# A Treatment Decision Model for Cutaneous Squamous Cell Carcinoma Based on Bayesian Networks

## TL;DR

This paper introduces a Bayesian network-based model to help doctors choose the best treatments for advanced skin cancer patients.

## Contribution

A novel Bayesian network decision support model for cutaneous squamous cell carcinoma treatment selection.

## Key findings

- The model achieved 95.5% accuracy in recommending treatments compared to tumor board decisions.
- The Bayesian network effectively handles missing or uncertain clinical and genetic data.
- The model's recommendations were statistically significant (p < 0.001).

## Abstract

Treatment decision-making has become increasingly challenging, especially in oncology, due to the growing number of available therapeutic options, particularly in advanced stages of disease. Cutaneous squamous cell carcinoma, one of the most common skin malignancies, is usually treated surgically. However, treatment selection may become more complex in advanced or unresectable cases. In recent years, immune checkpoint inhibition (e.g., Cemiplimab) has extended therapeutic options. In this study, we developed a Bayesian network–based decision support model to assist clinicians in selecting appropriate treatment strategies for patients with cutaneous squamous cell carcinoma.

Background: One of the most prevalent non-melanoma skin cancers (NMSCs) is cutaneous squamous cell carcinoma (cSCC), which is typically treated surgically. For patients with advanced or inoperable disease, systemic therapies—particularly immune checkpoint inhibitors—have become increasingly important. The anti-PD-1 monoclonal antibody Cemiplimab was approved for the treatment of advanced cSCC, providing patients who are unable to receive conventional therapy with additional options. Methods: In this study, we developed a clinical decision support tool based on Bayesian networks (BNs) to help clinicians choose the most suitable treatment strategies for cSCC. The model can manage missing or uncertain data and includes patient-specific clinical, histological, and genetic information, such as tumor type, stage, and PD-L1 expression. Results: Using data from 66 patients with either basal cell carcinoma (BCC) or cSCC, we retrospectively validated the model by comparing the treatment recommendations from the tool with the actual choices made by multidisciplinary tumor boards. The model demonstrated an overall accuracy of 95.5% and statistical significance with a p-value of <0.001. Conclusions: Our results suggest that BNs are a valuable tool for representing complex clinical decision-making processes.