Mortality After Radical Prostatectomy for Prostate Cancer
Mortality After Radical Prostatectomy for Prostate Cancer
Although surgery for localized PC has been compared with watchful waiting in randomized trials, there are no published survival results from randomized trials directly comparing RP with EBRT. We used data from the population-based observational PCOS to conduct a comparative effectiveness study of RP and EBRT. Using propensity scores as covariables in survival analyses, for stratification, for matched-pair analyses, and in an inverse probability of treatment weights analyses, we consistently found that overall and prostate cancer mortality were lower after RP than after EBRT.
The PCOS subjects were diagnosed in the mid-1990s when the majority of PC cases were detected with PSA screening. Only 6.3% of the subjects died from PC during 15 years of follow-up, reflecting the favorable survival associated with screen-detected localized PC. Subjects appear to be representative of cancer case subjects being diagnosed and treated for localized cancer during this time period. The cumulative PC mortality observed in PCOS subjects undergoing RP was 5.0%, which was just slightly lower than the 5.8% observed during a median 10-year follow-up (range = 7.3 to 12.6) in the RP arm of the PIVOT study. In contrast, the 15-year PC mortality in the prostatectomy arm of the SPCG-4 trial was 14.6%. This higher proportion is expected because just 12.4% of cancers in SPCG-4 were diagnosed at stage T1c.
The cumulative PC mortality observed in subjects undergoing EBRT was 17.2%, which is comparable with the 18% reported in abstract findings by the Scandinavian prospective randomized trial comparing external beam radiotherapy vs watchful waiting in early prostate cancer. The Scandinavian study enrolled 214 subjects from 1986 to 1997 and had a minimum follow up of 16 years. A more contemporary randomized trial of EBRT plus 4 months of ADT vs EBRT alone reported 10-year PC mortality in the study arms of 4% and 8%, respectively.
The survival advantage associated with RP needs to be interpreted cautiously. The outcomes reflect treatment modalities and approaches being offered in the mid-1990s and might be less applicable today, particularly in the case of EBRT for high-risk disease. More recent studies have shown that men with high-risk PC undergoing EBRT benefit substantially from prolonged (6 months to 3 years) courses of ADT. When we looked at the subset of men with high-risk PC who underwent RP or EBRT and ADT, we still found a statistically significant survival advantage with RP. However, this could reflect residual selection bias in the EBRT group due to unmeasured confounders or inadequate sample size. Another possibility is that the findings represent a true survival advantage.
Recent trial reports indicate that men with low-risk tumor features are unlikely to benefit from aggressive curative therapies. Observational data suggest that these men can be safely managed with active surveillance, with treatment avoided in the absence of clinical, biochemical, or histological evidence of disease progression. Indeed, when we limited our analysis to men with low-risk cancers, the treatment benefit associated with RP was markedly diminished. However, without an active surveillance control group, we cannot determine the absolute benefit of treatment or whether EBRT is better than no treatment for men with low-risk PC.
We found that RP was better than EBRT for men both older and younger than 65 years at the time of treatment, although the benefit was markedly attenuated in the older age group. The PIVOT trial also found no interaction between treatment arm and age. The SPCG-4 study found a survival benefit for receiving RP compared with watchful waiting only for men aged less than 65 years; however, these subjects had relatively more advanced stage clinically detected cancers.
Secular trends could also affect the generalizability of our results. For example, patients sometimes underwent pelvic lymph node dissection before radiation in the mid-1990s, either in the setting of aborted radical prostatectomy or for staging before EBRT. In an effort to control for this, we performed a sensitivity analysis in which we excluded the 12 patients in the radiation group (2.4%) who underwent pelvic lymph node dissection before EBRT. This did not statistically significantly alter our findings (data not shown). In addition, surgical and radiation techniques have evolved substantially over the past two decades. However, the advent of higher prostate dosing and three-dimensional conformal and intensity-modulation radiation therapy techniques has likely advanced radiation techniques to a greater degree than surgical advances. Conversely, patient selection has improved, particularly for men with limited life expectancy who are less likely to be treated surgically.
