A risk score alone doesn’t tell you what to change. This work explores counterfactual explanations — “what would have lowered this risk?” — as a way to make flare predictions actionable.
Interpretable models can show which inputs mattered, but feature importance is often hard to translate into everyday decisions. Counterfactuals frame explanations around plausible changes.
We generated counterfactual explanations for model-predicted flare risk, constrained to factors a person could realistically influence, and reviewed them for plausibility and clarity.
Constrained counterfactuals produced explanations that were easier to understand than raw feature attributions, and more often pointed to factors people recognized from their own experience.
Explanations framed as actionable changes may support shared decision-making, while making it clearer when a model’s reasoning does not match a patient’s experience.
Counterfactual explanations are a promising bridge between model output and everyday self-management. Validation with patients and clinicians is the next step.