RESEARCH ARTICLE

Lifestyle Factors and Parkinson Disease Risk

Korean Nationwide Cohort Study With Repeated Health Screening Data
Seo Yeon Yoon, MD, PhD, You Hyun Park, MPH, Hyo Jeong Lee, MD, Dae Ryong Kang, PhD,* and Correspondence
Yong Wook Kim, MD, PhD* Dr. Kang
[email protected]
®
Neurology 2022;98:e641-e652. doi:10.1212/WNL.0000000000012942 or Dr. Kim
[email protected]

Abstract MORE ONLINE

Background and Objectives Class of Evidence

Many previous studies, mostly performed in Western countries, on the effects of lifestyle factors Criteria for rating
on Parkinson disease (PD) used baseline lifestyle characteristics without directly accounting for therapeutic and diagnostic
studies
changes in covariate values over time. The objective of this study was to evaluate the association
of repeatedly measured lifestyle factors with PD risk in a Korean population. NPub.org/coe

Infographic
Methods links.lww.com/WNL/B737
We conducted a nationwide population-based cohort study. Among 512,836 Koreans in the
national health checkup database, we selected individuals who underwent health screening ≥3
times between 2002 and 2015 and followed up until December 31, 2015. PD was defined using
the ICD-10 code G20 (with ≥3 times clinic visits for PD, to increase the diagnostic validity).
Data on lifestyle factors such as smoking, alcohol consumption, and physical activity were
collected using self-reported questionnaires. Logistic regression analysis with time-dependent
covariates using generalized estimation equation models was performed to determine PD
development.

Results
During the 14-year follow-up, 2,665 patients developed PD. Smoking showed a dose–response
inverse association with PD only in men (ex-smoker, odds ratio [OR] 0.782, 95% confidence
interval [CI] 0.713–0.858; current smoker, OR 0.556, 95% CI 0.488–0.632). Alcohol con-
sumption and regular physical activity were related to reduced PD development in both sexes;
however, alcohol consumption in men (≤3 per week, OR 0.717, 95% CI 0.658–0.780; ≥4 per
week, OR 0.745, 95% CI 0.644–0.861) and physical activity in women (moderate, OR 0.792,
95% CI 0.748–0.840; vigorous, OR 0.830, 95% CI 0.756–0.911) had more consistent associ-
ations with PD development compared to those of the other sex. Participants with regular
health screening showed a consistent relationship between lifestyle factors and PD de-
velopment, whereas lifestyle factors in those without regular health screening had a decreased
relationship with PD, even smoking habit.

Conclusions
Analysis using repeatedly measured lifestyle factors showed an association between lifestyle
factors and PD development. Characteristics of lifestyle data including repeated measurements,
timing, or regularity might influence results, and future studies with appropriate lifestyle factors
could increase PD risk prediction.

Classification of Evidence
This study provides Class II evidence that smoking, alcohol use, and physical activity are
associated with reduced risk of PD in a Korean population.

*These authors contributed equally to this work.

From the Department of Physical Medicine and Rehabilitation (S.Y.Y.), Korea University Guro Hospital; Department of Biostatistics (Y.H.P.), Yonsei University, Seoul; Department of
Rehabilitation Medicine (H.J.L.), Bundang Jesaeng General Hospital, Gyeonggi-do; Department of Precision Medicine & Biostatistics (D.R.K.), Yonsei University Wonju College of
Medicine, Wonju; and Department and Research Institute of Rehabilitation Medicine (Y.W.K.), Yonsei University College of Medicine, Seoul, Korea.

Go to Neurology.org/N for full disclosures. Funding information and disclosures deemed relevant by the authors, if any, are provided at the end of the article.

Copyright © 2021 American Academy of Neurology e641

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
 Glossary
BMI = body mass index; CCI = Charlson Comorbidity Index; CI = confidence interval; GEE = generalized estimation equation;
ICD-10 = International Classification of Diseases, 10th Revision; NHIS = National Health Insurance Service; NHSP = National
Health Screening Program; OR = odds ratio; PD = Parkinson disease.

