- Email: [email protected]
Statistical Design Issues and Other Practical Considerations for Conducting Phase III Prostate Cancer Prevention Trials
0022-5347/04/1712-0064/0 THE JOURNAL OF UROLOGY® Copyright © 2004 by AMERICAN UROLOGICAL ASSOCIATION
Vol. 171, S64 –S67, February 2004 Printed in U.S.A.
DOI: 10.1097/01.ju.0000108260.33590.c8
STATISTICAL DESIGN ISSUES AND OTHER PRACTICAL CONSIDERATIONS FOR CONDUCTING PHASE III PROSTATE CANCER PREVENTION TRIALS CATHERINE M. TANGEN,*,† PHYLLIS J. GOODMAN,† JOHN J. CROWLEY† AND IAN M. THOMPSON‡ From the Southwest Oncology Group Statistical Center, Fred Hutchinson Cancer Research Center, Seattle, Washington, and the University of Texas Health Science Center at San Antonio, San Antonio, Texas (IMT)
ABSTRACT
Purpose: We briefly describe the purpose of phase II studies as a source of preliminary data for phase III prevention trials, and address statistical and other practical considerations for phase III prostate cancer prevention trials. Materials and Methods: Objectives of 2 types of phase II studies and general criteria for validating surrogate end points are described, and statistical considerations needed for planning a phase III prevention trial are explained, including selection of an appropriate study population and end point, the corresponding expected event rate, estimates of loss to followup, and death and compliance rates. If a preventive agent has an impact on the ability to detect prostate cancer, additional study design considerations are then made. Other practical issues are addressed, including collection of biological materials for correlative studies, assessment of quality of life measures and the addition of ancillary studies that may include the collection of additional end points unrelated to prostate cancer. Results: The Southwest Oncology Group is coordinating 2 recently reported phase III prostate cancer prevention trials. The Prostate Cancer Prevention Trial final results of 18,882 men randomized to either finasteride or placebo, and the Selenium and Vitamin E Cancer Prevention Trial has accrued more than half of its goal of 32,400 men. These studies are used to illustrate some statistical design features. Conclusions: The aforementioned trials provide valuable lessons in the successful design and conduct of phase III prostate cancer prevention trials. KEY WORDS: prostatic neoplasms, phase iii clinical trials, surrogate endpoint
We discuss statistical design considerations for phase III prostate cancer prevention trials. Before undertaking a phase III trial, sufficient evidence needs to be collected on a candidate agent in phase II trials. We briefly describe objectives of phase II prevention trials and the role that surrogate end points have in those studies before addressing some of the key statistical issues and other practical design issues for phase III prostate cancer prevention trials. Although phase II data typically are needed before phase III testing of an agent, this is not always the case. For example, in the Selenium and Vitamin E Cancer Prevention Trial (SELECT) there is adequate evidence from secondary analyses of previous phase III studies to indicate that the 2 agents may be effective at preventing prostate cancer.1, 2 However, a definitive phase III study with prostate cancer prevention as the primary objective is needed to confirm the results of secondary analyses of these trials. The 2 types of phase II studies used for cancer prevention trials proof-ofprinciple trials and intermediate-efficacy trials. PHASE II TRIALS
Proof-of-principle phase IIA trials. The primary objective of a proof-of-principle phase IIA study is to evaluate biological
activity by inhibiting, reversing or modulating the transition from normal prostatic epithelium to pre-cancer (eg high grade prostatic intraepithelial neoplasia [HGPIN]) or from pre-cancer to invasive cancer. Agents are currently being developed based on the molecular mechanism of pathogenesis and end point biomarkers are chosen to validate that mechanism.3 The patient population for these studies often consists of men with localized prostate cancer who receive intervention for 4 to 8 weeks before prostatectomy. End points of interest include regression of cancer and HGPIN, and modulation of intermediate biological markers in serum and tissue. This type of study often takes the form of a 2-arm randomized trial of prevention agent versus placebo or observation. It is important to avoid selecting a convenient sample of controls in place of randomization since this approach does not ensure comparability between treatment and control groups. An example of a proof-of principle study involved men with local prostate cancer randomized to selenium or observation for 14 to 31 days before prostatectomy.4 In addition to evaluating the change in tissue and serum selenium concentrations for the 2 groups from baseline to prostatectomy, this study also evaluated toxicities and collected information about any potential effect of selenium on prostate specific antigen (PSA) levels. This verification that selenium did not alter PSA levels was useful when designing the SELECT. Intermediate-efficacy phase IIB trials. The objective of a phase IIB intermediate-efficacy cancer prevention trial is to evaluate whether the agent will significantly modulate the intermediate end point biomarker. This type of trial provides
Supported by National Institutes of Health grants CA-38926 and CA-32102. * Correspondence: 1100 Fairview Ave. N., MP-557, P.O. Box 19024, Seattle, Washington 98109-1024 (telephone: 206-667-4623; FAX: 206667-4408; e-mail: [email protected]). † Financial interest and/or other relationship with Merck. ‡ Financial interest and/or other relationship with AstraZeneca, Mission Pharmacol and Abbott. S64
