Symptom Typology and Sub-grouping in Chronic Fatigue Syndrome Megan A. Arroll1*, PhD Victoria Senior1, PhD 1

Department of Psychology University of Surrey Guildford Surrey GU2 7XH Tel: +44 (0) 1483 682884 Fax: +44 (0) 1483 682913 Email: [email protected]

ABSTRACT Background: Chronic Fatigue Syndrome (CFS) is a condition of unknown aetiology with a heterogeneous population. The variability in symptomatology produces difficulties in studying CFS, therefore this study aimed to establish symptom typology and sub-groups within a sample of participants by use of data reduction techniques. Methods: Two-hundred and forty-six participants completed two symptom measures (one of which evaluated CFS-specific symptoms and the other a general symptom checklist) which were subsequently combined and analysed. Symptom types were established with factor analysis, whereas sub-groups within the sample were determined by cluster analysis. Results: Five symptom types resulted from the factor analysis which were labelled FMS-like, depression/anxiety, fatigue/post-exertional malaise, cognitive/neurological and IBS-like symptoms, with the FMS-like accounting for the majority of the variance in the data. Cluster analysis illustrated that the sample could be divided into three sub-groups based upon the symptom reports. The Bulletin of the IACFS/ME. 2009;17(2)

39

clusters that emerged were formed of a low symptomatology sub-group (LSS-G), a medium symptomatology sub-group (MSS-G) and a high symptomatology subgroup (HSS-G), which, as the names suggests, signified symptom severity. Notably, these sub-groups did not differ in respect to age, sex, illness duration or time taken to gain a diagnosis which infers that the groupings were not influenced by demographic concerns. Conclusion: This study illustrated that symptomatology in CFS can be divided into distinct categories that concur with the most recent guidelines for the condition. Additionally, the illness can be separated into discrete sub-groups, although these groupings are linked to overall severity, rather than symptom types.

INTRODUCTION Chronic Fatigue Syndrome (CFS) is a complex and disabling illness of unknown origin and although the overall CFS population appears to be heterogeneous in nature it has been suggested that within this condition specific sub-groups or clusters of patients exist. This idea was echoed in the most recent guidelines for CFS (1) that stated the import of research that investigates possible sub-groups. From a methodological point of view, to ascertain stable groupings within the population would be an asset to the research process as subsequent studies could investigate these distinct, homogenous groups. This would aid replication of studies and hence the validity of the research findings and may lead to the discovery of distinct clinical entities within the CFS population. Statistical procedures have been used to reduce large symptom checklist reports into discrete clusters of symptomatology in fatiguing illnesses and CFS; for instance Nisenbaum and colleagues (2;3) used a factor analysis technique in samples of chronically fatigued individuals that resulted in three-factor solutions (fatigue-moodcognition symptomatology, flu-like symptom and symptoms consistent with visual impairment in the earlier study and musculoskeletal, infection and cognition/mood/sleep symptom in the later work). In order to test the discriminating value of these clusters, hence their utility in identifying individuals with CFS, Nisenbaum et al. (3) investigated the scores from their factor analysis model in subsamples of participants with CFS and those that had chronic fatigue but did not meet all the criteria for diagnosis of CFS and found that the individuals with CFS had higher scores in all three factors, i.e. greater severity in every symptom area. Similarly, in an analysis of 780 chronically fatigued participants from a communitybased sample, four factors were discovered; ‘lack of energy’, ‘physical exertion’, ‘cognitive problems’ and ‘fatigue and rest’ (4). When looking at the utility of these symptom clusters, significant differences were found between participants with CFS, idiopathic chronic fatigue and medically explained chronic fatigue. Differing reports were also found on demographic variables in the overall sample which Bulletin of the IACFS/ME. 2009;17(2)

