Articles

Socioeconomic status and the 25 25 risk factors as

determinants of premature mortality: a multicohort study
and meta-analysis of 17 million men and women
Silvia Stringhini*, Cristian Carmeli*, Markus Jokela*, Mauricio Avendao*, Peter Muennig, Florence Guida, Fulvio Ricceri, Angelo dErrico,
Henrique Barros, Murielle Bochud, Marc Chadeau-Hyam, Franoise Clavel-Chapelon, Giuseppe Costa, Cyrille Delpierre, Silvia Fraga, Marcel Goldberg,
Graham G Giles, Vittorio Krogh, Michelle Kelly-Irving, Richard Layte, Aurlie M Lasserre, Michael G Marmot, Martin Preisig, Martin J Shipley,
Peter Vollenweider, Marie Zins, Ichiro Kawachi, Andrew Steptoe, Johan P Mackenbach, Paolo Vineis,Mika Kivimki, for the LIFEPATH consortium

Summary
Background In 2011, WHO member states signed up to the 25 25 initiative, a plan to cut mortality due to non- Published Online
communicable diseases by 25% by 2025. However, socioeconomic factors inuencing non-communicable diseases January 31, 2017
http://dx.doi.org/10.1016/
have not been included in the plan. In this study, we aimed to compare the contribution of socioeconomic status to S0140-6736(16)32380-7
mortality and years-of-life-lost with that of the 25 25 conventional risk factors.
See Online/Comment
http://dx.doi.org/10.1016/
Methods We did a multicohort study and meta-analysis with individual-level data from 48 independent prospective S0140-6736(17)30191-5
cohort studies with information about socioeconomic status, indexed by occupational position, 25 25 risk factors *These authors contributed
(high alcohol intake, physical inactivity, current smoking, hypertension, diabetes, and obesity), and mortality, for a equally to this work
total population of 1 751 479 (54% women) from seven high-income WHO member countries. We estimated the Joint last authors
association of socioeconomic status and the 25 25 risk factors with all-cause mortality and cause-specic mortality by Members are listed at end of
calculating minimally adjusted and mutually adjusted hazard ratios [HR] and 95% CIs. We also estimated the paper

population attributable fraction and the years of life lost due to suboptimal risk factors. Institute of Social and
Preventive Medicine and
Departments of Psychiatry and
Findings During 266 million person-years at risk (mean follow-up 133 years [SD 64 years]), 310 277 participants Internal Medicine, Lausanne
died. HR for the 25 25 risk factors and mortality varied between 104 (95% CI 098111) for obesity in men and University Hospital, Lausanne,
2 17 (206229) for current smoking in men. Participants with low socioeconomic status had greater mortality Switzerland (S Stringhini PhD,
C Carmeli PhD,
compared with those with high socioeconomic status (HR 142, 95% CI 138145 for men; 134, 128139 for Prof M Bochud PhD,
women); this association remained signicant in mutually adjusted models that included the 25 25 factors (HR 126, A M Lasserre MD,
121132, men and women combined). The population attributable fraction was highest for smoking, followed by Prof M Preisig MD,
physical inactivity then socioeconomic status. Low socioeconomic status was associated with a 21-year reduction in Prof P Vollenweider MD);
Institute of Behavioural
life expectancy between ages 40 and 85 years, the corresponding years-of-life-lost were 05 years for high alcohol Sciences, University of Helsinki,
intake, 07 years for obesity, 39 years for diabetes, 16 years for hypertension, 24 years for physical inactivity, and Helsinki, Finland
48 years for current smoking. (M Jokela PhD); Department of
Global Health and Social
Medicine, Kings College
Interpretation Socioeconomic circumstances, in addition to the 25 25 factors, should be targeted by local and global London, London, UK
health strategies and health risk surveillance to reduce mortality. (M Avendao PhD); Harvard T H
Chan School of Public Health,
Funding European Commission, Swiss State Secretariat for Education, Swiss National Science Foundation, the Boston MA, USA (M Avendao,
Prof I Kawachi PhD); Global
Medical Research Council, NordForsk, Portuguese Foundation for Science and Technology. Research Analytics for
Population Health, Health
Copyright The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY license. Policy and Management,
Columbia University, New
York, NY, USA
Introduction disease and injury attributable to 67 risk factors in (Prof P Muennig MD); MRC-PHE
The 201320 World Health Organization (WHO) Global 21 world-regions.2 Despite the fact that low socioeconomic Centre for Environment and
Action Plan for the Prevention and Control of Non- status is one of the strongest predictors of morbidity and Health, School of Public Health,
Department of Epidemiology
Communicable Diseases (NCDs) targets seven major premature mortality worldwide,36 poor socioeconomic
and Biostatistics, Imperial
risk factors, comprising the harmful use of alcohol, circumstances are not considered modiable risk factors College London, London, UK
insucient physical activity, current tobacco use, raised in these important global health strategies. (F Guida PhD,
blood pressure, intake of salt or sodium, diabetes, and Socioeconomic circumstances and their consequences M Chadeau-Hyam PhD,
Prof P Vineis PhD);
obesity (referred to as the 25 25 risk factors), with the are modiable by policies at the local, national, and
Epidemiology Unit, ASL TO3
overall aim of reducing premature mortality from non- international levels,79 as are risk factors targeted by Piedmont Region, Grugliasco,
communicable diseases by 25% by 2025.1 Similarly, the existing global health strategies. Evidence also suggests Italy (F Ricceri PhD,
Global Burden of Disease (GBD) Collaboration, the that the burden of most 25 25 risk factors is concentrated A dErrico MD);
EPIUnitInstitute of Public
largest study monitoring health changes globally, in lower socioeconomic groups worldwide.10,11 Inter-
Health, University of Porto,
performs an annual risk assessment of the burden of ventions to reduce premature mortality attributable to