STATISTICAL DESIGN ISSUES OF PHASE III PREVENTION TRIALS OF PROSTATE CANCER
preliminary efficacy estimates for a given dose of an agent in addition to information about adverse events, and it can be used to plan the subsequent phase III study. Randomization between the agent and placebo is essential to ensuring comparability. Because this type of trial needs to be completed in a timely fashion, ideal participants are men at risk for prostate cancer, including those with a family history of prostate cancer, increased PSA or abnormal digital rectal examination (DRE) but negative biopsy, or HGPIN. For a study to be conducted quickly, surrogate end points need to be used. Trock described criteria for a good surrogate endpoint, including 1) modulation by the preventive agent, 2) a strong correlation with prostate cancer incidence, 3) a short latency period relative to the development of cancer, 4) easy and reliable measurement and 5) provision of an accurate estimate of the magnitude of effect that can be expected in the phase III setting when the end point of interest will be prostate cancer incidence.5 Examples of surrogate (or intermediate) end points include prostatic intraepithelial neoplasia, PSA, Ki67, apoptosis and angiogenesis. If there are competing pathways of equal or greater importance than the one involving the specified surrogate end point, the estimate of efficacy from the chemopreventive agent will be biased and it may underestimate the magnitude of the true potential of the agent to reduce prostate cancer.5 A challenge in identifying appropriate surrogate markers is the unlikelihood that 1 surrogate marker will be applicable for multiple preventive agents. Each agent will have its own mechanism of action that may or may not target a specific surrogate marker. Additionally, each surrogate marker must be validated, which involves collecting the prostate cancer end point. In effect, a phase III trial is being conducted to validate each surrogate marker. PHASE III DESIGN CONSIDERATIONS
Before a phase III trial can be designed, adequate preliminary data indicating biological activity of a preventive agent in prostate cancer must have been collected. In part, phase II testing is conducted to provide answers to several questions before designing the phase III trial. 1) Does the agent have an impact on the detection of prostate cancer? For example, does it reduce the size or histological features of the prostate or affect PSA or some other diagnostic test? 2) Are there specific subpopulations that would be best suited for testing the agent?
S65
Ideally, an effective prevention agent will improve survival. However, that end point would take a massive sample size and many years to evaluate, making a study designed to evaluate survival not feasible to conduct. Instead, investigators typically must choose end points related to incidence/ period prevalence of cancer. If the agent itself impacts the ability to detect prostate cancer, then some type of mandated end of study (EOS) biopsy for all participants is probably the most reasonable solution because of a concern about a potential detection bias. For example, before planning the Prostate Cancer Prevention Trial (PCPT) investigators considered data from a previous phase III trial evaluating finasteride in men with benign prostatic hyperplasia which showed that finasteride reduced PSA levels and decreased prostate gland volume, which may influence subsequent biopsy rates and rates of prostate cancer detection.6 When the agent interferes with ascertainment, there is concern that interim prostate cancer cases may be identified with bias during the trial. It is difficult to specify a priori interim rules for terminating the study due to perceived extreme efficacy results based on interim prostate cancer cases. EOS biopsy results are expected to provide less biased estimates of treatment efficacy, which would give them more weight than the interim results. Designing a study that requires an EOS biopsy presents a number of challenges.7 A criticism of EOS biopsies as an endpoint is that the study may be detecting cancers that are not clinically relevant, and a reduction in these types of cases may not have a survival impact. A drawback of EOS biopsies is that subjects have to agree to undergo this procedure and urologists have to agree to perform the biopsy when there may be no clinical indication. Estimates of EOS biopsy compliance rates must be carefully considered, because sample size estimates are greatly affected by this assumption.8 Factors influencing the EOS biopsy rate include death, loss to followup, medical contraindications and refusal. Despite these potential concerns, when an agent interferes with identification of prostate cancer, an EOS biopsy is the best strategy for minimizing bias. Phase III studies for which the agent does not impact the detection of prostate cancer (eg SELECT9) may use clinically detected prostate cancer as an end point. Because there is no evidence that selenium or vitamin E affects the ability to diagnose prostate cancer, the primary end point of SELECT is time to clinically detected prostate cancer, and there is no EOS biopsy. Community surveillance standards are applied
Effect of expected incidence rate on statistical power. Assumptions are similar to PCPT except for reduced sample size for illustrative purpose. For sake of simplicity, compliance is not incorporated. Two-sided ␣ ⫽ 0.05, 4,000 men per arm, 60% known end point status and 25% reduction in prostate cancer on treated arm.
