An interpretable raw physiological–anatomical risk-stratification framework for 28-day mortality in ICU-admitted patients with major trauma: a retrospective cohort study

Early risk assessment in trauma must account for both physiological derangement and anatomical injury. Existing trauma scores are clinically useful, but they share component variables; entering several of them into one model can count the same information twice. We therefore built an interpretable framework that separates raw physiological abnormalities from anatomical injury burden and describes their combined mortality-risk strata. We conducted a retrospective cohort study of all 300 adult ICU records available in the curated dataset whose source records carried a local clinical polytrauma label. Per-region AIS severity was not recorded, so the Berlin definition could not be applied, and the cohort is described as major trauma. The primary outcome was 28-day all-cause mortality. A raw physiological derangement factor was derived from GCS, systolic blood pressure, respiratory rate and pulse after prespecified two-sided abnormality transformations. AIS-ISS represented anatomical injury burden. Age, the physiological factor and standardized AIS-ISS were entered into an additive logistic model, and median-defined anatomical–physiological strata were used descriptively. We assessed discrimination, calibration, Brier score and decision-curve performance with out-of-fold predictions. Standardization, PCA and model fitting were repeated within each training fold during 10 repetitions of stratified 10-fold cross-validation. Among 300 patients, 105 (35.0%) died within 28 days. The four descriptive strata had mortality rates of 4.4% (5/113), 32.4% (12/37), 28.6% (10/35) and 67.8% (78/115), respectively. The age-adjusted odds ratios relative to the low/low group were 10.03 (95% CI 2.98–33.69), 9.95 (2.84–34.85) and 77.63 (25.81–233.45). In the frozen repeated cross-validation analysis, the integrated model had a pooled out-of-fold AUC of 0.908, a Brier score of 0.113, a calibration intercept of − 0.008 and a calibration slope of 0.932. In a supplementary 277-patient blunt-injury comparison, AUCs were 0.932 for the integrated model, 0.901 for historical TRISS and 0.898 for recalibrated TRISS; this internal comparison does not establish external superiority. In this ICU-admitted major trauma cohort, an interpretable raw physiological–anatomical framework identified four descriptive risk strata with different 28-day mortality rates. The framework is a prognostic description, not a treatment or triage rule, and requires external validation before any clinical use.

Authors

Institutions

Publication Details

Journal
BMC Emergency Medicine
Published
2026-11-21
DOI
https://doi.org/10.1186/s12873-026-01847-7
Primary Topic
Trauma and Emergency Care Studies
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An interpretable raw physiological–anatomical risk-stratification framework for 28-day mortality in ICU-admitted patients with major trauma: a retrospective cohort study

Shizhen Zou, Zhenzhen Qin, Feng Zheng, Yiqing Jia et al.
BMC Emergency Medicine
Trauma and Emergency Care Studies
article

An interpretable raw physiological–anatomical risk-stratification framework for 28-day mortality in ICU-admitted patients with major trauma: a retrospective cohort study

Shizhen Zou, Zhenzhen Qin, Feng Zheng, Yiqing Jia, Yan Xu
article en

Abstract

Early risk assessment in trauma must account for both physiological derangement and anatomical injury. Existing trauma scores are clinically useful, but they share component variables; entering several of them into one model can count the same information twice. We therefore built an interpretable framework that separates raw physiological abnormalities from anatomical injury burden and describes their combined mortality-risk strata. We conducted a retrospective cohort study of all 300 adult ICU records available in the curated dataset whose source records carried a local clinical polytrauma label. Per-region AIS severity was not recorded, so the Berlin definition could not be applied, and the cohort is described as major trauma. The primary outcome was 28-day all-cause mortality. A raw physiological derangement factor was derived from GCS, systolic blood pressure, respiratory rate and pulse after prespecified two-sided abnormality transformations. AIS-ISS represented anatomical injury burden. Age, the physiological factor and standardized AIS-ISS were entered into an additive logistic model, and median-defined anatomical–physiological strata were used descriptively. We assessed discrimination, calibration, Brier score and decision-curve performance with out-of-fold predictions. Standardization, PCA and model fitting were repeated within each training fold during 10 repetitions of stratified 10-fold cross-validation. Among 300 patients, 105 (35.0%) died within 28 days. The four descriptive strata had mortality rates of 4.4% (5/113), 32.4% (12/37), 28.6% (10/35) and 67.8% (78/115), respectively. The age-adjusted odds ratios relative to the low/low group were 10.03 (95% CI 2.98–33.69), 9.95 (2.84–34.85) and 77.63 (25.81–233.45). In the frozen repeated cross-validation analysis, the integrated model had a pooled out-of-fold AUC of 0.908, a Brier score of 0.113, a calibration intercept of − 0.008 and a calibration slope of 0.932. In a supplementary 277-patient blunt-injury comparison, AUCs were 0.932 for the integrated model, 0.901 for historical TRISS and 0.898 for recalibrated TRISS; this internal comparison does not establish external superiority. In this ICU-admitted major trauma cohort, an interpretable raw physiological–anatomical framework identified four descriptive risk strata with different 28-day mortality rates. The framework is a prognostic description, not a treatment or triage rule, and requires external validation before any clinical use.

BMC Emergency Medicine
Chinese PLA General Hospital (CN)
Trauma and Emergency Care Studies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon