Abstract
Currently, there is little international consensus regarding the pathogenesis of chronic fatigue and related syndromes. As clinical syndromes are traditionally defined by characteristic symptoms and clinical course, over the last decade various national research and/or international committees have set out to define the key features of a more restricted group of patients, now termed as suffering from ‘chronic fatigue syndrome’ (CFS) [1–5]. Each of the proposed criteria sets for CFS have struggled to come to terms with a range of complex issues including: the relative weights that should be attached to common symptoms such as prolonged mental and physical fatigue, muscle pain, sleep disturbance, poor concentration and mood changes and/or clinical signs; severity of disability; duration of symptoms; acuity of onset; premorbid and/or concurrent psychological morbidity [1, 3, 5]; and the place of proposed laboratory markers (such as those for immune dysfunction) [2]. Those criteria sets which include many possible symptoms, or require a significant number of different symptoms, are biased towards the inclusion of patients with an increased number of somatic symptoms, longer illness duration and more functional disability. Such factors are, however, more likely to identify patients with a known psychiatric disorder [6–8] rather than a novel syndrome.
An alternative to expert consensus is to use statistical analyses to derive groups of subjects with more homogeneous characteristics [9]. There is no ideal method for the statistical ‘clustering’ of patients, although latent class analysis (in association with mixture analysis or principal components analysis) may be useful [6,10–12]. Importantly, this approach is free of clinician-bias, and factors such as illness onset, course and treatment response can be used to examine the validity of any derived typology. Given the subjective nature of the diagnosis of CFS and the consequent possibility of significant variation in sample characteristics, we set out: (i) to compare the clinical characteristics and functional disability in patients drawn from CFS research centres in Australia, the USA and the UK; (ii) to re-examine the earlier proposal [6] that relevant subgroups of CFS subjects could be identified empirically on the basis of symptom patterns alone; and (iii) to validate any derived classification against clinical data such as subject age and sex, duration of illness, functional disability, psychological morbidity and family history of psychiatric disorder.
Method
Study centres
Investigators from eight international centres with a known research interest in CFS agreed to participate in this study. Subjects therefore were enrolled from differing geographical and clinical specialty settings (immunology, infectious diseases, rheumatology and psychiatry), although each was predominantly university-based and either a secondary or tertiary referral centre.
Subjects
In order to avoid a priori judgements and as a result of the multicentre nature of the study, subjects were selected using the existing CFS diagnostic criteria employed at each centre. Subjects enrolled from Prince Henry Hospital (Australia) all met Lloyd et al. [2] clinical criteria (6 months chronic persisting or relapsing fatigue and neuropsychiatric dysfunction), subjects from the National Institutes of Health (USA) and Miami samples met Centers for Disease Control (CDC) criteria [1], whereas subjects from Michigan and Belfast were diagnosed according to ‘modified'CDC criteria [4, 5]. The study samples from London and Oxford both met UK criteria for CFS [3] and in addition, the sample from London excluded patients that met DSM-III-R [13] criteria for somatization disorder. All patients were evaluated to exclude other medical causes for current symptoms [4, 5] and laboratory investigations were carried out in accordance with the practice of each centre. Patients with an identified psychotic disorder and/or a diagnosis of drug or alcohol dependence (premorbid or current) were excluded. One centre (Boston) selected patients who each met clinical criteria for fibromyalgia [14], as well as satisfying Prince Henrycriteria [2] for a diagnosis of CFS.
Questionnaire
The subjects were asked to complete a 119-item self-report questionnaire comprising six sections. Section one recorded age, gender, marital status, educational achievement and employment status. Section two determined illness duration, character of symptom onset (sudden vs gradual) and factors considered relevant to onset (e.g. a viral or ‘flu-like illness’). In Section three subjects rated 48 symptoms as either present or absent. These were the physical, cognitive and neuropsychiatric symptoms considered likely features of the disorder and were an extension of the 38 symptoms utilized in an earlier epidemiological survey [15]. Ten further questions assessed more general, nonspecific somatic symptoms as well as evaluating the effects of sleep, heat, inactivity, emotional stress, mental exertion and vigorous physical activity on fatigue. These were rated on a five-point scale: 0, ‘no change or minimal improvement/deterioration’; 1, ‘moderate improvement’; 2, ‘complete or near complete improvement’; 3, ‘moderate deterioration’; and 4, ‘severe deterioration’. The extent to which fatigue fluctuated during the day was rated as were (for female subjects) changes in menstrual symptoms following the onset of the disorder, including whether fatigue levels fluctuated with the menstrual cycle.