S66
STATISTICAL DESIGN ISSUES OF PHASE III PREVENTION TRIALS OF PROSTATE CANCER
with respect to PSA and DRE monitoring with the recommendation for a biopsy in the event of increased PSA and/or suspicious DRE findings. Regardless of the primary end point, survival information should also be collected and centrally reviewed to identify prostate cancer deaths. Although this may require substantial effort in obtaining additional death documentation, it is necessary to get unbiased cause of death information. In most cases a certain amount of information regarding other cancer diagnoses and cardiovascular end points is critical for a thorough evaluation of agent toxicity and safety. Deciding on the target population will define a number of important study design features, including recruitment strategies, expected prostate cancer rate, duration of the study and generalizability of results. For example, since men with HGPIN have a greater risk of prostate cancer, a smaller sample size is needed, and the study can theoretically be completed more quickly. The figure illustrates how the event rate can affect statistical power. However, recruiting a specific subpopulation can be challenging, and results may not be generalizable to the larger population of men at risk for prostate cancer. It is important to recognize that men who are at risk for prostate cancer are older, and they may be taking a substantial amount of medication for other conditions during a prevention trial. Possible interactions between medications and the prevention agent must be considered because of safety concerns and the potential impact on compliance rates. Although appropriate exclusion criteria should be incorporated into the protocol, they do not address circumstances that may develop during the course of the trial. The fact that men who are candidates for a prostate cancer prevention trial are at risk of dying of competing causes, such as cardiovascular disease and other cancers, must be considered in sample size calculations. Typically, published race and age specific mortality rates are used to account for expected numbers of deaths during a study. These estimates may overstate the expected number of deaths since men who volunteer for prevention trials tend to be healthier than those in the general population, although this strategy tends to be safer and more conservative. When calculating sample size, it is desirable to use a statistical program that will allow for specifications of hazard of death over time, since risk typically starts low and gets higher as the men age on the study. Phase III prevention trials can take a number of years to complete. While on study some men will stop taking the active agent to which they were randomized (drop-outs) and others who were randomized to placebo will start taking the active agent (drop-ins). Both of these rates need to be minimized to the greatest extent possible because a reduction in compliance decreases the statistical power to detect a treatment effect. It is important to plan compliance monitoring during the course of a prevention trial. In the PCPT serum dihydrotestosterone (DHT) was measured annually on a small subset of men. Because finasteride substantially reduces DHT levels, this is a convenient way to evaluate study drug compliance. If the Data and Safety Monitoring Committee had found during the accrual portion of PCPT that a priori assumptions regarding compliance were not accurate based on the DHT data, updated sample size calculations would have been recommended. Pill counts are often used as a simple method of measuring compliance, but accurate collection of this information can be difficult.10 Designing a trial with adequate statistical power is crucial. If the top accruing site from PCPT (1,400 participants) had chosen to conduct its own version of this study, there would have been only a 20% power to detect a 25% reduction in prostate cancer with finasteride. If 5 similar sized institutions each conducted its own PCPT, an incorrect conclusion that finasteride was not effective would have been reached
94% of the time (ie 3 or more of the 5 trials would have provided nonsignificant tests of treatment). One adequately powered study is far superior to multiple under powered studies. Participants of prevention clinical trials tend to be healthy or at least asymptomatic. When a patient has a disease, he may be willing to trade some aspects of quality of life (QOL) for the possibility of a treatment benefit. However, this tradeoff decision may be quite different in healthy men who are free of disease. For this reason, it is important to evaluate any changes in QOL which may potentially be attributed to the prevention agent. Use of a standardized instrument, such as the Medical Outcomes Study-36 Item Short Form that has been shown to have good properties in healthy individuals, is a useful approach.11 Obtaining complete longitudinal QOL assessment can be challenging, especially for studies that may last for many years. Strategies and incentives for QOL compliance should be planned for, and