Basu et al.

BMC Medicine (2016) 14:122

DOI 10.1186/s12916-016-0662-y

RESEARCH ARTICLE Open Access

Fatigue is associated with excess mortality

in the general population: results from the
EPIC-Norfolk study
Neil Basu1*, Xingzi Yang1, Robert N. Luben2,4, Daniel Whibley1, Gary J. Macfarlane1, Nicholas J. Wareham3,4,
Kay-Tee Khaw3,4 and Phyo Kyaw Myint1

Abstract
Background: Significant fatigue is a frequent reason for seeking medical advice in the general population. Patients,
however, commonly feel their complaint is ignored. This situation may be because clinicians perceive fatigue to be
benign, unrelated to traditional biomedical outcomes such as premature mortality. The present study aimed to
investigate whether an association between significant fatigue and mortality actually exists, and, if so, to identify
potential mechanisms of this association.
Methods: A population-based cohort of 18,101 men and women aged 40–79 years who completed a measure of
fatigue (Short Form 36 vitality domain, SF36-VT) in addition to providing information on possible confounding
factors (age, sex, body mass index, marital status, smoking, education level, alcohol consumption, social class,
depression, bodily pain, diabetes, use of β blockers, physical activity and diet) and mechanisms (haemoglobin,
C-reactive protein and thyroid function) were followed up prospectively for up to 20 years. Mortality from all causes,
cancer and cardiovascular disease was ascertained using death certification linkage with the UK Office of National
Statistics.
Results: During 300,322 person years of follow-up (mean 16.6 years), 4397 deaths occurred. After adjusting for
confounders, the hazard ratio (HR) for all-cause mortality was 1.40 (95 % confidence interval [CI] 1.25–1.56) for those
reporting the highest fatigue (bottom SF36-VT quartile) compared with those reporting the lowest fatigue (top
SF36-VT quartile). This significant association was specifically observed for those deaths related to cardiovascular
disease (HR 1.45, 95 % CI 1.18–1.78) but not cancer (HR 1.09, 95 % CI 0.90–1.32). Of the considered mechanisms,
thyroid function was most notable for attenuating this association. The risk of all-cause mortality, however,
remained significant even after considering all putative confounders and mechanisms (HR 1.26, 95 % CI 1.10–1.45).
Conclusions: High levels of fatigue are associated with excess mortality in the general population. This commonly
dismissed symptom demands greater evaluation and should not automatically be considered benign.
Keywords: Fatigue, Mortality, Cardiovascular, Cancer

* Correspondence: neilbasu@abdn.ac.uk
1
Epidemiology Group, Institute of Applied Health Sciences, School of
Medicine & Dentistry, University of Aberdeen, Foresterhill, Aberdeen,
Scotland AB25 2ZD, UK
Full list of author information is available at the end of the article

© 2016 The Author(s). Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Basu et al. BMC Medicine (2016) 14:122 Page 2 of 8

Background Participants attended for baseline health assessments be-

Fatigue is a normal experience, but for some persons its tween 1993 and 1997 and have since been followed up by
impact is profound, resulting in concerns regarding their way of additional health assessments, questionnaires and
health. In fact, fatigue is one of the commonest com- national registry data linkage. Ethical approval was ob-
plaints reported in primary care, constituting up to 20 % tained from the Norwich Research Ethics Committee.
of all presentations [1, 2] and the principal reason for
consultation in 5-7 % of cases [1–3]. Prognostic factor of interest
The psychosocial consequences of fatigue are becom- Investigating the determinants of mortality is one of the
ing increasingly recognised. It is strongly associated with aims of this cohort. In this specific study, the prognostic
poor long-term quality of life [4] and work ability [5]. In- factor of interest was fatigue, defined according to the
deed, excessively fatigued workers are estimated to cost UK version of the Short Form 36 vitality domain (SF36-
employers $136.4 billion/year in the USA alone [6]. Des- VT). Although there is no single accepted measure of fa-
pite this, patients’ commonly consider the symptom to tigue, the SF36-VT is commonly used and validated in
be ignored by medical practitioners [7] and approxi- studies of both diseased and general populations [13, 14].
mately 75 % of these patients are not followed up [8]. It In fact, it is employed as a gold standard tool to support
may be that doctors are justifiably selective in their the development of other fatigue assessment instruments
follow-up of these patients; however, this paucity of at- [15]. Collected within the 18-month follow-up study ques-
tention may also reflect the perception that fatigue does tionnaire, the SF36-VT consists of four questions which
not affect traditional biomedical outcomes such as pre- refer to the previous 4 weeks: (1) Did you feel full of life?
mature mortality. (2) Did you have a lot of energy? (3) Did you feel worn
A few studies have reported a possible association be- out? (4) Did you feel tired? These were summed and
tween fatigue and mortality, although these suffer from transformed into a 0–100 scale (where a high score repre-
significant methodological limitations: (1) small sample sented low fatigue levels) using established scoring algo-
sizes which prohibit the inspection of cause-specific rithms [14]. For the purposes of analysis, the final scores
mortality, (2) non-generalisable samples, (3) the use of were categorised according to quartiles derived from the
non-validated single item measures of fatigue which pre- whole sample.
clude evaluation of dose response and (4) a limited con-
sideration of potential confounders and mechanisms, Outcome ascertainment
clearly crucial in the context of such a multifactorial Mortality was ascertained using death certification link-
symptom [9, 10]. age with the UK Office of National Statistics (ONS) at
So why could fatigue result in premature mortality? follow-up (censored in 2013). In addition to data on all-
Some have plausibly hypothesised that fatigue leads to cause mortality, deaths specifically attributed to CVD
harmful changes in behaviour [9]. Alternatively, the (defined by International Classification of Disease [ICD],
underlying pathophysiology of the symptom itself could 9:401–448 or ICD 10:I10-I79) or cancer (defined by ICD
lead to poor outcomes – fatigue is a commonly recognised 9:140–208 or ICD 10:C00-C97) were recorded.
feature of dysfunctional physiology, for example it is a
cardinal feature of low rates of metabolism due to Putative confounders/mechanisms
hypothyroidism, which in turn is an established risk factor Multiple putative covariates, commonly considered by
of cardiovascular disease (CVD)-related mortality [11]. clinicians when evaluating the complaint of fatigue, were
The aim of this study was to describe and then under- measured at baseline.
stand the reasons for an association between fatigue and The mechanisms of fatigue are poorly described and
mortality (both all-cause and cause-specific) in a large some may also be considered confounders. We under-
prospective population-based cohort. took a conservative approach by assuming that a factor
fulfilling the statistical criteria for a confounding variable
Methods (i.e. a variable that is significantly associated with both fa-
Study design and sample tigue and mortality) was a confounder. A covariate not
Participants were drawn from the European Prospective fulfilling such criteria was considered to be a possible
Investigation into Cancer and Nutrition (EPIC)-Norfolk mechanism.
study. The study methodology and sample distribution The covariates were collected by trained staff accord-
have been previously reported in detail [12]. In brief, this ing to standardised protocols:
prospective cohort study recruited members of the general
population, aged 40–79 years at the baseline, using general Self-reported measures
practice age-sex registers from the region of Norfolk – a A survey capturing demographic and lifestyle information
relatively stable and immobile population in the UK. was conducted at baseline. This included age, gender and
Basu et al. BMC Medicine (2016) 14:122 Page 3 of 8