Parkinson disease (PD) is the second most common neuro- Methods

degenerative disease, and its incidence is increasing worldwide
because of the aging global population.1,2 With disease pro- Data Source
gression, nonmotor symptoms such as gastrointestinal dys- Korea has maintained a nationwide health insurance system, the
function and psychiatric symptoms develop in addition to Korean National Health Insurance Service (NHIS), since 1963,
typical motor symptoms, leading to mobility problems and and almost all health system data are centralized in large databases.
disability that serve as a considerable burden of disease to the The databases contain information on demographic characteris-
health care system.3-5 Although the etiology of PD remains tics such as age, sex, or type of insurance, visit to health care
uncertain, multifactorial origins, including both environ- facilities, diagnoses, procedural codes, medical expenses claimed,
mental and genetic factors, have been associated with PD.3-5 and prescribed medications. The NHIS implements a biannual
Lifestyle factors including physical activity, alcohol and caffein National Health Screening Program (NHSP) without cost to all
consumption, and smoking habit are important in that they recipients aged ≥40 years. The NHSP includes a self-report health
are modifiable and can affect PD onset and progression.6,7 behavior questionnaire, anthropometric measurements, and lab-
oratory tests on hemoglobin, glucose, and cholesterol levels. The
Several previous studies have reported the association be- NHIS health checkup cohort database was used for this study,
tween lifestyle factors and the incidence, progression, and with approximately 510,000 randomly selected individuals from
mortality of PD.8 Although causal inference needs to be in- those aged ≥40 years in 2002 and 2003.
vestigated further, an inverse association between smoking
and PD has been suggested.9 Physical activity and exercise Standard Protocol Approvals, Registrations,
have protective effects on PD,10 whereas alcohol consumption and Patient Consents
This research was approved by the Institutional Review Board at
is related to a decreased rate of PD, with some controversy.11
Lifestyle factors such as smoking, alcohol, and physical activity Bundang Jesaeng General Hospital and the requirement for in-
can change over time. However, many previous studies ex- formed consent was waived. This study was conducted in ac-
amining the relationship between lifestyle factors and PD cordance with the guidelines of the Strengthening the Reporting
used baseline lifestyle characteristics of participants and did of Observational Studies in Epidemiology Cohort study.
not directly consider changes in covariates over time.8,9 Some Study Population
studies with follow-up data on lifestyle factors had relatively Among the data of 512,836 Koreans in the health checkup da-
low number and longer duration of follow-up on lifestyle tabase, we selected data of individuals who had participated in the
factors.12,13 On the other hand, most of the studies on the NHSPs more than 3 times between 2002 and 2015. To define the
effect of lifestyle factors on PD were performed in Western diagnosis of PD, we first selected newly diagnosed patients with
countries.8,9,12,13 Thus, it is difficult to generalize these results PD with a primary or secondary diagnosis code G20 based on the
to Asian countries’ populations. ICD-10. The Korean government began a registration program
for rare intractable diseases in 2004, which included PD. There-
In this study, we attempted to evaluate the association of fore, individuals diagnosed with PD between 2004 and 2005 were
lifestyle factors, including smoking, alcohol consumption, and excluded, and only individuals with new-onset PD were
regular physical activity, with PD development in a Korean included.14-16 To ensure diagnostic validity, only individuals who
population using a nationwide population-based health had attended clinics more than 3 times with a diagnosis of PD
checkup cohort data. We enrolled participants with at least 3 were included and those with a combined diagnosis of secondary
longitudinal health screening data during the follow-up pe- parkinsonism or atypical parkinsonism (ICD-10 code: G21-23)
riod. Time-dependent covariates are an essential tool in data were excluded. Individuals without a diagnosis of PD who received
analysis to model the influence of a factor whose value NHSP more than 3 times before the last health care visit date were
changes over time. We performed logistic regression analysis also included as a comparison group. We longitudinally followed
with time-dependent covariates using generalized estimation these 455,939 individuals to determine the association of lifestyle
equation (GEE) models. The cumulative average at each time factors with PD development until December 31, 2015 (Figure 1).
point was used as a time-dependent covariate for the repeated
measures data to assess the association between lifestyle fac- Health Checkup Data Including
tors and PD development. We also evaluated how lifestyle Lifestyle Factors
factors were related with PD development according to the All individuals in the NHSP were required to complete self-
timing and regularity of health screening. reported questionnaires consisting of questions on their smoking

e642 Neurology | Volume 98, Number 6 | February 8, 2022 Neurology.org/N

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
status, alcohol consumption, and physical activity. Smoking and diastolic) were assessed. Body mass index (BMI) was cal-
status was categorized, based on the Centers for Disease Control culated as weight divided by height squared (kg/m2), then
and Prevention, into the following 3 groups: current smoker grouped into 5 categories based on the Asia–Pacific BMI criteria
(had smoked ≥100 cigarettes in their lifetime and currently set by the Western Pacific Region of the World Health Orga-
smoke); ex-smoker (had smoked ≥100 cigarettes in their life- nization as <18.5, 18.5–23.0, 23–25, 25–30, and ≥30 kg/m2.
time, do not currently smoke); and never smoker.17 Alcohol After an overnight fast, IV samples were collected to assess
consumption was grouped according to the weekly frequency of fasting plasma glucose, total cholesterol, and hemoglobin levels.
drinking (none, ≤3 times/week, or ≥4 times/week). Physical
activity was grouped according to the weekly frequency and Other Variables
intensity of exercise (inactive, moderate, vigorous). Anthropo- Age was divided into 5 groups: 40–49, 50–59, 60–69, 70–79,
metric data such as height, weight, and blood pressure (systolic and ≥80 years. Residential areas were categorized as urban

Figure 1 Flowchart for Sample Selection

NHIS = National Health Insurance Service.

Neurology.org/N Neurology | Volume 98, Number 6 | February 8, 2022 e643

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
 Table 1 Characteristics of Study Participants
Parkinson disease cohort (n = 2,665) Comparison cohort (n = 453,274)

Variables N % N % p Value

Age, y, mean (SD) 70.98 (8.24) 64.55 (8.88) <0.0001

40–49 19 0.71 79 0.02

50–59 263 9.87 164,100 36.20

60–69 723 27.13 158,401 34.95

70–79 1,298 48.71 98,036 21.63

≥80 362 13.58 31,946 7.05

Sex <0.0001

Male 1,327 49.79 245,821 54.23

Female 1,338 50.21 207,453 45.77

Regions <0.0001

Urban 1,004 37.67 190,038 41.93

Rural 1,661 62.33 263,236 58.07

Insurance type <0.0001

NHI, self-employees 746 27.99 112,676 24.86

NHI, employees 1,860 69.79 334,336 73.76

Medical aid 59 2.21 6,262 1.38

Income level <0.0001

Lowest 385 14.45 70,168 15.48

Low–middle 479 17.97 94,048 20.75

Middle–high 731 27.43 13,296 2.93

Highest 1,070 40.15 156,362 34.50

Body mass index, kg/m2 <0.0001

<18.5 148 5.55 12,652 2.79

18.5–23 1,034 38.80 159,238 35.13

23–23.5 633 23.75 123,116 27.16

25–30 779 29.23 143,828 31.73

≥30 71 2.66 14,440 3.19

Smoking <0.0001

No 2,017 76.20 291,529 64.62

Ex-smoker 447 16.89 95,670 21.21

Current smoker 183 6.91 63,911 14.17

Systolic blood pressure, mm Hg, mean (SD) 128.1 (16.45) 126.0 (15.02) <0.0001

Diastolic blood pressure, mm Hg, mean (SD) 77.14 (10.12) 77.11 (9.72) 0.8735

Fasting glucose, mg/dL, mean (SD) 107.0 (31.41) 104.1 (27.37) <0.0001

Total cholesterol, mg/dL, mean (SD) 185.6 (40.34) 195.0 (39.03) <0.0001

Hemoglobin, g/dL, mean (SD) 13.27 (1.58) 13.85 (1.54) <0.0001

Alcohol consumption/wk <0.0001

Continued

e644 Neurology | Volume 98, Number 6 | February 8, 2022 Neurology.org/N

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
Table 1 Characteristics of Study Participants (continued)
Parkinson disease cohort (n = 2,665) Comparison cohort (n = 453,274)