frequency of administration and length of instrument need to be considered. It may be adequate to evaluate QOL end points in a representative subset of patients instead of all randomized patients. It is important to think to the future when designing a phase III prevention trial of prostate cancer. There will be hypotheses of interest, laboratory and genetic tests available, and other considerations at the end of the study which were not even imagined at the beginning of the study. Adequate collection of serum, tissue and white blood cells should be planned for in the protocol. It is important to describe the potential future conduct of currently unspecified tests in the informed consent and to accurately collect the subject’s level of consent. For many prevention trials sample sizes are so large and mechanisms of prevention tested are so general that hypotheses can be developed which assess whether the specific prevention strategy affects the risk of other diseases or other neoplasms. For this reason, it is wise to consult with physicians in other specialties regarding additional potential hypotheses. For example, SELECT is also collecting end point information on colorectal and lung cancers.9 Serious consideration should be given to the prospective collection of such end points and biological samples for development of new surrogates. For some end points a small increase (or no change) in sample size may allow an adequately powered secondary analysis of another disease end point. Finally, phase III prostate cancer prevention trials represent a unique opportunity for ancillary studies targeting a healthy aging male population. The existing study management structure is an ideal environment for additional research at minimal cost. However, there are issues that must be addressed before adding an ancillary study. A commitment is needed to both studies from the investigators and clinical research associates at the local site and the participants. Without such a commitment, 1 study could adversely affect the accrual or compliance to the other study. Additional funding is also needed to cover extra effort for recruiting and conducting the study. Another consideration is whether the data collection will be too burdensome for the participant or clinical research associates to complete. Complications with institutional review boards can be expected when ancillary studies are added.12
CONCLUSIONS
The PCPT and SELECT are 2 examples of phase III prostate cancer prevention trials. The design of the PCPT requires an EOS biopsy due to the effect of finasteride on prostate cancer detection. The SELECT uses time to clinically diagnosed prostate cancer as the primary end point. Each study provides useful design information that should be examined when planning future phase III trials.
STATISTICAL DESIGN ISSUES OF PHASE III PREVENTION TRIALS OF PROSTATE CANCER REFERENCES
1. Clark, L. C., Dalkin, B., Krongrad, A., Combs, G. F., Jr., Turnbull, B. W., Slate, E. H. et al: Decreased incidence of prostate cancer with selenium supplementation: results of a double-blind cancer prevention trial. Br J Urol, 81: 730, 1998 2. The Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group: The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. N Engl J Med, 330: 1029, 1994 3. Lieberman, R.: Prostate cancer chemoprevention: strategies for designing efficient clinical trials. Urology, suppl 4A, 57: 224, 2001 4. Sabichi, A. L., Lee, J. J., Taylor, R. J., Thompson, I. M., Jr., Miles, B. J., Basler, J. W. et al: Selenium accumulates in prostate tissue of prostate cancer patients after short-term administration of l-selenomethionine. Proc Am Assoc Cancer Res, 43: 1007, abstract 4993, 2002 5. Trock, B. J.: Validation of surrogate endpoint biomarkers in prostate cancer chemoprevention trials. Urology, suppl., 57: 241, 2001 6. Gormley, G. J., Stoner, E., Bruskewitz, R. C., Imperato-McGinley, J., Walsh, P. C., McConnell, J. D. et al:
7.
8.
9. 10.
11. 12.
S67
The effect of finasteride in men with benign prostatic hyperplasia. J Urol, 167: 1102, 2002 Goodman, P. J., Tangen, C. M., Crowley, J. J., Carlin, S., Ryan, A., Thompson, I. M. et al: Implementation of the Prostate Cancer Prevention Trial (PCPT). Unpublished data Feigl, P., Blumenstein, B., Thompson, I., Crowley, J., Wolf, M., Kramer, B. S. et al: Design of the Prostate Cancer Prevention Trial (PCPT). Control Clin Trials, 16: 150, 1995 Klein, E. A., Thompson, I. M., Lippman, S. M., Goodman, P. J., Albanes, D., Taylor, P. R. et al: SELECT: the next prostate cancer prevention trial. J Urol, 166: 1311, 2001 Pauler, D. K., Gower, K. B., Goodman, P. J., Crowley, J. J. and Thompson, I. M.: Biomarker-based methods for determining noncompliance in a prevention trial. Control Clin Trials, 23: 675, 2002 Sloan, J. A. and Varricchio, C.: Quality of life endpoints in prostate chemoprevention trials. Urology, suppl., 57: 235, 2001 Chay, T., Goodman, P. and Hartline, J. A.: Challenges of implementing ancillary studies within multi-site clinical trials. Unpublished data
Vol. 171, S64 –S67, February 2004 Printed in U.S.A.