We recognize that residual treatment selection bias could have led to spurious estimates of survival differences. We attempted to address this bias with propensity score analyses based on the characteristics most strongly associated with treatment selection. We found statistically significantly decreased hazard ratios for PC mortality when using propensity scores as a covariable, when stratifying by propensity-score quintiles, when matching by propensity score, and when using an inverse probability of treatment weights approach. However, an SEER–Medicare analysis of men with localized PC used similar techniques in comparing overall survival for men receiving aggressive vs conservative treatment and was criticized for inadequate adjustment. By performing medical record abstractions and patient surveys, we included more detailed information on baseline covariables than can be obtained from claims data alone. Nonetheless, we may have missed capturing potentially important prognostic information, such as the number of positive biopsy cores and extent of biopsy tumor involvement.
Although we used a comorbidity score based on the Charlson index, this is a relatively crude measure. We may have failed to accurately characterize the severity of baseline comorbid conditions or suboptimally weighted these conditions, resulting in residual confounding. Since the initiation of the PCOS study, researchers have developed tools that not only better capture the severity of a particular condition but also are specifically designed to predict other-cause mortality in localized prostate cancer. Some authors have even suggested that the weighting of conditions in the original Charlson index could be changed to better predict outcomes in prostate cancer. It is possible that we missed other important factors, such as frailty or functional status, which may have resulted in residual bias. Indeed, we still found a survival benefit for RP even among men with no documented baseline comorbidities.
Population-based observational data on men diagnosed with localized PC in the mid-1990s suggest that treatment with RP was associated with a statistically significant reduction in overall and PC-specific mortality when compared with EBRT. The benefit was most notable in healthier men, those aged less than 65 years, and men with higher-risk cancers. Possible explanations include residual selection bias or a true survival advantage for RP in this cohort. In the absence of randomized controlled trials, our data provide the best estimate for the long-term comparative effectiveness of RP and EBRT. Further comparative effectiveness research is needed to compare treatment outcomes among more contemporary cohorts of men with localized PC.
Discussion
Although surgery for localized PC has been compared with watchful waiting in randomized trials, there are no published survival results from randomized trials directly comparing RP with EBRT. We used data from the population-based observational PCOS to conduct a comparative effectiveness study of RP and EBRT. Using propensity scores as covariables in survival analyses, for stratification, for matched-pair analyses, and in an inverse probability of treatment weights analyses, we consistently found that overall and prostate cancer mortality were lower after RP than after EBRT.
The PCOS subjects were diagnosed in the mid-1990s when the majority of PC cases were detected with PSA screening. Only 6.3% of the subjects died from PC during 15 years of follow-up, reflecting the favorable survival associated with screen-detected localized PC. Subjects appear to be representative of cancer case subjects being diagnosed and treated for localized cancer during this time period. The cumulative PC mortality observed in PCOS subjects undergoing RP was 5.0%, which was just slightly lower than the 5.8% observed during a median 10-year follow-up (range = 7.3 to 12.6) in the RP arm of the PIVOT study. In contrast, the 15-year PC mortality in the prostatectomy arm of the SPCG-4 trial was 14.6%. This higher proportion is expected because just 12.4% of cancers in SPCG-4 were diagnosed at stage T1c.
The cumulative PC mortality observed in subjects undergoing EBRT was 17.2%, which is comparable with the 18% reported in abstract findings by the Scandinavian prospective randomized trial comparing external beam radiotherapy vs watchful waiting in early prostate cancer. The Scandinavian study enrolled 214 subjects from 1986 to 1997 and had a minimum follow up of 16 years. A more contemporary randomized trial of EBRT plus 4 months of ADT vs EBRT alone reported 10-year PC mortality in the study arms of 4% and 8%, respectively.
