Background:
OHCA patients use lots of hospital resources, and predicting prognosis is important in decision making for patient treatment. However, no algorithm can predict survival to hospital discharge rates in real-time.
Aim:
This study aimed to develop and validate the time adaptive model for real-time outcome prediction of OHCA patients.
Methods:
We performed a retrospective observational study using data from the Korea OHCA Registry in South Korea. In this study, we exclude patients with trauma, experienced ROSC before arriving in the ED, and patient who did not execute CPR in ED to select patients who executed CPR in ED. To develop the time adaptive prediction model, we organize training dataset as ongoing CPR patients by the minute. We used XGBoost as a machine-learning method and find the area under the receiver operating characteristic curve (AUROC) and predict the probability of the time adaptive prediction model.
Results:
The entire study population is 67270 and the majority were male patients (64%) with a median age of 70 years (IQR 23 years); 2632 (4.0%) had a shockable first documented rhythm at the ED. The subject was split into derivation and validation datasets at a ratio of 8 to 2. The AUROC of the model is 0.72 when the CPR starts, 0.68 after 30 minutes, and 0.62 after 60 minutes. Prediction probability of the time adaptive prediction model is shown in Fig. 1.
Conclusions:
We developed and validated the time adaptive prediction model by training ongoing CPR patients by minute to predict the CPR outcome of OHCA patients in real-time. This study showed the potential of a machine-learning-based algorithm model for decision making of patients about the termination of resuscitation.
Adaptive Prediction Model in Prospective Molecular Signature–Based Clinical Studies
