Longitudinal Trajectories of Depression and Anxiety in a Prospective Community Study: Title and subTitle BreakThe Zurich Cohort Study FREE

CURRENT KNOWLEDGE regarding the course and stability of depression is based primarily on follow-up of clinical samples. With the shift to identification of depression in primary care, however, there is now increasing information regarding the longitudinal manifestations of depression in systematic primary care samples. Knowledge regarding the course of depression has major implications for understanding the etiology, phenomenology, treatment outcome, and sequelae of depression. For example, the high relapse rate observed in clinical samples^{1 - 8} has led to a shift from the primary treatment of acute episodes of depression to the reduction of recurrence as the major thrust of contemporary management of depression.

There have only been 2 long-term prospective studies with multiple follow-ups of depression or anxiety in community settings . In a 40-year prospective study on the prevalence of depression in a Canadian sample, Murphy^{9 - 10} found that depression had a worse prognosis than anxiety in terms of chronicity and recurrence. Likewise, in an investigation of anxiety states and depression in a general population sample from Sweden during a 25-year period, Hagnell and Grasbeck¹¹ found that pure depression and pure anxiety states remained quite stable over time, whereas mixed depression and anxiety manifested more heterogeneous outcomes.¹¹ The results of several short-term follow-up studies of community samples have also yielded evidence of at least a moderate degree of stability of depression and, to a lesser extent, anxiety.^{12 - 13}

With respect to the development of comorbidity, the results of prospective community studies have shown that the emergence of depression among those with anxiety is far more common than the converse.^{11 ,14 - 15} The International Classification of Diseases, 10th Revision (ICD-10), even incorporates a specific category for mixed anxiety-depression based on concurrent expression of subthreshold-level anxiety and depression.¹⁶

There are several issues that complicate the longitudinal evaluation of the course and stability of depression. The first issue concerns the lack of inclusion of comorbid anxiety and depression in the characterization of the course of each of these disorders alone. A second issue is the classification of subjects who continue to manifest symptoms but fail to meet full criteria for depression at follow-up.^{17 - 18} The third problem is the consideration of methodologic effects such as missing observations, age distribution of the sample, and varying observation periods.

The dearth of long-term follow-up studies of community-based samples and differences in methodology across existing studies highlights the need for additional research designed to examine the stability of depression and anxiety in the community that systematically incorporates comorbid and noncomorbid subtypes and threshold and subthreshold levels of pathologic manifestation. The present study uses data from a prospective study of a cohort of young adults followed up for 15 years. The major goals of this article are (1) to examine the stability of anxiety and depression over time; (2) to investigate the role of comorbidity in the stability of and transitions between anxiety and depression; and (3) to examine whether inclusion of subthreshold manifestations of anxiety and depression increases the stability of depression and anxiety.

The Zurich Cohort Study is composed of a subset of a cohort of 4547 subjects (2201 men and 2346 women), representative of the canton of Zurich in Switzerland, who were screened at the age of 19 years (men) and 20 years (women) in 1978 with the Symptom Checklist 90-R (SCL-90R).¹⁹ A 2-stage sampling of 591 subjects (292 men and 299 women) was used to enhance the probability of cases in a general population sample consisting of two thirds of high scorers (defined by ≥85th percentile on the SCL-90R) and the remainder selected from a random sample of those who had scores below the 85th percentile.^{20 - 21}

All subjects provided written informed consent after receiving a complete description of the study and having an opportunity to ask questions. Table 1 displays the sample size, age, and sex at the 5 interviews of the study.

Ninety percent of the sample participated in 2 or more interviews, and 69% of the original sample remained in the cohort across the 15 years of the study. Those who dropped out did not differ significantly in their baseline measures in terms of demographic characteristics or risk status at study entry. Data are weighted to yield estimates of the population rates with the use of coefficients that reflect the representation of the subjects with respect to the entire population assessed.

