Predictors of In-Hospital Mortality in Severe Abdominal Trauma

Background and rationale Trauma is among the leading causes of death and disability worldwide and the foremost cause of mortality in the first four decades of life. The abdomen is involved in a substantial proportion of severe injuries, and abdominal trauma contributes disproportionately to early haemorrhagic death and to late mortality from sepsis and multi-organ dysfunction. Outcomes are determined by the interplay of injury anatomy, physiological reserve, the timeliness and adequacy of resuscitation, and the surgical and critical-care response. Existing instruments that quantify this risk - the anatomical Injury Severity Score (ISS), the physiological Revised Trauma Score (RTS), the combined TRISS, and simpler bedside tools such as MGAP - were predominantly derived and calibrated in high-income trauma systems, and their transferability to high-volume centres in resource-constrained settings is uncertain. Reliance on imported risk estimates may misclassify risk, distort benchmarking, and provide limited guidance for triage, activation of massive transfusion, and the choice between definitive repair and damage-control surgery.

Knowledge gap Despite an extensive trauma-scoring literature, there are few prospective single-centre data from high-volume tertiary trauma units in low- and middle-income settings that (i) focus specifically on severe abdominal trauma rather than polytrauma in aggregate, (ii) capture early dynamic physiological and biochemical markers (lactate, base deficit, coagulopathy indices) alongside anatomical grading, and (iii) directly compare locally derived predictors against established scores using contemporary prediction-model methodology with internal validation.

Objectives Primary objective: to identify independent predictors of in-hospital all-cause mortality among patients presenting with severe abdominal trauma at a tertiary trauma centre. Secondary objectives: to describe the demographic, mechanistic, physiological, biochemical and anatomical-injury profile of the cohort and compare survivors with non-survivors; to quantify the discrimination (AUROC) and calibration of a multivariable model built from routinely available admission and early in-hospital variables; to compare the derived model with ISS, RTS, TRISS and MGAP; to describe secondary outcomes (24-hour mortality, intensive-care mortality, massive transfusion, damage-control surgery, major complications and length of stay) and their association with candidate predictors; and, where statistically supported, to derive a simplified, internally validated bedside risk score.

Hypothesis A defined set of admission and early in-hospital variables - in particular age, depth of physiological derangement (systolic blood pressure, shock index, Glasgow Coma Scale), metabolic markers of shock (lactate, base deficit), coagulopathy, transfusion intensity and anatomical injury burden - are independently associated with in-hospital mortality, and a model combining these variables achieves discrimination at least comparable to established trauma scores in this population.

Design and setting Single-centre, prospective observational cohort study conducted in the Emergency Trauma Unit and Department of Surgery of Minia University Hospitals, Minia, Egypt - a tertiary referral centre with 24-hour emergency surgical, interventional-radiology, blood-bank and intensive-care services, receiving both primary presentations and referrals from secondary facilities. Consecutive eligible patients are enrolled at admission and followed until hospital discharge or in-hospital death. No study-mandated intervention is applied. Reporting follows the STROBE statement, and the prediction-model component additionally follows TRIPOD.

Variables and data sources Candidate predictors were selected a priori on the basis of biological plausibility, established association with trauma mortality, and routine availability at admission or during the first hours of care. They span six domains: demographic (age, sex, relevant comorbidity); mechanism and prehospital (blunt versus penetrating and sub-type, time from injury to arrival, direct versus referred arrival); arrival physiology (systolic blood pressure, heart rate, shock index, Glasgow Coma Scale, respiratory rate, oxygen saturation, temperature); admission biochemistry and haematology (arterial lactate, base deficit, pH, haemoglobin, platelet count, INR); anatomical injury (AIS-abdomen, ISS, number of injured intra-abdominal organs, organ-specific AAST grade, major vascular injury, associated extra-abdominal injury, eFAST and estimated haemoperitoneum); and resuscitation and operative management (time from arrival to operation, massive transfusion, 24-hour packed-red-cell units, damage-control surgery, vasopressor requirement). Data are collected prospectively on a standardised case-report form and cross-checked against the emergency record, operative notes, laboratory and imaging systems, blood-bank record and intensive-care chart, with daily verification by a dedicated coordinator. AAST grading is assigned by the operating surgeon and verified against imaging by a second reviewer, and ISS, RTS, TRISS and MGAP are computed by an investigator blinded to eventual outcome wherever feasible.