marital status. As with previous EPIC analyses, social class fatigue and subsequent all-cause mortality over the follow-
was categorised according to the Registrar General’s occu- up period. First, an unadjusted model was developed
pation classification scheme and further collapsed into (model A). Then, participants with a history of significant
manual or non-manual categories [16]. Education status CVD (stroke and/or myocardial infarction) or cancer at
of participants was defined according to whether or not baseline were excluded in order to avoid major confound-
they reached A-level standard (the senior secondary edu- ing (model B).
cation examination in England). Self-reported total alcohol Using simple descriptive statistics, as appropriate, the
consumption was quantified as the total units of alcohol associations between the covariates of interest and fatigue
(1 unit equates about 8 g) consumed in a week and smok- and mortality, respectively, were calculated. Those covari-
ing as current/non-current. In order to ascertain the ates statistically associated (p < 0.05) with both SF36 vital-
prevalence of baseline morbidity, participants were asked ity and mortality were considered to be confounders and
‘Has your doctor ever told you have the following?’ those not were considered as possible mechanisms. Mech-
followed by a list of illnesses which included cancer, dia- anistic factors are on the causal pathway between an ex-
betes, depression, heart attack and stroke. The SF36 bodily posure and an outcome; if adjusted for in the analysis,
pain domain (simultaneously collected with the SF36-VT) they will attenuate the observed exposure–outcome asso-
was scored and transformed to a 0–100 scale employing ciation. We tested our proposed ‘putative mechanisms’ by
validated methods [14]. As with SF36-VT, a high score adjusting for them in the analysis with the expectation
represented a good health state, that is, low pain. Physical that if they are indeed mechanisms then the magnitude of
activity was assessed using the EPIC-validated question- association between fatigue and mortality will be reduced.
naire [17] which rated participants as inactive, moderately Confounders were grouped and introduced (model C)
inactive, moderately active or active. The consumption of and then the influence of each possible mechanism was
fruit and vegetables (g/day) was determined by the EPIC evaluated by submitting these, in isolation, to this model.
Food Frequency Questionnaire, also a previously well- Subsequently, all putative confounders and mechanisms
described and validated tool [18] which asks participants were advanced together in a single model.
to quantify their average consumption of a number of food To better determine the precise causes of any ob-
items over the past year. Medication use, including aspirin, served mortality, the modelling process was repeated to
was self-reported and captured in the baseline survey. examine the longitudinal relationship between quartiles
of fatigue and subsequent (1) CVD-related mortality and
Clinical measures (2) cancer-related mortality.
Height and weight were assessed and the body mass Last, models were re-run with the exclusion of all
index subsequently calculated (weight[kg]/(height[m2]). participants who died within 2 years of completing
This has been categorised according to World Health SF36-VT in order to investigate the potential for reverse
Organisation definitions [19]. causality, that is, the possibility of a significant subclin-
Non-fasting blood samples were collected. AutoDELFIA ical disease causing fatigue.
time-resolved fluoroimmunoassay kits (Wallac, Finland) Analyses were undertaken using SPSS version 22.0
were used to assay thyroid-stimulating hormone (TSH) (IBM, Chicago, IL, US). The proportional hazard as-
levels (hypothyroid status: >4.0 mU/l) in order to evaluate sumption was checked by introducing an interaction
thyroid status; serum concentrations of the inflammatory term of time and vitality and was found not to be vio-
marker C-reactive protein (CRP) were measured using a lated. All reported p-values are for two-sided significance
high-sensitivity assay on an Olympus AU640 Clinical tests and effect sizes are expressed as hazard ratios (HR)
Chemistry Analyser (Olympus UK Ltd, Watford, UK); and with 95 % confidence intervals (CI).
a Coulter MD18 Haematology Analyzer (Addenbrookes
Hospital, Cambridge, UK) measured haemoglobin (Hb) Results
status to enable assessment for anaemia (moderate an- Of the 25,639 participants who attended the baseline
aemia defined as Hb < 110 g/L [20]). health assessment, complete SF36-VT data were available
for 18,101. In this sample, there were a total of 4397
Statistical analysis deaths (CVD: 1410, cancer: 1573, other causes: 1414) with
The cross-sectional relationship between quartiles of fa- over 300,322 person years of follow-up (mean 16.6 years).
tigue severity and other putative confounders was first Sample characteristics by SF36-VT quartile, where
described using Chi-squared tests (for categorical con- quartile 1 (SF36-VT: 76–100) represents low levels of fa-
founder data) and ANOVA (for continuous confounder tigue and quartile 4 (SF36 VT: 0–49) represents high
data), as appropriate. levels of fatigue, are described in Table 1. Quartile 1 par-
Cox proportional hazards models were then employed to ticipants were statistically more likely to be older, of fe-
examine the longitudinal relationships between quartiles of male sex, married, obese, physically inactive, depressed,
Basu et al. BMC Medicine (2016) 14:122 Page 4 of 8

