Work

Incremental prognostic information from first day archived ST summaries for hospital mortality in a multicentre landmark cohort

Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Background Archived monitor measurements can be associated with outcomes without adding useful prognostic information. We evaluated the additional information from simple first-day ST temporal summaries and its sensitivity to channel availability. Methods This retrospective eICU study evaluated adults with sufficient first-day ST records who remained in ICU and hospital beyond a 24-hour landmark. One hospitalization per patient was selected without using outcomes. Hospital-blocked five-fold validation compared clinical and Acute Physiology and Chronic Health Evaluation (APACHE) IVa models with mean and operational burden, then slope and root mean square of successive differences (rMSSD). Paired intervals used 1000 hospital resamples of fixed out-of-fold predictions. Exploratory full-refit analyses used 100 hospital-frequency draws per family with original folds retained. A post hoc extension repeated comparisons in patients with a stable valid-channel set. Results Among 39020 patients in 127 hospitals, 3435 died. Temporal addition increased clinical area under the receiver operating characteristic curve (AUC) by 0.00034 (conditional 95% confidence interval − 0.00009 to 0.00081). Its APACHE increment was 0.00045 (− 0.00009 to 0.00095; n = 38646). Channel sets changed in 34.73% of patients. Among 25468 stable-channel patients with 2251 deaths, the clinical temporal increment was 0.00089 (0.00011 to 0.00160); the APACHE increment was 0.00066 (0.00015 to 0.00115; n = 25216). Temporal full-refit ranges spanned zero in both model families in both populations, and Brier-score improvement was not established. Joint clinical associations persisted in the original analysis. Conclusions Simple temporal ST summaries contributed small discrimination increments, with positive conditional intervals in the stable-channel sensitivity but uncertainty when training was repeated. The findings depend on population, comparator and uncertainty scope; they do not establish equivalence, clinical ischaemia or a validated ST alarm. Trial registration Not applicable to this retrospective noninterventional database analysis.
Language: English
Open Access status: green
OpenAlex ID: W7236678162
Cites: 19
Cited by:
Related to: 10