PEARL-AI framework for clinicians, academics, administrators and developers
| Recommendation | Action | Fulfillment (Complete/Partial/Absent/Not Applicable) with Elaboration |
|---|---|---|
| Ensure AI used in healthcare promotes children's development and wellbeing | AI development/implementation prioritizes benefitting children and supporting children's well-being | |
| There is no or minimal potential risk of harm to children from use of the AI-enabled system | ||
| Ensure safety for children when AI is used in healthcare | The AI-enabled system used for pediatric healthcare has appropriate human agency, oversight and control measures with a human in the loop as far as possible | |
| A fail-safe design is adopted to assure that the performance of the AI-enabled system does not degrade in unforeseen circumstances | ||
| There is thorough testing for accuracy, safety, security, reproducibility and robustness before the AI-enabled system is deployed on children | ||
| There is a process to continuously assess and monitor the impact of the AI-enabled system on children | ||
| There is clear documentation as to when AI models are created, validated, and set to expire | ||
| The AI-enabled system is secure from unauthorized access, modification and disruption whilst maintaining confidentiality, integrity and availability | ||
| Ensure inclusion of and for children during the design and development of healthcare AI | There is meaningful child participation in AI policies and in the design and development processes | |
| AI development utilizes randomized controlled trials in children where feasible | ||
| Ensure AI used in healthcare for children prioritize fairness, non-discrimination and equitable access | Datasets include children as subjects | |
| Datasets are not biased against certain groups of children based on age, ethnicity, region or socioeconomic status | ||
| Tests for fairness (in children and in particular sub-groups of children) using suitable metrics are performed | ||
| There is equitable access to the AI-enabled system for all children, regardless of socioeconomic status | ||
| Ensure AI-enabled healthcare systems protect children's data and privacy | Data collection is minimized with only necessary data collected | |
| Consent is required from parents and guardians for the collection, use or storage of younger children's data | ||
| Children are able to access, understand the use of, and delete their data in accordance with maturity | ||
| Children who reach the age of consent are able to reverse parental consent previously given | ||
| Ensure that AI in healthcare supports transparency, explainability and accountability for children | Medical professionals, parents and children are informed about the use of an AI-enabled system, how the AI-enabled system makes its decisions, and its limitations | |
| Parents and children are informed regarding how the AI-enabled system uses and maintains confidential data | ||
| The AI-enabled system includes a statement on authorizing its use in children, including evidence that supports use in children or a lack thereof | ||
| Developers use interpretable and not black-box AI for building AI-enabled systems for children, in particular for irreversible decisions with far reaching consequences | ||
| Developers and manufacturers provide accessible mechanisms to medical professionals and families for reporting and escalating concerns on the AI-enabled system |