Table 1 Sample characteristics by SF36-VT quartiles of 18,101 EPIC-Norfolk participants

Characteristic N Total Q1 (76–100) Q2 (65–75) Q3 (50–64) Q4 (0–49) p-value
Agea (years) 18,101 59.0 ± 9.1 59.2 ± 8.7 59.0 ± 9.1 59.2 ± 9.3 58.6 ± 9.4 0.008
Sex 18,101 <0.001
Male 7943 (43.9) 2186 (49.7) 2322 (46.0) 1842 (40.9) 1593 (38.3)
Female 10,158 (56.1) 2209 (50.3) 2725 (54.0) 2659 (59.1) 2565 (61.7)
2
Body mass index (kg/m ) 18,064 <0.001
<18.5 83 (0.5) 13 (0.3) 19 (0.4) 23 (0.5) 28 (0.7)
18.5–25 7234 (40.1) 1891 (43.1) 2078 (41.2) 1721 (38.3) 1544 (37.3)
25–30 8123 (45.0) 2029 (46.3) 2289 (45.4) 2021 (44.9) 1784 (43.1)
>30 2624 (14.5) 451 (10.3) 654 (13.0) 732 (16.3) 787 (19.0)
Education 18,094 <0.001
Lower than A-level 8255 (45.6) 1912 (43.5) 2232 (44.3) 2065 (45.9) 2046 (49.2)
At least A-level 9839 (54.4) 2481 (56.5) 2812 (55.8) 2435 (54.1) 2111 (50.8)
Married 18,009 <0.001
Yes 14,681 (81.5) 3613 (82.6) 4184 (83.4) 3648 (81.5) 3236 (78.2)
Nob 3328 (18.5) 763 (17.4) 834 (16.6) 830 (18.5) 901 (21.8)
Social Class 17,757 <0.001
Manual 6725 (37.9) 1561 (36.2) 1778 (35.9) 1696 (38.4) 1690 (41.5)
Non-manual 11,032 (62.1) 2753 (63.8) 3178 (64.1) 2716 (61.6) 2385 (58.5)
Physical activity 18,101 <0.001
Inactive 5226 (18.7) 1019 (23.2) 1321 (26.2) 1372 (30.5) 1514 (36.4)
Moderately inactive 5287 (23.3) 1246 (28.4) 1519 (30.1) 1324 (29.4) 1198 (28.8)
Moderately active 4210 (29.2) 1092 (24.8) 1230 (24.4) 1022 (22.7) 886 (20.8)
Active 3378 (28.9) 1038 (23.6) 977 (19.4) 783 (17.4) 580 (13.9)
Depression 18,101 <0.001
Yes 2554 (14.1) 315 (7.2) 547 (10.8) 645 (14.3) 1047 (25.2)
No 15,547 (85.9) 4080 (92.8) 4500 (89.2) 3856 (85.7) 3111 (74.8)
Diabetes 18,101 <0.001
Yes 392 (2.2) 77 (1.8) 92 (1.8) 101 (2.2) 122 (2.9)
No 17,709 (97.8) 4318 (98.3) 4955 (98.2) 4400 (97.8) 4036 (97.1)
Bodily pain (SF36) 18,080 74.5 ± 23.0 87.7 ± 15.5 79.7 ± 18.8 70.7 ± 21.8 57.7 ± 24.6 <0.001
Fruit and vegetable consumption (g/day) 17,621 522.6 ± 257.0 539.8 ± 270.6 527.8 ± 245.2 517.8 ± 252.9 503.0 ± 259.3 <0.001
Alcohol consumption (g/day) 17,621 8.6 ± 12.7 9.3 ± 13.4 9.1 ± 12.8 8.2 ± 12.3 7.7 ± 12.4 <0.001
Beta blocker use 18101 <0.001
Yes 1138 (6.3) 215 (4.9) 278 (5.5) 323 (7.2) 322 (7.7)
No 16,963 (93.7) 4180 (95.1) 4769 (94.5) 4178 (92.8) 3836 (92.3)
Thyroid stimulating hormone (mU/L) 9761 0.265
High (≥4) 8857 (90.7) 2191 (90.35) 2445 (90.2) 2246 (91.7) 1975 (90.8)
Low (<4) 904 (9.3) 234 (90.7) 266 (9.8) 204 (8.3) 200 (9.2)
CRP (mg/L) 12,993 3.1 ± 6.2 3.1 ± 5.6 3.1 ± 6.4 2.9 ± 5.4 3.3 ± 7.3 0.223
Haemoglobin (g/dL) 12,971 0.886
High (≥11) 12,793 (98.6) 3138 (98.7) 3553 (98.5) 3220 (98.7) 2882 (98.6)
Low (<11) 178 (1.37) 41 (1.3) 54 (1.5) 43 (1.3) 40 (1.4)
p values indicate levels of significance across all quartiles
Data presented as mean (SD) for all continuous variables and number (%) for categorical variables
a
at SF36 completion bincludes single, widowed, seperated and divorced
Basu et al. BMC Medicine (2016) 14:122 Page 5 of 8

