Experiments: AI streamlined everyday patient work, reduced paperwork and freed up professionals’ time for care work

With Sitra’s funding support, Pohde, the Wellbeing Services County of North Ostrobothnia, developed an AI application in Oulu that uses various language models for patient record keeping. During the experiment, the application progressed to pilot use in patient work.

What was achieved?

Patient record keeping is an important and necessary part of care work, but it is often a time-consuming routine for doctors. If documentation were done faster and with high quality using technology, doctors would have more time for what matters most: patient care.

“Doctors have a calling to get their patients well. The more the computer can ease the burden of documentation or creation of referrals, the more time doctors have for care work. Interaction with patients is smoother when there is no need to simultaneously record information on paper or into the system,” describes Acting Chief Information Officer Heikki Mikkonen at Pohde.

The experiment resulted in a functional AI solution to assist healthcare staff and potentially support social welfare services as well.

During a doctor’s appointment, the application monitors the conversation between the doctor and the patient and produces a draft patient record for the patient information system based on the conversation. The doctor reviews the draft. The final documentation is the doctor’s responsibility, and AI does not automatically enter anything into the system.

Dozens of professionals from fields such as ear, nose and throat diseases, hand surgery, neurology, oncology, psychiatry, social services and primary health care participated in the development work. The care personnel felt that the topic was important.

In collaboration with the healthcare professionals, the experiment examined the clarity and comprehensibility of the drafts produced by AI. For example, AI must be able to summarise the essential points from the patient’s narrative. The components of the documentation include the reason for the visit, the patient’s preliminary information, current status, care plan and diagnoses.

“AI surprised us all, it was that good. It was a great moment when we were able to truly produce patient documentation,” says Mikkonen.

What was learned?

Based on the experiment, it was assessed that not only doctors but also nurses and other social welfare and healthcare professionals would benefit from the dictation tool. AI could be capable of compiling patients’ preliminary information into a readable format, potentially saving several minutes of doctors’ working time per visit. AI might also successfully compile and update care plans.

Originally, the experiment intended to use language models operating locally in Pohde’s network so that client documents would be kept in the wellbeing services county’s own network. This would have made the processing of personal data easier than when using language models operating in a cloud service.

However, the locally running non-fine-tuned language models did not function as desired, and there were deficiencies in their performance. The developers of the experiment decided to use cloud-based language models of a major technology company. Their use required a data protection impact assessment for the processing of patient data, which is a demanding and laborious process. Lessons were shared by another wellbeing services county that had already addressed similar challenges in its own AI experiments.

“We received tips on data protection impact assessment from the Wellbeing Services County of Western Uusimaa. That sped up our work by a few weeks,” says Development Manager Timo Alalääkkölä at Pohde.

What is next?

The development work will continue in the wellbeing services county. The aim is to improve the reliability and performance of the application and expand its use to new specialised fields.

Participants in the experiment

Pohde – the Wellbeing Services County of North Ostrobothnia, Esko Systems Oy, the University of Oulu

Oma Häme, the Wellbeing Services County of Kanta-Häme, aimed to automate the production of patient visit summary texts and the creation of statistical entries based on the text. The goal of the automation was to save the care personnel’s time and make it easier for the professionals to focus on patient interaction. The aim was also to improve the quality of documentation and broaden its content.

What was achieved?

In the experiment, Oma Häme’s virtual scribe was piloted in Riihimäki in the primary health care services.

Currently, documentation is a multiphase process in which the doctor first takes notes based on the patient’s narrative and then dictates a summary of the visit based on those notes, which then is transcribed, submitted for review and finally entered into the systems by another professional. The patient may only be able to read the text in the MyKanta service several days later. AI-assisted documentation saves multiple steps and can produce an almost complete entry during the consultation itself.

Oma Häme’s virtual assistant records an audio file of the consultation, which is converted into text. The AI produces a summary of the text for documentation purposes, and the doctor checks that it corresponds to the course of the visit and the discussion that took place during the appointment.

