QSpace at Queen's University >
Theses, Dissertations & Graduate Projects >
Queen's Theses & Dissertations >
Please use this identifier to cite or link to this item:
|Title: ||An Exploration of Life Expectancy Calculation Methods to Aid in Prostate Cancer Screening and Treatment Decision-Making|
|Authors: ||WYKES, Wykes, Dylan|
|Issue Date: ||2011|
|Series/Report no.: ||Canadian theses|
|Abstract: ||Background: Life expectancy (LE) estimation is an important part of both screening and treatment decision-making for potentially curable prostate cancer. Clinicians’ estimation of patient life expectancy is typically made using population-based life tables and intuition and it is often inaccurate. This study explores methods to improve LE prediction by formally considering patient co-morbid illness status, in addition to age, in the development of a LE prediction tool.
Methods: We conducted a population-based retrospective cohort study of patients from the Ontario Cancer Registry who were curative treatment candidates, identified between 1990-1998. We analyzed data on three sub-populations of this cohort, and we used LE estimates from the Ontario Life Tables. Each model utilized Cox proportional hazards analysis, and/or the declining exponential approximation of LE, to estimate the survival experience of potential curative treatment candidates, including the impact due to both age and co-morbid illness status. We developed five separate models, tested them using a random subset of the cohort study sample, and compared their predictive accuracy by measuring both discriminative ability and calibration to determine the ‘best’ model. We also conducted a supplementary analysis using logistic regression to develop a model to predict the probability of 10-year survival.
Results: The ‘best’ of our models demonstrated a c-index of 0.65 and very good calibration. Further analysis revealed that our ‘best’ model violated the Cox PH assumption for age and it’s predictions consistently over-estimated observed LE. Supplementary analysis of the logistic regression prediction model demonstrated a c-index of 0.70.
Conclusions: Our exploration of methods to predict LE resulted in modest predictive accuracy. However, based on the results of the logistic regression model, we conclude that the results of our LE prediction models are reasonable, and obtaining a high level of predictive accuracy may not be possible given just age and co-morbidities as predictors. Further studies should continue to explore these and other methods for LE prediction. External validation of the ‘best’ model from the current study is required before the model and its accompanying LE reference tables can be recommended for use in a clinical setting for screening or treatment decision-making.|
|Description: ||Thesis (Master, Community Health & Epidemiology) -- Queen's University, 2011-04-07 19:11:34.216|
|Appears in Collections:||Queen's Theses & Dissertations|
Community Health & Epidemiology Graduate Theses
Items in QSpace are protected by copyright, with all rights reserved, unless otherwise indicated.