Variables N % N % p Value

No 2,323 88.46 301,041 67.56

≤3 236 8.99 122,137 27.41

≥4 67 2.55 22,384 5.02

Physical activity/wk <0.0001

No 596 22.36 46,262 10.21

≤3 1,629 61.13 292,341 64.50

≥4 440 16.51 114,671 25.30

Hypertensiona <0.0001

Yes 1,924 72.20 248,326 54.78

No 741 27.80 204,948 45.22

Dyslipidemiaa 0.0021

Yes 1,276 47.88 230,579 50.87

No 1,389 52.12 222,695 49.13

Osteoporosisa <0.0001

Yes 1,074 40.30 133,805 29.52

No 1,591 59.70 319,469 70.48

CCI, mean (SD) 6.18 (2.97) 3.58 (2.73) <0.0001

<5 818 30.69 299,103 65.99

≥5 1,847 69.31 154,171 34.01

Frequency of visiting hospital, mean (SD) 303.0 (245.5) 277.8 (231.2) <0.0001

Abbreviations: CCI = Charlson Comorbidity Index; NHI = National Health Insurance.

a
Other comorbidities not included in CCI.

and rural. Income levels were grouped into 4 categories (Q1, comorbidities, CCI, and frequency of hospital visits. GEE can
Q2, Q3, and Q4 for all enrolled medical aid + 0–20, 21–50, be used to appropriately model correlated binary outcomes
51–80, and 81–100 percentile of NHIS subscribers, re- when there are time-dependent covariates.18 The standard
spectively). Comorbidities were investigated using the logistic regression is appropriate for cross-sectional data, while
Charlson Comorbidity Index (CCI). Among the comorbid- methods such as GEE are commonly used to fit population-
ities included in the CCI, we were unable to include diagnoses averaged models for longitudinal data.19-21 Logistic regression
related to dementia, tumor, and acquired immunodeficiency analysis using GEE models was conducted to estimate the
syndrome due to unavailability of data in the NHIS health odds ratio (OR) and 95% confidence interval (CI) of the
checkup database. In addition to CCI, several diseases (hy- association between lifestyle factors and the development of
pertension, dyslipidemia, and osteoporosis) that are prevalent PD. All statistical analyses were performed using SAS (version
or might be associated with PD were extracted based on their 9.4; SAS Institute Inc.). Statistical significance was set at
ICD-10 codes. p < 0.05.

Statistical Analyses Data Availability

Baseline clinical characteristics were analyzed using Student t The source NHIS data do not belong to the researchers
test and χ 2 tests where appropriate. We performed logistic and we are not permitted to transfer the data file to a third
regression analysis with time-dependent covariates using GEE party under Korean law. The data can be used after approval
models including variables such as age, sex, socioeconomic of the institutional review board and the Korea NHIS
status, BMI, blood pressure, fasting glucose, total cholesterol, Big Data Operations Department (nhiss.nhis.or.kr/bd/ay/
hemoglobin, smoking, alcohol consumption, physical activity, bdaya001iv.do).

Neurology.org/N Neurology | Volume 98, Number 6 | February 8, 2022 e645

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
 e646
Neurology | Volume 98, Number 6 | February 8, 2022

Table 2 Crude and Adjusted Association Between Lifestyle Factors and Parkinson Disease Development Using Generalized Equation Estimation Models
Model 1 Model 2 Model 3 Model 4
Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.

Variables Crude OR 95% CI p Value Adjusted OR 95% CI p Value Adjusted OR 95% CI p Value Adjusted OR 95% CI p Value

Smoking

No 1.000 1.000 1.000 1.000

Ex-smoker 0.676 0.618–0.738 <0.0001 0.797 0.729–0.873 <0.0001 0.802 0.733–0.878 <0.0001 0.803 0.733–0.879 <0.0001

Current smoker 0.446 0.400–0.498 <0.0001 0.589 0.522–0.665 <0.0001 0.562 0.498–0.634 <0.0001 0.568 0.503–0.642 <0.0001

Alcohol consumption/wk

No 1.000 1.000 1.000 1.000

≤3 0.489 0.454–0.526 <0.0001 0.693 0.643–0.746 <0.0001 0.699 0.649–0.753 <0.0001 0.699 0.649–0.752 <0.0001

≥4 0.744 0.649–0.854 <0.0001 0.746 0.649–0.857 <0.0001 0.714 0.621–0.820 <0.0001 0.715 0.622–0.822 <0.0001

Physical activity

Inactive 1.000 1.000 1.000 1.000

Moderate 0.678 0.649–0.709 <0.0001 0.795 0.761–0.830 <0.0001 0.841 0.805–0.879 <0.0001 0.838 0.802–0.876 <0.0001

Vigorous 0.768 0.721–0.819 <0.0001 0.884 0.830–0.941 0.0001 0.929 0.872–0.989 0.0213 0.917 0.862–0.976 0.0065

