Abstract Early recognition of high-risk or deteriorating patients with COVID-19 allows timely treatment escalation and optimises allocation of scarce resources across overstretched healthcare systems. Since the late 1990s, physiological scoring systems have been used in hospital settings to provide an objective signal of clinical deterioration prompting urgent clinical review. Several early warning scores (EWS) accurately predict the need for intensive care unit admission and survival in hospitalised patients with sepsis and other acute illnesses, and their routine use is now recommended in secondary care settings in high and low income countries alike. However, there are widespread concerns that existing EWS, which place a premium on the cardiovascular instability seen in severe sepsis, may fail to identify the deteriorating COVID-19 patient. Dozens of research groups have now assessed the predictive value of existing EWS in hospitalised adults with COVID-19, and used sophisticated statistical methods to develop novel early warning and prognostic scores incorporating vital signs, laboratory tests and imaging results. However, many of these novel scores are at high risk of bias and few have been adopted in routine clinical practice. In this education and learning article, we will discuss key pitfalls of existing prognostic and EWS in hospitalised adults with COVID-19; outline promising novel scores for this patient group; and describe the ideal properties of scoring systems suitable for use in low and middle income settings.