www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7 1

 Articles
Porto, Portugal
(Prof H Barros PhD, S Fraga PhD);
Department of Clinical
Epidemiology, Predictive
Medicine and Public Health,
University of Porto Medical
School, Porto, Portugal
(Prof H Barros); Center for
Research in Epidemiology and
Population Health, INSERM
U1018, Villejuif, France
(F Clavel-Chapelon PhD);
Department of Biological and
Clinical Sciences, Universtiy of
Turin, Turin, Italy
(Prof G Costa MD); INSERM,
UMR1027, Toulouse, France
(C Delpierre PhD,
M Kelly-Irving PhD); Universit
Toulouse III Paul-Sabatier,
UMR1027, Toulouse, France
(C Delpierre, M Kelly-Irving);
Population-based
Epidemiological Cohorts Unit,
INSERM UMS 11, Villejuif,
France (Prof M Goldberg MD,
M Zins MD); Paris Descartes
University, Paris, France
(Prof M Goldberg, M Zins);
Cancer Epidemiology Centre,
Cancer Council Victoria,
Melbourne, VIC, Australia
(Prof G G Giles PhD);
Epidemiology and Prevention
Unit, Fondazione IRCCS Istituto
Nazionale dei Tumori, Milan,
Italy (V Krogh MD); Department
of Sociology, Trinity College
Dublin, Dublin, Ireland
(R Layte PhD); University College
London, Department of
Epidemiology and Public
Health, London, UK
(Prof M G Marmot FRCP,
M J Shipley MSc,
Prof A Steptoe Dsc,
Prof M Kivimki PhD);
Department of Public Health,
Erasmus University Medical
Center, Rotterdam, Netherlands
(Prof J P Mackenbach PhD); and
Clinicum, Faculty of Medicine,
University of Helsinki, Finland
(Prof M Kivimki)
Correspondence to:
Dr Silvia Stringhini, Institute of
Social and Preventive Medicine
(IUMSP), Lausanne University
Hospital, 10 Route de la
Corniche, Lausanne 1010,
Switzerland
silvia.stringhini@chuv.ch
See Online for appendix
For more on the Lifepath
project see http://www.
lifepathproject.eu/
For more on ESEC see https://
www.iser.essex.ac.uk/archives/
esec/user-guide
2 www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7
Research in context
Evidence before this study
Low socioeconomic status is one of the strongest predictors of
morbidity and premature mortality worldwide. However, global
health strategies do not consider poor socioeconomic
circumstances as modiable risk factors. The WHO Global Action
Plan for the Prevention and Control of Non-Communicable
Diseases, for example, targets seven major health risk factors,
including insucient physical activity, current tobacco use. and
raised blood pressure, for reducing premature mortality from
non-communicable diseases by 25% by 2025. Low socioeconomic
status is not included among the 25 25 risk factors.
Added value of this study
We used data from more than 17 million individuals in
48 independent cohort studies from seven countries, and
found that the independent association between
socioeconomic status and mortality is comparable in strength
the 25 25 and other risk factors might therefore benet
from greater focus on socioeconomic adversity so that
the preventive toolkit for addressing NCDs can be
expanded. To examine this hypothesis, we collated
individual-level data from 48 independent prospective
cohort studies from Europe, the USA, and Australia and
aimed to determine the population attributable fraction
(PAF) and years of life lost (YLLs) due to low socio-
economic status and compared these with mortality and
YLLs attributable to the 25 25 risk factors.
Methods
Study population
This study is part of an EC Horizon 2020 consortium, the
Lifepath project, which includes ten cohort studies. We
have complemented those data with publicly available
data from 38 additional cohort studies from the Inter-
University Consortium for Political and Social Research
and the UK Data Service. Our analyses were based on
participants whose occupational position was assessed at
baseline between 1965 and 2009, dependent on the study
(appendix). The 48 studies comprised a total population
of 1 751 479 men and women from seven WHO member
countries (UK, France, Switzerland, Portugal, Italy, USA,
Australia). All studies included baseline data for
socioeconomic status and a mortality follow-up of a
minimum of 3 years. Each study was approved by the
relevant local or national ethics committees and all
participants gave informed consent to participate. We
assessed the quality of included studies using the
Cochrane Risk of Bias Tool for cohort studies.12 We
analysed a selection of exposed and non-exposed groups,
assessment of exposure, exclusion of the outcome of
interest at study baseline, adjustment for confounding
variables, assessment of confounding variables,
assessment of outcome, and adequacy of the follow-up.
and consistency to those of six 25 25 risk factors (tobacco use,
alcohol consumption, insucient physical activity, raised blood
pressure, obesity, diabetes). Our study is one of the largest
studies to date to examine the association between
socioeconomic status and premature mortality and the rst
large-scale investigation to directly compare the importance of
socioeconomic circumstances as determinants of health with
six major risk factors targeted in global health strategies for the
reduction of premature mortality.
Implications of all the available evidence
By showing comparable health impact of low socioeconomic
status to that of major risk factors, our study suggests that
socioeconomic adversity should be included as a modiable risk
factor in local and global health strategies, policies, and
health-risk surveillance.
Two reviewers (SS and MK) independently assessed the
studies. The quality of the study was judged as high if all
domains were assessed favourably (appendix).
Denitions and data collection
Our measure of socioeconomic status is a social class
measure based on an individuals last known
occupational title at study enrolment, coded into the
European Socio-economic Classication (ESEC). This
variable was predened and harmonised across the
study cohorts before statistical analyses were done.
Occupational position was categorised as high (higher
professionals and managers, higher clerical, services,
and sales workers [ESEC class 1, 2, and 3]), intermediate
(small employers and self-employed, farmers, lower
supervisors, and technicians [ESEC class 4, 5, and 6]), or
low (lower clerical, services and sales workers, skilled
workers, and semi-skilled and unskilled workers [ESEC
class 7, 8, and 9]). For one study (E3N), occupational
position was current occupation 2 years after baseline.
We used ESEC as a classication because it eliminates
the need to adjust for dierences in earnings and
standards of living across dierent national contexts. We
used individuals occupational class only because most
cohorts did not collect information about partners
occupation. This decision could have led to some
misclassication of socioeconomic status particularly for
older women with low labour force participation rates.
Each 25 25 risk factor comprised two or three
categories to allow a balanced comparison with socio-
economic status, which was grouped into three categories
(appendix). Self-reported smoking was categorised into
current smoker, former smoker, and never smoked.
Alcohol consumption was measured in alcohol units per
week and participants were categorised as abstainers
(0 units per week), moderate (121 units per week for

