Association between Long-term Exposure to PM 2.5 and Incidence of
Type 2 Diabetes in Taiwan
Chia-Ying Li
1, Huey-Jen Su
1,*, Chih-Da Wu
2, Wen-Chi Pan
3, Yi-Chen Chen
11 Department of Environmental and Occupational Health, Medical College, National Cheng Kung University
2Department of Forestry and Natural Resources, College of Agriculture, National Chiayi University
3Institute of Environmental and Occupational Health Sciences, National Yang-Ming University
*Corresponding author: hjsu@mail.ncku.edu.tw
ISEE-ISES AC2016
June 26-29, 2016, Sapporo, Japan
Conflict of Interest
I have no financial relationships to disclose.
PM 2.5 in Taiwan
2006-2012
Characteristic Concentration (µg/m
3)
Mean (± SD) 32.59 (±8.15)
Maximum 47.20
Minimum 13.31
IQR 12.03
Air pollution such as PM
2.5is a big problem in Asia.
16/10/17 3
Air quality for PM
2.5 Concentration limit (µg/m3)24-hour Annual
WHO 25 10
Japan 35 15
Republic of Korea 50 25
China Class 1 35 15
Class 2 75 35
Taiwan 35 15
Effects of PM 2.5 on Type 2 Diabetes - Hypothesized Mechanism-
Metabolic abnormali.es
PM 2.5
Insulin
TLRs, NLRs receptors as
sensors
(Weichenthal et al., 2014; Laing et al., 2010; Xu X et al., 2010;Mills et al., 2005; Sun et al., 2009)
(Brook et al., 2013; Chen Jet al., 2008; Dubowsky et al., 2006;
Xu et al., 2011; Steinvil et al., 2008)
(Laing et al., 2010;
Kampfrath et al., 2011;
Zheng et al., 2012)
Brown adipose .ssue
altera.ons
Endothelial dysfunc.on
Endoplasmic re.culum stress
Inflamma.on (visceral adipose .ssue)
Mitochondrial dysfunc.on
Increase makers of inflamma.on (CRP, IL-‐6)
(Dominici et al., 2007; Liberda et al., 2010; Li et al., 2011;
Shi et al., 2006; Vandanmagsar et al., 2011)
(Xu et al., 2011; Zhong et al., 2011; Lowell et al., 2005)
(Kampfrath et al., 2011; Zheng et al., 2012; Laing et al., 2010)
(Knight-‐Lozano et al., 2002;
Xu et al., 2011)
(Weichenthal et al., 2014; Dubowsky et al., 2006; Zeka et al., 2006;
Lee et al., 2011; Hoffmann et al.,2009)
Type 2 diabetes
Objectives
• To investigate the impact of PM 2.5 on the incidence of type 2 diabetes, utilizing retrospective cohort study design.
• To assess the dose–response relationship between PM 2.5 exposure level and incidence of type 2
diabetes.
5
16/10/17
Study Design and Frame Work Retrospective Cohort
Statistical Analysis
Cox proportional hazards model
Individual Exposure Concentration National Health Insurance
Research Database
Longitudinal Health Insurance Database 2000 (LHID 2000)
Study Population
Newly diagnosed with diabetes
Nondiabetic adults
Exposure Database
Taiwan Air Quality Monitoring Network
Adjusted Factors
16/10/17 7
Longitudinal Health Insurance Database 2000 (LHID 2000)
Study Population
(recruited adults without diabetes) 2001 – 2012
n=505,151
Exclude
1. Subjects with diagnosed diabetes before 2001
2. Subjects who were < 20 years old
Case subjects: Newly diagnosed with diabetes
• Concurrent use of antihyperglycemic medications
• Subjects with diagnosis code 250 three times in a year
Control subjects: without diabetes diagnosis
Selection Criteria
PM 2.5 Exposure
• Taiwan Air Quality Monitoring Network
• Hourly data during 2006-2012 (data was available since 2006)
• 71 stationary sites located around the Main Taiwan Island
• Individual Exposure Concentration
– Ordinary Kriging and zonal statistics (ArcGIS 10.2)
– According to the registered location of medical facilities of each
subject
Statistical Analysis
• Cox proportional hazards model
– Hazard Ratio (95% CI)
• Adjusted factors
– Individual-level variables (age, sex, insurance amount, and occupational type)
– Comorbidities (hypertension and hyperlipidemia)
– Area-level variables (township urbanization level and county- level income)
16/10/17 9
λ(T=t|X=x)= λ
0(t)exp(β
1X
1+ β
2TX
2+β
3TX
3+β
4TX
4+
β
5TX
5+
β
6TX
6+
β
7TX
7)
Baseline characteristics
• Of the 505,151 eligible participants enrolled in the study, a total of 48,611 newly developed diabetes cases were
identified.
