Summary of findings

Hypothesis testing of the individual risk markers identified similar premorbid risk markers for self-reported CFS/ME and CFS-like illness, suggesting aetiological roles for childhood adversity, adulthood physical inactivity, adulthood BMI and some childhood illnesses but not for childhood BMI or activity levels. Multivariable models, adjusted for premorbid and comorbid psychopathology identified parental physical abuse, childhood gastrointestinal symptoms and reporting many colds as independent risk markers for self-reported CFS/ME. Only female gender and adult premorbid psychopathology were risk markers for CFS-like illness after adjustment for comorbid psychopathology. Parental physical abuse and early adulthood psychopathology were premorbid risks for both CFS/ME and CFS-like illness, strengthening the evidence for their aetiological role, while also finding this association to be partially confounded by comorbid psychopathology. Overall, the statistically significant risk markers affected only the minority of individuals, however defined, so that, for example, only 16% of individuals reported

Table 3

Multivariable odds ratios between risk markers and chronic fatigue syndrome (CFS)-like illness by 42 years

	CFS-like illness: model adjusted for coexisting risk markers (n = 9368)		CFS-like illness: model adjusted for coexisting risk markers and premorbid psychopathology at 23 and 33 years (n = 9368)		CFS-like illness: model adjusted for coexisting risk markers, premorbid psychopathology at 23 and 33 years, and comorbid psychopathology at 42 years (n = 9368)
Risk/protective marker	OR	95% CI	OR	95% CI	OR	95% CI
Female	2.05***	1.52–2.77	1.62**	1.17–2.24	1.59**	1.14–2.21
Illness in the household (7–16)	1.32	0.94–1.84	1.20	0.84–1.72	1.15	0.79–1.65
Neglected/underfed appearance (7–11)	1.49	0.81–2.71	1.38	0.77–2.47	1.34	0.76–2.39
Maternal absence (7–16)	1.43	0.89–2.30	1.27	0.77–2.07	1.39	0.83–2.33
Paternal absence (7–16)	1.23	0.74–2.03	1.22	0.72–2.08	1.14	0.64–2.02
In care (7–16)	1.05	0.54–2.05	0.76	0.33–1.76	0.76	0.34–1.70
Divorce of parents	0.99	0.60–1.61	0.86	0.50–1.47	0.88	0.51–1.53
Parental physical abuse	2.44***	1.60–3.70	1.94*	1.17–3.20	1.56	0.96–2.52
Parental sexual abuse	1.48	0.69–3.17	1.16	0.55–2.42	1.09	0.48–2.46
Many colds at 7 or 11 years	1.19	0.81–1.74	1.07	0.69–1.64	1.07	0.70–1.64
More than 3 throat infections in previous year at 16 years	1.07	0.69–1.66	1.08	0.69–1.69	1.05	0.65–1.70
Headaches or migraine at 16 years	1.20	0.83–1.75	1.02	0.69–1.51	0.88	0.59–1.32
School absence at 16 years
Less than 1 week	1.00		1.00		1.00
1 week to 1 month	1.28	0.72–2.27	1.21	0.68–2.15	1.20	0.68–2.11
Over 1 month	1.69	0.95–3.02	1.37	0.73–2.55	1.33	0.70–2.52
Any report of family psychiatric illness	1.18	0.65–2.11	1.09	0.56–2.11	1.25	0.61–2.54
BMI at 23 years
Underweight	0.71	0.34–1.48	0.67	0.31–1.44	0.53	0.24–1.16
Overweight	1.17	0.75–1.84	1.04	0.65–1.67	1.04	0.64–1.69
Obese	1.75	0.64–4.78	1.68	0.60–4.75	1.81	0.66–4.99
BMI at 33 years
Underweight	1.97	0.93–4.19	1.92	0.84–4.38	1.94	0.83–4.57
Overweight	1.14	0.81–1.61	1.08	0.77–1.52	1.14	0.80–1.62
Obese	1.36	0.76–2.43	1.44	0.79–2.64	1.53	0.80–2.91
Leisure-time physical activity at 23 years
Never or once in last 4 weeks	1.92***	1.31–2.81	1.64*	1.09–2.47	1.47	0.98–0.23
3–4 times /5+ times a week	1.14	0.61–2.12	1.25	0.64–2.44	1.24	0.64–2.42
Leisure-time physical activity at 33 years
Never/less often	1.51*	1.10–2.08	1.27	0.90–1.78	1.18	0.81–1.71
4+ times a week	1.01	0.71–1.45	0.95	0.65–1.38	0.91	0.62–1.35
Psychopathology at 23 years			2.31***	1.61–3.31	1.31	0.91–1.89
Psychopathology at 33 years			6.26***	4.37–8.96	2.83***	1.94–4.11
Psychopathology at 42 years					12.84***	9.08–18.17

