Prediction model for mutation detection via circulating tumor DNA analysis in pancreatic cancer.

BACKGROUND

Comprehensive genomic profiling (CGP) is essential for precision medicine in pancreatic cancer (PC) patients. In unresectable cases, obtaining sufficient tissue is challenging, making plasma-based liquid CGP a common choice. The successful detection of KRAS mutations, which are present in more than 90 % of pancreatic cancers, indicates that liquid CGP has successfully detected circulating tumor DNA. This study aimed to develop a prediction model for detecting KRAS mutations in liquid CGP among patients with PC using the Japanese National Database.

METHODS

We evaluated the detection of potentially targetable gene mutations, including BRAF V600E, BRCA1/2, and PALB2. Multivariate logistic regression analysis was performed to identify independent predictors of KRAS mutation, and a scoring system was developed.

RESULTS

Targetable gene mutations were significantly more frequent in PC patients with KRAS mutations than in those without (4.8 % vs. 0.5 %, P < 0.001). A predictive model was created using four clinical variables (metastasis to the liver, bone, or peritoneum and treatment response before blood collection) in the training cohort (n = 1574). The percentage of patients with KRAS mutations was predicted to be 29.6 % at low probabilities (-1 to 3 points) and 84.4 % at high probabilities (9 or more points). In the validation cohort (n = 1052), the model showed moderately good discrimination, with a diagnostic accuracy of 0.70. We developed a web-based application for this predictive model (https://pdac-KRAS-detection.shinyapps.io/KRAS-prediction-tool/).

CONCLUSIONS

We developed and validated a predictive model for KRAS mutations. This model may be a useful clinical decision-making tool for liquid CGP in patients with PC.