Body mass index

<18.5 1.000 1.000 1.000 1.000

18.5–23 0.692 0.592–0.810 <0.0001 0.886 0.754–1.041 0.1413 0.932 0.793–1.097 0.3990 0.932 0.792–1.096 0.3940

23–25 0.654 0.554–0.772 <0.0001 0.844 0.711–1.003 0.0538 0.891 0.749–1.060 0.1940 0.888 0.746–1.057 0.1822

25–30 0.708 0.599–0.837 <0.0001 0.885 0.743–1.054 0.1720 0.921 0.771–1.100 0.3624 0.918 0.769–1.096 0.3456

≥30 0.670 0.532–0.859 0.0016 0.787 0.610–1.016 0.0657 0.774 0.598–1.001 0.0512 0.772 0.597–0.999 0.0494

Systolic blood pressure, mm Hg 1.128 1.110–1.146 <0.0001 1.056 1.035–1.078 <0.0001 1.037 1.016–1.058 0.0004 1.004 1.002–1.006 0.0003

Diastolic blood pressure, mm Hg 1.041 1.016–1.065 0.0010 0.968 0.940–0.997 0.0301 0.967 0.939–0.996 0.0261 0.997 0.994–0.999 0.0345

Fasting glucose, mg/dL 1.034 1.029–1.040 <0.0001 1.028 1.021–1.034 <0.0001 1.005 0.997–1.013 0.2357 1.001 1.000–1.001 0.1958

Total cholesterol, mg/dL 0.972 0.964–0.980 <0.0001 0.979 0.971–0.987 <0.0001 0.994 0.986–1.002 0.1365 0.999 0.999–1.000 0.1499

Hemoglobin, g/dL 0.744 0.715–0.776 <0.0001 0.900 0.857–0.945 <0.0001 0.920 0.878–0.965 0.0007 0.957 0.934–0.980 0.0003
Neurology.org/N

Abbreviations: CI = confidence interval; OR = odds ratio.

Model 1: Unadjusted. Model 2: Adjusted for age, sex. Model 3: Adjusted for age, sex, comorbidities, Charlson Comorbidity Index, number of health care visits. Model 4: Adjusted for age, sex, comorbidities, Charlson Comorbidity
Index, number of health care visits, socioeconomic status.
Table 3 Adjusted Association Between Lifestyle Factors and Parkinson Disease Development Using Generalized
Equation Estimation Models According to Sex
Male Female

Variables Adjusted OR 95% CI p Value Adjusted OR 95% CI p Value

Smoking

No 1.000 1.000

Ex-smoker 0.782 0.713–0.858 <0.0001 0.963 0.650–1.427 0.8513

Current smoker 0.556 0.488–0.632 <0.0001 0.708 0.481–1.043 0.0805

Alcohol consumption/wk

No 1.000 1.000

≤3 0.717 0.658–0.780 <0.0001 0.674 0.579–0.785 <0.0001

≥4 0.745 0.644–0.861 <0.0001 0.709 0.422–1.190 0.1929

Physical activity

Inactive 1.000 1.000

Moderate 0.894 0.837–0.955 0.0009 0.792 0.748–0.840 <0.0001

Vigorous 1.003 0.921–1.093 0.9441 0.830 0.756–0.911 <0.0001

Body mass index

<18.5 1.000 1.000

18.5–23 1.026 0.811–1.298 0.8279 0.847 0.678–1.058 0.1441

23–25 0.991 0.771–1.274 0.9443 0.793 0.624–1.009 0.0591

25–30 0.981 0.757–1.272 0.8870 0.852 0.668–1.086 0.1961

≥30 0.886 0.572–1.373 0.5892 0.661 0.479–0.913 0.0119

Systolic blood pressure, mm Hg 1.011 0.982–1.041 0.4638 1.062 1.033–1.091 <0.0001

Diastolic blood pressure, mm Hg 0.963 0.924–1.004 0.0773 0.975 0.935–1.017 0.2374

Fasting glucose, mg/dL 0.989 0.975–1.002 0.0982 1.022 1.013–1.032 <0.0001

Total cholesterol, mg/dL 0.979 0.967–0.992 0.0011 1.007 0.996–1.017 0.1994

Hemoglobin, g/dL 0.922 0.864–0.984 0.0138 0.907 0.845–0.973 0.0063

Abbreviations: CI = confidence interval; OR = odds ratio.

Adjusted for age, sex, comorbidities, Charlson Comorbidity Index, socioeconomic status, number of health care visits.

Results without PD. Patients with PD had a higher prevalence of hy-

pertension and osteoporosis and an increased number of health
Patient Characteristics care visits during the follow-up period, whereas dyslipidemia was
During the follow-up period, 2,665 patients developed PD. less prevalent in patients with PD compared to the values in
Table 1 shows the demographic and medical characteristics of those without PD. The person-years of follow-up duration was
persons with PD at the time of diagnosis and those who were not 23,883 for PD group and 4,784,835 person-year for those who
diagnosed with PD at the final follow-up. The mean age of never diagnosed with PD. The incidence rate of PD per 100,000
persons with PD at the time of diagnosis was 70.98 ± 8.24 years, person-year was 55.4.
which was approximately six years higher than those without a
diagnosis of PD at last healthcare visit. Patients with PD showed Logistic Regression With Time-Dependent
a lower proportion of current smokers and were less likely to Covariates Using GEE Models for
drink alcohol or perform regular physical activities than those PD Development
without PD. The mean values of systemic blood pressure and Table 2 displays the ORs for PD development using the GEE
fasting glucose were higher, whereas those of total cholesterol models. After adjusting for age, sex, comorbidities, number of
and hemoglobin were lower in patients with PD than in those health care visits, and socioeconomic status, smoking habit

Neurology.org/N Neurology | Volume 98, Number 6 | February 8, 2022 e647

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
 measured based on repeated health screening data was signifi-
cantly related to a decreased rate of PD development (ex- Figure 2 Percentages of Lifestyle Factors According to
smoker, OR 0.803, 95% confidence interval [CI] 0.733–0.879; Parkinson Disease Presence
current smoker, OR 0.568, 95% CI 0.503–0.642), and a dose–
response relationship was observed. Alcohol consumption (≤3
per week, OR 0.699, 95% CI 0.649–0.752; ≥4 per week, OR
0.715, 95% CI 0.622–0.822) and regular physical activities
(moderate, OR 0.838, 95% CI 0.802–0.876; vigorous, OR 0.917,
95% CI 0.862–0.976) were also related with reduced PD de-
velopment. However, the dose–response relationship was not
clear.

