alcohol dependence was diagnosed if the individual showed either a pattern of pathological alcohol use or impairment in social or occupational functioning and if tolerance or withdrawal symptoms were present. Diagnoses for alcohol use disorders were broken down into broadband diagnosis (alcohol abuse and dependence) and narrow-band diagnosis (alcohol dependence alone). Results based on each level of diagnosis are presented.

RESULTS

Effects of Family History, Sex, and Time on the Prevalence of Anxiety Disorders and Alcohol Use Disorders

We examined anxiety disorders, alcohol dependence or alcohol abuse and dependence diagnoses separately using two family history (family-history-positive versus family-history-negative) by two sex (male versus female) by three assessment period (years 1 versus 4 versus 7) repeated measures, weighted least squares analysis for categorical data (analogous to a repeated measures ANOVA) (27). (Note that none of the two- and three-way interactions were significant and are, therefore, not reported herein.)

Results showed that anxiety disorder diagnoses were more common among family-history-positive versus family-history-negative subjects across the three data collection waves (year 1=26.6% versus 13.6%; year 4=11.2% versus 5.9%; year 7=12.0% versus 8.6%, respectively) (χ²=11.72, df=1, N=454, p<0.01). Anxiety disorder diagnoses were also more common at the earlier data collection waves (year 1=20.3%; year 4=8.6%; year 7=10.4%) (χ² =31.17, df=2, N=454, p<0.001). Anxiety disorder diagnosis rates were not statistically different between men and women (χ²=0.26, df=1, N=454, p>0.50).

Not surprisingly, alcohol dependence diagnoses were more common among family-history-positive than family-history-negative subjects across the three data collection waves (year 1=9.0% versus 3.6%; year 4=8.2% versus 4.5%; year 7=8.2% versus 3.6%) (χ² =4.55, df=1, N=454, p<0.05). Also expected was the finding that alcohol dependence diagnoses were more common among men than women for the three data collection waves (year 1=7.3% versus 5.5%; year 4=10.6% versus 2.5%; year 7=9.2% versus 3.0%) (χ²=7.34, df=1, N=454, p<0.01). However, alcohol dependence diagnosis rates did not differ across the three data collection waves (χ²=0.09, df=2, N=454, p>0.50).

Paralleling findings for alcohol dependence or alcohol abuse and dependence diagnoses were more common for family-history-positive than family-history-negative subjects across the three data collection waves (year 1=31.3% versus 15.8%; year 4=23.2% versus 14.0%; year 7=21.0% versus 10.9%) (χ²=18.76, df=1, N=454, p<0.0001), and men were diagnosed with alcohol abuse or dependence significantly more than were women across the three data collection waves (year 1=33.5% versus 14.8%; year 4=30.3% versus 8.1%; year 7=24.8% versus 8.1%) (χ² =48.84, df=1, N=454, p<0.0001). Unlike the findings for alcohol dependence, however, alcohol abuse or dependence diagnoses were more common in the early data collection waves (year 1=23.8%; year 4=18.7%; year 7=16.0%) (χ² =14.25, df=2, N=454, p<0.001).

Cross-Sectional Analysis of Comorbidity

For any given assessment period, the actual percentage of individuals with both an anxiety disorder and an alcohol use disorder was modest (range for anxiety disorders with alcohol dependence=1.5 to 2.2; range for anxiety disorders with alcohol abuse or dependence=2.9 to 7.9). However, these proportions do not take into account disorder-specific base rates and are, therefore, not informative as to the strength of the association between these conditions. Nor are these proportions comparable to comorbidity rates reported in clinical samples where all cases are assumed to have an index diagnosis (e.g., comorbidity rates among alcoholism treatment patients). Table 1 provides several indices that provide information about the magnitude of cross-sectional comorbidity in this sample.

The odds ratios shown in table 1 provide a base-rate controlled index of comorbidity that refers to the ratio of the odds of being diagnosed with either an anxiety disorder or an alcohol use disorder when the other condition is present versus the odds when the other condition is absent. As can be seen, significant nonadjusted odds ratios between anxiety disorders and alcohol dependence were obtained for years 4 and 7, which remained significant even after controlling for family history and sex (adjusted odds ratio). However, this association only approached significance at year 1 (p=0.06); controlling for sex and family history further weakened this association (p values <0.13) at year 1.