report higher pain, and consume more fruit and vegeta- thyroid function (model E: HR 1.31, 95 % CI 1.03–1.68)
bles when compared to those within quartile 4. In con- and CRP (model F: HR 1.38, 95 % CI 1.07–1.79) attenuated
trast, laboratory measures such as TSH, CRP and Hb the association between high fatigue and mortality over the
were not statistically different across fatigue quartiles period of follow-up. Despite adjusting for all putative con-
and so were not considered to be confounders. All founders and mechanisms, the risk remained pronounced
other covariates fulfilled the criteria for confounding for both all-cause mortality (fully adjusted: HR 1.24, 95 %
(Additional file 1: Table S1). CI 1.05–1.46) and a similar magnitude of risk for CVD-
Table 2 shows the absolute risks and hazard ratios for related mortality (fully adjusted: HR 1.26, 95 % 0.93–1.70).
all-cause mortality by quartiles of SF36-VT. All models Finally, the effect of excluding all deaths occurring
consistently indicated that participants reporting the within 2 years of follow-up (n = 4) upon the final models
highest fatigue (SF36-VT quartile 4) were at significantly was negligible.
higher risk of mortality over the follow-up period than
those reporting the lowest fatigue (SF36-VT quartile 1).
The magnitude of this risk following exclusion of those Discussion
with a history of CVD (n = 1533) and cancer (n = 1410) In this large population-based prospective study with
and then adjustment of other confounders (model C: HR long-term follow-up, fatigue was significantly associated
1.40, 95 % CI 1.25–1.56) was attenuated by thyroid func- with premature mortality, even after adjusting for multiple
tion (model E: HR 1.26, 95 % CI 1.10–1.45) and less so potential confounders. Compared with those in the lowest
by CRP (model F: HR 1.39, 95 % CI 1.21–1.60). quartile of reported fatigue levels, participants in the high-
On further inspection of disease-specific causes of est quarter of reported fatigue levels had a 40 % higher
mortality, participants with high fatigue were at greater risk of premature mortality. This association was observed
risk of CVD-related death (Table 3) (model C: con- for CVD-related, but not cancer-related, deaths. Thyroid
founder adjusted HR 1.45, 95 % CI 1.18–1.78). status and CRP level attenuated these associations al-
In contrast, no significantly increased risk of cancer- though these do not entirely explain the mechanisms
related deaths was observed in any of the adjusted which underpin this concerning observation.
models (Table 4). Because cancer and CVD accounted As with all studies of this nature, the results must be
for the majority of deaths, the apparent excess of mortal- considered in the context of some possible limitations.
ity associated with fatigue appears to be mostly driven First, although general practices in the UK are consid-
by CVD-related deaths. As with all-cause mortality, ered ideal population sample frames (because almost all

Table 2 The risks of all-cause mortality by SF36 vitality score in the EPIC-Norfolk study
Mortality N (%) SF36 vitality score quartile p-valuea
76–100 65–75 50–64 0–49
Model A: unadjusted full dataset 4397/18,101 (24.3 %) 928/4395 (21.1 %) 1103/5047 (21.9 %) 1154/4501 (25.6 %) 1212/4158 (29.1 %) <0.001
HR (95 % CI)
1 (reference) 1.04 (0.96–1.14) 1.27 (1.17–1.39) 1.49 (1.36–1.62)
Model B: unadjusted with removal 3637/16,409 (22.2 %) 825/4097 (20.1 %) 945/4657 (20.3 %) 944/4049 (23.3 %) 923/3606 (25.6 %) <0.001
of baseline CVD/cancer
HR (95 % CI) 1 (reference) 1.01 (0.92–1.11) 1.20 (1.10–1.32) 1.34 (1.22–1.47)

