Marieke van Son

34 CHAPTER 2 Figure 2 shows the PSW-adjusted Kaplan Meier survival curves estimating FFS (Figure 2a) and OS (Figure 2b) for the three separate treatment groups. After six years, estimated FFS was 87.4% (95% CI 79.9-93.9) in the EBRT group, 73.9% (68-80.9) in the LRP group and 74.4% (68.4-81.5) in the focal group ( p <0.001). Estimated 6-year OS was 92.3% (83.5-95.8), 95.3% (88.9-98.3) and 97.5% (94.9-100), respectively ( p =0.05). DISCUSSION Within the limitations of a cohort-based analysis, our study provides comparative ef- fectiveness data on cancer control showing no clear difference between FT and radical therapies after 6 years follow-up. Due to the observational nature of the data, system- atic baseline differences between groups may affect treatment outcomes. To minimize this effect, we used PSW to equalise the distribution of measured baseline covariates. The first assumption of a PSW-analysis is that the set of observed pre-treatment covariates is sufficiently rich such that the propensity score is constructed without missing important unmeasured or unknown confounders 27 . To this end, this study had limitations. We had no data of important characteristics such as PSA doubling time and a robust measurement of tumour volume. Instead we used MCCL, which appears to be an independent predictor of cancer volume 28 . We also used simplified T-stage categories (stage T1 or T2) due to a large proportion of missing data (40-65%) on sub-classifications of T2. Furthermore, we had no data on comorbidity profiles or so- cioeconomic status. EBRT patients were more likely to have comorbidities, considering that they were (on average) 5-8 years older and had higher mortality rates than LRP or FT patients. Although we did have data on history of neo-adjuvant ADT, this was not used for construction of propensity scores because the difference between groups (96% before EBRT versus 4% and 11% before LRP and FT) was too large to achieve sufficient balance. These differences in use of neoadjuvant ADT are likely to account for the FFS rates favouring EBRT, considering that residual effects of LHRH agonist use is known to continue in approximately 25% of men for many months after cessation 29,30 . The second assumption is that each patient has a probability of receiving each treat- ment and that there are no values of pre-treatment variables that could occur only among patients receiving one of the treatments 27 . We therefore chose inclusion criteria (PSA <20 ng/mL, ≤ISUP 3 and T-stage ≤T2c) that represent patients who could have been eligible for all treatments. Baseline variables that were used to construct propen- sity scores (age, PSA, ISUP grade, MCCL, T-stage and year of treatment) generally have no values that are exclusively seen in one of the treatment groups. The demonstrated FFS advantage for patients treated with EBRT was most surprising. From randomized comparative trials, there is evidence that at least prostatectomy and radiotherapy are comparable in terms of oncologic outcomes 3,31 . Although these trials were conducted between 1989-2009 and both treatment techniques have markedly improved since, updated results from recent observational studies have only confirmed oncologic equivalence 32 . There are several concerns potentially causing biased results