Statistical Aspects of Surgical Audit

Hayes, Catriona E (1995) Statistical Aspects of Surgical Audit. PhD thesis, University of Glasgow.

Abstract

The emphasis of this thesis is on "Comparative Audit", an area which has become increasingly popular in recent years. Firstly, the history of Surgical Audit is explored, from the first data collection to the publication of League Tables. It is evident that any comparisons of outcome measures must involve adjustments for patient case mix. This must be done using a statistical model. The features required of a model for this use are described, and a review of available predictive systems in medicine shows that none of the widely known predictive models satisfy all of the criteria which are appropriate for use in audit. An interest in Comparative Audit has led to the acquisition of several large data sets. The first of these is from the Royal College of Surgeons of England (RCS) Comparative Audit Service. The methods of presentation they used originally did not utilise the full potential of the data. The change to presentation of death rates as relative mortality confidence intervals rather than ranked bar charts allows consultants to see at a glance whether their rates are significantly different from the mean value. Adjusting these for some aspects of case mix makes a substantial difference to the rank order and, at the very least, highlights the folly of publishing League Tables of raw mortality rates. The adjustments as they stand are crude for three main reasons. The data are of poor quality, and they were never intended for this type of analysis. Also, data are collected as totals for consultants rather than for individual patients, which limits the potential for modelling. Through a case study, the problems of extracting data in the format required by the RCS are investigated, and based on this, together with consideration of more technical issues related to statistical modelling, suggestions are made as to how the RCS data collection exercise could be improved. The issue of modelling based on aggregate data is also explored, using both case studies and simulated data. These investigations show that the analyses are still of worth, as the adjustments have the desired effect if there are a reasonable number of consultants and patients, and if there is no, or only weak, correlation between the explanatory variables. To progress from the methods introduced for the RCS, it has been suggested that the POSSUM system be used for adjustment. This is currently the most widely used system, but it fails to meet several criteria for models for use in audit. The second data set explored is from a large audit study in Portsmouth, in which the POSSUM variables were collected. It emerges that a greatly reduced model of only four variables performs as well as the POSSUM model. Using this would not only greatly facilitate data collection, but would substantially reduce the amount of missing data, and thus diminish the difficulties of bias which arise from this. It may be that general surgery is too broad and heterogeneous to model, and that specific areas should be tackled separately. Analysis of colorectal cancer audit data shows that good models can be acquired, and used to compare surgeons. The relationship between patient volume and outcome, a related topical question, is also investigated with these data.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Adviser: Gordon Murray
• Keywords: Biostatistics, Medicine
• Date of Award: 1995
• Depositing User: Enlighten Team
• Unique ID: glathesis:1995-74937
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 27 Sep 2019 15:05
• Last Modified: 27 Sep 2019 15:05
• URI: https://theses.gla.ac.uk/id/eprint/74937