An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
David F. Penson, Daniel A. Barocas; the Comparative Effectiveness Analysis of Surgery and Radiation (CEASAR), Investigators.
Author Information and AffiliationsThe 2008 evidence report of the Agency for Healthcare Research and Quality (AHRQ) called for high-quality, prospective cohort studies that identified men at the time of diagnosis of localized prostate cancer (PCa) and collected comprehensive patient, tumor, and treatment selection characteristics to generate critical patient-centered information to aid in decision-making for localized disease. Although new evidence has been generated since 2008 that has advanced the understanding of outcomes following treatment for PCa, the knowledge gap identified in the AHRQ report remains unfilled.
(1) Compare the effectiveness of contemporary surgical and radiation techniques for localized PCa; (2) identify patient-level characteristics that may influence comparative effectiveness; and (3) assess how the comparative effectiveness of the different therapies varies by the quality of care received.
The Comparative Effectiveness Analysis of Surgery and Radiation (CEASAR) study is a prospective, population-based observational cohort study of 3269 men younger than age 80 with clinically localized PCa diagnosed in 2011-2012 and enrolled in the study within 6 months of diagnosis. We followed the cohort longitudinally, collecting clinical and key patient-reported outcome data at baseline and at 6, 12, and 36 months after enrollment. In addition, we performed a full medical record review for each participant within a year of enrollment. We used multivariate statistical methods to achieve the proposed specific objectives and standard imputation methods to address missing data.
To achieve our first objective, we compared baseline, 6-month, 1-year, and 3-year health-related quality-of-life outcomes in men undergoing surgery, radiotherapy (RT), or active surveillance (AS). The analytic cohort for this analysis contained 2550 men: 1523 (59.9%) radical prostatectomy (RP); 598 (23.6%) external beam radiotherapy (EBRT); and 429 (16.6%) AS. At 3 years, sexual function for men undergoing RP was significantly worse than for those undergoing EBRT (−17.1 points; P < .001). The difference in sexual function between EBRT and AS at 3 years, while statistically significant, was not clinically significant (−5.9 points; P = .023). RP was associated with worse urinary continence scores than EBRT (−18 points; P < .001) or AS (−13.4 points; P < .001). No clinically significant differences in bowel or hormone function were noted. We explored the independent relationship of race on patient-reported outcomes at 1 year and found no significant associations. Similarly, we explored the relationship between commonly used quality of care metrics in PCa and patient-reported outcomes at 1 year and failed to find an association.
Significant differences exist in patient-reported outcomes following modern treatments for localized PCa, and these differences may vary in certain subgroups of patients. Future research should focus on exploring these potential associations and detecting longer-term differences in cancer control between treatments.
All observational studies face considerable risk for confounding by indication. We have attempted to control for this risk by using appropriate methods of statistical adjustment and by trying to capture previously unmeasured potential confounders at baseline. Our results may still contain some bias, although probably less than previous studies.
Prostate cancer (PCa) is the most common solid tumor among American men and the second leading cause of cancer death in men.1 Currently 2.7 million men are living with PCa in the United States, and the number is expected to increase to 3.9 million by 2022.2 Therapeutic options include surgical removal of the prostate (radical prostatectomy [RP]); irradiation of the prostate (which generally can be delivered as external beam radiotherapy [EBRT] or brachytherapy); and active surveillance ([AS] close monitoring of the tumor with serial prostate-specific antigen [PSA] measurement and repeated biopsies, reserving aggressive interventions for tumors that show signs of progression).3,4 Building on these decades-old approaches, new and costly technologies (including robotic surgery, dose-escalated intensity modulated radiotherapy [IMRT], stereotactic body radiotherapy [SBRT], and proton beam therapy) are now routinely offered to patients, without high-quality evidence that they are meaningfully better than existing therapies.5 All options have potential adverse effects that can have a significant impact on the patient's quality of life and day-to-day existence.6 In choosing a therapy, patients face the challenge of balancing survival, cancer control, and quality of life in the setting of imperfect information and evidence regarding outcomes. In addition, patients must make the difficult choice of a provider and a facility to deliver the treatment, in the presence of intense marketing and known variation in quality of care.7,8 Simply put, men newly diagnosed with localized PCa face a myriad of therapeutic options as well as provider and facility choices, and urgently need high-quality, reliable data that can be individualized to reflect their personal characteristics and preferences and guide their clinical decision-making.
The critical need to improve our knowledge concerning the comparative effectiveness of therapeutic strategies to treat localized PCa has been highly prioritized by both the Institute of Medicine (IOM) 9 and the Agency for Healthcare Research and Quality (AHRQ).10 In the IOM's report on Initial National Priorities for Comparative Effectiveness Research, the need to “compare the effectiveness of management strategies for localized PCa on survival, recurrence, side effects, quality of life, and costs” was prioritized in the highest quartile of topics.9 Similarly, AHRQ's 2008 systematic review of the literature on the comparative effectiveness of therapies for localized PCa10,11 concluded that uncertainty about the comparative effectiveness and harms of the primary treatments for localized PCa represented a major gap in knowledge and called for
high-quality, large, prospective cohort studies … that identify men at the time of diagnosis and … collect comprehensive patient, tumor, and treatment decision selection characteristics.10
The AHRQ report noted that few randomized clinical trials (RCTs) are available that have compared surgery with radiation and that most of the evidence to guide decision-making is of low quality.10 When the report was published, only 1 completed RCT in the literature had directly compared surgery with radiation.12 In addition to being so underpowered that it could not reach any meaningful conclusions, this trial was limited by methodological flaws regarding randomization and follow-up. Two completed RCTs have compared surgery with observation. Scandinavian Prostate Cancer Group Study Number 4 (SPCG-4) randomized 695 men with clinical, localized, well-differentiated or moderately differentiated PCa to receive either open RP (n = 347) or watchful waiting (n = 348) from 1989 through 1999. With a median follow-up of 12.8 years, the study found a significant advantage for surgery in terms of both overall survival (P = .007) and PCa-specific survival (P = .01).13 However, this study may not be helpful to patients diagnosed in 2013, as most of the patients in the study were not detected through PSA screening, and surgical technologies have evolved considerably since the 1990s. In the Prostate Intervention Versus Observational Therapy (PIVOT) study, Wilt and colleagues14 randomized 731 men from the Department of Veterans Affairs (VA) system diagnosed with localized PCa from 1994 to 2002 to either open surgery (n = 364) or watchful waiting (n = 367). Unlike SPCG-4, PIVOT failed to show an overall or disease-specific survival advantage for surgery, although certain patient-level subgroup analyses (such as PSA greater than 10 ng/ml) did favor surgery. However, PIVOT has been criticized for inadequate sample size, high numbers of participants not receiving their assigned intervention, inclusion of an infirm population not representative of today's patients, and inadequate follow-up.15 The Prostate Testing for Cancer and Treatment (ProtecT) study, a single 3-arm RCT comparing surgery to radiation to AS in the United Kingdom, recently reported out 10-year results.16 ProtecT included roughly 500 men in each arm but failed to show a statistically significant difference in overall mortality among the groups. However, most study participants had low-risk disease, and 10-year follow-up is inadequate to show significant differences in survival in these patients. Certainly, the findings in a cohort of predominantly low-risk men may not be applicable to intermediate- and high-risk patients. While ProtecT also included patient-reported outcomes,17 the study interventions may not reflect modern surgical or radiation techniques. Radiation patients in ProtecT underwent 3-D conformal radiotherapy (RT) with androgen deprivation therapy (ADT), whereas the contemporary norm is IMRT, with selective use of ADT.18-20 Similarly, 87% of surgery patients in ProtecT underwent open RP, while robot-assisted surgery is used in more than 80% of RPs in the United States.21,22 Furthermore, the ProtecT trial population was 99% White and nearly 80% low risk, limiting generalizability.16 Thus, we still need comparative effectiveness data regarding the outcomes after contemporary treatments for localized PCa.
