This article is a Preprint
Preprints are preliminary research reports that have not been certified by peer review. They should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Preprints posted online allow authors to receive rapid feedback and the entire scientific community can appraise the work for themselves and respond appropriately. Those comments are posted alongside the preprints for anyone to read them and serve as a post publication assessment.
A Deep Recurrent Reinforced Learning model to compare the efficacy of targeted local vs. national measures on the spread of COVID-19 in the UK
Preprint
in En
| PREPRINT-MEDRXIV
| ID: ppmedrxiv-20248630
ABSTRACT
ObjectivesWe have developed a deep learning model that provides predictions of the COVID-19 related number of cases and mortality in the upcoming 5 weeks and simulates the effect of policy changes targeting COVID-19 spread. MethodsWe developed a Deep Recurrent Reinforced Learning (DRRL) based model. The data used to train the DRRL model was based on various available datasets that have the potential to influence the trend in the number of COVID-19 cases and mortality. Analyses were performed based on the simulation of policy changes targeting COVID-19 spread, and the geographical representation of these effects. ResultsModel predictions of the number of cases and mortality of COVID-19 in the upcoming 5 weeks closely matched the actual values. Local lockdown with social distancing (LD_SD) was found to be ineffective compared to national lockdown. The ranking of effectiveness of supplementary measures for LD_SD were found to be consistent across national hotspots and local areas. Measure effectiveness were ranked from most effective to least effective 1) full lockdown; 2) LD_SD with international travel -50%; 3) LD_SD with 100% quarantine; 4) LD_SD with closing school -50%; 5) LD_SD with closing pubs -50%. There were negligible differences observed between LD_SD, LD_SD with -50% food & Accommodation and LD_SD with -50% Retail. ConclusionsThe second national lockdown should be followed by measures which are more effective than LD_SD alone. Our model suggests the importance of restrictions on international travel and travel quarantines, thus suggesting that follow-up policies should consist of the combination of LD_SD and a reduction in the number of open airports within close proximity of the hotspot regions. Stricter measures should be placed in terms travel quarantine to increase the impact of this measure. It is also recommended that restrictions should be placed on the number of schools and pubs open. O_TEXTBOXStrengths and limitations of this study - The proposed Deep Recurrent Reinforced Learning (DRRL)-based model takes into account of both relationships of variables across local authorities and across time, using ideas from reinforcement learning to improve predictions. - Whilst, predicting the geographical trend in COVID-19 cases based on the simulation of different measures in the UK at both the national and local levels in the UK has proved challenging, this study has provided a methodology by which useful predictions and simulations can be obtained. - The Office for National Statistics only released data on UK international travel up to March 2019 at the time of this study, and therefore this study used the amount of UK tourists in Spain as a reference variable for understanding the effect of international travel on COVID-19 spread. C_TEXTBOX
cc_by_nc_nd
Full text:
1
Collection:
09-preprints
Database:
PREPRINT-MEDRXIV
Type of study:
Cohort_studies
/
Prognostic_studies
Language:
En
Year:
2021
Document type:
Preprint