Leveraging Large Language Models and Survival Analysis for Early Prediction of Chemotherapy Outcomes
arXiv:2603.11594v1 Announce Type: new Abstract: Chemotherapy for cancer treatment is costly and accompanied by severe side effects, highlighting the critical need for early prediction of treatment outcomes to improve patient management and informed decision-making. Predictive models for chemotherapy outcomes using real-world data face challenges, including the absence of explicit phenotypes and treatment outcome labels such as cancer progression and toxicity. This study addresses these challenges by employing Large Language Models (LLMs) and ontology-based techniques for phenotypes and outcome label extraction from patient notes. We focused on one of the most frequently occurring cancers, breast cancer, due to its high prevalence and significant variability in patient response to treatment, making it a critical area for improving predictive modeling. The dataset included features such as vitals, demographics, staging, biomarkers, and performance scales. Drug regimens and their combina
arXiv:2603.11594v1 Announce Type: new Abstract: Chemotherapy for cancer treatment is costly and accompanied by severe side effects, highlighting the critical need for early prediction of treatment outcomes to improve patient management and informed decision-making. Predictive models for chemotherapy outcomes using real-world data face challenges, including the absence of explicit phenotypes and treatment outcome labels such as cancer progression and toxicity. This study addresses these challenges by employing Large Language Models (LLMs) and ontology-based techniques for phenotypes and outcome label extraction from patient notes. We focused on one of the most frequently occurring cancers, breast cancer, due to its high prevalence and significant variability in patient response to treatment, making it a critical area for improving predictive modeling. The dataset included features such as vitals, demographics, staging, biomarkers, and performance scales. Drug regimens and their combinations were extracted from the chemotherapy plans in the EMR data and shortlisted based on NCCN guidelines, verified with NIH standards, and analyzed through survival modeling. The proposed approach significantly reduced phenotypes sparsity and improved predictive accuracy. Random Survival Forest was used to predict time-to-failure, achieving a C-index of 73%, and utilized as a classifier at a specific time point to predict treatment outcomes, with accuracy and F1 scores above 70%. The outcome probabilities were validated for reliability by calibration curves. We extended our approach to four other cancer types. This research highlights the potential of early prediction of treatment outcomes using LLM-based clinical data extraction enabling personalized treatment plans with better patient outcomes.
Executive Summary
This article presents a novel integration of Large Language Models (LLMs) and survival analysis to improve early prediction of chemotherapy outcomes, particularly in breast cancer. By leveraging LLMs and ontology-based extraction techniques, the authors address the lack of explicit phenotypes and outcome labels in real-world clinical data. The study employs Random Survival Forest to achieve a C-index of 73% for time-to-failure prediction and demonstrates accuracy and F1 scores above 70% for outcome classification. The methodology extends to additional cancer types, indicating broader applicability. The study’s use of clinical documentation to bridge data gaps and enhance predictive modeling is a significant contribution to precision oncology.
Key Points
- ▸ Use of LLMs for phenotype and outcome extraction from clinical notes
- ▸ Application of survival analysis (Random Survival Forest) to predict chemotherapy outcomes
- ▸ Extension of methodology across multiple cancer types
Merits
Innovation
The integration of LLMs with clinical data extraction is a novel approach to overcoming data sparsity in predictive modeling.
Validation
Calibration curves and statistical metrics (C-index, accuracy, F1) provide evidence of model reliability and performance.
Demerits
Generalizability Concern
While extended to four other cancer types, the study’s focus on breast cancer may limit applicability to other malignancies with distinct clinical characteristics.
Data Quality Dependency
Predictive accuracy is contingent upon the quality and completeness of EMR note documentation; gaps in clinical documentation may still impede performance.
Expert Commentary
The convergence of natural language processing and survival analytics represents a paradigm shift in oncology analytics. The authors effectively mitigate a persistent challenge in clinical data—absence of labeled phenotypes and outcomes—by deploying LLMs as data intermediaries. This is particularly compelling in breast cancer, where heterogeneity in patient response necessitates more nuanced predictive models. The use of NCCN and NIH standards as anchors for drug regimen extraction adds a layer of clinical credibility. While the C-index of 73% is commendable, it remains below the threshold of high-risk medical decision-making; therefore, clinical validation in prospective cohorts is essential before widespread implementation. Moreover, the scalability of this framework across oncology subspecialties warrants further investigation. If validated, this approach could significantly reduce clinician burden and enhance patient autonomy by enabling more accurate early prognosis, thereby shifting the paradigm from reactive to proactive care.
Recommendations
- ✓ Conduct prospective clinical trials to validate predictive accuracy in real-time patient cohorts.
- ✓ Explore integration with electronic health record (EHR) platforms to automate LLM-based data extraction and real-time decision support.