If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
We hypothesized that unexpected deterioration among high-risk patients could be predicted using basic clinical information added to NEWS parameters using a machine learning method.
We conducted a multicenter-based, retrospective, consecutive cohort study at five hospitals associated with Hallym Medical Center. The study period was from February 2020 to December 2020, including three COVID-19 pandemic periods in Korea (February, August, and December). High-risk patients were defined as patients with a National Early Warning Score (NEWS) over 7. Outcomes included unexpected ICU admission, which was defined as transfer to the ICU within 24 h after being classified as high-risk, and in-hospital mortality.
The random forest method was used to predict outcomes with 500 trees using the ‘randomForest’ package of R, and the accuracy of the developed model was compared with the NEWS. During development, age, sex, body mass index, comorbidities, and unscored NEWS parameters were included. Training and test sets were assigned at a 7:3 ratio. Comorbidities consisted of cardiovascular, pulmonary, gastrointestinal, neurological, genitourinary disease, and cancer based on the 10th International Classification of Disease code. All statistical analyses were performed using R 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria).
A total of 6576 NEWS was recorded from 2118 patients. The median (interquartile range) of age, body mass index, and NEWS were 73 (64–82) years, 21 (18–24), and 8 (7–9), respectively. The proportion of male, cardiovascular, pulmonary, gastrointestinal, neurological, genitourinary disease, and cancer patients was 60.7%, 4.6%, 47.7%, 9.1%, 2.6%, 4.4%, 28.8%, respectively. The random forest model showed a good predictive ability for ICU admission (area under the receiver operating curve [AUROC] 0.787, 95% confidence interval [CI] 0.748–0.825) and excellent predictive ability for in-hospital mortality (AUROC 0.881 95% CI 0.866–0.895, which were higher than that of NEWS (p < 0.001) (Fig. 1).
Our study found that a random forest-based model had a good predictive ability for unexpected ICU admission and in-hospital mortality in high-risk patients. Decision-making for prioritizing intensive care among acute illness patients would be facilitated by our model. Also, our model used immediately measurable parameters of hospitalized patients. The algorithm was also intuitive and straightforward. Therefore, we believe that medical institutions with limited resources can prioritize intensive care using this method.
Ethics and patient consent
This study was conducted following the Helsinki declaration and was approved by the Institutional Review Board (IRB) of Hallym University Chuncheon Sacred Heart Hospital (IRB number: 2021-02-007). The need for written informed consent was waived because this was a retrospective cohort study.
Conceptualization, Formal analysis and Writing-original draft: S.H.K., Review and editing: H.C., H.L. and J.Y.H, Supervision and validation: Y.K. All authors read and approved the final manuscript.
The data of our study can be made fully available when the manuscript is accepted for publication.
Conflicts of interests
The authors do not have any competing interests to declare.
Reduced health-care utilization among people with chronic medical conditions during coronavirus disease.