The diagnostic instrument used in the Zurich Cohort Study was the Structured Diagnostic Interview for Psychopathologic and Somatic Syndromes (SPIKE), a semistructured instrument that was developed for epidemiologic studies.²² The SPIKE was designed for administration by trained psychiatric residents and clinical psychologists. Information was collected on childhood characteristics, treatment history, psychiatric and somatic syndromes, and use and abuse of various substances.

Screening probes based solely on the major phenomenologic features of each syndrome were administered for each diagnostic category. Positive endorsement of the entry probe was followed first by queries about specific symptoms and second by their duration, frequency, and severity, and treatment history and impairment. The core phenomenologic probe and treatment for all of the syndromes was asked about each of the interim years between interviews to cover the entire assessment period. At the first 2 interviews, the criteria for generalized anxiety disorder and panic were assessed in a single section of the interview. Fewer symptoms were collected in these sections than in later interviews, which included separate diagnostic sections as the DSM criteria for these 2 syndromes evolved. The interrater reliability of the SPIKE was found to be very high, with κ of 0.89 and 0.91 for the symptoms of depression and anxiety (generalized anxiety disorder) and of 0.90 for the corresponding syndromal diagnoses.²³ The validity of the SPIKE has also been assessed by comparing physician ratings and medical records with the SPIKE administered by another clinician among 140 patients drawn from psychiatric clinics or social-psychiatric services in the canton of Zurich^{24 - 26} and from a local hospital.²⁷ The SPIKE rating of the diagnostic level of depression was found to have high sensitivity and specificity (0.95 and 0.59, respectively, for major depression and 0.83 and 0.63, respectively, for minor depression). Likewise, the SPIKE had high sensitivity for detecting subthreshold depression, anxiety, and mania (ie, respective κ of 0.90, 0.83, and 0.67), confirming its ability to capture subclinical medical and psychiatric complaints.

The diagnoses from the 1979 to 1993 interviews were made according to the DSM-III or DSM-III-R criteria for most of the major categories. Because the SPIKE sections did not restrict collection of information based on specific diagnostic thresholds, we were also able to examine subthreshold manifestations of diagnostic categories.²⁸ The definitions of these new syndromes were based on classic psychopathologic criteria: number of symptoms, duration, and frequency of episodes. Evidence of the existence of an anxiety spectrum and depressive spectrum has been previously presented.^{1 ,28 - 36}

For the present analyses, the sample was classified into 4 levels of anxiety and depression disorder: threshold, subthreshold, symptoms present, and no symptoms.^{30 ,36 - 38} The diagnostic level of depression was based on DSM-III-R major depression or dysthymia. Subthreshold depression included "minor depression," which required 3 or 4 symptoms of depression with a minimum duration of 2 weeks, and "recurrent brief depression," which required 5 of 9 criterion symptoms with a monthly recurrence during a year and impairment. "Depressive symptoms" required depressed mood with only 1 or 2 criterion symptoms.

Anxiety disorders included symptoms of either panic or generalized anxiety disorder. Threshold-level panic was defined according to DSM-III-R (ie, 4 or more panic symptoms and 3 attacks in 3 weeks). Subthreshold panic was defined as 4 or more panic symptoms plus at least 2 attacks in 1 year. The symptom level of panic was defined as a panic attack with at least 1 of the concomitant panic symptoms.

The diagnostic level of generalized anxiety was defined according to the DSM-III-R criteria. Subthreshold anxiety included subthreshold generalized anxiety disorder and brief and recurrent generalized anxiety. Subthreshold generalized anxiety required excessive worrying, 1 to 3 symptoms plus duration of 4 or more weeks; subthreshold brief anxiety required excessive worrying plus 1 or more anxiety symptoms and a duration of 2 to 3 weeks; and recurrent brief anxiety required excessive worrying, 3 or more anxiety symptoms, duration less than 2 weeks, and more than 1 episode per month.

Our analysis applied log-linear models³⁹ with the use of SAS PROC CATMOD software (SAS Institute Inc, Cary, NC). The model estimation procedure first requires preparation of observed cell counts, which is used as the response variable in PROC CATMOD. Odds ratios were estimated on the basis of the parameters from the log-linear model. We evaluated the stability and switching of anxiety and depression with odds ratios between and within anxiety and depression diagnoses and assessment points. All analyses in the present report are based on 1-year prevalence rates for the year preceding each wave of assessment. In addition to diagnoses, sex and SCL-90 risk group were included as covariates.

