Therapeutic monoclonal antibodies have been successful in protecting vulnerable populations against SARS-CoV-2. However, their effectiveness has been hampered by the emergence of new variants. To adapt the therapeutic landscape, health authorities have based their recommendations mostly on in vitro neutralization tests. However, these do not provide a reliable understanding of the changes in the dose-effect relationship and how they may translate into clinical efficacy. Taking the example of EvusheldTM (AZD7442), we aimed to investigate how in vivo data can provide critical quantitative results and project clinical effectiveness. We used the Golden Syrian hamster model to estimate 90 % effective concentrations (EC90) of AZD7442 in vivo against SARS-CoV-2 Omicron BA.1, BA.2 and BA.5 variants. While our in vivo results confirmed the partial loss of AZD7442 activity for BA.1 and BA.2, they showed a much greater loss of efficacy against BA.5 than that obtained in vitro. We analyzed in vivo EC90s in perspective with antibody levels measured in a cohort of immunocompromised patients who received 300 mg of AZD7442. We found that a substantial proportion of patients had serum levels of anti-SARS-CoV-2 spike protein IgG above the estimated in vivo EC90 for BA.1 and BA.2 (21 % and 92 % after 1 month, respectively), but not for BA.5. These findings suggest that AZD7442 is likely to retain clinical efficacy against BA.2 and BA.1, but not against BA.5. Overall, the present study illustrates the importance of complementing in vitro investigations by preclinical studies in animal models to help predict the efficacy of monoclonal antibodies in humans.