Background: Sum of Diameters (SoD) of up to 5 target lesions is the continuous metric used for Response Evaluation Criteria In Solid Tumors (RECIST) classifications. As demonstrated in many studies, RECIST based outcomes such as progression free survival and response rate do not always correlate well with overall survival, warranting the need to explore alternative ways of assessing response, particularly in the setting of novel therapies. Many studies have correlated SoD rates of change to overall survival (OS) in solid tumor indications such as non-small cell lung cancer (NSCLC). To test the hypothesis that a patient's survival could be influenced more strongly by their most poorly responding or quickly progressing lesion, we compared the ability of SoD growth rate and the growth rate of the fastest growing target lesion to predict OS. Methods: CT scans from 84 adult (18+) stage III/IV NSCLC patients with varying baseline mutation status (N=19 non-Exon-20 EGFR mutation, 15 ALK rearrangements, 5 other mutations, 44 No driver mutation), treated in Finland in secondary care setting were analyzed by a radiologist to identify and track the diameters of up to 5 target lesions in each patient. Lesion diameters from the last two CT scans of each treatment line (representing 224 distinct patient-lines of treatment: 78 1 st Line, 63 2 nd Line, 99 3 rd -6 th Line, 4 unknown) were analyzed in two ways: EGRmax = max of the (signed) exponential growth rates (EGR) of target lesions; EGRSoD = EGR of SoD of target lesions. EGRmax and EGRSoD values were then clustered into three categories: low (L), medium (M) and high (H). Kaplan-Meier (KM) and Cox Proportional Hazards (CPH) analyses were then performed to determine the correlation of EGRmax or EGRSoD categories to OS. Results: Clustering analysis (independent of OS) resulted in EGRmax cutoffs of 0.06/month between L and M and 0.25/month between M and H (resulting in N=137/75/12 patient-lines in L/M/H categories), and EGRSoD cutoffs of -0.015/month between L and M and 0.095/month between M and H (resulting in N=46/151/27 patient-lines in L/M/H). Directionally, the KM analysis was as expected, with higher EGRs correlating with lower OS. However, KM analysis revealed consistent separation between L, M, and H categories for EGRmax (median OS of 33, 17.8 and 15.4 months respectively), while the KM curves for EGRSoD were overlapping for L and M and only separated for H (median OS of 19.5, 27.7, and 17 months respectively). Accordingly, CPH analysis indicated a spread of statistically significant hazard ratios (HRs) for EGRmax (HR = 1.97 for M/L and 3.28 for H/L), while only significant differences between H and L for EGRSoD (HR = 0.98 for M/L and 1.99 for H/L). Conclusions: In this population of NSCLC patients across several treatment lines, EGRmax appears to be a better predictor of OS than RECIST-based EGRSoD.