The Taiwan National Health Insurance (NHI) program is known to offer coverage to 99% of Taiwan’s 25 million residents and to contains information regarding contracts with over 90% of the country’s national health care facilities (https://nhird.nhri.org.tw/en/index.html) [15,16]. The corresponding electronic database of this program, namely the NHIRD, comprises the claims data of insurants. Published studies have validated the high reliability of NHIRD diagnostic data [15,16]. The NHIRD includes detailed information, such as outpatient visits, hospital admissions, prescriptions, procedures, and disease diagnoses executed on the basis of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) . An exclusive personal identification number (PIN) is assigned to every individual in Taiwan. In the NHIRD, for patient privacy protection, data concerning patient identities are scrambled cryptographically [15,17]. The entirety of NHI data sets can also be cross-referenced with each individual’s PIN. This study utilized an NHIRD-derived data file, namely the Children file, that comprises information from half (chosen at random) of all insured children in Taiwan . The data set was determined to afford an adequate sample for pursuing the study objectives. This study was ratified by China Medical University Hospital’s Institute Review Board (CMUH104-REC2-115), and it complied with the principles outlined in the Declaration of Helsinki.
Study population, outcome of interest, endpoints, and confounding factors
This was a retrospective cohort study. From the Children file, we formed a child cohort by selecting individuals aged <18 (0-17) years on of January 1, 2000 (baseline). The study period was from January 1, 2000, to December 31, 2012. The follow-up period of each participant began from baseline until migraine/headaches, end of the follow-up, termination of insurance, or death. Individuals who were missing information such as their address, sex, and air pollution data and individuals that had ever been diagnosed with migraine/headaches before the baseline were excluded. Migraine/headaches were defined as≧3 times diagnoses of ICD-9-CM code 346 and/or 784.0 in any diagnosis field during any inpatient or ambulatory claim process during study period. By the end of the study period, some participants would be entering adulthood. The final study population contained 218,008 participants. Our mean (standard deviation) follow-up years for patients with migraine was 10.7 (2.6). Urbanization level of residence, age, number of consultations/visits with a physician per year, monthly income, sex, and allergic diseases constituted the confounding factors. On the basis of the method realized by Liu et al , we classified the study patients’ residential areas, encompassing 365 townships of Taiwan, into seven urbanization levels, with Levels 1 and 7 representing the “most urbanized” and “least urbanized,” respectively. We stratified the townships for defining urbanization levels by using several variables, including the following: population ratio of people with an educational level of college or higher, population density (people/km2), population ratio of agricultural workers, number of physicians per 100,000 people, and population ratio of elderly people aged older than 65 years . Because Levels 4-7 were determined to have low sample size, we combined these four levels into a single group (Level 4). Thus, we stratified the factor urbanization level into four levels, with Levels 1 and 4 representing the highest density and lowest density, respectively. We also classified monthly income into the following three groups: >NT$20,000, NT$15,000-19,999, and <NT$15,000.
Ambient air monitoring of monthly average data for SO2, NO2, THC, CH4, and PM2.5 are available for 74 Taiwan Environmental Protection Administration (EPA) monitoring stations on Taiwan’s main island since 1994. Concentrations of each pollutant are measured hourly—CO by nondispersive infrared absorption, NO2 by chemiluminescence, SO2 by ultraviolet fluorescence, THC and CH4 by flame ionization detector, and PM2.5 by beta-gauge—and are reported hourly.
