This project has been shortlisted for the DPH 2025 Innovation Prize – Best Data Driven Innovation
Team: Lakshya Sharma, Vaishnavi Balaji, Alvin Katumba, Esha Mohan and Sola Adeleke (University College London, Curenetics Ltd, University Hospital Dorset)
Outline: Immune checkpoint inhibitor (ICI) related adverse events (irAE) are a common concern among patients with melanoma. Experiencing irAEs can negatively impact a patient’s quality of life and may necessitate hospital admission. However, there is a lack of precision medicine tools that accurately predict which patients will experience irAEs. Such tools could be used to support informed, shared decision-making between patients and clinicians.
Curenetics is a healthcare company that aims to use machine learning (ML) to predict irAEs and subsequent hospital admissions in patients with melanoma receiving ICI treatment, to enable a precision medicine approach to patient care and resource management.
Methods: 455 datasets were included for patients initiated on ICIs between 2014-2024 in a single cancer centre in Kent, U.K. Raw data was preprocessed using a standard scaling function and one hot encoding, followed by Synthetic Minority Oversampling Technique (SMOTE) to balance the imbalanced classes. Six ML algorithms were developed using 80% of the data for training, and 20% for validation. Grid search cross-validation technique was used for hyperparameter optimization. Shapley Additive Explanations (SHAP) explainable artificial intelligence (AI) was used to interpret the irAE prediction model.
Results: Of the 455 patients, 121 patients (27%) experienced irAEs. The most frequently occurring class of irAE (CTCAE v5) was gastrointestinal (38, 31%), followed by skin (25, 21%) and subsequently endocrine (22, 18%).
Logistic Regression model performed best at predicting irAEs (accuracy = 0.92, AUC = 0.90, sensitivity/recall = 0.95, precision = 0.95, F1 score = 0.90). Key features that predicted the development of irAEs included increasing age, female gender and exposure to combination therapy (ipilimumab and nivolumab) or monotherapy with pembrolizumab.
Among the 121 patients with irAEs, 37 (31%) required hospital admission due to irAEs. 49 patients (40%) stopped 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) increased admission risk, while endocrine irAEs (OR = 0.09, p =0.03) were associated with reduced risk of hospital admission.
Conclusion: We at Curenetics have developed machine learning algorithms that accurately predicted irAEs and related hospital admissions in melanoma patients. These predictive models could be incorporated into discussions about the risks and benefits of ICI therapy between patients and clinicians. Thus, achieving a precision medicine approach that tailors ICI treatments to the needs of individual patients. Additionally, the ability to predict hospital admissions can help healthcare systems to improve resource allocation and care delivery planning. Future work will aim to externally validate these models across diverse populations and clinical settings.