Model C: confounders 3388/15,577 (21.8 %) 771/3899 (19.8 %) 886/4442 (19.9 %) 880/3837 (22.9 %) 851/3399 (25.0 %) <0.001
HR (95 % CI)
1 (reference) 1.02 (0.93–1.13) 1.21 (1.10–1.34) 1.40 (1.25–1.56)
Model D: confounders + Hb 3388/15,577 (21.8 %) 771/3899 (19.8 %) 886/4442 (19.9 %) 880/3837 (22.9 %) 851/3399 (25.0 %) <0.001
HR (95 % CI)
1(reference) 1.01 (0.91–1.13) 1.24 (1.12–1.39) 1.40 (1.24–1.58)
Model E: confounders + TSH 3388/15,577 (21.8 %) 771/3899 (19.8 %) 886/4442 (19.9 %) 880/3837 (22.9 %) 851/3399 (25.0 %) <0.001
HR (95 % CI)
1 (reference) 0.97 (0.86–1.09) 1.23 (1.09–1.39) 1.26 (1.10–1.45)
Model F: confounders + CRP 3388/15,577 (21.8 %) 771/3899 (19.8 %) 886/4442 (19.9 %) 880/3837 (22.9 %) 851/3399 (25.0 %) <0.001
HR (95 % CI)
1 (reference) 1.06 (0.94–1.19) 1.24 (1.10–1.40) 1.39 (1.21–1.60)
Model G: confounders + Hb, 3388/15,577 (21.8 %) 771/3899 (19.8 %) 886/4442 (19.9 %) 880/3837 (22.9 %) 851/3399 (25.0 %) 0.001
CRP, TSH
HR (95 % CI) 1 (reference) 1.00 (0.87–1.15) 1.26 (1.09–1.46) 1.24 (1.05–1.46)
a
Across quartiles
Model A: unadjusted; Model B: unadjusted with removal of baseline CVD/cancer; Model C: Model B + statistically proven confounders – age, gender, body mass
index, marital status, smoking, education, alcohol, social class, diet, physical activity, beta blocker, depression, pain, diabetes; Model D: Model C plus Hb adjusted;
Model E: Model C plus TSH adjusted; Model F: Model C plus CRP adjusted; Model G: Model C plus Hb, CRP and TSH adjusted
CI confidence interval, CRP C-reactive protein, CVD cardiovascular disease, Hb haemoglobin, HR hazard ratio, TSH thyroid-stimulting hormone
Basu et al. BMC Medicine (2016) 14:122 Page 6 of 8

Table 3 The risks of cardiovascular disease related mortality by SF-36 vitality score in the EPIC-Norfolk study
Mortality SF-36 vitality score quartiles p-valuea
76–100 65–75 50–64 0–49
Model A: unadjusted full dataset 1 (reference) 0.98 (0.84–1.15) 1.25 (1.07–1.46) 1.63 (1.40–1.89) <0.001
HR (95 % CI)
Model B: unadjusted with removal 1 (reference) 0.92 (0.77–1.09) 1.18 (1.00–1.40) 1.41 (1.19–1.67) <0.001
of baseline CVD/cancer
HR (95 % CI)
Model C: confounders 1 (reference) 0.93 (0.78–1.12) 1.21 (1.00–1.45) 1.45 (1.18–1.78) <0.001
HR (95 % CI)
Model D: confounders + Hb 1(reference) 0.93 (0.77–1.13) 1.21 (1.00–1.48) 1.43 (1. 15–1.78) <0.001
HR (95 % CI)
Model E: confounders + TSH 1 (reference) 0.85 (0.68–1.05) 1.22 (0.98–1.52) 1.31 (1.03–1.68) 0.001
HR (95 % CI)
Model F: confounders + CRP 1 (reference) 0.96 (0.77–1.20) 1.16 (0.92–1.45) 1.38 (1.07–1.79) 0.021
HR (95 % CI)
Model G: confounders + Hb, CRP, TSH 1 (reference) 0.84 (0.64–1.09) 1.20 (0.92–1.56) 1.26 (0.93–1.70) 0.025
HR (95 % CI)
a
Across quartiles
Model A: unadjusted; Model B: unadjusted with removal of baseline CVD/cancer; Model C: Model B + statistically proven confounders – age, gender, body mass
index, marital status, smoking, education, alcohol, social class, diet, physical activity, beta blocker, depression, pain, diabetes; Model D: Model C plus Hb adjusted;
Model E: Model C plus TSH adjusted; Model F: Model C plus CRP adjusted; Model G: Model C plus Hb, CRP and TSH adjusted
CI confidence interval, CRP C-reactive protein, CVD cardiovascular disease, Hb haemoglobin, HR hazard ratio, TSH thyroid-stimulting hormone

people are registered to a general practitioner through Overall, these results appear generalisable when applied
the National Health Service), only 40–45 % of registrants to middle- to old-aged members of the general popula-
provided baseline data, of which approximately 30 % did tion, although, owing to the study design, they are un-
not complete the SF36-VT questionnaire. Nonetheless, likely to fully apply to typical healthcare attendees
previous descriptions of this cohort have failed to iden- principally presenting with significant fatigue. It is antici-
tify significant differences in population characteristics pated that such persons will complain of more severe
when compared to other national samples, except for a levels of fatigue compared to this study’s modest defin-
slightly lower prevalence of smokers [12], and no mater- ition of high fatigue (a self-reported score within the top
ial differences in background characteristics were identi- 25 % of the general population over a period of 1 month).
fied between SF36-VT responders and non-responders. Chronic fatigue disorders, for example, are frequently