* P = 0.05

** P = 0.01

*** P = 0.001.

parental physical abuse. This may reflect that CFS/ME is a heterogeneous condition ^{Reference White22} and future studies will need to define subgroups to further elucidate aetiological pathways.

Premorbid psychopathology and childhood adversity

This study confirms the importance of premorbid adulthood psychopathology in the aetiology of CFS/ME and CFS-like illness ^{Reference Harvey, Wadsworth, Wessely and Hotopf3} and demonstrates the need to take premorbid and comorbid psychopathology into account when examining risk markers for CFS/ME and CFS-like illness. The only childhood adversity that was a risk marker for CFS/ME and CFS-like illness was parental abuse confirming Heim et al's findings ^{Reference Heim, Wagner, Maloney, Papanicolaou, Solomon and Jones5} of higher reports of childhood trauma in individuals with clinically confirmed CFS/ME compared with healthy controls. Our effect sizes were more modest than Heim et al's and only 16% of ours reported parental physical abuse. This could reflect the use of individual multipleinformant adversity measures in the current study that may lack the sensitivity of the multiple-item Childhood Trauma Scale in Heim et al's study; the lack of measures of non-parental abuse in the current study; or recall bias in studies that use retrospective uncorroborated self-reports of childhood trauma.

Childhood physical abuse is associated with a wide-range of adverse physical health outcomes in adulthood, ^{Reference Walker, Gelfand, Katon, Koss, Von Korff and Bernsterin23} suggesting that distress associated with abuse may be expressed through later physical symptoms. Childhood adversity is also associated with altered stress responses in adulthood, as evidenced by hypothalamic-pituitary-adrenal axis dysfunction, ^{Reference Tarullo and Gunnar24} decreased hippocampal glucocorticoid receptor expression ^{Reference McGowan, Sasaki, D'Allessio, Dymov, Labonte and Szyf25} and inflammation, ^{Reference Danese, Pariante, Caspi, Taylor and Poulton26} which have also been implicated in the aetiology of CFS/ME. ^{Reference Rajeevan, Smith, Dimulescu, Unger, Vernon and Heim27-Reference Buchwald, Wener, Pearlman and Kith29} These findings suggest that childhood adversity should be explored when people present with CFS/ME, as a history of childhood adversity is more likely to be found in CFS/ME, particularly when accompanied by concurrent psychopathology.

Childhood illness

Evidence for childhood illness as a risk marker from previous studies is equivocal. ^{Reference Viner and Hotopf9,Reference Harvey, Wadsworth, Wessely and Hotopf10} We found few associations, using more detailed measures of childhood illness. Risk markers included gastrointestinal symptoms, many colds, headaches/migraines, throat infections and chronic illness. However, few associations were independent of psychopathology and physiological pathways between these illnesses and CFS/ME remain unclear. Early illness experience, vicarious or the child's own, is linked with the development of medically unexplained symptoms in adulthood, ^{Reference Hotopf, Mayou, Wadsworth and Wessely30} reflecting possible pathways for CFS/ME via social learning and/or reinforcement of illness behaviour. ^{Reference Fisher and Chalder31} The childhood illness measures may reflect the overprotectiveness of the parent, rather than illness per se. Overprotective parenting may be relevant to the formation of belief systems, such as negative schemata associated with response to illness, that may predispose to the development of CFS/ME in adulthood. ^{Reference Fisher and Chalder31} Overall, evidence from population cohort studies regarding childhood illness as a risk marker is not strong.