40

suggests that the separate dimensions were valuable in differentiating symptomatology within fatiguing illnesses. In an investigation of participant groupings within a chronic fatigue sample, rather than symptom types in chronically fatigued participants as above, Jason & Taylor (5) used a hierarchical cluster analysis to cluster the symptom factor scores of 166 medically-evaluated participants. This analysis resulted in a three-cluster solution with clusters labelled ‘low post-exertional fatigue’, ‘high post-exertional fatigue alleviated by rest’ and ‘high post-exertional fatigue not alleviated by rest’ (5). Subsequent investigation of the participants in each cluster showed the majority of CFS participants were defined by high post-exertional fatigue, severe fatigue not alleviated by rest, high levels of cognitive problems and low levels of physical and social functioning. Therefore, CFS appears to be distinct from other types of fatiguing illnesses on the basis of the high symptomatology and low functioning. This finding was similar to that of Nisenbaum et al. (3) where CFS participants differed from participants with chronic fatigue on the basis of severity, indicating that chronic fatigue is a spectrum with CFS at the most severe end of this continuum. Data reduction techniques have also been used in typology and sub-grouping studies of self-reported and medically evaluated CFS participants. In a study with a mixed sample of self-diagnosed patients and physician referrals, principle components analysis demonstrated a three-factor solution consisting of symptom types in terms of cognitive problems, flu-like symptoms and neurologic symptoms (6). Furthermore, in a study with only medically-evaluated CFS participants, six symptom factors emerged from theoretically driven symptom checklists; neurocognitive symptoms, vascular symptoms, symptoms consistent with inflammation, muscle/joint symptoms, infectious symptoms and sleep/postexertional symptoms (7). In a subsequent cluster analysis, four CFS sub-groups were unearthed and differed primarily in terms of symptom severity although there were key differences in reports of inflammatory/infectious symptoms and cognitive problems. In sum, previous research has illustrated the existence of specific symptom types in samples of chronically fatigued individuals and CFS participants. In addition, these symptom types have been compared across differing fatiguing illnesses and within CFS samples which have shown that individuals with CFS report greater symptomatology than those with other types of fatigue. Furthermore, when looking specifically at those with CFS, there appears to be key differences not only in severity levels but also in symptom typology, especially with regard to inflammatory/infectious symptoms and cognitive difficulties. The present study will attempt to add to this body of literature by utilising symptom items not only specific to CFS, but also to include additional symptom items within the data reduction techniques that may have been excluded from previous research. By including a general symptom checklist measure, this study aims to capture a wider range of symptomatology within a self-reported sample of CFS participants and compare typology and sub-grouping findings to previous work. Bulletin of the IACFS/ME. 2009;17(2)

41

METHODS Participants Participants for the present study were recruited through national and local (South West of England) Chronic Fatigue Syndrome/Myalgic Encephalomyelitis Support Groups. General inclusion criteria included an established diagnosis of CFS (this was self-reported and based on the Fukuda criteria (8)) and a minimum age limit of 18 years. Two hundred and forty-six individuals completed the questionnaire. Participants’ ages ranged from 18-80 years, with a mean of 47.82 years (SD = 13.21). The sample consisted of 200 females (81.3%) and 46 males (18.7%). Ninety-five per cent of participants were of White-British origin. One-third of the current sample was unemployed and just over a quarter of participants was retired, with the remaining 40% in part- or full-time employment or education. The average illness duration in the sample was 11.71 years (SD = 8.69), with a range from 1-69 years and the mean time taken to reach a diagnosis of CFS was 3.52 years (SD = 5.88), ranging from 6-months to 58 years. The predominance of females in this study concurs with previous estimates of gender ratio (9;10). However, the racial representation is undoubtedly skewed in favour of White-British which may be due to the fact that participants were recruited from self-help groups rather than randomly sampled. The fairly high rate of participants in employment or in education was contradictory to previous findings (11;12) suggesting that this sample consisted of relatively high-functioning CFS sufferers. The sample’s mean age and range did reflect epidemiological findings (10). Procedure Seventeen local CFS/ME support groups were contacted regarding participation in the study, two of which declined with the remaining 15 agreeing either to send an invitation letter via an email group or post letters directly to their members. Approximately 1,000 invitation letters were sent to members of CFS/ME support groups which included an information sheet and pre-paid envelope. Following an expression of interest in the study, a questionnaire pack and consent form were posted to participants with a pre-paid envelope; participants were allowed to complete the questionnaire pack in their own time. Three-hundred and twenty-four replies were received and 246 completed questionnaires were returned illustrating a 76% response rate within the group that replied to the invitation letter, although there was only about a quarter total response from the initial invitation letter.