The survival advantage associated with RP needs to be interpreted cautiously. The outcomes reflect treatment modalities and approaches being offered in the mid-1990s and might be less applicable today, particularly in the case of EBRT for high-risk disease. More recent studies have shown that men with high-risk PC undergoing EBRT benefit substantially from prolonged (6 months to 3 years) courses of ADT. When we looked at the subset of men with high-risk PC who underwent RP or EBRT and ADT, we still found a statistically significant survival advantage with RP. However, this could reflect residual selection bias in the EBRT group due to unmeasured confounders or inadequate sample size. Another possibility is that the findings represent a true survival advantage.
Recent trial reports indicate that men with low-risk tumor features are unlikely to benefit from aggressive curative therapies. Observational data suggest that these men can be safely managed with active surveillance, with treatment avoided in the absence of clinical, biochemical, or histological evidence of disease progression. Indeed, when we limited our analysis to men with low-risk cancers, the treatment benefit associated with RP was markedly diminished. However, without an active surveillance control group, we cannot determine the absolute benefit of treatment or whether EBRT is better than no treatment for men with low-risk PC.
We found that RP was better than EBRT for men both older and younger than 65 years at the time of treatment, although the benefit was markedly attenuated in the older age group. The PIVOT trial also found no interaction between treatment arm and age. The SPCG-4 study found a survival benefit for receiving RP compared with watchful waiting only for men aged less than 65 years; however, these subjects had relatively more advanced stage clinically detected cancers.
Secular trends could also affect the generalizability of our results. For example, patients sometimes underwent pelvic lymph node dissection before radiation in the mid-1990s, either in the setting of aborted radical prostatectomy or for staging before EBRT. In an effort to control for this, we performed a sensitivity analysis in which we excluded the 12 patients in the radiation group (2.4%) who underwent pelvic lymph node dissection before EBRT. This did not statistically significantly alter our findings (data not shown). In addition, surgical and radiation techniques have evolved substantially over the past two decades. However, the advent of higher prostate dosing and three-dimensional conformal and intensity-modulation radiation therapy techniques has likely advanced radiation techniques to a greater degree than surgical advances. Conversely, patient selection has improved, particularly for men with limited life expectancy who are less likely to be treated surgically.
We recognize that residual treatment selection bias could have led to spurious estimates of survival differences. We attempted to address this bias with propensity score analyses based on the characteristics most strongly associated with treatment selection. We found statistically significantly decreased hazard ratios for PC mortality when using propensity scores as a covariable, when stratifying by propensity-score quintiles, when matching by propensity score, and when using an inverse probability of treatment weights approach. However, an SEER–Medicare analysis of men with localized PC used similar techniques in comparing overall survival for men receiving aggressive vs conservative treatment and was criticized for inadequate adjustment. By performing medical record abstractions and patient surveys, we included more detailed information on baseline covariables than can be obtained from claims data alone. Nonetheless, we may have missed capturing potentially important prognostic information, such as the number of positive biopsy cores and extent of biopsy tumor involvement.
Although we used a comorbidity score based on the Charlson index, this is a relatively crude measure. We may have failed to accurately characterize the severity of baseline comorbid conditions or suboptimally weighted these conditions, resulting in residual confounding. Since the initiation of the PCOS study, researchers have developed tools that not only better capture the severity of a particular condition but also are specifically designed to predict other-cause mortality in localized prostate cancer. Some authors have even suggested that the weighting of conditions in the original Charlson index could be changed to better predict outcomes in prostate cancer. It is possible that we missed other important factors, such as frailty or functional status, which may have resulted in residual bias. Indeed, we still found a survival benefit for RP even among men with no documented baseline comorbidities.
Population-based observational data on men diagnosed with localized PC in the mid-1990s suggest that treatment with RP was associated with a statistically significant reduction in overall and PC-specific mortality when compared with EBRT. The benefit was most notable in healthier men, those aged less than 65 years, and men with higher-risk cancers. Possible explanations include residual selection bias or a true survival advantage for RP in this cohort. In the absence of randomized controlled trials, our data provide the best estimate for the long-term comparative effectiveness of RP and EBRT. Further comparative effectiveness research is needed to compare treatment outcomes among more contemporary cohorts of men with localized PC.
