David Muscatello presented to the NSW Branch to discuss the epidemiology of seasonal influenza in his talk “Estimating the unmeasurable: how do we see what influenza does to populations”, with a particular focus on the difficulty in estimating population-based measures.
David begins with an overview of influenza, a well-known virus of the orthomyxovirus family spread by the inhalation of infected respiratory droplets. There are 4 types of influenza (A, B, C and D), where type A and B are predominantly responsible for population outbreaks. Type A mutates quickly while type B is slower to evolve. New vaccines are released yearly to combat against mutations. In 2017, Australia experienced an influenza epidemic with more than triple the number of cases than in the previous year.
David highlighted there are several difficulties in estimating influenza in the population, including other causes may cause similar symptoms, confirmation relies upon specimen yet most infected with influenza do not require health care and lack of recognition of the role of influenza in hospitalizations or deaths. David was involved in a data-linkage study which linked the registry for virological notifications of influenza, deaths, hospital admissions and emergency department admissions in NSW. While the vast majority of influenza notifications had attended the hospital or emergency department, influenza diagnosis was only recognised in 7% of cases at the emergency department and 38% of cases in the hospital. In contrast, a flu tracking study conducted in England found that of those infected with influenza three-quarters were asymptomatic and less than 1% were hospitalized.
David then discusses how we can estimate influenza attributable deaths even when they are not captured in the medical certificate. He highlighted that there are several methods, where nearly all involve some kind of time series analysis that estimate the time varying fraction of deaths that occur when influenza is circulating. Typically, this relies on respiratory deaths but can specify influenza, pneumonia or non-specific deaths as total. The classical approach is to model excess mortality of influenza based on a Serfling (1963) model, which incorporates a sine wave of weekly mortality trend and influenza incidence to estimate the attributable mortality of influenza. David has applied methods similar to those used air pollution studies that replaces the sine wave with a smoothing spline using generalization additive models.
You can read more about David’s time series approach in his publication.
Dr Nicole De La Mata, Biostatistician, Sydney School of Public Health, The University of Sydney, ([email protected])
Dr Michael Stewart, Lecturer, School of Mathematics and Statistics, The University of Sydney, ([email protected])