How Large Language Models Are Reshaping Modern Healthcare Industry
Artificial intelligence is stepping into the doctor’s surgery – and not merely as an assistant, but as part of the care process itself. Large language models are rapidly transforming how patients are treated, research is conducted, and information is processed. Whilst challenges abound, experts predict a radical transformation in healthcare digitalisation over the next five years.
Martti Asikainen & Elisa Laatikainen 2.6.2025
The integration of language models and artificial intelligence into healthcare is progressing at an accelerating pace. Some compare its disruptive effects to mapping the human genome, or even the birth of the internet. The impacts are expected to be visible in doctor-patient interactions, physicians’ administrative workload, the operational management of hospitals and GP surgeries, medical research, and medical education (Powell 2025).
On the other hand, there’s also healthy scepticism surrounding the adoption of new technologies. For instance, a 2024 American study found that standard large language models (LLMs), such as ChatGPT, Claude, or Gemini, were unable to provide doctors with sufficiently relevant or evidence-based answers to medical questions. By contrast, ChatRWD, a retrieval-augmented generative system (RAG) that combines language models with information retrieval systems, produced useful answers to as many as 58% of questions, compared to conventional language models’ 2-10% (Low et al. 2024).
Despite the challenges, winds of change are blowing strong. For example, in the Western Uusimaa wellbeing area, an AI programme is currently being developed with emphasis on multilingualism and reducing paperwork for healthcare professionals. The region’s R&D director and medical specialist Johan Sanmark predicts that AI will revolutionise working methods, and automatic documentation will become commonplace throughout Finnish healthcare, possibly within five years (Seppänen 2024).
Practical applications set to transform clinical practice
Large language models have also proven promising in answering medical questions. Med-PaLM was the first model to pass US medical licensing examination questions. However, there remain challenges in producing lengthy responses and handling real-world illness and workflow management, which the next version of the model aimed to address through improvements to the underlying language model and domain-specific fine-tuning. The results are impressive. Med-PaLM 2 answered MedQA database-based tests with up to 86.5% accuracy, whilst its predecessor achieved 67.5% accuracy (e.g., Singhal et al. 2025).
Specialised language models have also been developed in Finland for healthcare needs. Researchers have successfully applied large language models to, for example, identifying diabetic retinopathy – the eye disease caused by diabetes – and classifying severity levels from unstructured reports in follow-up studies. The DR-GPT model was trained using over 40,000 patient records from HUS specialist healthcare visits between 2016-2019 (Jaskari et al. 2024). According to the researchers, DR-GPT analyses free-form Finnish medical reports with remarkable accuracy.
Meanwhile, language models could bring expected relief to busy consultation work. In future, consultations might begin with generative AI first reviewing the patient’s history and creating a summary of essential matters (Seppänen 2024). During the consultation, it would listen to the conversation and create a documentation suggestion, which the doctor would then verify. This would significantly reduce the time doctors spend on computers, which in turn would improve both customer experience and wellbeing at work, as well as the physical presence of the doctor or nurse (e.g., Sanmark & Sanmark). Currently, over eight hours of doctors’ total monthly work goes towards documenting reports (Richardt 2024).
Support for data collection and patient record quality
AI can also reduce repetition in data collection and improve patient data quality. When a patient’s previously provided information is systematically available and pre-processed into a clear format, unnecessary repetition of questions is avoided and the risk of errors is reduced. Simultaneously, the patient’s experience of the consultation may improve when matters can be addressed directly without unnecessary administrative delay (Padakanti et al. 2024).
Language model-based systems can also be utilised to support assessment of care urgency and follow-up treatment. Predictive analysis can help, for instance, with prioritising consultations and directing resources (Varnosfaderani et al. 2024). However, not everything is purely futuristic. AI applications are already being utilised in the healthcare sector.
For example, radiologists’ work is supported by an application that helps identify potential findings in X-rays. AI-analysed data can also be used for disease prevention. Additionally, there’s preliminary research evidence suggesting that machine learning might be able to predict mental health diagnoses based on personal patient questionnaires (Haavisto et al. 2023).
Implementation challenges and opportunities
The implementation of AI in healthcare also presents challenges. AI solution development is significantly complicated by the availability of necessary data. Furthermore, due to the special characteristics of healthcare professional vocabulary and practices, genuinely useful and conversational AI would need to be trained not only with professional terminology but also with comprehensive, authentic speech datasets, patient data, and other healthcare sector documentation (Kaartinen 2025).
Although health data has long been digital, obtaining datasets securely for research and product development use remains legally challenging, even though anonymised and consent-based datasets would benefit all parties (ibid. 2025). Regulation also presents its own challenges for AI utilisation. The EU’s data protection regulation, medical device regulation, and the AI Act coming into force in stages from 2024 may slow development (DigiFinland 2024).
