Development of a Risk Prediction Model for In-Hospital Cardiac Arrests Using Continuous Electrocardiographic Telemetry Monitoring Data
- Author(s): Do, Duc
- Advisor(s): Qu, Zhilin
- et al.
Background: Many electrocardiographic (ECG) changes occur before in-hospital cardiac arrest (ICHA) due to pulseless electrical activity (PEA)/asystole. It is not known if these are unique to this period.
Objective: Evaluate whether ECG changes provide predictive information for IHCA from PEA/asystole.
Methods: We collected continuous ECG data from case and control patients. We selected three consecutive 3-hour blocks (block 3, 2, 1 in that order) preceding IHCA in cases and randomly in controls; only case block 1 immediately preceded IHCA. In each block, we measured dominant positive and negative trends in ECG parameters and determined the presence of arrhythmias. We also compared features between consecutive blocks. We created one random forest and two logistic regression models, and tested them on differentiating case vs. control patients (case block 1 vs. control block 1), and temporal relationship to ICHA (case block 1 vs. case block 2).
Results: We evaluated 77 cases (age 62.5ï¿½17.3, 57% male) and 1783 control patients (age 63.5ï¿½14.8, 67% male). We found many significant differences in ECG trends between case and control block 1, particularly in QRS duration, QTc, RR, and ST. New episodes of atrial fibrillation and bradyarrhythmias were more common before ICHA. The optimal model developed using only ECG changes was the random forest, achieving an AUC of 0.810, 57.9% sensitivity, 95.7% specificity at differentiating case vs. control, and AUC 0.942, 88.3% sensitivity, 90.1% specificity at differentiating case block 1 vs. block 2.
Conclusions: ECG parameters during the 3-hour window immediately preceding ICHA differ significantly from other time periods, and provides very good predictive information for IHCA.