The basis for future personalized therapy approaches - Machine learning-generated 1-year survival rate, metastatic status and therapy-dependent survival in pancreatic cancer patients.

BACKGROUND

Pancreatic adenocarcinoma (PDAC) remains one of the most lethal types of cancer, characterized by its unspecific symptoms, aggressive nature and late-stage diagnosis.

PATIENTS AND

METHODS

In this study, Machine Learning (ML) models were trained and applied on the largest international, monocentric database, comprising over 23 clinical variables from 1040 PDAC patients. In this study, the potential of ML models was assessed for the 1-year survival rate, metastatic M0/M1 status and therapy-dependent survival time generating predictive and prognostic ML-based biomarkers.

RESULTS

Consistent predictive performance was achieved for the 1-year survival rate (accuracy, internal/external cohort 72 % / 70 %) and radiological metastatic M0/M1 status (Accuracy, internal/external cohort 79 % / 72 %). Moreover, distinct ML-based survival models were trained for approved first-line adjuvant and palliative chemotherapy regimen (mFOLFIRINOX, gemcitabine with or without nab-paclitaxel) across resected (C-index = 0.60), locally advanced (C-index = 0.71) and metastatic (C-index = 0.71) PDAC subgroups, demonstrating that predictive performance is best for palliative regimens.

CONCLUSIONS

The models developed in this study may serve as a foundation for supporting tumor board decisions and demonstrate that personalized, ML-based therapy concepts are feasible in the near future. In addition, the models proposed in this study are helpful for generating ML-based synthetic control arms in future clinical trials.