Developing a common language for tumor response to immunotherapy: immune-related response criteria using unidimensional measurements

Abstract

Purpose: Immune-related response criteria (irRC) was developed to adequately assess tumor response to immunotherapy. The irRC are based on bidimensional measurements, as opposed to unidimensional measurements defined by Response Evaluation Criteria in Solid Tumors, which has been widely used in solid tumors. We aimed to compare response assessment by bidimensional versus unidimensional irRC in patients with advanced melanoma treated with ipilimumab.

Experimental design: Fifty-seven patients with advanced melanoma treated with ipilimumab in a phase II, expanded access trial were studied. Bidimensional tumor measurement records prospectively conducted during the trial were reviewed to generate a second set of measurements using unidimensional, longest diameter measurements. The percent changes of measurements at follow-up, best overall response, and time-to-progression (TTP) were compared between bidimensional and unidimensional irRC. Interobserver variability for bidimensional and unidimensional measurements was assessed in 25 randomly selected patients.

Results: The percent changes at follow-up scans were highly concordant between the 2 criteria (Spearman r: 0.953-0.965, first to fourth follow-up). The best immune-related response was highly concordant between the 2 criteria (κw = 0.881). TTP was similar between the bidimensional and unidimensional assessments (progression-free at 6 months: 70% vs. 81%, respectively). The unidimensional measurements were more reproducible than bidimensional measurements, with the 95% limits of agreement of (-16.1%, 5.8%) versus (-31.3%, 19.7%), respectively.

Conclusion: irRC using the unidimensional measurements provided highly concordant response assessment compared with the bidimensional irRC, with less measurement variability. The use of unidimensional irRC is proposed to assess response to immunotherapy in solid tumors, given its simplicity, higher reproducibility, and high concordance with the bidimensional irRC.

Fig. 1

The percent changes according to bidimensional and unidimensional measurements at each follow-up scan from the 1^st to 17^th follow-up scans. The orange dashed lines represent the cutoff values for response and progression (−50% and +25% for bidimensional measurements, −30% and +20% for unidimensional measurements). The observations within the lower left, middle center, and upper right boxes have concordant assessment between tow measurements, while observations in other boxes have discordant assessment. The purple dashed line represents +44% change for bidimensional measurements, which corresponds to +20% change for unidimensional measurements, which was given to visually demonstrate that more observations are concordant if this cut-off value is used. The percent changes presented in the figure are in comparison with baseline measurements when tumors are decreasing to assess response, and in comparison with the nadir (the smallest measurement since baseline) when tumors are increasing to assess progression. These values are displayed since they are used to define response/progression in patients at the time of response assessment.

Fig. 2

The waterfall plot of the percent change of bidimensional and unidimensional measurements at the first follow-up. Dark gray bars represent the percent changes by bidimensional measurements, and light gray bars represent the percent change by unidimensional measurements. Dashed lines demonstrate cut-off values for bidimensional response and progression (−50% and +25%). Dotted lines demonstrate cut-off values for unidimensional response and progression (−30% and +20%). Response assessment at the 1^st follow-up by two assessments had almost perfect agreement (weighted κ = 0.844). Eight patients with discordant assessment are marked with asterisks (*). The first 5 patients (Δ) had bidimensional changes >200% (range: 238-768%).

Fig. 3

Time to progression according to bidimensional vs. unidimensional assessment.

Fig. 4

Interobserver variability of bidimensional and unidimensional measurements Bland-Altman plots demonstrate interobserver variability of bidimensional and unidimensional measurements on baseline scans in 25 patients. The 95% limits of agreement of bidimensional measurements were (−31.3%, 19.7%; Fig 4A, dashed lines), that were twice wider compared to those of unidimensional measurements (−16.1%, 5.8%; Fig. 4B, dashed lines). The dotted lines represent the mean relative difference (%).

Fig. 4

Interobserver variability of bidimensional and unidimensional measurements Bland-Altman plots demonstrate interobserver variability of bidimensional and unidimensional measurements on baseline scans in 25 patients. The 95% limits of agreement of bidimensional measurements were (−31.3%, 19.7%; Fig 4A, dashed lines), that were twice wider compared to those of unidimensional measurements (−16.1%, 5.8%; Fig. 4B, dashed lines). The dotted lines represent the mean relative difference (%).