Seasonal asthma in Melbourne, Australia, and some observations on the occurrence of thunderstorm asthma and its predictability

Abstract

We examine the seasonality of asthma-related hospital admissions in Melbourne, Australia, in particular the contribution and predictability of episodic thunderstorm asthma. Using a time-series ecological approach based on asthma admissions to Melbourne metropolitan hospitals, we identified seasonal peaks in asthma admissions that were centred in late February, June and mid-November. These peaks were most likely due to the return to school, winter viral infections and seasonal allergies, respectively. We performed non-linear statistical regression to predict daily admission rates as functions of the seasonal cycle, weather conditions, reported thunderstorms, pollen counts and air quality. Important predictor variables were the seasonal cycle and mean relative humidity in the preceding two weeks, with higher humidity associated with higher asthma admissions. Although various attempts were made to model asthma admissions, none of the models explained substantially more variation above that associated with the annual cycle. We also identified a list of high asthma admissions days (HAADs). Most HAADs fell in the late-February return-to-school peak and the November allergy peak, with the latter containing the greatest number of daily admissions. Many HAADs in the spring allergy peak may represent episodes of thunderstorm asthma, as they were associated with rainfall, thunderstorms, high ambient grass pollen levels and high humidity, a finding that suggests thunderstorm asthma is a recurrent phenomenon in Melbourne that occurs roughly once per five years. The rarity of thunderstorm asthma events makes prediction challenging, underscoring the importance of maintaining high standards of asthma management, both for patients and health professionals, especially during late spring and early summer.

Fig 1. seasonal cycle in population-normalised asthma-related hospital admissions.

The seasonal cycle in population-normalised asthma-related hospital admissions, normalised by the population size in each age-gender category. The individual panels show the cycle for the full population (A), children and teenagers (B), working-aged adults (C) and retiree-aged adults (D). The dashed lines show the effect plus or minus one standard error of the fitted cyclical cubic spline. Note the different scales on the y-axis in the four panels.

Fig 2. Proportion of admissions by age and gender over the year and for HAADs.

The proportion of admissions by gender and five-year age-group (stacked bar-chart) and the mean age of admitted individuals per gender (triangles), shown for HAADs in February and November and each month of the year. The left y-axis applies to the stacked bar-charts, while right y-axis applies to the points. The x-axis labels show the set of days and the gender of the individuals. To the left of the vertical black-grey dashed line is the same representation of the age-gender distribution for HAADs only or all days. Vertical grey dashed lines separate data for different months.

Fig 3. Modelled and observed population-normalised asthma admission rates (above), and deseasonalized equivalents (below).

Upper row: Time-series of observed population-normalised asthma-related hospital admissions (black) and the corresponding predicted values from the model (red). Lower row: the same time-series minus the seasonal mean. Left column: a time-series of these data is shown for a shorter period (2010-2015) to highlight the seasonality. Right column: a scatter plot of modelled versus observed values for the full data series (2000-2015), shown as a two-dimensional density plot (with outliers given as points); these panels also show some summary statistics and the least-squares linear model fit.