guest ::
| Home > Publications database > Confounder control in biomedicine necessitates conceptual considerations beyond statistical evaluations |
| Home > Publications database > Confounder control in biomedicine necessitates conceptual considerations beyond statistical evaluations |
| Preprint | FZJ-2024-07530 |
; ; ; ;
2024
This record in other databases:
Please use a persistent id in citations: doi:10.1101/2024.02.02.24302198
Abstract: Machine learning (ML) models hold promise in precision medicine by enabling personalized predictions basedon high-dimensional biomedical data. Yet, transitioning models from prototyping to clinical applications poseschallenges, with confounders being a significant hurdle by undermining the reliability, generalizability, andinterpretability of ML models. Using hand grip strength (HGS) prediction from neuroimaging data from theUK Biobank as a case study, we demonstrate that confounder adjustment can have a greater impact on modelperformance than changes in features or algorithms. An ubiquitous and necessary approach to confounding isby statistical means. However, a pure statistical viewpoint overlooks the biomedical relevance of candidateconfounders, i.e. their biological link and conceptual similarity to actual variables of interest. Problematically,this can lead to biomedically not-meaningful confounder-adjustment, which limits the usefulness of resultingmodels, both in terms of biological insights and clinical applicability. To address this, we propose a two-dimensional framework, the Confound Continuum, that combines both statistical association and biomedicalrelevance, i.e. conceptual similarity, of a candidate confounder. The evaluation of conceptual similarityassesses on a continuum how much two variables overlap in their biological meaning, ranging from negligiblelinks to expressing the same underlying biology. It thereby acknowledges the gradual nature of the biologicallink between candidate confounders and a predictive task. Our framework aims to create awareness for theimperative need to complement statistical confounder considerations with biomedical, conceptual domainknowledge (without going into causal considerations) and thereby offers a means to arrive at meaningful andinformed confounder decisions. The position of a candidate confoudner in the two-dimensional grid of theConfound Continuum can support informed and context-specific confounder decisions and thereby not onlyenhance biomedical validity of predictions but also support translation of predictive models into clinicalpractice.
|
The record appears in these collections: |
