Article Text

Download PDFPDF

Phenotypes of chronic fatigue syndrome in children and young people
  1. Margaret May1,
  2. Alan Emond2,
  3. Esther Crawley2
  1. 1Department of Social Medicine, Bristol University, Bristol, UK
  2. 2Centre for Child and Adolescent Health, Bristol University, Bristol, UK
  1. Correspondence to Dr Esther Crawley, Centre for Child and Adolescent Health, Hampton House, Cotham Hill, Bristol BS6 6JS, UK; esther.crawley{at}bristol.ac.uk

Abstract

Objective To investigate the heterogeneity of chronic fatigue syndrome (CFS/ME) in children and young people.

Setting Regional specialist CFS/ME service Patients Children and young people aged <19 years old.

Methods Exploratory factor analysis was performed on symptoms present at assessment in 333 children and young people with CFS/ME. Linear and logistic regression analysis of data from self-completed assessment forms was used to explore the associations between the retained factors and sex, age, length of illness, depression, anxiety and markers of severity (fatigue, physical function, pain and school attendance).

Results Three phenotypes were identified using factor analysis: muscoloskeletal (factor 1) had loadings on muscle and joint pain and hypersensitivity to touch, and was associated with worse fatigue (regression coefficient 0.47, 95% CI 0.25 to 0.68, p<0.001), physical function (regression coefficient −0.52, 95% CI −0.83 to −0.22, p=0.001) and pain. Factor 2 (migraine) loaded on noise and light hypersensitivity, headaches, nausea, abdominal pain and dizziness and was most strongly associated with physical function and pain. Sore throat phenotype (factor 3) had loadings on sore throat and tender lymph nodes and was not associated with fatigue or pain. There was no evidence that phenotypes were associated with age, length of illness or symptoms of depression (regression coefficient for association of depression with musculoskeletal pain −0.02, 95% CI −0.27 to 0.23, p=0.87). The migraine phenotype was associated with anxiety (0.40, 95% CI 0.06 to 0.74, p=0.02).

Implications CFS/ME is heterogeneous in children with three phenotypes at presentation that are differentially associated with severity and are unlikely to be due to age or length of illness.

Statistics from Altmetric.com

Introduction

Chronic fatigue syndrome or myalgic encephalopathy (CFS/ME) is defined by the Royal College of Paediatrics and Child Health (RCPCH) in the UK as ‘generalised fatigue, causing disruption of daily life, persisting after routine tests and investigations have failed to identify an obvious underlying cause’.1 The National Institute for Health and Clinical Excellence recently recommended that a minimum of 3 months of fatigue is required before a diagnosis of CFS/ME is made in children.2

CFS/ME in children is a relatively common3,,6 and potentially serious condition with over 50% of children bed bound at some stage and a mean time off school of one academic year.7 However, despite the publication of three adult and one paediatric consensus case definitions,1 attempts to identify a single cause for CFS/ME have failed. One reason for this might be that it is a heterogeneous condition and indeed there is considerable support for this in adult patients with CFS/ME.8,,13

What is already known on this topic

  • CFS/ME is common in children and young people and causes significant morbidity including time off school.

  • No single cause for CFS/ME has been discovered.

What this study adds

  • CFS/ME in children and young people is a heterogeneous condition.

  • There are potentially three disease phenotypes at presentation, labelled musculoskeletal, sore throat and migraine.

  • Both the musculoskeletal and migraine phenotypes are strongly associated with fatigue, physical function and pain.

Factor analysis is a statistical technique that exploits the correlations between measured variables (in this example symptoms) to express the data as a reduced number of uncorrelated factors that are linear combinations of the observed variables and which account for most of the variation in the data. The resulting factors can be used to understand more about the heterogeneity of an illness and to define groups of children with different phenotypic presentation.

This study investigated whether distinct phenotypes of paediatric CFS/ME could be identified at presentation and whether these were associated with markers of severity.

Methods

Patient cohort

The Bath specialist paediatric CFS/ME service covers a region in the south west of England with a population of some 400 000 children aged 5–19 (2001 census). Children and young people are assessed and offered treatment in outpatient clinics unless they are severely affected in which case they are seen at home. Data were collected on all children assessed between April 2005 and November 2008, who were under 19 years of age at the time of the assessment and were given a clinical diagnosis of CFS/ME using the RCPCH definition.

Inventories

Prior to clinical assessment, inventories measuring fatigue, physical function and pain were collected. Fatigue was measured using the 11-item Chalder Fatigue Scale which was scored using the 0–3 method for scoring each question (0 for ‘less than usual’, 1 ‘no more than usual’, 2 ‘for more than usual’ and 3 for ‘much more than usual’).14 Physical function was measured using the 10-item physical function subscale of the SF36. This is scored between 1 (‘yes limited a lot’) and 3 (‘no not limited at all’) for each question so that children with the worst physical function scored 10 while those with good physical function scored 30. A visual analogue scale was used to measure pain, with a score of 0 for ‘no pain’ and 100 for ‘pain as bad as possible.’ The Hospital Anxiety and Depression Scale was used to screen children aged 14 years and older for low mood and anxiety symptoms.15 School attendance was collected as a single item inventory as the percentage of possible time at school.