Table 1 also shows the significant odds ratios obtained between alcohol abuse or dependence and anxiety disorders at each of the three assessments. These effects remained significant even after controlling for sex and family history (again, shown as adjusted odds ratios in table 1). Taken together, these findings show that having either an anxiety disorder or an alcohol use disorder significantly increases the likelihood (from two- to fivefold) of concurrently having the other diagnosis and, further, this association is not dependent on family history or sex effects.

Prospective Analyses of Comorbidity Through the Use of Logistic Regression

Disorder onset. In one approach to understanding longitudinal relationships, we examined the effect of a diagnosis of either an anxiety disorder alone or an alcohol use disorder alone at an earlier assessment on the odds of being newly diagnosed with the other diagnosis at a later assessment. For example, we evaluated whether the likelihood of a first-time (“new”) onset of an alcohol use disorder at year 4 is affected by the presence of an anxiety disorder diagnosis at year 1. This, again, can be expressed as an odds ratio indicating the ratio of the odds of being newly diagnosed with either an alcohol use disorder or an anxiety disorder at a later assessment in the presence versus the absence of the other diagnosis (alone) at an earlier assessment.

Table 2 shows the raw data used to calculate these odds ratios, along with the results of logistic regression analyses controlling for sex and family history of alcoholism (shown in table 2 as adjusted odds ratio). Note that part of table 2 deals with the prospective association of anxiety disorders and alcohol dependence and that part deals with the prospective association of anxiety disorders and alcohol abuse or dependence. Finally, note that because we are examining prospective associations, analyses use subsamples that include only those participants who had not been diagnosed with the criterion diagnosis (i.e., the diagnosis being predicted) at a previous assessment.

The cross-tabulations in table 2 show that the odds of being newly diagnosed with alcohol dependence at year 4 (see the odds ratios) were more than double among those with versus without an anxiety disorder diagnosis at year 1; however, both the raw and adjusted logistic regression model failed to reach statistical significance (p values <0.17). By contrast, having an anxiety disorder diagnosis alone at either year 1 or 4 did significantly predict a new onset of alcohol dependence at year 7 (see cross-tabulations in table 2). These effects remained significant even after controlling for family history and sex.

Other cross-tabulations in table 2 show that the odds of being newly diagnosed with anxiety disorders at year 4 (see the odds ratios) were well over twice as great among those with versus without an alcohol dependence diagnosis at year 1; again, however, this effect did not reach statistical significance before or after controlling for family history and sex. Other cross-tabulations in this table, by contrast, show that having a diagnosis of alcohol dependence at either year 1 or year 4 did significantly increase the odds of developing a new onset of anxiety disorders in year 7. These effects remained significant even after controlling for family history and sex.

Also in table 2, other cross-tabulations show that having an anxiety disorder diagnosis at year 1 did not significantly increase the odds of developing a new alcohol abuse or dependence diagnosis at either years 4 or 7. Likewise, having an anxiety disorder diagnosis at year 4 did not significantly increase the odds of developing a new alcohol abuse or dependence diagnosis at year 7 (see cross-tabulations); however, note that the odds ratios here were more than two. The lack of statistical significance for these associations was not altered by controlling the effects of family history and sex.

Table 2 shows that having a diagnosis of alcohol abuse or dependence at year 1 did not increase significantly the odds of developing a new anxiety disorder diagnosis at year 4 (see cross-tabulations); however, year 1 alcohol abuse or dependence did increase significantly the odds of a new anxiety disorder diagnosis at year 7 (see cross-tabulations). Controlling for sex and family history, however, left this latter effect only marginally significant (p<0.10). Finally, other cross-tabulations in table 2 show that having alcohol abuse or dependence in year 4 significantly predicted a new onset of anxiety disorders in year 7, whether or not family history and sex were controlled.

Disorder persistence. We also examined whether a comorbid diagnosis prospectively affects the persistence or, alternatively, recovery from a disorder. With an analytic strategy similar to that described earlier, logistic regression was used to contrast the odds that a diagnosis at an earlier assessment would persist to a latter assessment in the presence versus the absence of a comorbid diagnosis at the earlier assessment. For example, does the likelihood that a year 1 diagnosis of alcohol dependence will persist to year 4 increase when a comorbid diagnosis of anxiety disorders is also present in year 1?

Note that whereas the previous analyses used subsamples that included only those participants who where not given the criterion diagnosis (i.e., the diagnosis to be predicted) at the earlier assessment, the current analyses used subsamples of only those participants who had been given the criterion diagnosis at the earlier assessment. The results of these comparisons are shown in table 3.

