Abstract A015: Precision medicine approach to melanoma immunotherapy: Predicting response, adverse events, and hospital admissions using machine learning and explainable artificial intelligence
Bibliographic record
Abstract
Abstract Introduction: Immune checkpoint inhibitors (ICI) are a key component of melanoma treatment but identifying which patients will respond to ICI and which will experience ICI-related adverse events (irAEs) can be challenging. Experiencing irAEs can negatively impact a patient’s quality of life and may necessitate hospital admission or stopping treatment. Currently, there is a lack of precision medicine tools that accurately predict both ICI response and irAEs in melanoma patients. Such tools could support informed, shared decision-making between patients and clinicians. This study aimed to use machine learning (ML) and explainable artificial intelligence (AI) to predict ICI response, irAEs and subsequent hospital admissions in patients with melanoma. Methods: 455 datasets were included for patients initiated on ICI between 2014-2024 in a U.K. cancer center. Six ML algorithms were developed using 80% of the data for training, and 20% for validation. Synthetic Minority Oversampling Technique (SMOTE) was used to address class imbalances. Shapley Additive Explanations (SHAP) explainable AI was used to interpret the prediction models. Results: Logistic regression predicted positive response to ICI treatment with 74% accuracy (AUC=0.72, sensitivity/recall=0.61, specificity=0.77, precision=0.38, F1 score=0.47). Key predictors of positive response included developing irAEs (OR=3.40, p<0.001), prior imaging suggestive of stable disease (OR=3.55, p<0.001) or prior imaging suggestive of positive response (OR=6.99, p<0.001). Receiving prior radiotherapy was negatively associated with response (OR=0.23, p<0.001). Among 455 patients, 121 (27%) experienced irAEs. The most common irAE classes (CTCAE v5) were gastrointestinal (38, 31%), skin (25, 21%) and endocrine (22, 18%). Of the ML models, logistic regression predicted irAEs with the greatest accuracy of 92% (AUC=0.90, recall=0.95, precision=0.95, F1 score=0.95). SHAP value analysis highlighted older age, female sex and treatment with combination therapy (ipilimumab and nivolumab) or monotherapy with pembrolizumab as predictors for irAE. Of the 121 patients with irAEs, 37 (31%) required hospital admission and 49 patients (40%) discontinued ICI treatment due to irAEs. Logistic regression predicted hospital admission with 77% accuracy (AUC=0.80, sensitivity/recall=0.38, specificity=0.94, precision=0.74, F1 score=0.50). Combination therapy (OR=3.74, p=0.004) and gastrointestinal irAEs (OR=4.34, p=0.002) was associated with an increased risk of admission, while endocrine irAEs (OR=0.09, p=0.03) were associated with reduced risk of hospital admission. Conclusion: These machine learning algorithms accurately predicted ICI response, irAEs and related hospital admissions in melanoma patients. Future work will aim to externally validate these models in diverse clinical settings. Ultimately, these predictive tools could be integrated into clinical discussions about the risks and benefits of ICI therapy to enable a precision medicine approach that tailors ICI treatments to the needs of individual patients. Citation Format: Lakshya Sharma, Vaishnavi Balaji, Alvin Katumba, Esha Mohan, Sola Michael. Adeleke. Precision medicine approach to melanoma immunotherapy: Predicting response, adverse events, and hospital admissions using machine learning and explainable artificial intelligence [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr A015.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.013 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".