men, 114 per week for women), or heavy (>21 units per
week for men, >14 per week for women) drinkers.
Although physical activity was measured with dierent
questions in each study, a dichotomised variable
indicating the presence or absence of physical activity
was dened (appendix). Body-mass index (BMI) was
categorised as normal (185<25 kg/m), overweight
(25<30 kg/m), or obese (30 kg/m). Hypertension
was dened as the presence of at least one of the
following conditions: systolic blood pressure more than
140 mm Hg, diastolic blood pressure more than
90 mm Hg, current intake of anti-hypertensive
medication, or self-reported hypertension. Diabetes was
dened as the presence of at least one of the following
conditions: fasting glucose more than 7 mmol/L, 2 h
post-load glucose above 111 mmol/L, glycated
haemoglobin A1c more than 65%, or self-reported
diabetes. Data for salt intake were only available from
less than a third of the cohort studies; we therefore
omitted this risk factor from our analysis.
We considered age, sex, race or ethnicity, and marital
status as potential confounders. Race or ethnicity was
categorised as white and non-white individuals. Marital
status was categorised as married or cohabiting versus
living alone.
Participants were linked to national mortality registries
that provided information about vital status with the
exception of the COLAUS study in which vital status was
ascertained through active follow-up. Mean follow-up for
mortality ranged between 32 years in the National
Health Interview Survey 2009, and 270 years in men and
295 years in women of the Alameda County Study 1965,
with a mean across cohorts of 133 years [SD 64 years].
All-cause mortality, cancer mortality, cardiovascular
disease mortality, and mortality from other causes of
death were examined separately. We focus on cancer and
cardiovascular disease as these diseases are the most
common causes of death in our samples. We used codes
from the International Classication of Diseases, 10th
Revision (ICD-10) to dene cancer (C00C97) and
cardiovascular disease (I00I99) mortality. Other causes
of death include all remaining deaths not classied as
cancer or cardiovascular disease.
Statistical analysis
Analyses were rst performed separately in each study;
estimates were subsequently combined in a meta-
analytical framework. In study-specic analyses, we
considered the maximum number of participants
without missing values for each exposure. To estimate
the association between risk factors and mortality, hazard
ratios (HR) and 95% CIs were generated using exible
parametric survival models on the cumulative hazards
scale,13 which, in addition to the HRs, allow direct
estimation of the conditional cumulative hazard function.
Within these models, we used restricted cubic splines
with 0 to 4 (depending on the cohort) internal knots to
www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7
Articles
model the baseline hazard using age as the timescale.
Separate models were tted for men and women and
included marital status and race or ethnicity (minimally
adjusted models). To check for the proportional hazard
assumption, we performed tests based on Schoenfeld
residuals and inspected log-log plots of Kaplan-Meier
curves. Age stratication in 5-year intervals was
conducted in all cohorts as a sensitivity analysis to adjust
for age calendar eects (results not shown).
In further analyses combining men and women, we
examined the association of socioeconomic status with
cause-specic mortality before and after adjustment for
the 25 25 risk factors. The mutually adjusted models
included age, sex, race or ethnicity, marital status,
socioeconomic status, and all 25 25 risk factors as
independent variables with total mortality and deaths
from cardiovascular disease, cancer, and other causes as
outcomes. To enable balanced comparisons between
socioeconomic status and 25 25 risk factors as
predictors of cause-specic mortality, these analyses
were restricted to a subgroup of participants with
complete data for socioeconomic status and the
25 25 risk factors.
To examine whether the association between socio-
economic status and mortality is attributable to the
higher prevalence of the 25 25 risk factors among low
socioeconomic status individuals, we repeated the
analyses in a subgroup of participants without any
25 25 risk factors. Analyses were also repeated
specically focusing on premature mortality (<70 years)
and by restricting the population to cohorts in which
height and weight as well as blood pressure were
measured objectively using standard procedures.
To further evaluate the eects of socioeconomic status
and the 25 25 risk factors on mortality, we computed the
population attributable fraction. The population
attributable fraction is based on the HR and the
proportion of participants exposed assuming the
association between exposure and outcome is causal.14
The variance of population attributable fraction was
estimated via bootstrapping using 1000 independent
replications. The proportion of participants exposed
(prevalence) was calculated as the mean prevalence
across all cohorts for each risk factor.
YLLs were calculated as the dierence of the areas
under the survival curves (from age 40 years to 85 years)
comparing the population exposed to a given risk factor
with the reference population with no exposure. Area
under the curve was computed via numerical integration
with a spline-based method. Life expectancies were
estimated conditional on survival to age 40 years. In view
of the truncation at age 85 years, the theoretical maximal
life expectancy at 40 years old is 45 years. Variance
of YLLs was estimated via bootstrapping using
1000 independent replications.
Study-specic HRs, PAF, and YLLs estimates were
meta-analysed using the Hartung-Knapp random-eects
3
 Articles

