Emily Jones' PhD thesis review seminar, "Novel application of logistic regression modelling to predict canine and feline urinary bladder disease diagnosis based on histological features"

Emily's advisors are A/Prof Rachel Allavena, A/Prof Chiara Palmieri, Dr John Alawneh, A/Prof Mary Thompson (Murdoch University) and Dr Karen Jackson

Background: There is increasing work on the utility of logistic regression in quantitative evaluation of disease, and predictive disease modelling. Logistic regression models have potential to predict disease risk in the human and veterinary medical fields, by providing a more objective probability of the disease occurring given the combination of variables. Here, we describe a dataset of archived canine and feline bladder tissues, and we explore the novel use of logistic regression modelling to apply a more quantitative approach to veterinary pathology.

Methods: Archived canine and feline bladder histology tissue from the University of Queensland School of Veterinary Science Veterinary Laboratory Services and the Murdoch University School of Veterinary and Life Sciences, as well as some prospective samples from local veterinary clinics, was included in this project. Detailed histological assessment was undertaken on all 338 samples obtained for this study. A novel application of logistic regression modelling was applied to describe in detail the UQ bladder pathology dataset, and to explore associations between microscopic features and bladder disease diagnosis. Following logistic regression analysis of the microscopic features, predictive probabilities were generated for each of the significant variables identified in the model, and these were then used to build a bladder tissue assessment tool which is currently being tested on four veterinary pathologists.

 Results: The histology dataset consisted of 338 cases (267 canine and 71 feline) including 102 cystitis, 84 neoplasia, 42 urocystolithiasis, 63 normal bladders and 47 other diagnoses. Logistic regression modelling revealed six significant variables associated with disease outcome: species, urothelial ulceration, urothelial inflammation, submucosal lymphoid aggregates, neutrophilic submucosal inflammation, and marked submucosal haemorrhage.

 Conclusion: This work has demonstrated the potential use of logistic regression modelling for descriptive analysis of pathology data, and for quantitively evaluating microscopic findings and developing predictive disease models.

Attending staff and students please submit your audience feedback here