As seen in table 3, the rates of persistence for alcohol dependence in years 4 and 7 were not affected by the presence or absence of a comorbid anxiety disorder diagnosis at the earlier assessments. However, the odds ratios for years 1 to 4 and years 1 to 7 (see cross-tabulations) were moderate in size, ranging from just under 2 to just under 3. A similar pattern of findings was found for predictions of anxiety disorder persistence, given the presence versus the absence of alcohol dependence (table 3). Specifically, odds ratios are moderate when looking at years 1 to 4 and 1 to 7 (see cross-tabulations), whereas those for years 4 to 7 are trivial. Again, however, none of these effects reached statistical significance.

Persistence effects associated with alcohol abuse or dependence (table 3) also failed to achieve statistical significance. Further, and unlike persistence analyses involving alcohol dependence, these odds ratios were uniformly trivial in size, with one exception. As shown in table 3, there was a modestly elevated risk of persistence for alcohol abuse or dependence diagnosis from years 1 to 4 among those who had a comorbid anxiety disorder diagnosis at year 1.

Prospective Analyses of Comorbidity Through the Use of Path Analysis

In a second, conceptually distinct, approach to understanding longitudinal relationships, we used path analysis to account for potential reciprocal relations over time, as well as to model how both measured third variables (i.e., sex, family history) and unmeasured third variables (e.g., temporally specific situational variables such as living situation, interviewer bias, and definitional overlap) may affect these associations. Specifically, we estimated two path models (one for alcohol dependence, one for alcohol abuse or dependence ) to model autoregressive and cross-lagged reciprocal effects simultaneously, while controlling for both sex and family history of alcoholism as exogenous variables and incorporating covariances between contemporaneous error variances. In this way, path analysis was used to examine whether anxiety disorders at an earlier time could be used to predict later alcohol use disorders or vice versa and whether sex or family history of alcoholism predicted either of these disorders at each of the three measurement occasions.

Tetrachoric correlations were calculated with PRELIS 2 (28) among family history, sex, anxiety disorders, and alcohol use disorders at years 1, 4, and 7 (listwise deletion of subjects with missing data affected less than 1% of the total sample). (Note that we used tetrachoric correlations, rather than product-moment correlations, to adjust for distributional problems associated with skewed, dichotomous data; i.e., the diagnoses.) Two just-identified path models were estimated, one in which the alcohol use disorder was alcohol dependence and one in which the alcohol use disorder was alcohol abuse or dependence. Each model included a) family history of alcoholism and sex as exogenous variables with direct effects to all endogenous variables (i.e., alcohol use disorders and anxiety disorders); b) alcohol use disorders and anxiety disorders treated as separate autoregressive (of lags 1 and 2) and cross-lagged (of lags 1 and 2) processes; and c) covariances between contemporaneous errors. Because each model was just-identified, no fit indices could be computed (i.e., the fit of these models is, by definition, perfect).

Figure 1 illustrates the model when alcohol use disorders are defined as alcohol dependence. Family history of alcoholism was significantly associated with anxiety disorders at years 1, 4, and 7 (0.29, 0.09, and –0.13, respectively) and alcohol dependence at year 1 (0.28). Sex was associated with alcohol dependence at years 4 and 7 (–0.29 and –0.09) and anxiety disorders at year 7 (0.13). In addition, against the background of moderate autoregressive stabilities, prospective effects from anxiety disorders to alcohol dependence and alcohol dependence to anxiety disorders were found. Anxiety disorders at year 1 predicted alcohol dependence at years 4 and 7 (0.14 and 0.15), and alcohol dependence at year 1 prospectively predicted anxiety disorders at years 4 and 7 (0.23 and 0.22). In sum, on the basis of path analysis, anxiety disorders and alcohol dependence reciprocally predict each other.

As mentioned earlier, an identical path model was analyzed with alcohol abuse or dependence in place of alcohol dependence. Consistent with results from the logistic regressions, the prospective relations between anxiety disorders and alcohol abuse or dependence were not as strong as between anxiety disorders and alcohol dependence. Anxiety disorders prospectively predicted alcohol abuse or dependence but not vice versa. In addition, a suppression effect was found between anxiety disorders at year 1 and alcohol abuse or dependence at year 7. Although the bivariate relation between these two variables is positive (r_tet=0.18), the path between them was negative (–0.08). This is, most likely, due to the collinearity between a number of other variables in the model (e.g., family history of alcoholism) and these two variables, leading to a negative path to successfully model the correlations (for a discussion of suppression effects, see reference 29). In sum, anxiety disorders prospectively predict alcohol abuse or dependence but not vice versa.