Men
datasets. SS, PV, and MK had nal responsibility for the
Deaths Mean HR (95% CI) Weight decision to submit for publication.
follow-up (years)
COLAUS 55 616 208 (098438) 01%
NHIS 2009 86 322 123 (072211) 02%
Results
NHIS 2008 111 418 125 (078200) 02% 48 studies were included (appendix). After excluding
MIDUS 133 1161 121 (081181) 03% 27 392 (15%) of 1 778 871 participants who had missing
EPIPORTO 144 688 164 (094286) 02%
NHIS 2007 148 513 130 (085199) 03% data for the covariates or mortality, 1 751 479 participants
NCDS 159 745 174 (115264) 03% were included in the analysis (appendix). Mean age at
NHIS 2006 183 609 182 (129258) 04%
NHANES 2007 190 386 121 (085173) 04% study entry was 478 years (SD 148) and 54% of
NHANES 2005 234 573 117 (085160) 05% participants were women. The proportion of participants
NHIS 2005 290 700 109 (080148) 05%
NHIS 2003 291 910 114 (086150) 06%
with low occupational position ranged from 69% to
WLSS 360 1272 131 (104165) 09% 669% across studies (mean 414% [SD 125] for men
NHIS 2002 372 1005 190 (147247) 07% and 271% [SD 149] for women). The proportion of
NHANES 2003 381 739 119 (094149) 09%
NHIS 2001 463 1101 189 (149239) 09% people with a high occupational position varied between
NHANES 1999 479 1051 133 (107166) 10% 59% and 848% (mean 325% [SD 117] men and 261%
NHANES 2001 483 912 146 (119180) 11%
WLSG 502 1495 149 (123182) 12%
[SD 123] women). Age stratication revealed no age
NHANES II 528 1374 132 (105164) 10% calendar eects (data not shown).
NHIS 2000 530 1194 147 (119181) 11% During 12 025 208 person-years at risk for men,
NHIS 1999 540 1288 166 (134204) 11%
NHANES III 656 1447 139 (114169) 12% 161 524 men died; during 14 580 862 person-years at risk
WHITEHALL II 708 2040 157 (121204) 07% for women, a total of 148 753 women died (mean follow-
NHIS 1998 719 1378 147 (123175) 15%
EPIC Italy 758 1604 140 (105188) 06%
up for men and women 133 years [SD 34]). In men,
NHIS 1997 829 1473 145 (123170) 17% 43 765 (152% of total) with low occupational position
ELSA 840 730 146 (122174) 15% died and 17 160 (115%) with high occupational position
NHIS 2004 1115 815 153 (132177) 21%
NHANES I 1147 1858 148 (127172) 19% died. In women, 11 835 (94% of total) with low
NHIS 1996 1247 1545 155 (136178) 24% occupational position died and 8292 (68%) with high
HRS 1279 1728 150 (131171) 24%
HALS 1359 2023 145 (125168) 20% occupational position died. Participants with low
Alameda County 1547 2696 129 (112147) 24% occupational position had a higher mortality risk than
GAZEL 1935 2534 168 (148190) 28% did those with high occupational position, in both men
NHIS 1995 2293 1631 138 (125152) 39%
NHIS 1994 3029 1718 146 (134159) 48% (HR 142, 95% CI 138145; gure 1) and women
NHIS 1993 3090 1808 144 (132157) 49% (134, 128139; gure 2). Participants with inter-
NHIS 1986 3331 2369 141 (129153) 51%
NHIS 1992 3898 1983 136 (126147) 57% mediate occupational position had a higher mortality
NHIS 1991 4152 1975 132 (122142) 60% risk compared with participants with high occupational
NHIS 1990 4590 2059 137 (128148) 63%
NHIS 1989 4848 2141 140 (131150) 66%
position (meta-analytic HR 121, 95% CI 118124 for
NHIS 1988 5564 2221 137 (129146) 71% men and 117, 112122 for women). A graded
NHIS 1987 6018 2293 138 (129146) 74% association between occupational position and mortality
WHIP 21 049 1160 147 (136160) 50%
was observed in both men and women (HR for one unit
Pooled HR 142 (138145) 100% decrease in SES 119, 95% CI 117120 in men and
Prediction interval 133151
I=145%, =00008
115, 113118 in women, p<00001 for both).
05 10 25
Heterogeneity in study-specic estimates was low for
Figure 1: Mortality for low versus high occupational position in men in 46 cohort studies men (I=145% [041%], p=02034, =00008) and
HRs are adjusted for age, marital status, and race or ethnicity. Pooled HR is represented with a grey diamond and moderate for women (I=298% [0512%], p=00309,
the 95% prediction interval with a black bar. I statistic is the percentage of between study heterogeneity;
=00048).
statistic measures the inter-study variance. The prediction interval provides a predicted range for the true
association between occupational position and mortality. HR=hazard ratio. Figure 3 shows mortality associated with the 25 25 risk
factors (minimally adjusted models). The greatest
increases in mortality associated with the 25 25 risk
method.15 To assess heterogeneity between cohorts, we factors were for current smoking and diabetes, although
computed I and statistics; I to assess heterogeneity physical inactivity, high alcohol intake, and hypertension
attributable to variation in the true association and to were also associated (gure 3). The eect of low
measure the inter-cohort variance. To account for in occupational position appeared greater than that of
the uncertainty around the pooled estimates, we further hypertension or obesity (gure 3); the eect of low
calculated 95% prediction intervals for hazard ratios.16 occupational position on mortality was greater than that
of obesity even when the obesity analysis was restricted
Role of the funding source to cohorts with a mean follow-up more than 10 years
The funding sources had no role in the study design; in (>10 years; HR 112, 95% CI 105121 for men and 124,
the collection, analysis, and interpretation of data; in the 118131 for women). 33 of 48 studies had complete
writing of the report; or in the decision to submit the data for occupational position and all 25 25 risk factors
paper for publication. CC and MJ had full access to the and had cause-specic mortality data, for a total of