We identified the map coordinates of the monitoring stations and air pollution sources. The ultraviolet fluorescence in these recording stations were automatically monitoring and recording readings of PM2.5, THC, CH4, SO2, and, NO2. The daily air pollution data were averaged based on these recording stations. Yearly average concentrations of pollutants were calculated from the baseline to the date of migraine and recurrent headaches occurrence, the withdrawal of patients, or the end of the study period, and the data were categorized into quartiles. The participants were assigned to residential districts based on the clinic where they most frequently sought treatment for acute upper respiratory infection (ICD-9-CM code 460). We divided the annual average air pollutant concentrations into quartiles: Q1, Q2, Q3, and Q4. We categorized annual average PM2.5 into Q1 (<11120 μg/m3), Q2 (11120-12652 μg/m3), Q3 (12652-15056 μg/m3), and Q4 (>15056 μg/m3); THC into Q1 (<835 ppm), Q2 (835-877 ppm), Q3 (877-949 ppm), and Q4 (>949 ppm); CH4 into Q1 (<735 ppm), Q2 (735-754 ppm), Q3 (754-770 ppm), and Q4 (>770 ppm) ; SO2 into Q1 (<1346 ppb), Q2 (1346-1914 ppb), Q3 (1914-2338 ppb), and Q4 (>2338 ppb); NO2 into Q1 (<7896 ppb), Q2 (7896-8894 ppb), Q3 (8894-10214 ppb), and Q4 (>10214 ppb). The air pollutant measurements from Taiwan EPA monitoring stations were integrated into monthly point data and interpolated to pollutant surfaces using inverse distance weighting (IDW). For the IDW approach, we used inverse squared distance (1/squared distance) weighted average of the three nearest monitors to compute monthly mean concentration. IDW predicts values of unknown points based on the similarity of two objects by its distance. When the unknown point to be estimated is closer to the known measuring point, the weighted value of the unknown point will be higher. We used the air pollution exposure in 2-year before and current year of diagnosis headaches to predict the monthly air pollution. And used IDW method to estimate the air pollution concentrations between the measured values of the air monitoring stations around the household registered by each patient according to the distance. Then explore the association between air pollutant and headaches (All data were managed by a geographic information system (ArcGIS version10.3; ESRI, Redlands, CA, USA)).
The sociodemographic factors in the current study included residential area urbanization level, sex, monthly income, age, and daily average exposure to air pollutants. To test the differences in daily average concentration distributions for each air pollutant by quartile and urbanization, we executed χ2 testing. Moreover, we calculated the incidence density rate of migraine/headaches (per 1000 person-years) according to each quartile of daily average concentrations for the five air pollutants. By employing Cox proportional hazard regression, we also derived estimates of the hazard ratios (HRs) as well as 95% confidence intervals (CIs) corresponding to migraine/headaches at the Q2-Q4 levels for air pollutant concentrations relative to the lowest level (Q1). To address the concern of constant proportionality, we examined the proportional hazard model assumption using a test of scaled Schoenfeld residuals. In the model evaluating the migraine and recurrent headaches risk throughout overall follow-up period, results of the test revealed a significant relationship between Schoenfeld residuals for PM2.5, THC, CH4, SO2, and NO2 and follow-up time (p-value<0.001, respectively), suggesting the proportionality assumption was violated. To deal with non-proportional hazards, we were used extended Cox models with time-dependent terms shows results. We adjusted the applied the multivariable model for allergic diseases, sex, number of consultations/visits with a physician per year, urbanization level, age, and monthly income. We also added the exposures as a continuous variable to estimate the risk of migraine and recurrent headaches as sensitivity testing. Further, we calculate the month average air concentration to estimate the month exposed air concentration for each patient by Inverse Distance Weighting Method (IDW methods). The IDW method is one of the most commonly used spatial interpolation methods in Geosciences, which calculates the prediction values of unknown points by weighting the average of the values of known points (This data were analyzed with ArcGIS version 10.3). We accessed the air pollution in 2-year before and current year the diagnosis of migraine and recurrent headaches and used IDW method to estimate the air pollution concentrations between the measured values of the air monitoring stations around the household registered by each patient according to the distance as sensitivity testing. The Statistical Package for the ArcGIS version 10.3 as well as SAS 9.3 (SAS Institute Inc, Cary, NC) constituted the platforms for all the executed analyses in this study. Additionally, for all executed statistical analyses, we deemed 2-tailed P values of <0.05 to indicate statistically significant tests.