Background: The relative treatment effectiveness of surgery versus radiotherapy for early-stage prostate cancer is uncertain and randomized clinical trials are unlikely to be performed. This study describes the difference in cause-specific survival between patients treated with radiotherapy versus surgery, using a number of design and analytic steps to mitigate confounding by indication within an observational study. Methods: We conducted a population-based case-cohort study, sampling patients from the Ontario Cancer Registry who were treated or were candidates for cure by radiotherapy or surgery. Cases were those who died of prostate cancer within 10 years. Cause-specific survival was analyzed using Cox-proportional hazard regression, with variance adjustment for the case-cohort sampling. Analysis using intent to treat was compared to that using treatment received. Propensity scores were also calculated and Cox-proportional hazard regression was conducted within each propensity score quintile. We formed instrumental variable groups based on radiotherapy rates in Cancer Care Ontario Regions (CCORs) using the study population sampling frame and checked the instrumental variable assumption of equal distribution of covariates by comparing those covariates across these groups using data from the subcohort. Results: The adjusted hazard ratios for risk of prostate cancer death for radiotherapy compared to surgery were 1.44 (95% CI 0.86-2.40) and 1.84 (1.06-3.17) using intent to treat and treatment received respectively. Stratified hazard ratios comparing radiotherapy to surgery for death from prostate cancer from the lowest propensity quintile to the highest propensity quintile were 0.30 (0.04-2.28), 1.54 (0.35-6.77), 0.90 (0.29-2.82), 2.71 (1.01-7.31) and 1.08 (0.41-2.81). Differences among these hazard ratios were not statistically significant (p=0.13). The distributions of all prognostic indicators were statistically significantly different between instrumental variable groups. Conclusion: Analysis by intent to treat produced a hazard ratio closer to the null than analysis by treatment received, indicating that uncontrolled confounding toward more serious cases getting radiotherapy was present in the analysis by treatment received. Future studies should focus on obtaining enough numbers for subgroup analysis such as the stratification by risk groups. Caution should be used when using the instrumental variable approach in this population, as prognostic indicators were not as equally distributed as expected. / Thesis (Master, Community Health & Epidemiology) -- Queen's University, 2009-09-24 17:54:09.955