Bulletin of the IACFS/ME. 2009;17(2)

42

Measures The Profile of Fatigue-Related Symptoms (PFRS (13;14) is a symptom measure specifically designed for use within CFS/ME research and consists of four subscales (emotional distress, cognitive difficulties, fatigue and somatic symptoms). It is scored on a 7-point scale with 0 representing ‘not at all’ to 6 which signifies ‘extremely’. Generally, the sub-scales are scored by summing the responses and producing a mean value for the individual sub-scale, but in this instance the scores were kept as total sums to be more in line with the PILL (see below). Although the instrument originally asked participants to complete the items based on the past week, here this has been extended to the past month to create uniformity among the measures used in the questionnaire pack. Internal consistencies and test-retest reliability have been shown to be high within the PFRS (13). This instrument was chosen as does not suffer from a low ceiling effect compared to other CFS measures and as such allows for a more accurate representation of the variability in CFS symptomatology (15). The Pennebaker Inventory of Limbic Languidness (PILL (16) consists of 54 symptoms and was designed to evaluate symptom reporting. In the original scale each symptom was rated on a 5-point scale from ‘have never or almost never experienced the symptom’ to ‘experienced the symptom more than once every week’, however for the purpose of this study the scale was altered to concur with the PFRS scale. Therefore, the scale was scored on a 7-point scale with 0 representing ‘not at all’ to 6 which signifies ‘extremely’. There are two methods of scoring the measure; summing all the items and summing just the high scoring items (i.e. C, D and E on the scale from A to E with E the most frequent option). The present study used the former method. The PILL has been validated in a number of laboratory studies where high PILL scores were associated with greater physical sensations and has also been shown to correlate with numerous illness behaviours (16). This measure was included in addition to the PFRS to capture the wide range of symptoms that have been documented in patient narratives (17).

RESULTS Principle component analysis In an attempt to account for the specific CFS symptoms, as included in the PFRS, as well as more general symptoms, a principle components analysis was carried out on the merged PFRS and PILL. This type of analysis was utilised as a means of scaling down a relatively large number of variables into a small number of components or factors. To improve the interpretation of the results, oblique rotation (direct oblimin) was used in this analysis as it allows the factors to correlate. Another important issue that was considered before this analysis was carried out was sample size. A sample size of 200-300 is believed to be ‘fair’ for principle Bulletin of the IACFS/ME. 2009;17(2)

43

components analysis (18); however Tabachnick & Fidell (19) state that as a general rule of thumb, it is best to have a sample size of at least 300 cases. Furthermore, another rule of thumb that states one should have 5-10 participants for each variable in the analysis if the sample size is below 300 cases (20). The current analysis contains 246 cases and 96 symptom variables, hence not quite reaching the 300 figure and with many variables compared to cases. But, according to MacCallum et al. (21) it is the magnitude of structural coefficients along with the sample size that is important and if the analysis renders factors with four or more loadings of .60 or more then the factors should be reliable despite the total sample size. As can be seen in Table 3, every factor apart from the FMS factor contained four or more loadings above .60. In fact, the only item in the FMS factor with a loading below .60 was ‘stiff joints’ with a loading of .54, i.e. approaching .60. Another basis for carrying out the analysis with the current sample size was that the Kaiser-Meyer-Olkin measure of sampling adequacy (KMO, (22)) had a value of .89 (i.e. close to 1) indicating that the correlations in this analysis were compact and hence the resulting factors should be discrete and reliable (20). Bartlett’s test of sphericity proved to be significant (p < 0.001) indicating that there are relationships between the symptom scores. Also, as there were only 4 variable communalities below .60 (‘strong reactions to insect bites’, ‘insomnia’, ‘your face flushes’, and ‘sensitive to alcohol’ with values of .52, .54, .59 and .59, respectively) and the factor solution contained a relatively low number (five) with a small number of indicator variables in each (the greatest being ‘fatigue/post-exertional malaise’ with 12 indicators) the sample size was deemed as acceptable for principle components analysis (21;23). To determine the number of meaningful factors produced in the analysis, a multimethod approach was used. Initially, the scree plot was investigated and this indicated five components in the data set. In addition, the individual Eigen values were all above 1 but as the sample size was less than 250 and there were more than 30 variables in the analysis this method of extracting factors, known as Kaiser’s criterion, was not ideal (20). Therefore, a random data set was generated and an identical principal components analysis was carried out on the resulting numbers. To achieve this, the sample data set was compared to an equivalent sized data set consisting of normally distributed random numbers (produced by via SPSS (24) syntax). The principal components analysis was then conducted on both data sets and the Eigen values associated with the resulting factors compared; only the factors where the Eigen value in the sample data set exceeded the values in the random data set were extracted. This parallel approach confirmed that five factors could be confidently obtained as the Eigen values in the present data exceeded those in the random set. Factor 1 consisted of four items originating from the PILL; ‘sore muscles’, ‘stiff muscles’, ‘tender muscles’ and ‘stiff joints’ which relate to physical manifestations consistent with Fibromyalgia Syndrome (FMS), hence were labelled FMS-like. In a reliability analysis, every item in the FMS-like factor had item total correlations Bulletin of the IACFS/ME. 2009;17(2)