To maximize the number of observations, we fit 4 log-linear models, each of which included data from 2 adjacent waves. Consequently, data in the intermediate years (1981, 1986, and 1988) were used in 2 log-linear models, and the associations were estimated repeatedly. To evaluate the impact of missing data, we also fit a log-linear model simultaneously by using the complete data from all waves, and consequently many observations were removed from this overall model. This led to some minor changes in the model parameter estimates that did not alter the interpretation of the results.

Two methods were used to classify anxiety and depression across waves of assessment, and models were fit to both approaches. First, anxiety and depression were treated as non–mutually exclusive categories, such that both anxiety and depression were defined as 2 dichotomous variables (ie, no anxiety vs anxiety, no depression vs depression). Next, the 2 disorders were considered simultaneously, yielding a 4-level variable.

In addition to modeling comorbidity in the 2 ways described above, we also modeled associations between anxiety and depression according to the different definitional thresholds described above. Comparison of the 3 cutoff points showed few differences in the patterns of transitions and stability of anxiety and depression, although significance levels were higher with successive decreases in the level of the threshold, most likely attributable to increased statistical power. All figures reflect analyses using the lower threshold of symptoms, subthreshold, or threshold disorder vs no symptoms.

To present the magnitude of the transitions between comorbid and noncomorbid anxiety and depression, we also provide the specific probabilities of changes between each of the 3 affected groups across the duration of the study. The transition probabilities between categories across waves of assessment were calculated from the cell frequencies and marginal totals of the 4 × 4 contingency tables between each pair of waves of the study.

The 1-year frequencies of comorbid and noncomorbid anxiety and depression by level of diagnostic threshold are presented in Table 2. The rates of threshold-level comorbid anxiety and depression and anxiety alone tended to increase over time, whereas the rates of depression alone tended to stabilize over time. When stratified by sex, there was an approximate 1½- to 2-fold female preponderance for both combined and pure forms of anxiety and depression. When subthreshold and symptom levels were considered, the patterns varied for comorbid anxiety and depression. Subthreshold levels of anxiety alone tended to be higher at later assessments, whereas subthreshold levels of depression alone tended to be lower at later assessments.

Figure 1 displays the stability, the concurrent comorbidity, and the predictive association across subsequent waves of interviews between symptom level and higher anxiety and depression compared with those with or without symptoms alone. The stability estimates of anxiety across successive waves are generally 2-fold or greater (range, 1.8-3.8); however, the estimates and significance tend to decrease over time. The stability estimates of depression are also generally 2-fold or greater (range, 1.2-2.5) but are highly statistically significant across all but the last 2 assessment periods.

The concurrent comorbidity between anxiety and depression throughout the study was also highly significant and stable across the 15 years of the study (range, 3.0-6.0). The estimates of stability and comorbidity for the intermediate waves included in overlapping models were remarkably similar. Despite significant concurrent comorbidity between anxiety and depression and the stability of the 2 disorders alone, the cross-associations between anxiety (or depression) at earlier waves with depression (or anxiety) in subsequent waves were nonsignificant across nearly all waves of assessment. The effects of alterations in the diagnostic threshold levels were remarkably similar; however, estimates of stability were generally higher and confidence intervals smaller as thresholds were lowered.

Not shown here, we also assessed the impact of missing observations by including all 5 waves of assessment in 1 log-linear model, as recommended by Gibbons and Hedeker.⁴⁰ Despite the reduction in sample size in the 2 models, there was remarkable consistency in both the magnitude and significance of the associations. Thus, estimates reported in Figure 1 are not strongly influenced by missing observations.