Inventories were coded as missing if more than one question was missing. On the Hospital Anxiety and Depression Scale, each seven item subscale for anxiety and depression was coded as missing if there was more than one question missing. Questions for which two answers were given were coded as missing. Total scores were corrected for the number of missing items.

At assessment children were asked by the clinician about the presence or absence of symptoms (producing a binary measure) using a symptom list derived from the Centres for Disease Control and Prevention criteria and the RCPCH guidelines.1 16

The clinician recorded symptoms as being present (no/yes) if they significantly affected the child or young person, were present most or all of the time and had only been there since the child or young person had become unwell. The symptom list included subjective memory impairment, sore throat, tender lymph nodes, muscle pain, joint pain, headaches, unrefreshing sleep, post-exertional malaise, abdominal pain, nausea, hypersensitivity to touch, hypersensitivity to noise, hypersensitivity to light and dizziness.

The symptoms ‘post exertional malaise’ and ‘unrefreshing sleep’ were excluded from the analysis as they were present in 96% and 94% of children and young people respectively and were therefore not discriminatory for this analysis leaving 12 symptoms included in the analysis.

Statistical methods

Factor analysis

We used factor analysis to reduce the symptom list (using the correlation between occurrence of symptoms) to a smaller number of independent continuous latent factors that hypothetically might correspond to pathological disease processes.

An exploratory factor analysis of the correlation matrix of the symptoms of CFS/ME was performed using Mplus V.5. This software uses maximum-likelihood algorithm to fit factor analysis to binary observed data (the symptoms) using probit regression. We used a varimax rotation of the factor loadings to enhance interpretability. In the full model, there is the same number of factors as symptoms. However, the usefulness of factor analysis as a data reduction technique is based on retaining only the first few factors, which are ranked according to the proportion of the variance in the data that is explained by each factor, to produce a more parsimonious model. We chose the number of factors (and therefore the number of underlying latent factors) to retain in the model using the Akaike information criterion (AIC). The AIC is a measure of statistical fit to the data that penalises the number of parameters in the model and hence results in a more parsimonious model than a full model. A series of factor models was fitted to the data, starting with the full factor model and reducing the number of retained factors by one in each subsequent model until all possible models had been fitted. We selected the model with the lowest value of AIC, which is the model with the best statistical properties in terms of trade-off between fit to the data and parsimony.

The coefficients of the factor model are called loadings and range from –1 to +1 and are used to estimate weights to multiply the symptoms (0=absent, 1=present) to produce a score for each factor for each child. In order to interpret the factors as latent factors associated with the list of symptoms, only factors that contributed substantially to the variance and had high loadings on individual symptoms were retained in the model. We did this by fitting a confirmatory factor analysis in which the factors were associated only with symptoms with magnitude of loadings greater than 0.3, as loadings have been considered meaningful when their magnitude is greater than 0.3 or 0.4.17 We applied this model to the children to see if the latent factors were associated with measures of disease severity.

Regression of factors on demographic variables and measures of severity

We used linear regression for continuous variables and logistic regression for binary variables to explore the associations between the retained factors from the selected model and sex, age (years), length of illness (months) and markers of severity (total fatigue score and SF36 score). To make the regression coefficients for the different inventories comparable they were rescaled so that the range for each was approximately 10. Thus, the SF36 score was divided by 2, and fatigue score was divided by 3. Factors were standardised to have mean 0 and SD1. All analyses were restricted to children with no missing data in any of the variables investigated. Regression analyses were performed using Stata V.9.2.

Ethics

The North Somerset & South Bristol Research Ethics Committee decided that the collection and analysis of these data were part of service evaluation and as such did not require ethical review by a NHS Ethics committee or approval from the NHS R&D office.

Results

Between April 2005 and November 2008, 374 children and young people under the age of 19 were assessed and given a diagnosis of CFS/ME using the RCPCH criteria. Of those seen, 41 children were excluded because they had 10 or more out of 12 possible symptoms and therefore were not discriminatory leaving 333 children for inclusion in the factor analysis. Table 1 shows characteristics of the 333 children who were included. The median age was 14.9 years (minimum age 2 years 2 months, maximum age 18 years 11 months) and 69.4% were female. Median length of illness (onset of symptoms to assessment) was 18 months. Most of the children were severely disabled by fatigue, with a median attendance at school of 40% of maximum possible.

Table 1

Characteristics of 333 children and young people with CFS/ME included in analysis

Factor analysis and identification of phenotypes

Initial factor analysis demonstrated that CFS/ME is a heterogeneous disease with either a three or four latent factors (AIC for the 2, 3 and 4 factor models were 170, 136.4 and 107.4, respectively). The 4 factor solution was rejected because the fourth factor had no positive loadings with magnitude >0.3. Table 2 shows the three factors with the loadings on the symptoms. Factor 1 had loadings on muscle and joint pain and was therefore called musculoskeletal. Factor 2 had loadings on headaches, abdominal pain, nausea and dizziness as well as noise, light and touch hypersensitivity, and was called migraine. Factor 3 had positive loadings on sore throat and tender lymph nodes and was called sore throat.

