OUP user menu

Chronic fatigue following infection by Coxiella burnetii (Q fever): ten‐year follow‐up of the 1989 UK outbreak cohort

M.J. WILDMAN, E.G. SMITH, J. GROVES, J.M. BEATTIE, E.O. CAUL, J.G. AYRES
DOI: http://dx.doi.org/10.1093/qjmed/95.8.527 527-538 First published online: 1 August 2002

Abstract

Background: Some patients exposed to Q fever (Coxiella burnetii infection) may develop chronic fatigue.

Aim: To determine whether subjects involved in the West Midlands Q fever outbreak of 1989 had increased fatigue, compared to non‐exposed controls, 10 years after exposure.

Design: Matched cohort study comparing cases to age‐, sex‐ and smoking‐history‐matched controls not exposed to Q fever.

Methods: A postal questionnaire was sent to subjects at home, followed by further assessment in hospital, including a physical examination and blood tests.

Results: Of 108 Q‐exposed subjects, 70 (64.8%) had fatigue, 37 idiopathic chronic fatigue (ICF) (34.3%), vs. 29/80 (36.3%) and 12 (15.0%), respectively, in controls. In 77 matched pairs, fatigue was commoner in Q‐exposed subjects than in controls: 50 (64.9%) vs. 27 (35.1%), p<0.0001. ICF was found in 25 (32.5%) of Q‐exposed patients and 11(14.3%) of controls (p=0.01). There were 36 (46.8%) GHQ cases in Q‐exposed subjects, vs. 18 (23.4%) controls (p=0.004). A matched analysis of those more intensively studied showed fatigue in 48 (66.7%) Q‐exposed patients and 25 (34.7%) controls, (p<0.0001), ICF in 25 (34.7%) Q‐exposed and 10 (13.9%) controls (p=0.004), and chronic fatigue syndrome (CFS) in 14 (19.4%) Q‐exposed patients and three (4.2%) controls (p=0.003). Thirty‐four (47.2%) Q‐exposed patients were GHQ cases compared to 17 (23.6%) controls (p=0.004).

Discussion: Subjects who were exposed to Coxiella in 1989 had more fatigue than did controls, and some fulfilled the criteria for CFS. Whether this is due to ongoing antigen persistence or to the psychological effects of prolonged medical follow‐up is uncertain.

Introduction

Q fever is an infectious disease caused by Coxiella burnettii.1 In the acute form, recovery is usual,2 but uncontrolled studies in Australian abattoir workers3 have suggested that some Q‐exposed patients may develop prolonged profound fatigue. In 1989, a windborne outbreak of Q fever affected 147 patients in the West Midlands region of the UK,4 some of whom subsequently complained of symptoms suggestive of a post‐infectious chronic fatigue syndrome (CFS). A questionnaire‐based cohort study in 1994 revealed that symptoms of persistent fatigue were present in 42.3% of subjects,5 but the frequency of the same symptoms was also high in the control group (26%) perhaps suggesting that the questionnaire used was over‐sensitive.

Persistent fatigue after infections (e.g. Epstein‐Barr virus6 and enteroviruses7) has been described, but studies have been plagued by inadequate standardization of the definition of fatigue. The International Chronic Fatigue Syndrome study group's structured approach to fatigue syndromes8 (1994 Centre for Disease Control CDC definition) was operationalized by Wessely to investigate the potential for viral infections to produce chronic fatigue in primary care.9 Wesseley found fatigue to be normally distributed in the population, with 5799 out of 15283 subjects having fatigue (38%, 95%CI 37.2–38.7) and 2793 (18.3%, 95%CI 17.7%–18.9%) subjects having idiopathic chronic fatigue.

This study set out to investigate whether the finding of excess fatigue following Q fever suggested by the five‐year follow‐up study was maintained at ten years, using the comprehensive 1994 CDC fatigue definition as operationalized by Wessely.

Methods

Records from the initial Q fever outbreak were used to trace the affected patients. A case of acute Q fever had been defined by a titre of ≥1:256 or a four‐fold rise in phase II antibodies. Controls were identified by random selection from two local general practices, matching for age, sex and smoking history (never, ex‐ or current) of the cases when they were last contacted in 1989 or 1994. Controls were excluded if they had serological evidence of prior Q‐fever exposure.

