PROCC: a predictive score to identify KRAS wild type metastatic colorectal cancer patients who are likely to obtain survival benefit from panitumumab treatment.

Abstract

Background: Practice guidelines recommend panitumumab with chemotherapy to treat KRAS wild-type (WT) metastatic colorectal cancer (mCRC) patients. However, not all patients respond to this therapy. We propose a score termed "PROCC" to identify likely panitumumab responders.

Methods: The training (TRDS) and validation (VALDS) datasets included KRAS WT mCRC patients treated with panitumumab (P) plus chemotherapy (TRDS, FOLFOX4; VALDS, FOLFIRI). TRDS included 36 diverse features used to generate synthetic representations analyzed via machine learning (ML) to identify patient subgroups, which were correlated with PFS and OS. A multivariable logistic regression model identified independent predictors to develop a predictive score.

Results: ML identified two subpopulations in TRDS: SP-A (n = 162) and SP-B (n = 298). Only SP-A patients under P/FOLFOX4 showed improved OS versus FOLFOX4 (HR 0.68; p = 0.04). CEA, ALP, LDH, and platelets at baseline were used to create a predictive score ("PROCC"). For TRDS, the score had an area under the curve of 0.81. PROCC ≥8.5 correlated with lower risks of progression (HR 0.67; p = 0.03) and death (HR 0.65; p = 0.04) for P/FOLFOX4 versus FOLFOX4. Validation in VALDS confirmed similar results with FOLFIRI.

Conclusions: Based on four baseline parameters, PROCC may guide the selection of KRAS WT mCRC patients most likely to benefit from panitumumab.