This project focused on developing a risk prediction model for patients admitted to hospital with COVID-19 that updates as new clinical information (vital signs, lab tests etc) is collected during a hospital visit. While there are many COVID-19 prediction models that use only the information collected available at the point when a patient is admitted to hospital, we are not aware of reliable existing COVID-19 prediction models that are able to update their prediction as new clinical information is collected.
We used de-identified data extracted from the hospital medical records at Addenbrookes to develop our model. Addenbrookes is one of only six European hospitals to reach the highest digital maturity, meaning we have been able to use highly-granular data. We included demographic data, observations, vital signs, laboratory tests, comorbidities and information about interventions and treatments.
Using recent advances in statistical methodology for medical applications, we were able to develop a model founded on sound statistical underpinnings including: accounting properly for patients who leave hospital because they have recovered or require transfer to another hospital for more specialised treatment; and acknowledging that e.g. extra blood tests may be requested for sicker patients.
The resulting prediction model has the potential to provide physicians with an assessment of a patient’s evolving prognosis throughout the course of active hospital treatment.
Preprint currently under review: https://www.medrxiv.org/content/10.1101/2021.02.15.21251150v1
a Intensive Care Medicine and Acute Medicine, Addenbrooke’s Hospital, Cambridge
b Infectious Diseases and General Medicine, Department of Medicine, University of Cambridge
c Pathology, Cambridge University Hospitals
d Department of Medicine for the Elderly, Cambridge University Hospitals
e Clinical Informatics Integration Lead, Cancer Research UK, Cambridge University Hospitals
f Department of Infectious Diseases, Cambridge University Hospitals
Joint research with Sarah Cowana, Claire Waddingtonb, David Halsallc, Victoria Keevild, Vince Taylore, Effrossyni Gkrania-Klotsasf and Jacobus Prellera