A postal questionnaire was sent to subjects at home, followed by further assessment in hospital, including investigations to explore the possibility that a subclinical cardiomyopathy might underlie the Coxiella‐exposed patient's fatigue. These results are reported elsewhere.10

Figure 1 outlines the follow‐up of the cohort from the original outbreak, up to and including the current study.

Questionnaire

The home questionnaire incorporated the 11‐item fatigue questionnaire used by Wessely.9,11,,12 Psychological morbidity was assessed using the twelve‐item general health questionnaire (GHQ).13 For the purposes of designating fatigue or psychological caseness, the traditional scoring systems were used (0,0,1,1), a score of 4 or more indicating psychological caseness.

Figure 1.

History of the cohort from the 1989 outbreak to the current study.

Hospital assessment

Each subject was informed that the purpose of the study was to assess health status 10 years after Q fever. Written consent was obtained. A structured history was taken followed by a physical examination and blood tests with the aim of detecting co‐morbid physical conditions that might account for fatigue.8 Blood tests comprised a biochemical profile, full blood count, immunoglobulins, thyroid function tests, C‐reactive protein and serology for C. burnetti. Spirometry using a dry bellows spirometer (Vitalograph), an electrocardiogram and a shuttle walk, and a standardized, incremental exercise test14 were performed. A further self‐completion questionnaire was carried out that included the Medical outcome study 20‐item questionnaire (MOS short form15), which was used to define functional impairment in the construction of the CFS definition. Subjects were asked to complete a checklist of 25 somatic symptoms derived from the Somatic Discomfort Questionnaire.16 Somatic symptoms are important minor symptoms in the construction of the CFS definition. The smoking burden was measured in pack‐years, since the patients contracting Q fever had a high prevalence of smoking, and smoking has the potential to be a confounder of cardiovascular morbidity.

Current psychiatric illness was investigated using a self‐completed computerized version of the Revised Clinical Interview Schedule (CIS‐R)17 which has been designed to record psychiatric morbidity in primary care and was used in the Wessely studies.

Outcome measures

Wessely's method of analysing the data from the questionnaires was used to construct the Centers for Disease Control (CDC) 1994 criteria for chronic fatigue syndrome.8 Subjects scoring four or more using the traditional scoring system for the fatigue questionnaire were designated as fatigued. Subjects designated as fatigued on the fatigue questionnaire and describing fatigue >50% of the time for 6 months were described as having idiopathic chronic fatigue (ICF). Patients satisfying the criteria for idiopathic chronic fatigue, and in addition having functional impairment and four or more of the CDC, defined additional symptoms were designated as having CFS. Strict operationalization of the CDC definition of fatigue requires that comorbidities that might explain fatigue are sought and that ‘pure’ fatigue syndromes are those occurring in comorbidity‐free subjects. Parametric comparisons of means were made by paired t tests, and comparisons of proportions by the McNemar test. In addition to measuring psychological distress with the GHQ, the Revised Clinical Interview Schedule was used, with a cut‐off of 12 indicating likely psychological ill health, and attempts were made to identify patients with prior mental health problems by asking if patients had had ‘previous nerve trouble’.

The study protocol was approved by the East Birmingham Local Research and Ethics committee.

Results

The home questionnaire

The original outbreak in 1989 involved 147 Coxiella‐exposed subjects (Q‐exposed) and Figure 2 shows the cohort's follow‐up to 1999. By 1999, 19 (12.9%) had died leaving 128 survivors. Table 1 shows the age at death and cause of death in the patients who died. Of the 128 surviving Q subjects, three had emigrated, leaving 125 survivors in the UK of whom 115 were traced and contacted: 110 replied and agreed to take part in the study (response rate from traced subjects 96%). Of these, 108 returned the home questionnaire, allowing 108/128 survivors (84%) to have an initial fatigue classification. Two subjects attended hospital but never returned the home questionnaire, so their fatigue status could not be determined. Of the 108 Q subjects, 70 (64.8%) were fatigued and 38 (35.2%) were fatigue free.

