Batchelor, John Stephen;
(2003)
Trauma scoring models using logistic regression.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
Trauma_scoring_models_using_lo.pdf Download (26MB) | Preview |
Abstract
Introduction. Trauma scoring models form an important part in trauma audit. Limitations of the TRISS model however has led investigators to search for new models which more accurately predict trauma deaths. Logistic regression has played an integral role in the development of such models in view of the fact that most investigators in this field use a dichotomous dependent variable (death/survival). Limitations in the methods used to evaluate new models currently makes it difficult to assess their true worth. Aims. The aim of this thesis was to evaluate current methods of model validation and also to evaluate alternative options. Methods and Results. A data set of 7069 complete trauma cases from the Los Angeles Trauma Registry was used for the study. Five goodness of fit tests were studied and were compared to the Hosmer Lemeshow test. The Copas test was found to be the most viable alternative to the Hosmer Lemeshow test. The two main methods of validating a prognostic model (i.e. data splitting and cross-validation) were evaluated using a series of simulation studies. Both methods were found to be unreliable with regard to their ability to reduce over-fitting of a prognostic model. Bootstrapping was also evaluated as a means of generating confidence intervals for the Hosmer Lemeshow test and the Copas test. Two models were used to assess the accuracy of the percentile method. Accurate bootstrap confidence intervals were developed for the Copas test but not for the Hosmer Lemeshow test. Conclusions. The Copas test with bootstrap confidence intervals provides a superior approach to validating a trauma model compared to the currently employed methods.
| Type: | Thesis (Doctoral) |
|---|---|
| Qualification: | Ph.D |
| Title: | Trauma scoring models using logistic regression. |
| Open access status: | An open access version is available from UCL Discovery |
| Language: | English |
| Additional information: | Thesis digitised by ProQuest. |
| URI: | https://discovery.ucl.ac.uk/id/eprint/10105109 |
Archive Staff Only
 |
View Item |
