Answers are grounded in this paper and cite their sources.
The study reveals limitations of current clinical prediction models in precision medicine, suggesting a need for causation-based improvements.
Current clinical prediction models are static models of association, rather than correctly specified counterfactual prediction models based on modifiable causes. They do not tell us that, if we change a predictor variable, this will alter the outcome for an individual. Instead, the intervention recommended based on current prediction models is unrelated to the model and only shown to work on average across the population. Consequently, current clinical prediction models are not fully actionable for individuals. To improve actionability, clinical prediction models should be based on causation rather than association, model the individualised treatment effect and predict counterfactuals incorporating the proposed intervention into the model, and provide dynamic predictions which react to changes in treatment. Then prediction models will be closer to meeting the precision medicine ideal.
No takes yet. Share an insight, caveat, or question.
Leighton et al. (2025) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: