Update on Mortality in Australia

The ABS has released the weekly mortality statistics up to 25 August in Australia (for doctor-certified deaths only, excluding coroner-referred deaths). This analysis is the fourth in a series of monthly updates on Australian mortality in 2020 (find previous versions here).[i]

In addition, the ABS released an analysis of excess mortality in Australia during 2020 which can be found here. They have reached very similar conclusions to our previous work, including that there are likely to have been further excess deaths due to the pandemic than were identified as deaths from COVID-19, particularly during late March and early April.

After a period where total deaths in Australia were much lower than predicted, deaths have increased in August to be within the expected range, albeit still a bit low.  This reflects a number of factors:

  • We continued to have very low deaths from all respiratory causes
  • The second wave in Victoria resulted in significant number of COVID-19 deaths
  • Deaths due to dementia and ‘other’ increased. In the absence of respiratory illness in the community and noting that the timing coincided with the Victorian second wave, we surmise that this may be either undiagnosed Covid19 and/or connected to difficulties in providing adequate care in aged care homes in Victoria at the time.

 

Across the year to 25 August, there have been almost 2,000 (2%) fewer doctor-certified deaths in Australia than predicted based on the previous five years’ deaths.

August Deaths

After around an eight-week period where deaths in Australia were much lower than prior years, deaths in August are higher and back to within our expected range. If we look at the five weeks to 25 August, the average deaths per week are 287 higher than the average of the previous 8 weeks. These higher deaths are almost entirely explained by males over 65 and females over 75, as shown in the table below.

While overall deaths have increased in August, this hides some quite material differences by cause of death.

Deaths in August due to respiratory disease (comprising influenza, pneumonia, chronic lower respiratory disease and other respiratory disease) continued to be much lower than predicted, due to continued lower levels of respiratory illness in the community.

There were a large number of deaths from COVID-19 as a result of the second wave in Victoria.

And deaths from dementia and other causes are materially higher than in the previous recent months.

In total, in the four weeks to 25 August, there were around 760 fewer deaths from respiratory disease than predicted based on previous years’ deaths.

The ABS has reported on doctor-certified COVID-19 deaths for the first time in this data release.  There were 424 such deaths in the four weeks to 25 August 2020.

For deaths caused by dementia and other causes (i.e. those not specifically reported on by the ABS), there is usually an increase in deaths over the winter months, likely brought on by respiratory illness.  In 2020, we did not see this seasonal impact in late-June/July – deaths were lower than predicted.  However, deaths in the last week of July and most of August have been back at more normal levels.  Given that respiratory illness continues to be low, and noting the simultaneous spike in COVID-19 cases in Victoria, we suspect that:

  • some of these deaths may be due to undiagnosed COVID-19; and
  • some may be the result of the provision of less than ideal levels of care in aged care homes that were dealing with an influx of COVID-19 cases.

 

For all other specified causes (cancer, heart disease, cerebrovascular disease and diabetes), where there are only minor seasonal impacts, deaths in August were close to predicted.

Deaths in the Year to 25 August

The following table summarises the deaths in the year to date compared with our prediction (based on deaths in 2015 to 2019).

Overall, there were around 2,000 (2%) fewer deaths in the year to 25 August than predicted based on the five-year history of deaths in Australia.

This is driven by lower numbers of deaths from respiratory illness.  Higher than expected deaths from this cause at the end of March (likely undiagnosed COVID-19) have been more than offset by the much lower numbers of deaths since mid-April.

Deaths from non-respiratory causes are also lower than predicted by around 600 deaths:

  • deaths from cancer and heart disease, two of Australia’s biggest killers, are a little lower than expected;
  • deaths from cerebrovascular disease (stroke, etc) are a little higher than predicted, driven by the experience in the first wave where we suspect that deaths from this cause may be due to either undiagnosed COVID-19 or people delaying treatment due to fears of visiting hospitals at that time;
  • deaths from dementia and ‘other’ are lower than predicted, driven by the lower than predicted deaths at the end of May and in June as a result of lower levels of respiratory disease; and
  • deaths from diabetes are higher than expected, driven by higher than expected deaths in April (likely undiagnosed COVID-19).

 

The lower than predicted deaths from non-respiratory causes are offset by deaths from diagnosed COVID-19.