Given the shortcomings of available RCTs in localized PCa, researchers have turned to well-designed observational studies in an attempt to fill knowledge gaps. To date, observational studies have mostly focused on health-related quality-of-life and functional outcomes in PCa, as these endpoints are important to patients and are often the primary factor in decision-making.23-26 Studies show that sexual dysfunction (specifically, erectile and orgasmic dysfunction) is common after both surgery and radiation,6,27,28 although it is worse after surgery.6,29,30 Urinary dysfunction is also common after both therapies, although the type of dysfunction differs: Surgery tends to be associated with stress urinary incontinence,31,32 while radiation is often associated with irritative symptoms.32,33 Both sexual and urinary function have been shown to be associated with other patient characteristics, such as age,31,33 race/ethnicity,34,35 and comorbidity.33,36 Finally, it appears that there is more bowel dysfunction following RT.6,37 While informative, the observational comparative effectiveness research literature in PCa is limited by frequent use of cross-sectional design, inclusion of older therapies, lack of adequate follow-up, and inadequate risk adjustment owing to unmeasured confounders.
Given the limitations of the existing research, AHRQ's 2010 report Future Research Needs for Comparative Effectiveness of Treatments of Localized Prostate Cancer reiterated the need for additional high-quality studies to aid patients in decision-making in the face of this common malignancy.38 The authors prioritized an updated list of potential research studies based on AHRQ's 2008 literature review and any new publications since its release. The 2010 report's technical experts panel agreed that while RCTs would be the best study design to address many of the research gaps, previous experience indicated the difficulty of conducting such trials in the United States for early stage disease. Thus, other research designs, including observational studies, were deemed acceptable. One of the 3 important gaps identified by the report was the need to develop a better understanding of the comparative effectiveness of different treatments for localized PCa. To fill the knowledge gap regarding factors that affect treatment decision-making, the AHRQ recommended studies that explored both the comparative effectiveness of primary treatments and the effectiveness of alternative strategies within a modality; studies that assessed the long-term sequelae of treatments; and studies of the psychological impact of diagnosis and treatment, especially for those under AS.38 The Comparative Effectiveness Analysis of Surgery And Radiation (CEASAR) study was specifically designed to generate critical new information on the comparative effectiveness of modern treatments and to fill these research gaps.
Throughout the investigative process, we engaged stakeholders representing the population of interest as research team members and sought their input. Specifically, we included a patient advocate on the research team from day 1 of the process, soliciting input and guidance in writing the original funding application and maintaining this engagement through the research process up to the present day. Initially, we partnered with Dan McCollum, a Vanderbilt employee who is both a PCa survivor and a leader in the local PCa advocacy community. McCollum worked with the principal investigator and the investigative team to craft the specific objectives and the research plan of the grant proposal, ensuring from its onset that the study would generate information that was meaningful to patients and other stakeholders.
After we were funded, we worked with McCollum and the Vanderbilt–Ingram Cancer Center (VICC) to identify interested PCa survivors to serve on a patient advisory council, which met as soon as we received formal notice of funding. VICC is one of the largest providers of cancer care in middle Tennessee and is committed to patient-centered health care and improving cancer survival. A regular PCa support group was already in existence, as well as a cadre of motivated and interested PCa patient advocates (and their partners) who were committed to helping other men who suffer from this common malignancy. We identified a group of roughly 15 ethnically diverse PCa survivors of varying ages to serve on our advisory council. The council met quarterly during the project and helped us focus the research questions, choose patient surveys, identify ways to maximize recruitment and adherence, and choose methods for disseminating study results.
During these meetings, we identified a second patient advocate, Ralph Conwill, who wished to become more involved in the day-to-day activities of the study. Conwill and McCollum began to attend our weekly research meetings and provided us with invaluable feedback on the conduct of the study and the analysis of the results. Both are particularly interested in how we will disseminate our findings and have maintained full engagement to the present day. The unique input of the patient advisory council, coupled with the regular and intense involvement of Conwill and McCollum, ensured that we met PCORI methodology standard PC-1 for patient-centeredness and added great value to the work. An in-depth description of their involvement can be found at http://www.pcori.org/research-in-action/prostate-cancer-voices-experience.
CEASAR is a prospective observational cohort study. The study population consists of a population-based cohort augmented by a small sample of men from an observational disease registry to ensure representation of AS and emerging technologies. The 2010 AHRQ report on future research directions in PCa drew a number of important conclusions that led us to select this study design.38 First, the report noted that although RCTs would be the best study design to address many of the research gaps, previous experience illustrated the difficulty of conducting such trials in the United States for early stage disease. Second, the AHRQ report noted that emerging technologies such as IMRT, proton beam radiation, laparoscopic and robotic assisted prostatectomy, and cryotherapy were increasingly being used in practice despite the absence of RCTs demonstrating their effectiveness and that completion of these trials would be challenging. Third, and most important, AHRQ identified a need for large, high-quality prospective cohort studies that enrolled men at the time of diagnosis and collected comprehensive data on patients, tumors, and treatment decisions.
We used a Rapid Case Ascertainment System (RCAS) to identify men with localized PCa diagnosed from January 2011 through February 2012 in 5 population-based Surveillance, Epidemiology, and End Results (SEER) registry catchment areas (Atlanta, GA: Los Angeles, CA; LA; NJ; and UT). RCAS leverages electronic and written communication between pathology laboratories and the SEER registries to identify newly diagnosed cases rapidly, so patients could be consented, and baseline data gathered before the initiation of therapy or immediately thereafter. The SEER registries have been used successfully as a cohort inception tool in other studies39 because this methodology minimizes selection bias, ensures that the cohort is a nationally representative sample, and leverages the RCAS. In addition, the SEER sites employ skilled investigators and professional abstractors who have demonstrated the ability to execute these studies and yield high-quality data. The inclusion of SEER sites in Atlanta, Louisiana, and Los Angeles ensured a racially and ethnically diverse population, including a significant proportion of socioeconomically disadvantaged men. We also accrued patients from the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) observational disease registry, launched in 1995 by the University of California–San Francisco, which includes >13 800 men diagnosed with PCa at 1 of 40 community and academic practice sites nationwide.40 By including CaPSURE patients, we intended to enrich the population with patients who were undergoing novel therapies and AS, since many of the CaPSURE sites are large urology group practices, which tend to embrace emerging management strategies early on. The central organizing site at Vanderbilt, each of the SEER sites, and CaPSURE obtained approval from the relevant local IRBs.
Inclusion criteria were as follows: men younger than 80 years of age with newly diagnosed, pathologically confirmed, clinically localized adenocarcinoma of the prostate diagnosed within 4 months of enrollment; PSA <50 ng/ml; English or Spanish speaking; and able and willing to give informed consent. We excluded men who did not meet all the inclusion criteria.