• The incidence of type 2 diabetes in the study sample was 1.14%
Baseline characteristics Statistics
Age 41.6 ± 15.80 years
Male 48.74%
Female 51.26%
Insurance amount NTD$18,092 ± 15,331.40 (about USD$ 560.97)
Baseline characteristics of study population (n=505,151)
16/10/17
Long-term PM 2.5 & diabetes incidence
11
The results suggest that per 12.03 µg/m
3increase of long-term exposure to PM
2.5was contributing to a 4.9% increase of the risk for diabetes incidence in Taiwan.
Association between long-term PM
2.5exposure and incidence of type 2 diabetes
Model HR (95% CI) per IQR increment of
PM
2.5§P-value
Crude model 1.048 (1.033-1.064) <0.001
Addition of individual-level covariates
a1.036 (1.021-1.051) <0.001 Addition of comorbid conditions
b1.056 (1.041-1.071) <0.001 Addition of area-level covariate
c1.049 (1.033-1.065) <0.001
§Interquartile range (IQR)=12.03 µg/m3
aAdjusted for age group, sex, level of insurance amount and occupational type.
bAlso adjusted for hypertension and hyperlipidemia.
cAlso adjusted for county level income and township level urbanization.
Sensitivity analysis
Sensitivity analysis for the associations of incident diabetes with quartile of PM
2.5Model No. of cases Adjusted HR (95% CI)
aP-value
Quartile of PM
2.5First 105363 1
Second 180233 0.974 (0.946-1.002) 0.0652
Third 98628 1.042 (1.009-1.077) 0.0126
Forth 120927 1.082 (1.050-1.114) <0.0001
Restricted to participants with
No relocation 267267 1.046 (1.025-1.067) <0.0001
Diabetes diagnosis with using antihyperglycemic
medications
484960 1.095 (1.072-1.118) <0.0001
aModel was adjusted for age group, sex, level of insurance amount, occupational type, hypertension, hyperlipidemia,
16/10/17
Stratified analysis
13
Stratified analysis for the hazard ratio of incident diabetes and per IQR increment of PM2.5
Characteristic No. of cases Total number HR (95% CI) P-value Sexa
male 26,312 246,196 1.063 (1.040-1.086) <0.0001
female 22,299 258,955 1.028 (1.004-1.053) <0.05
Ageb
<65 33519 445,094 1.026 (1.006-1.046) 0.011
≧65 15092 60,057 1.035 (1.006-1.064) 0.016
Insurance amountc
low 19259 222617 1.071 (1.044-1.098) <0.0001
high 29352 282534 1.037 (1.016-1.058) 0.001
Hypertensiond
no 19751 367009 1.038 (1.012-1.065) 0.004
yes 28860 138142 1.034 (1.013-1.055) 0.016
Hyperlipidemiad
no 29733 419515 1.027 (1.006-1.048) 0.0107
yes 18878 85636 1.051 (1.024-1.079) <0.001
aModel was adjusted for age group, level of insurance amount, occupational type, hypertension, hyperlipidemia, county, county level income, and township urbanization.
bModel was adjusted for sex, level of insurance amount, occupational type, hypertension, hyperlipidemia, county, county level income, and township urbanization.
cModel used the median amount of the insurance (19200 NTD) as cutoff value and adjusted for age group, sex, occupational type, hypertension, hyperlipidemia, county, county level income, and township urbanization.
dModel was adjusted for sex, age, level of insurance amount, occupational type, county, county level income, and township urbanization.
