Comorbidity of personality disorders in anxiety disorders: A meta-analysis of 30 years of research

Abstract

Background

A comprehensive meta-analysis to identify the proportions of comorbid personality disorders (PD) across the major subtypes of anxiety disorders (AD) has not previously been published.

Methods

A literature search identified 125 empirical papers from the period 1980–2010 on patients with panic disorders, social phobia, generalised anxiety, obsessive-compulsive (OCD) and post-traumatic stress disorder (PTSD). Several moderators were coded.

Results

The rate of any comorbid PD was high across all ADs, ranging from .35 for PTSD to .52 for OCD. Cluster C PDs occurred more than twice as often as cluster A or B PDs. Within cluster C the avoidant PD occurred most frequently, followed by the obsessive-compulsive and the dependent PD. PTSD showed the most heterogeneous clinical picture and social phobia was highly comorbid with avoidant PD. A range of moderators were examined, but most were non-significant or of small effects, except an early age of onset, which in social phobia increased the risk of an avoidant PD considerably. Gender or duration of an AD was not related to variation in PD comorbidity.

Limitations

Blind rating of diagnoses was recorded from the papers as an indication of diagnostic validity. However, as too few studies reported it the validity of the comorbid estimates of PD was less strong.

Conclusions

The findings provided support to several of the proposed changes in the forthcoming DSM-5. Further comorbidity studies are needed in view of the substantial changes in how PDs will be diagnosed in the DSM-5.

Introduction

Anxiety disorders (ADs) are one of the most common mental disorders with a 12-month prevalence of 18% and a lifetime prevalence of about 30% (Kessler et al., 2005a, Kessler et al., 2005b). Depending on the nature and severity of an AD, the concurrent debilitation of daily functioning and quality of life may become substantial (Ghaedi et al., 2010, Lochner et al., 2003). As avoidance of situations and/or other people are the cardinal behavioural symptoms of ADs, the immobilisation and isolation tend to increase over time. This eventually impedes the quality of social relationships and support (Hickey et al., 2005, Stein and Kean, 2000), which in combination with the primary AD symptoms contribute to the chronic course of the disorder (Beard et al., 2010, Perkonigg et al., 2005, Rubio and Lopez-Ibor, 2007, Wittchen, 2002), particularly if left untreated.

The comorbidity of other mental disorders appears to be the rule rather than an exception (Barlow et al., 1986), which partly derives from the secondary functional problems arising over time. In a study by Sanderson et al. (1990), up to 70% of the patients had at least one additional axis I diagnosis. The high comorbidity is striking. However, such estimates may be inflated as the current diagnostic system (i.e., DSM-IV or ICD-10) largely neglects communality in symptom clusters across diagnostic groups. Nevertheless, mental disorder comorbidity is considered negative for the natural course of illness by implying additional kinds of dysfunction (Coryell et al., 1988) depending on whether it is a disorder of mood, substance abuse or personality (Ansell et al., 2011, Brown and Barlow, 1992). These three are the most common comorbid disorders.

Comorbid depression mainly involves anhedonia and reduced motivation. Such clinical features may strongly temper the effects of treatment (Spijker et al., 2001), and in particular psychotherapy treatments requiring own efforts, such as behavioural homework tasks commonly used in for instance cognitive-behavioural therapy. A substance abuse disorder may represent a coping behaviour strategy and a negative reinforcement factor counteracting change in automated avoidance behaviour. This paper focuses however on the proportions of comorbidity of axis II personality disorders (PD). Such comorbidity may more profoundly affect the prognosis and outcome of an AD compared to any other axis I comorbid diagnoses (Reich, 2003, Telch et al., 2011).