Specific instructions were given not to rate symptoms that were present only in the first 3 months of the disorder. This was to minimize the reporting of acute symptoms present only following a precipitating event such as a viral infection, and not necessarily part of the ongoing symptom complex. For each positive symptom, subjects then rated the overall severity and frequency of that feature. Severity was rated on a three-point scale: 0, ‘mild’; 1, ‘moderate’; and 2, ‘severe’. Frequency was rated on a four-point scale: 0, ‘infrequent’; 1, ‘frequent’; 2, ‘very frequent/constant symptoms’; and 3, ‘cyclical’. The overall severity CFS: A MULTICENTRE STUDY 522 and frequency of depressed mood was also rated using the above scales. The relevant questions from the Diagnostic Interview Schedule (DIS) [16] were used to delineate the presence of an episode diagnosis of DSM-III-R major depression [13] and/or panic disorder.
Information on significant past medical history, including health problems which required specialist attention or hospitalization, as well as a previous history of treatment for anxiety or depression were recorded (section four). A family history of treatment for depression or other psychiatric illness was also sought (section five). Functional impairment (section six) was evaluated in several ways. Work participation and social activity were rated on a three-point scale: 0, ‘no change or minimal reduction in activity’; 1, ‘moderate reduction in activity’; and, 2, ‘completely unable to participate’. Subjects recorded the frequency of visits to a medical practitioner as well as levels of current and premorbid physical activity. Data from section three (symptom check-list) were used as the basis for the subclassification of study subjects. Demographic information, course of illness, past and family history and data on levels of functional impairment were used to test its validity.
Statistical methods
The statistical methodology used in the study is similar to that reported previously by Hickie et al. [6]. Fifty-five items from section three (symptom check-list) were selected for which a 0–1 codingof ‘absent’ or ‘present’ was possible. These were the 48 symptom items and seven items (all coded dichotomously to share the same metric) which assessed the effect of moderating factors (e.g. sleep or inactivity) and fatigue fluctuation.
The analysis had two steps: first, the reduction of the individual 55 items to a small number of scores on continuous dimensions; and, second, the use of combinations of these dimensions to identify patient subclasses. The first step comprised a principal components-like analysis where the matrix analysed contained sums-of-squares and crossproducts rather than the usual correlations or covariances. This modification allowed item differences in means and variances to influence the results. The second step used the method called latent profile analysis (LPA) [9], which assumes that the sample comprised a mixture of patients from one or more different classes, that the patients’ class memberships are unknown and they have been measured on a number of continuous measures (in this case the dimensions from step one). For a specified number of classes, LPA estimates the proportions in each class, the means of each measure for subjects within a particular class, and allows each individual to be allocated to a particular class. This procedure is similar to latent class analysis in that the underlying variable is categorical, except that instead of categorical measures (manifest variables) there are continuous measures (in this case scores on the principal components).
Results
Subjects and clinical characteristics
Complete symptom data sets were available from 744 patients. The demographic characteristics and comparative length of illness of subjects are shown in Table 1. In the total sample there was little correlation between illness duration and age (r = 0.17), but male subjects had a shorter illness duration than females (6.7 years for males vs 8.1 years for females, p = 0.030). Forty-six per cent of subjects reported that their symptoms began suddenly, with 72% recording a ‘viral illness’ at onset. Twenty-one per cent believed that a specific environmental factor (e.g. exposure to toxic fumes or physical injury) was associated with their illness onset. The majority of subjects (71%) also reported a fluctuating course, with 21% of these having experienced periods (days to weeks) of complete symptom remission.