DOI: 10.1097/01.ju.0000108260.33590.c8
STATISTICAL DESIGN ISSUES AND OTHER PRACTICAL CONSIDERATIONS FOR CONDUCTING PHASE III PROSTATE CANCER PREVENTION TRIALS CATHERINE M. TANGEN,*,† PHYLLIS J. GOODMAN,† JOHN J. CROWLEY† AND IAN M. THOMPSON‡ From the Southwest Oncology Group Statistical Center, Fred Hutchinson Cancer Research Center, Seattle, Washington, and the University of Texas Health Science Center at San Antonio, San Antonio, Texas (IMT)
ABSTRACT
Purpose: We briefly describe the purpose of phase II studies as a source of preliminary data for phase III prevention trials, and address statistical and other practical considerations for phase III prostate cancer prevention trials. Materials and Methods: Objectives of 2 types of phase II studies and general criteria for validating surrogate end points are described, and statistical considerations needed for planning a phase III prevention trial are explained, including selection of an appropriate study population and end point, the corresponding expected event rate, estimates of loss to followup, and death and compliance rates. If a preventive agent has an impact on the ability to detect prostate cancer, additional study design considerations are then made. Other practical issues are addressed, including collection of biological materials for correlative studies, assessment of quality of life measures and the addition of ancillary studies that may include the collection of additional end points unrelated to prostate cancer. Results: The Southwest Oncology Group is coordinating 2 recently reported phase III prostate cancer prevention trials. The Prostate Cancer Prevention Trial final results of 18,882 men randomized to either finasteride or placebo, and the Selenium and Vitamin E Cancer Prevention Trial has accrued more than half of its goal of 32,400 men. These studies are used to illustrate some statistical design features. Conclusions: The aforementioned trials provide valuable lessons in the successful design and conduct of phase III prostate cancer prevention trials. KEY WORDS: prostatic neoplasms, phase iii clinical trials, surrogate endpoint
We discuss statistical design considerations for phase III prostate cancer prevention trials. Before undertaking a phase III trial, sufficient evidence needs to be collected on a candidate agent in phase II trials. We briefly describe objectives of phase II prevention trials and the role that surrogate end points have in those studies before addressing some of the key statistical issues and other practical design issues for phase III prostate cancer prevention trials. Although phase II data typically are needed before phase III testing of an agent, this is not always the case. For example, in the Selenium and Vitamin E Cancer Prevention Trial (SELECT) there is adequate evidence from secondary analyses of previous phase III studies to indicate that the 2 agents may be effective at preventing prostate cancer.1, 2 However, a definitive phase III study with prostate cancer prevention as the primary objective is needed to confirm the results of secondary analyses of these trials. The 2 types of phase II studies used for cancer prevention trials proof-ofprinciple trials and intermediate-efficacy trials. PHASE II TRIALS
Proof-of-principle phase IIA trials. The primary objective of a proof-of-principle phase IIA study is to evaluate biological
activity by inhibiting, reversing or modulating the transition from normal prostatic epithelium to pre-cancer (eg high grade prostatic intraepithelial neoplasia [HGPIN]) or from pre-cancer to invasive cancer. Agents are currently being developed based on the molecular mechanism of pathogenesis and end point biomarkers are chosen to validate that mechanism.3 The patient population for these studies often consists of men with localized prostate cancer who receive intervention for 4 to 8 weeks before prostatectomy. End points of interest include regression of cancer and HGPIN, and modulation of intermediate biological markers in serum and tissue. This type of study often takes the form of a 2-arm randomized trial of prevention agent versus placebo or observation. It is important to avoid selecting a convenient sample of controls in place of randomization since this approach does not ensure comparability between treatment and control groups. An example of a proof-of principle study involved men with local prostate cancer randomized to selenium or observation for 14 to 31 days before prostatectomy.4 In addition to evaluating the change in tissue and serum selenium concentrations for the 2 groups from baseline to prostatectomy, this study also evaluated toxicities and collected information about any potential effect of selenium on prostate specific antigen (PSA) levels. This verification that selenium did not alter PSA levels was useful when designing the SELECT. Intermediate-efficacy phase IIB trials. The objective of a phase IIB intermediate-efficacy cancer prevention trial is to evaluate whether the agent will significantly modulate the intermediate end point biomarker. This type of trial provides
Supported by National Institutes of Health grants CA-38926 and CA-32102. * Correspondence: 1100 Fairview Ave. N., MP-557, P.O. Box 19024, Seattle, Washington 98109-1024 (telephone: 206-667-4623; FAX: 206667-4408; e-mail: [email protected]). † Financial interest and/or other relationship with Merck. ‡ Financial interest and/or other relationship with AstraZeneca, Mission Pharmacol and Abbott. S64
