Electrocardiographic and electrolyte abnormalities in the early phase of admission predict in-hospital mortality: a MIMIC-IV-ECG cohort study
J W Park, S A Bae, K M Kim, J W Roh, O H Lee, Y C Kim, E Im, I H Jung, D K ChoAbstract
Introduction
Electrocardiography (ECG) reflects cardiac conditions, and electrolyte imbalances can promptly induce ECG abnormalities, potentially leading to adverse outcomes. We hypothesized that both the timing of ECG and electrolyte measurements and their recorded values may be associated with in-hospital mortality.
Method
Among 800,035 ECGs from 151,352 patients in the MIMIC-IV-ECG dataset and 431,231 admission records from 180,733 patients in the MIMIC-IV clinical dataset, a total of 65,084 ECG–electrolyte pairs met the inclusion criteria, which required machine-interpreted sinus rhythm. We categorized these pairs into two groups based on timing: those obtained within 24 hours of admission (ECG-E<24h) and those obtained more than 24 hours after admission (ECG-E>24h). We then compared logistic regression models and their predictive performance between the two groups.
Results
A total of 3,863 in-hospital deaths (6.38%) occurred in the study cohort. Among ECG–electrolyte pairs associated with in-hospital mortality, patients were older at admission (68 years [IQR 59–77] vs. 64 years [IQR 53–74], p<0.001), had a lower prevalence of PR prolongation ≥200 ms (9.1% vs. 10.5%, p=0.008), a higher prevalence of wide QRS ≥120 ms (14.8% vs. 13.2%, p=0.005), and more frequent QTc prolongation ≥500 ms (5.8% vs. 3.5%, p<0.001) compared with survivors. Potassium levels did not differ significantly between groups (p=0.214). Calcium (p<0.001) and sodium (p<0.001) levels were lower in the in-hospital death group, whereas magnesium levels were higher (p<0.001). In logistic regression analyses for ECG-E<24h, hypermagnesemia (OR 2.21, p<0.001), hyperkalemia (OR 1.21, p=0.001), hyponatremia (OR 0.95, p<0.001), hypocalcemia (OR 0.44, p<0.001), and QTc prolongation (OR 1.67, p=0.02) were associated with in-hospital mortality. In contrast, potassium (p=0.245) and sodium levels (p=0.272) were not significantly associated with mortality in ECG-E>24h. The logistic model using ECG-E<24h demonstrated superior predictive performance (AUC 0.711) compared with the ECG-E>24h model (AUC 0.641).
Conclusion
Early recognition of ECG and electrolyte abnormalities after admission can provide prognostic information for in-hospital mortality, particularly in patients with hypermagnesemia, hyperkalemia, hypocalcemia, or QTc prolongation.AUROC of risk modelsLogistric regression of risk model <24hr