The 5 population-based SEER registries participating in CEASAR are Atlanta, Los Angeles, Louisiana, New Jersey, and Utah. These registries are mandated by their respective states to routinely collect and report information on all cancers diagnosed in their catchment areas, including sociodemographic information and tumor characteristics, such as final pathologic stage and histologic grade. We chose these 5 registries because of their ability to recruit and retain participants in general and their specific ability to recruit from underrepresented populations. In addition, we recruited participants from the CaPSURE observational disease registry. The goal of including CaPSURE was to selectively capture participants undergoing treatment at centers that were high users of novel technologies such as cryotherapy, high-intensity focused ultrasound, and image-guided radiotherapy. We applied the same inclusion and exclusion criteria used for participants enrolled from the SEER registries to the CaPSURE participants.
As this was an observational study that explored all potential treatments for localized PCa, there were no study-directed interventions, and treatment comparisons were driven by the most commonly performed treatments for the disease. The primary interventions of interest (and those most commonly performed) were RP (primarily using a robotic-assisted laparoscopic approach); external beam radiation (usually in combination with androgen deprivation therapy using an intensity modulated approach); and AS, which was considered the primary comparator. AS was first ascertained by patient report and then confirmed by medical record review. AS was defined as documentation in the patient survey or the medical record that (1) the patient had proactively selected AS; (2) serial PSA monitoring and, when appropriate, repeat prostate biopsy were performed; and (3) neither surgery nor radiation was administered within the first 6 months of diagnosis. We did encounter some use of other therapies, including interstitial brachytherapy, image-guided radiotherapy (such as cyberknife), proton beam therapy, cryotherapy (and other types of ablation), and various other experimental treatments. However, the total number of participants treated with these modalities was relatively small, so we focused on comparing surgery, external beam radiotherapy, and AS. All primary treatments were captured in the analysis and no patient had received treatment for PCa prior to enrollment in the study.
At the time of enrollment, patients were asked to complete a baseline questionnaire that collected clinical and sociodemographic information, health-related quality-of-life data, and other patient-reported outcomes information for use in our analyses. Participants were contacted at 6, 12, and 36 months after enrollment and asked to complete surveys that assessed clinical outcomes, health-related quality-of-life outcomes, and other patient-reported endpoints. At roughly 12 months after enrollment, we attempted to obtain and review each participant's detailed medical records related to his PCa care from hospitals, free-standing radiologic or surgical centers, VA hospitals, health maintenance organizations, and private physician offices. For participants who received radiotherapy, we obtained records from the treating radiation oncologist to ensure that we could accurately assess the type of radiation delivered and the quality of the treatment. Records from multiple physicians (primarily urologists and radiation oncologists) were abstracted on the basis of lists of treating physicians obtained from participants and from registry sources.
The primary outcome measures analyzed were 36-month domain scores on the 26-item Expanded Prostate Index Composite (EPIC-26), a validated instrument for measuring disease-specific function in sexual, urinary incontinence, urinary irritative, bowel, and hormonal domains after treatment for PCa.41 We elected to focus on these outcomes at the recommendation of our patient stakeholder partners and after discussion with various experienced providers of PCa care. Domain scores on these instruments range from 0 to 100, with higher scores representing better function. The minimally important difference (MID), representing the magnitude of change that is clinically meaningful to patients, has been estimated for each domain, using one-third to half a standard deviation and anchoring to satisfaction with treatment.42 The sexual function domain focuses on the quality and frequency of erections (MID, 10-12 points). The urinary incontinence (MID, 6 points) and irritative (MID, 5 points) domains ask questions about frequency and amount of urinary leakage and symptoms such as dysuria, hematuria, and urinary frequency. The bowel function domain (MID, 4 points) focuses on bowel frequency, urgency, bleeding, and pain. The hormonal domain (MID, 4 points) assesses symptoms such as hot flashes, gynecomastia, low energy, and weight change. Although we collected information on a number of other patient-reported outcomes of interest (including PCa–related anxiety and participatory decision-making style), they were identified as secondary by our stakeholder panel; in this report we focus on the analysis of primary outcomes.
After identifying a potential study participant through the RCAS, we contacted the diagnosing physician to obtain his or her consent to contact the patient. This ensured that the patient was aware of the diagnosis before we contacted him. If consent to contact the patient was not actively refused, the patient was invited by mail to participate. As per our IRB-approved study protocol, return of the baseline survey constituted initial implied consent for participation. If the patient did not return the survey within 2 weeks, he was contacted by telephone and baseline data collection was performed by phone at that time, assuming he was willing to participate in the study. Verbal consent to participate was collected at the time of the phone interview. To ensure the quality control of the telephone interviews, we developed a standardized telephone script and provided training for the interviewers. Once data collection began, we monitored the number of telephone interviews required and reviewed results at our monthly teleconferences. We compared the results from phone interviews to patient-completed surveys on a quarterly basis and did not note any major discrepancies.
At the 6-month point, a new survey was mailed to each participant. At this time, participants were also required to sign written informed consent for the 12-month medical record review and further study participation. If the participant failed to respond to the mailed survey within 2 weeks, a member of the site research staff called him to determine whether he had received the study materials and to answer any questions he might have about the study. The staff member offered to complete the survey with the respondent over the telephone then or at another mutually agreed on time. If the participant requested a second copy of the survey, we mailed one. The questionnaire was translated into Spanish, and interviews were conducted in Spanish when necessary. We made follow-up phone calls as needed to complete the questionnaire or determine that the patient did not want to participate. If a patient actively refused participation in the study, we did not attempt to collect further information and excluded him from the analysis. We made extensive efforts to locate nonrespondents in order to achieve the highest response rate possible. We used internet search engines, Department of Motor Vehicle records, and other public records, as well as specialized tracking resources at each of the SEER registries. We used a similar approach at the 12- and 36-month follow-up points.
In general, we used a unified, generalized regression approach to investigate the association between the response (eg, health-related quality of life, complications of therapy) and explanatory variables (eg, treatment types) of interest for each objective. In all analyses, we chose dependent and independent variables in advance. The distributions of the primary variables of interest and other explanatory variables were summarized graphically and numerically. We performed all statistical analyses using R version 3.2.2.
Compare the effectiveness of contemporary surgical and radiation techniques for localized PCa.
We compared baseline demographic characteristics across treatments using Kruskal-Wallis tests for continuous variables and chi-square tests for categorical variables. To describe typical trajectories of function over time, we fit longitudinal regression models to predict EPIC-26 domain scores as a function of treatment, time since treatment, and their interaction. For each domain, we fit a single model incorporating domain scores from all time points. We used generalized estimating equations (GEEs) with an independent weight matrix because of the correlation between observations on the same patients. Modeling time using regression splines allowed for a flexible relationship between function and time. Variability in the interval between treatment and survey completion allowed for estimation of domain scores between rounds of data collection.
