Development of systematic review
The Genitourinary Cancer Disease Site Group of the Cancer Care Ontario Program in Evidence-Based Care (CCOPEBC) has developed an evidence-based clinical practice guideline on the use of non-hormonal systemic therapy for HRPC using the methodology outlined in the practice guideline development cycle by Browman et al [6
]. The guideline was derived from a systematic review and input from practitioners in Ontario, Canada. This report describes the systematic review portion of the guideline, which addressed the following question: which non-hormonal systemic therapies are most beneficial for the treatment of men with HRPC and clinical evidence of metastases?
Articles were selected for inclusion in the systematic review if they were randomized controlled trials (RCTs) or meta-analyses of RCTs studying a first-line non-hormonal systemic (cytotoxic and non-cytotoxic) therapy in patients with HRPC and metastases that reported on at least one of the following outcomes: overall survival, disease control (i.e., progression-free survival [PFS], time-to-progression [TTP], time-to-treatment failure, objective tumor response, and PSA response), palliative or symptomatic response, quality of life (QoL), and/or toxicity. Previous systematic reviews or evidence-based guidelines that addressed non-hormonal systemic therapy in HRPC were also eligible for inclusion.
Articles were excluded from the systematic review if they studied second-line hormonal therapies, bisphosphonates or radiopharmaceuticals, or randomized fewer than 50 patients per trial arm. When formulating the protocol for this review, the authors were aware of dozens of small randomized trials comparing the activity of various drugs in HRPC. Sometimes these were identified explicitly as phase II trials, but often they were not. Although valuable for identifying potential anti-tumor activity, such trials are by definition underpowered to address the patient outcomes of interest to this review, and often did not report these. Although these limitations could be overcome by statistical pooling, it is also recognized that such RCTs are associated with more variability and are more likely to be reported and published if positive [7
]. Theoretically, positive results from such trials require subsequent confirmation by larger pragmatic RCTs, but this does not always occur. After considering the outcomes of interest for this review, a minimum sample size of 50 randomized patients per trial arm was chosen, in order to be as inclusive as possible while still minimizing inclusion of randomized phase II trials. This sample size was based on a requirement of at least 80% power for an RCT to reliably detect the difference between an endpoint response rate of 10% versus 30% with one-tailed α = 0.05. Reliable assessment of outcomes such as disease-free and overall survival would require even larger sample sizes. RCTs with lesser discriminating ability were considered underpowered and potentially misleading with regard to those endpoints.
Literature search strategy
The MEDLINE (1966 through March 2004), EMBASE (1980 through 2004, week 10), and Cochrane Library (2003, Issue 4) databases (Central Register of Controlled Trials [CCTR] and Database of Systematic Reviews [DSR]), and the conference proceedings of the American Society of Clinical Oncology (1999 through 2004) were searched for abstracts of relevant trials. MEDLINE was searched using the following medical subject headings (MeSH): "prostatic neoplasms", "drug therapy", "antineoplastic agents", and "drug therapy, combination"; and EMBASE was searched using the following Excerpta Medica tree terms: "prostate tumor", "prostate cancer", "drug therapy", "antineoplastic agent", "drug combination", and "combination chemotherapy". In each database, those subject headings were combined with disease and treatment-specific text words (e.g., "prostate cancer", "prostate tumor", "prostate carcinoma", and "chemotherapy"). Those terms were then combined with the search terms for the following publication types and study designs: practice guidelines, systematic reviews, meta-analyses, reviews, randomized controlled trials, and controlled clinical trials. The CCTR and DSR databases were searched using a combination of the aforementioned MeSH and keywords. The reference lists from eligible articles were searched for additional trials, as were the reference lists from relevant review papers.
Methods of review
Citations identified by the literature search strategy were screened for eligibility by four of the authors (EW, DSE, SB, SH) and discrepancies were handled by consensus. Data pertaining to trial design, participants, interventions, and outcomes were extracted from each eligible trial by one reviewer (TW) and were audited by a second reviewer (EW) independently. Information indicative of trial quality, including methods of randomization, absence or degree of blinding, completeness of patient follow-up, and whether statistical analyses were performed by intent-to-treat were also extracted from each trial report. The quality parameters assessed on each trial were used to identify any trials with serious flaws in trial methodology that could lead to potentially misleading results.
Synthesis of the evidence
This review is based on data provided by published reports. When pooling of data is to be undertaken and individual patient data are not available, methods are available for generating summary statistics from published data. However, statistical pooling (meta-analysis) of data from RCTs is only possible if control arms are similar and is only valid if those data are statistically homogeneous. Reports of RCTs studying systemic therapies in HRPC date back 30 years and have studied heterogeneous patient populations, interventions, and outcomes. Numerous drug interventions have been tested, including a variety of single-agent and combination chemotherapy regimens including estramustine phosphate (EMP), and non-cytotoxic drugs such as liarazole, suramin, and atrasentan. What constitutes standard therapy in control arms has been controversial and has included placebo, corticosteroids, EMP, and cytotoxics. On the basis of those observations, large-scale quantitative pooling of RCT data was considered neither possible nor appropriate, and so an interpretive summary of the data was planned. The natural history and management of HRPC has changed in the last three decades; therefore, more contemporary studies were emphasized in the interpretive summary to provide clinicians with the evidence most relevant to current practice. Furthermore, more emphasis on the results of RCTs demonstrating internally consistent benefits in survival, palliation, and quality of life outcomes was planned.