From data to insight Applying statistical and machine learning methods to real-world data in esophageal and gastric cancer

The overarching aim of this thesis was to apply statistical and machine learning methods to real-world data from patients with esophageal and gastric cancer, both to deepen understanding of these diseases and to advance the application of modern analytics in clinical research. By integrating epidemiological analyses, quality-of-life research, real-world evidence, and prediction modeling, this work demonstrates how routine clinical data can be transformed into meaningful insights for patients and clinicians alike.
Analyses of health-related quality of life showed that, when appropriately adjusted, real-world cohorts can provide representative and clinically relevant outcomes. Building on these findings, the first prediction models of quality of life for this patient group were developed, illustrating the potential to forecast post-treatment well-being and to support shared decision-making.
The thesis further highlights the growing role of real-world evidence alongside randomized clinical trials. By combining registry data with single-arm studies, it was possible to evaluate novel therapies and generate insights of value to both clinical practice and regulatory decision-making.
Finally, new approaches to survival prediction were explored. By examining existing models, enriching them with additional features, and developing new models using causal inference, this work takes steps toward addressing the critical “what if” question of treatment choice—laying the foundation for more personalized and clinically actionable predictions.
Altogether, this thesis illustrates how the thoughtful application of statistical and machine learning methods to real-world data can enhance understanding, improve communication of outcomes, and advance the future of personalized cancer care.