BACKGROUND: During the early phase of the COVID-19 pandemic in England, people with pre-existing conditions at severe clinical risk were advised to drastically reduce face-to-face contacts in a policy known as "shielding". The impact of shielding in preventing COVID-19 hospitalisations and deaths has not been evaluated nationally using transmission-dynamic modelling. METHODS: With the approval of NHS England, we present a retrospective cohort evaluation of the shielding policy, drawing data from electronic health records (EHRs) for 24 million patients in England accessed through the OpenSAFELY platform. The study is from 1 January-1 December 2020, prior to vaccination and SARS-CoV-2 variants. We used a dynamic model of SARS-CoV-2 transmission, infection, and hospitalisation, stratified by age and shielding status for the general population (excluding care homes). We estimated transmission rates in the shielding and non-shielding groups using data from the CoMix social contact survey and fitted the model to hospitalisations and deaths in and outside hospital. RESULTS: We found the risk of hospitalisation was higher for shielding people at all ages and increased with age. The hospitalisation fatality risk was similar between shielding and non-shielding people from January to June 2020 and greater in shielding people from July 2020 onward. By comparing the observed epidemic to a counterfactual scenario without shielding, we projected that between 7800 and 10,600 hospitalisations and 2300 to 3500 deaths due to COVID-19 were directly averted by the policy, corresponding to reductions of 25% (24, 28%) and 23% (21, 25%), respectively, in the shielding population in England up to 1 December 2020. Including also the indirect effect in the non-shielding population, we projected 14,700 - 21,800 hospitalisations and 3700-5500 deaths due to COVID-19 were averted by the policy in the total population, each corresponding to reductions of 13% (11, 16%). CONCLUSIONS: Based on our data and assumptions, we estimated the shielding policy reduced severe illness and mortality in clinically-extremely vulnerable shielding patients in England up to 1 December 2020, and, through indirectly-reduced exposure, also in the non-shielding population. Similar policies for other infections could have a comparable public health impact in reducing both mortality and pressure on health services.