Effect of target lesions selection on between-reader variability of response assessment according to RECIST 1.1.

2528Background: Response-classification-systems, e.g. RECIST1.1, and dedicated oncology software-tools (DOST) are used to standardize response assessment. Expectation is that different readers should yield the same response-classification for any given patient. We investigated real-life variability between readers who, as in clinical practice, were free to select target-lesions (TL). Methods: Prospective study on 316 patients with metastatic disease who underwent 932 CT-studies, yielding a total 616 follow-up occasions (baseline vs. follow-up) for analysis. All CT-studies were independently evaluated by 3 radiologists who used state-of-the-art DOST (MintLesion). Readers were free to select TL in the respective baseline study, and did so independently. Kappa-statistics were used to analyse agreement for RECIST1.1 response-class-assignment depending on whether readers had selected the same or different TL.To investigate possible impact on treatment decisions, agreement was also determined after aggregating ...