Finally, note that the correlated error terms of anxiety disorders and alcohol use disorders at each year are an important component of the models. This statistical control adjusts for the effects of third variables, present at any wave, that would (linearly) influence the prevalence of both disorders. The high correlation between the “error” for anxiety disorders and alcohol use disorders at year 1 suggests that variables beyond those indexed by a family history of alcoholism potentially contribute to a shared vulnerability process. We use “potentially” because this correlation could reflect processes other than shared vulnerability, including shared method variance and unidirectional or reciprocal influences that occurred before baseline diagnosis. Correlations between contemporaneous disturbances at later years are useful for assessing the effect of temporally specific third variables (e.g., changes in social networks or living conditions, the effect of legal drinking age). The correlations between error terms at years 4 and 7 are consistent across models generated by using either alcohol dependence or alcohol abuse or dependence.

DISCUSSION

The co-occurrence of alcohol use disorders and anxiety disorders is extremely well documented (8, 12). However, the extent to which one disorder promotes the other or whether the influence of other third variables actually account for some or all of this association remains highly speculative (13, 14). For some time now, epidemiologist and clinical researchers have pointed to the necessity of conducting prospective studies to better understand the causal paths linking comorbid disorders (9, 10). This is the only prospective study of which we are aware that addresses the longitudinal association of alcohol use disorders and anxiety disorders by using modern DSM diagnostic criteria, a large sample size, and a powerful data analytic approach.

The prospective analyses concerning disorder onset (table 2 and Figure 1) provide strong evidence of reciprocal causal influences between anxiety disorders and alcohol dependence. We found that having an anxiety disorder at years 1 or 4 quadrupled the risk for a new onset of alcohol dependence in year 7. Conversely, we found that alcohol dependence at years 1 or 4 increased by three to five times the risk for a new onset of anxiety disorders in year 7. These longitudinal findings may help reconcile alternative views as to which comorbid condition causes the other (8, 12–14).

For example, the view that anxiety disorders promote alcohol disorders appears to derive support from studies showing that alcohol dampens stress responses (30), reduces clinical anxiety (31), and is used by those who are comorbid to manage such feelings (32, 33). On the other hand, the view that alcohol disorders promote anxiety disorders appears to derive support from studies showing that alcohol can increase stress responses (34) and exacerbate clinical anxiety states (35). Somewhat problematic for both views are studies showing that cases tend to be (roughly) evenly split in terms of which condition begins first (7, 36).

However, all of these findings are consistent with the reciprocal causal influence between anxiety disorders and alcohol use disorders found in the present study. For example, as suggested by Kushner and colleagues (8, 11), short-term anxiolytic alcohol effects (possibly mediated by γ-aminobutyricergic mechanisms [30]) may combine with longer-term anxiogenic alcohol effects (34) to produce a vicious cycle of upwardly spiraling alcohol use and anxiety symptoms. This feed-forward, vicious-cycle view of comorbidity development suggests that risk for developing the comorbid condition is increased by means of the same interactive process, regardless of which condition began first.

Alternatively, however, it could be that the process whereby anxiety disorders increase the risk for alcohol use disorders is different than that whereby alcohol use disorders increase the risk for anxiety disorders. This separate-path view of comorbidity suggests that those who start with an anxiety disorder and then develop an alcohol use disorder do so through mechanisms that are distinct from those who start with an alcohol use disorder and then develop an anxiety disorder. Unfortunately, however, our data cannot clarify this issue.

Although data concerning the stress-dampening and anxiolytic properties of alcohol are extensive (30, 31), there are fewer data concerning the mechanisms whereby alcohol use disorders could be anxiogenic. One possibility is that the negative psychosocial sequelae of pathological alcohol use (e.g., loss of relationships, educational or vocational failure, legal consequences) trigger anxiety reactions in vulnerable individuals. Research on life events and anxiety disorders (37, 38) is consistent with this hypothesis, although most of this research is cross-sectional and based on clinical samples. Another possibility, and particularly relevant for considering the long-term effects of alcohol dependence, is that repeated withdrawal-related symptoms associated with the hyperadrenergic state (e.g., nervousness, tremulousness) contribute to a cognitive or physiological diathesis toward experiencing anxiety symptoms (11). This “withdrawal” view is quite consistent with the fact that we obtained our strongest effects for analyses involving alcohol dependence (Table 2 table 2 and Figure 1). (Note that in DSM-III, physical dependence symptoms [i.e., withdrawal or tolerance] are required for diagnosis). With this said, however, reasonably strong (and, in some cases, significant) trends were obtained in analyses by using the alcohol abuse or dependence grouping as well. Unfortunately, limited statistical power reduced our ability to establish the reliability of effects of this magnitude.