One distinction between axis I and II comorbid disorders seems to be in terms of severity and complexity. First, the degree of fear and phobic avoidance (Dreessen et al., 1994, Kose et al., 2009) is usually more pronounced in patients with axis II compared with axis I comorbidity, as well as the level of general psychopathology (Dreessen et al., 1994, Ozkan and Altindag, 2005). This is also reflected in analyses of health care costs showing that axis II PD patients represent a markedly higher economic burden to the health care system than for example depression and anxiety patients (Soeteman et al., 2008). Moreover, having a PD represents a strong vulnerability factor for developing other axis I disorders (Stein et al., 1993). Second, treatment of comorbid PDs are normally more complex and less optimistic than treatment of axis I comorbidities due to less favourable outcomes (Reich, 2003, Slaap and den Boer, 2001, Telch et al., 2011), higher drop-out rates (Sanderson et al., 2002), less positive patient expectations (Martino et al., 2012) and more challenges with establishing a durable and flexible therapeutic alliance (Bienenfeld, 2007, Martino et al., 2012). Furthermore, the degree of comorbid psychosocial impairment depends on the type of PD. For example in a study of 668 PD patients, Skodol et al. (2002) reported a higher degree of impairment among schizotypal and borderline patients than among obsessive-compulsive or avoidant PD patients.

Several reviews of general associations between PDs and ADs have been published (Bienvenu and Stein, 2003a, Brandes and Bienvenu, 2009, Brooks et al., 1989, Stein et al., 1993, Van Velzen and Emmelkamp, 1999). The general conclusions from these reviews are that the comorbid frequency of PDs among patients with an AD diagnosis vary considerably, but that the proportions are highest in the avoidant, dependent and compulsive PDs (cluster C), and smaller in the schizoid, schizotypal and paranoid PDs (cluster A) and the dramatic, borderline and anti-social PDs (cluster B). Patients with cluster A or cluster B PDs also appear to have a poorer treatment response than cluster C PD patients (Hansen et al., 2007, Noyes et al., 1990).

Very few meta-analyses on the same comorbidity have been published. To the best of our knowledge there is only one previous meta-analysis by Ng and Bornstein (2005). They specifically examined the comorbidity of a dependent PD (mean r=11). However, the ratio varied considerably between AD diagnostic groups, with panic disorder, obsessive-compulsive disorder (OCD), agora- and social phobia showing a higher prevalence of PDs, while generalised anxiety disorder (GAD) and post-traumatic stress disorder (PTSD) showed no relationship with PDs. The study was also criticised (Holmbeck and Durlak, 2005) for using a PD rather than an AD as the main inclusion criteria. The lack of previous meta-analyses using AD as the prime inclusion criteria, as well as studies that summarises the comorbidity of a PD across all AD diagnostic groups was therefore the main impetus for the present study. Additionally, we present data as proportions (equal to percentages), which represents a more user-friendly and intuitive effect size statistic.

The patterns of dysfunction and severity of a comorbid PD depend of the type of PD (Ansell et al., 2011). Panic disorder patients with or without agoraphobia are usually strongly confined to their home due to rigid avoidance of public, enclosed or open spaces (Rodriguez et al., 2007), hence constricting mobility and social participation. As avoidance is one of the core symptoms of a panic disorder, the avoidant PD is also one of the most frequent comorbid PDs (Brooks et al., 1989). An avoidant personality preference easily intensifies the primary anxiety problems due to the loss of exposure to potentially corrective experiences. A dependent PD is less common, but if present, the patient may develop an undue reliance on another person, hence representing a safety behaviour strategy. The undue reliance also affects reciprocity in the relationships with other people, which over time constricts the social network. Patients with a borderline PD have more intense panic attacks, and they also attract a lot more attention and concern from other people. It is also regarded as the single PD with the poorest prognosis for a successful treatment of a panic disorder (Nurnberg et al., 1989).

Patients with social phobia are overly afraid of negative evaluations and social embarrassments. Hence, they avoid or at least endure with great pain any situations that involve visibility or responsibility. The protracted avoidance and social withdrawal is also presumed as the prime reason for the lower socioeconomic status often characterising this group (Stein and Kean, 2000, Vorcaro et al., 2004). The avoidant PD is the most frequently occurring comorbid condition (Tillfors and Ekselius, 2009), which is expected considering the strong phenomenological overlap between these two diagnoses (Brooks et al., 1989, Reich, 2000, Rettew, 2000). However, in a review by Rettew (2000) the comorbidity estimates varied extremely, from 25% to nearly 100%. Clearly, this variation must be due to other factors than sampling error alone. Hence, we expected the avoidant PD to be the most prevalent comorbid PD in social phobia. Moreover, this PD is particularly related to a poor treatment outcome (Turner, 1987).

