Auxiliary modalities add no detectable discrimination or calibration over structured EHR for 30-day COPD readmission: evidence from a leakage-audited MIMIC-IV cohort
- 1University of Padova: Universita degli Studi di Padova
Abstract
BackgroundHospital readmission is a common and costly outcome in patients with chronic obstructive pulmonary disease (COPD). Multimodal deep learning models combining structured electronic health record (EHR) data with physiological, imaging, and textual data have been proposed to improve its prediction, but whether such fusion outperforms EHR data alone, once temporal-leakage safeguards are applied, remains unclear.
MethodsUsing a stratified MIMIC-IV cohort of COPD patients enriched to about 50% readmission (natural prevalence 18.1%; n=1,279; 895/192/192 training/validation/test), we predicted 30-day, all-cause readmission from four frozen, pre-trained encoders (electrocardiogram, chest radiograph, clinical notes, structured EHR). We audited and corrected temporal leakage in each modalitys admission-linkage pipeline, then compared three fusion architectures (naive concatenation, scalar gated fusion, EHR-anchored residual fusion) against unimodal baselines and an EHR-alone reference, using bootstrap significance testing, a missingness stress test, and calibration analysis with and without post-hoc recalibration.
ResultsThe audit corrected three previously undetected temporal-alignment issues, affecting 5-9% of patients for clinical notes, 29-43% for chest radiographs, and 53-58% for ECG. After correction, every modality except EHR showed near-chance discrimination alone (ROC-AUC 0.46-0.53), confirmed with a non-linear adapter probe. No fusion architecture significantly outperformed the EHR-alone baseline (ROC-AUC 0.63); removing auxiliary modalities improved rather than degraded fusion performance under stress testing. The raw Brier score was lower for all three fusion architectures than for EHR-alone (about 0.24 versus 0.32), but this reflected overconfidence of the EHR-alone probe (calibration slope 0.12); after logistic recalibration fitted on the validation set, Brier scores were indistinguishable (EHR-alone 0.239; fusion 0.240-0.244; all paired bootstrap 95% CIs included zero).
ConclusionsStructured EHR data alone matched the discrimination and, after recalibration, the calibration of more complex multimodal fusion for 30-day COPD readmission prediction. These findings highlight the importance of rigorous temporal-leakage auditing in multimodal clinical prediction research and suggest that reformulating this task around temporal proximity to the outcome, rather than refining the fusion mechanism further, may be more productive for future work.
Related
- Does Replacing SOFA-1 with SOFA-2 Change Mortality Discrimination and Calibration Across Patient Subgroups? A Paired Analysis in MIMIC-IV
- Discordance between reported discrimination and translational-readiness evidence in AI-based ASCVD prediction: a systematic review with structured methodological audit
- Parkinson’s disease is associated with increased 90-day readmission and reoperation after elective total hip arthroplasty
- Machine Learning Prediction of 90-Day Readmission and Mortality in General Internal Medicine: A Retrospective Cohort Study
- Inflammatory Biomarkers and Left Atrial Remodelling for Early Risk Stratification of 6-Month Readmission in Ischaemic and Idiopathic Dilated Cardiomyopathy
BibTeX
@misc{pozza2026auxiliary,
title = {{Auxiliary modalities add no detectable discrimination or calibration over structured EHR for 30-day COPD readmission: evidence from a leakage-audited MIMIC-IV cohort}},
author = {M. Pozza and N. Navarin and S. Gabrielli},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.14.26365503},
url = {https://www.medrxiv.org/content/10.64898/2026.10.14.26365503v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Auxiliary modalities add no detectable discrimination or calibration over structured EHR for 30-day COPD readmission: evidence from a leakage-audited MIMIC-IV cohort
AU - M. Pozza
AU - N. Navarin
AU - S. Gabrielli
PY - 2026
DA - 2026/10/14
PB - medRxiv
DO - 10.64898/2026.10.14.26365503
UR - https://www.medrxiv.org/content/10.64898/2026.10.14.26365503v1
AB - BackgroundHospital readmission is a common and costly outcome in patients with chronic obstructive pulmonary disease (COPD). Multimodal deep learning models combining structured electronic health record (EHR) data with physiological, imaging, and textual data have been proposed to improve its prediction, but whether such fusion outperforms EHR data alone, once temporal-leakage safeguards are applied, remains unclear. MethodsUsing a stratified MIMIC-IV cohort of COPD patients enriched to about 50% readmission (natural prevalence 18.1%; n=1,279; 895/192/192 training/validation/test), we predicted 30-day, all-cause readmission from four frozen, pre-trained encoders (electrocardiogram, chest radiograph, clinical notes, structured EHR). We audited and corrected temporal leakage in each modalitys admission-linkage pipeline, then compared three fusion architectures (naive concatenation, scalar gated fusion, EHR-anchored residual fusion) against unimodal baselines and an EHR-alone reference, using bootstrap significance testing, a missingness stress test, and calibration analysis with and without post-hoc recalibration. ResultsThe audit corrected three previously undetected temporal-alignment issues, affecting 5-9% of patients for clinical notes, 29-43% for chest radiographs, and 53-58% for ECG. After correction, every modality except EHR showed near-chance discrimination alone (ROC-AUC 0.46-0.53), confirmed with a non-linear adapter probe. No fusion architecture significantly outperformed the EHR-alone baseline (ROC-AUC 0.63); removing auxiliary modalities improved rather than degraded fusion performance under stress testing. The raw Brier score was lower for all three fusion architectures than for EHR-alone (about 0.24 versus 0.32), but this reflected overconfidence of the EHR-alone probe (calibration slope 0.12); after logistic recalibration fitted on the validation set, Brier scores were indistinguishable (EHR-alone 0.239; fusion 0.240-0.244; all paired bootstrap 95% CIs included zero). ConclusionsStructured EHR data alone matched the discrimination and, after recalibration, the calibration of more complex multimodal fusion for 30-day COPD readmission prediction. These findings highlight the importance of rigorous temporal-leakage auditing in multimodal clinical prediction research and suggest that reformulating this task around temporal proximity to the outcome, rather than refining the fusion mechanism further, may be more productive for future work.
ER -