Table 1. Checklist for ideal algorithms in healthcare.

Desiderata	Criteria	Yes	Location	No	N/A
Explainable     Yesa     Partiallyb     Noc     N/Ad	Feature importance: conveys the relative importance of features in determining algorithm outputs
	Descriptive accuracy: describes what the algorithm has learned (e.g., weights in a neural network)
	Simulatability: clinicians can understand and mentally simulate the model’s process for generating predictions
	Relevance: describes relevancy as judged by the algorithm’s target human audience
Dynamic     Yesa     Partiallyb     Noc     N/Ad	Temporality: captures temporal changes in physiologic signals and clinical events
	Continuous monitoring: performance is reassessed at several time points, including the point at which performance is expected to plateau
Precise     Yesa     Partiallyb     Noc     N/Ad	Data frequency: rate of data collection matches the rate of physiologic changes
	Complexity: algorithm complexity matches the complexity of the prediction or classification task
Autonomous     Yesa     Partiallyb     Noc     N/Ad	Efficiency: the algorithm executes without the need for time-consuming, manual data entry by the end user (i.e., patient, provider, or investigator)
Fair     Yesa     Partiallyb     Noc     N/Ad	Generalizability: algorithm is developed and validated across diverse patient demographics and practice settings
	Selectivity: excludes features that lack pathophysiologic or linguistic association with outcomes, but may introduce bias
	Objectivity: includes variables that are minimally influenced by clinician judgments (e.g., vital signs)
Reproducible     Yesa     Partiallyb     Noc     N/Ad	Generalizability: validated externally, prospectively
	Collaboration: algorithm is shared with the research community
	Compliance: fulfills SPIRIT-AI extension guidelines (if trial) and fulfills CONSORT-AI guidelines

^aOverall adjudication is “Yes” when all criteria are either met or not applicable.

^bOverall adjudication is “Partially” when some but not all criteria are either met or not applicable.

^cOverall adjudication is “No” when no criteria are met.

^dOverall adjudication is “N/A” when all criteria are not applicable.

CONSORT-AI, Consolidated Standards of Reporting Trials-Artificial Intelligence; N/A, not applicable; SPIRIT-AI, Standard Protocol Items: Recommendations for Interventional Trials-Artificial Intelligence.