Smoking and Parkinson's disease: using parental smoking as a proxy to explore causality.
ABSTRACT In epidemiologic studies and in studies of discordant twins, cigarette smoking has been consistently associated with a lower risk of Parkinson's disease, but whether this association is causal remains controversial. Alternatively, an infectious or toxic exposure in childhood or early adulthood could affect both the reward mechanisms that determine smoking behavior and the future risk of Parkinson's disease. If so, parental smoking, commonly established before the birth of the first child, would be unlikely to be related to Parkinson's disease risk. The authors assessed the association between Parkinson's disease and parental smoking during childhood in the Nurses' Health Study and the Health Professionals Follow-up Study conducted in the United States. During 26 years and 18 years of follow-up, respectively, 455 newly diagnosed Parkinson's disease cases were documented among those who provided information on parental smoking. The age-adjusted, pooled relative rate of Parkinson's disease was 0.73 (95% confidence interval: 0.53, 1.00; P-trend = 0.04) comparing participants who reported that both parents smoked with those who reported that neither did. Adjustment for caffeine and alcohol intake did not materially change the results. If the inverse association between smoking and Parkinson's disease were due to confounding by an environmental factor or were the result of reverse causation, it is unlikely that parental smoking would predict Parkinson's disease.
- [Show abstract] [Hide abstract]
ABSTRACT: Alcoholism has complex etiology and there is evidence for both genetic and environmental factors in its pathophysiology. Chronic, long-term alcohol abuse and alcohol dependence are associated with neuronal loss with the prefrontal cortex being particularly susceptible to neurotoxic damage. This brain region is involved in the development and persistence of alcohol addiction and neurotoxic damage is likely to exacerbate the reinforcing effects of alcohol and may hinder treatment. Understanding the mechanism of alcohol's neurotoxic effects on the brain and the genetic risk factors associated with alcohol abuse are the focus of current research. Because of its well-established role in neurodegenerative and neuropsychological disorders, and its emerging role in the pathophysiology of addiction, here we review the genetic and epigenetic factors involved in regulating α-synuclein expression and its potential role in the pathophysiology of chronic alcohol abuse. Elucidation of the mechanisms of α-synuclein regulation may prove beneficial in understanding the role of this key synaptic protein in disease and its potential for therapeutic modulation in the treatment of substance use disorders as well as other neurodegenerative diseases.Neurochemistry International 06/2013; · 2.65 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Parkinson's disease (PD) is a progressive neurodegenerative disorder, which is characterized by neuroinflammation, dopaminergic neuronal cell death and motor dysfunction, and for which there are no proven effective treatments. The negative correlation between tobacco consumption and PD suggests that tobacco-derived compounds can be beneficial against PD. Nicotine, the more studied alkaloid derived from tobacco, is considered to be responsible for the beneficial behavioral and neurological effects of tobacco use in PD. However, several metabolites of nicotine, such as cotinine, also increase in the brain after nicotine administration. The effect of nicotine and some of its derivatives on dopaminergic neurons viability, neuroinflammation, and motor and memory functions, have been investigated using cellular and rodent models of PD. Current evidence shows that nicotine, and some of its derivatives diminish oxidative stress and neuroinflammation in the brain and improve synaptic plasticity and neuronal survival of dopaminergic neurons. In vivo these effects resulted in improvements in mood, motor skills and memory in subjects suffering from PD pathology. In this review, we discuss the potential benefits of nicotine and its derivatives for treating PD.Frontiers in Aging Neuroscience 12/2014; 6(340). · 2.84 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Epidemiologic findings suggest that lipids and alteration in lipid metabolizing protein/gene may contribute to the development of neurodegenerative disorders. The aim of the current study was to determine the serum lipid levels and genetic variation in two lipid metabolizing genes, low-density lipoprotein receptor-related protein-associated protein (LRPAP1) and apolipoprotein E (APOE) gene in Parkinson's disease (PD). Based on well-defined inclusion and exclusion criteria, this study included 70 patients with PD and 100 age-matched controls. LRPAP1 and APOE gene polymorphism were analyzed by polymerase chain reaction and restriction fragment length polymorphism, respectively. Fasting serum lipid levels were determined using an autoanalyser. The logistic regression analysis showed that high levels of serum cholesterol [odds ratio (OR) = 1.101, 95 % confidence interval (CI95%) = 1.067-1.135], LRPAP1 I allelic variant alone (OR = 2.766, CI95% = 1.137-6.752) and in combination with APOE ε4 allelic variant (OR = 4.187, CI95% = 1.621-10.82) were significantly associated with increase in PD risk. Apart from that, the high levels of LDL cholesterol appears to have a protective role (OR = 0.931, CI95% = 0.897-0.966) against PD. The LRPAP1 I allelic variant may be considered a candidate gene for PD, predominantly in patients having the APOE ε4 allelic variant.Neurological Sciences 02/2014; · 1.50 Impact Factor
American Journal of Epidemiology
ª The Author 2009. Published by the Johns Hopkins Bloomberg School of Public Health.
