Read on Twitter
A thread about epidemic modelling, prompted by calls to "release the modelling" and the MJA paper. #COVID19
Disclaimer: I'm an epidemiologist and ID physician. I have published the occasional mathematical model. I wouldn't dare call myself a modeller - this is a highly specialised area within ID epidemiology.
https://www.ncbi.nlm.nih.gov/pubmed/30309313 https://onlinelibrary.wiley.com/doi/abs/10.1111/trf.12525
https://www.ncbi.nlm.nih.gov/pubmed/30309313 https://onlinelibrary.wiley.com/doi/abs/10.1111/trf.12525
This is by no means a criticism of the very hard working teams around the world that have been working on this, including Jodie McVernon, @j_mccaw and @rob_mathbio in Australia https://www.theguardian.com/world/2020/mar/22/flatten-the-curve-why-predicting-coronavirus-infections-and-deaths-is-so-tricky
We are used to reading results of studies as a best estimate, and a range of uncertainty around that. This is not how this works in mathematical models of disease spread.
Mathematical models simply tell you what could happen if you assume particular inputs. But there are simply too many uncertainties with COVID19 to have a single likely outcome.
We don't really know what the R0 would be in Australia, or the proportion asymptomatic, or mixing patterns currently, or how well physical distancing and other public health interventions might work.
The proportion requiring hospitalisation and ICU and how long patients stay on ventilation? How this might be different in different age groups or risk groups? That's before the wildcards of seasonality, interaction with flu and other viruses, susceptibility factors (eg smoking)
For example, Wu & Leung's early paper detailed various potential scenarios. They had to use parameter values from MERS and SARS but others were estimated from the available Chinese data at that time https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)30260-9/fulltext
None have come to pass in China so far. Other major cities haven't seen big outbreaks. That's not to say that the model is wrong. It just says that if China hadn't taken the actions they had, things would probably be very different.
There is certainly still a role for modelling to examine different strategies - to ask the "what if" questions. If we cohorted patients in hospitals, how much less PPE might this save? How many contacts do we need to trace and how quickly?
But the assumptions behind these models need to reflect the important processes. In the @theMJA paper, the authors assumed exponential growth and concluded we will run out of ICU beds on 5 April.
While a more conventional approach might use a SIR model, exponential growth is a good approximation in early epidemics.
But this also assumes that there are no interventions to bend the curve. Importantly, it ignores that there are two drivers of cases in Australia - importation and local transmission.
Currently, most cases are returned travellers (although this will change.) The growth in cases to date is mainly reflecting the incidence and volume of travel from other countries. https://www.health.gov.au/news/australian-health-protection-principal-committee-ahppc-coronavirus-covid-19-statement-on-22-march-2020
It also ignores that the proportion requiring ICU is strongly dependent on age, and so far, most cases in Australia are in younger travellers that haven't required hospitalisation. (this too, will change)
In terms of possible scenarios, there's enough going on elsewhere in the world that predictive modelling isn't required. In terms of countries with comparable health systems to Australia, the possible realistic scenarios are Singapore/Japan, South Korea, Italy/US
"Releasing the modelling" risks us declaring a preordained outcome, which it most certainly isn't. The danger with precise numbers is that they provide an illusion of certainty.
"Real" modellers have been writing on these and other limitations recently - in the UK @AdamJKucharski https://twitter.com/AdamJKucharski/status/1243138088685252608?s=20
A common aphorism, attributed to George Box, is "all models are wrong, but some are useful"