Childhood and adulthood activity levels and BMI

Childhood activity levels and childhood BMI were not premorbid risk markers for CFS/ME or CFS-like illness, contrary to the findings of the 1946 and 1970 cohorts that overactivity and inactivity were risk markers. ^{Reference Viner and Hotopf9,Reference Harvey, Wadsworth, Wessely and Hotopf10} Additionally, there were no associations between early adulthood activity and BMI once co-occurring risk markers and psychopathology were taken into account. This could reflect differential associations as a result of cohort differences in rates of physical activity; ^{Reference Goodwin, White, Hotopf, Stansfeld and Clark4} a lack of correction for non-response bias in the 1970 and 1946 analyses; a role for childhood inactivity for earlier but not later onset of CFS/ME; the lack of power in our data to detect an effect of low BMI; unadjusted confounding by other risk markers; or chance findings associated with the lack of sensitivity of the activity questions used in cohort studies.

Defining CFS/ME and CFS-like illness

We have examined two definitions of CFS: self-reported CFS/ME and operationalised CFS-like illness. These definitions have methodological limits that may have influenced the findings. The self-reported nature of the CFS/ME measure may influence associations with the risk markers: stronger associations may be observed in samples defined using clinically assessed CFS/ME. Our CFS-like illness outcome is limited as it is derived from symptom reports in the previous few weeks, ^{Reference Goldberg, Gater, Sartorius, Ustun, Piccinelli and Gureje14,Reference Rodgers, Pickles, Power, Collishaw and Maughan15} making it not directly comparable with our CFS/ME measure, which assessed having a diagnosis during adulthood. This difference could explain some of the variation in findings between the risk markers for the two outcomes. Such differences could also be explained by the lack of power for the self-reported CFS/ME analyses compared with the CFS-like illness analyses. Confidence in the associations is strengthened in those risk markers found with both definitions, such as physical abuse. Conversely, for other risk markers the effect sizes for the different outcomes do not look comparable, for example for physical activity and some childhood illness measures. Our speculations about the influence of these definitions on our findings is limited, but we note the effect of different definitions of CFS found in previous studies. ^{Reference Hamilton, Gallagher, Thomas and White32-Reference Buchwald, Herrell, Ashton, Belcourt, Schmaling and Sullivan34}

Strengths and limitations

Limitations include the use of non-clinically confirmed cases of CFS/ME; retrospective reports of parental abuse, which did not capture abuse by others; adversity measures that do not assess how the individual coped with adversity; a lack of power for some analyses; and an increased chance of type 1 errors. Unmeasured genetic or other individual confounding factors could account for the associations. However, this is the first study to examine prospectively measured childhood adversity and is also the first to use imputation to compensate for missing data and attrition. Other strengths include the use of two ways to define cases, examination of a wide range of childhood illness, a good sample size with a long-follow up capturing the mean onset age for CFS/ME; and premorbidly assessed risk markers.

Implications

Many risk markers for CFS/ME and CFS-like illness were related to premorbid and comorbid psychopathology. Prospectively measured childhood adversity was not a risk marker for CFS/ME, however defined. When measured retrospectively, parental abuse was associated in a minority but this was confounded by concurrent psychopathology for one definition of CFS/ME. No risk marker was associated with both definitions of CFS/ME in the final adjusted models, perhaps reflecting the heterogeneity of this illness. ^{Reference Walker, Gelfand, Katon, Koss, Von Korff and Bernsterin23} Female gender, childhood illness and premorbid psychopathology were risk markers for either CFS/ME or CFS-like illness, but were sometimes confounded by comorbid psychopathology. Potential pathways between risk markers, including psychopathology, and CFS/ME include childhood adversity and their consequently altered stress responses. These pathways should be the focus of further pathophysiological research, in pre-specified subgroups, to inform interventions for prevention and treatment. ^{Reference Heim, Nater, Maloney, Boneva, Jones and Reeves6,Reference Kato, Sullivan, Evengard and Pedersen7,Reference McGowan, Sasaki, D'Allessio, Dymov, Labonte and Szyf25,Reference Rajeevan, Smith, Dimulescu, Unger, Vernon and Heim27,Reference White, Thomas, Kangro, Bruce-Jones, Amess and Crawford33}