44

above .4 indicating that these items correlated with the overall score from the scale (20); the Cronbach’s alpha in this scale was within the acceptable range at .85 (25). Factor 2 was a condensed version of the PFRS ‘emotional distress’ sub-scale that was more related to depression and anxiety than other types of distress such as anger and annoyance which were included in the original PFRS sub-scale. The items in this factor were ‘feeling sad’, ‘feeling pessimistic’, ‘feeling depressed’, ‘feeling nervous’, ‘feeling worthless’, ‘feeling anxious’, ‘feeling tense’, ‘tearfulness’ and ‘worrying unnecessarily’ and was named depression/anxiety. This second factor again had high correlations between each item and the total sub-scale score ranging from .65 to .80 and a high alpha of .93. None of the items if deleted would increase the reliability statistic therefore the sub-scale remained the same with nine items. Factor 3 was identical to the PFRS ‘fatigue’ sub-scale and consisted of items relating to lack of energy and post-exertional malaise and as such was labelled fatigue/post-exertional malaise (see Table 1 for the full list of items). The eleven items in the fatigue/post exertional malaise factor had high item-total correlations (from .56 to .79) and a high Cronbach’s alpha of .93. Factor 4 contained the PFRS sub-scale of ‘cognitive difficulties’ which included the items ‘slow thought’, ‘forgetting conversation’, ‘mental fog’, ‘difficulty reasoning’, ‘difficulty finding words’, ‘absent-minded’, ‘difficulty understanding’, ‘difficulty concentrating’, ‘difficulty remembering’ and ‘difficulty following plots’ and was titled cognitive/neurological. This factor also retained all of its ten items and had a very high alpha of .94 with every item correlating with the overall total score at .70 or more. The final factor in this analysis was generated from two PFRS items (‘stomach pain and ‘diarrhoea/constipation’ from the ‘somatic symptoms’ sub-scale) and four items from the PILL (‘upset stomach’, ‘indigestion’, ‘heartburn’ and nausea’). This factor clearly comprised of gastric and irritable bowel symptomatology, thus was named IBS-like, had a Cronbach’s alpha of .87 and preserved its six items and had itemtotal correlations ranging between .60 and .75. Therefore, all of the new sub-scales had very respectable internal consistency in this sample. The sub-scales were not highly related to each other, with the highest correlation .35 between the fatigue/post-exertional malaise factor and the cognitive/neurological factor. Interestingly, this correlation, along with the correlations between the FMS-like factor and every other factor except cognitive/neurological was negative illustrating that as these symptoms increased the symptoms on the opposing factors decreased. None of the correlations were significant suggesting that each scale was measuring distinct symptom groups. The factor FMS-like accounted for the majority of the variance in the data (25.10%) with the factors depression/anxiety, fatigue/post-exterional malaise, cognitive/neurological and IBS-like accounting for 7.64%, 5.60%, 3.44% and 2.64%, Bulletin of the IACFS/ME. 2009;17(2)

45

respectively. Thus, the overall five factor solution accounted for 44.4% of variance. This model was a good fit as only 7% of the residuals had a value of .05 or more indicating the predicted correlation coefficients were very close to the observed coefficients.

