Use this URL to cite or link to this record in EThOS:
Title: Prediction of perioperative mortality after oesophagectomy using the Northern Oesophagogastric Cancer Unit clinical database
Author: Warnell, Ian
ISNI:       0000 0004 2739 6666
Awarding Body: University of Newcastle Upon Tyne
Current Institution: University of Newcastle upon Tyne
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Access from Institution:
Perioperative mortality after thoracoabdominal oesophagectomy for cancer is about 4%. Stratifying this risk may assist patients to make treatment choices, facilitate comparative audit, and enhance research. I aimed to explore prediction modelling of this risk, using the Northern Oesophagogastric Cancer Unit (NOGCU) database. The first section is a systematic review of prediction models and candidate predictors from ‘high surgical volume’ centres. Three models were externally validated but overestimated higher risk mortality; discrimination was moderate. Two groups used prediction models to reduce mortality in practise but there were no clinical impact studies. Candidate predictor definitions and associations with mortality were varied. Age predicts mortality and should be included as a continuous predictor in any model. Risk of bias in primary studies was poorly reported. In section two, I explored the risk of perioperative mortality using logistic regression on 1575 records from the NOGCU database, from 1991 to 2009. Comorbidity fields required extensive cleaning and recoding, and there were variable amounts of missing data, which caused spurious associations. I compared a prespecified model containing age, operation, albumen and cardiorespiratory comorbidity with a statistical stepwise elimination model and used split-sample validation. Age, gender, operation, white cell count, cardiac risk index, operation and weight loss were associated with mortality but only age, gender, operation and weight loss were significant in multivariate analysis. Discrimination was moderate, at best, for all models and the prediction range was only to a maximum 20%. The best calibrated models contained age, operation and gender, and originated from the most complete datasets. These models are not suitable for individual risk prediction but could be developed as risk adjusters for provider profiling and research. The sample sizes and high quality data required for further development are most likely to be achieved in larger scale studies, data syntheses or clinical databases.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Thesis (M.D.)) Qualification Level: Doctoral
EThOS ID:  DOI: Not available