STATISTICAL DESIGN ISSUES OF PHASE III PREVENTION TRIALS OF PROSTATE CANCER
preliminary efficacy estimates for a given dose of an agent in addition to information about adverse events, and it can be used to plan the subsequent phase III study. Randomization between the agent and placebo is essential to ensuring comparability. Because this type of trial needs to be completed in a timely fashion, ideal participants are men at risk for prostate cancer, including those with a family history of prostate cancer, increased PSA or abnormal digital rectal examination (DRE) but negative biopsy, or HGPIN. For a study to be conducted quickly, surrogate end points need to be used. Trock described criteria for a good surrogate endpoint, including 1) modulation by the preventive agent, 2) a strong correlation with prostate cancer incidence, 3) a short latency period relative to the development of cancer, 4) easy and reliable measurement and 5) provision of an accurate estimate of the magnitude of effect that can be expected in the phase III setting when the end point of interest will be prostate cancer incidence.5 Examples of surrogate (or intermediate) end points include prostatic intraepithelial neoplasia, PSA, Ki67, apoptosis and angiogenesis. If there are competing pathways of equal or greater importance than the one involving the specified surrogate end point, the estimate of efficacy from the chemopreventive agent will be biased and it may underestimate the magnitude of the true potential of the agent to reduce prostate cancer.5 A challenge in identifying appropriate surrogate markers is the unlikelihood that 1 surrogate marker will be applicable for multiple preventive agents. Each agent will have its own mechanism of action that may or may not target a specific surrogate marker. Additionally, each surrogate marker must be validated, which involves collecting the prostate cancer end point. In effect, a phase III trial is being conducted to validate each surrogate marker. PHASE III DESIGN CONSIDERATIONS
Before a phase III trial can be designed, adequate preliminary data indicating biological activity of a preventive agent in prostate cancer must have been collected. In part, phase II testing is conducted to provide answers to several questions before designing the phase III trial. 1) Does the agent have an impact on the detection of prostate cancer? For example, does it reduce the size or histological features of the prostate or affect PSA or some other diagnostic test? 2) Are there specific subpopulations that would be best suited for testing the agent?
S65
Ideally, an effective prevention agent will improve survival. However, that end point would take a massive sample size and many years to evaluate, making a study designed to evaluate survival not feasible to conduct. Instead, investigators typically must choose end points related to incidence/ period prevalence of cancer. If the agent itself impacts the ability to detect prostate cancer, then some type of mandated end of study (EOS) biopsy for all participants is probably the most reasonable solution because of a concern about a potential detection bias. For example, before planning the Prostate Cancer Prevention Trial (PCPT) investigators considered data from a previous phase III trial evaluating finasteride in men with benign prostatic hyperplasia which showed that finasteride reduced PSA levels and decreased prostate gland volume, which may influence subsequent biopsy rates and rates of prostate cancer detection.6 When the agent interferes with ascertainment, there is concern that interim prostate cancer cases may be identified with bias during the trial. It is difficult to specify a priori interim rules for terminating the study due to perceived extreme efficacy results based on interim prostate cancer cases. EOS biopsy results are expected to provide less biased estimates of treatment efficacy, which would give them more weight than the interim results. Designing a study that requires an EOS biopsy presents a number of challenges.7 A criticism of EOS biopsies as an endpoint is that the study may be detecting cancers that are not clinically relevant, and a reduction in these types of cases may not have a survival impact. A drawback of EOS biopsies is that subjects have to agree to undergo this procedure and urologists have to agree to perform the biopsy when there may be no clinical indication. Estimates of EOS biopsy compliance rates must be carefully considered, because sample size estimates are greatly affected by this assumption.8 Factors influencing the EOS biopsy rate include death, loss to followup, medical contraindications and refusal. Despite these potential concerns, when an agent interferes with identification of prostate cancer, an EOS biopsy is the best strategy for minimizing bias. Phase III studies for which the agent does not impact the detection of prostate cancer (eg SELECT9) may use clinically detected prostate cancer as an end point. Because there is no evidence that selenium or vitamin E affects the ability to diagnose prostate cancer, the primary end point of SELECT is time to clinically detected prostate cancer, and there is no EOS biopsy. Community surveillance standards are applied
Effect of expected incidence rate on statistical power. Assumptions are similar to PCPT except for reduced sample size for illustrative purpose. For sake of simplicity, compliance is not incorporated. Two-sided ␣ ⫽ 0.05, 4,000 men per arm, 60% known end point status and 25% reduction in prostate cancer on treated arm.