Demographic characteristics and symptoms reported in more than 80% of the subjects, latent profile analysis class allocations and history of mood disturbance across study centres
Across all centres subjects were predominantly female (mean = 74% range = 64–96%), middle-aged (mean = 40.8 years, range = 35.0–50.3) and chronically ill (mean illness = 7.9 years, range = 3.9–12.8). This female predominance was especially pronounced in the fibromyalgia patients, however, this centre population did not differ significantly from the other centres across symptom or other illness variables. The key symptoms of fatigue and malaise as well as neuropsychological symptoms (poor concentration and memory impairment) were among the most common features (see Table 1). Two items related to pain (generalized and/or muscular) and two items related to sleep dysfunction (non-restorative sleep and ‘global’ sleep disruption) were also prominent. Overall, the number of symptoms reported as present at some stage during the illness had a range of 3–47 (mean = 28, median = 27).
For female subjects, 39% reported more painful menstruation during their illness, with 33% noting increased menstrual irregularity. Fiftytwo per cent felt that their fatigue worsened premenstrually.
Levels of disability
Subjects across separate study sites reported similar levels of disability, symptom severity and concurrent psychiatric morbidity. Fiftytwo per cent reported being unable to work during the course of their illness and 61% stated that they had been unable to complete morethan 1 h of daily active work or exercise over the course of their illness. There were highly significant correlations between items which measured various domains of functional impairment (e.g. ‘severe’ symptoms overall correlating with inability to work r = 0.62, p < 0.001 and social impairment r = 0.45, p < 0.001).
Thirty-nine per cent of the sample had had an episode of major depression (MDE) during the illness, with 17% meeting criteria for panic disorder. Twenty per cent of subjects had had previous treatment for depression, however, data from the questionnaire did not allow for specific premorbid psychiatric diagnoses to be made. Surprisingly, an MDE (39% of sample) was not associated with a statistically increased risk of functional impairment: 53% of subjects with MDE were unable to work versus 51% without MDE (χ2 = 3.4, p = 0.180), and 56% with MDE unable to participate in ‘normal activities’ due to symptoms versus 48% without MDE (χ2 = 4.7, p = 0.140). However, an MDE was strongly associated with more frequent visits to a physician (51% with MDE attending more than four times a year vs 34% without MDE; χ2 = 20.9, p < 0.001). An MDE was associated with an average positive symptom report of 25 (SD = 7.5) versus an average positive symptom report without MDE of 31 (SD = 7.0, p < 0.001). A previous history of depression was associated with a positive symptom report of 28 (SD = 7.8) whereas those subjects without a history of depression recorded a positive symptom count of 29 (SD = 7.1, p = NS].
Latent profile analysis
From the 55 symptom items, five components were retained after principal components analysis (PCA). Using LPA, initial models assumed that the population included a mixture of two and three subclasses. The three-class solution generated class prevalences of 63%, 29% and 7% respectively meaning that most of the subjects represented one category of symptom profile and two other groups of patients were also represented but progressively less frequently. However, as a result of the inadequate representation of class three across study centres (two centres with no subjects in class three and two centres with less than 5% of their subjects in this class), this model was not pursued. The simpler two-class solution is utilized hereafter.
The means of the first five PCA components are shown in Table 2. Overall, 68% of the subjects were allocated to class one and 32% to class two. Both classes weighted heavily on component one (labelled ‘classical CFS symptoms’), which included items considered core features (fatigue, malaise, neuropsychological impairment and sleep disturbance). Class two subjects showed significantly higher mean scores for component two (‘multiple somatic symptoms’ e.g. swollen joints, dysphagia and painful eyes), component three (‘depression and anxiety’ e.g. panic attacks, 2 weeks of depression during episode) and component four (‘subjective nodal pain and swelling’ e.g. subjective report of swollen cervical or generalized glands, report of tender cervical or generalized glands). The mean scores obtained for component five (‘fatigue variability’), which represented items assessing the effect of various moderating factors on fatigue (e.g. sleep, temperature and emotional stress), did not differ significantly between the two subject groups.
