Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1733
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-17T18:14:01.689219Z
status: published
lens: p_space
sector: health
headline: Targeting delirium in gastric surgery: An interpretable predictive model and statistical analysis in critical care
dek: BackgroundPostoperative delirium is a serious complication in elderly ICU patients after gastric surgery, increasing mortality and healthcare burden. Early identification of high-risk patients may support timely prevention and optimise perioperative care.ObjectiveTo develop and externally validate an interpretable machine learning model for predicting postoperative delirium in elderly ICU patients undergoing gastric surgery.MethodsThis retrospective multicohort study used three perioperative and critical care da…
gain_title: (none)
problem_title: Low outcome prevalence and class imbalance produced low precision and a substantial false-positive burden, so the model is not ready for clinical implementation.
trace_subject: (none)
gain_reading: (none)
gain_evidence: (none)
problem_reading: Low outcome prevalence and class imbalance produced low precision and a substantial false-positive burden, so the model is not ready for clinical implementation.
problem_evidence: marked class imbalance and low outcome prevalence were associated with low precision and a substantial false-positive burden
quick_read: Researchers built a machine learning model to predict postoperative delirium in elderly intensive care patients after gastric surgery. They developed it on MIMIC-IV data, chose logistic regression among six algorithms, locked it, and tested it on two independent databases, eICU and MOVER.

Delirium raises mortality and healthcare burden, so early identification could support timely prevention. However, the model's low precision and many false alarms mean its real-world value is unproven until prospective validation, threshold tuning, and utility assessment are completed.
limitation: The model showed low precision and a substantial false-positive burden due to class imbalance and low prevalence, and the retrospective design requires prospective validation and clinical utility assessment before implementation.
tag: Evidence-backed problem
key_points: The study was a retrospective multicohort analysis using three perioperative and critical care databases: MIMIC-IV, eICU and MOVER. | Cohort sizes were 4316 for training, 1851 for internal validation, 1745 for eICU testing and 1152 for MOVER testing. | LASSO regression selected features, and logistic regression was chosen over five other algorithms for discrimination, recall and interpretability. | SHAP analysis identified chloride, sodium, bicarbonate and anion gap as dominant predictors across cohorts. | Recall was 0.827 in eICU and 0.936 in MOVER, but precision was low.
rundown: The team compared six machine learning algorithms after LASSO feature selection. Logistic regression was picked using only the MIMIC-IV internal validation cohort, based on discrimination, recall and clinical interpretability, and was then evaluated unchanged in the external cohorts.

On external testing the model reached AUCs of 0.819 in eICU and 0.827 in MOVER. Recall was 0.827 and 0.936 respectively, indicating it caught most delirium cases.

SHAP analysis quantified predictor contributions and found that electrolyte and acid-base markers, specifically chloride, sodium, bicarbonate and anion gap, dominated across cohorts, which adds transparency to the model's outputs.

The authors report that class imbalance and low outcome prevalence led to low precision and many false positives. They state that prospective validation, threshold optimisation and assessment of clinical utility are needed before implementation.
sources:
- peer_reviewed | Technology and Health Care | https://doi.org/10.1177/09287329261495038 | 2026-10-08
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

a53c08d5af47d4e23bc4d68f81a989fa942872a74a54241b9ca2485c878bc4f4

previous

0000000000000000000000000000000000000000000000000000000000000000