- Many patients are supportive of the use of their data to improve healthcare and research, especially when it’s in the hands of trusted physicians, medical professionals and researchers….but they feel they should give their permission first.
- That makes sense. What are some of the models by which we can do this? Different settings have different consent models which would appear to be the most ethically sound model.
- Certainly it’s an appropriate model for interventional studies, such as clinical trials or procedures in the clinical setting, but this can be cumbersome, or even unfeasible for large historical data sets of routinely collected data.
- There are also challenges in prospectively deploying such a consent model, particularly when seeking permission from patients for unforeseen future uses of their de-identified data. How can you truly have consent if the patients are asked to sign up for terms and conditions before their episodes of care, and to agree to some future unspecified use of their data about which they cannot reasonably refuse or be fully informed.
- Also, the requirement of consent may even impede or preclude that aggregation of clinical data to benefit populations, leading to additional to substantial costs. Added small costs can add up to a large aggregated sum that can hinder innovation and preclude the generalization of the knowledge that could be of substantial societal benefit.
- As a solution, opt-in models have been proposed. These require affirmative consent and the use of opt-in models have been proposed as almost a lighter touch to the informed consent models
- However, what that really means and translates into, is that only the most engaged patients, the ones who proactively and actively take steps to get involved would be included. And this may have imposed the risk for ethical concerns regarding racial bias in patients who may be excluded from such data sets.
- Opt-out models exists as a third path, and for example, as part of a review into the security and use of NHS data in the UK in 2016, the National Data Guardian recommended that a national opt-out policy and model should be introduced, rather than one based on opt-in consent.
- Yet this also holds potential for bias as people who opt out of a data collection may also be concentrated in particular demographics or in those with particular conditions who would then be subsequently unrepresented or skewed in the representation in data.
Justin Ko, MD. Ethical Considerations in AI. 8th World Congress of Teledermatology, Skin Imaging and AI in Skin diseases – November 2020