All rights reserved. For permissions, please e-mail: email@example.com.
Vol. 169, No. 6
Advance Access publication January 8, 2009
Smoking and Parkinson’s Disease: Using Parental Smoking as a Proxy to Explore
E´ilis J. O’Reilly, Honglei Chen, Hannah Gardener, Xiang Gao, Michael A. Schwarzschild, and
Initially submitted July 3, 2008; accepted for publication November 12, 2008.
In epidemiologic studies and in studies of discordant twins, cigarette smoking has been consistently associated
with a lower risk of Parkinson’s disease, but whether this association is causal remains controversial. Alternatively,
an infectious or toxic exposure in childhood or early adulthood could affect both the reward mechanisms that
determine smoking behavior and the future risk of Parkinson’s disease. If so, parental smoking, commonly estab-
lished before the birth of the first child, would be unlikely to be related to Parkinson’s disease risk. The authors
assessed the association between Parkinson’s disease and parental smoking during childhood in the Nurses’
Health Study and the Health Professionals Follow-up Study conducted in the United States. During 26 years and
18 years of follow-up, respectively, 455 newly diagnosed Parkinson’s disease cases were documented among
those who providedinformation onparentalsmoking. Theage-adjusted, pooled relativerate of Parkinson’s disease
was 0.73 (95% confidence interval: 0.53, 1.00; P-trend ¼ 0.04) comparing participants who reported that both
parents smoked with those who reported that neither did. Adjustment for caffeine and alcohol intake did not
materially change the results. If the inverse association between smoking and Parkinson’s disease were due to
confounding by an environmental factor or were the result of reverse causation, it is unlikely that parental smoking
would predict Parkinson’s disease.
causality; Parkinson disease; smoking; tobacco smoke pollution
Despite the well-documented adverse health effects of
cigarette smoking, cigarette smokers and users of other to-
bacco products have lower rates of Parkinson’s disease than
nonsmokers/nonusers do. Rates of Parkinson’s disease de-
crease as dose of cigarettes per day and number of years of
smoking increase, and they increase as number of years
since quitting increases (1–5). In longitudinal studies, cur-
rent smokers, compared with never smokers, were found to
have one-third the risk of Parkinson’s disease (5). Among
men never smokers, use of smokeless tobacco was also as-
sociated with lower rates of Parkinson’s disease (6, 7). In an
ecologic study, country-specific variations in gender ratios
of Parkinson’s disease over time were found to be correlated
with variations in gender ratios of smoking (8).
Whether the observed inverse associations between
smoking and Parkinson’s disease are causal remains con-
troversial; the association could be explained by a still-
unknown third factor that increases the risk of Parkinson’s
disease and also causes an aversion to smoking. This factor
is unlikely to be genetic because, in monozygotic and di-
zygotic twin studies, which control tightly for genetics and
shared environment, smoking remains associated with
Parkinson’s disease (9, 10). However, confounding by an
environmental factor cannot be eliminated by twin studies.
For example, a person could be exposed to a toxic chem-
ical or an infectious agent that causes subclinical damage
to the dopaminergic system, which is involved in novelty
seeking and addiction, and this exposure could reduce his
or her desire to smoke while independently increasing the
risk of Parkinson’s disease. This apparently intractable
confounding could be circumvented by examining the re-
lation between parental smoking and risk of Parkinson’s
disease because the hypothetical toxic chemical or infec-
tious agent is less likely to affect parents’ smoking behav-
ior, which is typically established before the birth of the
first child (11). Children of smokers are exposed to
Correspondence to E´ilis J. O’Reilly, Departments of Epidemiology and Nutrition, Harvard School of Public Health, 677 Huntington Avenue,
Boston, MA 02215 (e-mail: firstname.lastname@example.org).
