Quantifying Symptom Causality in Clinical Decision Making: An Exploration Using CausaLM
The work investigates how symptoms like 'chest pain' causally influence diagnostic predictions using CausaLM and TReATE to move beyond simple correlation. By creating counterfactual text representations, researchers quantify the causal effect of a symptom on a model's output, showing that certain diseases are significantly influenced by the presence or absence of specific symptoms.