Potential to free up time for other tasks

In the pilot use, doctors estimated that the AI produced an 80 per cent complete summary, which was almost ready to be transferred to the patient information system without major changes. This result was achieved using general language models that had not yet been trained with special terminology. When the models are further trained and tailored to be used in various special fields of social welfare and health care, the documentation can become even more accurate.

“Depending on the professional group and the type of visit, documentation currently takes 10 to 45 minutes per consultation. Even if the professional must make corrections, AI still brings significant time savings,” estimates Chief Data and Impact Officer Katja Antikainen at Oma Häme.

Oma Häme has estimated that if one minute of working time were saved from the documentation at each visit, it would free up a total of around 20 person-years annually for other tasks.

“If we can save even ten minutes per documentation, we are talking about really significant numbers,” Antikainen says.

More understandable texts for patients

The experiment also yielded a surprising result: AI was able to produce text that was more human-centred and understandable from the patient’s perspective than that of doctors as professionals.

“Doctors tend to write as professionals, but when the patient reads the text in MyKanta, it is not suitable for them. AI brought the documentation closer to everyday language, and this was an eye-opener for the professionals,” explains Chief Development Officer Toni Suihko  at Oma Häme.

What was learned?

The untrained AI identified diagnoses based on the disease classification from speech, but it was unable to recognise all medication, for example. This problem can be solved by teaching the AI special vocabulary.

The language model could leave out essential information, if it appeared only at the end of the conversation or as a side note. In such cases, the doctor must make a follow-up request to the language model and ask it to include a certain topic in the text. The challenge is that no patient data archive can be created when using AI. The original recording and files are therefore deleted almost immediately.

What is next?

The wellbeing services county of Kanta-Häme has received additional funding of EUR 360,000 as part of DigiFinland’s AI project and the AI Ecosystem in Social and Health Services (SOTE) established by the Ministry of Social Affairs and Health for the AI-based compilation of client background and risk information. The virtual scribe will be developed by including patients’ previous background materials such as patient record texts in the assistant’s work. So far, the assistant has only operated based on information spoken during appointments.

The pilot will be expanded to special fields such as physical medicine and psychiatry. Dozens of users in social welfare and health care will be able to test the solution and assess its suitability for their work.

Participants in the experiment

Oma Häme – the wellbeing services county of Kanta-Häme, Accenture, Digia.

In the wellbeing services county of Western Uusimaa, an AI solution for documenting client and patient information was piloted for the first time in the summer of 2024 at the Nummela Health Station. The wellbeing services county used an application developed by the Finnish start-up Gosta Labs to assist doctors in documenting consultations and telephone calls.

Following promising results, the wellbeing services county expanded the experiment to three new health stations in Kirkkonummi, and Leppävaara and Tapiola in Espoo in the autumn of 2024. Sitra funded this extended experiment.

The aim of the experiment was to gain a better understanding of the use of AI-assisted documentation in doctors’ consultations and the potential for saving working time. Another goal was to improve the experience of professionals and patient satisfaction: professionals could focus more on patient interaction while the information system would handle the documentation tasks under supervision. Increased efficiency could also shorten patient waiting times for treatment.

What was achieved?

The application was further developed in collaboration with the care personnel. The solution was intended to meet the users’ needs as precisely as possible. Different units had varying documentation practices, and the broader user group had different needs and uses for the application.

In practice, the application records the conversation between the doctor and the patient and transcribes it into text using speech recognition technology. An AI language model then creates a draft patient record based on the text that the doctor checks, edits and transfers to the patient information system.

“Many found it important to be involved in the development work and see that their feedback had an impact,” says Veera Vihula, Development Manager at the Western Uusimaa wellbeing services county.

In the expansion phase of the experiment, around 80 doctors participated. They viewed the assistance provided by the AI positively or neutrally. Some felt that they could better focus on interacting with patients during appointments, although the learning process and commitment to the new solution varied among new users.

The short experiment did not yet allow for the verification of actual working time savings.

What was learned?

Expanding the use of the application to new health stations provided more monitoring data and revealed different user needs.

“We were able to significantly strengthen our learning process when we expanded the experiment to new health stations, and the number of users multiplied,” describes Vihula.

Insights were also gained into change management and communicating it to the organisation.

