AbstractEmergency bowel surgery is a commonly performed procedure, with a high overall mortality rate compared to other types of major surgery. Pre-operative risk stratification has become a key part of its care pathway, as it facilitates the identification of high-risk patients. The aims of this thesis are to explore the care pathway for emergency bowel surgery, the risk models used to predict post-operative mortality, and to investigate the ability of variables that are routinely available pre-operatively to predict 30-day mortality after emergency bowel surgery.
First is a series of literature reviews which provide an overview of the three key topics. Second, a single centre retrospective cohort study performed using the local National Emergency Laparotomy Audit database (01/12/2013 to 31/01/2020). Further data were then extracted from electronic hospital records (n=1,508). Exploratory data analysis was conducted to investigate associations between candidate predictor variables and 30-day mortality.
The National Early Warning Score (NEWS), Laboratory Decision Tree Early Warning Score (LDTEWS), combined LDTEWS-NEWS risk score and Hospital Frailty Risk Score (HFRS), all displayed the strongest relationships with mortality in exploratory data analysis. They were then externally validated on the complete dataset, with the standalone scores being reasonable predictors of mortality (c-statistic 0.708, 0.724, 0.740 and 0.683 respectively), but poorly calibrated overall.
A logistic regression model (PRE-OP) was developed using age, NEWS, LDTEWS and HFRS as predictor variables. The PRE-OP model demonstrated good discrimination (c-statistic 0.827) and calibration on internal validation; but was outperformed by the NELA score (c-statistic 0.861). SORT, APACHE II and P-POSSUM all displayed inferior calibration and fair-to-good discrimination (c-statistic 0.808, 0.734 and 0.796 respectively).
In summary, patient vital signs, blood tests and markers of frailty can be used to accurately predict 30-day mortality after emergency bowel surgery and perform comparably to other risk models. The variables are reliably available pre-operatively and free from inter-observer bias. Further research is needed to externally validate these findings and investigate more detailed use of raw predictor variables with more complicated modelling techniques.
|Date of Award||5 Sep 2022|
|Supervisor||Jim Briggs (Supervisor), David Prytherch (Supervisor) & Rebecca Stores (Supervisor)|