On February 25,2020, the Consumer Technology Association (CTA) launched the “first-ever industry-led standard for AI in health care”. These standards are the result of the collective working group set up less than one year ago, which includes 52 organizations and member companies such as the American Medical Association Amazon, IBM, Google Inc. or Fitbit.
According to Gary Shapiro, President and CEO, CTA, “this is a major first step – convening some of the biggest players in the digital health world – to help create a more efficient health care system and offer value-based health care to Americans”.
The CTA released its two standards applying to AI in HealthCare:
- ANSI/CTA-2089, providing “Definitions and Characteristics of Artificial Intelligence”; and
- ANSI/CTA-2089.1, providing “Definitions/Characteristics of Artificial Intelligences in Health Care”
Pat Baird, regulatory head of global software standards at Philips, explained the main two ideas provided by these standards: “transparency and a common language (which) will be key to enable the proper and safe functioning of AI”.
For instance, ANSI/CTA-2089.1 defines “de-identified data” as a data that meets two cumulative conditions : 1) “it does not identify an individual”, 2) “there is no reasonable basis to believe the data can be used to identify an individual”. However, it also underlines that “de-identification does not necessarily eliminate all risk of re-identification even if all personally identifiable information is removed”.
These standards also highlight the “unique set of challenges with AI that are specific to applications in health care”, including:
- Data availability and data operability;
- Incorrect predictions;
- Data distribution shift (e.g. when “the performance of AI algorithms is impacted by distribution shift- a change in the statistical properties of the data encountered when the algorithm is deployed compared to the data used to train the algorithm”);
- Clinician trust; and