678 Am J Epidemiol 2009;169:678–682
environmental smoke and are more likely to become
smokers themselves, so if smoking is truly protective, they
would be expected to have a lower risk of Parkinson’s
disease than children of parents who do not smoke. On
the other hand, if the association between smoking and
Parkinson’s disease were due to exposure to a toxic chem-
ical or an infectious agent, no relation would be expected
between parental smoking, which was established prior to
the offspring being exposed, and Parkinson’s disease risk.
We examined the association between parental smoking
and Parkinson’s disease rates in the Nurses’ Health Study
and Health Professionals Follow-up Study in the United
MATERIALS AND METHODS
The Nurses’ Health Study enrolled 121,701 female nurses
aged 35–55 years who returned a mailed questionnaire in
1976 regarding lifestyle and medical history (12). The
Health Professionals Follow-up Study enrolled 51,529
males aged 40–75 years who returned a similar question-
naire in 1986 (13). Participants of both cohorts have re-
ceived follow-up questionnaires biennially to record newly
diagnosed illnesses and to update lifestyle and dietary
Health Professionals Follow-up Study and Nurses’
Health Study participants were first asked about a lifetime
diagnosis of Parkinson’s disease on the 1988 and 1994
questionnaires, respectively, and on every follow-up ques-
tionnaire thereafter. Confirmation of self-reported Parkinson’s
disease has been described elsewhere (14). Briefly, permis-
sion to contact the treating neurologist was sought from
participants who reported a new diagnosis of Parkinson’s
disease. Neurologists (or internists if the neurologists did
not respond) were then asked to complete a questionnaire
to provide their judgment on the certainty of the diagnosis,
the date on which symptoms were first noticed, and the
date of diagnosis; copies of medical records were also
sought. Confirmed cases of Parkinson’s disease included
those for which 1) the level of certainty of the diagnosis
reported by the treating neurologist or internist was definite
or probable, 2) the medical record indicated a final diag-
nosis of Parkinson’s disease by a neurologist, or 3) there
was evidence at a neurologic examination of at least 2 of
the 3 cardinal signs of Parkinson’s disease (rest tremor,
rigidity, or bradykinesia) in the absence of features sug-
gesting other diagnoses. A movement disorder specialist,
blinded to exposure status, reviewed the medical records.
Overall, the diagnosis was confirmed by the treating neu-
rologist in 84.6% and by the treating internist in 9.7%
Health Study (1976) and the Health Professionals Follow-up Study (1986), United States
Age-adjusted Baseline Risk Factors for Parkinson’s Disease by Parental Smoking Status in the Nurses’
Parental Smoking Status
Nurses’ Health Study
(no. of cases)
30,182 (104)45,562 (154) 17,177 (29)11,907 (81) 13,633 (66)5,128 (21)
44 (30–63)44 (30–57) 39 (30–60)53 (39–77) 53 (39–77)47 (39–77)
365397 407214 243248
Never 54.043.0 31.458.847.042.7
<10 pack-years 17.619.2 18.510.7 10.810.6
10–24 pack-years16.0 20.224.6 17.221.2 21.7
25–44 pack-years8.812.518.0 9.614.917.5
2.53.7 6.23.7 6.1 7.5
Reference0.95 0.89 Reference0.90 0.87
aPredicted relative rates of Parkinson’s disease for parental smoking were calculated as an average of the relative
rates for pack-years published in the study by Herna ´n et al. (2) weighted by the distribution of smoking history at
baseline within each level of parental smoking. (Compared with never smoking, the published relative rates for 1–9,
10–24,25–44,and ?45pack-yearswere1.0, 0.8,0.4, and 0.4in the Nurses’Health Studyand 0.6,0.5, 0.5,and 0.3 in
the Health Professionals Follow-up Study, respectively.)