The study observed stronger associations between diabetes and PM
2.5among participants who were males, > 65 years of age, or had lower insurance
amount.
Dose-response curve
Quartile of PM2.5 Adjusted HR (95% CI) P-value
First 1
Second 0.974 (0.946-1.002) 0.0652 Third 1.042 (1.009-1.077) 0.0126 Forth 1.082 (1.050-1.114) <0.0001
16/10/17 15
Summary of previous findings
Location Subject Air pollution Exposure
Time Result
Taiwan
(This study) 505,151 adults PM2.5
(IQR=12.03 µg/m3) 12 years HR=1.05 (1.033-1.065) 1.005 for 1 µg/m3
Ruhr,
Germany 1,775 women PM
(IQR=10 µg/m3) 16 years HR=1.15 (1.04–1.27) 1.014 for 1 µg/m3 Los Angeles 3,992 black women
PM2.5
(10 µg/m3) 10 years IRR=1.63 (0.78–3.44) 1.051 for 1 µg/m3
U.S.
NHS:
74,412 female nurses
HPFS:
15,048 male health
professionals
PM2.5 PM10 PM10–2.5
(IQR=4 µg/m3)
12 months
HR=1.03 (0.96–1.10) for PM2.5 1.007 for 1 µg/m3
Ontario,
Canada 62,012 adults PM2.5
(10 µg/m3) 6 years HR=1.11 (1.02-1.21) for PM2.5 1.010 for 1 µg/m3
6 sites, U.S. 6,814 adults PM2.5
(IQR=2.43 µg/m3) 9 years HR=1.05 (0.87-1.26) 1.020 for 1 µg/m3 Ruhr,
Germany 3,607 adults All PM2.5
Traffic-related PM2.5 (1 µg/m3)
5 years
HR=1.03 (0.95–1.12) for all PM2.5 HR=1.36 (0.97–1.89) for traffic- related PM2.5
• Incidence of both hypertension and diabetes is almost twice as high in African American women as in white women.
(Downey et al., 2009)
• Risk for developing diabetes was increased for Asians, Hispanics, and African Americans compared with whites
(Shai et al., 2006; Iris et al., 2006)
Strengths
1. The study represents the general population in Taiwan, and is the largest population-based observational study around the world to evaluate the association between PM
2.5exposure and incident diabetes.
(National Health Insurance Research Database 2014)
2. Assessment of dose-response relationship between PM
2.5exposure level and incidence of type 2 diabetes.
Limitations
1. Misclassification of exposure
The PM
2.5exposure status was based on the concentration derived from the registered location of medical facilities, which may not
representative enough to the real exposure level for each subject.
2. Individual diabetes risk factors
Fail to consider factors including BMI, smoking status, drinking, diet, physical activity, individual education level, marital status, family history and employment status in data analysis, which may lead to
overestimate the risk in the current finding.
3. Undiagnosed cases of diabetes
The study could not identify undiagnosed cases of diabetes in the cohort. It may underestimated the true effect of PM
2.5exposure.
16/10/17 17
Conclusions
1. Long-term exposure to PM
2.5increased the risk of type 2 diabetes in the population of Taiwan.
2. The study observed stronger associations between diabetes and PM
2.5exposure among participants who were < 65 years of age, males, or had lower insurance amount.
3. A dose-response relationship between PM
2.5exposures and
incident diabetes in higher exposure concentration is found.
Thank You for Your Attentions!
Chia-Ying Li
Email: s76034023@mail.ncku.edu.tw
Acknowledgement
This work was supported by Ministry of Science and Technology, Taiwan, R.O.C.
This study is based in part on data from the National Health Insurance Research Database provided by the National Health Insurance Administration, Ministry of Health and Welfare and managed by National Health Research Institutes.
