Mathematics

# Maharashtra’s missing COVID-19 deaths: technical details

Below are plots from simulations used to estimate the scale of missing COVID-19 fatalities in Maharashtra. The agent-based modelling framework is described at https://maths.mdx.ac.uk/research/modelling-the-covid-19-pandemic/, with code at https://github.com/muradbanaji/COVIDAGENT. Parameter values are given in the Appendix.

In every case the recorded data is taken from https://www.covid19india.org/. The key quantity which varies between simulations is the “case-to-fatality reporting delay” or “C-F delay” for short discussed previously. This is the difference between the time_to_death and testdate parameters. As this is increased, we need to lower case detection in order to match the early cases and fatalities data. Other parameters are tweaked to get a better visual match between the modelled and measured data. All matching is currently by eye – no optimisation is carried out, although this is intended in the future.

Each estimate is based on 10 simulations at the given parameter values. However, only point estimates are given. A more proper treatment would involve putting uncerainties on many of the parameters in the model.

Appendix: parameter values for the simulations

Parameter values are given for the simulations in the order in which they appear above

number_of_runs 10
death_rate 0.5
geometric 1
R0 3.1
totdays 150
inf_start 3
inf_end 10
time_to_death 16, 16, 18, 18, 16
dist_on_death 6
time_to_recovery 20
dist_on_recovery 6
initial_infections 10
percentage_quarantined 6.0
percentage_tested 84, 60, 42, 30, 84
testdate 14, 12, 12, 10, 14
dist_on_testdate 6
herd 1
population 40000000
physical_distancing 0
pd_at_test N/A
pdeff1 N/A
haslockdown 1
lockdownlen 150
infectible_proportion 0.11, 0.1, 0.11, 0.12, 0.22
lockdown_at_test 110, 110, 110, 120, 110
pdeff_lockdown 53.5, 50, 49, 46, 52
popleak 50000, 50000, 140000, 150000, 60000
popleak_start_day 10, 10, 20, 20, 10
sync_at_test 1000
sync_at_time 24