4 www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7

 Articles

275 973 participants with 21 923 deaths during the follow- Women
up (gure 4). The association between low socioeconomic Deaths Mean HR (95% CI) Weight
follow-up (years)
status and mortality was consistent across causes of
COLAUS 22 619 172 (0221315) 00%
death and remained signicant in the minimally adjusted EPIPORTO 68 642 247 (0591037) 01%
models and the mutually adjusted models (gure 4). The NCDS 80 772 117 (061223) 04%
MIDUS 91 1169 108 (063186) 05%
highest minimally adjusted HR was current smoking NHIS 2009 98 322 116 (071190) 06%
(gure 4). NHIS 2008 120 420 214 (131351) 06%
We assessed the PAF for socioeconomic status and the NHANES 2007 145 391 090 (056147) 06%
NHIS 2007 153 517 120 (080179) 09%
25 25 risk factors, assuming the associations with NHANES 2005 166 584 165 (102268) 07%
mortality are causal and that the risk could be reduced to NHIS 2003 177 916 160 (096266) 06%
NHANES II 187 1429 134 (076238) 05%
the level of the most favourable category for each factor NHIS 2006 219 611 135 (096191) 12%
(gure 5). We estimated the achievable reduction in WLSS 241 1306 105 (075145) 13%
mortality during the follow-up period should the death NHIS 2002 250 1013 161 (106244) 09%
NHIS 2001 284 1112 174 (121251) 11%
risk in the whole population equate that of high NHANES 2003 294 762 144 (105199) 14%
occupational position or the reference group for each of NHIS 2000 308 1209 125 (084185) 09%
NHANES III 322 1457 153 (106220) 11%
the 25 25 risk factors. The PAF for low SES was 1894% WHITEHALL II 328 2034 104 (074145) 12%
(95% CI 17632024) for men and 1533% (12761790) NHIS 2005 339 708 140 (106186) 17%
for women. The highest PAF was for smoking for men NHANES 1999 344 1095 144 (107195) 15%
GAZEL 367 2581 123 (087174) 12%
(2904%, 26903118) and for physical inactivity for WLSG 374 1524 172 (131226) 18%
women (2341%, 20422639). NHIS 1999 390 1302 173 (123244) 12%
NHANES 2001 402 939 119 (092153) 20%
In men and women combined, partial life expectancy NHIS 1998 446 1396 110 (079152) 13%
at 40 years was reduced by more than 2 years because of NHANES I 472 2030 108 (078149) 13%
low socioeconomic status (gure 6). All other NHIS 1997 496 1496 160 (120215) 16%
EPIC Italy 565 1530 109 (069174) 07%
25 25 factors assessed were associated with decreased HRS 686 1850 179 (144223) 25%
life expectancy, apart from BMI (gure 6). NHIS 1996 728 1563 166 (133207) 24%
ELSA 736 757 132 (108160) 29%
Additional sensitivity analyses including only western Alameda County 767 2947 107 (087130) 28%
European cohorts, restricting the analysis to premature NHIS 2004 1076 823 148 (123179) 31%
NHIS 1995 1307 1661 155 (130186) 32%
mortality (<70 years), to a subset of participants without HALS 1490 2128 157 (136182) 41%
the 25 25 risk factors (HR for low SES vs high SES 126, NHIS 1994 1725 1751 133 (114154) 41%
95% CI 112142), and to high quality studies or to NHIS 1993 1794 1843 134 (115155) 41%
NHIS 1986 1864 2460 130 (113150) 42%
cohorts with height and weight or blood pressure NHIS 1992 2138 2032 129 (113147) 46%
measured using standard procedures, yielded similar NHIS 1991 2278 2027 130 (114148) 47%
WHIP 2430 1060 096 (068136) 12%
results (appendix). NHIS 1990 2598 2116 128 (114145) 49%
NHIS 1989 2766 2203 118 (104132) 50%
Discussion NHIS 1988
NHIS 1987
3173
3292
2292
2382
136 (122153)
123 (110137)
53%
54%
We used individual-level data from more than 17 million E3N 6621 1683 128 (118139) 65%
individuals in 48 independent cohort studies to compare 134
Pooled HR (128139) 100%
the association of low socioeconomic status with Prediction interval 115155
mortality to those of six WHO 25 25 risk factor targets I=298%, =00048
05 10 25
for the reduction of premature mortality. We found that
the independent association between socioeconomic
Figure 2: Mortality for low versus high occupational position in women in 47 cohort studies
status and mortality is comparable in strength and HRs are adjusted for age, marital status, and race or ethnicity. Pooled HR is represented with a grey diamond and
consistency across countries to those for the 25 25 risk the 95% prediction interval with a black bar. The prediction interval provides a predicted range for the true
factors. Low socioeconomic status was associated with association between occupational position and mortality. HR=hazard ratio.
21 YLLs between ages 40 and 85 years, while the
corresponding years of life lost were 05 for high alcohol the eect of low socioeconomic status on premature
intake, 07 for obesity, 39 for diabetes, 16 for hyper- mortality and is to our knowledge the rst large-scale
tension, 24 for physical inactivity and 48 for current study to directly compare the importance of socio-
smoking in men and women combined. These ndings economic circumstances as determinants of health with
are largely consistent with previous studies,1719 which the six major risk factors targeted in global health
used income or education as a measure of socioeconomic strategies for the reduction of premature mortality. The
status. association between low socioeconomic status and
The strong inuence of socioeconomic factors on premature mortality was consistent across causes of
health, morbidity and mortality is well established,3,2025 death, whereas the 25 25 risk factors were generally
with studies showing a widening in inequalities in more strongly associated with cardiovascular disease
mortality22,25 despite absolute inequalities falling in some mortality than with cancer and with mortality of
countries.22,23 Our study is one of the largest to examine other causes.

www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7 5

 Articles

Risk factor Deaths Participants Time at risk HR (95% CI)

