568P - Single cell AI-based detection of DNA mismatch repair deficiency in 1,988 colorectal cancers reveals prognostic and predictive value in the SCOT trial

Background

Testing for DNA mismatch repair deficiency (MMRd) is recommended for all colorectal cancers (CRC). Automation of this would facilitate precision medicine, particularly if it provided information on likely aetiology.

Methods

We developed AIMMer, an AI-based method for determination of MMR protein expression at single cell level from IHC-stained routine pathology samples. We applied it to over 2,000 cases from the SCOT clinical trial, which compared 3 vs 6 months of oxaliplatin-based adjuvant chemotherapy (mFOLFOX6 or CAPOX) for stage II/III CRC. AIMMer performance was evaluated against ground truth established by blinded review by two expert pathologists, and MMRd prognostic and predictive value was determined by Cox proportional hazards models.

Results

Benchmarking of AIMMeR against pathologist ground truth MMR calls revealed AUROC of 0.98 (bootstrap 95% CI = 0.97–0.99) and Youden index (sensitivity plus specificity) of 1.87 at cutpoint of 10.7% MMR positive cells. This cutpoint gave positive predictive value (PPV) of >95% for both the commonest pattern of somatic MMRd (MLH1 and PMS2 loss) and for MMR proficient (MMRp) status, and agreement with pathologist calls approaching that between individual pathologists (k = 0.79–0.82 vs 0.88). Analysis of CRC recurrence-free interval (RFI) confirmed MMRd prognostic value in oxaliplatin-treated patients (multivariable-adjusted HR (aHR) = 0.62, 95% CI = 0.44–0.88, P=0.007), with effect size larger in those <70 years and those with right-sided tumours. MMRd did not predict differential benefit from chemotherapy duration. Patients with MMRp tumors had similar RFI whether treated with CAPOX or FOLFOX (aHR = 0.95, 95% CI= 0.77-1.16, P=0.6), while those with MMRd tumours treated with FOLFOX had shorter RFI (aHR = 2.08, 95% CI= 1.09-3.97, P=0.027, P _INTERACTION=0.04).

Conclusions

AIMMeR holds promise to reduce pathologist workflow and streamline clinical diagnostics in CRC. The possible predictive value of MMRd for chemotherapy regimen should be investigated in additional cohorts.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

D.N. Church.

Funding

Cancer Research UK, National Institute for Health Research (NIHR), Promedica.

Disclosure

All authors have declared no conflicts of interest.