Validating Time‐Series Regression Model Predictions of Influenza Mortality: An Interrupted Time‐Series Analysis
Jessie J. Goldsmith, David J. Price, Katherine B. Gibney, Patricia Therese Campbell, Sheena G. SullivanABSTRACT
Background
In Australia, non‐pharmaceutical interventions introduced early in 2020 to curtail the spread of COVID‐19 resulted in the temporary elimination of influenza. We used data from this influenza‐free period (2021) to validate influenza mortality burden predictions from time‐series models.
Methods
Time‐series models were parameterised with routinely collected health data from 2010 to 2019 (pre‐pandemic) and from 2010 to 2019 plus 2021 (pandemic‐inclusive). These models were used to predict influenza A‐ and influenza B‐associated mortality rates in Queensland between 2010 and 2019. Mortality estimates were made in persons < 65 and ≥ 65 years for pneumonia and influenza (P&I) and respiratory illnesses as contributory mortality causes and for all‐cause mortality.
Results
The models predicted consistent mortality rates for influenza B. However, the pandemic‐inclusive model for individuals ≥ 65 years predicted influenza A‐associated P&I mortality of 32.1 per 100,000 person‐years (95% prediction interval [PI]: 25.9, 39.6), significantly higher than the pre‐pandemic model estimate of 17.6 (95% PI: 13.4, 23.6). A similar, non‐significant difference occurred for respiratory mortality.
Conclusions
These results suggest that time‐series modelling may underestimate the true burden of influenza‐A mortality among the elderly. Further studies are needed to confirm this finding and establish the conditions under which the time‐series regression approach may be reliable.