To measure the association between treatment choice and domain score over time, we fit a similar set of models that adjusted for age, race, comorbidity,43 disease risk stratum,44 physical function,45,46 social support,47 depression,48 medical decision-making style,49 site, and baseline EPIC-26 domain score.41 We designed this multivariable modeling approach to minimize bias associated with known differences in baseline characteristics associated with functional outcomes (ie, confounding). Because ADT is a standard component of EBRT for high-risk disease and an option in intermediate-risk disease, it was not controlled for in the models.50 Instead, we fit exploratory models for sexual and hormonal function as dependent variables with 5 treatment groups: nerve-sparing RP, non–nerve-sparing RP, EBRT without ADT, EBRT with ADT, and AS. Using GEEs, we fit unadjusted and adjusted longitudinal logistic regression models for responses to individual EPIC-26 items, using the same covariates as above. Differences in domain scores between treatments were statistically significant if the tailed P value was <.05 and were interpreted as clinically meaningful if the differences were as large as the MID.
We addressed missing data as follows: We excluded 199 CEASAR participants with no postbaseline outcome measures whatsoever. This left 2550 participants in the analytic cohort. We compared baseline characteristics of the analytic cohort with those of the 199 participants with no follow-up information whatsoever and noted no statistically significant differences, suggesting that the analytic cohort was representative of the general population. According to the EPIC-26 scoring algorithm,51 patients who answered at least 80% of questions in a particular domain received a score for that domain (computed as an average of the scores on questions that were answered, scaled from 0-100). We considered those who answered less than 80% of questions in a particular domain to have missing data for that domain and handled them as described below.
We imputed missing values of regression model covariates (baseline clinical, sociodemographic, and baseline EPIC domain scores or individual EPIC-26 items) using the multiple imputation using chained equations procedure.52,53 We used imputation only for independent variables that were measured before treatment; thus, we imputed no outcome variables measured after treatment. In this procedure, we imputed missing values of covariates by modeling each covariate as an outcome in a regression model, using all other model covariates as predictors. In this case, we used only baseline data (excluding treatment). This is described by Harrell and implemented using the rms package in R.54
We used multiple imputation to avoid casewise deletion of all observations with at least 1 missing value of the independent variables. The imputation and regression model fitting with imputed data involves 3 steps: (1) imputing the data using the imputation models, (2) estimating coefficient parameters and their SEs in the analysis regression models using the imputed data, and (3) adjusting the SE estimates to account for the variability associated with the imputation procedure.
To impute missing data on a covariate (X), we first imputed the missing data using a random sample of nonmissing X. Then we fit a flexible additive model on a bootstrap resampled data set using X as the outcome. Using this model, we obtained fitted values for the variable. To impute a missing value, we found nonmissing X whose fitted value was closest to the fitted value of the missing observation and imputed the missing value with the matching nonmissing X. This approach is referred to as predictive mean matching.54
We repeated this resample-model-impute step 15 times after a burn-in period of 10 iterations, with the missing values updated with the imputed values after each step. Then we fit the final model using the complete data set using the values imputed at the last iteration, and adjusted the standard error estimates to account for additional uncertainty associated with the imputation, as described in Harrell.54
Identify patient-level characteristics that may influence comparative effectiveness.
We planned numerous subgroup analyses. The variables of primary interest for these analyses include race/ethnicity, age, disease risk strata, number and severity of comorbid conditions, education level, and annual household income. We hypothesized that all these variables would have an independent effect on patient-reported outcomes following treatment for PCa but that this effect would not vary by treatment received. To date, we have completed the subgroup analyses comparing patient-reported outcomes by race/ethnicity group while controlling for treatment received; we present these results herein. The other subgroup analyses are ongoing, so we do not have final results to present. For the subgroup analyses exploring the effect of race/ethnicity on outcomes, we compared patients' baseline demographic and clinical characteristics across racial/ethnic groups using Kruskal-Wallis and chi-square tests. To characterize typical changes in patient-reported function over time in each treatment group among each racial/ethnic group, we fit longitudinal regression models predicting each EPIC-26 domain score using only race/ethnicity, treatment type, and their interaction as independent variables. We used all of each patient's scores in a particular domain over time and modeled time since beginning treatment (time since baseline survey in AS patients) continuously. We modeled the relationship between time and mean function as restricted cubic spline permitting nonlinearity. We used GEE with an independence working covariance matrix to calculate standard errors for each regression coefficient.
To identify racial/ethnic differences in the effect of treatment on functional outcomes following treatment, we fit a set of longitudinal models with interactions between race/ethnicity and treatment type, adjusted for the following baseline factors: time since beginning treatment; pretreatment function; patient age; comorbidity tumor characteristics (PSA corrected for 5-alpha reductase inhibitor use, Gleason score [≤6, 3 + 4, 4 + 3, or ≥8], and T-stage [T1 or T2]); psychosocial measures (educational attainment, insurance type, employment type, marital status, physical function score, social support, depression score, and participatory decision-making index); receipt of hormone therapy; and study site. We fit adjusted models using the same approach as the unadjusted models (GEE). We characterized treatment effects by differences in function score between treatment groups at 1 year after treatment, and characterized racial differences in treatment effects by the difference between races in these treatment effects. Thus, our estimate of interest is a difference in differences (DID) accompanied by a 95% CI. In a sensitivity analysis, we used a propensity score regression adjustment as an alternative means of accounting for systematic pretreatment differences between patients receiving different treatments. We used a multinomial logistic regression model to estimate the log odds of receiving each of the 3 treatments. We then included the fitted values from this model in a second version of our main analysis model as independent variables.
We first imputed missing data using multiple imputation via predictive mean matching to avoid casewise deletion of patient records missing any model covariates.55 Because African American men were more likely to undergo open RP than robotic RP (and less likely to receive a nerve-sparing operation and IMRT), we performed a second sensitivity analysis to assess the impact of treatment technique on the results by excluding men who did not have a robotic nerve-sparing operation and those who did not receive IMRT.
Assess how the comparative effectiveness of the various therapies varies by the quality of care received (quality measures).
We chose 6 quality measures on the basis of their endorsement by the National Quality Forum, Physicians Consortium for Performance Improvement, and Physician Quality Reporting System. For the measures that describe documentation at new diagnosis and prior to treatment, adherence to PSA documentation was not available and was either omitted from the measure (documentation in newly diagnosed) or exchanged for digital rectal examination (documentation before primary therapy). Measures containing multiple elements (eg, complete pathology documentation) required compliance with every element to be compliant with the quality measure.
We calculated physician compliance with each measure in the relevant patient population. We used D'Amico risk stratification56 to determine the proper patient groups for guideline concordant imaging use. We calculated pretreatment EPIC-26 scores for both compliant and noncompliant groups and compared them using appropriate parametric statistical tests. We performed multivariable linear regression analysis to determine the effect of measure compliance on EPIC-26 domain scores, adjusting for baseline EPIC score; treatment type (surgery vs radiation); age (<65 vs ≥65 years); race (White vs other); household income level (≤$50 000 vs > $50 000); insurance status (Medicare, private, or other); comorbidity (Total Illness Burden Index for Prostate Cancer)43; D'Amico risk classification; and SEER site. We performed multivariable logistic regression with adjustment for patient and disease characteristics to determine the effect of measure compliance on complication rates.
We submitted the final study protocol to the IRB at each study site and the coordinating center; it was approved in all cases. We made no changes to the study protocol after submission, with the exception of amendments to allow for 36- and 60-month data collection. IRB approval was sought and granted at each of the 5 SEER sites, the University of California– San Francisco (in its role as the supervising site for CaPSURE), and Vanderbilt University Medical Center (in its role as the coordinating center and primary analytic site).