S66
STATISTICAL DESIGN ISSUES OF PHASE III PREVENTION TRIALS OF PROSTATE CANCER
with respect to PSA and DRE monitoring with the recommendation for a biopsy in the event of increased PSA and/or suspicious DRE findings. Regardless of the primary end point, survival information should also be collected and centrally reviewed to identify prostate cancer deaths. Although this may require substantial effort in obtaining additional death documentation, it is necessary to get unbiased cause of death information. In most cases a certain amount of information regarding other cancer diagnoses and cardiovascular end points is critical for a thorough evaluation of agent toxicity and safety. Deciding on the target population will define a number of important study design features, including recruitment strategies, expected prostate cancer rate, duration of the study and generalizability of results. For example, since men with HGPIN have a greater risk of prostate cancer, a smaller sample size is needed, and the study can theoretically be completed more quickly. The figure illustrates how the event rate can affect statistical power. However, recruiting a specific subpopulation can be challenging, and results may not be generalizable to the larger population of men at risk for prostate cancer. It is important to recognize that men who are at risk for prostate cancer are older, and they may be taking a substantial amount of medication for other conditions during a prevention trial. Possible interactions between medications and the prevention agent must be considered because of safety concerns and the potential impact on compliance rates. Although appropriate exclusion criteria should be incorporated into the protocol, they do not address circumstances that may develop during the course of the trial. The fact that men who are candidates for a prostate cancer prevention trial are at risk of dying of competing causes, such as cardiovascular disease and other cancers, must be considered in sample size calculations. Typically, published race and age specific mortality rates are used to account for expected numbers of deaths during a study. These estimates may overstate the expected number of deaths since men who volunteer for prevention trials tend to be healthier than those in the general population, although this strategy tends to be safer and more conservative. When calculating sample size, it is desirable to use a statistical program that will allow for specifications of hazard of death over time, since risk typically starts low and gets higher as the men age on the study. Phase III prevention trials can take a number of years to complete. While on study some men will stop taking the active agent to which they were randomized (drop-outs) and others who were randomized to placebo will start taking the active agent (drop-ins). Both of these rates need to be minimized to the greatest extent possible because a reduction in compliance decreases the statistical power to detect a treatment effect. It is important to plan compliance monitoring during the course of a prevention trial. In the PCPT serum dihydrotestosterone (DHT) was measured annually on a small subset of men. Because finasteride substantially reduces DHT levels, this is a convenient way to evaluate study drug compliance. If the Data and Safety Monitoring Committee had found during the accrual portion of PCPT that a priori assumptions regarding compliance were not accurate based on the DHT data, updated sample size calculations would have been recommended. Pill counts are often used as a simple method of measuring compliance, but accurate collection of this information can be difficult.10 Designing a trial with adequate statistical power is crucial. If the top accruing site from PCPT (1,400 participants) had chosen to conduct its own version of this study, there would have been only a 20% power to detect a 25% reduction in prostate cancer with finasteride. If 5 similar sized institutions each conducted its own PCPT, an incorrect conclusion that finasteride was not effective would have been reached
94% of the time (ie 3 or more of the 5 trials would have provided nonsignificant tests of treatment). One adequately powered study is far superior to multiple under powered studies. Participants of prevention clinical trials tend to be healthy or at least asymptomatic. When a patient has a disease, he may be willing to trade some aspects of quality of life (QOL) for the possibility of a treatment benefit. However, this tradeoff decision may be quite different in healthy men who are free of disease. For this reason, it is important to evaluate any changes in QOL which may potentially be attributed to the prevention agent. Use of a standardized instrument, such as the Medical Outcomes Study-36 Item Short Form that has been shown to have good properties in healthy individuals, is a useful approach.11 Obtaining complete longitudinal QOL assessment can be challenging, especially for studies that may last for many years. Strategies and incentives for QOL compliance