Introduction
It is well documented that co-morbidity in major depressive disorder (MDD) is more a rule than an exception. Epidemiological (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994, Reference Kessler, Chiu, Demler and Walters2005; Demyttenaere et al. Reference Demyttenaere, Bruffaerts, Posada-Villa, Gasquet, Kovess, Lepine, Angermeyer, Bernert, de Girolamo, Morosini, Polidori, Kikkawa, Kawakami, Ono, Takeshima, Uda, Karam, Fayyad, Karam, Mneimneh, Medina-Mora, Borges, Lara, de Graaf, Ormel, Gureje, Shen, Huang, Zhang, Alonso, Haro, Vilagut, Bromet, Gluzman, Webb, Kessler, Merikangas, Anthony, Von Korff, Wang, Brugha, Aguilar-Gaxiola, Lee, Heeringa, Pennell, Zaslavsky, Ustun and Chatterji2004; Grant et al. Reference Grant, Hasin, Stinson, Dawson, Chou, Ruan and Huang2005; Hasin et al. Reference Hasin, Stinson, Ogburn and Grant2007) and clinical studies (Zimmerman et al. Reference Zimmerman, McDermut and Mattia2000; Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002) suggest that both current and lifetime anxiety (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994, Reference Kessler, Chiu, Demler and Walters2005; Zimmerman et al. Reference Zimmerman, McDermut and Mattia2000; Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002; Demyttenaere et al. Reference Demyttenaere, Bruffaerts, Posada-Villa, Gasquet, Kovess, Lepine, Angermeyer, Bernert, de Girolamo, Morosini, Polidori, Kikkawa, Kawakami, Ono, Takeshima, Uda, Karam, Fayyad, Karam, Mneimneh, Medina-Mora, Borges, Lara, de Graaf, Ormel, Gureje, Shen, Huang, Zhang, Alonso, Haro, Vilagut, Bromet, Gluzman, Webb, Kessler, Merikangas, Anthony, Von Korff, Wang, Brugha, Aguilar-Gaxiola, Lee, Heeringa, Pennell, Zaslavsky, Ustun and Chatterji2004; Grant et al. Reference Grant, Hasin, Stinson, Dawson, Chou, Ruan and Huang2005) and alcohol use disorders (AUDs) (Kessler et al. Reference Kessler, McGonagle, Zhao, Nelson, Hughes, Eshleman, Wittchen and Kendler1994, Reference Kessler, Chiu, Demler and Walters2005; McDermut et al. Reference McDermut, Mattia and Zimmerman2001; Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002; Demyttenaere et al. Reference Demyttenaere, Bruffaerts, Posada-Villa, Gasquet, Kovess, Lepine, Angermeyer, Bernert, de Girolamo, Morosini, Polidori, Kikkawa, Kawakami, Ono, Takeshima, Uda, Karam, Fayyad, Karam, Mneimneh, Medina-Mora, Borges, Lara, de Graaf, Ormel, Gureje, Shen, Huang, Zhang, Alonso, Haro, Vilagut, Bromet, Gluzman, Webb, Kessler, Merikangas, Anthony, Von Korff, Wang, Brugha, Aguilar-Gaxiola, Lee, Heeringa, Pennell, Zaslavsky, Ustun and Chatterji2004; Grant et al. Reference Grant, Hasin, Stinson, Dawson, Chou, Ruan and Huang2005; Hasin et al. Reference Hasin, Stinson, Ogburn and Grant2007) are significantly more frequent in individuals with MDD than in the general population. The diagnostic classifications in psychiatry, including the DSM-IV (APA, 2000) and the forthcoming DSM-5, are based on the premise that mental disorders can be discriminated from each other. The descriptive validation of these disorders is crucial for developing the classifications of mental disorders and for elucidating the etiological mechanisms. Because of a high degree of co-morbidity (i.e. covariation of disorders), the descriptive validity of many current diagnostic concepts has been questioned (Wittchen, Reference Wittchen1996; Maj, Reference Maj2005), and more prospective data on longitudinal covariation, that is the degree to which disorders follow independent courses longitudinally, are needed (Wittchen et al. Reference Wittchen, Beesdo, Bittner and Goodwin2003; Rutter et al. Reference Rutter, Kim-Cohen and Maughan2006). In particular, there is a dearth of prospective studies investigating longitudinal covariation among psychiatric patients in clinical psychiatric settings.