Compare the effectiveness of contemporary surgical and radiation techniques for localized PCa.
The analytic cohort contained 2550 men: 1523 (59.7%) RP, 598 (23.5%) EBRT, and 429 (16.8%) AS, as shown in Figure 1. Baseline characteristics are shown in Table 1. Of the cohort, 26% was non-White. EBRT patients were older, had a higher comorbidity burden, and had higher-risk disease features compared with RP patients. Of AS patients, 77% were low-risk. Among RP patients with sufficient data, 79% had bilateral nerve-sparing surgery and 76% had robotic surgery. Of EBRT patients, 81% had IMRT, and 45% had ADT within the first year. By the 3-year survey, 24.2% of AS patients had undergone treatment, and 90.2% of the remainder had had their PSA checked within the past 12 months.
Diagram of the Assembly of the Analytic Cohort in the CEASAR Study.
Baseline Clinical and Sociodemographic Characteristics of the CEASAR Analytic Cohort.
Men undergoing RP had higher baseline sexual domain scores than men undergoing EBRT and comparable scores to those of men on AS (Table 2). RP and EBRT were associated with declines in sexual function, resulting in a similar domain score at 3 years on unadjusted trajectories (Figure 2A). However, when controlling for baseline domain scores and other covariates (Table 2, Figure 2B), the decline in sexual domain score was significantly worse after RP compared with EBRT (−11.9 points; 95% CI, −15.1 to −8.7) or AS (−16.2 points; 95% CI, −20.6 to −11.7), relative to the MID of 10 to 12. Adjusted domain score after EBRT was significantly worse than AS at 12 months (−10.5), but the magnitude of difference at 3 years was no longer clinically significant (−4.3).
Sexual Function Outcomes.
Unadjusted Mean and Adjusted Mean Point Difference in EPIC-26 Domain Scores (95% CI), Longitudinally by Treatment.
The factors most strongly associated with sexual function outcomes were treatment and baseline domain score (Figure 2B). We then performed an exploratory analysis comparing 5 different forms of treatment: AS, EBRT with ADT, EBRT alone (without ADT), nerve-sparing RP, and non–nerve-sparing RP. The difference between EBRT alone and AS was not statistically significant (−3.0 points; P = .266), and the difference between RP and EBRT + ADT was attenuated (−8.2 points; 95% CI, −13.2 to −3.2) below the MID (Figure 3).
Adjusted Mean EPIC-26 Domain Score (95% CI), Longitudinally by Treatment.
In a separate analysis of 991 men within CEASAR, we explored the independent effectiveness of the nerve-sparing technique on 3-year sexual function outcomes following surgery. In a multivariate analysis, men who underwent bilateral nerve sparing had significantly higher mean sexual function scores than those who underwent unilateral or non–nerve-sparing surgery (6.1 points; 95% CI, 2.0-10.3; P = .004). The effect of nerve sparing was highly dependent on baseline function: In a subgroup analysis, men with high baseline sexual function showed a greater independent benefit from bilateral nerve sparing than men with low baseline sexual function (8.23 points, P = .014 vs 4.0 points, P = .09).57
More men who underwent RP were bothered by sexual dysfunction 3 years after diagnosis (44% vs 35% for EBRT and 28% for AS; P < .001 on multivariable analysis) (Table 2). Erection insufficient for intercourse was common at 3 years (70% for RP, 71% for EBRT, and 51% for AS on raw percentages) but, controlling for baseline sexual function and other factors, odds were significantly higher for RP vs AS (odds ratio [OR], 3.4 [2.5-4.6]) and RP vs EBRT ([OR] [1.5-2.9]). Among men who had erections sufficient for intercourse at baseline, sufficient erections at 3 years were reported in 43% (40-47) for RP, 53% (45-60) for EBRT, and 75% (68-80) for AS in raw percentages.
Baseline urinary incontinence and irritative domain scores were similar across groups. However, RP was associated with a significant decline in urinary continence score after treatment compared with EBRT or AS (Table 3, Figure 2C). Despite some improvement in incontinence domain scores 12 months after RP, adjusted incontinence scores were still significantly worse for RP compared with AS (−12.7 points [−16.0 to −9.3]) and EBRT (−18.0 points [−20.5 to −15.4]) at 3 years, differences greater than the MID (6 points). By contrast, urinary incontinence was not significantly different between EBRT and AS on unadjusted or adjusted analyses. Irritative function scores were slightly better for men undergoing RP compared with AS at 1 and 3 years, with adjusted differences at the threshold of clinical significance (4.5 and 5.2 points). Other comparisons, while statistically significant, were below the MID of 5 (Table 3, Figure 2E, Figure 2F).
Urinary Function Outcomes.
The factors most strongly associated with incontinence were treatment (RP) and baseline domain score (Figure 2D). The same factors were associated with urinary irritative scores, but RP was associated with improvement rather than decline compared with AS (Figure 2F).
Urinary function bother scores were not demonstrably different for RP vs AS and EBRT vs AS at 3 years but were higher for RP vs EBRT (12% vs 10%; OR 1.7 [1.1-2.5]). Moderate or large problem with urinary leakage was more common after RP vs AS (14% vs 6%; OR, 2.9 [1.8-4.7]) and RP vs EBRT (14% vs 5%; OR, 4.5 [2.7-7.3]). Burning with urination was uncommon (2% in each group). Moderate or large problem with frequent urination was lower for RP vs AS (13% vs 18%; OR, 0.6 [0.4-0.8]) and for EBRT vs AS (15% vs 18%; OR, 0.6 [0.4-0.8]) at 3 years, but not demonstrably different between RP and EBRT.
Decline in bowel domain score was not common (Table 4, Figure 2G, and Figure 2H). Six months after treatment, domain scores were higher in men who underwent RP vs EBRT (4.6 points [3.2-6.1]) and lower for EBRT vs AS (−5.8 points [−10.3 to −1.2]). However, by 12 months, these differences were near the MID of 4 (RP vs EBRT: 4.2 points; EBRT vs AS: 3.5 points), and by 36 months, they were even smaller (RP vs EBRT: 3.0 points; EBRT vs AS: −1.6 points). Unadjusted and adjusted results were similar.
Bowel Function Outcomes.
The frequency of moderate or large problem with bowel bother, bloody stools, or bowel urgency was between 1% and 8% across all treatments at 3 years. Nonetheless, the odds of bowel urgency at 3 years were lower for RP than for EBRT (3% vs 7%; OR, 0.3 [0.2-0.6]) and for RP vs AS (3% vs 5%; OR, 0.5 [0.3-0.9]).
Androgen deprivation therapy (using agents such as leuprolide, goserelin, or bicalutamide, among others) is often part of the treatment of localized PCa, particularly in men undergoing EBRT. These agents can cause a number of distinct symptoms, such as fatigue, decreased libido, and hot flashes, to name a few. These patient-reported symptoms are captured in the hormone domain score of the EPIC. Not surprisingly, we found that hormone domain scores were worse for EBRT compared with AS and RP at 6 months (RP vs EBRT: 5.0 points [3.3-6.6]; EBRT vs AS: −6.5 points [−11.1 to −1.9]), but these differences are no longer significant at 3 years on unadjusted or adjusted analyses (Table 4, Figure 2I, and Figure 2J). In the models that separated EBRT into with and without ADT, the only group with decrements in hormone function was the EBRT + ADT group, and these associations were limited to the first year (Figure 3).