Reported versus Certified COVID-19 deaths

The number of doctor-certified Covid19 deaths is higher than the number reported in the daily press.  The total reported in the press to 25 August was 525, whereas the doctor-certified deaths to that date totalled 608.  We believe that this difference is mostly due to timing, as there were 83 COVID-19 deaths reported on 31 August and 4 September as having occurred in July and August.  Some COVID-19 deaths reported in April and May are not included in the doctor-certified COVID-19 deaths for those months.  This suggests that the extra reported deaths may have been referred to the coroner and thus will not be included in the ABS doctor-certified deaths.  Some of the deaths in relation to the Ruby Princess may fall into this category.

Standardised Death Rates – providing confidence in our modelling approach

The monthly reports from the ABS compared 2020 deaths with those in prior years, without adjustment for changes in size and age structure of the population.  These changes mean that 2020 deaths would naturally be expected to exceed the 2015-19 average.  Our analysis, on the other hand, does allow for these demographic changes.

In its current release, the ABS has for the first time produced standardised death rates (SDRs) by year and cause of death.  These SDRs allow us to test the robustness of our demographic adjustments, noting that the SDRs will be more precise that our adjustments as the ABS has access to more granular data than what is published.

The residuals of our scaled deaths (before allowance for late reported deaths) are compared with the residuals of the SDRs in the graph below.

The residuals are almost identical, giving us confidence in our demographic adjustments.

Excess Mortality – Comparison to ABS analysis

Perhaps inspired by our lead, the ABS has now produced a paper that analyses the excess mortality in 2020.  They have done this using a detailed regression analysis that deliberately dampens the contribution of epidemics in order to highlight them when they arrive.  We have compared the results with those from our model.   

The ABS graph below compares actual and expected deaths, clearly showing the serious flu epidemic in 2017.  It also shows the impact of flu in other winters.

Looking at 2020, there is one week when the overall deaths were slightly higher than the expected range of deaths (the blue shaded area).


We’ve reproduced our model below, just showing the picture for 2020, with deaths outside the shaded area in late March/early April.  Our shaded area represents a narrower probability range (90% v 95%), we have a smaller standard deviation than the ABS, and we allow for late-reported deaths, but the picture is essentially the same.

This chart compares our model with the ABS model and actual deaths (adjusted for late reporting):

Although the two models are broadly similar, we can see that:

  • the ABS model is smoother, with less week-by-week variation than our model;
  • the ABS model has slightly lower deaths modelled at the beginning of 2020; and
  • the ABS model does not allow as much for seasonality in the winter months as our model (so our model has predicted more deaths in winter than the ABS model).

 

We also note that our model has been built from individual cause of death statistics, rather than starting from the total deaths in previous years (so our model will allow for trends in mortality by cause of death in a more robust way).

In addition, the ABS has looked at individual causes of death, with similar results – here is the ABS graph of pneumonia deaths during 2020.

Overall, the key conclusions are similar between the two analyses.

Excess Mortality – UK

The Office of National Statistics in England and Wales has published weekly deaths for most of 2020, as you can see in this chart from Our World in Data:

The UK, being in the northern hemisphere, has its flu season in winter, but this graph clearly shows the enormous impact of COVID-19 there.  Deaths in Australia from COVID-19 (whether identified or indirect) are well within an average year’s seasonal variation. The worst flu season in the past five years in the UK (which was 2015) is dwarfed by the extra deaths from COVID-19, even without adjusting for the population size and distribution.  The Continuous Mortality Investigation of the Institute and Faculty of Actuaries shows this another way after standardising for the population size and distribution:

In Australia, we have to look quite closely to see a COVID-19 effect. The success of our non-pharmaceutical measures for controlling COVID-19 (border closures, quarantine, lockdowns and other social distancing measures) is unmistakeable.

Lifeline is a national charity providing all Australians experiencing a personal crisis with access to 24 hour crisis support and suicide prevention services. Call 13 11 14 for confidential support.

[i] As with our previous analysis, we built a model predicting weekly deaths in 2020, based on the information provided by the ABS about weekly deaths during 2015-2019.  For each cause of death:

  • we start with weekly deaths from 2015-2019;
  • we scale those deaths for population (an 8% increase over five years), age distribution (a 4% increase over five years) and late-reported deaths; and
  • we fit a model to predict deaths in 2020, incorporating allowance for average mortality improvement over the last five years (a 3.5% decrease).

 

CPD: Actuaries Institute Members can claim two CPD points for every hour of reading articles on Actuaries Digital.