The strongest cross-sectional (Krueger, Reference Krueger1999; Krueger & Markon, Reference Krueger and Markon2006; Beesdo-Baum et al. Reference Beesdo-Baum, Höfler, Gloster, Klotsche, Lieb, Beauducel, Bühner, Kessler and Wittchen2009; Røysamb et al. Reference Røysamb, Kendler, Tambs, Orstavik, Neale, Aggen, Torgersen and Reichborn-Kjennerud2011) and longitudinal association (Krueger et al. Reference Krueger, Caspi, Moffitt and Silva1998; Vollebergh et al. Reference Vollebergh, Iedema, Bijl, de Graaf, Smit and Ormel2001; Merikangas et al. Reference Merikangas, Zhang, Avenevoli, Acharyya, Neuenschwander and Angst2003; Fergusson et al. Reference Fergusson, Horwood and Boden2006; Moffitt et al. Reference Moffitt, Caspi, Harrington, Milne, Melchior, Goldberg and Poulton2007; Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2008, Reference Olino, Klein, Lewinsohn, Rohde and Seeley2010; Hale et al. Reference Hale, Raaijmakers, Muris, van Hoof and Meeus2009) has been consistently documented between MDD and other internalizing disorders. Although the division of internalizing versus externalizing disorders results largely from genetic rather than environmental risk factors (Kendler et al. Reference Kendler, Prescott, Myers and Neale2003, Reference Kendler, Aggen, Knudsen, Røysamb, Neale and Reichborn-Kjennerud2011), both genetic and environmental risk factors have an impact on the etiological pathways to symptoms of co-morbid depression and anxiety (Middeldorp et al. Reference Middeldorp, Cath, Van Dyck and Boomsma2005; Kendler & Gardner, Reference Kendler and Gardner2011). Longitudinal studies on adolescents and young adults indicate that anxiety and depressive disorders share genetic and environmental etiological vulnerability to internalizing traits, but onset of symptoms may still be due to unique environmental, disorder-specific factors (Fergusson et al. Reference Fergusson, Horwood and Boden2006; Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2008), and symptoms of depression and anxiety may develop independently as distinct entities (Hale et al. Reference Hale, Raaijmakers, Muris, van Hoof and Meeus2009). A study using individual-centered statistical methods found subgroups of adolescents with distinct longitudinal trajectories of depressive and anxiety disorders associating with different risk factors (Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2010). However, studies investigating longitudinal intra-individual stability and covariation of depressive and anxiety symptoms among adolescents or young adults (Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2008, Reference Olino, Klein, Lewinsohn, Rohde and Seeley2010; Hale et al. Reference Hale, Raaijmakers, Muris, van Hoof and Meeus2009) focus on trajectories of early phases of disorders during a major developmental period. By contrast, among adults, the onsets of internalizing disorders have mostly already occurred, and covariation of the proper disorders can be investigated. A significant proportion of MDD has onset in adulthood or in the elderly, but whether these patients have similar patterns of covariation with anxiety disorders (Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2008, Reference Olino, Klein, Lewinsohn, Rohde and Seeley2010; Beesdo-Baum et al. Reference Beesdo-Baum, Höfler, Gloster, Klotsche, Lieb, Beauducel, Bühner, Kessler and Wittchen2009) remains unknown. In addition, co-occurrence of depression and anxiety may vary between individuals and be related to age at onset. Early onset illness may be more genetically predisposed, which might, because of the high genetic correlation between the disorders, result in differences in patterns of covariation depending on age at onset.
Large epidemiological studies in adult populations (Vollebergh et al. Reference Vollebergh, Iedema, Bijl, de Graaf, Smit and Ormel2001; Slade & Watson, Reference Slade and Watson2006) and a large clinical cross-sectional study in an out-patient population (Kotov et al. Reference Kotov, Ruggero, Krueger, Watson, Yuan and Zimmerman2011) have suggested a broad internalizing factor, including both MDD and anxiety disorders. We are, however, unaware of any longitudinal clinical studies in adult psychiatric patients with MDD investigating stability and change in the course of co-occurring Axis I disorders. Clinical studies vary in the degree to which their populations represent variation in the general population and always involve some selection bias, but nevertheless provide an important complementary view to general population findings. More importantly, such clinical studies are necessary to confirm the clinical validity of findings made in general population studies, in which it usually remains uncertain. Thus, whether the relationship across time between MDD and its co-morbid disorders is different among adult psychiatric patients remains unknown, and the possibility of some co-morbid anxiety disorders as totally independent disorders cannot be excluded. Moreover, because it is possible that distinct subgroups of individuals may have different trajectories between depressive and anxiety disorders over time (Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2010; Kendler & Gardner Reference Kendler and Gardner2011), it is important to examine these at the level of the individual. In statistical analysis, this translates to the application of random effects models, such as latent curve models (LCMs; Bollen & Curran, Reference Bollen and Curran2006). In an LCM, the trajectories of MDD and co-morbid disorders can be estimated simultaneously.
