A Case Study On Optimizing Immunotherapy with Excelra’s AI-powered Patient Stratification

Overview

Immunotherapy has transformed cancer treatment by enabling targeted immune responses against tumors. However, variability in patient response remains a major challenge, often leading to ineffective treatment outcomes and increased risk of adverse effects. Leveraging AI-powered patient stratification and advanced bioinformatics solutions delivered through Excelra’s Bioinformatics Services and Scientific Informatics capabilities, Excelra developed a data-driven approach to accurately identify patients most likely to benefit from immunotherapy, supporting precision medicine initiatives and improved clinical decision-making.

Our client

Our client is a California-based pharmaceutical company focused on developing innovative immunotherapy drugs for cancer treatment. The organization is committed to improving therapeutic effectiveness through advanced research and clinical innovation supported by modern digitization and data science in drug discovery approaches but faced challenges in accurately selecting suitable patients for immunotherapy interventions.

Client’s challenge

A leading California-based pharmaceutical company developing immunotherapy drugs faced a critical challenge: identifying the most suitable patients for their therapy. Immunotherapy is promising in treating various cancers, but patient response can vary greatly. Accurately classifying patients who would benefit most from the therapy is crucial for maximizing treatment effectiveness and minimizing side effects.

Traditionally, patient selection relied on broad clinical factors, leading to suboptimal outcomes. The client sought a more precise approach aligned with modern AI & Machine Learning solutions and advanced Clinical Data Services to identify ideal candidates for their immunotherapy treatment.

Client’s goals

The client aimed to develop a robust and efficient method for classifying patients into two distinct groups:
By achieving this goal, the client hoped to:

• Responders: Patients most likely to experience a positive response to the immunotherapy treatment.
• Non-responders: Patients unlikely to benefit from the treatment and potentially at risk of side effects.

By achieving this goal, the client hoped to:

• Improve treatment efficacy: Matching the right patients to the right therapy would lead to better overall treatment outcomes supported by AI-driven precision medicine strategies.
• Reduce side effects: Minimizing unnecessary administration of the therapy would decrease the risk of adverse reactions in non-responding patients.
• Optimize clinical trials: A more precise selection process could streamline trials using structured datasets similar to approaches described in analysis-ready clinical datasets.

Our Approach

Excelra partnered with the client to leverage artificial intelligence (AI) and machine learning (ML) for patient stratification, combining expertise from Data Science Services and Scientific Data Management.

• Data Preparation: We meticulously cleaned, normalized, and integrated complex patient data from various sources, including electronic health records, genomic data, and clinical trial data.
• Feature Selection: We employed a robust feature selection methodology, combining multiple techniques to identify the most relevant diagnostic parameters for patient classification. This ensured the model focused on the most informative data points for accurate prediction.
• Model Building: We developed and evaluated various machine learning models using the selected features. This involved training the models on a portion of the data and testing their performance on unseen data to ensure generalizability.
• Evaluation and Model Interpretation: We rigorously evaluated the models’ performance metrics, including accuracy, sensitivity, and specificity. We also employed advanced interpretability techniques to understand how the models arrived at their classifications. This transparency provided valuable insights into the factors driving patient response.

Results

Through our collaborative efforts, we achieved significant progress in patient stratification for the client’s immunotherapy treatment:

• Highly Accurate Classification: The XGBoost model achieved an accuracy of 89% in differentiating responders from non-responders. This can lead to improved treatment efficacy and potentially reduce unnecessary side effects.
• Improved Response Identification: The XGBoost model demonstrated a sensitivity of 85%, indicating it can correctly identify 85% of patients who will respond positively to the immunotherapy treatment.
• Reduced False Positives: The model achieved a specificity of 90%, meaning it can accurately exclude 90% of patients unlikely to benefit from the therapy, potentially minimizing unnecessary side effects.