Moving beyond correlation to estimate how treatment choices may change outcomes for individual patients — combining causal inference, counterfactual reasoning and foundation AI models.
Many clinical AI systems learn useful associations from historical data. Yet a treatment decision is an intervention: the research challenge is to estimate the effect of choosing one option instead of another for a particular patient, under uncertainty and real-world data constraints.
Finds patterns between patient characteristics, treatments and outcomes.
Models how an intervention may alter an outcome for a given individual or subgroup.
The research investigates whether pretrained representations can make causal modelling more practical when biomedical datasets are limited, heterogeneous or noisy — without treating a foundation model as a substitute for causal assumptions.
Estimate the contrast between alternative treatment outcomes for a specific patient profile.
Investigate how pretrained representations can help under constrained sample sizes.
Test how recommendations behave under confounding, missingness and population shift.
This interactive laboratory uses a small synthetic dataset created for demonstration. The values are not clinical recommendations and are not derived from real patients.
Synthetic demo · educational/research prototype only.
Ask about the research problem, causal inference, counterfactual reasoning, foundation models or how the proposed system could be evaluated.
Map causal recommendation, treatment-effect estimation and foundation-model literature.
Build reproducible prediction and causal baselines across representative biomedical settings.
Evaluate representations, adaptation strategies and data efficiency.
Develop and test methods for personalised treatment-effect reasoning.
Stress-test confounding, missing data, subgroup shifts and uncertainty.
Translate model outputs into interpretable research demonstrations.
Connect the research concept with the project supervisors and explore their official university profiles.