Methodological issues in pooled analysis of biomarker studies