Estonian hospitals put AI scribes into doctor-patient visits

Artificial intelligence that records conversations between patients and doctors, writes summaries and helps physicians with diagnoses and next steps is already in use in a number of Estonian hospitals.

Confido chief executive Edvard Garder told ERR that the company began testing AI last year and was among the first in Estonia to do so. He said the system is now used in several departments and that the company plans to extend it across the organisation. Patients have responded positively, he said, because doctors have more time to spend with them.

Narva Hospital is seeking Estonian-language speech recognition and transcription software through a public procurement. The software would turn conversations between doctors or nurses and patients into electronic text that can be processed, automatically translated, played back and stored. The hospital also wants an AI assistant to suggest medical codes and support clinical decisions. The procurement requires doctors to remain the decision-makers, and AI-generated content that a healthcare professional has not reviewed and approved must not be entered into patients' medical records.

Tartu University Hospital began testing AI in its emergency department in mid-October and plans to run the trial for three months. If the technology proves useful, the hospital will extend the trial to confirm the application's benefits. Aime Keis, head of the emergency department, said the EMO24 application is meant to cut the time doctors and nurses spend on documentation and to reduce cases in which patients describe their condition to several staff members.

Heba Clinic uses AI to compile medical histories before appointments, document conversations during visits and prepare summaries and explanations for patients. In an opinion article published by ERR last week, chief executive Ain Aaviksoo said AI cannot make decisions on behalf of doctors. He said it should support tools that help people manage their own health and help doctors assist them in doing so.

Nele Labi, deputy secretary general for innovation at the Ministry of Social Affairs, said the ministry has not carried out a comprehensive assessment of AI use across all hospitals, though it knows of several initiatives. She said private hospitals have generally moved fastest in testing new solutions. There is no single national system for this work, and hospitals are largely choosing tools according to their own needs.

Labi said two major state AI projects are under development: a transcription tool for family medicine and a transcription tool for the electronic ambulance system. She said these two are expected to have the greatest near-term effect on healthcare professionals' administrative workload, although the state has other AI projects as well. The ministry also said projects in other European countries have shown a clear reduction in administrative work. Estonia's approach, Labi said, is to integrate general-purpose tools such as speech recognition and transcription into hospital workflows so clinicians spend time on patients rather than paperwork.

Garder said healthcare professionals generally welcome AI and that scepticism is relatively rare. He sees no significant difference between specialties: automated transcription and clinical decision support can lighten documentation in virtually every field, give doctors more time with patients and get discharge summaries and other reports onto the national health portal more quickly. He said adoption does not bring immediate savings because hospitals must invest first. He believes the technology can reduce workload and burnout among doctors and nurses, while noting that the financial value of better staff well-being cannot always be measured precisely.

Confidentiality is a separate issue. Garder said Confido regards its current solutions as secure and continues to limit risks, and that patients may decline AI during appointments. Tartu University Hospital said participation in the pilot is voluntary; if an emergency patient opts out, information gathering, diagnosis and treatment continue as before. Labi confirmed that strict data-protection rules apply and that patients must be told about recordings and their purpose. Because the technology is moving quickly, she said, greater legal clarity is needed, patients must understand the use, and the process must be transparent. Ministry working groups have already been set up.

Garder said the state should set a standard for security criteria. He does not want one AI system imposed on every provider, arguing that tying all institutions to a single system would create a monopoly, stifle innovation and may not be the most efficient use of public money. He would rather see all data in the national health portal made usable by AI so appointment preparation is faster. In his view, the state is moving far too slowly on AI.