MedFusionBench: A Multimodal Clinical Dataset and Benchmark for Patient Outcome Prediction

Abstract

The development of multimodal healthcare AI is constrained by the limited availability of standardized datasets that integrate heterogeneous clinical information and support reproducible evaluation. This study introduces MedFusionBench, a multimodal clinical dataset and benchmarking framework designed for patient outcome prediction. The dataset organizes longitudinal patient information into four complementary modalities: structured demographics and diagnoses, laboratory and vital-sign time series, medication histories, and clinical text. A unified preprocessing pipeline performs temporal alignment, missing-value characterization, terminology normalization, and patient-level data quality assessment. Benchmark experiments evaluate conventional machine learning, recurrent neural networks, Transformer-based models, and multimodal fusion architectures across mortality, readmission, and length-of-stay prediction tasks. Additional experiments quantify model robustness under 10%, 20%, and 40% modality missingness and evaluate performance across demographic and temporal subgroups. The results reveal substantial differences between models that are not apparent under standard aggregate evaluation, particularly under incomplete-data conditions. MedFusionBench provides a systematic framework for studying multimodal representation learning, robustness, and generalization in clinical AI.

References

[1] J. N. Acosta, G. J. Falcone, P. Rajpurkar, and E. J. Topol, “Multimodal biomedical AI,” Nature Medicine, vol. 28, pp. 1773–1784, 2022.

[2] M. Xiao, Y. Li, X. Yan, M. Gao, and W. Wang, “Convolutional neural network classification of cancer cytopathology images: Taking breast cancer as an example,” pp. 145–149, 2024.

[3] H. Harutyunyan et al., “Multitask learning and benchmarking with clinical time series data,” Scientific Data, vol. 6, Art. no. 96, 2019.

[4] H. Zheng, L. Zhu, W. Cui, R. Pan, X. Yan, and Y. Xing, “Selective knowledge injection via adapter modules in large-scale language models,” ICAIDE, pp. 373–377, 2025.

[5] A. E. W. Johnson et al., “MIMIC-III, a freely accessible critical care database,” Scientific Data, vol. 3, Art. no. 160035, 2016.

[6] X. Yan, J. Du, X. Li, X. Wang, X. Sun, P. Li, and H. Zheng, “A Hierarchical Feature Fusion and Dynamic Collaboration Framework for Robust Small Target Detection,” IEEE Access, vol. 13, pp. 92953–92964, 2025.

[7] X. Yan, W. Wang, M. Xiao, Y. Li, and M. Gao, “Survival prediction across diverse cancer types using neural networks,” pp. 134–138, 2024.

[8] A. E. W. Johnson et al., “MIMIC-IV, a freely accessible electronic health record dataset,” Scientific Data, vol. 10, Art. no. 1, 2023.

[9] J. Wei, Y. Liu, X. Huang, X. Zhang, W. Liu, and X. Yan, “Self-Supervised Graph Neural Networks for Enhanced Feature Extraction in Heterogeneous Information Networks,” ICMLCA, pp. 272–276, 2024.

[10] Y. Li, X. Yan, M. Xiao, W. Wang, and F. Zhang, “Investigation of Creating Accessibility Linked Data Based on Publicly Available Accessibility Datasets,” pp. 77–81, 2024.

[11] H. Zheng, Y. Ma, Y. Wang, G. Liu, Z. Qi, and X. Yan, “Structuring low-rank adaptation with semantic guidance for model fine-tuning,” ICECAI, pp. 731–735, 2025.

[12] A. Rajkomar et al., “Scalable and accurate deep learning with electronic health records,” npj Digital Medicine, vol. 1, Art. no. 18, 2018.

[13] W. Wang, Y. Li, X. Yan, M. Xiao, and M. Gao, “Breast cancer image classification method based on deep transfer learning,” pp. 190–197, 2024.

[14] B. Shickel et al., “Deep EHR: A survey of recent advances in deep learning techniques for electronic health record analysis,” IEEE Journal of Biomedical and Health Informatics, vol. 22, no. 5, pp. 1589–1604, 2018.

[15] X. Yan, Y. Jiang, W. Liu, D. Yi, and J. Wei, “Transforming Multidimensional Time Series into Interpretable Event Sequences for Advanced Data Mining,” ICHCI, pp. 126–130, 2024.

[16] Y. Li, W. Zhao, B. Dang, X. Yan, M. Gao, W. Wang, and M. Xiao, “Research on adverse drug reaction prediction model combining knowledge graph embedding and deep learning,” MLISE, pp. 322–329, 2024.

[17] X. Yan, J. Du, L. Wang, Y. Liang, J. Hu, and B. Wang, “The Synergistic Role of Deep Learning and Neural Architecture Search in Advancing Artificial Intelligence,” ICEDCS, pp. 452–456, 2024.

[18] R. Miotto et al., “Deep Patient: An unsupervised representation to predict the future of patients from the electronic health records,” Scientific Reports, vol. 6, Art. no. 26094, 2016.

Cover Image
Downloads
Published
2026-04-01
Section
Articles