Related Experiment Video
Updated: Jun 7, 2026
A Low Mortality Rat Model to Assess Delayed Cerebral Vasospasm After Experimental Subarachnoid Hemorrhage
Published on: January 17, 2013
A predictive model of delayed pseudoaneurysm formation in paediatric patients with isolated blunt splenic injury
Haruka Taira1, Hideto Yasuda1,2,3,4, Morihiro Katsura5,6
1Department of Emergency and Critical Care Medicine Jichi Medical University Saitama Medical Center Saitama Japan.
Aim:
To develop and evaluate a predictive model for delayed pseudoaneurysm formation after non-operative management (NOM) in children with blunt splenic injuries.
Methods:
A post hoc analysis of a multicenter cohort study in Japan included patients aged ≤16 years who underwent NOM for isolated blunt splenic injuries. The outcome was the formation of a pseudoaneurysm, which was not identified on admission and confirmed at least 24 h after admission. Predictors were determined from data available within 24 h of hospital arrival. Five predictive models were developed using logistic regression analysis and evaluated using discrimination (receiver operating characteristic [ROC] and precision-recall curve [PRC]), calibration (calibration plot and Brier score) and decision curve analysis (DCA) with bootstrap resampling data.
Results:
Pseudoaneurysms developed in 41 (9.4%) of 434 cases of isolated splenic injury in our cohort. Model 1 (19 predictors) had the highest ROC (0.828) and PRC (0.358), followed by model 5 (8 predictors; ROC 0.805, PRC 0.295). Calibration was similar across models, indicating good calibration. Models 1 and 5 outperformed the other DCAs. Overall, model 5, incorporating factors such as age, sex, Injury Severity Score, American Association for the Surgery of Trauma-Organ Injury Scale, contrast extravasation on computed tomography, concomitant injuries, cryoprecipitate dose and NOM details, was simpler and showed better predictive ability than the other models.
Conclusion:
A predictive model for delayed pseudoaneurysm formation was developed with moderate discrimination and calibration. Further improvement using different modelling methods, such as machine learning, may be necessary.
