# A Streamlined Protocol for Developing a Clinicopathological Prediction Model for Patient Survival of Post‐Resection of Pancreatic Cancer

## TL;DR

This paper presents a streamlined protocol to predict survival of pancreatic cancer patients after surgery by combining immune features and clinical data.

## Contribution

A novel clinicopathological prediction model integrating tumor immunological features and clinical data for post-resection survival prediction in pancreatic cancer.

## Key findings

- Cancer cell MHC I intensity, CD4+ T cell to tumour cell ratio, resection margin status, and tumour T stage were key variables in the model.
- The model showed moderate discrimination with time-dependent AUC values of 0.698, 0.765, and 0.825 at 1, 3, and 5 years, respectively.
- Internal validation by bootstrapping showed a slight decrease in the overall c-index from 0.67 to 0.652.

## Abstract

Pancreatic ductal adenocarcinoma (PDA) is one of the most malignant solid cancers. As surgery is the only cure, prediction of long‐term survival post‐resection is critical to guide patient selection for the subsequent treatment. Tumour immune evasion plays a key role in PDA tumorigenesis.

Using a streamlined protocol, we developed a clinicopathological prediction model for the overall survival of patients with PDA after resection by integrating tumour immunological features and clinical data. Multiplex immunohistochemistry was performed using human tumour microarray samples. The results were combined with retrospectively collected clinical data of 79 patients. Variables were selected by least absolute shrinkage and selection operator (LASSO) regression with 10‐fold cross‐validation to develop the prediction model. The performance of the model was assessed using the concordance index, receiver operating characteristic curve, calibration plot and decision curve analysis. The model was validated by bootstrap resampling.

Cancer cell MHC I intensity, CD4+ T cell to tumour cell ratio, resection margin status and tumour T stage were identified for prediction model development using Cox proportional hazard regression. Discrimination of developed model was moderate on the time‐dependent area under curve at one (0.698), three (0.765) and five (0.825) years. A small decrease in the overall c‐index from 0.67 to 0.652 was shown in the internal validation by bootstrapping.

Our protocol provided a framework for developing a complex model that will significantly contribute to clinical practice.

We have identified the spatial distribution of MHC‐I, CD4, CD8, CD19, PAK4, and LC3B using multiplex immunostaining. We then combined the data of multiplex immunostaining with clinical pathological data for the software analysis to create a model that can be used to predict the post‐resection survival of pancreatic cancer patients.