Australian data answers key questions about COVID-19 mortality

The risk of death after the diagnosis of a COVID-19 infection during Victoria’s 2020 outbreaks was 4 percent overall but was estimated to be 10 times higher among the elderly, a study from the University of Sydney’s NHMRC Clinical trials Centre has found.

The data analytics study, published in BMC Medical Research Methodology, is a comprehensive assessment of COVID-19 mortality prior to the introduction of vaccines in the state of Victoria, Australia, focusing particularly on the impact of age.

The findings demonstrate how modelling and data analytics can help answer key questions on COVID-19 infection in an Australian context. This can help guide global surveillance and the development of strategies to forecast and control the pandemic.

Professor Ian Marschner from the Faculty of Medicine and Health and NHMRC Clinical Trials Centre says the results, particularly the high mortality among the elderly, emphasise the importance of Australia’s vaccination program.

He performed the analysis with publicly available COVID-19 surveillance data from Victoria’s two waves of infection during 2020 (January to December). 

Data from Victoria were used as the Victorian health department publishes comprehensive data on age, and the Victorian outbreak comprised the majority of Australian cases in 2020. 

“Although it is well known that older age is a key risk factor for COVID-19 death, there are many related questions for which we have not had good answers,” he said.

“What is the mortality risk among young people and how much does it increase in older people? How long do people survive prior to death? Does the length of survival differ between people of different ages?

“These questions and others are key questions for understanding the natural progression of COVID-19 infection, and the answers impact construction of models to forecast healthcare needs and guide pandemic control strategies.”

The study analysed age-specific data on over 800 deaths and 20,000 diagnoses in Victoria. The data were in the public domain from the Department of Health and Human Services of the Victoria state government.

• Overall population-wide risk of death after the diagnosis of a COVID-19 infection was 4 percent

However, this overall mortality risk hides wide disparities in age groups:

COVID-19 mortality risk was 40 percent for the elderly (90+), which was 10 times the population-wide risk

• 32 percent COVID-19 mortality risk for people in their 80s
• 14 percent COVID-19 mortality risk for people in their 70s
• 3 percent COVID-19 mortality risk for people in their 60s
• Less than 1 percent COVID-19 mortality risk for people in their 50s or younger.

Data also suggest COVID-19 infections in Australia experience a longer survival time than had previously been assumed based on international data.

Key findings included:

• In fatal cases the average time to death was almost 3 weeks (18 days)
• In 10 percent of cases this survival time was almost 5 weeks (33 days)
• Despite the wide disparities in risk of death, there was no evidence that the time to death among fatal cases was affected by age.

A key innovation of the study was the use of data analytics algorithms that implement a technique called statistical deconvolution.

Previous studies of COVID-19 mortality have used smaller data sets which required a clear patient history, tracking both the onset and outcome of COVID-19 infection in an individual.

Deconvolution bypasses that limitation by taking the population’s daily case counts and determining what the mortality pattern must have been to produce the population’s daily death counts. The study was therefore able to use complete data on all deaths and diagnoses within the Victorian population, without the need to link onset and outcome at an individual level.

“It is therefore a powerful analytical tool that could be widely applied to extract substantial mortality information from existing global surveillance data,” said Professor Marschner

“The analytics algorithms effectively compared the daily pattern of case diagnoses with the subsequent daily pattern of deaths, to unravel population-level information on survival time and mortality risk. This allowed a more comprehensive analysis of mortality.”

Professor Marschner says the findings provide key information to help understand the natural progression of COVID-19 infection, and for calibrating mathematical models to the Australian context.

“Modelling assumptions need to be continually updated on the basis of new data. The analysis of Australian data presented here estimates a longer delay to death than has previously been assumed for the Australian context,” he says.

The study also underlines the need for age-specific analysis.

Models of COVID-19 spread in Australia have previously assumed the age-distribution of cases remains stable over time, whereas in the Victorian data there was strong evidence that it evolved over time, with the first wave having 2.7 percent of cases in people older than 80 years compared to 9.5 percent in the second wave.

These changes in the age distribution affect the overall mortality risk and make age-specific mortality analysis essential.

“Overall, the study highlights the importance of close collaboration between data scientists, statisticians and mathematical modellers in developing quantitative tools to inform pandemic control strategies,” says Professor Marschner.

Disclaimer: Professor Ian Marschner declares no competing interests.