Methods
Reference library generation. A survey of the PubMed database was undertaken to identify all reports that correlated biomarkers with clinical responses and outcomes (i.e., time to progression, progression-free survival, overall survival) following specific chemotherapy. The search terms used were “[drug]”, “predictive” and “marker”, where the drug is an approved or off-label drug used in ovarian cancer. For selected biomarkers, additional queries were carried out with “[marker]” and “ovarian cancer”, where the marker was identified as having potential correlation with any of the drugs of interest. Initial searches recovered ~5000 references, from which 585 were selected following review of the abstracts for relevancy. Using a reference manager program, End Note, each abstract was annotated to reflect cancer type, specific drug combinations used in the study, predictive ability, and marker type (protein, RNA, DNA). Total references by drug class: 234 platinum, 155 taxanes, 43 topoisomerase 1 inhibitor, 111 topoisomerase 2 inhibitor, 45 gemcitabine, 106 fluoropyrimidines/anti-folates. No specific correlative markers were found in searches for ifosfamide, cyclophosphamide, and altretamine.
Marker selection criteria. Candidate markers were identified from reports of clinical (not pre-clinical) research showing evidence for association with response/outcome and where protein, mRNA, or DNA amplification or function-altering mutations were measured. With the exception of BRCA1/2, germline DNA alterations or polymorphisms were not considered. Markers were considered predictive if >= 50% of the studies for that marker showed consistent and statistically significant (p value <0.05) associations with response/outcome. At least 5 references were required for any marker that had only 50% of the studies with consistent results. In other cases, a minimum of two concordant studies was required.