We used occupational position as a proxy of
(years) socioeconomic status and social circumstances in
Low SES (reference high SES) general. This measure is one of the most commonly
Men 87 716 619 402 9 835 775 142 (138145) used indicators of socioeconomic status, data for this
Women 48 791 592 157 9 538 159 134 (128139)
Current smoking (reference never smoking)
indicator were widely available across the cohort studies
Men 37 238 276 686 3 150 820 217 (206229) included in our analysis and occupational position is
Women 46 447 423 861 5 271 704 202 (191214) comparable between countries. Occupational position
Diabetes
Men 39 655 262 745 3 089 811 169 (156183) also has the advantage of reducing reverse causalitywe
Women 38 162 325 540 3 749 493 188 (173203) assessed last known occupation, which is less likely to
Physical inactivity
Men 39 794 259 265 3 029 468 160 (150170)
change with illness than is ones income. However,
Women 45 353 398 992 4 941 600 158 (148167) socioeconomic status is a complex factor that comprises
High alcohol intake (reference moderate alcohol intake) several dimensions and by using a single indicator of
Men 33 151 235 245 2 808 575 150 (138164)
Women 37 864 363 666 4 649 162 169 (149192) socioeconomic status we might have underestimated its
Hypertension full eect on mortality. Addressing several components
Men 41 034 273 190 3 184 326 130 (124136)
Women 44 340 391 681 4 752 337 128 (121136) of socioeconomic status (ie, low occupational position,
Obesity (reference normal BMI) income poverty, low education) could be important for
Men 131 882 636 779 17 632 210 104 (098111)
Women 136 680 815 005 22 310 188 117 (110124)
population health improvement.
This study has some important limitations. First, risk
05 10 15 20 15
factors (ie, hypertension, physical activity, obesity, and
Figure 3: Pooled hazard ratios of socioeconomic status and 25 25 risk factors for mortality diabetes) are interconnected making it dicult to
HRs are adjusted for age, marital status, and race or ethnicity. SES=socioeconomic status. BMI=body-mass index. establish their independent contribution. For example,
low socioeconomic status might induce changes in one
or more risk factors, but risk factors for chronic diseases
Risk factor and Minimally adjusted Mutually adjusted might also reduce labour supply and earnings, thereby
outcomes HR (95% CI) HR (95% CI) lowering socioeconomic status. Furthermore, factors
Low SES (reference high SES) other than those considered in the 25 25 list could be
All-cause 146 (139153) 126 (121132) involved in the pathways between socioeconomic status
CVD 152 (137167) 129 (116143)
Cancer 143 (134152) 126 (119134) and mortality. In view of these complex relationships,
Other 145 (135156) 125 (117133) our estimates of the population attributable fraction,
Current smoking (reference never smoking)
All-cause 227 (214239) 221 (210233)
assuming unidirectional causal associations, should be
CVD 219 (198242) 221 (200244) interpreted with caution. Second, dierent measures of
Cancer 264 (240291) 252 (232274) socioeconomic status can themselves be intertwined,
Other 205 (191220) 199 (185214)
Diabetes and can inuence risk factors for health or disease at
All-cause 187 (172203) 173 (160188) dierent points over a persons life. For example,
CVD 218 (186255) 192 (164227)
Cancer 121 (106138) 118 (104134)
increased educational levels might contribute to
Other 221 (201242) 208 (191226) increased life expectancy via multiple pathways
Physical inactivity including better occupational position, higher income,
All-cause 143 (134153) 128 (119137)
CVD 154 (143165) 135 (125146) less smoking, reduced occupational hazard, more
Cancer 125 (115136) 114 (106123) physical activity, healthier diet, increased self-care, and
Other 150 (137164) 134 (122147)
High alcohol intake (reference moderate intake) adherence to medical treatments.26 However, the
All-cause 164 (144187) 136 (123151) nding that socioeconomic status is associated with
CVD 145 (126166) 119 (108132)
Cancer 170 (144199) 138 (121156)
death risk independently of conventional risk factors
Other 176 (152203) 146 (130165) suggests that both socioeconomic adversity and
Hypertension 25 25 risk factors should be targeted by health
All-cause 138 (130146) 131 (124138)
CVD 183 (166203) 169 (153188) strategies. Third, with broad two-level or three-level
Cancer 108 (098118) 107 (099116) categorisations, the assessment of both socioeconomic
Other 138 (128147) 129 (121138)
Obesity (reference normal BMI)
status and risk factors was crude, potentially
All-cause 118 (109127) 105 (097114) underestimating the strength of associations with
CVD 146 (128166) 122 (106140) mortality outcomes. However, the comparison between
Cancer 101 (092110) 102 (094111)
Other 117 (108126) 101 (092110) risk factors should be balanced because they were all
05 10 15 20 25 30
measured with the same relative level of precision. The
observed associations of smoking, physical activity,
Figure 4: Pooled hazard ratios of socioeconomic status and 25 25 risk factors for all-cause mortality and high alcohol intake, diabetes, and hypertension with
cause-specic mortality
mortality were comparable with those of previous
The minimally adjusted models were only adjusted for sex, age, and race or ethnicity; in the mutually adjusted
models, SES and the 25 25 risk factors are mutually adjusted. BMI=body-mass index. CVD=cardiovascular disease. studies.2730 The non-signicant outcome observed
SES=socioeconomic status. between obesity and all-cause mortality in men might
be an underestimate due to pre-existing morbidity

6 www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7

 Articles

leading to weight loss and increased mortality risk Risk factor Prevalence (%) PAF (95% CI)
among lean or underweight individuals.31,32 Hetero-
Low SES (intermediate/low)
geneity in study-specic estimates was generally low Men 251/424 1894 (1763 to 2024)
for occupational position, but larger for some of the Women 458/281 1533 (1276 to 1790)
Current smoking (former/current)
risk factors (appendix). This dierence could be due to Men 328/271 2904 (2690 to 3118)
varying degrees of precision in the measurement of the Women 209/210 2104 (1902 to 2307)
25 25 risk factors in the dierent cohorts, and random- Diabetes
Men 94 593 (485 to 700)
eect meta-analysis partially takes this uncertainty into Women 87 688 (576 to 800)
account for the estimation of pooled eects. Physical inactivity
Men 395 2616 (2301 to 2931)
Finally, the cohort studies participating in the Women 462 2341 (2042 to 2639)
LIFEPATH consortium were from high-income High alcohol intake
countries. Thus, our results might not be generalisable Men 100 434 (326 to 542)
Women 48 327 (234 to 420)
to other populations. Previous studies suggest that Hypertension
socioeconomic factors and the 25 25 risk factors are Men 380 976 (792 to 1160)
Women 314 821 (622 to 1020)
also strong predictors of premature mortality in low Obesity (overweight/obese)
and middle income countries.33 Further research should Men 439/194 557 (884 to 231)
Women 289/220 355 (135 to 574)
assess socioeconomic status and 25 25 risk factors in
predicting mortality in dierent economic settings. 10 0 10 20 20 40
Population attributable fraction (%)
Despite these limitations, our study has important
implications. Our ndings suggest that existing global Figure 5: Population attributable fraction for socioeconomic status and 25 25 risk factors
strategies and actions dened in the 25 25 health plan Calculations assume risk in the population at the level of the least exposed group. SES=socioeconomic status.
PAF=population attributable fraction.
and the Global Burden of Diseases surveillance
programme potentially exclude a major determinant of
health from the agenda. A lack of consideration of the Men
Years of life lost (95% CI)
interrelation between social circumstances and health is 45 26 (28 to 24) 56 (61 to 52) 41 (47 to 34) 28 (34 to 22) 06 (11 to 01) 19 (23 to 16) 04 (08 to 01)
also evident in the Sustainable Development Goals
(SDGs): SDG 3 focuses on health but it makes no 40
Life expectancy (years)

mention of the role of social circumstances. Similarly,

35
SDG 1 and 4 focus on the elimination of poverty and the
achievement of universal primary education but they do 30
not mention reducing health inequalities as an explicit
goal. Similar to the risk factors targeted by existing 25
global health strategies, socioeconomic circumstances
are modiable by policies at the local, national, and 20

international levels,26,34 through interventions such as Women

promotion of early childhood development, poverty Years of life lost (95% CI)
15 (18 to 13) 40 (45 to 36) 39 (46 to 32) 20 (25 to 15) 04 (08 to 00) 13 (17 to 10) 09 (13 to 06)
reduction, improvements to access to high-quality 45

education, enacting of compulsory schooling laws, and

40
Life expectancy (years)

creation of safe home, school, and work environments.8,9

Over the past decade, socioeconomic factors have 35
started making their way into international agencies
and global reports, as evidenced in the report of the 30
WHO Commission on the Social Determinants of
25
Health (CSDH) in 200826 and in the Rio Political
Declaration on the Social Determinants of Health.35 20
Although these eorts have raised awareness of
h