Of 226 controls contacted to take part in the study, 86 (38.1%) returned the home questionnaire and 81 (35.8%) attended hospital for further assessment (Figure 3). The total number of controls approached included the individuals matched to cases who were initially approached but did not respond, with the final total number of controls invited to take part being the cumulative total needed to obtain the number of matched controls finally used. Of the 86 controls returning the home questionnaire, 32 (37.2%) were fatigued, and of the 81 who attended hospital, 30 (37%) were fatigued. Of the 86 controls returning the home questionnaire, 20 (23.3%) were psychologically distressed according to GHQ and of the 81 controls attending hospital, 19 (23.4%) were psychologically distressed (Table 2). The six controls with serological evidence of Q fever had a mean age of 60.2 (SD 13.8) years; five (83.3%) were male and three (50%) were fatigued. Of the three fatigued patients, one had ICF (with comorbidities) and two had CFS (with comorbidities). The other three were fatigue‐ and comorbidity‐free. Only one of the six Q‐exposed controls was a GHQ case.

On the basis of the home questionnaire alone and excluding the six controls with serological evidence of prior Q exposure, 108 Q‐exposed patients and 80 controls were available for analysis. This provided 77 matched pairs for analysis (Table 3). McNemar's test was used to test whether fatigue, ICF and GHQ caseness differed between the matched pairs. Fatigue was commoner in Q‐exposed subjects: 50 (64.9%) vs 27 (35.1%) (p⩽0.001). Idiopathic chronic fatigue (ICF) was commoner in Q‐exposed subjects than controls: 25 (32.5%) vs. 11(14.3%) (p=0.01). GHQ cases were also commoner in Q‐exposed subjects: 36 (46.8%) vs. 18 (23.4%) (p=0.004).

Figure 2.

Demographics, deaths and losses to follow‐up in Q exposed cohort.

View this table:
Table 1

Details of cases who died in the 10 years after developing acute Q fever

Sex and age (years)Year of deathCause of death
Male, 721990Pneumonia and dementia
Male, 781990Pneumonia and dementia
Male, 771991Myocardial infarction
Male, 611992COPD
Male, 671992Myocardial infarction
Male, 671993COPD
Male, 731994Myocardial infarction
Female, 491994Carcinoma of the cervix
Male, 701994Renal failure and diabetes
Female, 801994Ruptured aortic aneurysm
Male, 711994Cardiac failure
Male, 531995Sub‐acute bacterial endocarditis (aortic stenosis)*
Male, 771996Myocardial infarction
Female, 731997Pneumonia, aortic valve replacement*
Male, 751997CVA
Female, 761997Pulmonary fibrosis**
Male, 661997Carcinoma of the kidney
Male, 871998Myocardial infarction
Female, 851998Myocardial infarction
  • *In both cases, valve disease thought to pre‐date exposure. **Pre‐dated exposure.

Figure 3.

Demographics and response rate in controls.

View this table:
Table 2

Characteristics of Q‐exposed subjects and controls returning the home questionnaire

Q‐exposed subjectsControls
n Mean age (SD)Sex (male) (%)Fatigued (%)ICF (%)GHQ cases (%) n Mean age (SD)Sex (male) (%)Fatigued (%)ICF (%)GHQ cases (%)
All subjects returning home questionnaires10855.5 (12.1)94 (87.0%)70 (64.8%)37 (34.3)49 (45%)80*55.6 (11.8)70 (87.5%)29 (36.3%)12 (15.0%)19 (23.8%)
Subjects returning home questionnaires but not attending hospital2358.1 (12.6)21 (91.3%)11 (47.8%) 6 (26.1%) 7 (30.4%) 559.6 (17.1%) 5 (100%) 1 (25%) 1 (25%) 1 (20%)
Subjects returning home questionnaire and attending hospital8554.7 (12.0)73 (85.9%)59 (69.4%)31 (36.5%)42 (49.4%)7555.3 (11.5)65 (86.7%)27 (36.0%)11 (14.7%)18 (24%)
  • *86 controls returned the home questionnaire, but six controls had serological evidence of Q fever and their data are presented elsewhere.

View this table:
Table 3

Characteristics of successfully matched cases and controls returning home questionnaire used for matched analysis

Q‐exposed (n=77)Controls (n=77) p
Age (mean SD)55.6 (11.8)55.4 (11.7)
Sex (males)68 (88.3%)68 (88.3%)
Fatigue, all levels50 (64.9%)27 (35.1%)⩽0.001
ICF or worse25 (32.5%)11 (14.3%)0.01
GHQ case36 (46.8%)18 (23.4%)0.004
  • The matching process required cases and controls to have both a home questionnaire and to attend hospital for further assessment. There were 79 controls who attended hospital, but since 2 of these were matched to Q cases who did not complete the home questionnaire, 77 pairs were available for analysis. ICF, idiopathic chronic fatigue; GHQ, general health questionnaire.