Table 4 The risks of cancer related mortality by SF36 vitality score in the EPIC-Norfolk study
Mortality SF36 vitality score quartiles p valuea
76–100 65–75 50–64 0–49
Model A: unadjusted full dataset 1 (reference) 0.97 (0.84–1.16) 1.16 (1.01–1.33) 1.10 (0.95–1.27) 0.043
HR (95 % CI)
Model B: unadjusted with removal 1 (reference) 0.95 (0.82–1.10) 1.10 (0.95–1.28) 0.98 (0.83–1.15) 0.237
of baseline CVD/cancer
HR (95 % CI)
Model C: confounders 1 (reference) 0.96 (0.83–1.13) 1.14 (0.96–1.34) 1.09 (0.90–1.32) 0.185
HR (95 % CI)
Model D: confounders + Hb 1(reference) 0.94 (0.79–1.11) 1.16 (0.97–1.38) 1.12 (0.91–1.38) 0.086
HR (95 % CI)
Model E: confounders + TSH 1 (reference) 0.91 (0.75–1.10) 1.13 (0.93–1.38) 0.98 (0.77–1.24) 0.132
HR (95 % CI)
Model F: confounders + CRP 1 (reference) 1.06 (0.88–1.27) 1.18 (0.97–1.44) 1.09 (0.87–1.37) 0.385
HR (95 % CI)
Model G: confounders + Hb, CRP,TSH 1 (reference) 1.04 (0.84–1.29) 1.19 (0.95–1.51) 0.86 (0.65–1.15) 0.081
HR (95 % CI)
a
Across quartiles
Model A: unadjusted; Model B: unadjusted with removal of baseline CVD/cancer; Model C: Model B + statistically proven confounders – age, gender, body mass
index, marital status, smoking, education, alcohol, social class, diet, physical activity, beta blocker, depression, pain, diabetes; Model D: Model C plus Hb adjusted;
Model E: Model C plus TSH adjusted; Model F: Model C plus CRP adjusted; Model G: Model C plus Hb, CRP and TSH adjusted
CI confidence interval, CRP C-reactive protein, CVD cardiovascular disease, Hb haemoglobin, HR hazard ratio, TSH thyroid-stimulting hormone
Basu et al. BMC Medicine (2016) 14:122 Page 7 of 8

characterised as disabling and lasting for >6 months multidimensional construct consisting of emotional ex-
[21]. Because our data evidences a consistent dose re- haustion, cognitive weariness and physical fatigue. Intri-
sponse between fatigue severity and mortality, these pa- guingly, this is also associated with all-cause mortality, in
tients are likely to be at even greater risk of premature particular CVD [23]. This state has been characterised by
death in comparison to the population described in this reduced hypothalamic pituitary adrenal axis responses
study. These data may, however, not be used to infer the [24], low grade inflammation [25] and impaired fibrino-
prognosis of patients with chronic fatigue syndrome be- lytic capacities [26], all of which have also been independ-
cause information on total fatigue duration was not ently associated with CVD [27–29]. It is interesting that
measured and this may also represent a distinct disorder. our study also identified inflammation, as measured by
Indeed, recent evidence suggests that such patients have CRP, to be a potentially relevant mechanism. We have also
a different risk factor profile compared to the general identified the importance of TSH. Of the multiple func-
population in terms of mortality causes [22]. tions of this hormone, its positive influence on the basal
Second, there is the potential for pre-existing baseline metabolic rate and on catecholamine sensitivity best ex-
morbidities, especially CVD, to exaggerate the true associ- plain why low levels commonly manifest as fatigue [30].
ation between fatigue and premature mortality. We Furthermore, the hypothyroid state is associated with ac-
attempted to address such reverse causality within the celerated atherosclerosis, systemic vascular resistance and
analysis by excluding participants who reported baseline subsequent CVD [31]. Though this study does not directly
CVD and cancer. To further exclude the influence of pos- inform whether thyroxine replacement (in those with high
sible subclinical disease, deaths within 2 years were ex- abnormal TSH level) would attenuate the mortality risk of
cluded and baseline aspirin use, a surrogate marker of fatigue, the data in combination with existing biological
high CVD risk, was submitted as a covariate. Neither re- knowledge does strongly allude to this.
sulted in a meaningful modification of the final models. The persisting significant risk despite adjustment of al-
Third, there are potential measurement errors in the most all routinely considered associations highlights the
assessment of exposures. We only used a single measure need for greater therapeutic options in this field. This can-
at one time point to assess each confounder and so did not be achieved without further research in this area to
not take into account possible changes over the follow- unravel the likely complex mechanisms. Based on our
up period. That said, such random measurement error findings, such biological mechanisms may be shared with
would probably attenuate any associations observed, so CVD, providing a starting point for further investigation,
the estimated differences in risk are likely to be larger but other behavioural and psychosocial factors, for ex-
than those observed. ample work-shift patterns, should also be considered.
In relation to other prognostic factors of mortality, which Being in the highest quartile for self-reported fatigue is
are already considered key targets for intervention in significantly predictive of death and on a par with recog-
modern-day health services, the observed unadjusted mor- nised risk factors such as active smoking. The risk is
tality risk of high fatigue (HR 1.63, 95 % CI 1.40–1.89) is likely to be even greater for those presenting with more
similar to being a current smoker (HR 1.35, 95 % CI 1.24– severe symptoms and is especially associated with CVD-
1.48, see Additional file 1: Table S2) in this same sample related death. This study therefore indicates the poten-
(although direct comparisons of HRs are not possible tial seriousness of this frequently dismissed symptom
owing to differences in the measurement scale). These re- [7]. The attenuation of risk by both behavioural and clin-
sults are even more meaningful from a public health per- ical factors in our models suggests a potential benefit of
spective given the highly prevalent nature of fatigue. Hardy a comprehensive bio-psychosocial approach. In the fu-
et al. [10] and Moreh et al. [9] also reported similarly sized ture, fatigued persons may benefit from both non-
effects of fatigue on mortality (adjusted HR 1.40, 95 % CI pharmacological interventions, for example physical
1.08–1.93 and 1.31, 0.92–1.86, respectively) although their activity, and pharmacological interventions, for example
study populations were restricted to the elderly and were anti-inflammatory agents.
small (n = 492 and n = 605, respectively), so did not deter-
mine cause-specific mortality [9, 10]. Furthermore, both Conclusions
utilised an unvalidated single item question to measure fa- This study presents the first reliable evidence linking the
tigue, failed to investigate key putative mechanisms such as common symptom of fatigue with excess mortality. Patients
thyroid status and inflammation, and did not account for have long recognised the importance of fatigue [7]. The
the possibility of reverse causality as we have. biomedical community should now align with this pa-
Knowledge regarding the mechanisms which cause fa- tient priority and more optimally research this problem
tigue is limited and the precise pathophysiology of fatigue because further elucidation and targeting of the mecha-
has yet to be elucidated. Possible insights may be derived nisms may not only improve crucial patient reported
from research examining the concept of ‘burnout’ – a outcomes but conceivably improve life longevity as well.
Basu et al. BMC Medicine (2016) 14:122 Page 8 of 8

