Background Depression and anxiety are common in Parkinson's disease (PD) and although clinically important remain poorly understood and managed. To date, research has tended to treat depression and anxiety as distinct phenomena. There is growing evidence for heterogeneity in PD in the motor and cognitive domains, with implications for pathophysiology and outcome. Similar heterogeneity may exist in the domain of depression and anxiety.
Objective To identify the main anxiety and depression related subtype(s) in PD and their associated demographic and clinical features.
Methods A sample of 513 patients with PD received a detailed assessment of depression and anxiety related symptomatology. Latent Class Analysis (LCA) was used to identify putative depression and anxiety related subtypes.
Results LCA identified four classes, two interpretable as ‘anxiety related’: one anxiety alone (22.0%) and the other anxiety coexisting with prominent depressive symptoms (8.6%). A third subtype (9%) showed a prominent depressive profile only without significant anxiety. The final class (60.4%) showed a low probability of prominent affective symptoms. The validity of the four classes was supported by distinct patterns of association with important demographic and clinical variables.
Conclusion Depression in PD may manifest in two clinical phenotypes, one ‘anxious–depressed’ and the other ‘depressed’. However, a further large proportion of patients can have relatively isolated anxiety. Further study of these putative phenotypes may identify important differences in pathophysiology and other aetiologically important factors and focus research on developing more targeted and effective treatment.
Statistics from Altmetric.com
As well as causing distress, depression in Parkinson's disease (PD) has been associated with greater functional disability,1 rate of physical and cognitive decline,2 dementia risk3 and mortality,4 and reduced quality of life.5 Anxiety is also a prominent feature of PD6 7 although implications for outcome are less well understood. Evidence for the efficacy of pharmacotherapy remains weak and often inconsistent8 9 although there are a growing number of trials of pharmacological and psychological interventions for depression (eg, see http://ClinicalTrials.Gov). Trials are needed in PD anxiety although there is some evidence that anxiety symptoms improve in depressed PD patients treated with the selective serotonin reuptake inhibitor (SSRI) citalopram.10 Although depression and anxiety symptoms can occur in isolation in PD, they frequently coexist.11 12 While some have suggested that this reflects the comorbidity of distinct pathophysiological substrates,11 others argue that anxious depression represents a specific and common depressive subtype in PD.7
The relationship between depression and anxiety has important implications for treatment and clinical trials. Non-PD patients with major depression with marked anxiety (anxious depression) show poorer outcomes than those with major depression alone when treated with SSRIs and when switched/augmented with other agents.13 If anxious depression is common in PD, similar reduced treatment efficacy may be anticipated, while combining subtypes could produce aggregate results that do not adequately inform the response to treatment of either subgroup.
Broader clinical heterogeneity of PD has been examined using empirical approaches such as cluster analysis (CA) or latent class analysis (LCA).14 In contrast with prespecified clinical or theoretically defined subgroups, data driven methods seek to identify clusters of patients defined by low intragroup but high intergroup differences across a set of selected variables. Applying similar methods to depression and anxiety related symptoms may identify empirical subgroups that then become available for validation and further study. Differences in pathophysiology and other factors involved in symptom onset and maintenance could guide the development of more targeted treatment. The present study describes the first use of a data driven approach to explore possible heterogeneity in depressive and anxiety related symptoms in patients with PD.
Eligibility and recruitment
Patients were recruited over a 12 month period from specialist PD or movement disorder outpatient clinics (ethics ref 07/MRE01/9) as part of a longitudinal study (PROMS-PD) (UKCRN ID 2519). Inclusion criteria were a clinical diagnosis of idiopathic PD,15 the ability to provide informed consent at entry into the study and living within 2 h of travel time of a study research centre. Exclusion criteria were the presence of sensory loss or communication difficulty (including inadequate command of English) sufficient to interfere with assessment. Patients with cognitive impairment or severe psychiatric disorder were excluded only if these interfered with the capacity to consent. The eligibility criteria were deliberately chosen to include as broad a range of patients as possible in terms of age and clinical severity.
