Forecasting Likelihood of COVID-19 Mortality Using Machine Learning
Predicted critical illness and mortality up to 10 days prior to onset using admission information.
Due to the rapid spread of the virus, there has been a sharp increase in the demand for medical resources required to support infected people. Despite valiant efforts by governments around the world, according to WHO as of today, there are around 53M COVID-19 cases and 1.3M deaths.
The pandemic has caused worldwide chaos in hospitals as they try to squeeze in more and more patients with limited staff, PPE (personal protective equipment), and bed allocations. Therefore, accurately identifying the likely course of a medical condition (prognosis) is essential to ease off the burden on healthcare systems.
COVID-19 patients display a wide range of symptoms from none at all to severe acute respiratory distress syndrome or death. This makes it difficult for providers to be certain of what treatment to prescribe for patients.
It would be hugely beneficial for physicians if they could identify characteristics of patients that indicate the severity of the course of disease across large patient cohorts. Particularly given its potential to aid physicians and hospitals in predicting disease trajectory, allocating essential resources effectively, and improving patient outcomes.
Previously, researchers have tried to develop such a model. However, due to shortage of samples and the lack of data regarding minority populations led to the inaccuracy of the models.
Researchers from Mount Sinai created a machine learning that can predict the possibility of critical illness or mortality in COVID-19 patients. In this study, researchers had access to large volumes of data from diverse patient populations. They analyzed electronic health records(EHRs) from >4,000 patients who were tested positive were admitted to hospitals in the Mount Sinai Health System.
The team looked into characteristics of the patients, such as their medical history, vital signs, and test results at admission, to predict critical events and mortality. The machine learning model was used to predict critical events or mortality at four different time (3, 5, 7, and 10 days) periods since the date of admission. The model displayed its highest accuracy in the 7-day mark when it accurately identified the most crucial events.
The strongest drivers at the 7-day mark
The strongest drivers in predicting critical illness:
- Acute kidney injury (AKI) /acute renal failure (ARF)
- Fast breathing
- High blood sugar
- High levels of lactate dehydrogenase (LDH) — Indicates damage to tissue
The strongest drivers in predicting mortality:
- Older age
- blood level imbalance
- C-reactive protein levels — A high level of CRP in the blood is a marker of inflammation
As COVID-19 continues to run rampant, researchers will refine their model further to provide hospitals and providers with better analytic tools to support them allocate resources for the patients.
These validated machine learning models successfully predicted critical illness and mortality up to 10 days prior to onset in a diverse patient population using merely admission information. The team believes that this model also identified important markers for acute care prognosis that can be used by health care institutions to improve care decisions at both the physician and hospital level for management of COVID-19–positive patients.
The results of the study show the potential for machine learning to forecast patient outcomes and perhaps, more importantly, the impressive progress the healthcare sector has made in developing such technologies that are specifically catering to the pandemic.
Want to learn more about how machine learning can improve healthcare? Subscribe to ‘Science for Real’ to receive regular articles straight to your inbox!
Jessica Kent (2020).Machine Learning Models Forecast Likelihood of COVID-19 Mortality. Health IT Analytics. https://healthitanalytics.com/news/machine-learning-models-forecast-likelihood-of-covid-19-mortality
An, C., Lim, H., Kim, D. W., Chang, J. H., Choi, Y. J., & Kim, S. W. (2020). Machine learning prediction for mortality of patients diagnosed with COVID-19: a nationwide Korean cohort study. Scientific reports, 10(1), 1–11.https://www.nature.com/articles/s41598-020-75767-2
Vaid, A., Somani, S., Russak, A. J., De Freitas, J. K., Chaudhry, F. F., Paranjpe, I & Zhao, S. (2020). Machine Learning to Predict Mortality and Critical Events in a Cohort of Patients With COVID-19 in New York City: Model Development and Validation. Journal of Medical Internet Research, 22(11), e24018. https://www.jmir.org/2020/11/e24018
Passionate about the interaction between humans, science, and technology to provide better care. 🧬 Medium: eshansamaranayake.medium.com ✍🏽 Newsletter: scienceforreal.substack.com 📰