When analyzed separately by sex, there were some differences

in the relationship between lifestyle factors and PD occur-
rence (Table 3). Smoking was related to decreased occur-
rence of PD only in men (ex-smoker, OR 0.782, 95% CI
0.713–0.858; current smoker, OR 0.556, 95% CI
0.488–0.632). Alcohol consumption and physical activity
showed inverse association with PD in both men and women;
however, alcohol consumption in men (≤3 per week, OR
0.717, 95% CI 0.658–0.780; ≥4 per week, OR 0.745, 95% CI
0.644–0.861) and physical activities in women (moderate,
OR 0.792, 95% CI 0.748–0.840; vigorous, OR 0.830, 95% CI
0.756–0.911) showed more consistent relationship with PD
development compared to that of either sex.

Hemoglobin levels showed a negative relationship with PD

development in both sexes. Systolic blood pressure (OR
1.062, 95% CI 1.033–1.091) and fasting glucose level (OR
1.022, 95% CI 1.013–1.032) were associated with an in-
creased risk of PD in women, whereas total cholesterol level
(OR 0.979, 95% CI 0.967–0.992) was associated with a de-
creased risk of PD in men only.

Sensitivity Analyses
We performed sensitivity analyses according to the timing and
regularity of health screening and age. Figure 2 displays life- (A) Smoking. (B) Alcohol consumption. (C) Physical activity.

style factors, including participant’s smoking habit, alcohol

consumption, and regular physical activity, stratified by PD
development in the last 3 health screening points before the
PD diagnosis or last health care visit, respectively. The mean We also performed GEE models according to the regularity of
duration from the first health screening to PD onset during the health screening. Participants were divided based on
the 3 recent health screenings was 5.5 ± 2.4 years. When we whether more than 3 times the health screening data were
performed the logistic regression analysis with time- obtained over the 6-year period or not. Participants who
dependent covariates for PD development using GEE mod- underwent regular health screening showed a significant re-
els using the last 3 health screening data, smoking (ex-smoker, lationship with PD and lifestyle factors, including smoking
OR 0.822, 95% CI 0.765–0.883; current smoker, OR 0.567, habit, alcohol consumption, and regular physical activity.
95% CI 0.518–0.619) showed a similar inverse relationship However, in the analysis with participants without regular
with decreased PD development. Alcohol consumption (≤3 health screening, the relationship between lifestyle factors,
per week, OR 0.561, 95% CI 0.520–0.605; ≥4 per week, OR including that of smoking habit, and PD occurrence decreased
0.603, 95% CI 0.533–0.683) and physical activities (moder- (Figure 4).
ate, OR 0.749, 95% CI 0.711–0.788; vigorous, OR 0.767, 95%
CI 0.717–0.821) also showed inverse association with PD We analyzed the effects of lifestyle factors on PD development
risk. Furthermore, the ORs of those lifestyle factors were in participants aged ≥60 years separately, and the results
somewhat reduced in the analysis including the last 3 health showed an almost consistent pattern compared to the analysis
screening data compared to the overall follow-up period including those aged ≥40 years (eTable 1, links.lww.com/
health screening data (Figure 3). WNL/B613).

e648 Neurology | Volume 98, Number 6 | February 8, 2022 Neurology.org/N

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
We performed GEE models for the relationship between baseline + follow-up), and ethnicity. Lifestyle factors change
socioeconomic status and PD development; participants with over time; thus, it could be difficult to predict PD de-
high income level showed relationship with increased PD risk velopment using baseline lifestyle factors, especially in long-
(eTable 2, links.lww.com/WNL/B613). Sensitivity analysis term follow-up studies. Several studies examined follow-up
stratified by income level was performed and the association lifestyle factors to predict PD development.12,13 Alcohol
between lifestyle factors and PD development remained consumption was measured at baseline and after 14 years for
consistent (eTable 3, links.lww.com/WNL/B613). PD development, in a mean follow-up years of 17.9 years.12
Participants were followed up to assess smoking habit 6 times
during the 40-year follow-up period.13 The number of follow-
Discussion up visits was small considering the study period, and the
We longitudinally followed 455,939 participants who had re- follow-up interval seemed relatively long. In addition, many
peated measures of health screening data, at least 3 times, and previous studies focusing on the relationship between specific
performed logistic regression analysis with time-dependent lifestyle factors and PD did not consider covariates such as
covariates using GEE models to evaluate how repeated mea- comorbidities or socioeconomic status.9,13,22 Thus, we tried
sures of lifestyle factors were associated with PD development. to investigate how repeated measures of lifestyle factors were
Smoking showed a dose–response inverse association with PD associated with PD occurrence by adjusting for confounding
only in men. Alcohol consumption and regular physical activity variables such as comorbidities and socioeconomic status.
were related to reduced PD risk in both sexes; however, alcohol
consumption in men and physical activity in women presented We performed additional analyses according to the timing and
more consistent relationships with PD development compared regularity of the health screening in this study. Analysis in-
to those of either sex. Regarding the timing of health screening, cluding the last 3 health screening data showed more reduced
the latest health screening data before the onset of PD showed a ORs for PD development compared to those of the whole
greater association with PD development. Based on the regu- follow-up period, which implies that recent lifestyle charac-
larity of health screening, participants with regular health teristics are more related to PD development. On the other
screening showed consistent association of lifestyle factors with hand, reverse causality could exist in the results, when using
PD development, whereas lifestyle factors without regular more recent data from PD onset, and causal relevance needs
health screening showed reduced association with PD de- further investigation in future studies.
velopment, even for smoking habit.
The regularity of health screening showed a somewhat dif-
Several previous studies’ results showed the relationship be- ferent relationship with PD occurrence. Participants with
tween lifestyle factors and PD, with some controversy.8,9,12,22 regular follow-up (at least 3 times over the 6 years) showed a
These discrepancies across studies might have originated significant association between lifestyle factors and PD de-
from differences in study designs (case–control vs longitudi- velopment, whereas lifestyle data with irregular follow-up
nal cohort), characteristics of lifestyle factors (baseline vs showed decreased relationship with PD, even smoking habit.