Additional file 7. Meeuwesen L, Bensing J, van den Brink-Muinen A. Communicating fatigue in

general practice and the role of gender. Patient Educ Couns. 2002;48(3):233–42.
8. Kenter EG, Okkes IM, Oskam SK, Lamberts H. Tiredness in Dutch family
Additional file 1: Table S1. Associations between baseline covariates
practice. Data on patients complaining of and/or diagnosed with
and mortality. Table S2. The risks of all-cause mortality by smoking.
“tiredness”. Fam Pract. 2003;20(4):434–40.
(DOCX 33 kb)
9. Moreh E, Jacobs JM, Stessman J. Fatigue, function, and mortality in older
adults. J Gerontol A Biol Sci Med Sci. 2010;65(8):887–95.
Abbreviations 10. Hardy SE, Studenski SA. Fatigue predicts mortality in older adults. J Am
CI, confidence interval; CRP, C-reactive protein; CVD, cardiovascular disease; Geriatr Soc. 2008;56(10):1910–4.
EPIC, European Prospective Investigation into Cancer and Nutrition; Hb, 11. Rodondi N, den Elzen WP, Bauer DC, Cappola AR, Razvi S, Walsh JP, Asvold
haemoglobin; HR, hazard ratio; ICD, International Classification of Disease; BO, Iervasi G, Imaizumi M, Collet TH, et al. Subclinical hypothyroidism and the
ONS, Office for National Statistics; SF36-VT, Short form 36 vitality, TSH, risk of coronary heart disease and mortality. JAMA. 2010;304(12):1365–74.
thyroid-stimulating hormone 12. Day N, Oakes S, Luben R, Khaw KT, Bingham S, Welch A, Wareham N. EPIC-
Norfolk: study design and characteristics of the cohort. European
Prospective Investigation of Cancer. Br J Cancer. 1999;80(1):95–103.
Acknowledgements
13. Neuberger GB. Measures of fatigue. Arthritis Care Res. 2003;45(5S):175–83.
We gratefully thank the participants of the EPIC-Norfolk study. We also thank
14. Ware JE. SF-36 health survey: manual and interpretation guide. Lincoln:
the collaborating general practices, staff of the EPIC-Norfolk and our funders.
QualityMetric Inc; 2000.
15. Donovan KA, Jacobsen PB, Small BJ, Munster PN, Andrykowski MA. Identifying
Funding clinically meaningful fatigue with the Fatigue Symptom Inventory. J Pain
This work was supported by programme grants from the Medical Research Symptom Manag. 2008;36(5):480–7.
Council G1000143 and the Cancer Research UK 8257. Funders had no role in 16. Myint PK, Luben RN, Wareham NJ, Bingham SA, Khaw KT. Combined effect of
study design, analysis or interpretation of the findings. health behaviours and risk of first ever stroke in 20,040 men and women over
11 years’ follow-up in Norfolk cohort of European Prospective Investigation of
Authors’ contributions Cancer (EPIC Norfolk): prospective population study. BMJ. 2009;338:b349.
PKM and NB conceived the idea. KTK and NJW are the principal investigators 17. Myint PK, Luben RN, Wareham NJ, Welch AA, Bingham SA, Day NE, Khaw
of the EPIC-Norfolk study. RNL performed the record linkage. XY performed the KT. Combined work and leisure physical activity and risk of stroke in men
statistical analyses. NB and DW drafted the manuscript. PKM had full access to and women in the European prospective investigation into Cancer-Norfolk
all of the data in the study and takes responsibility for the integrity of Prospective Population Study. Neuroepidemiology. 2006;27(3):122–9.
the data and the accuracy of the data analysis. All authors contributed to the 18. Myint PK, Welch AA, Bingham SA, Surtees PG, Wainwright NW, Luben RN,
interpretation of results and in making an important intellectual contribution to Wareham NJ, Smith RD, Harvey IM, Day NE, et al. Fruit and vegetable
the manuscript. All authors read and approved the final manuscript. consumption and self-reported functional health in men and women in the
European Prospective Investigation into Cancer-Norfolk (EPIC-Norfolk): a
Competing interests population-based cross-sectional study. Public Health Nutr. 2007;10(1):34–41.
The authors declare that they have no competing interests. 19. Global database on body mass index [http://apps.who.int/bmi/index.
