Doctor of Philosophy
Epidemiology and Biostatistics
Rodrigues, George B.
We evaluated the ability of matching techniques to balance baseline characteristics between treatment groups using non-experimental data. We identified a set of balance diagnostics that assessed key differences in baseline covariates with potential for confounding. These diagnostics were used in a novel systematic approach to developing and evaluating models for use in propensity score matching that optimized balance and data retention. We then compared the performance of propensity score and coarsened exact matching strategies in optimizing balance and data retention, using non-experimental data from a pan-Canadian prostate cancer database. Both matching techniques balanced baseline covariates adequately and retained approximately 70% of the data. To further study the role of treatment selection and prostate cancer outcomes, we performed a systematic review and meta-analysis that examined the rate of prostate cancer-specific mortality among those with high-risk non-metastatic prostate cancer who were initially treated with radiation or surgery. No statistically significant difference was found between groups in this analysis. In follow-up to this analysis, we compared the rate of metastatic progression following treatment between those with unfavourable-risk non-metastatic prostate cancer and treated with radiation or surgery, using data acquired from two Ontario cancer centres. Results from this comparison showed no statistically significant difference between treatment groups. In summary, a systematic approach to matching can be effective in balancing baseline covariates and producing more accurate effect estimates from non-experimental data. Moreover, treatment selection in the realm of higher risk prostate cancer does not appear to significantly influence important oncological outcomes.
Summary for Lay Audience
Accounting for bias in research performed using nonrandomized data is necessary to validly quantify differences in treatment effectiveness. Although statistical techniques can reduce bias, they are of limited value when treatment groups vary substantially. Matching individuals between treatment groups can overcome this issue; however, depending on how matching is accomplished, different issues may persist. For example, matching directly on all patient characteristics can lead to too few matches from which to draw valid conclusions. Alternatively, using a simple score derived from patient characteristics (e.g., a health score as defined by the presence of multiple illnesses, health behaviors such as smoking, exercise and diet, and age) might be limited in its ability to differentiate between those with similar scores who might still vary considerably in important ways. As such, we compared the ability of different matching strategies to balance important patient characteristics between treatment groups, while generating enough matches. To accomplish this, a set of tests were identified from previous research that adequately quantify important differences between treatment groups when attempting to estimate treatment effects. We developed a systematic approach to matching that optimized similarity in characteristics between groups, while maximizing the number of matches made. Finally, the ability of two different matching strategies were compared using nonrandomized data obtained from a pan-Canadian radiotherapy database of men diagnosed with prostate cancer. Both strategies performed well, leading to minimal differences between treatment groups, while generating enough matches to validly estimate treatment effects. In follow-up to the matching project using prostate cancer data, we aggregated effect estimates from studies comparing the effectiveness of radiation and surgery in treating high-risk prostate cancer, using a research technique called a systematic review and meta-analysis. No difference was found in the effectiveness of these two treatment modalities in this patient population. The last project in this thesis used the matching strategies developed in earlier chapters to compare the effectiveness of radiation and surgery in the treatment of higher-risk prostate cancer with newly acquired patient data. Like other studies, no difference in effectiveness was identified between these treatment modalities. However, due to data limitations, these estimates could not account for several potential biases which are explored in this thesis.
Guy, David E., "Addressing Bias in Non-Experimental Studies Assessing Treatment Outcomes in Prostate Cancer" (2021). Electronic Thesis and Dissertation Repository. 7935.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.