tl;dr — A stochastic agent-based model of the SARS-CoV-2 epidemic in France*
This blog is a summary of the paper from Hoertel, N. et al.,[1] which was published in Nature Medicine[2] on 14th July, 2020.
The focus of the paper is on understanding the effects of lockdown extension by 8–12 weeks and post lock down NPIs (physical distancing, mask wearing etc) and particularly its impacts on — the cases, mortality and ICU requirements.
The Key takeaways
- Duration of the lockdown alone will not lead to a reduction in covid-19 fatality, overwhelming of ICUs etc → just delays the inevitable!
- Making use of physical distancing and further adherence of masks, the epidemic was showing 33% and 47% reduction in mortality respectively. This is when compared to the scenario when no measures were implemented post lockdown.
- Furthermore, shielding at risk people for (16 weeks or 32 weeks) showed a further decrease in the mortality rate → 62% more than with just masks+physical distancing post lock down (shielding for 32 weeks) and 80% more than no measures after lockdown.
- All these numbers are based on 100% adherence to the mentioned measure. If only 50% of the at-risk population is shielded for 16/32 weeks, the effects aren’t this profound.
- The results are robust as a variation of +/- 20% of the individual model parameter wasn’t causing much change to the numbers.
Limitations/caveats
- Does not capture the changing trends of seasonality associated with the virus.
- Data is based on what is available and would have missed out asymptomatic people who might be a large reason for the spread.
- Post-infection immunity is assumed to be long lasting.
- Adherence to the different measures enforced would have a heterogeneous nature (depends in the population, etc)
Model/Features
- Uses n = 194 features — individual/population level characteristics (n = 140, family structure, age structure etc), societal characteristic (n = 33, school size, colleague pool, shopping density etc), virus characteristics (21, ICU rates, asymptomatic population etc)
- 2 variables — contamination risk and undiagnosed cases were estimated through model calibration — i.e., the values are used to best replicate the observed epidemic
- Was run for 360 days, 500k individuals and all results are average across the 200 simulations run.
- The Github repo provided by the authors is linked here — [Code]
I have tried to extract the key points of the paper and present it in a succint form. All credits of the original papers belongs to the orginal authors.
Please let me know of any feedback about the blog. Furthermore, I strongly encourage reading the paper, it’s a very fascinating read!
References
[1] Hoertel, N., Blachier, M., Blanco, C. et al. A stochastic agent-based model of the SARS-CoV-2 epidemic in France. Nat Med 26, 1417–1421 (2020). https://doi.org/10.1038/s41591-020-1001-6