In our analysis, after rigorous adjustment, active treatments showed a significantly lower risk of overall and prostate cancer-specific mortality compared to observation, especially in intermediate- and high-risk disease. In comparison to our RWD study, three RCTs (SPCG-4, PIVOT, and ProtecT) showed different mortality rates by the initial treatment which may be partially explained by the periods of study, study design, quality of enrollment, the evolution of diagnostic and therapeutic approaches, and variable lengths of follow-up. Additionally, patients in clinical practice commonly differ from those enrolled in RCTs and thus, RCTs can have limited generalizability in some clinical settings or cohorts. Even the most recent RCT, the ProtecT trial, only included a small subset of high-risk patients (3%) 20 compared to 22.3% in our study, which better reflects the contemporary population at diagnosis.21 ProtecT also mainly included men identifying as white (ProtecT 99% vs. 78% in our study) which does not adequately represent the general population in the United States. ProtecT trial was limited with inadequate for the sample size,8 but our RDW study had enough sample size overcoming the shorter follow-up.
Population-based cohort studies allowed real-world evidence-based treatment strategies for men with localized PCa in the absence of generalizable RCT data. Indeed, many cohort studies have used the SEER database for competing-risk analysis,10,22 despite its limitations, including usage of “observation” as a proxy for both AS/WW and non-active treatments due to the absence of separate codes, or the risk of selection bias. In order to overcome this, Wong et al., used the quintiles of the estimated propensity score to balance observed covariates between treatment and observation groups,12 and concluded a better oncological outcome after RP - a finding which has been confirmed by other groups as mentioned above. However, the clinical T stage was not matched, and the absence of competing risk analyses made it challenging to draw the appropriate conclusion in terms of PCSM. In addition, men who underwent RT as primary treatment were excluded for analysis. Only three studies performed propensity score matching and competing risk analysis sequentially. Abdollah et al. performed two studies using SEER database (1992–2005),9,13 including a total of 44,694 patients. However, even after matching, the standardized difference of age was relatively high between the two groups (9.0 year vs. 0.1 year in our study). Furthermore, the exclusion of RT patients and the lack of risk stratification and/or PSA adjustment were substantial limitation of the study. In another study by Albertsen et al. using data from the Connecticut Tumor Registry, the authors concluded that men undergoing surgery for localized PCa may have an advantage in PCSM compared to those undergoing RT or observation.23 In this study, risk adjustment was not performed in this small cohort. Additionally, the authors acknowledged that their results may not be applicable to contemporary men as most men in this study were diagnosed between 1990 and 1992. This is an issue for many studies utilizing the SEER database before the 2010s, as mentioned above. Collectively, previous population-based research studies could not avoid the potential problem of patients’ risk adjustment or account for the effect of OCM. In most of the studies, traditional statistical methods, such as propensity score matching did not allow for a valid comparison among the three different treatment groups. Thus the RT group was usually excluded, despite it accounts for one-third of men with localized PCa.24 In this study, we used recently-a SEER-WW data (2010–2016), which has a newly created variable clearly defined as “AS/WW”. Unlike previous SEER databases, this new variable enables more accurate treatment group classification. To overcome the potential limitations of any observational study, we used the IPTW method which is one of the most advanced statistical adjustments, which provides flexible and valid three-way matching. In addition, IPTW-weighted competing risk analysis provides an unbiased estimate of the risk of PCSM in the presence of OCM. By rigorous adjustment for confounding factors and performing additional competing risk analysis using Fine and Gray method in this large study population, we could increase the statistical power in a more balanced way used before.
Since D’Amico et al. developed a combined modality staging system by stratifying patients into groups with a low-, intermediate-, and high-risk PCa in 1998,25 there has been a constant push for risk-based management of men with localized PCa.26 For low-risk men, observation as initial treatment has increased from around 30–60% from the early 2000s to 2010s, while RT rates trended down and surgery rates remained constant.2,4 For high-risk groups, on the contrary, treatment selection among RT, RP, and observation has remained relatively constant since the 2000s. Overall, between the early 2000s to 2010s, there has been a significant trend towards observation and surgery while decreasing RT decreased in localized PCa. Cumulative incidence rates of initial therapy for localized PCa stratified by risk group demonstrated that RP is performed more often in men with intermediate- and high-risk disease compared to low-risk (48%, 45%, and 10% for intermediate-, high-, and low-risk group, respectively).2 These findings are supported by our final competing risk analysis showing improved survival rates after RP for intermediate- and high-risk patients (RP vs. observation: sHR [95% CI] 0.44 (0.36–0.55) for intermediate-risk and 0.39 (0.33–0.46) for high-risk). Meanwhile, RT was chosen as primary treatment in 30–40% of men irrespective of risk classification in real-world clinical practice.2 Based on our study findings, RT did not confer a survival benefit in low-risk patients similar to RP. While high-risk patients were noted to have a significant survival benefit from RT compared to observation, there was no significant survival benefit in the intermediate-risk group.
There are several limitations that must be acknowledged. First, our observational study is inherently limited by lack of randomization and this may have biased survival outcomes. Despite every effort to control for all measurable parameters, it is still possible that insufficient balancing may occur due to unmeasured factors, leading to selection bias affecting the observed survival rates. This possibility was higher in RT vs. observation (E-value for sHR 1.5). However, it is not likely that unmeasured factors would have a greater effect on prostate cancer mortality than by having a sHR exceeding 3.77 in RP vs. observation. Second, the median follow-up period was relatively short (36.5-mon). In particular, the relatively shorter follow-up period in the observation group may have led to a lower cumulative cancer mortality rate compared to RT. To overcome this limitation, we attempted to increase statistical power through a large-sized IPTW-adjusted pseudo-population, with a more than 300-fold sample-sized cohort compared to previous RCTs. Third, for men in the observation group, we could not distinguish between WW or AS. Fourth, further information on treatment side effects and complications could not be taken into account, thereby limiting a comprehensive interpretation.
Despite these limitations, our analysis has several strengths. First, this study is performed in the contemporary era of PSA testing and treatment strategies. Second, this study is based on population-based RWD, which more accurately represents a real practice, contemporary patient distribution, especially in terms of racial/ethnic diversity, age range, risk groups, and geographic variation. Third, this is the first study of IPTW-based rigorous adjustment and competing risk analysis, demonstrating the association between initial treatment and survival outcomes supported by strong statistical power. Additionally, SEER-WW database analysis made study interpretation more precise by using a new, improved classification of the observation group (AS/WW) compared to prior SEER survival analyses, followed by the comparison of three independent treatment groups (observation, RT, and RP). Although the relatively short follow-up may limit a completely accurate assessment of survival outcomes for men with indolent, low-risk PCa, the large IPTW-adjusted sample size in this study leads to a more precise estimate of survival benefit in men undergoing active treatment.
After rigorous adjustment, active treatments including surgery, showed a significantly lower risk of overall and prostate cancer-specific mortality compared to observation in men with localized PCa - particularly in intermediate- and high-risk groups. However, observation (AS/WW) represents a safe option in men with low-risk PCa and should be the preferred treatment option in this subgroup.