Bulletin of the IACFS/ME. 2009;17(2)

46

Cluster analysis To investigate whether the factor scores could be a valuable way of classifying participants into sub-groups, a cluster analysis was performed. Total scores were calculated for each factor by simply summing the individual items in each of the five factors before a hierarchical agglomerative cluster analysis was used. Squared Euclidean distance was employed as the similarity measure to account for elevation of scores, i.e. those who scored highly were clustered with others who scored in this direction (26). This was deemed the most appropriate method as the aim of the analysis was to obtain symptomatology clusters based on reporting and severity of symptoms. Investigation of the dendrogram revealed six clusters although subsequent investigation of the agglomeration schedule revealed large increases in the coefficients in only three clusters. This 3-cluster solution included all of the Bulletin of the IACFS/ME. 2009;17(2)

47

participants, therefore the conservative 3-cluster solution was accepted and this number was used in a K-means cluster analysis to optimize the findings. Sixtyseven participants were contained in cluster 1, 86 in cluster 2 and 93 in cluster 3. Characteristics of clusters Cluster 1 contained participants with low scores in the FMS-like, fatigue/postexertional, cognitive/neurological, IBS-like symptom factors but had average scores on the depression/anxiety scale. Therefore, this cluster was labelled the ‘low symptomatology sub-group’ (LSS-G). Cluster 2 consisted of participants who reported high scores in all of the five factor scales; hence it was named the ‘high symptomatology sub-group’ (HSS-G). Finally, cluster 3 comprised of participants who stated average levels of symptoms on the FMS-like, fatigue/post-exertional, cognitive/neurological, IBS-like factors and low scores on the depression/anxiety scale compared to the group as a whole; therefore, this group was defined as the ‘medium symptomatology sub-group’ (MSS-G). A series of one-way analysis of variance tests was conducted to confirm that there were significant differences between the sub-groups on the symptom factors (Table 2). The factor cognitive/neurological did not meet the requirement of homogeneity of variance therefore the Welch’s F statistic was quoted in this analysis. Also, to make comparisons between the cluster groups in the variable ‘sex’ a chi-square test was used as the data was categorical.

Total symptom scores differed significantly between the sub-groups (F(2,243) = 139.54, p < 0.001) and post-hoc analyses revealed significant differences in mean scores between every group at the 0.001 level with a Bonferroni correction in the expected direction (i.e. significantly higher rates of symptoms in the HSS-G, Bulletin of the IACFS/ME. 2009;17(2)

48

followed by the MSS-G and the LSS-G). There were also significant differences in every symptom factor scale across every group (in the expected direction) except in the depression/anxiety factor between the LSS-G and the MSS-G. Furthermore, there were not significant differences between the sub-groups on any of the demographic variables which suggests that the sub-groups identified illness severity independent of demographic characteristics.

DISCUSSION This study investigated symptomatology and sub-grouping within a sample of selfreported CFS participants by using two data reduction techniques; principle components analysis and cluster analysis. The former method scaled-down the data set into five distinct factors pertaining to FMS-like, depression/anxiety, fatigue/post-exertional malaise, cognitive/neurological and IBS-like symptoms. The FMS-like factor accounted for most of the variance in the analysis which was consistent with findings from Nisenbaum et al. (3) although, unlike their study, the present findings illustrated low inter-factor correlations of -.35 and under suggesting that the factors here were more independent of one another. Unlike the findings of both Friedberg et al. (2) and Nisenbaum et al. (3), a flu-like or infection factor was not unearthed here. This was not due to the items that were entered into the analysis as the PILL has numerous symptoms relating to these categories (e.g. sneezing, coughing, sore throat). Further investigation concerning the onset of illness in the present sample may account for this difference. With regards to how the symptom factors related to the most recent diagnostic and research criteria for CFS (1), the findings here partially supported the view of six distinct categories of symptoms (fatigue, post-exertional malaise, pain, cognitive/neurological, sleep dysfunction and autonomic/neuroendocrine/immune manifestations). The first four Canadian criteria clearly map onto the FMS-like, fatigue/post-exertional malaise, cognitive/neurological factors found here and the autonomic/neuroendocrine/immune manifestations criterion is very closely related to the IBS-like factor. However, the sleep dysfunction criterion has not been accounted for here as only one symptom item related to this difficulty. Additionally, the Canadian guidelines (1) do not include an anxiety/depression category and this factor scale undoubtedly resulted from the items contained in the analysis (specifically the PFRS) and may have been due the self-reported, heterogeneous nature of the present sample. Future research utilising data reduction techniques may investigate this issue further by including the full list of symptom items consistent with the Canadian guidelines. The cluster analysis in this study illustrated that the sample could be divided into three sub-groups based upon symptomatology. The clusters that emerged were formed of a low symptomatology sub-group (LSS-G), a medium symptomatology sub-group (MSS-G) and a high symptomatology sub-group (HSS-G), which, as the Bulletin of the IACFS/ME. 2009;17(2)