Participants were assessed over two sessions, typically in their own home. Where possible patients were assessed in their motor ‘on’ state. Depressive, anxiety and related symptoms were assessed using a semi-structured interview based on the Geriatric Mental State (GMS),16 an instrument designed to assess psychopathology in older adults, including those with significant cognitive impairment. Items were rated 0=absent or ‘normal’; 1=present but not prominent (mild to moderate intensity, infrequent or fleeting); and 2=prominent (severe, frequent or persistent). Symptoms were rated as reported or observed.17 Self-reported depression and anxiety was assessed using the Hospital Anxiety and Depression Scale (HADS),18 with subscale scores of ≥11 indicating significant symptoms. Motor function and disability was assessed using the Unified Parkinson's Disease Rating Scale (UPDRS)19 parts II and III, Hoehn and Yahr Scale20 and Schwab and England Scale.21 The ratio of tremor to non-tremor motor symptoms was calculated as previously described.22 Cognition was assessed using the Addenbrooke's Cognitive Examination-Revised (ACE–R),23 with a score of ≤83 indicating significant impairment.24 Annual assessments are planned for up to 4 years.
LCA can identify unobserved (latent) subgroups of cases that share similar symptoms from a set of multivariate categorical data. Model based CA methods such as LCA avoid a criticism of conventional heuristic CA approaches such as linkage that do not provide a formal basis for determining the optimal number of clusters or indeed the existence of more than a single cluster. LCA provides inferences for statistically comparing the fit of solutions across different numbers of clusters, and is also better able to handle missing data. We employed LCA of binary symptom variables using M-Plus software.25 Subjects were allocated to classes based on the maximum estimate of a posteriori probability of cluster membership (maximised allocation probability method). Individual GMS symptoms were dichotomised as ‘prominent’ (score=2) or ‘absent or present but not prominent’ (score <2), to provide a classification based on the presentation of symptoms with clear clinical significance. Symptoms with a frequency of less than 2% (n<10) were excluded. In total, 28 items were identified as relevant to phenotyping (see figure 1).
Preliminary assessment of validity was conducted by examining the association between the defined classes and other variables not included in the LCA. One way analysis of variance with paired comparisons (Tukey's HSD, p<0.05) was used to determine whether LCA classes could be discriminated on continuous demographic and clinical variables, and χ2 for categorical data. Finally, multinomial logistic regression was used to examine clinical and demographic predictors of class membership. This latter analysis included a range of dichotomised measures associated with broad clinical heterogeneity in previous studies.14
Nine hundred and forty-one patients were invited to participate of whom 525 consented but 12 subsequently withdrew before completing the initial assessment. Complete assessment was obtained from 513 patients. The sample had a wide range of disease durations and severities although the majority were in Hoehn and Yahr stages II–III (table 1) with a non-tremor dominant motor profile.26 Motor symptom onset preceded the age of 50 years in 18.7% of cases. Approximately 30% of participants showed evidence of significant cognitive impairment on the ACE-R. On the HADS, 13% reported marked depressive symptomatology and 22% anxiety. Mild to moderate symptoms (HADS subscale score 8–10) were reported in a further 22% and 23% of cases, respectively. Almost all (94.9%) were taking antiparkinsonian drugs, and 24% reported taking antidepressant and/or anxiolytic medication.
Figure 1 shows the ranked profile of 28 GMS items. The most frequent prominent symptoms were worry (33.5%), subjective tension and restlessness (33.2%), and fatigue (27.6%). Symptoms such as excessive energy, belief about punishment and suicidal thinking were rare (<2%, data not shown) and not included in the LCA.
Empirically defined patient subgroups
Seven LCA solutions were evaluated, with a four class solution providing the best overall fit across a range of information indices (table 2). The median allocation probability was 99% with 90% of the sample allocated to their final class with a probability of 80% or more. The largest class (class 4) included 60.4% of the sample, class 3, 22% and classes 1 and 2, 8.6% and 9.0%, respectively.
Interpreting classes from cluster based methods is a largely non-empirical process and involves visually identifying those features which most clearly (a) define an individual class and (b) distinguish it from other classes. Figure 2 shows the symptom probability profiles for the 28 variables for each of the individual class. Features of classes 1 and 2 were sad and/or empty mood, loss of interest, slow thinking, loss of confidence, lack of energy and poor concentration. In contrast, loss of enjoyment was rare and symptoms of hopelessness, worthlessness and guilt relatively uncommon, particularly in class 2. In addition, and in contrast with class 2, patients in class 1 also showed high probabilities of prominent worry and tension/restlessness, irritability, general anxiety/panic, specific fear and autonomic symptoms of anxiety. Class 3 was marked by a profile of prominent worry and tension/restlessness, and moderate probabilities of other severe anxiety related symptoms but with a low probability of depressive symptomatology. Finally, class 4 was characterised by overall low probability of prominent symptoms.
