Medicine Discovery
How should cardiovascular multimodal AI account for incomplete or stale evidence when deciding whether to issue a prediction?
This correspondence separates conventional fusion, missing-modality robustness, temporal modeling, selective prediction and interpretation of why information is missing. It proposes representing observation state alongside patient state and assessing evidence sufficiency before producing a prediction. Figure 1 summarizes six principles. The authors suggest five candidate evaluation conditions, including incomplete inputs, natural co-missingness, temporal staleness and external data, and argue that deferral should lead to a predefined care pathway. This is a conceptual proposal that requires prospective evaluation and broader consensus; it is not a validated algorithm, clinical standard or demonstration of improved patient outcomes.
Yahui Han, Qiang Li, Taiqi Zhao, Shanshan Cai, Biyun Li, Kaisaierjiang Kadier, Tianshu Gu, Zhonghua Sun, Pengyuan Xu, Dezhi Liu, Moyan Li, Xufeng Huang, Qian Guo, Yuan Cheng, Honglin Zheng, Ge Zhang. Multimodal AI for cardiovascular care must learn when not to predict. Medicine Discovery. 2026; article 100004. https://doi.org/10.1016/j.medd.2026.100004
Multimodal AI for cardiovascular care must learn when not to predict
A correspondence proposes an intended-use-specific framework for missing modalities, asynchronous observations and deferral in cardiovascular AI; it reports no new dataset or model performance.
Numbers describe the conceptual framework, not patients, external cohorts, performance estimates or trial evidence.
Conceptual taxonomy of fusion, missing inputs, time, selective prediction and observation mechanisms.
Source locator: Corrected proof p. 1, opening paragraph
Figure-based conceptual proposal; deferral is distinct from calibration or confidence alone.
| Measure | Value |
|---|---|
| principles in figure | 6 |
| evidence type | Conceptual framework, not an evaluated intervention |
Source locator: Corrected proof pp. 2-3; Fig. 1 caption and paragraphs on abstention/deferral
Suggested minimum suite requiring future validation and stakeholder consensus.
Source locator: Corrected proof p. 3, paragraph beginning Evaluation should be defined by intended use
It addresses when multimodal cardiovascular AI has enough current and clinically relevant evidence to predict, and when it should defer.
Deferral follows prespecified evidence-adequacy criteria and should trigger an alternative care pathway; it is not synonymous with calibration or high model confidence.
Evidence: KF2
Five candidate conditions are complete inputs, a critical modality missing, natural co-missingness, temporal staleness and geographically or temporally external data.
Evidence: KF3
| Row ID | Component | Context | Finding | Evidence Level | Scope | Source Locator | Evidence refs |
|---|---|---|---|---|---|---|---|
| evidence-kf1 | The correspondence separates distinct problems that are often grouped under multimodal AI. | Conceptual taxonomy of fusion, missing inputs, time, selective prediction and observation mechanisms. | problem categories: 5 | Author perspective and methodological proposal | Attribution of the stated evaluation and deferral proposal to this correspondence. | Corrected proof p. 1, opening paragraph | KF1 |
| evidence-kf2 | The proposed framework makes evidence sufficiency and a defined downstream pathway central to deferral. | Figure-based conceptual proposal; deferral is distinct from calibration or confidence alone. | principles in figure: 6; evidence type: Conceptual framework, not an evaluated intervention | Author perspective and methodological proposal | Literature discussion of missingness, temporal context and prospective evaluation needs. | Corrected proof pp. 2-3; Fig. 1 caption and paragraphs on abstention/deferral | KF2 |
| evidence-kf3 | The authors propose intended-use-specific evaluation conditions for incomplete and temporally heterogeneous information. | Suggested minimum suite requiring future validation and stakeholder consensus. | candidate evaluation conditions: 5: complete inputs, critical modality missing, natural co-missingness, staleness, external data | Author perspective and methodological proposal | Literature discussion of missingness, temporal context and prospective evaluation needs. | Corrected proof p. 3, paragraph beginning Evaluation should be defined by intended use | KF3 |
It addresses when multimodal cardiovascular AI has enough current and clinically relevant evidence to predict, and when it should defer.
Evidence: KF1, KF2
Deferral follows prespecified evidence-adequacy criteria and should trigger an alternative care pathway; it is not synonymous with calibration or high model confidence.
Evidence: KF2
Five candidate conditions are complete inputs, a critical modality missing, natural co-missingness, temporal staleness and geographically or temporally external data.
Evidence: KF3
No. It is a correspondence presenting conceptual proposals, with no new dataset or validated performance result.
Evidence: KF1, KF3
Study-level author-maintained evidence summary. Not participant-level data and not a ready-to-pool meta-analysis dataset. For formal citation, use the original article DOI.
Download study-level evidence CSV (UTF-8)Formula safety: cells starting with =, +, -, @ (after leading whitespace), or tab/CR/LF are prefixed with an apostrophe. Treat all cells as text; remove only that safety prefix when recovering the original string.
These records identify the same original work; database inclusion is not evidence of scientific validity or clinical benefit.