49

names suggests, were severity groups within CFS. Notably, these sub-groups did not differ in respect to age, sex, illness duration or time taken to gain a diagnosis which infers that the groupings were not influenced by demographic concerns. This study followed procedures previously carried out by Nisenbaum and colleagues (2;3) and Jason and colleagues (4;5;7) in the sense that large amounts of symptom data were reduced to interpretable factor scores. However, in this study CFS-specific symptoms and general symptoms were included in the symptom checklist to account for complaints that may have been over-looked with the sole inclusion of CFS measures. Nevertheless, the outcome of the cluster analysis was in line with previous findings in the sense that participants were grouped on the basis of severity. This supports the theory that CFS is the severe end of a fatigue spectrum and is in accord with previous research that has shown that individuals who have been empirically classified as CFS had more severe fatigue, greater impairment and more frequent and severe minor symptoms than individuals categorised as suffering from medically unexplained fatigue that is not CFS (27). The sampling technique in this study was not as strong as that of Jason et al. (7); hence the sample can only be defined as ‘self-reported CFS’, not medically evaluated CFS as participants responded to requests for participants, were not randomly sampled from the community and did not go through an extensive screening process. This self-selection process (only about a quarter of individuals who were invited to take part did so) will have affected the results as those that took part must have had a certain level of functioning to do so, i.e. the very severely afflicted would not have had the capacity to take part. Additionally, there may have been characteristics of those who volunteered that differed from the general CFS population not solely because they had volunteered but also as they were approached via self-help groups; for instance there is a racial skew in favour of White-British participants and this is not representative of the CFS population (10). Another difficulty with this sample concerns co-morbidity; some of the participants may have been suffering from CFS with co-morbid conditions such as FMS, IBS and/or depression which would account for the corresponding factors in the analysis. There is also the chance that some of the participants may not have had CFS at all and therefore future work should concentrate on medically defined patient groups. This more heterogeneous sample may have led to the relatively small percentage of variance that was accounted for by the fatigue factor. Finally, future studies should attempt to gain access to larger samples for data reduction analyses so that the results may be considered with more confidence.

Acknowledgements This study was funded in part by ME Solutions, a registered UK charity, and Network MESH which supports those with ME/CFS. Bulletin of the IACFS/ME. 2009;17(2)

