Abstract
Predictive feature selection and importance scores identify variables associated with an outcome, but they do not by themselves indicate which features would change that outcome under plausible perturbations. Our long-term goal is a practical procedure for prioritizing causal features in high-dimensional tabular data; Conditional Intervention Analysis (CIA) is a step in that direction. CIA learns conditional distributions of each feature given the others, samples counterfactual values on the observed feature manifold, and estimates finitedifference outcome contrasts with cross-validated confidence intervals and significance tests. On synthetic data with known causal structure, ranking by effect magnitude recovers all true causal features while Pearson correlation misses one third. On ten age-corrected NIH Toolbox scores from the Adolescent Brain Cognitive Development (ABCD) Study
(n = 9,551), CIA highlights parent/caregiver education, neighborhood deprivation, demographics, and morphometry as features with consistent interventional evidence. We state conditions under which CIA estimates converge to a local interventional quantity, and observational applications remain hypothesis-generating rather than confirmatory. Code: https://github.com/marafathussain/Causality_Analysis.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MultiTab_023.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=zBURauCzPr
BibTex
@InProceedings{HusMoh_Conditional_MICCAISAT2026,
author = { Hussain, Mohammad Arafat AND Du, Liyan AND Grant, P. Ellen AND Ou, Yangming},
title = { { Conditional Intervention Analysis: Toward Identifying Causal Features in Tabular Multimodal Data } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17263},
month = {pending},
page = {pending}
}