Figure 2 presents the stability of anxiety only, depression only, and comorbid depression and anxiety with the use of the lower threshold of symptoms and higher. There was substantial stability of the comorbid syndromes across all waves of assessment (range, 7.1-20.7), suggesting that comorbid anxiety and depression is much more stable than noncomorbid anxiety or depression. Although the stability odds ratios were elevated for noncomorbid anxiety (range, 1.8-4.6), the only significant finding emerged for stability between 1986 and 1988. The lack of statistical significance for the earlier indexes was attributable in part to the small numbers in these categories.

By contrast, the magnitude of noncomorbid depression (range, 2.1-3.6) was quite stable over time, but, with the exception of the last interview, the estimates across all interviews were highly significant. Again, inclusion of higher thresholds yielded somewhat lower estimates and levels of significance but similar patterns of effects across waves of assessment.

Figure 3 depicts the patterns of transitions between comorbid and noncomorbid anxiety and depression with the use of the same diagnostic threshold as those used inFigures 1 and 2. Figure 3A presents the transitions from the pure states to alternative pure states (ie, pure anxiety to pure depression) and to comorbid anxiety and depression. Subjects with anxiety only had a moderate probability of developing depression only (range of odds ratios, 1.5-7.8), whereas those with depression only did not develop anxiety only at any of the interviews over time (odds ratio range, 0.3-1.8). In contrast, there were substantial transitions from both pure anxiety and pure depression to comorbid anxiety and depression across all of the interviews, with odds ratios ranging from 1.5 to 10.0. Figure 3B presents the transitions from the comorbid cases to pure cases of anxiety and depression. This figure shows that there was also fluctuation between comorbid and pure conditions of both anxiety and depression.

The empirical probabilities of transition are presented in Table 3 to demonstrate the magnitude of the findings described above. The rows and columns present the 1-year rates of the mutually exclusive cross-classification of anxiety and depression at adjacent pairs of interview. Both anxiety alone and depression alone were strongly associated with the development of comorbid anxiety and depression at subsequent interviews. Across all interviews, an average of 21% (range, 0.14-0.22) of those with depression alone and 24% (range, 0.17-0.32) of those with anxiety alone developed comorbid anxiety and depression at follow-up. Transitions from anxiety alone to depression alone were also quite common, particularly across earlier interviews (range, 0.05-0.29). Collectively, nearly half of those with anxiety alone developed depression either alone or comorbid with anxiety at subsequent interviews.

By contrast, the transition from depression to anxiety alone was quite rare; an average of only 9% (range, 0.04-0.13) of those with depression alone reported anxiety alone. The transition from the comorbid syndrome to anxiety alone was also uncommon; an average of only 11% (range, 0.07-0.14) developed anxiety alone at subsequent interviews. Transitions to depression occurred more frequently, with 24% (range, 0.18-0.31) of the comorbid cases manifesting depression alone at a subsequent interview.

We also examined the potential sex difference in the longitudinal trajectories of anxiety and depression by fitting separate log-linear models for men and women. Although there were no statistical differences between men and women on estimates of stability or comorbidity, men tended to exhibit somewhat greater concurrent comorbidity whereas women tended to manifest greater predictive comorbidity. That is, women were more likely than men to exhibit transitions from pure anxiety or depression to comorbid conditions at follow-up. Sex differences in patterns of transitions between comorbid and noncomorbid anxiety and depression could not be assessed because of reduced power when stratifying by sex.

Lewis⁴¹ noted that anxiety is an integral part of the depressive reaction. Although the observations of Lewis and others¹⁸ were based on clinical samples, subsequent studies of inpatients, outpatients, and community samples converge in verifying this astute clinical observation. There has been a surfeit of research on comorbidity during the past 2 decades since the term was first applied in psychiatry. The results of large community studies in the United States and elsewhere show that comorbidity is pervasive for most Axis I psychiatric disorders.^{42 - 45} Reviews of potential explanations for comorbidity in adults were presented by Kessler and Price,⁴⁶ Merikangas,⁴⁷ Kendler et al,^{48 - 49} and Maier et al,^{50 - 51} and in children by Caron and Rutter⁵² and Angold et al.⁵³ The bulk of evidence regarding mechanisms for comorbidity has been derived from cross-sectional data or from twin and family studies. There has been far less longitudinal research that investigates the stability of anxiety and depression over time, particularly using specified diagnostic categories. The present study helps to fill this gap in knowledge by investigating longitudinal patterns of expression of comorbid and noncomorbid anxiety and depression during a 15-year period. Moreover, the current study considers additional methodolologic issues concerning subthreshold-level symptomatology and missing data.