Table 3 shows the main demographic and clinical characteristics of the four classes. All variables showed highly significant differences (p<0.001), including when age was used as a covariate. Patients in class 4 tended to be older, with a high proportion of males, later age at onset and shorter disease duration, with lower levels of motor symptoms, disability and medication. Non-tremor symptoms were least common in this class. Class 2 was similar to class 4 in terms of age, age of onset, duration of PD and level of medication but showed more cognitive and motor impairment and disability, and somewhat higher levels of HADS depression and anxiety. This class also displayed the most pronounced non-tremor dominant symptom profile. Classes 1 and 3 were the youngest with more advanced disease and higher mean levels of antiparkinsonian medication than classes 2 and 4. Class 1 in particular showed early mean age of onset and long disease duration. Class 1 patients also showed greater disability and twice the level of significant cognitive impairment than class 3, and markedly increased levels and rates of anxiety and depression.
Predictors of class membership
Class membership was assessed using multinomial logistic regression with a set of dichotomised demographic and clinical variables: age (<70/≥70 years), gender (female/male), age at diagnosis (<55/≥55 years), PD duration (<5/≥5 years), levodopa equivalent daily dose (LEDD) (<600/≥600 mg/day), Schwab and England (<80/≥80), UPDRS III (<23/≥23) and ACE-R (<83/≥83). Class 4 was the reference category. Duration of PD did not independently predict class membership in any model and was therefore not included in the final analysis.
Relative to class 4, membership of class 1 was significantly (p<0.05) predicted by younger age (OR 2.6, 95% CI 1.0 to 6.6), PD onset less than 55 years (OR 3.9, 1.6 to 9.3), greater level of disability (Schwab and England) (OR 2.6, 1.2 to 6.0) and higher UPDRS III score (OR 2.4, 1.0 to 5.5). Female gender approached but did not reach significance (p<0.10). Membership of class 3 was predicted by a similar set of variables: younger age (OR 2.2, 1.2 to 4.0), female gender (OR 3.0, 1.8 to 4.8), younger age of PD onset (OR 1.7, 1.0 to 3.1), higher LEDD (OR 1.8, 1.1 to 2.9) and greater disability (OR 2.3, 1.3 to 4.1). Only cognitive impairment significantly predicted membership of class 2 (OR 3.7, 1.8 to 7.7).
This study provides the first detailed investigation of affective heterogeneity in PD using a broad range of indicator variables associated with (although not specific to) the psychiatric constructs of depression and anxiety. The results supported a four class solution with empirically defined subgroups that appear distinctive in terms of clinical and demographic features. We should not equate such subgroups with syndromal disorders, nor probabilistically defined class membership with diagnosis. Nevertheless, it is useful to suggest provisional clinically meaningful labels to the classes. Patients in the largest group (class 4) showed low levels of prominent symptomatology and can be labelled ‘psychologically healthy’, although the high symptom threshold does not rule out some problems. Class 3 showed prominent anxiety related features with a profile similar to generalised anxiety disorder (GAD)27 and can be labelled ‘anxious’. The profiles of classes 1 and 2 were more complex. Both showed elevated probabilities of core depressive symptoms of sad or empty mood and loss of interest, along with other supporting features such as poor concentration, slow thinking and fatigue, although loss of enjoyment was largely absent, and feelings of worthless, hopelessness and self-blame evident only in class 1. Patients in class 1 had a higher probability of prominent depression related symptomatology but also showed marked prevalence of subjective tension, worry and general anxiety or panic and irritability in class 1, suggesting the label ‘anxious depressed’. Class 2, together with depressive features, also showed high levels of cognitive impairment and may reflect a ‘depressed’ or possibly ‘apathetic depressed’ subtype.