Table 2

Factor analysis of symptoms with three retained factors

Association of phenotypes with demographic variables and measures of severity

The musculoskeletal phenotype was characterised by stronger association with fatigue compared with other phenotypes (table 3). The sore throat phenotype appeared to be the least severely affected group with no association with fatigue and pain and less strong association with physical function. Interestingly, this phenotype was also associated with female gender. The migraine phenotype had the strongest association with lower physical function and worse pain compared with other phenotypes. Unsurprisingly, this phenotype was also associated with lower school attendance.

Table 3

Regression coefficients (95% CI) for the association of gender, age at clinic, length of illness, school attendance, fatigue, physical function (SF36 score) and pain with each of the three factors adjusted for age and sex

The phenotypes were not associated with the age of the child. None of the phenotypes were associated with length of illness, suggesting that they are not merely due to de-conditioning.

Association of phenotypes with symptoms of anxiety and depression

There was no evidence that symptoms of depression were associated with any of the phenotypes after adjustment for other variables (table 4). However, there was some evidence that symptoms of anxiety were associated with migraine phenotype.

Table 4

Regression coefficients (95% CI) for the association of symptoms of depression and anxiety with each of the factors adjusted for age, sex, fatigue, physical function and pain (n=173)

Discussion

We have shown that the presentation of CFS/ME is heterogeneous in children and young people with three phenotypes that can be clearly differentiated from each other. Each phenotype was differentially associated with markers of severity. Musculoskeletal (factor 1) and migraine (factor 2) phenotypes were associated with worse fatigue, physical function and pain while the sore throat phenotype (factor 3) was not associated with fatigue and pain. The migraine phenotype (factor 2) was associated with worse school attendance. None of the phenotypes were associated with age or length of illness suggesting that they did not merely reflect de-conditioning. The lack of association with symptoms of depression or anxiety (with the exception of the migraine phenotype) suggests that they are unlikely to reflect mood disorders.

Strengths and weaknesses

This is the first study to describe the heterogeneity of phenotypic presentation of CFS/ME in children. Data were collected prospectively on a large well described, unselected group of children and young people attending a paediatric CFS/ME clinic. Symptom histories were collected during the initial assessment by clinicians using a symptom checklist. However as the cohort for this study was derived from a specialist CFS/ME clinic it is likely that these children were more severely affected than those attending a general paediatric clinic might be and that therefore the phenotypic factors described might not be generalisable to all cases of CFS/ME.

Factor analysis cannot prove the existence of separate subdivisions within a disease but merely suggest that subdivisions may exist given the list of symptoms analysed. The interpretation of the factor analysis makes an assumption that the underlying latent variables, or factors, group the measured variables (symptoms in this case) such that they represent phenotypic presentations of disease. This means that the analysis is dependent on the list of symptoms chosen. It is, however, an important method to generate hypotheses about potential disease mechanisms that may explain the heterogeneity of the phenotypic disease presentation in children. However, it is reliant on the number and nature of the symptoms available for analysis. Different latent factors of disease presentation might have been arrived at if alternative sets of symptoms had been collected.

Results in context with previous literature

This study is consistent with studies in adult patients with CFS/ME which have demonstrated that CFS/ME is heterogeneous in presentation and have described phenotypes which are not exclusively a marker of severity.8,9,18,19 We were unable to compare the phenotypes described in our study with those described in adult populations because the symptom lists and measures used were not sufficiently similar to allow meaningful comparison.

Previous studies in adults have shown that length of time of illness was related to different phenotypes in adults,13 20 but this does not appear to be the case in our cohort. In addition, our phenotypes (factors) were not associated with symptoms of depression nor anxiety (with the exception of migraine phenotype) once the analysis adjusted for markers of severity, age and gender unlike other studies in adults which have shown an association between phenotypes and psychiatric morbidity.12 13

Implications

Paediatricians who see children with CFS/ME need to appreciate that it is a heterogeneous condition with phenotypic presentations differentially associated with markers of severity. Children who present to paediatricians with the musculoskeletal phenotype (muscle pain and joint pain) or the migraine phenotype are likely to be more severely affected by fatigue have worse physical functioning and greater pain and may require more intensive rehabilitation. In contrast, children presenting with the sore throat phenotype appear to be less severely affected, can be expected to recover faster and may require less intensive rehabilitation.

Understanding more about the heterogeneity and different phenotypic presentations in paediatric CFS/ME is important for developing appropriate treatment protocols and generating hypothesis about underlying aetiology. Future work will examine whether the phenotypes described here can be used to prospectively define clinically useful groupings of children with different phenotypic disease presentation that have different prognoses and may require appropriately tailored treatments.

Acknowledgments

Thank you to the young people and their families who took part.

References

Footnotes

  • Funding We are grateful to The Linbury Trust who funded this study.

  • Competing interests Esther Crawley is a medical advisor for the Association of Young People with ME.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

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.