Figure 3 Adjusted ORs for Parkinson Disease Development Using the Latest Three Health Screening Data

CI = confidence interval; OR odds ratio.

Neurology.org/N Neurology | Volume 98, Number 6 | February 8, 2022 e649

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
 Figure 4 Adjusted ORs for Parkinson Disease Development According to the Regularity of Health Screening

(A) Regular checkup. (B) Irregular checkup. CI = confidence interval; OR odds ratio.

In this study, we included participants with at least 3 but it was not consistent in women. This finding is somewhat
follow-up lifestyle data, and followed them up for 14 years. in agreement with previous findings of a significantly de-
Thus, the mean interval between health screenings might creased rate of PD in current smokers but was nonsignificant
be less than 5 years, which is not that long compared to between ex-smokers and PD occurrence in women.9,24
those of previous studies. Whether irregularity or longer Smoking and PD are both prevalent in men compared with
follow-up interval affected the decreased association be- women25; thus, a relatively low number of participants with
tween lifestyle factors and PD is uncertain in this study. smoking habit or PD might be inadequate to detect the as-
Future studies of lifestyle factors and PD with repeated sociation in women. Another possible explanation for the sex
measures of lifestyle factors are needed, and the effects of difference could be the hormonal effects in women. Previous
timing and regularity of lifestyle factors need further in- studies have reported an attenuated risk-reducing effect for
vestigation for PD prediction. PD in women who have used postmenopausal hormones.26,27
Physical activity has shown protective effects on PD occur-
Smoking has shown an inverse dose–response relationship rence and progression.10,22,28 A meta-analysis in 2018 showed
with PD in many previous studies.9,13,23,24 Our results with an inverse dose-dependent relationship between physical ac-
repeated measures of smoking habit showed similar dose– tivity and PD development only in men,10 which is contrary to
response relationship with smoking and decreased risk of PD, our results. One suggested explanation for the difference

e650 Neurology | Volume 98, Number 6 | February 8, 2022 Neurology.org/N

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
between sexes in the physical activity effect on PD risk might information on these variables was not included in the data-
be the lower percentage of women included in most studies.7 base. Instead, we tried to adjust for other confounding vari-
On the other hand, all included studies in the meta-analysis ables, including comorbidities and socioeconomic status.
were performed in Western countries,22,29-31 and some of Third, selection bias could exist. The claims-based study only
them only used baseline physical activity for PD prediction. included patients with hospital visit and did not include pa-
There has been a report that additive interaction effect between tients with mild symptoms who might not use health care
caffeine and physical activity was found only in women.32 facilities. Fourth, reverse causality might exist between life-
Considering these results together with ours, we thought that style factors and PD occurrence in this observation study and
male predominance may not be conclusive to explain the effects future studies to elucidate causal relevance are warranted.
of physical activity on PD, and studies with increased number of Fifth, we could not analyze the effect of the intensity and
female participants with appropriate repeated measure of phys- duration of each lifestyle factor on PD development. In this
ical activity for PD risk are warranted. study, we used repeated measures of health screening data,
and the summation of baseline and follow-up data on each
Alcohol consumption’s effect on PD risk has been lifestyle factor was difficult. We evaluated how repeated
controversial.11,12,33,34 The inverse relationship between al- measures of lifestyle factors were associated with PD occur-
cohol consumption and PD was mainly observed in retro- rence differently according to the timing and regularity of
spective case–control studies, but has not been clear from health screening. The results of this study raise awareness
prospective studies.33,34 These inconsistent results might be about the importance of the characteristics of lifestyle factors
related to recall bias, reverse causality, or residual confounding in studies on PD risk. Future studies including appropriate
of smoking habit.34 In this study, we used the national health lifestyle factors could increase the prediction of PD risk, and
screening data that were prospectively collected and adjusted more studies in Asian countries are warranted.
for various confounding factors including smoking habit. In
the subgroup analysis with the last 3 health screening data, Study Funding
alcohol consumption reduced PD risk by 56%, whereas al- This work was supported by the National Research Founda-
cohol consumption during the entire follow-up duration re- tion of Korea (NRF) grant funded by the Korea government
duced PD risk by 70%. There could be a harm-avoiding (MSIT) (2020R1C1C1006867).
personality whereby patients with premorbid PD with pro-
dromal PD symptoms discontinued alcohol consumption but Disclosure
subsequently then developed PD. Thus, based on our results, The authors report no disclosures relevant to the manuscript.
the causal association between alcohol consumption and PD Go to Neurology.org/N for full disclosures.
is also unclear. In our results, the inverse relationship between
alcohol consumption and PD risk was more consistent in men Publication History
compared to women. Women who consumed alcohol ≥ 4 Received by Neurology April 16, 2021. Accepted in final form
times per week did not show significant association with PD September 24, 2021.
risk. Since there were only 67 persons who drank alcohol ≥ 4
times per week in PD group, these results must be interpreted
Appendix Authors
with caution until they can be replicated.
Name Location Contribution
This study has several limitations. First, the diagnoses of PD Seo Yeon Department of Physical Wring of the first draft,
and comorbidities were based on ICD codes of the NHIS Yoon, MD, Medicine and Rehabilitation, research project (execution),
health checkup cohort database. Thus, there might be inac- PhD Korea University Guro statistical analysis (execution)
Hospital, Seoul, Republic of
curacies of the claim code leading to misclassification of the Korea
diseases. We excluded patients with PD combined with di-
You Hyun Department of Biostatistics, Wring of the first draft,
agnosis of secondary parkinsonism or atypical parkinsonism Park, Yonsei University, Seoul, statistical analysis (design,
to more precisely define PD, which could also exclude par- MPH Korea execution)

