Time Trends in Use of Radical Prostatectomy by Tumor Risk and Life Expectancy in a National Veterans Affairs Cohort

Key Points Question How has the use of radical prostatectomy (RP) changed over time with respect to tumor risk and life expectancy (LE)? Findings In this cohort study of 5736 men treated with RP at 8 Veterans Affairs hospitals from 2000 to 2017, the proportion of low-risk tumors decreased 44%, the proportion of intermediate-risk tumors increased 29% (with favorable intermediate-risk tumors decreasing 20% and unfavorable intermediate-risk tumors increasing 11%), and the proportion of high-risk tumors increased 15%. During this period, the proportion of men treated with RP with LE less than 10 years increased 9%. Meaning These findings suggest that although use of RP has shifted away from low-risk and favorable intermediate-risk to higher-risk prostate cancer, its use among men with limited LE appears unchanged across tumor risk subgroups and increased overall.


Introduction
With the endorsement of active surveillance for low-risk and favorable intermediate-risk prostate cancer (PCa) by guideline-producing bodies during the last 10 years, 1-3 active surveillance has gradually replaced radical prostatectomy (RP) 4 as the most common treatment approach for low-risk PCa. Concurrent increases in the use of active surveillance, or conservative management, and decreases in the use of RP for low-risk disease have been observed across numerous practice settings. [4][5][6] These trends indicate that physicians and patients are appropriately paying greater attention to tumor risk before proceeding with potentially morbid local therapy, such as surgery or radiotherapy.
Guidelines also universally highlight the importance of life expectancy (LE) when considering aggressive vs conservative management for nonmetastatic PCa. In fact, the National Comprehensive Cancer Network guidelines 1 use LE as the primary branch point in triaging between aggressive and conservative management for all tumor risk subtypes. Considering that survival benefits associated with aggressive local therapy do not manifest until about 10 years after surgery 7 and treatment often worsens quality of life in older men and men with more severe illness, 8,9 National Comprehensive Cancer Network guidelines recommend observation or nonsurgical management for men with less than 10 years' LE and low-risk or intermediate-risk PCa.
Although epidemiologic studies have documented increasing use of active surveillance among subgroups of tumor risk and age, 4-6,10-13 few studies have explicitly looked at treatment trends by LE at diagnosis, to our knowledge. Although age can serve as a rough proxy for LE, both age and comorbidity have been shown to be independently associated with LE 14 ; therefore, both should be factored into clinical treatment decisions. Furthermore, several studies have shown that overtreatment of men with limited LE occurs most often in men with more severe illness, not older men. 15,16 Therefore, in this study, we sought to define treatment trends in the use of RP by tumor risk and LE in the active surveillance era using the Prostate Cancer Comorbidity Index (PCCI), a weighed scale based on age and comorbidity that was created and validated in men with PCa to predict longterm noncancer mortality (ie, LE). 17 Among those treated with RP, men with PCCI scores of 0 to 2, 3 to 6, 7 to 9, and 10 or higher have 10-year overall mortality rates of 7% to 15%, 19% to 28%, 41%, and 51%, respectively. 18 In this study, we analyzed the individual association of LE and tumor risk with time trends in the use of RP among subgroups of men with different PCCI scores (0-2, 3-6, 7-9, and Ն10) and tumor risk (

Data Source and Analytic Sample
We sampled 5736 men treated with RP for nonmetastatic PCa between January 1, 2000, and December 31, 2017, from the Shared Equal Access Regional Cancer Hospital (SEARCH) database, which includes data from 8

Life Expectancy
The PCCI is an externally validated, long-term, other-cause mortality prediction tool for early-stage PCa modeled after the Charlson Comorbidity Index. 24 It uses age and comorbidity at diagnosis to provide a weighted numerical score predicting long-term, other-cause mortality in men with PCa.
The PCCI is calculated by adding 1 point for each 6 years older than 60 years and additional points for the number and type of comorbidities, as previously described. 17 Prostate Cancer Comorbidity Index scores were grouped a priori into 4 categories (0-2, 3-6, 7-9, and Ն10) based on similar long-term mortality rates in a prior study of men with PCa in the VA Health System. 17