jsp?introPage=intro_3.html]. Accessed 1 Oct 2015.
20. Haemoglobin concentrations for the diagnosis of anaemia and assessment
Ethics approval and consent to participate
of severity [http://www.who.int/vmnis/indicators/haemoglobin.pdf].
Ethical approval was obtained from the Norwich Research Ethics Committee,
Accessed 1 Oct 2015.
UK, and written informed consent was obtained from patients according to
21. Afari N, Buchwald D. Chronic fatigue syndrome: a review. Am J Psychiatry.
the Declaration of Helsinki.
2003;160(2):221–36.
22. Roberts E, Wessely S, Chalder T, Chang CK, Hotopf M. Mortality of people
Author details
1 with chronic fatigue syndrome: a retrospective cohort study in England and
Epidemiology Group, Institute of Applied Health Sciences, School of
Wales from the South London and Maudsley NHS Foundation Trust
Medicine & Dentistry, University of Aberdeen, Foresterhill, Aberdeen,
Biomedical Research Centre (SLaM BRC) Clinical Record Interactive Search
Scotland AB25 2ZD, UK. 2Department of Public Health and Primary Care,
(CRIS) Register. Lancet. 2016;387:1638–43.
Strangeway Research Laboratory, University of Cambridge, Cambridge, UK.
3 23. Prescott E, Holst C, Gronbaek M, Schnohr P, Jensen G, Barefoot J. Vital
Department of Public Health and Primary Care, Clinical Gerontology Unit,
exhaustion as a risk factor for ischaemic heart disease and all-cause mortality in
University of Cambridge, Cambridge, UK. 4MRC Epidemiology Unit,
a community sample. A prospective study of 4084 men and 5479 women in
Cambridge, UK.
the Copenhagen City Heart Study. Int J Epidemiol. 2003;32(6):990–7.
24. Melamed S, Shirom A, Toker S, Berliner S, Shapira I. Burnout and risk of
Received: 29 May 2016 Accepted: 28 July 2016
cardiovascular disease: evidence, possible causal paths, and promising
research directions. Psychol Bull. 2006;132(3):327–53.
25. Black PH, Garbutt LD. Stress, inflammation and cardiovascular disease. J
References Psychosom Res. 2002;52(1):1–23.
1. Cathebras PJ, Robbins JM, Kirmayer LJ, Hayton BC. Fatigue in primary care: 26. Kop WJ, Hamulyak K, Pernot C, Appels A. Relationship of blood coagulation
prevalence, psychiatric comorbidity, illness behavior, and outcome. J Gen and fibrinolysis to vital exhaustion. Psychosom Med. 1998;60(3):352–8.
Intern Med. 1992;7(3):276–86. 27. Tracy RP. Inflammation, the metabolic syndrome and cardiovascular risk. Int
2. Cullen W, Kearney Y, Bury G. Prevalence of fatigue in general practice. Ir J J Clin Pract Suppl. 2003;134:10–7.
Med Sci. 2002;171(1):10–2. 28. Hansson GK. Inflammation, atherosclerosis, and coronary artery disease. N
3. Gallagher AM, Thomas JM, Hamilton WT, White PD. Incidence of fatigue Engl J Med. 2005;352(16):1685–95.
symptoms and diagnoses presenting in UK primary care from 1990 to 2001. 29. Rosmond R, Wallerius S, Wanger P, Martin L, Holm G, Bjorntorp P. A 5-year
J R Soc Med. 2004;97(12):571–5. follow-up study of disease incidence in men with an abnormal hormone
4. Nelson E, Kirk J, McHugo G, Douglass R, Ohler J, Wasson J, Zubkoff M. Chief pattern. J Int Med. 2003;254(4):386–90.
complaint fatigue: a longitudinal study from the patient's perspective. Fam 30. Kronenberg H, Williams RH. Williams textbook of endocrinology. 11th ed.
Pract Res J. 1987;6(4):175–88. Henry M. Kronenberg … [et al.] editor. Philadelphia: Elsevier Saunders; 2008.
5. Janssen N, Kant IJ, Swaen GMH, Janssen PPM, Schroer CAP. Fatigue as a 31. Gomberg-Maitland M, Frishman WH. Thyroid hormone and cardiovascular
predictor of sickness absence: results from the Maastricht cohort study on disease. Am Heart J. 1998;135(2 Pt 1):187–96.
fatigue at work. Occup Environ Med. 2003;60:71–6.
6. Ricci JA, Chee E, Lorandeau AL, Berger J. Fatigue in the US workforce:
prevalence and implications for lost productive work time. J Occup Environ
Med. 2007;49(1):1–10.