Although MDD and AUD have been found to belong to different latent dimensions (internalizing versus externalizing) of mental disorders in epidemiological and clinical studies (Krueger, Reference Krueger1999; Vollebergh et al. Reference Vollebergh, Iedema, Bijl, de Graaf, Smit and Ormel2001; Krueger & Markon, Reference Krueger and Markon2006; Slade & Watson, Reference Slade and Watson2006; Beesdo-Baum et al. Reference Beesdo-Baum, Höfler, Gloster, Klotsche, Lieb, Beauducel, Bühner, Kessler and Wittchen2009; Kotov et al. Reference Kotov, Ruggero, Krueger, Watson, Yuan and Zimmerman2011; Røysamb et al. Reference Røysamb, Kendler, Tambs, Orstavik, Neale, Aggen, Torgersen and Reichborn-Kjennerud2011), these disorders are often co-morbid in both epidemiological populations (Grant et al. Reference Grant, Hasin, Stinson, Dawson, Chou, Ruan and Huang2005; Hasin et al. Reference Hasin, Stinson, Ogburn and Grant2007; Kushner et al. Reference Kushner, Wall, Krueger, Sher, Maurer, Thuras and Lee2012) and psychiatric settings (McDermut et al. Reference McDermut, Mattia and Zimmerman2001; Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002). According to the meta-analysis of epidemiological studies by Boden & Fergusson (Reference Boden and Fergusson2011), co-morbidity cannot be fully accounted for by common factors, and the disorders seem to be linked in a causal manner. Although some prospective studies in an adult general population and population-based twin sample (Wang & Patten, Reference Wang and Patten2001; Kuo et al. Reference Kuo, Gardner, Kendler and Prescott2006) have provided support for a possible direct causal effect, such as self-medication of depressive symptoms, most longitudinal studies (Wang & Patten, Reference Wang and Patten2002; Fergusson et al. Reference Fergusson, Boden and Horwood2009, Reference Fergusson, Boden and Horwood2011; Boden & Fergusson, Reference Boden and Fergusson2011) have suggested that alcohol use increases the risk of MDD, rather than vice versa. However, to our knowledge, no previous longitudinal study of MDD patients has investigated the covariation of AUDs and depression. Thus, the generalizability of general population findings to clinical populations needs to be confirmed.
In a previous study of psychiatric co-morbidity of DSM-IV bipolar I and II disorders, we reported that no co-morbid disorder seems to follow a course fully independent of bipolar disorder, but concurrent disorders differ in how strongly and with what type of mood episodes they covary (Mantere et al. Reference Mantere, Isometsä, Ketokivi, Kiviruusu, Suominen, Valtonen, Arvilommi and Leppämäki2010). In the present study we examined the longitudinal association of the course of DSM-IV MDD with both categorical and dimensional co-morbidities in a clinical sample of depressive adults in psychiatric care followed prospectively for 18 months. Baseline cross-sectional associations of MDD with Axis I and II co-morbid disorders and longitudinal stability of Axis II disorders and MDD have been reported previously (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002, Reference Melartin, Haukka, Rytsälä, Jylhä and Isometsä2010). In the current study we tested the degree to which longitudinal observations fit two alternative hypotheses: (1) co-morbid Axis I disorders follow a course independent of major depressive episode (MDE), that is there is no consistent cross-sectional or longitudinal association between depression and co-morbid Axis I disorders; and (2) the course of MDE determines, or is determined by, the state of concurrent co-morbid psychiatric disorders in MDD, that is the spectrum of concurrent psychiatric disorders covaries consistently depending on the MDE course. To investigate the hypotheses empirically at the level of the individual, we used the LCM approach (Bollen & Curran, Reference Bollen and Curran2006). Based on earlier literature and our own studies (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002, Reference Melartin, Haukka, Rytsälä, Jylhä and Isometsä2010), we had an a priori expectation that anxiety disorders would covary more strongly with MDE than would AUDs.