Means of ‘principal components’ utilized in latent profile analysis by class allocation
The distribution of classes across centres is shown in Table 1. Comparative clinical characteristics are shown in Tables 3 and 4. Class one was characterized by younger age, lower female to male ratio, a shorter illness duration, reduced prevalence of an episode diagnosis of panic disorder or MDE, lower rates of previous treatment for depression and anxiety and less reported family history of affective or other psychiatric disorder. Class one members also recorded less functional impairment and had a lower rate of previous attendance to a physician for medical problems. There were no reported differences as to whether symptoms commenced suddenly or gradually (47% class one subjects with sudden onset vs 44% class two; χ2 = 1.5, p = 0.820).
Comparison of symptom report and fatigue modulating factors between Class one and Class two CFS subjects
Latent profile analysis class designation vs. demographic, psychiatric and disability data
By contrast, class two subjects reported their symptoms as being less responsive to factors expected to improve fatigue such as sleep and physical inactivity. As well as reporting a high prevalence of typical CFS symptoms (e.g. fatigue, malaise and sleep disturbance), class two subjects also reported high frequencies of ‘atypical’ symptoms such as loss of vision, incontinence, swollen joints and dysphagia. The mean number of positively rated symptoms for class one subjects was 24 which was significantly lower than the mean of 36 for class two subjects (p < 0.001).
Importantly, there were significant intersite differences in class allocation, with the proportion of class one subjects varying from 94% in Belfast to 52% in Michigan (Table 1).
Conclusions
In this study we have shown that patients currently diagnosed with CFS in university-based research centres in Australia, the USA and the UK have similar overall demographic and clinical characteristics and report comparable levels of psychological morbidity and functional disability. By utilizing a multivariate statistical procedure (LPA), however, we were able to identify at least two clinically important subgroups. One group, consisting of approximately one-third of the total sample (class two), displayed characteristics more suggestive of a somatoform disorder, namely: (i) greater numbers of non-specific or ‘atypical’ symptoms (including those traditionally associated with somatoform disorders such as loss of vision, bladder disturbance and dysphagia); (ii) higher psychological morbidity and health care utilization (both in the past and concurrently); (iii) a longer duration of illness; (iv) a higher female to male ratio; and (v) more functional disability. These international results replicate earlier findings based only on Australiansubjects [6]. The significant between-group differences for the prevalence of premorbid (as evidenced by treatment rates) and familial psychiatric disorder argues against the proposition that the groups differ only along some shared dimensional factor such as symptom severity, functional disability or concurrent psychological morbidity. Rather, it suggests a constitutional vulnerability to a somatoform disorder in class two subjects. An alternative possibility is that patients with more prolonged disorders develop some (but not all) of these features as a consequence of the chronicity of their condition.
Importantly, the presence of a diagnosis of a major depressive episode, although associated with increased health care utilization, was not associated with greater functional disability, nor did current or premorbid depression predict increased positive symptom reporting. From a traditional psychiatric perspective, these were unexpected findings and suggest that the presence of affective disorder alone does not account for the other symptomatic differences between the two groups. It is consistent, however, with the findings from other aetiological studies which suggest that the liability toward reporting such somatic syndromes is determined by independent genetic and/or environmentalfactors [17–21]. Previous attempts to explain such syndromes simply in terms of unrecognized depressive disorders are no longer consistent with either these aetiological studies or treatment studies utilizing common antidepressant agents [22, 23]. The clinical management of these patients is likely to be enhanced by modes of treatment based on sleep-wake cycle [24] and/or specific cognitive–behavioural approaches [25, 26].
