Thread by @globalhlthtwit

Let me summarise a really important new #COVID19 modelling paper from Rosalyn Moran, Karl Friston and colleagues from Kings and UCL, London. Published in ArXiV. (1) fil.ion.ucl.ac.uk/spm/covid-19/M…

"We sought to determine the current size of the effective (i.e.susceptible) population for 7 European countries—to estimate immunity following this first wave. We compare these nos. to total pop. sizes of these countries, in order to investigate the potential fr future peaks."(2)

"The (initial) susceptible size of a population (termed ‘S0’) is assumed to be the size of a particular city, e.g. 10 million in Wuhan (Prem, Liu et al. 2020) or—for a country—is assumed to comprise of multiples of smaller city sized outbreaks." (3)

They applied "two models to data from seven countries: including Ireland, the United Kingdom, Italy, Spain, France, Germany and Switzerland." The (published) ODE model is a 12-compartmental model—with one factor with 12 states). (4)

Their model (Markov Decision Process, MDP, or hidden Markov model, HMM) is a 256-compartment model—with four factors (location, infection and clinical) each with four levels, giving 44 = 256 states. (5)

"Taking a Bayesian average—across all models and
countries—the estimated proportion of people that were initially susceptible at the start of this
outbreak—and thus immune at the end of the outbreak—was 6.4% of the total population of each
country." (6)

"We plot this fall in susceptibility state S (increase in immunity) over time,from the initial size S0 for the ODE and HMM models separately for Ireland and the UK." (7)

"The projections under HMM (Moran/Friston's HMM/MDP model) are more pessimistic because the effective population size at the start of the first outbreak (i.e.susceptible population) is smaller (and more precise)." (8)

"Values suggest tht after the current wave of cases, between 3 (lowest estimates for Ireland/UK) + 12 (highest estimate for Germany) more cycles (with identical dynamics to those from Jan 22nd) would be required to bring the total population to probable herd immunity (60%) (9)

From this Figure their HMM model suggests that 8-12 more cycles of infectionin the UK might be needed to reach herd immunity. (10)

DEATHS. "For the UK the ODE and HMM were remarkably consistent, predicting a cumulative death toll of 49296 and 49785, respectively (from this wave)." (11)

IMMUNITY: "after the current wave, a large proportion of the total population will remain without immunity... in the absence of strong seasonal effects, new medications or more comprehensive contact tracing, further epidemic waves in different geographic centres are likely." (12)

"How do these estimates relate to previous predictions of
Covid-19 deaths in the UK? It was predicted (Ferguson, Laydon et al. 2020) that without interventions
510,000 deaths could occur, in the UK due to Covid-19." (13)

Their "analysis predicted, that even with an optimal mitigation scenario, death rates would reduce only by one half i.e. to 255,000...." (14)

"Thus, the predicted death cases of our models ~50,000 in the current cycle are in line with the predictions of mitigation effects, if we assume that several more cycles are possible." (15)

MY COMMENTS. These estimates assume all infected individuals will be 'immune'...but we don't yet know the nature, strength or duration of immunity. There are worries about reactivation of the virus in infected individuals in Korea and China. (16)

"These findings have implications for ‘exit strategies’ from any lockdown stage." Indeed they do. Potential options. 1) A cycle of lockdown/lift/lockdown. But LD/Lift/LD will severely damage the economy +cause many 'indirect' deaths thru poverty, unemployment and recession. (17)

2) A herd immunity policy could take 8-10 more cycles of infection with 250-500,000 deaths in the UK. This would be politically unacceptable, and overload the NHS causing more indirect deaths. (18)

3) Build a community shield through mass testing doing millions of tests per week. Paul Romer in the USA proposes 22 million tests per week for their pop of 330 million, == 4.5m in UK. Not feasible for now. (19)

4) Digital monitoring of the whole population by mobiles monitoring all case population movements, social distancing, contacts and quarantine. Major civil rights and privacy concerns. Not feasible now. (20)

5) Less intrusive policy of >community testing + case finding, w mobile monitoring of symptoms/quarantine by public health +volunteers contact tracing in less affected cities/rural areas (+partial LD if necessary) will suppress infection while we wait for a VACCINE. Feasible (21)

6) We have 115 candidate vaccines with 79 being studied, and 5 already in human volunteers. Usually <10% of vaccine candidates get to market. The challenge is scale-up. Flu vaccine grows in chicken eggs, but #COVID19 does not. (22)

Please do comment if I have made mistakes or misinterpreted this model. Or if you believe this paper makes errors. (When I retweet model findings it does not imply endorsement) (23)