Sample size The sample size is driven by the requirements of multivariable prediction modelling rather than a single hypothesis test. Institutional volume is approximately 250 patients with severe abdominal trauma annually, so 12-month enrolment is expected to yield close to 430 evaluable patients. Anticipating in-hospital mortality of approximately 20%, 430 patients generate an expected 86 deaths; applying at least ten events per variable supports a model containing up to eight independent predictors, and the final model will be constrained to eight or fewer variables. Using contemporary sample-size criteria for clinical prediction models (anticipated C-statistic approximately 0.80, outcome prevalence 0.20, up to 12 candidate parameters considered), the minimum required sample is below 400; the planned 430 provides a margin of roughly 10% for incomplete data.

Statistical analysis Continuous variables will be summarised as mean with standard deviation or median with interquartile range according to distribution (Shapiro-Wilk), and categorical variables as frequencies and percentages, overall and stratified by the primary outcome. Survivors and non-survivors will be compared using the independent-samples t-test or Mann-Whitney U test and the chi-square or Fisher exact test. Variables associated with mortality at p < 0.10, together with predictors judged clinically essential a priori, will enter a multivariable binary logistic regression model with in-hospital mortality as the dependent variable. Linearity of the logit will be checked and collinearity screened using variance-inflation factors. Adjusted odds ratios with 95% confidence intervals will be reported, with two-sided significance at p < 0.05.

Model performance, validation and score derivation Discrimination will be quantified by the area under the receiver-operating-characteristic curve and calibration by the Hosmer-Lemeshow test and a calibration plot. Internal validation will use bootstrap resampling with 1,000 replicates to derive optimism-corrected estimates. The derived model will be compared with ISS, RTS, TRISS and MGAP using the DeLong test for correlated AUROCs. If the model is robust and well calibrated, a simplified integer risk score will be derived by assigning points proportional to the regression coefficients, internally validated by the same bootstrap procedure, and reported with risk strata and corresponding predicted mortality.

Missing data, sensitivity and subgroup analyses The pattern and extent of missing data will be reported. Where a variable has less than 20% missing values assumed missing at random, multiple imputation by chained equations with pooling by Rubin's rules will be used, and a complete-case analysis reported as a sensitivity analysis. Further sensitivity analyses will examine 24-hour and intensive-care mortality as alternative endpoints and will repeat the primary analysis after excluding early deaths occurring within one hour. Pre-specified subgroup analyses will examine injury mechanism (blunt versus penetrating), age group (60 years and over versus under 60) and the presence of major vascular injury, assessed by interaction terms and interpreted with caution given reduced power. Analyses will be performed using SPSS version 27 or later and R version 4.3 or later (pROC, rms and mice packages).

Oversight and quality assurance As a non-interventional observational study that imposes no change to clinical care, a formal data and safety monitoring board is not required. A study oversight group comprising the principal investigator, a senior trauma surgeon and the study statistician will meet at least every six months to review recruitment, completeness and quality, and to approve protocol amendments before submission to the ethics committee. All investigators and data collectors complete structured training before enrolment begins. At least 10% of completed case-report forms will be independently source-verified by a team member not involved in the original entry, and inter-rater agreement for AIS and AAST grading will be quantified by Cohen's or weighted kappa on a random subset. Protocol deviations will be recorded in a deviation log.

Anticipated limitations As a single-centre study, findings may have limited generalisability to centres with different case mix, prehospital systems or resources, and external validation will be required before clinical adoption of any derived score. A prospective observational design cannot establish causation; identified predictors are markers of risk rather than necessarily modifiable causes of death. Some early deaths and prehospital variables may be incompletely captured. Internal validation by bootstrapping reduces but does not eliminate optimism. Referral patterns may introduce selection effects, as the most severely injured patients may die before reaching the centre.