As shown in table 3, odds ratios were between 2 and 3 for the alcohol dependence-anxiety disorders analyses for years 1 to 4 and 1 to 7. That is, the odds that an alcohol dependence or an anxiety disorder would persist from year 1 to years 4 or 7 was approximately two to three times greater if the comorbid diagnosis was also present at the earlier assessment. Although these effects did not reach statistical significance, low power resulting from the restricted Ns used in these analyses (table 3) raise the possibility of type II errors. On the other hand, there actually appears to be a modest, nonsignificant trend in the opposite direction for years 4 to 7 (table 3). Taken together, these results do not provide clear support that comorbidity affects persistence; however, they also do not convincingly rule out this possibility.

Notably, because DSM-III criterion C (duration of disturbance for at least 1 month) is not specifically determined in the interview, it is possible that the diagnosis of alcohol abuse was overestimated. However, visual inspection of all diagnostically relevant data for individuals (e.g., individual DIS items, questionnaire reports of alcohol use and problems) suggested that the DIS was reasonably successful in ruling out a diagnosis when symptoms were short-lived. Further, because DSM-III dependence criteria, but not those of DSM-III-R or DSM-IV, require withdrawal or tolerance, those diagnosed with alcohol dependence in this sample, as a group, were probably more severely affected than if the current diagnostic nomenclature had been applied.

Also concerning generalizability, our oversampling of subjects positive for a family history of alcoholism (see Method) could have increased comorbidity rates between anxiety disorders and alcohol use disorders. In fact, family history for alcoholism had a direct effect on the rate of both alcohol use disorder and anxiety disorder diagnoses in this study (Figure 1) as well as in an Epidemiologic Catchment Area site (39).Nonetheless, it does not appear that family history for alcoholism accounted for much of the comorbidity between diagnoses, either cross-sectionally or longitudinally. Specifically, most key effects remained significant after statistically controlling direct effects of family history (Table 1 and Table 2 2 and Figure 1).

Documenting the comparability of our longitudinal findings is made more difficult by the virtual absence of comparable prospective studies. However, cross-sectional data are comparable to those from other community-based studies (40, 41). For example, those studies and this show a two- to threefold increase in risk for either an alcohol disorder or an anxiety disorder given the presence of the comorbid condition (table 1). Also, the cross-sectional rates of anxiety disorders among those with alcohol dependence (year 1=34%; year 4=24%; year 7=30%) and of alcohol dependence among those with anxiety disorders (year 1=11%; year 4=18%; year 7=17%) are reasonably consistent with those reported in clinical samples where subjects all have an index diagnosis (8). In short—and lending confidence to the generalizability of the longitudinal findings—our cross-sectional results appear to be quite representative in terms of comorbidity rates.

Finally, although space constraints preclude a detailed discussion of the value of longitudinal path modeling, we believe that the approach we have employed provides a useful statistical approach for investigating comorbidity processes. That is, these path models provide probing analyses of the direction of the effect of one disorder on another, while controlling for a number of important statistical concerns (e.g., autoregressivity and the effects of both measured and unmeasured third variables). In a recent article, we describe in more detail the value of multivariate approaches to studying comorbidity prospectively (42).

Presented at the annual meeting of the Research Society on Alcoholism, Washington, D.C., June 19–24, 1996. Received Dec. 4, 1996; revision received Oct. 20, 1998; accepted Dec. 7, 1998. From the University of Minnesota, Minneapolis; and the University of Missouri-Columbia. Address reprint requests to Dr. Sher, Psychology Department, 200 South 7th St., Columbia, MO 65211; [email protected] (e-mail). Supported in part by grant AA-07231 from the National Institute on Alcohol Abuse and Alcoholism to Dr. Sher. Added protection of participants’ data was secured through a Certificate of Confidentiality from the Department of Health and Human Services.

TABLE 1

Enlarge table

TABLE 2

Enlarge table

TABLE 3

Enlarge table