Ethical Dimensions of AIMDs According to Primary Responsible Party: Clinicians During Deployment, Governance by Administrators and Professional Societies, and Governance by State and Federal Regulators
| Ethical dimension | Clinicians during deployment | Governance by administrators and societies | Governance by regulators |
|---|---|---|---|
| Welfare | Ensure familiarity with performance and limitations of AIMDs before use | Ensure clinicians are trained in AIMD performance and limitations | Ensure that clinical utility, not just in silico performance, is evaluated and declared |
| Avoid off-label or unevaluated use of AIMD | Avoid automation bias through technical and policy interventions | Ensure that legal liability for harm is clearly demarcated | |
| Autonomy | Ensure consent to direct data collection and that risks, benefits, and limitations of AIMD are declared to patient | Ensure clear protocols for declaring risks and benefits of AIMD to patients | Ensure that performance and limitations are declared in ways that enable patient and clinician decision-making |
| Consider notification for secondary reuse of data in learning AIMDs | Promote legitimacy by involving patients in policy priority setting for AIMD use | Promote techniques that explain salience of input features at population level | |
| Justice | Carefully justify use of sensitive attributes (race, sex, etc.) as input features | Ensure regular auditing of performance disparities by race, sex, etc. | Require evaluation of performance disparities by race, sex, etc. |
| Avoid conflicts of interest as clinician/researcher | Ensure clear mechanisms for managing clinician/researcher conflicts of interest | Promote broad access to AIMDs through subsidy and regulation |
Use of term ensure or required denotes legally or ethically required conduct; use of term promote or avoid denotes ethically ideal conduct.