The findings of the current study confirm those of previous primary care and community studies of adults and children regarding the stability of depression and the comorbidity of anxiety and depression over time. The increased stability of depression (either comorbid or noncomorbid) over anxiety alone and the pattern of transitions between anxiety alone and depression alone were also observed in the Lundby study¹¹ despite major differences in the diagnostic criteria and assessment methods. These findings are consistent with prospective and retrospective studies suggesting that anxiety is much more likely to predict the subsequent onset of depression than the converse.^{10 - 11 ,54 - 55} The finding that stability estimates of anxiety tend to decrease over time further suggests development-specific expression of anxiety as individuals progress through adulthood. Retrospective and prospective studies of youth drawn from both clinics and the community have also shown stability of anxiety and depression^{55 - 61} ;however, few have examined the specificity of comorbid and noncomorbid subtypes.

Although not a major goal of these analyses, inspection of sex differences in the stability and transitions between anxiety and depression disclosed few differences in the longitudinal patterns of stability between men and women. Although both men and women exhibited comorbidity between anxiety and depression, men exhibited somewhat higher concurrent comorbidity, suggesting greater severity of anxiety or depression in men at the time of the actual episode. By contrast, women with anxiety tended to develop subsequent depression and women with depression tended to develop subsequent anxiety, more so than men with anxiety or depression. Thus, although the course of anxiety or depression alone may not be more chronic in women, the course of these syndromes in general may be more persistent in women than men. These patterns require further evaluation, especially since not all findings were significant and findings from previous studies suggest few sex differences in the course of anxiety and depression.^{62 - 63}

The finding that the associations between sex and risk group (SCL-90) and anxiety were significant at early waves but not at later waves suggests that these factors may be associated with the onset, but not the maintenance, of anxiety. In contrast, stronger associations between these factors and depression at the most recent interview suggest they may be important factors in the course of depression later in adulthood. Additional research is necessary to evaluate these explanations.

Previous work on this community sample has demonstrated that contemporary diagnostic systems fail to cover depressive and anxiety states among those who do not meet duration or impairment criteria yet exhibit recurrence and subjective distress and have a history of treatment.^{28 ,31 - 32 ,64 - 66} This conclusion is substantiated by accumulating evidence of the importance of subthreshold syndromes manifested in clinical, primary care, and community samples^{17 ,28 ,33 - 34 ,50} as well as by the emergence of similar patterns of longitudinal stability and comorbidity in the current study. Wittchen et al¹³ found similar results in a short-term follow-up (19.7 months) of those with threshold-level anxiety and depression at baseline. This suggests that there are fluctuations across threshold and subthreshold levels over time among those who experience anxiety and depression.¹⁷ Therefore, classification of stability by means of only threshold-level diagnostic criteria at each cross-sectional evaluation will fail to capture the majority of cases with persistent anxiety and depression across the life span. This conclusion is particularly important because we found that very few persons experience time-limited symptoms; individuals who meet subthreshold diagnostic criteria at an early age continue to manifest symptoms with impairment throughout adult life. This indicates that mood and anxiety disorders are chronic diseases rather than transient perturbations in response to stressful life events.

The major finding of the present study is the remarkable longitudinal stability of comorbid anxiety and depression. Once comorbidity develops, the probability of recurrence of either disorder alone, and particularly anxiety, is far lower than that of comorbidity. These findings were generally consistent across differential diagnostic cutoff points. When taken together with the results of family and twin studies demonstrating common etiologic factors underlying anxiety and depression with age-specific (or development-specific) manifestations across the life span that are in part attributable to invalid diagnostic thresholds and boundaries for these syndromes,^{48 - 51} the results of this study suggest that the term comorbidity may be a misnomer. That is, anxiety and depression may be manifestations of the same underlying diathesis, perhaps with age-dependent expression, rather than comprising independent disorders. Despite the stability of comorbid anxiety and depression in the present study, there may also be specific anxiety and depression subtypes without a shared diathesis. For example, Sartorius et al⁶⁷ noted that the more severe the anxiety and depression, the better the 2 syndromes could be discriminated in a primary care study of 15 international sites.

