Utility of public Knowledge Bases (KB) for comprehensive tumor molecular profiling (CTMP) result interpretation.

Abstract

e15119 Background: With the spread of CTMP in cancer patient management and expansion of molecular-targeted treatment opportunities, resources for quick and efficient molecular profile interpretation are in high demand. Number of public KBs have been introduced. We analysed their utility to compile per patient treatment recommendations. Methods: Analysis was performed across patients, who underwent CTMP. Based on the reported molecular alterations, interpretation and critical evidence evaluation was performed not taking into account the recommendations provided in the original CTMP reports. Therapy recommendations were ranked according to ESCAT. Resistance biomarkers were ranked according to OncoKB V2 level of evidence system. OncoKB, CIViC, CGI, CGA, VICC and MolecularMatch were assessed. KBs were assessed through API or local copy of KB. Results: A total of 222 CTMP reports from 222 patients (35% - NSCLC; 24% - CRC; 8% - pancreas; 7% - breast; 4% - gastric; 22% - other) were analysed. CTMP was performed in Foundation Medicine (82%) or Atlas Oncology Diagnostics (18%). Across 222 patients 932 molecular alterations were identified. A total of 1394 therapy recommendations were compiled associated with 368 molecular alteration across 171 (77%) patients (112 recommendations ESCAT level I; 908 - ESCAT III-IV; 70 - OncoKB R1). Across 181 ESCAT-I/R1 therapy recommendations 172 (95%) were present in at least 1 KB (92% for ESCAT-I recommendations, 100% - R1) with the VICC providing the highest rate of retrieval (80%), followed by OncoKB (78%), MolecularMatch (70%), CIViC - (69%) and CGI (69%). Only 92 (53%) of ESCAT-I/R1 therapy recommendations were present in all KBs. Across 27 patients with ESCAT-I recommendations, at least 1 therapy recommendation was retrieved in any KB for all patients. ESCAT II-IV therapy recommendations were provided for 161 patients and grouped by drug class, resulting in 457 [biomarker]-[drug class] associations. Of them 346 (75%) were present in at least 1 KB with the VICC providing the highest rate of retrieval (69%), followed by CGI (61%), CGA (56%), OncoKB (48%), CIViC (48%) and MolecularMatch (45%). Of 161 patients, for 24 (15%) no recommendations were retrieved in any KB for any biomarker identified. For random 15 (a total of 68 molecular alterations) patients KBs were assessed manually in order to estimate the average time in use. CGI was the most easily accessible (3.2 min per patient in average), following by CGA (5.6 min), OncoKB (7.2 min), CIViC (26.2 min), and VICC (35.9 min). Conclusions: Public KBs provide substantial information on ESCAT-I/R1 biomarkers and decent information on ESCAT II-IV biomarkers, reasoning manual curation. VICC provided the most complete information, though was thorough for manual use. The use of multiple KBs may significantly improve retrieval results, though is time-consuming and results in excess of misleading/outdated recommendations.