Health · Medical Research · 5 hrs ago
Hospitals lack capacity to assess healthcare AI’s real-world effects, expert says
Emmanuel Bacry says hospitals cannot reliably measure whether AI improves patient outcomes after the technology is deployed.
He told the Bengaluru Tech Summit that only six or seven hospitals worldwide are equipped to evaluate medical devices in real-world use.
Hospitals often struggle to link patient information and track outcomes such as recovery and deaths after discharge.
An alert system was linked to twice as many deaths in two of six US hospitals where it was used in 2022, though the exact cause was not known.
Bacry said AI needs tested response procedures and must fit into clinical work, rather than simply replacing a human alert.
Laboratory accuracy scores may not show how AI performs in real care, and hidden patterns in data could affect medical decisions in ways clinicians cannot explain.
Bacry estimates that about 70% of healthcare AI training data comes from male, white and Western populations, and he called for smaller models trained on representative local data.
Hospitals worldwide lack the capacity to assess whether healthcare AI systems improve patient outcomes, Emmanuel Bacry said.
Bacry, scientific director of France’s Health Data Hub, said only six or seven hospitals worldwide can evaluate medical devices deployed in their systems.
He cited an alert system deployed in six U.S. hospitals in 2022 that was associated with a doubling of deaths in two of them.
He said laboratory accuracy scores may not predict how AI performs in clinical settings.
Bacry also warned that healthcare AI training data is heavily skewed towards male, white and Western populations.
- Who
- Emmanuel Bacry, scientific director of France’s Health Data Hub.
- What
- He warned that hospitals cannot reliably assess the real-world effects of healthcare AI.
- When
- At the Bengaluru Tech Summit. The article says it was published on November 20, 2026, while its publication metadata gives November 22, 2026.
- Where
- The Bengaluru Tech Summit; the remarks concern hospitals worldwide.
- Why
- Hospitals struggle to access and integrate patient information to track outcomes, making it difficult to measure AI’s clinical and economic impact.
This story does not have two clearly opposing sides.
France’s Health Data Hub was established under French law to facilitate secondary use of health data for research.
An alert system for severe kidney failure was deployed in six U.S. hospitals and was associated with a doubling of deaths in two of them, Bacry said.
Bacry warned that hospitals lack the capacity to assess whether AI systems improve patient outcomes.
- Hospitals able to evaluate devices
- Six or seven worldwide, according to Bacry
- Alert system deployment
- Six U.S. hospitals in 2022
- Deaths associated with the system
- Doubled in two of the six hospitals
- Estimated training-data skew
- Nearly 70% from male, white and Western populations
- Health Data Hub established
- 2019
1 source
What people are saying
There are maybe six or seven hospitals in the whole world today that are able to evaluate medical devices deployed in their system
When you deploy in real life, you can forget about these scores
We are developing an Occidental AI and even narrow types of Occidental AI