Merlijn Sevenster PhD, Abstract Co-Author: Employee, Koninklijke Philips NV

Paul J. Chang MD, Presenter: Co-founder, Stentor/Koninklijke Philips Electronics NV Technical Advisory Board, Amirsys, Inc Research Contracts, Koninklijke Philips NV Medical Advisory Board, lifeIMAGE Inc Medical Advisory Board, Merge Healthcare Incorporated

Jeffrey Bozeman BA, Abstract Co-Author: Nothing to Disclose

Andrea Cowhy BS, Abstract Co-Author: Nothing to Disclose

Piotr Roman Obara MD, Abstract Co-Author: Nothing to Disclose

Joost Peters, Abstract Co-Author: Employee, Koninklijke Philips NV

Manish Sharma MD, Abstract Co-Author: Nothing to Disclose

Will Trost BA, Abstract Co-Author: Nothing to Disclose

Lauren Wall MS, Abstract Co-Author: Nothing to Disclose

Charles William Westin MD, Abstract Co-Author: Nothing to Disclose

RECIST guidelines objectivize treatment response assessment in oncology trial patients by standardizing comparison of index lesions. Clinical research associates (CRAs) typically select index lesions that are described in a narrative radiology report. We describe a Lesion Dashboard (LD) tool that presents index candidates that were identified by the radiologist as selectable items leveraging an electronic based workflow orchestration. The LD tool presents these index lesions in a longitudinal view on patient disease progression. We measure agreement in index lesion selection comparing the traditional manual and LD workflows.   

RECIST worksheets of 9 trial patients were obtained under IRB 13-0397. Images and reports of each worksheet's first two cycles were obtained and deidentified. Part 1: Mimicking the traditional workflow, 4 CRAs reviewed the 2 deidentified radiology reports for each patient, selected index lesions out of the candidates described, and transcribed these into a blank RECIST worksheet Part 2: Radiologists measured and annotated all candidates discussed in the reports on the images. Annotation data was persisted in AIM-consistent format and electronically sent to LD. 3 oncologists and 3 CRAs subsequently selected and marked candidates as index lesions in LD and completed RECIST response assessment consistent with anticipated use of LD. F-measure statistic was used to quantify agreement between selections.  

Part 1: Agreement between CRAs for selecting index lesions was 75%. Part 2: LD contained 42 candidates. Participants selected 94% of LD candidates as index lesions (oncologists: 92%; CRAs: 97%). 22% of CRA selected index lesions in part 1 were not candidate lesions in LD. Agreement was 98% for CRAs, 94% for oncologists, 96% for all participants.

Substantial variability exists in index lesion selection by end users in the traditional workflow, which is nearly eliminated by utilizing LD as users generally comply with the radiologist’s index lesion suggestions. By consuming computer-interpretable data generated upstream, LD leverages the radiologist’s expertise in the oncology workflow. 

Oncology index lesion selection variability using traditional methods is significant. Electronic based workflow orchestration can reduce variability and better leverage the radiologist's expertise and relevance  in oncologic patient management. 

Sevenster, M, Chang, P, Bozeman, J, Cowhy, A, Obara, P, Peters, J, Sharma, M, Trost, W, Wall, L, Westin, C, Reducing RECIST (Response Evaluation Criteria in Solid Tumors) Index Lesion Selection Variability through Radiology-driven Candidate Selection.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14019368.html