Computer simulated comparison of neoadjuvant versus upfront surgery for resectable pancreatic cancer: the application of machine-learning algorithms to support personalised decision-making

Aims: To create a hybrid discrete event simulation (DES) model that performs patient-level decision-analysis of neoadjuvant versus upfront surgery for resectable PDAC. Methods: A hybrid DES model was created using TreeAgePro software to simulate 10000 patients with resectable PDAC treated in neoadjuvant and upfront surgery pathways. Model probabilities were drawn from the entirety of the data distributions contained within randomised-controlled-trials and prospective phase II/III trials, as determined by the Anderson-Darling statistic, for each parameter contained within the model. Model outcomes were validated against a prospectively maintained patient database. Results: Neoadjuvant pathway gave 20.01months versus 16.55months in upfront surgery pathway with a superior pathway selection frequency 40.6% and pathway indifference frequency 59.4%. Threshold analysis showed that the probability of resection and R0 resection in the neoadjuvant pathway had to be greater than 38% and 15.4% respectively for neoadjuvant pathway to be superior. Patients in the neoadjuvant pathway that did not proceed to resection had an expected incremental value of 1.5-5.5months over a resection probability range of 47-94% if treated within upfront surgery pathway. Implementation of PRODIGE trial results showed upfront surgery to be superior if the probability of resection and receiving adjuvant therapy within the upfront surgery pathway was greater than 54% and 8% respectively. Conclusion: This study suggests that superior pathway selection depends on individual patient factors. Future research must focus on harnessing advances in computational statistics to support better decision-making through personalised predictive medicine that can engage with real-world practice where, unlike in randomised-controlled-trials, complexity cannot be controlled for.