Did the loss of Q‐exposed subjects during the matching process introduce imbalance in fatigue status that might bias the matched analysis?

At the level of the analysis of the home questionnaire, the loss of patients due to the matching process did not significantly alter the prevalence of fatigue in the Q‐exposed (64.8% total cohort vs. 64.9% subgroup‐matched), or in the controls (36.3% total cohort vs. 35.1% subgroup‐matched). Considering the Q‐exposed cases who were matched, 50 (64.9%) of the 77 matched Q‐exposed subjects were fatigued, a proportion that is not significantly different to the proportion of fatigued patients in the original 108 (64.8% vs. 64.9%). Of the 77 controls available for matching, 27 (35.1%) were fatigued. Hence the matched controls had a prevalence of fatigue of 35.1% vs. the eligible controls as a whole of 36.3%.

Hospital assessment

In addition to receiving the home questionnaire, subjects were invited to attend hospital for further assessment. Table 4 shows the characteristics of subjects attending hospital. The more detailed hospital assessment allows the further categorization of fatigue to include chronic fatigue syndrome (CFS) and the identification of patients with comorbidities. Data are also presented for the numbers of cases and controls who had a history of prior nerve troubles, the number of somatic symptoms and the accumulated pack‐year smoking burden.

Patients who were fatigued on home questionnaire showed a trend toward being more likely to attend hospital for further assessment. The relative risk of a fatigued Q subject attending hospital compared to a non‐fatigued Q subject was 1.23 (95%CI 0.97–1.56) χ2 3.7, p=0.054).

The relative risk of a Q‐exposed subject with psychological morbidity attending hospital compared to a psychological‐morbidity‐free Q‐exposed subject was 1.18 (95%CI 0.97–1.43, NS).

Of the 80 controls returning the home questionnaire, 29 (36.3%) were fatigued, and of the 75 who attended hospital, 27 (36%) were fatigued. Of the 80 controls returning the home questionnaire, 19 (23.8%) were psychologically distressed according to GHQ, and of the 75 controls attending hospital, 18 (24%) were psychologically distressed. There was no significant difference in fatigue scores or GHQ scores in controls classified by subsequent hospital attendance.

Hospital attendance allowed investigations to establish whether a subclinical cardiomyopathy might explain the fatigue in Q‐exposed subjects: these data are presented elsewhere.10 In addition. hospital assessment allowed subjects to be assessed for the co‐morbidities defined by the CDC 1994 fatigue definition as possible explanations for fatigue (Table 5). Comorbidities were present in 25 (29.4%) of the 85 Q‐exposed patients attending hospital and 20 (26.7%) of the 75 controls. There was no statistically significant difference in the proportion of subjects with co‐morbidities in the 72 matched pairs: 23 (31.9%) Q‐exposed vs. 19 (26.4%) controls (p=0.57).

Table 6 shows the characteristics of the 72 matched Q‐exposed subjects and controls seen in hospital. Tabulations are provided for all 72 of the matched pairs, including both homogeneous (both members having co‐morbidity or both members being co‐morbidity free) and heterogeneous pairs (pairs discordant with respect to co‐morbidity) together and for 37 homogeneous pairs where both pair members were co‐morbidity free and seven homogeneous pairs where both pair members had co‐morbidities.

In the 72 matched pairs taken together, fatigue, (Q‐exposed 48 (66.7%) vs. controls 25 (34.7%), p⩽0.001), ICF (Q‐exposed 25 (34.7%) vs. controls 10 (13.9%), p=0.004), CFS (Q‐exposed 14 (19.4%) vs. controls 3 (4.2%), p=0.003) and GHQ caseness (Q‐exposed 34 (47.2%) vs. controls 17 (23.6%), p=0.004) were all significantly commoner in the Q‐exposed patients than in controls. In addition, the mean number of somatic symptoms was higher in Q‐exposed subjects than controls (Q‐exposed 6.7 vs. controls 4.6, p=0.002) (paired t test).

