Review Approach

The methods for this evidence review follow the Methods Guide for the Evidence-based Practice Center (EPC) Program. Appendix A provides more detail on the methods. Appendix B provides the list of excluded and background studies. Appendix C provides more details on the results and Appendix D provides the evidence tables. The topic of this report was developed by the Patient-Centered Outcomes Research Institute (PCORI) in consultation with the Agency for Healthcare Research and Quality (AHRQ). Initially a panel of Key Informants provided input on the Key Questions to be addressed. The Key Questions were posted on AHRQ’s Effective Health Care (EHC) website for public comment for 3 weeks in July 2019, and PCORI conducted a stakeholder call to discuss the Key Questions in August 2019. The EPC revised the questions in response to comments. A panel of Technical Experts provided high-level content and methodological expertise throughout development of the review protocol. Further details regarding expert guidance and review are provided in Appendixes E and F. The final protocol is posted on the EHC website at https://effectivehealthcare.ahrq.gov/products/radiation-brain-metastases/protocol. The PROSPERO registration is CRD42020168260.

Analytic Framework

The analytic framework (Figure 1) outlines the patient population, the interventions, and the outcomes that are addressed in the evidence synthesis.

Figure 1

Analytic framework for radiation therapy for brain metastases. Abbreviations: KQ = Key Question, SRS = stereotactic radiosurgery, WBRT = whole brain radiation therapy

Study Selection

We included randomized controlled trials (RCTs) evaluating radiation therapy, including WBRT and SRS alone or in combination, as initial or postoperative treatment, with or without systemic therapy (immunotherapy, chemotherapy or targeted therapy) for adults with brain metastases due to non-small cell lung cancer, breast cancer, and melanoma. Studies had to report on effects of radiation therapy in the 1990s or later, and we included studies published to July 2020. We also included large (N≥200) clinical controlled trials and cohort studies comparing two cohorts to address adverse effects of the interventions. The details of the sources, search strategies, screening procedure, and the eligibility criteria are described in detail in Appendix A.

Data Extraction and Risk of Bias Assessment

We abstracted study, patient, intervention and comparator details, and documented the results for clinical and patient-centered outcomes as well as adverse events. Publications reporting on the same patient group were consolidated. To facilitate comparisons across studies, we standardized descriptions (e.g., intervention characteristics) and converted study characteristics to proportions. Results were converted to measure-independent variables such as relative risks and standardized mean differences and effect estimates were presented together with 95-percent confidence intervals. Time to event data were analyzed as the hazard ratio.

Risk of bias assessed selection bias and risk of bias arising from the randomization process, performance bias and bias due to deviations from intended interventions, attrition bias and bias due to missing outcome data, detection bias and bias in measurement of the outcome, reporting bias and bias in selection of the reported results, and other sources of bias (lack of use of validated measures). In addition, we evaluated the data collection of adverse events and the reporting of adverse events.

The procedures are described in detail in Appendix A.

Data Synthesis and Analysis

We synthesized the effects of WBRT (Key Question [KQ] 1), SRS (KQ2), post-surgery treatment (KQ3), and any adverse events (KQ4) associated with the interventions. Where outcomes, interventions, and comparators allowed, we determined pooled effects across studies for the following outcomes: overall survival, disease-free survival, deaths due to brain metastases, intracranial progression, quality of life, functional status, cognitive effects, serious adverse events, number of adverse events, radiation necrosis, headaches, fatigue, seizure, and vomiting. We assessed statistical heterogeneity with the I-squared statistic and explored publication bias (Begg, Egger test). To address the subquestions, we conducted meta-regressions to detect effect modifiers such as the role of prognosis and the primary cancer site in indirect analyses across studies. The analytic methods are documented in detail in Appendix A.

Grading the Strength of the Body of Evidence

We formulated evidence statements for the interventions and outcomes of interest. We then graded the strength of evidence to describe our confidence in effect estimates as high, moderate, low, and insufficient evidence. The assessment is based on our analysis of the study limitations, directness, consistency, precision, and reporting bias (see Appendix A for more details).