Smoking and Parkinson’s Disease679
Am J Epidemiol 2009;169:678–682
of the cases and by review of the medical records in the
In both cohorts (women in 1982 and men in 2004), par-
ticipants reported whether their parents (neither, mother
only, father only, both) smoked while living with them dur-
ing childhood. To gain statistical power, the categories
‘‘mother only’’ and ‘‘father only’’ were collapsed to ‘‘one
Participants contributed person-time of follow-up from
the date of return of the baseline questionnaire to the date
of Parkinson’s disease diagnosis, death from any cause, or
end of follow-up (June 2002 for the Nurses’ Health Study
and January 2004 for the Health Professionals Follow-up
Study). Mantel-Haenzel age-adjusted incidence relative
rates and 95% confidence intervals were obtained relative
to the incidence rates for thosewhose parents did not smoke.
To adjust for other risk factors for Parkinson’s disease, we
used Cox proportional hazards analysis. A Wald test deter-
mined whether the relative rate from each study was hetero-
geneous. A pooled relative rate was calculated by weighting
the study-specific log relative rate by the inverse of their
variances using a random-effects model (15). All reported
P values are 2-sided.
During 26 years offollow-up in the Nurses’ Health Study,
328 newly diagnosed Parkinson’s disease cases were docu-
mented. The 1982 questionnaire was completed by 110,163
eligible women; of these, 92,921 gave information on pa-
rental smoking (287 cases). During 18 years of follow-up in
the Health Professionals Follow-up Study, 384 newly diag-
nosed Parkinson’s disease cases were documented. The
(1976–2002) and the Health Professionals Follow-up Study (1986–2004), United States
Parental Smoking Status and Risk of Parkinson’s Disease in the Nurses’ Health Study
Health Professionals Follow-up Study
Neither parent smoked 208,563 81Reference
At least 1 parent smoked328,860 870.73 0.54, 0.99
1 Parent smoked 238,65566 0.72 0.52, 0.99
Both parents smoked 90,205 210.800.49, 1.30
Nurses’ Health Study
Neither parent smoked755,123 104Reference
At least 1 parent smoked 1,569,776 1830.94 0.74, 1.20
1 Parent smoked 1,137,4801541.01 0.79, 1.30
Both parents smoked432,296 290.68 0.45, 1.03
Neither parent smoked963,686 185Reference
At least 1 parent smoked1,898,636 2700.85 0.67, 1.070.2
1 Parent smoked1,376,135 2200.870.64, 1.200.1
Both parents smoked522,50150 0.730.53, 1.000.6
P for trend0.040.6
Abbreviations: CI, confidence interval; RR, relative rate.
Parkinson’s disease is not causal. The observed association is due to
confounding. C represents one or more factors causally related to
both Parkinson’s disease and smoking status that, given it exists,
would confound the association between smoking and Parkinson’s
disease. Under this paradigm, parental smoking would not be
associated with Parkinson’s disease risk because there is no open
path between parental smoking and Parkinson’s disease. Such a
path could be provided by an influence of parental smoking on
caffeine intake under the assumption that caffeine is causally related
to Parkinson’s disease (not shown in figure) (5, 14). Because
adjustment for caffeine intake did not attenuate the relation between
parental smoking and Parkinson’s disease, such an explanation is
Paradigm I: The inverse relation between smoking and
680 O’Reilly et al.
Am J Epidemiol 2009;169:678–682
2004 questionnaire was completed by 34,884 eligible men;
of these, 30,668 gave information on parental smoking (168
cases). The average age at diagnosis of Parkinson’s disease
was 68.8 years (range: 46.6–87.8) for the men and 69.3
(range: 46.3–81.3) for the women.
As expected, participants’ baseline smoking status was
predicted by parental smoking. Ever smokers constituted
69% of the women and 57% of the men whose parents were
both smokers but only 46% of the women (P < 0.0001) and
41% of the men (P < 0.0001) whose parents were both
nonsmokers. We calculated expected relative rates of
Parkinson’s disease for each level of parental smoking
status as the averages of published study-specific relative
rates for pack-years (2) weighted by the distribution of
pack-years of smoking at baseline within each category of
parental smoking. We conservatively predicted a reduction
in risk of Parkinson’s disease of 11% for men and 13% for
women who reported that both parents smoked compared
with those who reported that neither parent smoked. Daily
caffeine and alcohol intake was higher among those whose
parents smoked (Table 1).
