Is Environmental and Occupational Particulate Air Pollution Exposure Related to Type-2 Diabetes and Dementia? A Cross-Sectional Analysis of the UK Biobank

Human exposure to particulate air pollution (e.g., PM2.5) can lead to adverse health effects, with compelling evidence that it can increase morbidity and mortality from respiratory and cardiovascular disease. More recently, there has also been evidence that long-term environmental exposure to particulate air pollution is associated with type-2 diabetes mellitus (T2DM) and dementia. There are many occupations that may expose workers to airborne particles and that some exposures in the workplace are very similar to environmental particulate pollution. We conducted a cross-sectional analysis of the UK Biobank cohort to verify the association between environmental particulate air pollution (PM2.5) exposure and T2DM and dementia, and to investigate if occupational exposure to particulates that are similar to those found in environmental air pollution could increase the odds of developing these diseases. The UK Biobank dataset comprises of over 500,000 participants from all over the UK. Environmental exposure variables were used from the UK Biobank. To estimate occupational exposure both the UK Biobank’s data and information from a job exposure matrix, specifically developed for UK Biobank (Airborne Chemical Exposure–Job Exposure Matrix (ACE JEM)), were used. The outcome measures were participants with T2DM and dementia. In appropriately adjusted models, environmental exposure to PM2.5 was associated with an odds ratio (OR) of 1.02 (95% CI 1.00 to 1.03) per unit exposure for developing T2DM, while PM2.5 was associated with an odds ratio of 1.06 (95% CI 0.96 to 1.16) per unit exposure for developing dementia. These environmental results align with existing findings in the published literature. Five occupational exposures (dust, fumes, diesel, mineral, and biological dust in the most recent job estimated with the ACE JEM) were investigated and the risks for most exposures for T2DM and for all the exposures for dementia were not significantly increased in the adjusted models. This was confirmed in a subgroup of participants where a full occupational history was available allowed an estimate of workplace exposures. However, when not adjusting for gender, some of the associations become significant, which suggests that there might be a bias between the occupational assessments for men and women. The results of the present study do not provide clear evidence of an association between occupational exposure to particulate matter and T2DM or dementia.


Figures
The red arrow represents assumed causal relationship between particulate air pollution exposure and T2DM. Blue arrows represent implied confounding paths before any covariate adjustment has been applied. Adjustments were made for all the variables, except for BP that was a potential mediator. The red arrow represents assumed causal relationship between particulate air pollution exposure and dementia. Blue arrows represent implied confounding paths before any covariate adjustment has been applied. Adjustments were made for all the variables, except for BP that was a potential mediator. Tables   Table S1. Percentages used to create exposure score (P×L) [1][2][3]

Recategorisation of covariates
Age was constructed from variable 34 from the UK Biobank.
Townsend deprivation index was based on variable 189 and on the preceding national census output areas. Each participant was assigned a score corresponding to the output area in which their postcode was located, and 5 levels were constructed, with level 1 corresponding to the least deprived areas and level 5 to the most deprived areas.
Cholesterol levels is a risk factor for dementia, and blood pressure medication is a risk factor for both diabetes and dementia, therefore the variables 6177 (information only for males) and 6153 (information only for females) from UK Biobank were used to extract this information. Firstly, all the columns of each variable were merged to one, as 12 different measurements were available, and then recategorised, to obtain one new variable that represented cholesterol medication (Yes/No/Unknown) and another one that represented blood pressure medication (Yes/No/Unknown).
To investigate diet as a risk factor, we recategorised variable 1538, which represents major dietary changes in the last 5 years. We merged the information from all the instances, by keeping firstly the answer "Yes, because of illness" and then the answer "Yes, because of other reasons" and then the "No" and lastly the "Prefer not to answer" responds. Therefore, we created a variable with 4 levels.
Stroke, which is a risk factor for dementia, was extracted from data fields 42006. This information was algorithmically defined from Hospital Episode data, so it was gathered at specific instances, but it is continually updated from the NHS linkage systems and offers the participant's information from 1975 up to date. Thus, the dates in this variable were replaced with a "Yes" and all the other with a "No".
Physical activity was extracted from variable 22040, which is the summed Total Metabolic Equivalent Task (MET) minutes per week for all activity, including walking, moderate and vigorous activity. We created 3 different levels, low, moderate, and high according to IPAQ (International Physical Activity Questionnaire) guidelines.
From variables 20107 and 20110 we extracted information about parental illnesses (illnesses of father and mother accordingly). From 31 columns, that provided a follow-up information for parental diabetes and dementia history, we created 2 columns for parental diabetes history (father and mother) and 2 columns for parental history of dementia. "Yes" corresponds to diabetes or dementia family history, "No" corresponds to other conditions and then "Prefer not to answer" and "Do not know".