The utility of the Liverpool Uveal Melanoma Prognostication Online ‘LUMPO’ as a prognostication algorithm in determining metastatic risk in uveal melanoma

Uveal melanoma (UM) is the most common primary intraocular malignancy in adults. Approximately 50% of patients with UM develop metastatic disease, usually occurring in the liver. Patient survival is directly related to the presence of hepatic metastases and how they affect liver function. After the identification of metastases, most patients die within a year, and the differing forms of existing treatment rarely extend life considerably. It has been proposed that prognostication can improve the quality of life, even when the probability of survival is reduced. Certain prognostic factors have been identified in UM that are associated with an increased risk of metastatic disease: these are clinical parameters as well as the histomorphological and genetic features of the primary UM. A team from Liverpool designed in 2011 a multiparameter algorithm called the “Liverpool Uveal Melanoma Prognostication Online” (LUMPO), to establish the prognosis for UM patients, stratifying them according to their relative risk of metastatic death. This was based on data collected over 20 years, and was validated externally by other ocular oncology centres. It was later updated to LUMPO3, and included new parameters and functions. The objective of this thesis was to perform a retrospective study, analysing the liver scan reports of patients with UM; examining whether LUMPO3 is able to predict the appearance of metastases in patients in Liverpool, and to determine the cost analysis of liver screening for the detection of metastases in patients with UM; and also to validate LUMPO3 externally. Following the Introductory chapter, the characteristics of the scan reports of 615 patients diagnosed with UM were analysed in Chapter 2. The data were collected over an eleven-year period (2008-2018). Data of the characteristics of liver scanning and the metastases detected were analysed, and later combined with the demographic, histological, genetic and patient outcome data. It was estimated that 37% of UM patients treated in Liverpool developed metastases at different stages of the disease. The analysis of the characteristics of the primary UM demonstrated that increasing tumour size and hence TNM staging category of the UM, was associated with an increasing risk of metastatic disease and reduced survival. Most metastatic UM were associated with monosomy 3 and gain of chromosome 8q. Many previous studies have identified different risk groups (low, intermediate and high) for UM patients to develop metastases: in this study, I identified 3 groups of patients that were divided according to when and whether they developed metastatic disease. Liver surveillance was most frequent in patients who ultimately developed metastases and the most frequent modality employed was magnetic resonance imaging (MRI). In chapter 3, the same cohort of 615 UM patients were analysed, here focussing on the estimated costs for all scans performed for which both the minimum and maximum costs were calculated using the NICE costing for clinical imaging. A feature of LUMPO3 called “linear predictor of death due to metastases” (lpmd) was extracted to predict the onset of liver metastases and to determine the hypothetical costs of liver surveillance in UM patients. Model performance was determined in terms of discrimination and calibration. Results showed consistent discrimination performances over a 5-year time period from date of initial treatment. Calibration performance was determined visually with good agreement between observed and predicted onset of UM metastases up to 10 years following initial treatment. The results suggested that unnecessary radiological examinations could be avoided in the UM patients with low metastatic risk, without any adverse effects to patient management. These results suggested that this lpmd could save costs by minimizing the number of examinations for metastases screening, and likely decrease the patient angst associated with liver screening. Chapter 4 study was a multicentre project that allowed the recruitment of a sufficient number of patients from 7 external collaborative centres, using the existing clinical network within the Ocular Oncology Group and the USA. Data were collected from patients diagnosed and / or treated for UM at these collaborating centres. The results showed that LUMPO3 is a reasonably accurate and valuable tool for predicting all-cause mortality in UM patients, despite the differences in the recording of clinical, histopathological and genetic data between the centres and hence the various cohorts studied. The calibration graphs presenting the expected probabilities of actuarial survival showed a good agreement between the observed and the predicted probabilities. In conclusion, I examined in detail “real world” data of UM patients treated in various ocular oncology centres that allowed the construction of the LUMPO3 model, and its validation in Liverpool. These data are valuable when considering the costs of surveillance programs, as well as potential modifications and revisions to the predictive algorithms for UM