Cardiologists invent patients so heart notes can be shared
Doctors wrote artificial hospital reports in seven languages, so AI can learn from clinical notes with no real patient behind them.



Picture made for ABUREAU.
Real patient notes are the data health AI most needs and the hardest to share. A European team found a way round it: write the patients. Cardiologists from seven hospitals wrote CliTe, 699 artificial cardiology reports in seven languages, covering outpatient, discharge and inpatient notes.
The reports cover ischaemic heart disease, heart failure and valvular disease. A second set, ParaCliTe, holds reports from 94 fictitious patients, each in all seven languages with sentence-level alignment, 658 documents in total. A third part is machine-translated and labelled as such.
The authors say the sets are meant for building AI tools in cardiology across languages. The paper page does not state a licence for the sets themselves or give a download link, so anyone planning to use them should check the supplement first.
For people who own or use data, this shows a middle path between locking records away and sharing them. Expert-written, fictional records carry clinical skill without carrying a patient. The open question is how closely they match real notes, and that is what buyers will want measured.