Method
The methodology of the Vantaa Depression Study (VDS) at baseline (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002) and at follow-up (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2004) has been described in detail elsewhere. In brief, the VDS is a collaborative research project between the Department of Mental Health and Substance Abuse Services of the National Institute for Health and Welfare, Helsinki, Finland (previously the Department of Mental Health and Alcohol Research at the National Public Health Institute) and the Department of Psychiatry, Peijas Hospital, Helsinki University Central Hospital (HUCH), Vantaa, Finland. Peijas Hospital provides secondary-care psychiatric services to all residents of Vantaa (169 000 inhabitants in 1997). The ethics committee of HUCH approved the study protocol.
In the first phase of the study, 806 psychiatric subjects were screened for the presence of depressive symptoms during an 18-month period starting from 1 February 1997. Of the 703 eligible subjects, 542 (77%) agreed to participate and gave their written informed consent (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2004). In the second phase, a researcher using the World Health Organization (WHO) Schedules for Clinical Assessment in Neuropsychiatry (SCAN) 2.0 (Wing et al. Reference Wing, Babor, Brugha, Burke, Cooper, Giel, Jablenski, Regier and Sartorius1990) interviewed the consenting patients, 269 of whom were subsequently diagnosed as having DSM-IV MDD and included in the study. The patients who were currently abusing alcohol were interviewed after 2–3 weeks of abstinence to exclude those with substance-induced mood disorder. All baseline interviewers received the relevant training by a WHO-certified training center; this was supervised by the last author (E.I.). Diagnostic reliability was investigated using 20 videotaped diagnostic interviews; the κ coefficient for MDD was 0.86 (confidence interval 0.58–1.0), with a 95% observed agreement rate (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002).
MDD course, Axis I diagnoses and symptomatology were evaluated at three time points: at baseline (T1), 6 months (T2) and 18 months (T3). All medical and psychiatric records were available. A psychiatrist or psychologist researcher assigned full DSM-IV Axis I diagnoses using SCAN 2.0 (Wing et al. Reference Wing, Babor, Brugha, Burke, Cooper, Giel, Jablenski, Regier and Sartorius1990). At baseline, concurrent co-morbid psychiatric diagnoses were evaluated during an acute phase of MDD. In the follow-up, co-morbidity was assessed simultaneously with severity and/or state of MDD, which was followed by using a DSM-IV-based life chart (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2004). The outcome of MDE was evaluated as defined in the DSM-IV. Longitudinal course information was obtained by gathering information from medical and psychiatric records and interviewing patients about changes in psychopathological states by using calendars and personal important life events as probes. The data were then integrated into a graphic life chart based on DSM-IV criteria. In addition, self-report and observer scales were used. Severity of depressive state was measured during interviews using the Hamilton Depression Rating Scale (HAMD; Hamilton Reference Hamilton1960). The patients also completed the 21-item Beck Depression Inventory (BDI; Beck et al. Reference Beck, Ward, Mendelson, Mock and Erbaugh1961) and the Beck Anxiety Inventory (BAI; Beck et al. Reference Beck, Epstein, Brown and Steer1988). Our life-chart method was similar to the Longitudinal Interval Follow-up Evaluation (LIFE) methodology used in the National Institute of Mental Health (NIMH) Collaborative Depression Study (CDS), developed by Keller et al. (Reference Keller, Lavori, Friedman, Nielsen, Endicott, McDonald-Scott and Andreasen1987).
Sociodemographic and clinical characteristics, co-morbidity and treatment (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002, Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2005) of the total cohort (n = 269) at baseline, and outcome at the 18-month follow-up (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2004), have been reported previously. Of the total of 269 patients with current MDD initially included in the cohort (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002), 229 participated in the 6-month and 207 in the 18-month follow-up. Patients whose diagnosis switched to bipolar disorder between the follow-ups (n = 9) were excluded. The 193 living patients for whom complete information on the SCAN was available at all three evaluation points (baseline, 6 months and 18 months) were included in this study. Of the baseline cohort, the majority were female (72%), out-patients (83%), currently employed (60%), married/cohabiting (54%) and had co-morbid disorders (79%). Patients dropping out were younger (median 32.9 v. 42.9 years, Z = −3.61, p < 0.001), had higher scores on the BAI (25.3 v. 21.0, Z = −2.24, p = 0.025) and more often lived alone (46/76, 60.5% vs. 88/193, 45.6%, χ 2 = 4.283, df = 1, p = 0.038) than participants. At intake, the median duration of MDE was 3.5 months and the number of episodes two. At baseline, the cohort included more out-patients than in-patients (n = 163, 84%). The median times to the 6- and 18-month follow-up interviews were 6.5 and 18.8 months respectively.
