Management Summary
The AI4ANNA study shows which clinical endpoints yield predictive signals with machine learning and where the evaluated models reach their limits. Cross-study validation revealed moderate signals for the objective response rate and progression-free survival, while predictive performance was limited for most safety outcomes. The study therefore provides a solid basis for assessing suitable applications of machine learning and for focusing further analyses on appropriate research questions.
Project summary
Novartis
Industry: Pharmaceuticals
Focus: AI & Machine Learning
Use Case: Predictive Machine Learning
Project Objective
Evaluate the predictive potential of clinical trial data
An Overview of the Key Project Aspects
- Analysis of data from the RIBECCA and RIBANNA studies
- Comparison of LASSO and XGBoost for clinical predictive models
- Internal and cross-study validation of the models
Tech Stack
The Starting Point
Pharmaceutical companies need reliable methods for analyzing clinical trial data. Until now, traditional statistical methods have been the primary tools used for this purpose. While these methods yield important and easily interpretable results, they are sometimes based on linear assumptions.
Machine learning can complement established methods by identifying additional patterns in clinical data. Time-to-event data poses a particular challenge. In such cases, the analysis must account for the fact that events—such as tumor progression—may not occur in all patients during the observation period or may not be fully captured.
Our Strengths, Your Benefits
Combining Statistical Methods and Machine Learning
HMS helped Novartis explore the potential of machine learning for clinical trial data. In this project, we combined statistical analysis with ML-based predictive models, cross-study validation, and methods for assessing relevant baseline factors.
What You Can Expect from HMS
When analyzing clinical data, model selection, validation, and interpretability must work in tandem. To achieve this, HMS combines established statistical methods with machine learning and expert interpretation.
- Selection and comparison of suitable models for clinical questions
- Internal and cross-study validation of predictive performance
- Transparent analysis of relevant clinical baseline characteristics
Are you planning a similar project?
Talk to us about your current situation and the requirements for a potential implementation. We’ll help you assess the project from a business and technical perspective and determine the best next steps.