Median follow-up time among censored patients was 40 months. There were 78 deaths, including 3 PCa deaths. On Kaplan-Meier analysis, estimated 3-year disease-specific survival was similar across groups (99.7%-100%). Unadjusted 3-year overall survival was slightly better for RP (99% [98%-99%]) compared with other groups (EBRT: 96% [94%-98%]; AS: 97% [95%-99%]; P < .001), commensurate with the younger age and lower comorbidity of men undergoing RP.
Identify patient-level characteristics that may influence comparative effectiveness (in this case, race/ethnicity).
As proposed in the original protocol, we sought to identify patient-level characteristics that might influence comparative effectiveness. We planned to explore the relationship of race/ethnicity, age, comorbidity count, and disease risk strata on outcomes in our study population. All these variables were included in the analyses for Objective 1. As shown in Figure 4, all these variables were statistically associated with outcomes in various health-related quality of life domains. Given these observations, we are currently developing more detailed models to specifically explore the effect of disease risk strata, comorbidity, and age on the comparative effectiveness of various therapies for PCa. We expect to complete these models in the next 6 months and prepare numerous manuscripts to disseminate our findings to the research community. We have completed the more detailed analysis of the relationship between race/ethnicity and patient-reported outcomes, and we present those findings below.
Forest Plots Showing the Covariate-Adjusted Effect of Treatment, Age, Race/Ethnicity, and D'Amico Risk Stratum on Domain Score.
Among the 2550 CEASAR participants in the analytic cohort, we identified participants who self-reported their race/ethnicity as White, African American, or Hispanic. Given the relatively small number of participants who self-identified in other racial/ethnic groups, we excluded those men from this analysis. Of the remaining 2338 participants, 1835 (78.5%) were White, 324 (13.9%) were African American, and 179 (7.7%) were Hispanic. Table 5 presents the distributions of selected demographic, socioeconomic, and clinical characteristics by race/ethnicity. In general, White patients had a higher level of educational attainment and were more likely to be married than African American and Hispanic men. Hispanic and African American men were more likely to be uninsured or insured by Medicaid and more likely to make less than $30 000 per year. African American men were more likely to harbor high-risk disease according to the D'Amico criteria and were more likely to receive an open rather than robotic prostatectomy. Baseline function also varied significantly by race and ethnicity (Table 6). African American and Hispanic men reported lower EPIC domain scores for sexual function at baseline than did White patients, and Hispanic patients reported lower scores for urinary irritative symptoms and the urinary incontinence domain. We noted no clinically significant differences in the baseline domain scores for bowel function.
Pretreatment Demographic and Clinical Characteristics by Race/Ethnicity.
Pretreatment Function Quartiles by Race/Ethnicity.
Overall, RP was associated with lower adjusted mean scores for urinary incontinence compared with AS and EBRT. The adjusted mean score for urinary incontinence was 19.9 (95% CI, 17.2-22.7; P < .001) points lower for RP compared with AS and 21.9 (95% CI, 19.2-24.6; P < .001) points lower compared with EBRT. While we observed this association between RP and incontinence across all race/ethnic groups, the decline was larger for African American men than for Whites (adjusted DID: 8.4 points; 95% CI, 2.0-14.8; P = .011) (Table 7). Despite this result, baseline function and primary treatment appeared far more important than race/ethnicity in predicting posttreatment urinary incontinence (Figure 5).
Adjusted Mean Differences in Function at 12 Months Between Treatment Types by Race/Ethnicity.
Proportion of R2 Explained by Different Factors and Interactions.
Because African American men reported a greater decline than White men in urinary incontinence function after RP, we tested whether African American men had greater odds of reporting moderate or severe bother secondary to overall urinary function compared with White men after RP (data not shown). There were no apparent between-race differences in the odds of moderate or severe bother by overall urinary function, despite lower scores on the continuous domain (P = .152).
There was no evidence of any clinically significant differences by race/ethnicity in the effects of treatment on EPIC scores for sexual, bowel, or hormone function (Table 7) or for bother scores in these domains (data not shown). There were slight differences between White and African American men in urinary irritative symptoms following RP, but these differences were small and not likely to be clinically important (DID = 4.4; 95% CI, 0.8-8.0; P = .016) (Table 7).
Because there was evidence of differential adoption of modern treatment modalities among minority populations, we performed a sensitivity analysis excluding patients who did not undergo a robotic nerve-sparing operation and those who did not receive IMRT. The results of this analysis were similar to those of our main analysis with respect to racial differences in treatment effects. However, we found an even larger decline in the post-RP incontinence domain among African American patients compared with Whites (14.1 points; 95% CI, 5.4-22.9; P = .002).
Because this is a nonrandomized clinical trial, we performed a second sensitivity analysis using propensity score adjustment as an alternative method to account for pretreatment differences between patients receiving different treatments. After propensity score adjustment, we did not note any substantial differences from our primary results.
Assess how the comparative effectiveness of the various therapies varies by the quality of care received.
In an effort to determine how the quality of care impacted the effectiveness of therapy, we elected to validate 6 commonly used measures of quality of care in PCa. Our reasoning was that if these 6 measures could not be validated, it would be difficult, if not impossible, to assess the relationship between the true quality of care and the effectiveness of therapy. For this initial analysis, we included the 2601 men in CEASAR who completed a 12-month survey and for whom full medical chart abstraction was performed. Compliance with the 6 quality measures studied ranged from 63.8% to 87.5% (Table 8).
Compliance With Quality Measures.
We calculated baseline EPIC-26 domain summary scores for measure-compliant and measure-noncompliant groups. For most measures, baseline scores were similar among groups; however, mean baseline scores in the EPIC-26 sexual domain for measure 2 (ADT for high-risk patients) were markedly lower in the compliant group (mean 47 vs 74; P = .001). This could be expected, as patients with erectile dysfunction at baseline might be more likely to be offered or to accept ADT without concerns about its adverse effects on libido and sexual performance. There were small but statistically significant differences in EPIC-26 bowel for measure 1 (bone scan avoidance in low-risk patients) and in EPIC-26 sexual for measure 5 (discussion of treatment options).
In 21 of the 24 multivariable models predicting 1-year EPIC-26 domain scores, compliance with the quality measure was not associated with any difference in functional outcome (Figure 6). The remaining 3 models showed small magnitude but statistically significant effects. EPIC bowel scores were slightly better among patients whose physicians complied with measure 5 (documentation of the discussion of treatment options; β = 1.69; P = .01). EPIC-26 sexual scores were slightly worse among patients whose physicians were compliant with measure 1 (avoidance of bone scan in low-risk patients; β – 4.6; P = .04) and slightly better for patients whose physicians were compliant with measure 4 (documentation of disease characteristics prior to treatment; β = 2.93; P = .04). The forest plot (Figure 6) displays the difference in EPIC domain score by quality measure compliance with 95% CI.
Relationship Between Quality Measure Compliance and EPIC-26 Instrument Summary Scores.
We performed subgroup analyses using age, race, and income level, with interaction terms between these variables and measure compliance (data not shown). Of these subgroups, patients with an annual income less than $50 000 were found to have a significant reduction in EPIC-26 sexual domain score with compliance with adjuvant ADT with EBRT in high-risk patients (β – 23.85; P = .001). This may be related to the known impact of ADT on sexual function, although it is not clear why the effect is less significant for men with higher income.
