AI-Assisted Chemotherapy Side Effect Management

Breast Cancer, Colorectal Cancer

Large Language Model, LLM, Breast cancer, Side effect, Supportive care, colorectal cancer

This two-stage adaptive randomized controlled trial evaluates the feasibility and preliminary efficacy of large language model (LLM)-assisted intervention for managing chemotherapy side effects in patients with solid tumors. Adults with histologically confirmed breast or colorectal cancer scheduled for at least 3 months of systemic chemotherapy will be randomly assigned (1:1) to receive either LLM-assisted care or standard care. The study employs an adaptive design with initial enrollment of 60 patients (30 per arm), followed by interim analysis. Unless enrollment is stopped for safety reasons identified at interim analysis, an additional 114 patients will be enrolled for a maximum total of 174 patients (87 per arm). In the intervention group, healthcare providers input anonymized patient symptom data into an LLM system using sessions where data is not retained, which generates evidence-based management recommendations. Physicians critically review these recommendations and use them as reference for clinical decision-making, with final treatment decisions remaining under physician discretion. The control group receives standard supportive care without LLM assistance. The primary outcome is change in health-related quality of life measured by EORTC QLQ-C30 global health status/QoL scale from baseline to end of treatment. Secondary outcomes include proportion achieving clinically meaningful improvement (≥8-point increase), treatment adherence, dose intensity, healthcare resource utilization, and physician acceptance of LLM recommendations.

Chemotherapy-induced side effects affect patient quality of life and treatment outcomes. Studies indicate 86% of patients receiving chemotherapy experience at least one side effect, with 30% reporting moderate to severe symptoms. Common adverse effects including nausea, vomiting, fatigue, anorexia, pain, diarrhea, constipation, mucositis, and peripheral neuropathy lead to dose reductions, treatment delays, or discontinuation. Care gaps persist despite advances in supportive care due to clinician time constraints and variable attention to specific side effects. Digital health interventions have demonstrated improved quality of life, reduced emergency department visits, and enhanced survival in cancer patients. A 2025 umbrella review reported positive effects of digital health interventions on multiple mental health indicators in cancer patients. LLMs offer natural language processing and medical knowledge bases. Clinical application remains limited, with only 5% of LLM healthcare studies involving actual patient care according to recent systematic reviews. Reliability concerns and hallucination require physician oversight. While healthcare providers manage life-threatening chemotherapy toxicities effectively, less attention may be given to symptoms that impact quality of life without immediate life-threatening consequences. This adaptive randomized controlled trial employs a two-stage design. Stage 1 enrolls 60 patients (30 per arm) with breast or colorectal cancer without malignancy-related symptoms at baseline. Stratification occurs by cancer type. The intervention involves healthcare providers collecting patient symptoms during clinic visits, excluding information that could identify patients, anonymizing information according to standardized prompts, and inputting data into the LLM system via temporary chat sessions where no records are retained. The LLM generates symptom analysis and management recommendations. Physicians review recommendations and adjust treatment plans accordingly, with all final decisions remaining under physician responsibility. Interim analysis after 60 patients are enrolled evaluates safety, feasibility, and efficacy. Safety assessment includes stopping for serious LLM-related issues. Interim analysis after 60 patients are enrolled evaluates safety and feasibility. Safety assessment includes stopping for serious LLM-related adverse events. Feasibility assessment examines LLM system usage rate and physician satisfaction. No formal efficacy stopping rule is applied; therefore, no alpha adjustment is required for the final analysis. Based on interim results, up to 114 additional patients may be enrolled for a maximum total of 174 participants. The primary analysis uses linear mixed-effects models adjusting for baseline scores and stratification factors. A clinically meaningful difference of 8 points on the EORTC QLQ-C30 scale with standard deviation of 20 points provides \>85% power at the final sample size. Secondary analyses include logistic regression for clinically meaningful improvement rates, appropriate statistical tests for adherence and toxicity outcomes, and Poisson/negative binomial regression for healthcare utilization. This study aims to evaluate whether LLM-assisted intervention improves patient outcomes and healthcare resource utilization in oncology practice, providing evidence for the clinical application of LLMs as decision support tools in patient care.

No related papers found yet.

South Korea