This study shows the difficulties which arise when a traditional approach to diagnosis and classification is adopted in this patient group. We used a method without prior clinical assumptions (e.g. the significance of certain clinical symptoms such as fever, lymphadenopathy or course of illness factors such as onset aftera ‘viral’ illness) to define subgroups. Expert-derived classification systems [3, 7, 27] do not discriminate between subjects, as those in class two (likely somatoform disorder) report ‘classical'CFS features as frequently as class one subjects. While such systems (and their inevitable revisions) will continue to drive much clinical and aetiological research [5], it cannot necessarily be assumed that they have yet succeeded in describing distinct or valid entities. Similarly, the use of impaired cell-mediated immunity as an additional laboratory criteria [2] is unlikely to identify a more homogeneous group, given the similar class frequencies of the Australian sample (the majority of whom had some evidence of impaired cell-mediated immunity during their illness). Rather, a multiplicity of non-specific or ‘atypical’ somatic symptoms in association with higher rates of psychological morbidity may more clearly differentiate patients with a primary somatoform disorder and also predict higher levels of functional disability. This is in keeping with earlier reports [6, 7] that current CFS classificatory systems are unfortunately biased towards the inclusion of such subjects, and emphasizes that only sophisticated forms of substratification are likely to reveal critical patient differences. While our findings suggest aetiological heterogeneity within CFS populations, further studies are needed to determine whether the proposed subgroups lie on a continuum (of severity or chronicity) or truly represent aetiologically distinct groups.
In contrast to the comparable clinical characteristics across the centres, there were significant intersite differences in subclass distributions. In particular, the sample from Belfast was characterized by a very low frequency of class two subjects (6% vs 35% overall, p < 0.001), implying the importance of local selection factors. This factor is generally ignored in multisite international comparison studies. As a consequence of working in very different health systems and of the different clinical orientations of the investigators, the sites did not generate directly comparable patient groups. That is, while clinicians may assume that they are applying current classification systems any resultant patient cohorts are likely to be quite heterogeneous.
It has been common in the medical literature to speak of ‘fibromyalgia’ and ‘CFS’ as different but related conditions. In this study, there were no significant differences in subclass allocations in subjects with a diagnosis of fibromyalgia (68% c1ass one Boston vs 65% overall, p = 0.970). Although, since all these subjects also met Prince Henry criteria for a diagnosis of CFS, this finding adds weight to the view that the two diagnostic labels are largely interchangeable [28]. This has important implications for patient education, clinical management and aetiological research.
This was a highly selected sample, derived from university-based, secondary- and tertiary-referral centres. Although uniform diagnostic criteria were not utilized, the exclusion of patients with concurrent medical disorders (and, to a lesser extent, those with more overt psychiatric disorder) should have reduced the number of subgroups identified. A CFS sample collected from a primary care setting, however, might identify quite different patient subclasses (e.g. a ‘depressive-anxious’ subset may be more likely). It is likely that referral to a university-based research centre has a filtering effect, favouring those with a pattern of excess health care utilization, increased care-eliciting behaviour (thereby increasing the female to male ratio), more chronicity and functional disability. All these factors are likely to increase the chances of detecting somatoform disorders.
Although the use of a self-report methodology may increase the number of positive symptom responses, it is the pattern of symptom response which best helps to identify the somatizing patients. The self-report methodology also allowed for a standardized symptom inventory to be collected internationally, independent of interviewer-induced biases. Data pertaining to antecedent factors should be viewed cautiously, although these data do provide a guide as to the approximate prevalence of possible initiating factors. It must also be noted that psychiatric diagnoses were made using selfreport data only, and therefore should only be seen as approximating prevalence rates able to be determined by structured diagnostic interview.
The present study suggests that CFS research samples are heterogeneous. Attempts to improve case definition by simply refining clinical features already considered characteristic of the disorder will do little to identify patient subgroups. A definition requiring fewer symptoms would be likely to diminish class two size. Alternative strategies such as subtyping and categorizing CFS patients according to presence of a psychiatric diagnosis, hypothalamic-pituitary-adrenal axis activation or immune abnormality may be of use. Future studies to test aetiological or treatment hypotheses should incorporate specific strategies to identify patient subtypes.
Footnotes
Acknowledgements
Dr Wilson was supported by a New South Wales Institute of Psychiatry Research Fellowship and a grant from the Royal Australian and New Zealand College of Psychiatrists Board of Research. The New South Wales Myalgic Encephalomyelitis/Chronic Fatigue Syndrome Society also provided financial assistance.