A recent systematic review and meta-analysis14 summarised studies using similar approaches to define broad clinical heterogeneity in PD. Two studies22 28 independently identified four subgroups of patients labelled ‘young onset’, ‘tremor dominant’, ‘non-tremor dominant’ and ‘rapid disease progression’. Only the ‘non-tremor dominant’ group (including akinesia and postural instability and gait symptoms) was associated with increased levels of depression. A young onset cluster, with typically slow rate of progression, was associated with low levels of depression, although another study29 found the opposite result. Other studies, not included in the systematic review, have examined non-motor heterogeneity. One study30 analysed 10 broad psychiatric and behavioural symptoms of the Neuropsychiatric Inventory,31 identifying a large ‘low total Neuropsychiatric Inventory’ subgroup and four others exhibiting predominant apathy, psychosis, depression or anxiety. However, substantial levels of depression and particularly anxiety were evident in all of the groups except for the first. Another study considered depression, anxiety, apathy, daytime somnolence and cognition,32 and defined four subgroups, with the largest showing good cognitive and psychiatric function. The remainder showed cognitive impairment, psychiatric impairment or both. No evidence of heterogeneity was found within the psychiatric symptoms themselves although the use of a single global index for each precluded detailed assessment.
Cluster based methods are subject to a variety of methodological factors that can influence the robustness and therefore potential validity of the solutions. As noted, LCA has advantages over traditional CA methods in providing a range of indices to inform decisions about the optimal number of clusters. The present results strongly supported a four class solution over one with more or fewer subgroups. The high allocation probabilities further suggest that patients were assigned unequivocally to one of the four classes. Ideally, we would have liked to replicate the classification in an independent sample to further confirm its reliability, as done in one similar study using cluster analysis.30 Unfortunately, our sample size, although large, was insufficient given the number of parameter estimates being modelled (n=115).
Face validity was high and the subgroups clinically interpretable. Validity was further supported by the associations of the classes with a range of variable not included in the LCA. These converged on evidence from previous studies describing associations between depression and broad clinical and motor phenotypes. Our results support a link between psychopathology and non-tremor dominant motor phenotype, with the ‘healthy’ subgroup showing the highest proportion of tremor symptoms and the ‘apathetic depressed’ group the least. Younger age of onset emerged as a distinguishing feature in class 3 but particularly in class 1, along with female gender and higher LEDD. However, unlike the apparently benign psychiatric profile of the ‘young onset’ groups of some previous studies,22 28 younger age of onset in the present study was associated with marked anxiety (classes 1 and 3). The present study also suggests that at least some patients with younger onset (class 1) experience significant depressive symptomatology, particularly those with longer duration, advanced disease and greater disability, similar to a previous report.29
An important question is whether subtypes remain stable or patients move between subtypes. Pending longitudinal study, a partial answer can be found in evidence on the duration of the individual symptoms that define class membership (see supplementary information available online). Almost all prominent symptoms reported had been present for most days in the past 6 months in over 80% of patients, and for many, most days in the past 2 years. Although not proving that the subtypes are stable, it suggests that that the symptoms on which they are based are persistent in this sample.
A criticism of cross sectional studies is that empirically distinct clusters may simply reflect observations of patients at different stages of a homogenous but progressive disease.14 A number of facts suggest that this is not the case with the present findings. The between group differences were modest compared with the range observed within the classes, with the ‘anxious’ and ‘depressed’ subgroups showing almost identical mean disease durations and levels of disability. Also, the logistic regression analysis failed to indicate a role of disease duration in predicting subtype. In contrast, age of onset emerged as a more significant influence, with younger onset (independent of duration) associated with the two anxiety related subtypes.
The high level of anxiety related symptomatology in this sample, and specifically in two of the subgroups, is consistent with emerging evidence.6 7 33 Even when sub-syndromal, GAD-like symptoms of subjective tension, worry and irritability can be a significant source of distress, particularly if chronic, but may be missed without adequate screening. Anxiety related symptoms are a common feature of depressive disorder although the depressive symptoms are generally held to be primary and the target of treatment. However, anxiety symptoms may turn out to have a high prominence in some PD patients with depression34 while symptoms such as anhedonia and guilt may be less characteristic with implications for assessment and diagnosis. If anxious depression is common in PD it would also have important implications for management and outcome. In the elderly, remission rates of comorbid depression and GAD are only 27% over 3 years compared with 41% for depression and 48% for GAD.35 In a major trial of citalopram (STAR*D), time to remission was longer for anxious depression, remission rates were significantly lower and reported side effects higher.13 The broad phenotypic similarity of patients in classes 1 and 3 suggests that these two subtypes may be closely associated. Longitudinal study will test the hypothesis that persistent GAD-like anxiety symptoms are a risk factor for anxious depression. If so, then early identification and treatment of the chronic anxiety symptoms such as worry and tension may be indicated.