No

No

te

No

al
w

nt

gh

e
ve

Ye

Ye

Ye
g

es
rm
a
Lo

rre
Hi

Hi
Ne

er

Ob
No
od
Cu

socioeconomic inequalities in health, global prevention

M

SES Smoking Diabetes Physical Alcohol intake Hypertension BMI

strategies still appear to be centred on the treatment of
proximal risk factors. Such approaches fail to address
powerful upstream structural solutions such as
investment in early education programmes for children
(allowing parents to work while their children are cared
for) and work incentive programmes (ie, earned income
tax credit) that might be a cost-eective way to reduce
inequalities in health.10,3638 By showing low
socioeconomic status has a comparable health eect to
that of major risk factors, the results of our study
inactivity
Figure 6: Life expectancy from age 40 years to 85 years and years of life lost due to low socioeconomic status
and 25 25 risk factors
SES=socioeconomic status. BMI=body-mass index.
suggest that socioeconomic circumstances, in addition
to the 25 25 factors, should be treated as a target for
local and global health strategies, health risk
surveillance, interventions, and policy.

www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7 7

 Articles
Contributors
MK, SS, and PV conceived the study. SS wrote the rst and successive
drafts of the manuscript. CC and MJ modelled and analysed the data.
CC, MA, JPM, PM, CD, IK, MK-I, RL, AS, MC-H, and AdE contributed
to study conception and design. FG and FR contributed to data analysis.
HB, MB, FC-C, GC, SF, MG, GGG, VK, AML, MGM, MP, MJS, AS, PV,
MZ, MK, PV, and MJ collected the data. All authors revised the
manuscript for important intellectual content.
LIFEPATH Consortium
Harri Alenius, Mauricio Avendano, Henrique Barros, Murielle Bochud,
Cristian Carmeli, Luca Carra, Raphaele Castagn, Marc Chadeau-Hyam,
Franoise Clavel-Chapelon, Giuseppe Costa, Emilie Courtin,
Cyrille Delpierre, Angelo DErrico, Pierre-Antoine Dugu, Paul Elliott,
Silvia Fraga, Valrie Gares, Graham Giles, Marcel Goldberg,
Dario Greco, Allison Hodge, Michelle Kelly Irving, Piia Karisola,
Mika Kivimki, Vittorio Krogh, Thierry Lang, Richard Layte,
Benoit Lepage, Johan Mackenbach, Michael Marmot, Cathal McCrory,
Roger Milne, Peter Muennig, Wilma Nusselder, Salvatore Panico,
Dusan Petrovic, Silvia Polidoro, Martin Preisig, Olli Raitakari,
Ana Isabel Ribeiro, Ana Isabel Ribeiro, Fulvio Ricceri, Oliver Robinson,
Jose Rubio Valverde, Carlotta Sacerdote, Roberto Satolli, Gianluca Severi,
Martin J Shipley, Silvia Stringhini, Rosario Tumino, Paolo Vineis,
Peter Vollenweider, and Marie Zins.
Declaration of interests
PV reports grants from GlaxoSmithKline. JPM reports grants from
European Commission. MK reports grants from the Medical Research
Council and NordForsk, the Nordic Research Programme on Health and
Welfare. MP reports grants from GlaxoSmithKline and Swiss National
Science Foundation, during the conduct of the study. All other authors
declare no competing interests.
Acknowledgments
This study was partly supported by the European Commission
(Horizon 2020 grant number 633666) and the Swiss State Secretariat
for Education, Research and Innovation SERI. SS is supported by an
Ambizione Grant (PZ00P3_167732) from the Swiss National Science
Foundation. SF is supported by the Portuguese Foundation for Science
and Technology (SFRH/BPD/97015/2013). AML is supported by an
MD-PhD grant (323530_151479) from the Swiss National Science
Foundation. Various sources have supported recruitment, follow-up,
and measurements in the 48 cohort studies contributing to this
collaborative analysis. MK is supported by the UK Medical Research
Council (K013351) and NordForsk.
References
1 WHO. Global action plan for the prevention and control of
noncommunicable diseases 20132020. Geneva, Switzerland: World
Health Organization, 2013.
2 Lim SS, Vos T, Flaxman AD, et al. A comparative risk assessment
of burden of disease and injury attributable to 67 risk factors and
risk factor clusters in 21 regions, 19902010: a systematic analysis
for the Global Burden of Disease Study 2010. Lancet 2012;
380: 222460.
3 Mackenbach JP, Stirbu I, Roskam AJ, et al.
Socioeconomic inequalities in health in 22 European countries.
N Engl J Med 2008; 358: 246881.
4 Stringhini S, Sabia S, Shipley M, et al. Association of socioeconomic
position with health behaviors and mortality. JAMA 2010; 303: 115966.
5 Stringhini S, Rousson V, Viswanathan B, Gedeon J, Paccaud F,
Bovet P. Association of socioeconomic status with overall and cause
specic mortality in the Republic of Seychelles: results from a
cohort study in the African region. PLoS One 2014; 9: e102858.
6 Hosseinpoor AR, Bergen N, Mendis S, et al. Socioeconomic
inequality in the prevalence of noncommunicable diseases in
low- and middle-income countries: results from the World Health
Survey. BMC Public Health 2012; 12: 474.
7 Rasella D, Aquino R, Santos CA, Paes-Sousa R, Barreto ML.
Eect of a conditional cash transfer programme on childhood
mortality: a nationwide analysis of Brazilian municipalities.
Lancet 2013; 382: 5764.
8 Lleras-Muney A. The relationship between education and adult
mortality in the United States. Rev Econ Stud 2005; 72: 189221.
8 www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7
9 Heckman JJ. Skill formation and the economics of investing in
disadvantaged children. Science 2006; 312: 190002.
10 Lopez-Arana S, Avendano M, van Lenthe FJ, Burdorf A. The impact
