Abstract:
HTML
Purpose: To develop strategies for sample size estimates in a prospective radiologic clinical trial to compare the accuracy and reproducibility of multiple sequences for cartilage imaging.
Methods and Materials: A prospective musculoskeletal MR trial was designed to compare a standard 3DSPGR and novel 3D pulse sequences (3DFSE, 3DSSFP, 3DSPGR with water selective spectral-spatial pulse), correlated with knee arthroscopy as the gold standard. Size and grade scores were taken in 12 segments per subject (anterior, central, posterior in the medial and lateral femoral condyles and tibial plateau). Standard ordinal rating scales were used to measure cartilage loss: Grade 0 normal to 6 full thickness loss; Size A to E based on size categories (from <1 cm2 to >4cm2). We hypothesized that 3D acquisitions: (1) provide more accurate and sensitive assessment of cartilage defects than standard sequences, (2) are more reproducible based on quantitative measurements of cartilage loss, and (3) are more reproducible in evaluating disease progression over 2 years. Pilot data utilized kappa scores of repeated cartilage readings from our previous reproducibility study on volumetric cartilage assessment and rate of change in cartilage volume over time. Statistical methods under each hypothesis were (1) contingency table of sensitivity (Sens); Wilcoxon test of area under the ROC curve (Az); (2) kappa of repeated measures (k); (3) Chi-square test for coefficient of variation (CV). Significance level was 5% at 85% power, with a 15% annual attrition rate per year due to bleeding during arthroscopy thereby obscuring the visual field for cartilage grading or withdrawing from the arthroscopy.
Results: The sample sizes to test changes in parameter values from each null to its corresponding alternative hypothesis were: (1) 35 (Sens: 87% to 92%); 43 (Az: 90% to 95%); (2) 39 (k: .40 to .50); (3) 37 (CV: 3.9% to 1.5%). Final sample size was 43 yielding 57 at an added 15% attrition rate per year for 2 years.
Conclusion: Metrics were adopted for designing a radiologic clinical trial to compare multiple pulse sequences. Recommended approaches are ROC analysis for assessing the accuracy of MRI pulse sequences, observer agreement for reproducibility assessments, and coefficient of variation for long-term followup.
Questions about this event email: zou@bwh.harvard.edu
Zou PhD, K,
Statistical Methods for Estimating Sample Sizes in a Clinical Trial Designed to Compare Different Cartilage Sensitive MRI Pulse Sequences. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3104739.html