ticipants with PD. The patients with PD diagnosis less than Hyo Jeong Department of Rehabilitation Statistical analysis (design,
three times were also excluded. Although we tried to exclude Lee, MD Medicine, Bundang Jesaeng review, and critique)
General Hospital, Gyeonggi-
patients who were misdiagnosed with PD, it is possible that do, Korea
patients with PD who had a follow-up loss or died before next
Dae Department of Precision Research project (conception),
clinic visit have been also excluded. In addition, some Ryong Medicine & Biostatistics, statistical analysis (design)
comorbidities, such as dementia and acquired immunodefi- Kang, PhD Yonsei University Wonju
College of Medicine, Wonju,
ciency syndrome, were not available in this NHIS cohort data Korea
and could not be included in the analysis. Second, caffeine
consumption and dietary habits such as dairy food or vitamin Yong Department and Research Research project (conception,
Wook Institute of Rehabilitation organization, execution,
E intake that have shown effects on PD development were not Kim, MD, Medicine, Yonsei University review, and critique)
included in this analysis.32-34 This study used the national PhD College of Medicine, Seoul,
Republic of Korea
health checkup cohort data from the NHSP; however,

Neurology.org/N Neurology | Volume 98, Number 6 | February 8, 2022 e651

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
 References 18. Hu FB, Goldberg J, Hedeker D, Flay BR, Pentz MA. Comparison of population-
1. Martinez-Martin P, Macaulay D, Jalundhwala YJ, et al. The long-term direct and averaged and subject-specific approaches for analyzing repeated binary outcomes. Am
indirect economic burden among Parkinson’s disease caregivers in the United States. J Epidemiol. 1998;147(7):694-703.
Mov Disord. 2019;34(2):236-245. 19. Sullivan Pepe M, Anderson GL. A cautionary note on inference for marginal re-
2. Weir S, Samnaliev M, Kuo TC, et al. Short- and long-term cost and utilization of gression models with longitudinal data and general correlated response data. Commun
health care resources in Parkinson’s disease in the UK. Mov Disord. 2018;33(6): Statistics-Simulation Comput. 1994;23(4):939-951.
974-981. 20. Lalonde TL, Nguyen AQ, Yin J, Irimata K, Wilson JR. Modeling correlated binary
3. Müller-Nedebock AC, van der Westhuizen FH, Kõks S, Bardien S. Nuclear genes outcomes with time-dependent covariates. J Data Sci. 2013;11.
associated with mitochondrial DNA processes as contributors to Parkinson’s disease 21. Hedeker D, Gibbons RD. Longitudinal Data Analysis. John Wiley & Sons; 2006.
risk. Mov Disord. 2021;36(4):815-831. 22. Yang F, Trolle Lagerros Y, Bellocco R, et al. Physical activity and risk of Parkinson’s
4. Heckman MG, Labbé C, Kolicheski AL, et al. Fine-mapping of the non-coding var- disease in the Swedish national March cohort. Brain. 2015;138(pt 2):269-275.
iation driving the Caucasian LRRK2 GWAS signal in Parkinson’s disease. Parkin- 23. Kim R, Yoo D, Jung YJ, Han K, Lee JY. Smoking cessation, weight change, and risk of
sonism Relat Disord. 2021;83:22-30. Parkinson’s disease: analysis of national cohort data. J Clin Neurol. 2020;16(3):455-460.
5. Nag N, Jelinek GA. A narrative review of lifestyle factors associated with Parkinson’s 24. Kim R, Yoo D, Jung YJ, Han K, Lee JY. Sex differences in smoking, alcohol con-
disease risk and progression. Neurodegen Dis. 2019;19(2):51-59. sumption, and risk of Parkinson’s disease: a nationwide cohort study. Parkinsonism
6. Belvisi D, Pellicciari R, Fabbrini G, Tinazzi M, Berardelli A, Defazio G. Modifiable risk Relat Disord. 2020;71:60-65.
and protective factors in disease development, progression and clinical subtypes of 25. Wooten GF, Currie LJ, Bovbjerg VE, Lee JK, Patrie J. Are men at greater risk for
Parkinson’s disease: what do prospective studies suggest? Neurobiol Dis. 2020;134: Parkinson’s disease than women? J Neurol Neurosurg Psychiatry. 2004;75(4):637-639.
104671. 26. Simon KC, Chen H, Gao X, Schwarzschild MA, Ascherio A. Reproductive factors,
7. Marras C, Canning CG, Goldman SM. Environment, lifestyle, and Parkinson’s dis- exogenous estrogen use, and risk of Parkinson’s disease. Mov Disord. 2009;24(9):
ease: Implications for prevention in the next decade. Mov Disord. 2019;34(6): 1359-1365.
801-811. 27. Gatto NM, Deapen D, Stoyanoff S, et al. Lifetime exposure to estrogens and Parkinson’s
8. Paul KC, Chuang YH, Shih IF, et al. The association between lifestyle factors and disease in California teachers. Parkinsonism Relat Disord. 2014;20(11):1149-1156.
Parkinson’s disease progression and mortality. Mov Disord. 2019;34(1):58-66. 28. Amara AW, Chahine L, Seedorff N, Caspell-Garcia CJ, Coffey C, Simuni T. Self-
