Machine learning models can now predict the probability of in-hospital mortality by gathering information on patient age, sex and daily clinical state.
A recent study in the Journal of the American Medical Informatics Association revealed how a machine learning algorithm can use individual characteristics of patients to predict their illness trajectory. The model works by predicting the patient’s disease course in terms of clinical states like mild, moderate, severe or critical, as well as hospital utilization.
According to the researchers of the study, there is an urgent requirement for tools that can aid decision-makers plan allocation on the unit, hospital, and national level due to the unprecedented burden COVID-19 has placed on healthcare systems around the world.
The researchers found that hospitalized COVID-19 patients do not always transition between clinical states in a linear manner. For instance, a hospitalized patient might be in a moderate state for a week and then deteriorate to “severe” before suddenly recovering. Therefore, they worked on developing a multi-state model that can keep track of all these characteristics.
By tracking the day-to-day clinical state of each patient, along with information about their age and sex, researchers found that they were able to predict hospital occupancy as well and the likelihood of in-hospital mortality and critical illness.
Such multistate models can accurately predict healthcare utilization for a given patient arrival process by using straightforward and readily available patient characteristics. Such models can also be used to simulate utilization under different patient influx scenarios, which can then be utilized to plan resource allocation and the opening or closing of COVID-19 wards in a more effective manner.
The increasing strain on hospital resources caused by the pandemic has resulted in researchers turning to innovative and cutting-edge prediction models that have the ability to predict COVID-19 patient outcomes. Because of the unprecedented nature of the virus and the clinical uncertainty about COVID-19 patient trajectories, accurate predictions can significantly help improve clinical decision-making when the prediction is made.