Among both men and women, the risk of Parkinson’s
disease was lower for individuals who reported that both
parents were smokers than for those who reported that nei-
ther parent smoked (Table 2). In a pooled analysis, those
reporting that both parents were smokers had a 27% lower
risk of Parkinson’s disease than those reporting that neither
parent smoked (relative rate ¼ 0.73, 95% confidence inter-
val: 0.53, 1.00; P-trend ¼ 0.04) (Table 2). We further ad-
justed the pooled data for caffeine and alcohol intake at
baseline and found that the effect estimate did not materially
change but, as expected, that confidence intervals were
wider (pooledrelativerate ¼ 0.76,95%confidenceinterval:
We then restricted the multivariable pooled analysis to
never smokers at baseline. There was no association be-
tween parental smoking and Parkinson’s disease, as would
be expected if the effect of parental smoking is primarily
mediated through smoking by the offspring.
In this pooled analysis of 2 large cohorts, risk of
Parkinson’s disease was 27% lower for individuals whose
parents smoked than for individuals whose parents were
nonsmokers. An inverse association between smoking and
Parkinson’s disease has been found in numerous epidemio-
logic studies, but consensus is lacking as to whether the
relation is causal, due to confounding, or a result of reverse
causation. If the desire to initiate or continue smoking and
Parkinson’s disease were both independently affected by
a third factor, then parental smoking should not be related
to Parkinson’s disease risk. This causal structure is illus-
trated with directed acyclic graphs (Figure 1) (16, 17).
Under the paradigm of reverse causation, where subclinical
also not predict Parkinson’s disease in offspring (Figure 2).
Were smoking truly protective against Parkinson’s disease,
and parental smoking predicted smoking, then parental
smoking would be expected to be inversely associated with
Parkinson’s disease, as we found (Figure 3).
observedassociationis due to reversecausation.Because thereis no
open path between parental smoking and Parkinson’s disease, no
association would be expected.
Paradigm II: Parkinson’s disease prevents smoking. The
Parkinson’s disease is causal. C represents one or more factors
causally related to Parkinson’s disease and smoking status that,
given it exists, could confound the association between smoking
and Parkinson’s disease. The effect of parental smoking on
Parkinson’s disease is probably mediated through its effects on indi-
viduals’ smoking status because there are no alternative open paths
between parental smoking and Parkinson’s disease even in the pres-
ence of C; the association, as illustrated, is causal.
Paradigm III: The inverse relation between smoking and
are confounded. C represents a factor causally related to both paren-
tal smoking and Parkinson’s disease. An assumption underlying the
conclusion that smoking could be causally related to Parkinson’s dis-
ease is the nonexistence of C.
Paradigm IV: Parental smoking and Parkinson’s disease
dependent cause of Parkinson’s disease. C represents a factor that is
causally related to Parkinson’s disease and is associated with paren-
tal smoking, although not causally. U represents a predictor of both
parental smoking and C. Under this paradigm, the inverse association
betweenparentalsmokingand Parkinson’sdiseasewouldnotbea re-
sult of a causal relation between smoking and Parkinson’s disease.
Paradigm V: Parental smoking is associated with an in-
Smoking and Parkinson’s Disease 681
Am J Epidemiol 2009;169:678–682
An inverse association between parental smoking and
Parkinson’s disease could exist under other scenarios. The
possibility cannot be ruled out, for example, that an expo-
sure such as a toxic or infectious agent in a family could
affect parental smoking behavior as well as the future risk of
Parkinson’s disease for both parents and children (Figure 4).
in a household where there is parental smoking (Figure 5).
Such an exposure would have to be transgenerational, affect-
the settings in which the association between smoking and
Parkinson’s disease has been found.
Our analysis provides a novel approach to assessing
causality, and, when taken with the results of previous
longitudinal (1–5) and ecologic (8) studies, suggests a caus-
al interpretation of the relation between smoking and
Parkinson’s disease over confounding or reverse causation.