Psychological caseness measured by the Revised Clinical Interview Schedule (CIS‐R) (Q‐exposed 16 (22.2%) vs. controls 8(11.1%), p=0.134), and comorbidity (23 (31.9%) Q‐exposed vs. 19 (26.4%) controls, p=0.57) were not significantly commoner in Q‐exposed patients.

Analysing only homogeneous pairs leads to a loss of power, but in the 37 pairs in which both members are free from comorbidities, fatigue remained significantly commoner in Q‐exposed subjects than in controls (p=0.003). In the comorbidity free pairs, ICF showed a trend to being more common in Q‐exposed patients (p=0.07). None of the other measures were significantly different in the homogeneous pairs. Though Q‐exposed patients had more fatigue and psychological morbidity than did controls, the small numbers in the homogeneous pairs are accompanied by a loss of statistical significance. In the seven homogeneous pairs with comorbidities, controls had a significantly higher mean number of pack‐years smoking history (Q‐exposed mean 40.0 years vs. controls mean 51.7 years; p=0.013).

View this table:
Table 4

Characteristics of cases and controls attending hospital

Q‐fever‐exposedControls
All attending hospital (n=85)Attending hospital, co‐morbidity‐free (n=60)Attending hospital with co‐morbidities (n=25)All attending hospital (n=75)Attending hospital, co‐morbidity‐free (n=55)Attending hospital with co‐morbidities (n=20)
Mean age (SD)54.7 (12.0)52.7 (12.6)59.7 (8.8)55.3 (11.5)52.8 (11.6)62.5 (7.5)
Sex (males)73 (85.9%)51 (85.0%)22 (88.0%)65 (86.7)48 (87.3%)17 (85%)
Fatigue, all levels59 (69.4%)37 (61.7%)22 (88.0%)27 (36.0%)13 (23.6%)14 (70%)
ICF or worse31 (36.5%)16 (26.7%)15 (60.0%)11 (14.7%) 3 (5.5%) 8 (40.0%)
CFS17 (20%) 7 (11.7%)10 (40.0%) 4 (5.3%) 0 (0%) 4 (20%)
GHQ case42 (49.4%)27 (45.0%)15 (60.0%)18 (24.0%)13 (23.6%) 5 (25.0%)
CIS case22 (25.9%)12 (20.0%)10 (40.0%) 9 (12.0%) 5 (9.1%) 4 (20.0%)
Mean somatic symptoms (SD) 6.9 (4.6) 6.0 (4.3) 9.2 (4.8) 4.7 (3.2) 4.2 (2.9) 6.1 (3.8)
Previous ‘nerve troubles’28 (32.9%)16 (26.7%)12 (48.0%)17 (22.7%)12 (21.8%) 5 (25.0%)
Pack‐years smoking23.2 (19.1)18.3 (15.2)35.1 (22.4)28.2 (20.2)23.7 (17.3)40.8 (22.6)

View this table:
Table 5

Significant co‐morbidities of the subjects in both exposed and control groups

NumbersCo‐morbidity
Cases
4Chronic obstructive pulmonary disease, bronchiectasis, occupational asthma
5Ischaemic heart disease (including uncontrolled AF) (including one patient with heart transplant for ischaemic cardiomyopathy)
10Other: renal transplant (1), Non‐Hodgkin's lymphoma (1), metastatic cancer (2), sleep apnoea (1), myeloid leukaemia (1), major psychiatric disorder (1), uncontrolled diabetes (1), severe arthritis (2)
6Alcohol excess
Controls
4Chronic obstructive pulmonary disease
7Ischaemic heart disease
6Other: Severe urological problems and arthritis (1), chronic renal failure (1), prostate cancer (1), severe arthritis (2), uncontrolled diabetes (1)
3Alcohol excess