Statistical analysis
Descriptive statistics were computed to describe the general characteristics of the cohort at baseline (Table 1) along with the normality of measures used in the follow-up. Pearson's χ 2 tests, the Student t test, the Mann–Whitney U test and the Kruskal–Wallis test were used as appropriate. For descriptive purposes, we present all p values significant at the < 0.05 level, irrespective of the high number of statistical tests and the risk for type I errors.
Table 1. Characteristics at intake of patients (n = 193) in the Vantaa Depression Study (VDS) with three evaluations of co-morbidity
MDE, Major depressive episode; AUD, alcohol use disorder; BDI, Beck Depression Inventory; HAMD, Hamilton Depression Rating Scale; BAI, Beck Anxiety Inventory; s.d., standard deviation.
a At baseline, all patients in a current MDE were included.
We investigated the hypotheses in two steps. First, we evaluated the extent of change in depression, anxiety and alcohol use. This was initially carried out based on a sequence analysis of prevalences of distinct disorders. Then, to allow for individual differences in change between the three evaluation points, we modeled individual-level change using a piecewise linear approach (Bollen & Curran, Reference Bollen and Curran2006). This approach allows for the possibility of all different change trajectories: improvement in both time windows, improvement in the first time window but regression in the second, no change in either or both of the time windows, and so on. Figure 1 shows that improvements in the first time window are not necessarily linked to improvements in the second time window.
Fig. 1. Heterogeneity of individual-level change trajectories: 20 illustrative sample cases of both improvement–improvement (dark gray lines) and improvement–recurrence (light gray lines) trajectories. BDI, Beck Depression Inventory.
Second, we extended the LCM approach (Bollen & Curran Reference Bollen and Curran2006) to incorporate two variables at a time. Co-morbidity was examined by analyzing the statistical associations between the individual change trajectories in the two time windows. Based on the previous step, we used categorical diagnosis to describe a disorder with no change in status during the follow-up, and change curves to describe changing trajectories. When changes in one variable in a given time window correlate with changes in the other variable in the same time window, the two disorders are interpreted as being co-morbid. A positive association between BDI and BAI, for instance, means that declines (increases) in depression symptoms are linked to declines (increases) in anxiety symptoms. In the change models, it is also important to model longitudinal autoregressive effects of individual variables. For instance, the regression-to-the-mean effect would predict that the higher the baseline BAI, the more likely a greater decline in the 0–6-month time window. Figure 1 illustrates the heterogeneity in individual change trajectories. This heterogeneity calls for the modeling of individual-level change through the application of random effects modeling.
For descriptive statistics, PASW version 18.0 (SPSS Inc., USA, 2009), was used. LCM analyses were performed using Mplus 5.21, with robust maximum likelihood (MLR) estimation for non-normal dimensional variables and the robust weighted least squares mean- and variance-adjusted (WLSMV) estimator for categorical variables (Muthén & Muthén, Reference Muthén and Muthén2010).
Results
Characteristics of the cohort
At intake, all patients had a current MDE. In the two follow-up evaluations, about one-third of the patients had some concurrent Axis I disorder, about one-fifth had an anxiety disorder and one-tenth an AUD (Table 1). The stability of the disorders was evaluated by describing specific sequences of having a particular categorical diagnosis (Table 2). One-fifth (17.6%, 34/193) of patients had a co-morbid Axis I disorder at all evaluation points whereas a quarter (25.9%, 50/193) had no co-morbid disorder at any evaluation point. Most patients (65.3%) had an anxiety disorder at least at one evaluation point but few (10%) exhibited the disorder at all evaluation points; thus, change in status seemed to be the rule for anxiety disorders. By contrast, nearly three out of four (72.5%) patients had no AUD at any evaluation point and, of the remaining patients, nearly a quarter (13/53, 24.5%) exhibited the disorder at all evaluation points. The average decrease in BDI score (Table 1) was 14.7 points. In the 6–18 months time window, about one-third of patients showed an improvement (the average decline in BDI score being 1.5 points), but one-fifth experienced a significant increase in depressive symptoms, the majority of whom had shown an improvement in the earlier time window. The same individual variability was observed in anxiety symptoms. In summary, heterogeneous changes over time in depression and anxiety were observed, but there was relative stability in AUDs.
