Population and data collection
We established a cohort of AD high-risk population in Zhejiang Province. Twelve administrative districts of Zhejiang province were divided into 4 groups based on economic levels [17]. From each of these 4 groups, 1 district was systematically selected. Then 1 community was randomly chosen from each district. Subjects met the following criteria were invited to participate: age ≥ 60, living in the selected community for more than 5 years, with normal cognitive function. Exclusion criteria: other types of neurodegenerative diseases such as vascular cognitive impairment, dementia with Lewy bodies, parkinson’s dementia, using anticonvulsants, neuroleptics, antiemetic drugs. Follow-up Flow Chart of the Cohort was showed in Fig. 1.
Questionnaire and basic physical examination
The survey was conducted by trained general practitioners in community health service centers or participants' residences. Data collection included questionnaires and basic physical examinations. The questionnaires was designed by the National Center for Cardiovascular Disease, which was used in sub-centers throughout the country. There were mainly two parts of the questionnaires included of basic information and preliminary screening questionnaire and follow-up questionnaire, which contained hundreds of questions. We collected basic information of participants including age, education level, profession, and geocoded residential addresses, and so on. Basic physical examination included height, weight, BMI, waist, blood pressure, heart rate, blood tests including high-density lipoprotein (HDL), triglyceride (TG), cholesterol (TC), low density lipoprotein (LDL), glucose (GLU) and urine tests. Scales such as Mini-Mental State Examination (MMSE), Montreal cognitive assessment scale (MoCA) used in this study were the most common rapid screening tools for cognitive dysfunction in the world. The scale were translated and revised into many languages and widely used in clinical practice. Chinese version of them were confirmed having good reliability and validity.
The cognitive status was assessed combining MMSE, MoCA and the Hospital Anxiety and Depression Scale (HAD) [18]. Particularly, MMSE, the most commonly used instrument to screen cognitive impairment, showed education and language/cultural bias and it usually took about 5 min to complete [19]. MoCA was more sensitive in the screening of mild cognitive impairment, especially in cases with impairment of a single cognitive domain, such as amnestic cognitive impairment, and it took about 10 min to complete [20]. Using the combination of MMSE and MoCA had brought a number of benefits: firstly, it improved the accuracy of cognitive impairment screening, and could initially perform simple screening for various cognitive disorders. Secondly, it improved the detection rate of cognitive impairment, increased sensitivity and specificity, and reduced false positive rate and false negative rate. The HAD scale was a primary survey of depression and anxiety [18]. The diagnosis process was conducted by specifically trained psychiatrists based on guidelines [20], combined with the results of the MMSE and MoCA, magnetic resonance imaging (MRI) was used when needed.
Hypertension was defined as a mean systolic pressure (SBP) of at least 140 mm Hg or a mean diastolic pressure (DBP) of at least 90 mm Hg, or use of an antihypertensive drug in the past 2 weeks. Physical examination was conducted in accordance with standard procedures, mainly including height and weight. Body mass index (BMI) was defined as weight (kg) divided by height 2 (m2). Normal was defined as BMI < 24 kg/m2, overweight was defined as BMI ≥ 24 kg/m2 and < 28 kg/m2, obesity was defined as BMI ≥ 28 kg/m2[21].Smoking was defined as continuous or cumulative smoking for 6 months or more, while alcohol consumption was defined as drinking at least 2 times a week.
Quality checks were carried out on the measures. In fact, when measuring height and weight, participants were asked to wear light clothing, no shoes. Blood pressure was measured in a sitting position, resting for at least 5 min before measurement, which was measured twice on the right upper arm using a standard electronic sphygmomanometer (Omron HEM-7430). If the difference between two readings was greater than 10 mm Hg, a third measurement was taken and the average of the last two readings was used. MMSE test including of Chinese version was confirmed that having good reliability and validity [22].
Air pollution exposure assessment
The geocoded residential addresses of 1,545 participants were linked to average PM2.5 concentrations between 2013 and 2017, which were estimated from a satellite based spatial statistical model developed by Ma et al. [23]. Briefly, this model was established using the collection 6 aerosol optical depth (AOD) retrieved by the US National Aeronautics and Space Administration (NASA) Moderate Resolution Imaging Spectroradiometer (MODIS), assimilated meteorology data, land use data (fire spots, urban and forest cover, etc.) and PM2.5 concentrations from Chinese ground monitoring network [24]. This model was validated to have little bias in the monthly estimates on PM2.5. For a certain grid cell, the model could not predict the PM2.5 value if the AOD value was missing. A minimum of 6 data points of AOD in a month was showed to be sufficient to appropriately represent a monthly average [23].
The geocoding and exposure assignment was conducted in ArcMap (Version 10.2) [24]. Specifically, we merged the grid cells of modeled data over the study period with the boundaries of Chinese administrative divisions. Each grid cell had a spatial resolution of 1 km × 1 km, with individuals who resided in the same cell sharing the same exposure levels. The modeled exposures were recorded as monthly averages, and we calculated the average concentrations one to five years before the first physical examination (2013 to 2017). And the average concentrations was used as indicators of the historical (long-term) exposure. For sensitivity analysis, we obtained another source of modeled PM2.5 concentrations from the Global Burden of Disease database, which generated yearly average estimates by combining the satellite-based estimates, chemical transport model simulations and ground measurements [25].
Ethics
This study was approved by the Scientific and Ethical Committee of Zhejiang Hospital. All participants were informed of the purpose and method of the study and signed the informed consent.
Statistical analysis
Epidata 3.0 was used for data entry, and SAS 9.4 was used for data management and analysis. Frequency (percentage) description was used for counting data, and mean and standard deviation description was used for measuring data, such as socio-demographic characteristics including of age, BMI, waistline, laboratory test results including of SBP, DBP, HDL, TG, TC, LDL and cognitive function of the subjects. T test and chi-square test were used to compare the statistical differences for measuring data and measuring data, respectively. A common health association analysis method used with follow-up data in prospective cohort studies is cox proportional hazards regression model, the ratio of any two risk functions refers to the relative hazard (HR).
We built a unadjusted model and three adjusted models. Model 1 was an unadjusted model; Model 2 included PM2.5, age, gender; Model 3 added smoking and environmental tobacco smoke (ETS) exposure based on Model 2; Model 4 added educational degree, family income, BMI, and occupation before retirement based on Model 3.
To test the possible effect modification, we conducted several stratification analyses by age groups, sex, educational level, marital status, occupation and BMI, using adjusted models including variables above except stratification variables. All analyses were conducted by bilateral significance test, and the significance level of hypothesis test was set to P < 0.05.