Study Design and Frame Work Retrospective Cohort
Statistical Analysis
• Cox proportional hazards model
Individual Exposure Concentration
• Ordinary Kriging (ArcGIS 10.2)
• based on medical treatment locations of subjects
National Health Insurance Research Database
• Longitudinal Health Insurance Database 2000 (LHID 2000)
• From 2000 to 2012, retrieving millions of people's medical records
Study Population
2001-2012 N=505,151
Newly diagnosed with diabetes
• antihyperglycemic medications
• diagnosis code 250 three times in a year
Nondiabetic adults
Exposure Assessment
Taiwan Air Quality Monitoring Network 2006-2012
Adjusted Factors
• Individual-level variables
• Contextual variables
• Comorbidities
16/10/17 21
Summary of previous finding
Location Subject Air pollution Exposure Time Result
Taiwan
(This study) 505,151 adults PM2.5
(IQR=12.03 µg/m3) 12 years Adjusted HRs:
1.05 (1.033-1.065) Ruhr,
Germany 1,775 women PM
(IQR=10 µg/m3) 16 years Adjusted HRs:
1.15 (1.04–1.27) Los Angeles 3,992 black women
PM2.5
(10 µg/m3) 12 months Adjusted IRRs:
1.63 (0.78–3.44)
U.S.
NHS:
74,412 female nurses
HPFS:
15,048 male health
professionals
PM2.5 PM10 PM10–2.5
(IQR=4 µg/m3)
12 months Adjusted HRs:
1.03 (0.96–1.10) for PM2.5
Ontario,
Canada 62,012 adults PM2.5
(10 µg/m3) 6 years Adjusted HRs:
1.11 (1.02-1.21) for PM2.5 6 sites, U.S. 6,814 adults PM2.5
(IQR=2.43 µg/m3) 9 years Adjusted HRs:
1.05 (0.87-1.26)
Ruhr,
Germany 3,607 adults All PM2.5
Traffic-related PM2.5 (1 µg/m3)
5 years
Adjusted RRs:
1.03 (0.95–1.12) for all PM2.5
Adjusted RRs:
1.36 (0.97–1.89) for traffic- related PM2.5
Location Subject Air pollution Result
Taiwan
(This study) 505,151 adults PM2.5
(1 µg/m3) Adjusted HRs: 1.004 Ruhr,
Germany 1,775 women PM
(1 µg/m3) Adjusted HRs: 1.014 Los Angeles 3,992 black women
PM2.5
(1 µg/m3) Adjusted IRRs: 1.051
U.S.
NHS:
74,412 female nurses
HPFS:
15,048 male health
professionals
PM2.5 PM10 PM10–2.5
(IQR=1 µg/m3)
Adjusted HRs:
1.007 for PM2.5
Ontario,
Canada 62,012 adults PM2.5
(1 µg/m3) Adjusted HRs: 1.010 6 sites, U.S. 6,814 adults PM2.5
(1 µg/m3) Adjusted HR: 1.020
Ruhr, 3,607 adults All PM2.5
Traffic-related PM
Adjusted RRs:
1.03 (0.95–1.12) for all PM2.5
• Incidence of both hypertension and diabetes is almost twice as high in African American women as in white women.
(Downey et al., 2009)
• Risk for developing diabetes was increased for Asians, Hispanics, and African Americans compared with whites
(Shai et al., 2006; Iris et al., 2006)
Summary of previous finding
16/10/17
Introduction
23
PM
2.5and Type 2 Diabetes
ü Long-term PM
2.5exposure may contribute to elevated incidence of type 2 diabetes, but the results were not always significant.
(Kramer et al., 2010; PueZ et al., 2011; Coogan et al., 2012; Chen et al., 2013)
ü Empirical research evidence relating PM
2.5and diabetes still lack in Asia where the PM
2.5concentration is higher and the diabetes burden is greater.
(Ramachandran et al., 2013; Thiering et al., 2015)
PM
2.5Increase makers of inflammation
resistance Insulin Type 2 diabetes Cumulative adverse
health effect
The Ranking Major Causes of Deaths in Taiwan Year
12 7
5 5 5 4 4
5 5 5 4
0 2 4 6 8 10 12 14
1981 1991 2001 2007 2009 2011
Ranking