Early Prediction of Mortality in ICU Patients
Author Information
Author(s): Yeh Yu-Chang, Kuo Yu-Ting, Kuo Kuang-Cheng, Cheng Yi-Wei, Liu Ding-Shan, Lai Feipei, Kuo Lu-Cheng, Lee Tai-Ju, Chan Wing-Sum, Chiu Ching-Tang, Tsai Ming-Tao, Chao Anne, Chou Nai-Kuan, Yu Chong-Jen, Ku Shih-Chi
Primary Institution: National Taiwan University Hospital
Hypothesis
Can machine learning models predict ICU mortality upon admission?
Conclusion
The ADM models can provide crucial information on ICU mortality as early as upon ICU admission.
Supporting Evidence
- The ICU mortality rate was 7% in the testing dataset for the admission models.
- 38.4% of patients were discharged alive or dead within 1 day of ICU admission.
- The area under the receiver operating characteristic curve for the admission GBT model was 0.856.
- The performance was highest for the 24-hour GBT model with an AUROC of 0.910.
Takeaway
Doctors can use special computer models to guess if very sick patients in the ICU might not survive, right when they arrive.
Methodology
Data from two ICU databases were used to train and validate machine learning models for predicting ICU mortality.
Potential Biases
Potential bias due to missing data and lack of subgroup classification based on diagnosis.
Limitations
Missing vital signs and laboratory data prior to ICU admission may affect model performance.
Participant Demographics
Adult patients from two teaching hospitals, with exclusions for age and data errors.
Statistical Information
P-Value
<0.001
Confidence Interval
95% CI 0.845–0.867
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website