Additionally, the phased implementation of AI may create inequality between wellbeing areas and health units. Smaller healthcare entities may not have the resources and necessary expertise to invest in new technologies and integrate them into daily routine procedures. However, if productivity growth were realised even moderately in the public healthcare system through AI, its benefits could easily rise to hundreds of millions of euros annually (Heinäsenaho et al. 2023).
This article has been published as part of the AI Health project coordinated by Haaga-Helia University of Applied Sciences.
Martti Asikainen
RDI Communications Specialist
+358 44 920 7374
martti.asikainen@haaga-helia.fi
Haaga-Helia University of Applied Sciences
Elisa Laatikainen
Project Specialist, Customer Manager
+358 294471359
elisa.laatikainen@haaga-helia.fi
Haaga-Helia University of Applied Sciences
Lähteet
DigiFinland. (2024). Tekoäly hyvinvointialueilla: sosiaali- ja terveydenhuollon käyttötapaukset ja kansallinen edistäminen.
Haavisto, O., Väänänen, A., Varje, P., Taimela, S., Taalas, A., Niemenoja, O. & Nieminen, N. (2023). Mielenterveysdiagnooseja voidaan ennustaa koneoppimisen avulla. Published in Lääkärilehti on 24.10.2023. Accessed 2.5.2025.
Heinäsenaho, M., Äyräs-Blumberg, O. & Lähesmaa, J. (2023). Tekoäly mullistaa terveydenhuoltoa – mahdollisuudet hyödynnettävä viipymättä. Published on the Ministry of Social Affairs and Health website on 14.4.2023. Helsinki. Accessed 2.5.2025.
Jaskari, J., Sahlsten, J., Summanen, P., Moilanen, J., Lehtola, E., Aho, M., Säpyskä, E., Hietala, K. & Kaski, K. (2024). DR-GPT: A large language model for medical report analysis of diabetic retinopathy patients. PLOS ONE, 19(10). Public Library of Science. San Francisco.
Jaskari, J. & Kaski, K. (2024). Tekoäly auttaa tunnistamaan diabeteksen aiheuttaman silmäsairauden – kielimallien käyttö lääketieteessä lisääntyy. Published on Aalto University website on 25.11.2024. Accessed 2.5.2025.
Kaartinen, J. (2025). Tekoäly hoitajien apuna: Uusi tutkimushanke paljastaa teknologian mahdollisuudet terveydenhuollossa. PROFIT project press release published on VTT Technical Research Centre website on 23.1.2025. Accessed 2.5.2025.
Low, Y.S., Jackson, M. L., Hyde, R., Brown, R., Sanghavi, G, H., Alexander, N., Hassan, H., Nene, R. V., Pike, M., Pokrzywa, C., Vedak, S., Yan, A.P., Yao, D-h, Zipursky, A., Dinh, C. ym. (2024). Answering real-world clinical questions using large language model based systems. Cornellin yliopisto.
Padakanti, S., Kalva, P. and Kommidi, V. R. 2024. Transforming Healthcare: Cloud Technologies and AI in Patient Care. International Journal of Engineering and Technology Research. https://iaeme-library.com/index.php/IJETR/article/view/IJETR_09_02_038
Powell, A. (2025). Artificial intelligence is up to the challenge of reducing human suffering, experts say. Are we?. Published in The Harvard Gazette on 20.3.2025. Harvard Magazine. Cambridge. Accessed 1.5.2025.
Sanmark, J. & Sanmark, E. 2024. Mitä tiedämme generatiivisen tekoälyn hyödyistä terveydenhuollossa? Medical journal Duodecim. Accessed 2.5.2025.
Seppänen, A. (2024). Tekoäly mullistaa lääkärin vastaanoton. Published in Lääkärilehti on 1.10.2024. Accessed 2.5.2025.
Singhal, K., Tu, T., Gottweis, J., Sayres, R., Wulczyn, E., Amin, M., Hou, Le., Clark, K., Pfohl, S.R., Cole-Lewis, H., Neal, D., Rashid, Q.M., Schaekermann, M., Wang, A., Dash, D., Chen, J.H., Shah, N.H., Lachgar, S., Mansfield, P.A. & Prakash, S. ym. (2025). Toward expert-level medical question answering with large language models. Nature Medicine, 2025 (31), s. 943-950). Nature Portfolio. London.
Richardt, I. (2024). Tässä ovat terveydenhuollon digitrendit 2024. Published on Terveystalo website on 11.1.2024. Accessed 2.5.2025.
Varnosfaderani, S. M and Forouzanfar, M. (2024). The Role of AI in Hospitals and Clinics: Transforming Healthcare in the 21st Century. Bioengineering. https://www.mdpi.com/2306-5354/11/4/337Medical journal Duodecim
Finnish AI Region
2022-2025.
Media contacts