Table 2. Sequences of categorical co-morbid Axis I diagnoses in the Vantaa Depression Study (VDS) (n = 193)
MDE, Major depressive episode; T1, baseline; T2; 6 months; T3, 18 months; +, the disorder was diagnosed; −, the disorder was not diagnosed.
Modeling co-morbidity
The piecewise linear LCM is depicted in Fig. 2 and consisted of baseline measures (BDI0 and BAI0) and individual-level changes in these measures (ΔBDI and ΔBAI) in the 0–6 and 6–18 months time windows.
Fig. 2. Results of the change models. The Beck Anxiety Inventory (BAI)/Beck Depression Inventory (BDI) model is a bivariate piecewise linear change model where absolute levels at baseline and changes in the two time windows are modeled. The BDI-ALCO model is a bivariate model where BDI is modeled as a piecewise linear change model and alcohol abuse as a categorical variable. * p < 0.05, ** p < 0.01, *** p < 0.001. Dashed lines are paths included in the model that were not significant at the 0.05 level.
Depression and anxiety correlated cross-sectionally at baseline, and higher baseline anxiety predicted less improvement in depressive symptoms in the 0–6 months time window. Longitudinally, ΔBDI and ΔBAI correlated in both the 0–6 and 6–18 months time windows (Fig. 2).
Because changes in AUD over time were rare in the sample, we modeled it as a binary variable (Fig. 2). A positive correlation between AUD at 6 months and both BDI0 and ΔBDI0–6 was found. Higher baseline depression increased the likelihood of AUD at 6 months, and AUD at 6 months was linked to less improvement in depressive symptoms in both the 0–6 months and the 6–18 months time window. All autoregressive effects were found to be significant. The correlations between BAI and BDI remained highly significant after controlling for these regression-to-the-mean effects. In both change models, age and sex were entered as control variables at all time points. For clarity, however, these are omitted from Fig. 2.
We also estimated two models with alternative operationalizations for cross-validation purposes. Specifically, we estimated an alternative depression model by substituting the HAMD for the BDI, and another anxiety model by substituting a diagnostic categorical anxiety measure for the BAI. These alternative models yielded essentially the same results as the original ones.
Finally, because of a risk of attrition influencing our findings, we considered the roles of lower age and higher BAI scores among patients not included (see Method). However, age did not correlate with the individual change trajectory components. Baseline BAI correlated with some of the change components, so we split the sample at the median BAI baseline score, and ran the model for both high and low baseline BAI groups separately. The results were identical with one exception: the path from ΔBDI0–6 to ΔBAI6–18 became significant and positive in the low baseline BAI group. The improvement in BDI in the 0–6-month time window linked to improvement in BAI in the 6–18 months time window in this subgroup. In the high baseline BAI group and in the total cohort, this path was not significant.
Discussion
In adult secondary-care psychiatric patients with co-morbid MDD, we investigated the degree to which the longitudinal course of current co-morbid Axis I disorders followed the course of MDE. We found that the longitudinal course of co-morbid anxiety symptoms was associated moderately strongly with depressive symptoms, and the longitudinal course of co-morbid AUDs moderately weakly. In addition, there was a significant individual variability in courses of anxiety symptoms and disorders, but not categorical AUDs. Patients with higher depressive symptom scores at an acute phase of depression were more likely to have an AUD in follow-up. Thus, the courses of both internalizing and externalizing disorders in adult patients with MDD seemed to be dependent, albeit to differing degrees, on the course of depressive symptoms.
Among mainly psychiatric out-patients with MDD, both the cross-sectional and longitudinal associations of symptoms of depression and anxiety were moderately strong. A change in the co-morbid internalizing status was common, and the change trajectories of depressive and anxiety symptoms exhibited clear intra-individual covariation independent of the direction of change. Thus, at the level of individual patients, symptoms of depression and anxiety seemed to be waxing and waning together. Our results confirmed that anxiety and depressive symptoms also show longitudinal covariation in a clinical psychiatric setting after illness onset, in accordance with most longitudinal studies on adolescent and young adult populations, suggesting a developmental association between anxiety and depression (Fergusson et al. Reference Fergusson, Horwood and Boden2006; Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2008). These studies have speculated that anxiety and depressive disorders share genetic and environmental etiological vulnerability factors for the internalizing trait, but also have environmental, disorder-specific factors affecting the current state (Fergusson et al. Reference Fergusson, Horwood and Boden2006; Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2008). Our results are not fully concordant with those of a population study investigating longitudinal trajectories of depressive and anxiety disorders from adolescence to young adulthood and reporting the probabilities of disorders over time to be largely disparate (Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2010). Thus, in adult psychiatric patients with MDD, longitudinal covariation of depression and anxiety symptoms seems to be a consistent but individually varying finding. Overall, most of the available evidence suggests that depression and anxiety may, to some degree, be interpreted as manifestations of the same illness propensity and, to some degree, as distinct entities.