9. Gallo V, Vineis P, Cancellieri M, et al. Exploring causality of the association between reported physical activity levels and clinical progression in early Parkinson’s disease.
smoking and Parkinson’s disease. Int J Epidemiol. 2019;48(3):912-925. Parkinsonism Relat Disord. 2019;61:118-125.
10. Fang X, Han D, Cheng Q, et al. Association of levels of physical activity with risk of 29. Chen H, Zhang SM, Schwarzschild MA, Hernán MA, Ascherio A. Physical activity and
Parkinson disease: a systematic review and meta-analysis. JAMA Netw Open. 2018; the risk of Parkinson disease. Neurology. 2005;64(4):664-669.
1(5):e182421. 30. Logroscino G, Sesso HD, Paffenbarger RS Jr., Lee IM. Physical activity and risk of
11. Bettiol SS, Rose TC, Hughes CJ, Smith LA. Alcohol consumption and Parkinson’s Parkinson’s disease: a prospective cohort study. J Neurol Neurosurg Psychiatry. 2006;
disease risk: a review of recent findings. J Parkinsons Dis. 2015;5(3):425-442. 77(12):1318-1322.
12. Kim IY, Yang TO, Heath AK, et al. Alcohol intake and Parkinson’s disease risk in the 31. Thacker EL, Chen H, Patel AV, et al. Recreational physical activity and risk of
million women study. Mov Disord. 2020;35(3):443-449. Parkinson’s disease. Mov Disord. 2008;23(1):69-74.
13. Mappin-Kasirer B, Pan H, Lewington S, et al. Tobacco smoking and the risk of 32. Kim IY, O’Reilly É J, Hughes KC, et al. Integration of risk factors for Parkinson disease
Parkinson disease: a 65-year follow-up of 30,000 male British doctors. Neurology. in 2 large longitudinal cohorts. Neurology. 2018;90(19):e1646-e1653.
2020;94(20):e2132–e2138. 33. Jiménez-Jiménez FJ, Alonso-Navarro H, Garcı́a-Martı́n E, Agúndez JAG. Alcohol
14. Bae JM, Chung KY, Yun SJ, et al. Markedly reduced risk of internal malignancies in consumption and risk for Parkinson’s disease: a systematic review and meta-analysis.
patients with vitiligo: a nationwide population-based cohort study. J Clin Oncol 2019; J Neurol. 2019;266(8):1821-1834.
37:903-911. 34. Peters S, Gallo V, Vineis P, et al. Alcohol consumption and risk of Parkinson’s disease:
15. Han TY, Lee JH, Noh TK, et al. Alopecia areata and overt thyroid diseases: A data from a large prospective European cohort. Mov Disord. 2020;35:1258-1263.
nationwide population-based study. J Dermatol 2018;45:1411-1417. 35. Hong CT, Chan L, Bai CH. The effect of caffeine on the risk and progression of
16. Moon HI, Shin J, Kim YW, Chang JS, Yoon S. Diabetic polyneuropathy and the risk of Parkinson’s disease: a meta-analysis. Nutrients. 2020;12(6).
developing carpal tunnel syndrome: A nationwide, population-based study. Muscle 36. Hughes KC, Gao X, Kim IY, et al. Intake of dairy foods and risk of Parkinson disease.
Nerve 2020;62:208-213. Neurology. 2017;89(1):46-52.
17. Bondy SJ, Victor JC, Diemert LM. Origin and use of the 100 cigarette criterion in 37. Etminan M, Gill SS, Samii A. Intake of vitamin E, vitamin C, and carotenoids and the
tobacco surveys. Tob Control. 2009;18(4):317-323. risk of Parkinson’s disease: a meta-analysis. Lancet Neurol 2005;4(6):362-365.

e652 Neurology | Volume 98, Number 6 | February 8, 2022 Neurology.org/N

Copyright © 2021 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
Lifestyle Factors and Parkinson Disease Risk: Korean Nationwide Cohort Study With
Repeated Health Screening Data
Seo Yeon Yoon, You Hyun Park, Hyo Jeong Lee, et al.
Neurology 2022;98;e641-e652 Published Online before print October 14, 2021
DOI 10.1212/WNL.0000000000012942

This information is current as of October 14, 2021

Updated Information & including high resolution figures, can be found at:
Services http://n.neurology.org/content/98/6/e641.full

References This article cites 34 articles, 7 of which you can access for free at:
http://n.neurology.org/content/98/6/e641.full#ref-list-1
Subspecialty Collections This article, along with others on similar topics, appears in the
following collection(s):
Class II
http://n.neurology.org/cgi/collection/class_ii
Cohort studies
http://n.neurology.org/cgi/collection/cohort_studies
Parkinson's disease/Parkinsonism
http://n.neurology.org/cgi/collection/parkinsons_disease_parkinsonism
Permissions & Licensing Information about reproducing this article in parts (figures,tables) or in
its entirety can be found online at:
http://www.neurology.org/about/about_the_journal#permissions
Reprints Information about ordering reprints can be found online:
http://n.neurology.org/subscribers/advertise

Neurology ® is the official journal of the American Academy of Neurology. Published continuously since
1951, it is now a weekly with 48 issues per year. Copyright © 2021 American Academy of Neurology. All
rights reserved. Print ISSN: 0028-3878. Online ISSN: 1526-632X.