18 Sep 2021
Statement on the Doherty Institute modelling sensitivity analysis
Following the release of the Technical Report and Addendum to National Cabinet on 10 August, the Doherty Institute-led modelling consortium was asked to provide “sensitivity analyses” of the scenarios represented in the Report.
Specifically, the consortium was asked to test the robustness of the recommendation to transition to Phases B and C of the National Plan at 70% and 80% vaccination coverage if COVID-19 infection was already established in the community. The original report compared outbreaks that were seeded with 30 cases at a time when “COVID-zero” was the goal. However, some states are now reporting significantly more than this number of infections each day. At these higher case numbers, it’s also recognised that test-trace-isolate-quarantine (TTIQ) responses are likely to be partially, rather than optimally, effective.
The modelling consortium considered three levels of case numbers at the time of transition to Phase B – low (tens, approximately 10-100), medium (hundreds approximately, 300-1000) and high (thousands, approximately 1,000-4,500) infections.
In the majority of cases the model’s conclusions were unchanged, with one important exception. If transitioning to Phase B at the 70% coverage threshold, seeding of thousands of infections in the context of partial effective TTIQ, with only baseline public health and social measures (PHSMs), resulted in earlier and larger outbreaks. While vaccine rollout continued throughout all the simulations, in the time window between 70% and 80% coverage, the epidemic was still growing from thousands to tens of thousands. As coverage increased beyond 80% coverage, the epidemic came under control. But starting from a point this high led to more cases overall.
Much less impact on the overall size of epidemics was observed when these seeding scenarios (tens, hundreds, thousands) were introduced from the 80% coverage timepoint. It is therefore recommended that case numbers continue to be strongly suppressed through Phase B until 80% coverage is reached.
In all scenarios, health outcomes were improved if a maximally effective TTIQ response (“optimal” TTIQ) were maintained. Similar impacts were achieved by combining partially effective TTIQ with a “low” level of PHSMs (above baseline restrictions). At high caseloads, additional “medium” level PHSMs may be needed while coverage increases from 70% to 80% to achieve target outcomes and protect the health system. The requirement for PHSMs above baseline will depend on how the epidemic is growing at a local or regional level and the anticipated impacts on the health system based on that epidemiological assessment.
These findings confirm our earlier strategic advice that even high levels of vaccination will not be sufficient to stop COVID-19 in its tracks. Maintaining at least partial (and ideally optimal) TTIQ responses will be essential for ongoing COVID-19 control. These measures will need to be supported by ongoing application of “low” PHSMs (above baseline) to keep case numbers down and achieve best health and economic outcomes for Australia.
Furthermore, as evidence continues to emerge regarding the Delta variant, it has become clear that our initial assumptions on transmissibility and vaccine effectiveness remain robust, however it also suggests Delta is more severe than was assumed in our modelling. This new evidence further endorses our recommendations for a multi-pronged approach to disease control to keep case numbers as low as possible so the health system is able to manage anticipated clinical loads. It therefore does not change our conclusions in any way.
This model is a high-level abstraction of the Australian context to guide strategic policy making. In reality, the national COVID-19 epidemic has been and will continue to be focal in nature, a “fire” fought on multiple fronts.
As stated in our original report, ongoing situational assessment of measured transmission potential and circulating SARS-CoV-2 variants in the Australian population over the coming months will allow benchmarking of these hypothetical scenarios to guide real time policy decision making about the transition to Phase B of the National Plan.
We provide such assessments to the Australian Health Protection Principal Committee every week, ensuring continuous updating of advice based on emerging local and international evidence to inform health responses.
Watch our press conference on the Doherty Institute modelling sensitivity analysis
Baseline PHSM – No stay-at-home orders, low density requirements, no retail restrictions, schools open.
Low PHSM – As per baseline PHSM, but with capacity limits on recreational activities, limits on retail group sizes and restrictions on workplace capacity.
Medium PHSM – Stay-at-home except for work, study and essential purposes, retail and cafes/restaurants open subject to density restrictions, working from home if possible with density restrictions in workplaces. Indoor recreational venues closed, small numbers of household visitors allowed. Closed or graduated return to schools.
High PHSM – As per medium PHSM but with no recreational gatherings, movement distance limits, no household visitors, two person rule for exercise and curfew.
Partial TTIQ – The observed reduction in transmission (43%) resulting from test-trace-isolate-quarantine Reponses at the height of the Victorian ‘second wave’ when case numbers were in the hundreds per day and the system was under strain resulting in delays.
Optimal TTIQ – The observed average reduction in transmission (54%) resulting from test-trace-isolate-quarantine responses implemented in NSW over a period of several months including the Christmas/New Year outbreak. Responses remained timely over variable case loads.
Seeding – The initial number of daily cases present in the population at a given vaccination threshold.
Low seeding – ranging from 10-100 cases per day
Medium seeding – ranging from 300-1000 cases per day
High seeding – ranging from 1,000-4,500 cases per day
Small area effects – The change in transmission caused by relatively small groups, such as geographic or cultural differences, that are not explicitly modelled in this work.
Source code for modelling
Source code for the analyses described in the reports, including simulation models, scripts for linking model components, plotting scripts and instructions for running the analyses are available at: https://github.com/aus-covid-modelling/NationalCabinetModelling
Please note that some of the analyses require access to epidemiological data provided through the National Notifiable Disease Surveillance System (NNDSS), which is not able to be shared under the terms of our data access agreement. Further information on the NNDSS, including how to request access to data from the Commonwealth, is available at: