A Bayesian Approach to Biological Variation Analysis

摘要: BACKGROUND: Biological variation (BV) data have many applications for diagnosing and monitoring disease. The standard statistical approaches estimating BV are sensitive to “noisy data” assume homogeneity of within-participant CV. Prior knowledge about is mostly ignored. aims this study were develop Bayesian models calculate that (a) robust data,” (b) allow heterogeneity in the CVs, (c) take advantage prior knowledge. METHOD: We explored with different degrees robustness using adaptive Student t distributions instead normal when possibility CV was allowed. Results compared more chloride triglyceride from European Variation Study. RESULTS: Using most approach on a raw set gave results comparable outlier assessments removal. posterior distribution fitted model gives access credible intervals all parameters can be used assess reliability. Reliable relevant priors proved valuable prediction. CONCLUSIONS: recommended clear picture degree heterogeneity, ability crudely estimate personal CVs explore subgroups. Because experiments expensive time-consuming, estimates should considered high value applied accordingly. By including reliable knowledge, precise possible even small sets.