Statistical Analysis
Statistical analysis was performed from June 30, 2018, to August 20, 2020. Patient demographic and clinical characteristics were compared across tumor risk strata (low, intermediate, and high risk) using the Kruskal-Wallis test and the Pearson χ 2 test for continuous and categorical variables as appropriate. The primary outcome of our study was the rate of patients receiving RP over time.
Patient-level data were summarized on an annual level to estimate the proportion of patients receiving RP by PCCI, tumor risk, and by PCCI × tumor risk stratification. Univariate ordinary leastsquares regression was used to estimate the absolute change in the proportion of patients receiving RP by each tumor risk and PCCI stratification. Furthermore, a log-linear Poisson regression model with an offset was used to estimate the annual percentage change in the rate of RP across our study period by each PCCI × tumor risk stratification. A test of interaction between year of surgery and PCCI subgroups by each tumor risk stratification was performed to test for differences in the annual percentage change of RP over time between PCCI subgroups. A total of 21 comparisons were performed, with 3 comparisons within each tumor risk stratification. A Bonferroni-adjusted significance level of (.05/21) P < .002 was used to assess for significance of interactions.
We also conducted sensitivity analyses checking for interactions between race/ethnicity and year of surgery with annual percentage change of RP over time by each tumor risk stratification. We also conducted a parallel analysis with region (Northwest, Southern California, and South) instead of race/ethnicity. All statistical analyses were performed in R, version 3.5.1 (R Group for Statistical Computing) 25 using 2-sided tests with a significance level of P < .05.
When stratified by both tumor risk and PCCI scores, there were no significant differences in the use of RP across PCCI scores within strata of tumor risk (eFigure in the Supplement). In other words, within each tumor risk category, there were no trends away from treating those with the highest PCCI scores with RP. Ranges of annual percentage changes in the use of RP across PCCI subgroups within the same tumor risk strata were narrow ( Table 3). Tests for interaction between year of surgery and PCCI within each tumor risk subgroup were not statistically significant (Table 3). Tests for interaction between year of surgery and race or region within each tumor risk subgroup were also not statistically significant. Year The blue points and blue line show the trend in the proportion of patients undergoing radical prostatectomy with the given PCCI score range. The orange trend line shows the estimated proportion of patients undergoing radical prostatectomy with the given PCCI score range using the log-linear Poisson regression model. The shaded area represents the 95% CI of the estimates from the log-linear Poisson regression model.

Discussion
The present study shows a marked decrease in the use of RP for low-risk and favorable intermediate-   However, our study also shows that there has been very little change in treatment patterns of Other groups have created population-based LE prediction tools based on age and comorbidity, 30 and yet others have created treatment-specific tools. 29

Limitations
Several limitations of the current study are worth noting. First, the study sample consisted entirely of patients with PCa treated at 8 VA medical centers. Therefore, the results of the present study may not be readily generalizable to health care systems outside of the Veterans Health Administration, which have different payment models and patient sociodemographic characteristics that may both impact treatment choice. Second, the freedom to seek medical care outside of the Veterans Health Administration system may have led to the misclassification of men into smaller PCCI score subgroups owing to incomplete VA medical records on comorbidity status; this scenario could potentially diminish the difference in trends between PCCI subgroups. Third, we did not have a sufficient sample size to investigate differences in RP trends among the 8 VA medical centers.
Immense variation in the primary treatment of localized PCa at both the clinical level and clinician level 26,32 emphasizes the need for further studies on variation of cancer treatments in an effort to address discrepancies, improve treatment decision supports, and establish best-practice standards that align with national clinical practice guidelines.

Conclusions
In this study, we found that the tumor risk case mix among men receiving RP in the VA health care system has drastically changed in the active surveillance era, with increasing use of RP for high-risk and unfavorable intermediate-risk tumors and dramatically decreasing use for low-risk tumors.
Despite these changes, there is little difference in treatment trends by LE across tumor risk subgroups, and there is an overall increase in the use of RP among men with limited LE. Ideally, we would see sharper decreases in rates of RP use for lower-risk cancers and similar increases for highrisk cancers among those with limited LE, which were not observed in our data. Although the active surveillance era has shifted treatment paradigms based on tumor risk, there appears to be a persistent problem with appropriate management of men with limited LE. Adherence to guidelines and use of validated tools to predict LE may assist in improving management of these men.