Data: sources and preprocessing

This study uses ozone and health outcome data from the EPA Air Quality Data Mart and Texas Department of State and Health Services (Texas DSHS) for the years 2007–2016. Health data from DSHS was obtained from the statewide emergency room visit (ERV) public use data file. This study required an Institutional Review Board (IRB) approval, which was obtained through the University of North Texas Cayuse IRB system (approval IRB 18-477) (Department of State Health Services of Texas (DSHS), 2018). Pre-processing of the health data was necessary and is presented in Fig. 4. The DFW study region ERV data was processed and sorted at the zip code administrative level and organized in a database. Zip code level data is suppressed from the state in multiple ways and was removed from the dataset. State suppressed data includes fewer than 30 discharges by ED, alcohol and drug use, HIV diagnosis, hospital has fewer than 50 discharges or if a hospital has fewer than five discharges of a particular gender, including unknown. Thresholds for data suppression reflect numbers per quarter.

ICD-9 and ICD-10 main diagnosis codes were first clustered into 17 major groups representing all ER diagnoses (Table 1). Then, the data was refined to focus on Category 8, diseases of the respiratory system (Table 2). This category contains seven ICD-9 and ICD-10 codes: Asthma, Bronchiectasis, Bronchitis, COPD, Emphysema, External Agents and Other. The records were grouped by zip code and the total number of patient diagnosis were tallied per major diagnosis code.

Some health data zip codes were suppressed (white areas) while others were added due to access to available sensor data for a more complete ozone surface.

A limitation of the health data is that diagnosis codes were tallied instead of patients. Respiratory ERV incidence rates may be overestimated in cases where patients exhibited multiple diagnosis and were counted more than once. However, due to inconsistent practices in how I believe its ICD 9/10 codes are reported, all data points including those patients with multiple ERV codes were retained in the analysis. Further, the proportion of records with multiple diagnosis codes was small.

Our analysis was divided into three major components—the first component models ozone exposure values and ERV for respiratory outcomes across the DFW metroplex, the second component identifies spatial patterns of ozone exposures and ERV for respiratory health, and the third component uses concentration response curves to examine associations between ozone exposure and respiratory ERV.