1670P - Evaluation of alternate tumor metrics and cut-points for response categorization using the RECIST 1.1 data warehouse

Date 30 September 2012
Event ESMO Congress 2012
Session Poster presentation II
Topics Anti-Cancer Agents & Biologic Therapy
Presenter Sumithra Mandrekar
Authors S.J. Mandrekar1, M. An2, J. Meyers3, A. Grothey4, J. Bogaerts5, D. Sargent3
  • 1Health Sciences Research, Mayo Clinic, 55905 - Rochester, NY/US
  • 2Mathematics, Vassar College, Poughkeepsie/US
  • 3Health Sciences Research, Mayo Clinic, 55905 - Rochester/US
  • 4Medical Oncology, Mayo Clinic, US-55905 - Rochester/US
  • 5Statistics Dept, EORTC, 1200 - Brussels/BE



We previously showed that the trichotomized response metric (TriTR: Complete or Partial Response (CR or PR) vs. Stable (SD) vs. Progression (PD)) had better predictive ability than the Response Evaluation Criteria in Solid Tumors (RECIST) best response (BR) metrics (CR/PR vs. others) (An et al., 2011). We sought to test and validate TriTR, disease control rate (DCR: CR/PR/SD vs. PD) and BR metrics utilizing alternate cut-points for PR and PD using the RECIST data warehouse.


Data from 13 trials (5994 patients with breast, lung or colorectal cancer) were randomly split (60:40) into training and test sets stratified by tumor type, survival and progression status. 21 pairs of cut points for (PR, PD) were considered: PR (20-50% decrease, by 5% increments) and PD (10-20% increase, by 5% increments), where the pair (30%, 20%), corresponds to the RECIST categorization. Cox proportional hazards models with a flexible landmark analysis at 12- and 24-weeks post-baseline, adjusted for baseline tumor burden, were used to assess the impact of the metrics (using alternate cutoffs) on overall survival (OS). Model discrimination was assessed using the concordance (c-) index [c = 0.5: no association; c = 1.0: perfect association].


Standard RECIST cutoffs demonstrated similar predictive ability as the alternate (PR, PD) cutoffs for all metrics at both time points. Regardless of tumor type, the last TriTR and DCR metrics (assessed at the landmark time point) had higher (or similar) c-indices to BR, and best TriTR and DCR metrics (assessed any time before the landmark time point). The 24-week metrics were no better than those at 12-weeks. None of the metrics did particularly well for breast cancer.


Alternative cut-points to RECIST standards provided no meaningful improvement in OS prediction. TriTr and DCR metrics assessed after 12-weeks have good predictive performance. These results are currently being validated.

Table: 1670P

C-Index at 12 and 24 weeks (Training Set)
Breast Colon Lung
Metric (based on RECIST cutoffs) 12 wks (N = 1048) 24 wks (N = 1060) 12 wks (N = 710) 24 wks (N = 727) 12 wks (N = 1816) 24 wks (N = 1819)
BR 0.57 0.58 0.61 0.64 0.62 0.62
Last TriTR 0.58 0.59 0.63 0.66 0.64 0.63
Best TriTR 0.58 0.58 0.62 0.64 0.63 0.62
Last DCR 0.58 0.58 0.62 0.65 0.61 0.61
Best DCR 0.57 0.56 0.59 0.59 0.62 0.59


All authors have declared no conflicts of interest.