The wellbeing services county’s AI experiments discovered good practices for innovation co-operation with companies, as the development work was carried out in collaboration with a local start-up.

“When you develop something as a pioneer, you must carefully assess risks and consider ethical aspects. Responsible development also means daring to think and act in new ways and being willing to investigate and experiment,” Vihula explains.

“The work is not meant to remain at the level of experiments but to proceed systematically to the next steps of implementation,” she points out.

The wellbeing services county has also identified the need for a national discussion on the target level of adopting AI.

What is next?

The wellbeing services county of Western Uusimaa has received a grant of EUR 550,000 from the Ministry of Social Affairs and Health’s AI project to expand AI-assisted documentation and to carry out the procurement of the solution.

The use of the application will be expanded from doctors to other social welfare and healthcare professionals involved in the documentation.

Participants in the experiment

Wellbeing services county of Western Uusimaa, Gosta Labs

The wellbeing services county of Kanta-Häme piloted an AI application that helps predict the development of older people’s functional capacity. Based on the input data, the application provides predictions of events such as potential declines in functional capacity or falls.

The goal of the pilot was to enable predictive care and rehabilitation for older people and to make the work of home care workers more efficient. Another aim was to find ways to help older people maintain and improve their functional capacity, enabling them to live at home for as long as possible. This could reduce emergency care visits or expensive procedures.

“The realisation of these goals could ultimately be seen in that older people would much later transfer to 24-hour care services, or similarly the need for multiple home care visits per day would be delayed to an older age. It is not only about financial benefits for society but also about quality of life for people,” says Katja Antikainen, Chief Data and Impact Officer at the Kanta-Häme wellbeing services county.

What was achieved?

The functional capacity of older people is typically assessed every six months using an established measurement tool in inpatient care or during home care visits. Based on the data entered into the system by a professional, the AI application provides a prediction of the patient’s functional capacity for the next six-month period.

In the pilot, the application was used by practical nurses, nurses and rehabilitation staff in home care units. The staff received predictive assessments from the application and reviewed them. At the same time, the staff was expected to teach the AI by informing it whether the given assessment was accurate or not. This proved challenging, and not enough feedback was collected to support the development of the AI model.

What was learned?

The experiment did not immediately lead to significant changes in staff practices. The Kanta-Häme wellbeing services county’s Chief Development Officer Toni Suihko assessed that it was possible that professionals did not yet directly see the benefits in their work.

“Our observation was that the application works, but its data base must be expanded so that it would function better and more sensitively predict the development of older people’s condition,” says Suihko.

What is next?

The Wellbeing Services County of Kanta-Häme has received funding of EUR 120,000 as part of DigiFinland’s AI project and the AI Ecosystem in Social and Health Services (SOTE) established by the Ministry of Social Affairs and Health to continue the development of functional capacity prediction.

The application’s data base will be expanded and combined with another more versatile functional capacity assessment tool that covers different age groups and provides a more comprehensive picture of an person’s condition. With the expansion of the data base, the experiment can be extended to rehabilitation units and services for children and young people, for example.

Participants in the experiment

Oma Häme – the wellbeing services county of Kanta-Häme, Avaintec, Raisoft

• The doctor must familiarise themselves with the patient’s preliminary information, which tends to be more extensive the more illnesses the patient has, and the older they are. The information may be located in several places in the patient information systems. During an appointment, the patient describes their current condition.
• The doctor may record their observations into the information system during the appointment. The data may also be entered into the system using dictation, in which case a third party transcribes and saves the information for the doctor to review.
• The doctor may then determine the diagnosis of the patient’s condition and recommend further treatment.
• Finally, all the data generated during the appointment are entered into the patient information system in the required format.

AI can assist in all stages of an appointment with a patient

• AI can compile all preliminary information about the patient into a simple, quickly reviewable format for the doctor.
• During the appointment, AI can convert the information provided by the patient into text.
• Based on this, AI can perform various tasks such as summarising the patient’s condition and making diagnostic and treatment recommendations for the doctor.
• The doctor is responsible for assessing the quality of the conclusions and recommendations produced by AI and for editing them as necessary. Finally, the information can be entered into the patient information system with the help of AI.