We repeated multivariable linear regression analyses with satisfaction composite scores as the outcome measure (Figure 7). Compliance with quality measures did not have a significant effect on composite satisfaction scores at 1 year.
Relationship Between Quality Measure Compliance and Service Satisfaction Scale for Cancer Care Satisfaction Scale Scores.
We calculated treatment-related complication rates in the early and delayed periods. We performed logistic regression, with odds ratios calculated to summarize the likelihood of complication with measure compliance. Documented complication rates were low overall, with 19 occurring within 30 days and 120 occurring within 1 year. There was no significant association between measure compliance and complication rates (Figure 8). This finding must be tempered owing to small-sample bias affecting the maximum likelihood estimation in the model.
Relationship Between Quality Measure Compliance and Treatment-Related Complications.
Few opportunities exist to investigate comparative functional outcomes after RP, external beam radiotherapy, or AS for localized PCa in a population-based cohort. As of May 2018, only a single randomized clinical trial has compared these 3 treatment options with 10-year follow-up, and that study failed to show a statistically significant difference in survival among the 3 treatments.16 Also, the study has been criticized for including primarily low-risk patients, focusing on older therapeutic modalities, and having inadequate follow-up. Due to a lack of information, many patients still view the 3 options as being relatively equivalent in terms of cancer control and cure.23 To fill this void, studies that focus on patient-reported outcomes using modern treatment techniques are desperately needed. In the CEASAR study, which began in 2011-2012, most RP patients underwent robotic surgery, most radiation patients underwent IMRT, and most AS patients were monitored with repeated testing. The results from CEASAR provide critical new information regarding modern therapeutic approaches that will help patients and providers make what is generally considered to be a highly preference-sensitive clinical decision. In addition, the study results demonstrate the frequency and severity of adverse effects of contemporary treatments, which provides a basis for shared decision-making regarding treatment. Second, in contrast to previously published studies,17,32,58,59 CEASAR may be more generalizable, as the cohort is racially diverse and population based, and includes a range of disease severity. Finally, results from this study may inform future research on personalized risk assessment, tools to facilitate shared decision-making, and other patient-centered outcomes.
Erectile dysfunction (erection insufficient for intercourse) was common (39%-56% at baseline and 51%-71% at 3 years) and declined in all groups. However, RP was associated with clinically significant decrements in sexual function compared with EBRT and AS. Urinary incontinence scores also declined significantly after surgery compared with EBRT and AS: 14% of RP patients reported a moderate or large problem with urinary leakage at 3 years, compared with 5% with EBRT and 6% with AS. There was a benefit to RP compared with AS for irritative voiding symptoms, which met the threshold for clinical significance. Mean scores in bowel and hormonal domains were significantly worse for EBRT vs RP and AS at 6 months, but by 3 years the differences were below the threshold for clinical significance. In addition to treatment, other factors associated with this outcome were baseline domain scores, disease risk stratum (a surrogate for aggressiveness of the intervention), and age. This information may facilitate patient counseling regarding the expected harms of contemporary treatments.
Previous studies have quantified the harms of PCa treatment. Chen and colleagues60 performed a population-based observational study of 1141 men diagnosed with localized PCa in North Carolina in 2014-2015. Participants in this analysis received 1 of 4 therapies: (1) AS (n = 314); (2) RP (n = 469); (3) external beam radiotherapy (n = 249); or (4) interstitial brachytherapy (n = 109). Follow-up was at 2 years and patient-reported outcomes were assessed using the Prostate Cancer Symptom Indices. Their analytic approach was substantially different from ours in that they handled differences in baseline function in the calculation of their propensity scores (as opposed to treating it as a dependent variable in the model as we did) and they compared each of the aggressive therapies with AS (as opposed to comparing the aggressive therapies with one another). While this approach resulted in some minor differences, the findings from Chen and colleagues60 generally concur with those from CEASAR.
A randomized clinical trial would eliminate many of the limitations associated with observational studies such as ours and that of Chen and colleagues.60 However, randomized trials in localized PCa have been difficult to execute, and those that have been completed focused on outmoded treatments; enrolled too few minority patients; lacked a range of disease severity; failed to collect baseline functional assessments; or included a preponderance of elderly, infirm patients or low-risk patients, for whom treatment is questionable.17,58,59 The ProtecT trial, for example, included 99% White patients and nearly 80% Gleason 6 (low-risk) patients. In ProtecT, 87% of surgical patients underwent open RP (compared with 76% robotic in our study), and patients undergoing EBRT had 3-dimensional conformal RT plus ADT (compared with 81% IMRT, with 45% receiving concurrent ADT in our study).16,17 Thus, ProtecT study findings may be difficult to apply to a racially diverse population with a range of disease risk strata, managed with contemporary treatments.
Despite these caveats, the ProtecT trial can provide some context for our study. For example, ProtecT and our study both reported erections insufficient for intercourse at 3 years as 59% vs 51% for AS, 79% vs 70% for RP, and 66% vs 71% for EBRT. Urinary incontinence (moderate or large problem) at 36 months was also similar between ProtecT and this study (7% vs 6% for AS, 12% vs 14% for RP, 4% vs 5% for EBRT). Bowel function (moderate or severe bother) at 36 months was similar for EBRT patients in ProtecT and this study (5% vs 6%), although 7.4% of ProtecT patients reported bloody stools compared with 2% in this study, perhaps reflecting the increased use of IMRT.17
While outcomes reported here are similar to outcomes in previous studies, comparisons between the CEASAR cohort and similar historical cohorts have shown subtle differences between modern techniques (robotic RP and IMRT) and older techniques (open RP and 3-dimensional conformal RT). For example, erectile function outcomes were slightly better for robotic surgery compared with open RP at 6 months (8.3 points on the EPIC sexual function domain) and 12 months (7.6 points).61 Additionally, bowel domain scores at the 6-month time point were better (by 4.3 points) for modern EBRT compared with EBRT delivered in the 1990s (unpublished data). These data suggest that contemporary treatments have similar associations with functional outcomes, but the magnitude may be slightly less.
When we explored the relationship between race/ethnicity and comparative effectiveness of therapy, we observed that the effect of treatment on patient-reported function did not vary dramatically by race/ethnicity. Only in the urinary incontinence domain did we find any evidence of a significant interaction between race and PCa treatment. African American men experienced a somewhat more pronounced post-RP decline in scores for urinary incontinence compared with Whites, but the corresponding changes in bother scores were not significantly different. We also found that race/ethnicity is not nearly as predictive of function at 1 year as treatment selection and baseline function. These findings are new to the PCa literature and may be leveraged to inform evidence-based treatment decisions.
