On the Relevance of ECG Features for Survival Prediction. Application to Septic Shock

Raphael aditya Chalard1, Amar Kachenoura2, Guy Carrault2, GE Di3, Antoine Kimmoun4, Alexandre Mebazaa5, Ahmad KARFOUL6
1Univ Rennes, Inserm, LTSI - UMR 1099, Rennes, F-35000, France, 2Univ Rennes, Inserm, LTSI - UMR 1099, 3LTSI, 4CHRU de Nancy, 5Inserm UMR-S 942 MASCOT, University of Paris Cité, Lariboisière Hospital, Paris, France, 6Université de Rennes


Abstract

Aims: In Intensive Care Units (ICUs), survival predictions essentially based on clinical data, while ElectroCardioGram (ECG) data, which are sys-tematically recorded upon patient admission, are often overlooked. This study aims to demonstrate that features derived from ECG can provide val-uable insights into patient prognosis.

Methods: We included 468 patients admitted ICUs for septic shock, using data from the French and euRopean Outcome reGistry in Intensive Care Units (FROG-ICU) study. In addition to comprehensive clinical information (e.g., medical history), which can be difficult to collect, 12-lead ECG features were collected. Here, only features extracted from the first ECG recorded during the ICU stay were included. To explore the prognostic value of ECG-derived features, patients were grouped into two clusters using the K-means algorithm, first based on ECG features alone, and then on non-ECG clinical variables. Survival analysis with the log-rank test was used to discriminate clusters.

Results: Based on ECG-derived features, a 12% difference (p-value = 0.0028) in survival at ICU discharge and 20% (p-value < 1e-4) one year later were observed between the two clusters. Compared to the high-risk group which have survival probabilities of 67% at ICU discharge and 44% after one year, the low-risk group shows 80% and 62%, respectively. For non-ECG clinical variables, a 12% difference (p-value = 0.0025) in survival at ICU discharge and 26% (p-value < 1e-4) at one year were observed between the two clusters. The high-risk group had survival probabilities of 70% at ICU discharge and 42% one year later, compared to 82% and 68% for the lower-risk group.

Conclusion: With only the first ECG recorded at ICU admission, high-risk patients can be identified as accurately as with clinical variables that are harder to obtain. Thus, the ECG offers great potential as an initial tool for assessing prognosis.