However, causality cannot be proven until a mechanism is
Our interest in smoking and Parkinson’s disease is etio-
logic. Smoking is hugely damaging to health; any benefit
derived from a reduction in risk of Parkinson’s disease is
outweighed by the increased risks of cancer and cardiovas-
cular disease. This, however, should not be an impediment
to evaluating tobacco components for possible neuroprotec-
Author affiliations: Department of Epidemiology, Har-
vard School of Public Health, Boston, Massachusetts (E´ilis
J. O’Reilly, Alberto Ascherio); Department of Nutrition,
Harvard School of Public Health, Boston, Massachusetts
(E´ilis J. O’Reilly, Xiang Gao, Alberto Ascherio); Epidemi-
ology Branch, National Institute of Environmental Health
Sciences, Research Triangle Park, North Carolina (Honglei
Chen); Department of Neurology, Miller School of Medi-
cine, University of Miami, Miami, Florida (Hannah Gar-
dener); Department of Neurology, Massachusetts General
Hospital, Boston, Massachusetts (Michael A. Schwarzs-
child); and Channing Laboratory, Department of Medicine,
Brigham and Women’s Hospital and Harvard Medical
School, Boston, Massachusetts (Alberto Ascherio).
The study was supported by a National Institutes of
Health/National Institute of Neurological Diseases and
Stroke grant (to A. A.) to study Parkinson’s disease in the
Health Professionals Follow-up Study and Nurses’ Health
Study cohorts and by the intramural program of the National
Institutes of Health, the National Institute of Environmental
Health Sciences (to H. C.).
Conflict of interest: none declared.
1. Grandinetti A, Morens DM, Reed D, et al. Prospectivestudy of
cigarette smoking and the risk of developing idiopathic Par-
kinson’s disease. Am J Epidemiol. 1994;139(12):1129–1138.
2. Herna ´n MA, Zhang SM, Rueda-deCastro AM, et al. Cigarette
smoking and the incidence of Parkinson’s disease in two
prospective studies. Ann Neurol. 2001;50(6):780–786.
3. Thacker EL, O’Reilly EJ, Weisskopf MG, et al. Temporal re-
lationship between cigarette smoking and risk of Parkinson
disease. Neurology. 2007;68(10):764–768.
4. Ritz B, Ascherio A, Checkoway H, et al. Pooled analysis of
tobacco use and risk of Parkinson disease. Arch Neurol. 2007;
5. Herna ´n MA, Takkouche B, Caaman ˜o-Isorna F, et al. A meta-
analysis of coffee drinking, cigarette smoking, and the risk of
Parkinson’s disease. Ann Neurol. 2002;52(3):276–284.
6. O’Reilly EJ, McCullough ML, Chao A, et al. Smokeless to-
bacco use and the risk of Parkinson’s disease mortality. Mov
7. Benedetti MD, Bower JH, Maraganore DM, et al. Smoking,
alcohol, and coffee consumption preceding Parkinson’s dis-
ease: a case-control study. Neurology. 2000;55(9):1350–1358.
8. Morozova N, O’Reilly EJ, Ascherio A. Variations in gender
ratios support the connection between smoking and Parkin-
son’s disease. Mov Disord. 2008;23(10):1414–1419.
9. Tanner CM, Goldman SM, Aston DA, et al. Smoking and
Parkinson’s disease in twins. Neurology. 2002;58(4):581–588.
10. Wirdefeldt K, Gatz M, Pawitan Y, et al. Risk and protective
factors for Parkinson’s disease: a study in Swedish twins. Ann
11. Goddard E. Why children start smoking. Br J Addict. 1992;
12. Colditz GA, Manson JE, Hankinson SE. The Nurses’ Health
Study: 20-year contribution to the understanding of health
among women. J Womens Health. 1997;6(1):49–62.
13. Rimm EB, Giovannucci EL, Stampfer MJ, et al. Reproduc-
ibility and validity of an expanded self-administered semi-
quantitative food frequency questionnaire among male health
professionals. Am J Epidemiol. 1992;135(10):1114–1126.
14. Ascherio A, Zhang SM, Herna ´n MA, et al. Prospective study
of caffeine consumption and risk of Parkinson’s disease in men
and women. Ann Neurol. 2001;50(1):56–63.
15. DerSimonian R, Laird N. Meta-analysis in clinical trials.
Control Clin Trials. 1986;7(3):177–188.
16. Greenland S, Pearl J, Robins JM. Causal diagrams for epide-
miologic research. Epidemiology. 1999;10(1):37–48.
17. Herna ´n MA, Robins JM. Estimating causal effects from epi-
demiological data. J Epidemiol Community Health. 2006;
18. Quik M. Smoking, nicotine and Parkinson’s disease. Trends
682O’Reilly et al.
Am J Epidemiol 2009;169:678–682