Only 1 other study has examined the interaction between race/ethnicity and treatment on functional outcomes after PCa treatment. In the Prostate Cancer Outcomes Study (PCOS), the authors demonstrated a significant interaction between race/ethnicity and urinary incontinence.34 However, while African American men in our study reported worse effects of RP related to urinary incontinence compared with White men, African American men in the PCOS reported better domain scores for urinary incontinence after RP. The precise reason for this difference between the studies is unknown, but there are several plausible explanations. First, both surgery and RT have undergone considerable changes since the collection of the PCOS data in 1994. Compared with the PCOS era, use of minimally invasive RP, IMRT, and image-guided radiotherapy is now widespread. These new approaches may affect men of varying races differently than the older treatments did. This theory is supported, in part, by the findings from the sensitivity analysis, in which the race–treatment interaction seems to be even stronger among patients who received robotic surgery or IMRT. Second, the original PCOS did not study AS patients. Having an AS cohort allowed us to estimate treatment effects compared with AS within a particular race (eg, the difference in mean EPIC-26 scores between RP and AS among African American men). Then we were able to formally test the interaction between race and treatment by estimating how the effects of treatment varied by race/ethnicity (eg, the difference in mean EPIC scores between RP and AS among African American men subtracted from the difference in mean EPIC scores between RP and AS among White men, which is the difference in differences). Using this systematic approach, we were able to precisely test the race–treatment interaction for all patient-reported functional outcomes after PCa treatment.
Finally, we had hoped to inform patients and providers regarding how the quality of care delivered might impact the comparative effectiveness of therapies for localized PCa, providing patients with critical information regarding which treatment and provider to choose. Unfortunately, we found only a few weak associations between compliance with nationally endorsed quality measures and patient-centered outcomes. Although some associations were statistically significant, we found no discernable pattern of association between compliance and improved outcomes, and the magnitude of these differences was not clinically important, according to published thresholds for clinically detectable change.42 Thus, we were not able to achieve our third objective, given the limitations of current quality metrics in PCa.
Results from CEASAR are among the most generalizable generated to date in the field of localized PCa. The mean age of the cohort is 64, which reflects the lower age of diagnosis noted in the PSA-screening era. The population itself is ethnically diverse, with 25% of patients identifying as minority, including 12% African American, 8% Hispanic, 3% Asian, and 2% other. The socioeconomic status of the population is also quite diverse, with 69% reporting some college education or more. Finally, 55% of the participants were diagnosed with intermediate- or high-risk disease, which is highly reflective of current disease patterns and represents patients who likely have clinically significant illness. The distribution of the various treatment modalities closely reflects that seen in the general US population in both stage and grade.1 Finally, the therapeutic approaches reflect modern techniques, making the information particularly meaningful to men diagnosed in 2017 and beyond.
The study included patients from 5 SEER registries and CaPSURE, an observational disease registry that we included to enrich the sample with patients who received less commonly used therapies, such as cryotherapy or high-intensity focused ultrasound. Not surprisingly, there were significant differences at baseline between the CaPSURE patients and those enrolled at the SEER sites (just as there were significant differences between patients enrolled at different SEER sites). Specifically, the CaPSURE patients were statistically significantly more likely to be White and to be more highly educated, after adjustment for multiple comparisons. To control for these baseline differences by site, we included site of enrollment as an independent variable in all analyses.
Analysis in this area is ongoing using the CEASAR data. We completed an analysis of the relationship between PCa treatment group and race/ethnicity as described. We controlled for various clinical and sociodemographic confounders in the primary analysis, including comorbidity, age, baseline disease stratification, and baseline functional status. Nearly all these covariates were significant, which underscores the importance of the ongoing work to help patients personalize the information from CEASAR to aid in clinical decision-making.
This study has several limitations. First, we did not measure other important outcomes to consider in localized PCa, including long-term functional outcomes and oncologic endpoints, overall quality of life, anxiety, satisfaction, and financial toxicity. Second, some adverse outcomes may present after 3 years, and their number and severity may differ by treatment. Third, we did not include data on patients who had other treatments, such as brachytherapy and ablation, because there were too few of them to generate sufficient statistical power for reliable comparisons. Fourth, aggregated data and average function scores may fail to capture the severity of adverse effects for individuals, and they do not yield personalized risk estimates (eg, accounting for one's baseline function, disease risk, or age). Fifth, we excluded approximately 200 patients (7% of the total cohort) from the analysis, owing to lack of follow-up data. This may have introduced some bias, as the excluded patients may not have been missing at random. To explore this possibility, we compared baseline characteristics between included and excluded patients; the latter were statistically significantly more likely to be African American (22% vs 14%; P = .002). Although we noted several statistically significant differences in various psychosocial scales and baseline functional status, the only clinically meaningful difference was in baseline sexual function: Excluded patients had significantly lower baseline function (54.2 vs 73.3; P < .001). We noted no other statistically significant or clinically meaningful differences. Sixth, the data were clustered by site. Although we took into account within-subject dependency using a generalized estimating equation, site was adjusted only as a fixed effect using 5 dummy variables. While some of the site effects reached statistical significance, none reached the prespecified limits of clinical significance, and the site clusters seem to have only a small effect; however, we cannot completely rule out unexplained site effects that could potentially bias our results. Finally, and perhaps most important, this study used an observational cohort rather than an experimental design, so there may be unmeasured sources of confounding. Nonetheless, we believe the adjustment for numerous known and previously unstudied hypothesized confounders provides relatively valid estimates of functional outcomes.
Much work remains to be done using the CEASAR data. We are currently collecting 60-month patient-reported outcomes, funded by a grant from the Agency for Healthcare Research and Quality. As mentioned earlier, ongoing subgroup analyses will provide useful information to individual groups of patients that will aid in decision-making. We also plan to explore outcomes following lesser-used treatment modalities, including brachytherapy and various combination therapies. Long-term follow-up is needed to understand the comparative effectiveness of therapies in terms of cancer control and survival. Ultimately, however, the information generated in the current study will serve as the basis for novel interventions that can help patients make more informed and personalized clinical decisions regarding treatment for this common disease.
Three-year results from the CEASAR study provide critical patient-centered information to aid in decision-making in localized PCa. We designed the study to help patients understand “what works, in which patients and in whose hands” in this common malignancy. Given the often long life expectancy of men newly diagnosed with localized disease, 3-year outcomes must be considered early to intermediate; however, findings from this study can still prove useful for patients who are making this difficult decision.
In comparing patient-reported outcomes after surgery, radiation, or AS, we found that RP was associated with worse sexual function and urinary incontinence than EBRT or AS after 3 years. There was a benefit to RP compared with AS for urinary irritative voiding symptoms, but no meaningful differences among treatments in bowel or hormonal function beyond 12 months. Subgroup analysis of other potential effect modifiers is ongoing, but we found that the effect of treatment on patient-reported function does not vary dramatically by race/ethnicity. Incontinence 1 year after RP may be worse for African American men, but the difference appears to be modest overall. Finally, we found that commonly used measures of quality in PCa do not correlate with short-term patient-reported outcomes, calling into question the validity of these measures and underscoring the need to find better ways to assess quality in this common condition.
Taken as a whole, these findings may help in counseling men about the comparative effectiveness of contemporary treatments for PCa and ultimately lead to more informed decision-making regarding this common malignancy.
Research reported in this report was [partially] funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (#CE-12-11-4667) Further information available at: https://www.pcori.org/research-results/2013/comparing-effects-surgery-radiation-therapy-and-active-surveillance-men
Penson DF, Barocas DA; CEASAR Investigators. (2019). Comparing the Effects of Surgery, Radiation Therapy, and Active Surveillance on Men With Localized Prostate Cancer—The CEASAR Study. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/1.2020.CE.12114667
The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.
This book is distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License which permits noncommercial use and distribution provided the original author(s) and source are credited. (See https://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
