Comparing two methods of meta-analysis in clinical research - individual patient data-based (IPD) and literature-based abstracted data (AD) methods: Analyzing five oncology issues involving more than 10,000 patients in randomized clinical trials (RCTs)
6512 Background: Meta-analyses are one of the highest recommendations levels in Evidence-Based Medicine (EBM). Recently, meta-analyses have increased using either IPD or AD methods. Controversy exists regarding reliability, applicability and feasibility of the different methods to draw conclusions from conflicting RCTs and to estimate magnitude of benefit of different treatments. Methods: As seen in the table , we selected 5 major issues in 3 malignancies subjected to IPD meta-analysis, and then conducted AD meta-analyses from publications of the individual studies, using published methods (Bria, Cancer Treat Rev 2006). We required that >90% of patient numbers for both IPD and AD analyses be available. Event-based relative risk ratios (RRs) with 95% confidence intervals (CI) were derived. Fixed- and random-effect models, and absolute benefits (AB) were calculated. Correlations between IPD Hazard Ratios (HRs) and AD-RRs were estimated using a linear regression model according to Pearson (r) and R2 coefficients (parametric) and Spearman (Rho) coefficient (non- parametric). Results: Results are below. A strong linear correlation exists between IPD-HRs and AD-RRs (r=0.994, R2=0.989; p<0.001; Rho=1.00). Conclusions: The strong correlation supports using high quality meta-analyses with either method to resolve major issues. Differences exist between the methods: IDP is well-suited for sensitivity analyses and for hypothesis generation involving issues not originally anticipated; AD is a practical method allowing EBM to be applied rapidly to major issues in oncology in a timely fashion. No significant financial relationships to disclose. [Table: see text]