Heterogeneity of affective disturbance in PD may have implications for pathophysiology. Apathetic depressive subtype, associated with non-tremor dominant PD and cognitive dysfunction, may have a mainly dopaminergic substrate with involvement of the ventral striatum and associated prefrontal regions, plus a higher burden of cortical pathology in addition to the more typical nigrostriatal changes.36 This subtype may have contributed to the previously reported associations between depression and rate of cognitive decline and dementia risk.2 3 In contrast, non-dopaminergic mechanisms may be the major factor in the anxious depressed and anxious subtypes. Evidence of serotinergic involvement in depression in PD is inconsistent37 (perhaps due to heterogeneity) although there is some support for the involvement of the noradrenergic system in PD depression,38 with monoamine uptake ([11C]RTI 32) in noradrenergic areas related to the severity of the anxiety but not depressive symptoms.
Demonstrating heterogeneity of depression in PD has implications for clinical trials. If studies combine patients with potentially different pathophysiology the outcome may be different response rates to treatment. This dilution of effect may account, at least in part, for the apparently poor overall response rates observed with conventional SSRI based antidepressants in PD.8 Selecting homogeneous samples with specifically targeted treatment may provide more reliable and useful evidence to focus clinical management.
The present study used convenience sample of outpatient clinic attenders. The rates of significant self-rated depressive (13%) and anxiety related (22%) symptomatology from the HADS were comparable with that reported in two recent validation studies, with figures of 17% and 20%,39 and 14% and 22%,40 respectively, and with very similar mean scores. However, the purpose of the present study was not to define the prevalence of the subtypes (the absolute values for which should be treated with caution) but to identify their characteristics and associated features. Studies in different populations may produce different prevalence rates of the four subtypes, just as the rate of the ‘rapid disease progression and old age at onset’ clinical subtype described in other studies has varied from 6% to 64%.14 Such variability in prevalence, however, does not challenge the validity of the subtype.
In conclusion, this study suggests that models of depression and anxiety as distinct unitary conditions in PD may be inaccurate. GAD-like anxiety symptoms were among the most common in the sample and occurred both independently and in association with significant depressive symptoms. The distinctive clinical and demographic profiles suggest that anxious depression may represent a subtype distinct from depression. Such heterogeneity may need to be considered when carrying out research into the pathophysiological substrate of affective disorder in PD and in planning current management and future clinical trials.
The authors acknowledge support from: Parkinson's UK; NIHR Biomedical Research Centre for Mental Health at the South London and Maudsley NHS Foundation Trust (SLaM) and Institute of Psychiatry, King's College London; NIHR Dementias and Neurodegenerative Diseases Research Network (DeNDRoN); Wales Dementias and Neurodegenerative Diseases Research Network (NEURODEM Cymru); NIHR Mental Health Research Network (MHRN); and British Geriatric Society. In addition to the listed authors, we thank the following members of the PROMS-PD Study Group who all made a significant contribution to the work reported in this paper. London: KR Chaudhuri, King's College Hospital NHS Foundation Trust, London (participant recruitment); C Clough, King's College Hospital NHS Foundation Trust, London (participant recruitment); B Gorelick, Parkinson's Disease Society, London (member of the study management group); A Simpson, Institute of Psychiatry, King's College London, London (data collection); R Weeks, King's College Hospital NHS Foundation Trust, London (participant recruitment). Liverpool and North Wales: M Bracewell, Ysbyty Gwynedd, Bangor (participant recruitment, data collection); M Jones, University of Wales Bangor, Bangor (participant recruitment, data collection); L Moss, Wythenshawe Hospital, Manchester (participant recruitment, data collection); P Ohri, Eryri Hospital, Caernarfon (participant recruitment); L Owen, Wythenshawe Hospital, Manchester (participant recruitment, data collection); G Scott, Royal Liverpool University Hospital, Liverpool (participant recruitment); C Turnbull, Wirral Hospitals NHS Trust, Wirral (participant recruitment). Newcastle: S Dodd, Institute for Ageing and Health, Newcastle University, Newcastle upon Tyne (participant recruitment, data collection); R Lawson, Institute for Ageing and Health, Newcastle University, Newcastle upon Tyne (participant recruitment, data collection).
↵* Members of the PROM-PD Study Group are listed in the Acknowledgements section.
Funding The study was funded by Parkinson's UK (previously Parkinson's Disease Society).
Competing interests None.
Ethics approval This study was conducted with the approval of the South East Research Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.