View this table:
Table 6

Characteristics of matched cases and controls seen in hospital

Matched with and without comorbidities attending hospitalMatched without comorbidities attending hospitalMatched with comorbidities attending hospital
Q‐exposed (n=72)Controls (n=72) p Q‐exposed (n=37)Controls (n=37) p Q‐exposed (n=7)Controls (n=7) p
Age (mean SD)55.3 (11.4)55.1 (11.3)50.5 (11.7)50.2 (11.6)64.0 (4.2)64.1 (5.1)
Sex (males)63 (87.5%) 6.3 (87.5%)32 (86.5%)32 (86.5%) 5 (71.4%) 5 (57.1%)
Fatigue all levels48 (66.7%)25 (34.7%)<0.00120 (54.1%) 9 (24.3%)0.003 6 (85.7%) 5 (71.4%)1.0
ICF or worse25 (34.7%)10 (13.9%)0.004 8 (21.6%) 2 (5.4%)0.070 5 (71.4%) 2 (28.6%)0.38
CFS14 (19.4%) 3 (4.2%)0.003 2 (5.4%) 0 (0%)0.500 4 (57.1%) 1 (14.3%)0.25
GHQ case34 (47.2%)17 (23.6%)0.00416 (43.2%) 9 (24.3%)0.118 3 (42.9%) 1 (14.3)0.5
CIS case16 (22.2%) 8 (11.1%)0.134 6 (16.2%) 2 (5.4%)0.289 3 (42.9%) 0 (0%)0.25
Mean somatic symptoms (SD) 6.7 (4.3) 4.6 (3.2)0.002 5.4 (3.5) 4.2 (2.9)0.27 8.7 (5.5) 7.0 (3.1)0.31
History previous ‘nerve troubles’24 (33.3%)17 (23.6%)0.281 9 (24.3%) 8 (22.2%)1.0 2 (28.6%) 2 (28.6%)1.0
Pack‐years smoking24.5 (19.9)28.3 (20.0)0.116.9 (16.6)21.9 (16.2)0.15940.0 (28.9)51.7 (25.7)0.013
Comorbidity present23 (31.9%)19 (26.4%)0.57 0% 0%100%100%
  • McNemar's test used for comparison of proportions, paired samples T‐test for comparison of means.

Discussion

This study has demonstrated significantly increased scores using instruments validated for the operationalization of the 1994 CDC fatigue definitions in Q‐exposed subjects 10 years after acute Q fever compared to age‐, sex‐ and smoking‐matched controls. Considering the 108 Q‐exposed patients who returned the home questionnaire, 64.8% had fatigue of any intensity, 34.3% had ICF and 45% were GHQ cases. Of the 80 controls who returned the home questionnaire, 36.3% had fatigue of any intensity, 15% had ICF and 23.8% were GHQ cases. Under statistical comparison of 77 matched pairs of cases and controls returning the home questionnaire, fatigue of any level (p⩽0.001), ICF (p=0.01) and GHQ caseness (p=0.004) was commoner in the Q‐exposed patients than controls.

Overall, 85 Q‐exposed cases and 75 controls were assessed in hospital, which allowed the identification of comorbidities. Sixty Q‐exposed patients and 55 controls were free of comorbidities, and the fuller hospital assessment allowed classification to be extended to include chronic fatigue syndrome. Of the comorbidity‐free, Q‐exposed cases, 61.7% had fatigue of any level, 26.7% had ICF and 11.7% had CFS, substantially higher rates than in controls for all levels of fatigue. Matched analysis of Q‐exposed cases with controls led to a loss of statistical power, with only 37 pairs homogeneously comorbidity‐free. Nevertheless, Q‐exposed patients had significantly more fatigue of any level (p=0.003).

These results show an excess of fatigue in Q‐exposed patients compared to controls, whether the comparison involves subjects returning the home questionnaires, subjects seen in hospital with and without comorbidities, or patients seen in hospital who are comorbidity‐free. In interpreting these results, it is important to consider whether they might be explained by bias, confounding or chance. Given that the controls only represent 38.1% of the controls contacted, and the 108 Q‐exposed only represent 84.3% of the Q‐exposed still alive in 1999, could the excess of fatigue seen in Q‐exposed result from selection bias? The results show that Q‐exposed patients with fatigue were more likely to attend hospital than those without fatigue and if we were to take this to the extreme, and propose that all the Q‐exposed alive in 1999 but not responding were fatigue‐free, we would have 70 (54.7% (95%CI 45.7%–63.4%)) fatigued patients out of 128 Q‐survivors. The prevalence of fatigue in controls returning the home questionnaire was 36.3% (95%CI 25.8%–47.8%), similar to the prevalence of fatigue of 38% (95%CI 37.2%–38.7%) found by Wessely in a community sample of 15 283 patients using identical instruments and methodology.18 In order for selection bias to explain the difference between cases and controls, the above sensitivity analysis demonstrates that we would need to assume that all Q‐exposed non‐responders were not fatigued and that selection bias operated in an opposite direction in controls, with non‐responding controls being more likely than responding controls to be fatigued.

