Association Between Asthma Exacerbations and daily NOx Concentrations in East London: A time Series Regression Model

Problem

Based on surveys by Asthma UK, about two thirds of people with asthma believe that poor air quality makes their asthma worse and puts them at risk of an asthma attack. Although several policies and regulations have improved air quality in London, studies show that 34 of the 86 comparable sites still exceeded the annual legal limit for NOx (30 μgm-3) in 2019.

In this research, time series analysis is used to understand the relationship between daily NOx concentrations and the pattern of exacerbations in people with asthma. Because asthma-related A&E attendance information is not available, this study uses the number of oral corticosteroid courses prescribed in general practice for patients with asthma as a marker for asthma exacerbation.

Approach

- Dataset

NOx daily concentration measurements were extracted from all available monitoring stations in east London for two years from February 2018 to January 2020. Anonymised prescribing data from all registered patients (5-80 yrs) with asthma in 83 practices in the London boroughs of Tower Hamlets and Newham was used. The number of oral steroid courses and the daily prescribed dose of oral corticosteroids in mg were extracted for the study period for all patients registered at least one year prior to the initial search date (01/02/2018).

- De-seasonalization

Time series analysis of oral corticosteroid courses/daily prescribed corticosteroid dose in mg shows evidence of two seasonal trends in the data. The most likely causes being respiratory infection/flu in winter and hay fever in spring. Before introducing this time series into the regression model, these seasonal cycles were removed to provide a clearer view of non-seasonal variations in prescribed corticosteroids.

- Multivariate time series regression model

The dynamic relationship between total prescribed oral steroid tablets in milligram and NOx concentrations at monitoring stations, is described by a lagged regression model. The components of this model include NOx concentrations at each station, lag of these variables up to 21 days (3 weeks), a constant, and a random error term with normal distribution (this model is presented in the appendix with details).

Findings

The results of fitting the lagged regression to the prescribing dataset shows that NOx concentration at the background monitoring stations is statistically significant in predicting prescribed corticosteroids 8 days in the future. This means it takes just over a week to see the negative impacts of NOx concentrations on patients with asthma. As expected, same day air pollution observations are not statistically significant in this model.

Consequences

This time series regression model estimates changes in oral corticosteroid prescriptions based on the daily NOx concentrations. This could be used to develop an early warning system for patients/GPs and AEDs that predicts asthma exacerbations, based on air pollution measurements.