Comorbidity has been linked to the chronicity and course of both anxiety and depression. Treatment studies suggest a poor prognosis for comorbid depression,⁶⁸ and community studies suggest that comorbid major depression is more persistent and severe than pure or primary depression.⁶¹ Moreover, primary anxiety is associated with a more persistent course of depression and with self-perceived interference with life and activities, suicide attempts, hospitalization for depression, and overall treatment rates.^{69 - 71} Similarly, comorbidity with depression confers a substantially negative impact on the outcome of anxiety disorders.^{72 - 73} Since these findings suggest that comorbid depression is both more common and more stable than either anxiety or depression alone in the community, far more effort should be devoted to studying the etiologic pathways and treatment strategies for comorbid anxiety-depression. Further research on the stability, familial specificity, treatment response, and biological markers of comorbid anxiety and depression, as well as of specific subtypes of these disorders, is therefore indicated.

The major strengths of this study are the prospective design, the long-term follow-up period of 15 years, and the community-based sample. The evaluation of a single age cohort also eliminates the potential confounding influence of age in the association between the order of onset and stability of anxiety and depression. Additional features including the nonhierarchical approach to diagnosis, the collection of data on symptoms and subthreshold manifestations of anxiety and depression, and the data analytic method enable advancement of our understanding of the association between anxiety and depression over time. To our knowledge, this is the first long-term prospective study of the patterns of comorbidity of anxiety and depression using contemporary diagnostic nomenclature.

Limitations of the study include the following: (1) differences in diagnostic information available across waves, since the clinical interview was expanded over time to capture the evolution of more extensive diagnostic systems; (2) the 15-year attrition rate of 30%; and (3) the relatively small sample size of this cohort. Finally, these analyses are based on aggregate anxiety and depressive disorders and may not apply to specific diagnostic subtypes of these syndromes (eg, panic, social phobia, dysthymia).

Findings of stability and comorbidity of anxiety and depression have important implications for treatment. In their classic articles, Feinstein⁷⁴ and Kaplan and Feinstein⁷⁵ argued that comorbidity must be integrated into the design of clinical trials, since its neglect can seriously bias conclusions regarding treatment efficacy. Contemporary clinical trials often exclude comorbid cases despite the overwhelming evidence shown here and elsewhere that comorbidity is more common than depression or anxiety alone.

There are also major implications of these findings for the diagnostic nomenclature and etiologic research. Advances in the tools that tap central nervous system function and structure have far surpassed our ability to define the clinical entities for investigation. Longitudinal studies such as the one described herein are critical to validate current diagnostic concepts that will inform future research on the pathogenesis of affective disorders. Developmental neurobiological approaches should examine the primarily unidirectional association between anxiety disorders and depression as well as the factors underlying the age-dependent expression of anxiety and depression. Likewise, the greater stability of depression alone and comorbid anxiety and depression suggests further investigation of the etiologic factors underlying comorbid states. Although both labor- and time-intensive, studies such as the Zurich Cohort Study are invaluable in informing diagnostic classification, clinical course, and etiologic pathways in psychiatry.

Corresponding author and reprints: Kathleen Ries Merikangas, PhD, National Institutes of Health, National Institute of Mental Health, Mood and Anxiety Disorders Program, 15K North Dr, MSC #2670, Bethesda, MD 20892-2670(e-mail: Kathleen.Merikangas@nih.gov).

Accepted for publication September 9, 2002.

This study was supported by grant 32-33980 from the Swiss National Foundation, Berne (Dr Angst); and grants AA12044 and DA12468 from the National Institutes of Health, Bethesda, Md (Dr Zhang).