Patients with GAD experience constant worry and tension as their prime symptoms. Procrastination, avoidance behaviour and frequent reassurance seeking from others are typical compensatory strategies (Clark and Beck, 2010). Patients with social phobia and GAD appears to have higher rates of comorbidity than patients with panic disorders (Blashfield et al., 1994, Jansen et al., 1994). Cluster C PDs appears to be the most prevalent in GAD, particularly the avoidant and the OCD PD subtypes (Garyfallos et al., 1999). We therefore expected these to have the highest comorbid estimates. Like for social phobia, the avoidant PD also appear to be the most relevant PD predicting a poorer treatment outcome (Massion et al., 2002).

PTSD-patients are particularly prone to lapse into unhealthy behaviours by overusing alcohol or drugs to alleviate traumatic memories (Hidalgo and Davidson, 2000, Leeies et al., 2010), which over time leads to a poorer physical health and an overuse of health care services (Greenberg et al., 1999). A connection between PTSD and the axis II disorders avoidant (Ansell et al., 2011), obsessive-compulsive (Pietrzak et al., 2011) and borderline PDs have been established (Bienvenu and Stein, 2003b, Pagura et al., 2010, Shea et al., 1999), but also with paranoid and schizotypal PDs (Crepulja and Franciskovic, 2010, Pietrzak et al., 2011). Paranoid traits are understandable as PTSD patients often feel a constant need to be vigilant and on guard. However, the reported proportions vary across studies, to which the present meta-analysis may provide more reliable estimates.

OCD-patients are chronic sufferers who experience major impairment especially related to physical/occupational and emotional functioning (Hollander et al., 1996, Hollander et al., 2010). The degree of any PD has varied from 66% (Stein et al., 1993) to as low as 4% (Joffe et al., 1988), with the OCD PD as the most frequent comorbid condition (Alnaes and Torgersen, 1988, Samuels et al., 2000, Skodol et al., 1995). Obviously, a comorbid OCD PD may introduce a “double burden” in terms of intensified needs for order and perfection when performing OCD related checking or rituals in order to reduce anxiety.

From this brief outline of the literature, it is evident that a concurrent PD may impede both treatment and outcome of an AD, as well as possibly raising health care costs. The lack of previous meta-analyses covering all ADs, call for an empirical review to summarise and evaluate the current standing in the literature. It is also evident that the large between-study variation in the comorbid proportions of PDs across the various ADs, invite for a search for moderators that may explain this variability.

In the current study we examined the impact of diagnostic system (DSM-III-R versus DSM-IV), diagnostic methods (interview versus questionnaire), sample characteristics (patients receiving inpatient versus outpatient treatment), mean age of onset of AD and duration of AD. We expected higher comorbidity estimates from studies based on questionnaire data compared to structured interviews as that has been reported from PD comorbidity studies on for example eating disorders (Ramklint et al., 2010, Rosenvinge et al., 2000). Literature comparing the PD comorbidity between in- and outpatient samples is very scarce, except some findings indicating a higher degree of cluster B comorbidity in inpatients (Melartin et al., 2002). Hence, a clear-cut prediction could not be made, but the degree of comorbidity was expected higher among inpatients based on the psychiatric worse condition that an inpatient admission implies, particularly among patients with cluster B PDs. An early age of onset of an AD appears negative as it is associated with a longer duration of illness and a higher general symptom load in OCD (Wang et al., 2012), a history of childhood abuse and worse clinical outcome in GAD (Goncalves and Byrne, 2012), as well as generally higher comorbidity with other ADs (Klein Hofmeijer-Sevink et al., 2012). Moreover, early onset has also been related to higher comorbidity estimates between OCD and OCD-PD (Coles et al., 2008), between GAD and any PD (Garyfallos et al., 1999), and between social phobia and avoidant PD (Holt et al., 1992). Since the duration of an AD is partly related to the age of onset, we expected that an early age of onset and a longer duration of AD would imply a higher degree of comorbid PDs.

Hence, the paper had two primary aims: First, to compute mean weighted proportions of comorbidity between the major axis I ADs and all axis II PDs. Second, to examine whether any of the moderators described above could explain the variability reported in the literature.