should be planned for, and frequency of administration and length of instrument need to be considered. It may be adequate to evaluate QOL end points in a representative subset of patients instead of all randomized patients. It is important to think to the future when designing a phase III prevention trial of prostate cancer. There will be hypotheses of interest, laboratory and genetic tests available, and other considerations at the end of the study which were not even imagined at the beginning of the study. Adequate collection of serum, tissue and white blood cells should be planned for in the protocol. It is important to describe the potential future conduct of currently unspecified tests in the informed consent and to accurately collect the subject’s level of consent. For many prevention trials sample sizes are so large and mechanisms of prevention tested are so general that hypotheses can be developed which assess whether the specific prevention strategy affects the risk of other diseases or other neoplasms. For this reason, it is wise to consult with physicians in other specialties regarding additional potential hypotheses. For example, SELECT is also collecting end point information on colorectal and lung cancers.9 Serious consideration should be given to the prospective collection of such end points and biological samples for development of new surrogates. For some end points a small increase (or no change) in sample size may allow an adequately powered secondary analysis of another disease end point. Finally, phase III prostate cancer prevention trials represent a unique opportunity for ancillary studies targeting a healthy aging male population. The existing study management structure is an ideal environment for additional research at minimal cost. However, there are issues that must be addressed before adding an ancillary study. A commitment is needed to both studies from the investigators and clinical research associates at the local site and the participants. Without such a commitment, 1 study could adversely affect the accrual or compliance to the other study. Additional funding is also needed to cover extra effort for recruiting and conducting the study. Another consideration is whether the data collection will be too burdensome for the participant or clinical research associates to complete. Complications with institutional review boards can be expected when ancillary studies are added.12
CONCLUSIONS
The PCPT and SELECT are 2 examples of phase III prostate cancer prevention trials. The design of the PCPT requires an EOS biopsy due to the effect of finasteride on prostate cancer detection. The SELECT uses time to clinically diagnosed prostate cancer as the primary end point. Each study provides useful design information that should be examined when planning future phase III trials.
STATISTICAL DESIGN ISSUES OF PHASE III PREVENTION TRIALS OF PROSTATE CANCER REFERENCES
1. Clark, L. C., Dalkin, B., Krongrad, A., Combs, G. F., Jr., Turnbull, B. W., Slate, E. H. et al: Decreased incidence of prostate cancer with selenium supplementation: results of a double-blind cancer prevention trial. Br J Urol, 81: 730, 1998 2. The Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group: The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. N Engl J Med, 330: 1029, 1994 3. Lieberman, R.: Prostate cancer chemoprevention: strategies for designing efficient clinical trials. Urology, suppl 4A, 57: 224, 2001 4. Sabichi, A. L., Lee, J. J., Taylor, R. J., Thompson, I. M., Jr., Miles, B. J., Basler, J. W. et al: Selenium accumulates in prostate tissue of prostate cancer patients after short-term administration of l-selenomethionine. Proc Am Assoc Cancer Res, 43: 1007, abstract 4993, 2002 5. Trock, B. J.: Validation of surrogate endpoint biomarkers in prostate cancer chemoprevention trials. Urology, suppl., 57: 241, 2001 6. Gormley, G. J., Stoner, E., Bruskewitz, R. C., Imperato-McGinley, J., Walsh, P. C., McConnell, J. D. et al:
7.
8.
9. 10.
11. 12.
S67
The effect of finasteride in men with benign prostatic hyperplasia. J Urol, 167: 1102, 2002 Goodman, P. J., Tangen, C. M., Crowley, J. J., Carlin, S., Ryan, A., Thompson, I. M. et al: Implementation of the Prostate Cancer Prevention Trial (PCPT). Unpublished data Feigl, P., Blumenstein, B., Thompson, I., Crowley, J., Wolf, M., Kramer, B. S. et al: Design of the Prostate Cancer Prevention Trial (PCPT). Control Clin Trials, 16: 150, 1995 Klein, E. A., Thompson, I. M., Lippman, S. M., Goodman, P. J., Albanes, D., Taylor, P. R. et al: SELECT: the next prostate cancer prevention trial. J Urol, 166: 1311, 2001 Pauler, D. K., Gower, K. B., Goodman, P. J., Crowley, J. J. and Thompson, I. M.: Biomarker-based methods for determining noncompliance in a prevention trial. Control Clin Trials, 23: 675, 2002 Sloan, J. A. and Varricchio, C.: Quality of life endpoints in prostate chemoprevention trials. Urology, suppl., 57: 235, 2001 Chay, T., Goodman, P. and Hartline, J. A.: Challenges of implementing ancillary studies within multi-site clinical trials. Unpublished data