The longitudinal course of depressive symptoms was associated moderately weakly with AUDs over time. However, the patients with higher depressive symptom scores at an acute phase of MDD were more likely to have an AUD in follow-up. These findings are in accordance with previous community and birth-cohort studies suggesting associations of MDD with AUD (Fergusson et al. Reference Fergusson, Boden and Horwood2009; Boden & Fergusson, Reference Boden and Fergusson2011; Fergusson et al. Reference Fergusson, Boden and Horwood2011) and heavy use of alcohol (Wang & Patten, Reference Wang and Patten2002). Although there was significant individual variability in anxiety disorders and symptoms over time, categorical AUDs were fairly stable. However, the majority of the patients (73%) had no AUD, and only 7% of patients had the disorder at all evaluation points. Overall, the covariation between MDE and AUD was weaker than between MDE and anxiety disorders, supporting the division of the internalizing and externalizing dimensions of mental disorders, in agreement with the data of epidemiological and clinical studies (Krueger, Reference Krueger1999; Vollebergh et al. Reference Vollebergh, Iedema, Bijl, de Graaf, Smit and Ormel2001; Krueger & Markon, Reference Krueger and Markon2006; Slade & Watson, Reference Slade and Watson2006; Beesdo-Baum et al. Reference Beesdo-Baum, Höfler, Gloster, Klotsche, Lieb, Beauducel, Bühner, Kessler and Wittchen2009; Kotov et al. Reference Kotov, Ruggero, Krueger, Watson, Yuan and Zimmerman2011; Røysamb et al. Reference Røysamb, Kendler, Tambs, Orstavik, Neale, Aggen, Torgersen and Reichborn-Kjennerud2011). Our findings of longitudinal covariation between MDD and AUD should, however, be interpreted with some caution. We assessed DSM-IV diagnoses of alcohol abuse and dependence in a semi-structured face-to-face interview and did not investigate the consumption of alcohol. Previous studies have pointed out that the dimensional approach should be linked to a categorical definition to provide more information (Hasin et al. Reference Hasin, Liu, Alderson and Grant2006; Helzer et al. Reference Helzer, van den Brink and Guth2006). Overall, our longitudinal data at the individual level indicate that alcohol use might be associated with an increase in the risk of MDD (Wang & Patten, Reference Wang and Patten2002; Fergusson et al. Reference Fergusson, Boden and Horwood2009, Reference Fergusson, Boden and Horwood2011; Boden & Fergusson, Reference Boden and Fergusson2011), and also suggest that co-morbid internalizing disorders, more consistently than externalizing disorders, may follow the longitudinal course of MDE. However, these findings need to be confirmed by future studies investigating alcohol consumption over time.
Methodological strengths and limitations
Our study has several strengths. To our knowledge, no previous prospective, longitudinal clinical study has investigated the covariation of MDD and Axis I current co-morbid disorders at three time points among adult psychiatric depressive patients. Our subject pool comprises a cohort of mainly psychiatric out-patients with MDD (n = 163, 84%) receiving free-of-charge secondary care in community mental health centers of a large Finnish city; two-thirds of all depressed subjects in the city of Vantaa are estimated to receive treatment in these community mental health centers (Rytsälä et al. Reference Rytsälä, Melartin, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2001). All patients were assessed for Axis I diagnoses by psychiatrists or psychologists using a semi-structured interview with excellent reliability for MDD diagnosis (κ = 0.86) (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002) and also self-reported depression and anxiety scales at three different time points. Unlike previous studies, in our analyses we used both categorical DSM-IV diagnoses and self-report symptom scores. Moreover, we applied the sophisticated statistical LCM method, which can reveal a correlational association between variables by analyzing concurrent change trajectories of individuals over time (Bollen & Curran, Reference Bollen and Curran2006); this method is being used increasingly in psychiatric follow-up studies describing intra-individual stability and change in co-morbidity and traits in adolescents and young adults (Olino et al. Reference Olino, Klein, Lewinsohn, Rohde and Seeley2008, Reference Olino, Klein, Lewinsohn, Rohde and Seeley2010; Hale et al. Reference Hale, Raaijmakers, Muris, van Hoof and Meeus2009). We are unaware of any previous longitudinal clinical studies in adult depressive patients that have investigated stability and change in courses of MDE and Axis I disorders using a similar methodology. Using the LCM, we modeled individual-level changes in depression, anxiety and alcoholism through a piecewise linear approach, and also the individual concurrent and longitudinal association between symptoms of two concurrent disorders. When applying latent factor analysis, correlations between variables do not bias the findings, but the strength and direction of these correlations and the extent to which changes in one variable prospectively predict changes in another can be estimated. We suggest that our findings can be generalized to other psychiatric settings, given the similarity of our baseline depression symptom ratings and patterns of co-morbidity to those of other studies (Zimmerman et al. Reference Zimmerman, McDermut and Mattia2000; McDermut et al. Reference McDermut, Mattia and Zimmerman2001; Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2002).