of a conditional cash transfer programme on determinants of child
health: evidence from Colombia. Public Health Nutr 2016; 19: 114.
11 Stringhini S, Viswanathan B, Gedeon J, Paccaud F, Bovet P.
The social transition of risk factors for cardiovascular disease in the
African region: Evidence from three cross-sectional surveys in the
Seychelles. Int J Cardiol 2013; 168: 120106.
12 Higgins JPT, Green S, (eds). Cochrane Handbook for Systematic
Reviews of Interventions Version 5.1.0 [updated March 2011].
The Cochrane Collaboration, 2011. http://www.handbook.cochrane.
org (accessed June 15, 2016).
13 Royston P, Parmar MK. Flexible parametric proportional-hazards
and proportional-odds models for censored survival data, with
application to prognostic modelling and estimation of treatment
eects. Stat Med 2002; 21: 217597.
14 WHO. Metrics: Population Attributable Fraction (PAF). Quantifying
the contribution of risk factors to the Burden of Disease. 2016.
http://www.who.int/healthinfo/global_burden_disease/metrics_
paf/en/2016 (accessed June 15, 2016).
15 IntHout J, Ioannidis JP, Borm GF. The Hartung-Knapp-Sidik-
Jonkman method for random eects meta-analysis is
straightforward and considerably outperforms the standard
DerSimonian-Laird method. BMC Med Res Methodol 2014; 14: 25.
16 IntHout J, Ioannidis JP, Rovers MM, Goeman JJ. Plea for routinely
presenting prediction intervals in meta-analysis. BMJ Open 2016;
6: e010247.
17 Muennig P, Fiscella K, Tancredi D, Franks P. The relative health
burden of selected social and behavioral risk factors in the United
States: implications for policy. Am J Public Health 2010; 100: 175864.
18 Muennig P, Franks P, Jia H, Lubetkin E, Gold MR.
The income-associated burden of disease in the United States.
Soc Sci Med 2005; 61: 201826.
19 Maki NE, Martikainen PT, Eikemo T, et al. The potential for
reducing dierences in life expectancy between educational
groups in ve European countries: the eects of obesity, physical
inactivity and smoking. J Epidemiol Community Health 2014;
68: 63540.
20 Townsend P, Davidson N. Inequalities in health: The Black report.
Harmondsworth, UK: Penguin Books, 1982.
21 Marmot MG, Shipley MJ, Rose G. Inequalities in deathspecic
explanations of a general pattern? Lancet 1984; 1: 100306.
22 Chetty R, Stepner M, Abraham S, et al. The Association between
income and life expectancy in the United States, 20012014.
JAMA 2016; 315: 175066.
23 Mackenbach JP, Kulhanova I, Artnik B, et al. Changes in mortality
inequalities over two decades: register based study of European
countries. BMJ 2016; 353: i1732.
24 Stringhini S, Dugravot A, Shipley M, et al. Health behaviours,
socioeconomic status, and mortality: further analyses of the British
Whitehall II and the French GAZEL prospective cohorts. PLoS Med
2011; 8: e1000419.
25 Mayhew L, Smith D. An investigation into inequalities in adult
lifespan. London, UK: Cass Business School, City University
London, 2016.
26 Commission for the Social Determinants of Health. Closing the gap in
a generation: health equity through action on the social determinants
of health. Final Report of the Commission on Social Determinants of
Health. Geneva: World Health Organization, 2008.
27 Gellert C, Schottker B, Brenner H. Smoking and all-cause mortality
in older people: systematic review and meta-analysis.
Arch Intern Med 2012; 172: 83744.
28 Di Castelnuovo A, Costanzo S, Bagnardi V, Donati MB, Iacoviello L,
de Gaetano G. Alcohol dosing and total mortality in men and
women: an updated meta-analysis of 34 prospective studies.
Arch Intern Med 2006; 166: 243745.
29 Nocon M, Hiemann T, Muller-Riemenschneider F, Thalau F, Roll S,
Willich SN. Association of physical activity with all-cause and
cardiovascular mortality: a systematic review and meta-analysis.
Eur J Cardiovasc Prev Rehabil 2008; 15: 23946.
30 Seshasai SR, Kaptoge S, Thomson A, et al, for Emerging Risk
Factors Collaboration. Diabetes mellitus, fasting glucose, and risk of
cause-specic death. N Engl J Med 2011; 364: 82941.

31 Berrington de Gonzalez A, Hartge P, Cerhan JR, et al. Body-mass
index and mortality among 146 million white adults. N Engl J Med
2010; 363: 221119.
32 Whitlock G, Lewington S, Sherliker P, et al, for the Prospective
Studies Collaboration. Body-mass index and cause-specic mortality
in 900 000 adults: collaborative analyses of 57 prospective studies.
Lancet 2009; 373: 108396.
33 Di Cesare M, Khang YH, Asaria P, et al. Inequalities in
non-communicable diseases and eective responses. Lancet 2013;
381: 58597.
34 Marmot MG, Atkinson T, Bell J, et al. Fair society, healthy lives:
a strategic review of health inequalities in England post-2010:
The Marmot Review. London: UCL Institute, 2010.
www.thelancet.com Published online January 31, 2017 http://dx.doi.org/10.1016/S0140-6736(16)32380-7
Articles
35 WHO. Rio Political Declaration on Social Determinants of Health.
Rio de Janeiro, Brazil: World Health Organization, 2011.
36 Levin H, Beleld C, Muennig P, Rouse C. The costs and benets of
an excellent education for Americas children. New York, NY:
Teachers College, 2006.
37 Elesh D, Lefcowitz MJ. The eects of the New Jersey-Pennsylvania
Negative Income Tax Experiment on health and health care
utilization. J Health Soc Behav 1977; 18: 391405.
38 Muennig PA, Mohit B, Wu J, Jia H, Rosen Z. Cost eectiveness of
the earned income tax credit as a health policy investment.
Am J Prev Med 2016; published online Aug 26. DOI:10.1016/
j.amepre.2016.07.001.
9