The matched design meant that statistical comparisons between cases and controls were only made between a proportion of subjects. The prevalence of fatigue in all home questionnaire responders was 64.8% in Q‐exposed patients, nearly twice that of controls, and the proportions were similar when regarding further sub groupings. However, in the 37 homogeneous, comorbidity‐free pairs, fatigue prevalence had fallen by 12.5% in Q‐exposed to 54.1%, but by a similar proportion in controls to 24.3%. The fall in fatigue in comorbidity‐free pairs is to be expected, since this process specifically seeks to exclude comorbidities that might cause fatigue. It can be seen that there was no marked difference between the prevalence of fatigue in the subjects contributing to the matched analysis compared to the sample as a whole. Since matching led to a loss of numbers and a consequent loss of statistical power to find a difference between cases and controls, the statistical difference in fatigue between cases and controls at every level supports the conclusion that the excess fatigue in Q‐exposed patients is a robust finding.

Response bias and confounding are conveniently considered together since they both require a careful consideration of the exposures of cases and controls and how that might influence response to the questionnaires. Figure 1 shows that since 1989 the Q‐exposed cohort have had repeated contact with the medical profession following on from an illness with frightening symptoms. We also see that the process of follow‐up has repeatedly informed patients that investigators are seeking to determine whether Q fever is accompanied by long‐term sequelae. Given that the diagnosis of fatigue is by self‐report questionnaires, it is possible that the responses to the questionnaire have become distorted, and that the pattern of patient response does not reflect fatigue but instead reflects patients' altered approach to the questionnaire. In this case the Wesseley instruments would no longer be measuring fatigue at all.

Confounding may also explain the higher scores observed in cases, but whereas response bias would mean that higher scores no longer represented fatigue, here confounding would mean that though fatigue was being measured, it was not due to Q fever. Once again the differential exposures of the prolonged follow‐up are important. Investigational contact between patients and the medical profession can cause psychological illness. Men receiving the results of screening tests revealing them to be at risk of cardiovascular disease show increased GHQ scores to case levels over time,19 an effect now recognized in numerous other clinical conditions.20 Our Q‐exposed subjects might be considered at risk of psychological distress because of the uncertainty accompanying their original dramatic illness, followed by subsequent repeated medical contacts reinforcing perceptions of ill health. Since fatigue is an ubiquitous symptom of psychological distress, repeated medical probing for symptoms could have produced the fatigue scores observed. Thus the excess fatigue observed in the Q‐exposed subjects might be considered an iatrogenic illness caused and/or perpetuated by the extensive follow‐up of the subjects following a severe illness.

If the fatigue is not a distortion caused by response bias or an iatrogenic illness caused by follow‐up, we are left with the possibility that the observed fatigue is in some way related to the Q fever 10 years previously. It has been suggested that fatigue may be related to the infection through persistent immune activation by persisting Coxiella antigen. Persistent antigen can be detected in buffy coat and bone marrow from abattoir workers with Q fever fatigue syndrome using PCR.21 In addition, their peripheral blood mononuclear cells show enhanced release of interleukin‐6 when exposed to Coxiella antigen in vitro. This suggests that persistence of antigen could produce chronic fatigue by cytokine release.22,,23

We conclude that subjects who were exposed to Coxiella in 1989 now score more highly on questionnaires designed to measure fatigue than does an unexposed control population. It is possible that these results do not represent fatigue at all, but represent a biased response to questionnaires caused by the process of follow‐up. If the higher scores do in fact represent fatigue, this may be due to ongoing antigen persistence, or alternatively may represent the psychological effects of prolonged medical follow‐up. The former possibility is currently being explored in this cohort, but this study has raised important concerns about the methodological and ethical difficulties in investigating the possible connection between rare disease exposures and long‐term symptoms, particularly subjective perceptions of fatigue.

Acknowledgments

We are grateful to Professor B.P. Marmion for his helpful and constructive comments throughout the conduct and writing‐up of this study. The study was supported by a LORS (Locally Organised Research) grant from the West Midlands NHS Executive.

Footnotes

  • Address correspondence to Professor J.G. Ayres, Birmingham Heartlands Hospital, Birmingham, West Midlands B9 5SS. e‐mail: ayresj{at}heartsol.wmids.nhs.uk

References

View Abstract