Our results should also be considered in the context of several methodological limitations. First, despite the moderate cohort size (n = 269) at intake and a high proportion of patients with three evaluations of co-morbid disorders (n = 193, 75% of the living patients), the number of patients within some subgroups was relatively small, predisposing to type II errors and increasing the risk of spurious findings. The risk for type I error also exists theoretically, but is diminished by hypothesis-driven analyses and the confirmatory nature of multiple tests. Second, the possibility of selective attrition biasing our findings must be considered. Patients who were younger, had a higher baseline BAI score or were living alone were found to be more likely to drop out. However, age did not correlate with the individual change trajectory components. In a median split sensitivity analysis, only one path from ΔBDI0–6 to ΔBAI6–18 became significant in the below median BAI subgroup, but not in the high BAI subgroup (or the total cohort). However, this analysis overestimated the potential bias due to attrition because the actual difference in BAI scores was smaller (25.3 v. 21.0). Third, some subgroups of anxiety disorders were small and could not be investigated separately. Whether our results could be generalized to clinical settings with different anxiety disorder spectra remains unknown. Fourth, we did not evaluate inter-rater reliability of the co-morbid disorders. However, the findings of both cross-sectional and longitudinal co-morbidity were essentially the same when anxiety symptoms (BAI) were replaced with a categorical DSM-IV anxiety disorder in the statistical models. Fifth, the self-report instrument (BAI) we used for measuring dimensional anxiety in our LCMs measures somatic anxiety well, but may underestimate the dimension of phobic anxiety. Sixth, we only assessed DSM-IV diagnoses of alcohol abuse and dependence and did not use dimensional symptom scales for alcohol consumption. It is also possible that the number of new cases during the 18-month follow-up is low because of the time frame of 1 year in the DSM-IV. Seventh, this was a naturalistic study, and therefore treatment was not controlled. However, we took into account the possible effects of receiving antidepressant treatment by adjusting it at 6 and 18 months in our analyses, and, as we have reported previously (Melartin et al. Reference Melartin, Rytsälä, Leskelä, Lestelä-Mielonen, Sokero and Isometsä2005), there is a tendency in naturalistic studies for more severely ill patients to receive more treatment for depression. However, there were no statistically significant differences in the associations between anxiety (BAI) and depression (BDI) in patients on antidepressants and patients not on antidepressants at either the 6- or 18-month evaluation points. Thus, we consider it unlikely that treatment would have markedly biased our findings.
Conclusions
In adult secondary-care psychiatric patients with co-morbid MDD, we investigated the degree to which the longitudinal course of current co-morbid Axis I disorders followed the course of MDE. We found that the longitudinal course of co-morbid anxiety symptoms was associated moderately strongly and the longitudinal course of co-morbid AUDs moderately weakly with depressive symptoms. A significant individual variability existed in the courses of anxiety symptoms and disorders, but not in the courses of categorical AUDs. Patients with higher depressive symptom scores at an acute phase of depression were more likely to have an AUD in follow-up. Thus, in accordance with general population studies, our study supports the longitudinal and individual-level courses of both internalizing and externalizing disorders in adult patients with MDD being dependent, albeit to differing degrees, on the course of depressive symptoms.
Acknowledgments
This study was supported by research grants from the Finnish Medical Foundation, the Finnish Medical Society Duodecim and the Department of Psychiatry at Helsinki University Central Hospital.
Declaration of Interest
None.