50

Reference List (1) Carruthers BM, Jain AK, de Meirleir KL, Peterson DL, Klimas NG, Lerner AM, Bested AC, Flor-Henry P, Joshi P, Powles A-CP, Sherkey JA, van de Sande MI. Myalgic Encephalomyelitis/Chronic Fatigue Syndrome: Clinical Working Case Definition, Diagnostic and Treatment Protocols. Journal of Chronic Fatigue Syndrome 2003;11(1):2003, 7-2003115. (2) Nisenbaum R, Reyes M, Mawle AC, Reeves WC. Factor analysis of unexplained severe fatigue and interrelated symptoms. American Journal of Epidemiology 1998;148:72-7. (3) Nisenbaum R, Reyes M, Unger ER, Reeves WC. Factor analysis of symptoms among subjects with unexplained chronic fatigue: What can we learn about chronic fatigue syndrome? Journal of Psychosomatic Research 2004;56(2):171-8. (4) Jason LA, Taylor RR, Kennedy CL, Jordan K, Huang CF, Torres-Harding S, Song S, Johnson D. A factor analysis of chronic fatigue symptoms in a community-based sample. Social Psychiatry and Psychiatric Epidemiology 2002;37(4):183-9. (5) Jason LA, Taylor RR. Applying cluster analysis to define a typology of chronic fatigue syndrome in a medically-evaluated, random community sample. Psychology and Health 2002;17(3):323-37. (6) Friedberg F, Dechene L, McKenzie MJI, Fontanetta R. Symptom patterns in long-duration chronic fatigue syndrome. Journal of Psychosomatic Research 2000;48(1):59-68. (7) Jason LA, Corradi K, Torres-Harding S. Toward an empirical case definition of CFS. Journal of Social Service Research 2007;34:43-54. (8) Fukuda K, Straus SE, Hickie I, Sharpe MC, Dobbins JG, Komaroff A. The chronic fatigue syndrome: A comprehensive approach to its definition and study. Annals of Internal Medicine 1994;121:953-9. (9) Jason LA, Richman JA, Rademaker AW, Jordan KM, Plioplys AV, Taylor RR, McCready W, Huang CF, Plioplys S. A community-based study of chronic fatigue syndrome. Annals of Internal Medicine 1999;159:2129-37. (10) Taylor RR, Jason LA, Richman JA, Torres-Harding SR, King C, Song S. Epidemiology. In: Jason LA, Fennell PA, Taylor RR, editors. Handbook of chronic fatigue syndrome.Hoboken, N.J.: Wiley; 2003. p. 3-25. (11) Taylor RR, Kielhofner GW. Work-related impairment and employmentfocused rehabilitation options for individuals with chronic fatigue syndrome: A review. Journal of Mental Health (UK) 2005;14(3):Jun-267. (12) Vercoulen J-HMM, Swanink C-MA, Fennis J-FM, Galama J-MD. Prognosis in chronic fatigue syndrome: A prospective study on the natural course. Journal of Neurology, Neurosurgery and Psychiatry 1996;60(5):489-94. (13) Ray C, Weir WR, Phillips S, Cullen S. Development of a measure of symptoms in chronic fatigue syndrome: The Profile of Fatigue-Related Symptoms (PFRS). Psychology and Health 1992;7(1):27-43.

Bulletin of the IACFS/ME. 2009;17(2)

51

(14) Ray C, Weir WR, Cullen S, Phillips S. Illness perception and symptom components in chronic fatigue syndrome. Journal of Psychosomatic Research 1992;36(3):243-56. (15) Stouten B. Identification of ambiguities in the 1994 chronic fatigue syndrome research case definition and recommendations for resolution/. BMC Health Services Research 2005;5(37):http://www.pubmedcentral.nih.gov/picrender.fcgi?artid=1175848&blobty pe=pdf. (16) Pennebaker JW. The psychology of physical symptoms. New York: Springer; 1982. (17) Arroll MA, Senior V. Individuals' Experience of Chronic Fatigue Syndrome/ Myalgic Encephalomyelitis: An Interpretative Phenomenological Analysis. Psychology and Health 2008;23:443-58. (18) Comrey RD, Lee HB. A first course in factor analysis. 2nd ed. Hillsdale, NJ: Erlbaum; 1992. (19) Tabachnick BG, Fidell LS. Using multivariate statistics. 4th ed. Needham Heights, MA: Allyn & Bacon; 2001. (20) Field A. Discovering statistics using SPSS. London: Sage; 2005. (21) MacCallum RC, Widaman KF, Zhang S, Hong S. Sample size in factor analysis. Psychological Methods 1999;4:84-99. (22) Kaiser HF. An index of factorial simplicity. Psychometrika 1974;39:31-6. (23) Guadagnoli E, Velicer W. Relation of sample size to the stability of component patterns. Psychological Bulletin 1988;103:265-75. (24) SPSS for Windows [computer program]. Version 13.0. Chicago: SPSS Inc.; 2002. (25) Kline P. The handbook of psychological testing. 2nd ed. London: Routledge; 1999. (26) Clatworthy J, Buick D, Hankins M, Weinman J, Horne R. The use and reporting of cluster analysis in health psychology: A review. British Journal of Health Psychology 2005;10:329-58. (27) Reeves WC, Wagner D, Nisenbaum R, Jones JF, Gurbaxani B, Solomon L, Papanicolaou DA, Unger ER, Vernon SD, Heim C. Chronic fatigue syndrome - A clinically empirical appraoch to its definition and study. BMC Medicine 2005;3(19):http://www.biomedcentral.com/content/pdf/1741-7015-3-19.pdf.

Bulletin of the IACFS/ME. 2009;17(2):39-52. © 2009 IACFS/ME

Bulletin of the IACFS/ME. 2009;17(2)

52