A Bayesian Network Approach to Investigate Breakthrough Cancer Pain

Abstract

INTRODUCTION
Cancer pain is a complex, multidimensional syndrome affecting approximately 55% of patients with cancer, depending on tumor stage and treatment, and is associated with poorer clinical outcomes and reduced quality of life. Pain assessment and management become particularly challenging when patients experience transient exacerbations of pain despite otherwise controlled baseline pain with opioid therapy, a condition known as breakthrough cancer pain (BTcP), which affects up to 70% of patients with chronic cancer pain. Artificial intelligence (AI) and advanced statistical approaches offer new opportunities to investigate the complex relationships among clinical variables contributing to BTcP. The Ruggi Study (ClinicalTrials.gov: NCT07038434) is a single-center, non-profit, observational-interventional study designed to characterize chronic pain through multimodal AI-based assessments in adult oncologic and non-oncologic patients.
OBJECTIVES
This preliminary analysis aimed to investigate the causal relationship between opioid therapy and BTcP, evaluating the mediating role of chronic pain phenotype in a subcohort of 99 cancer patients assessed at their first follow-up visit.
METHODS
A Bayesian network was constructed including age, smoking status, body surface area (BSA), comorbidities, cardiovascular therapy, cancer severity (presence of metastases or bone metastases), ongoing anticancer treatment at enrollment, time from diagnosis, treatment duration, opioid therapy, chronic pain phenotype, BTcP, and quality of life assessed by the Brief Pain Inventory (BPI, item 9g). The network structure was learned using the Bayesian Information Criterion, allowing the identification of statistical dependencies among variables. Missing data were preliminarily assessed using Little’s test and subsequently imputed through a model-based approach.
RESULTS
Overall, 22% of the dataset contained missing values. Little’s test excluded a missing-not-at-random mechanism (p = 0.14), supporting the imputation strategy, which mainly affected variables related to cancer therapies. The median patient age was 65 years, with a median BSA of 1.4 m²; approximately 98% had a sex-adjusted normal BSA. Most patients had comorbidities, 66% had a time from diagnosis exceeding 12 months, 17% presented with bone metastases, and 39% were receiving opioid therapy. Neuropathic pain was the most frequent chronic pain phenotype (>60%), while eight patients experienced BTcP. Twenty-six patients reported impaired quality of life (BPI item 9g >4). The Bayesian network did not identify a direct causal relationship between opioid therapy and BTcP. Instead, chronic pain phenotype emerged as a mediator: patients with nociceptive pain receiving opioids showed a higher probability of developing BTcP than those with neuropathic or mixed pain (25% vs 19%).
CONCLUSIONS
These preliminary findings suggest that the association between opioid therapy and BTcP is influenced by the underlying chronic pain phenotype rather than representing a direct causal relationship. Bayesian network analysis provides a valuable framework for disentangling complex clinical interactions and identifying patient subgroups at increased risk of BTcP, potentially supporting more personalized pain management strategies. Larger prospective studies are warranted to validate these findings and assess their clinical applicability.