Final report of the RJ Sabbatical project titled: Artificial Intelligence analyzes patient records. Is this possible and could this improve health care? By Hercules Dalianis

The project and textbook that has been written is the result of a 10 year long research project at DSV / Stockholm University. The research project began in 2007 by obtaining research funds from Vinnova to collaborate with Stockholm County Council to summarize patient records. The original idea was to facilitate the doctor to write a discharge letter of the journal. A discharge letter is a summary of the time of care that is written when the patient is printed. A discharge letter contains the conclusions of the health care and provides advice for home treatment after the discharge from the clinic.
The project grew and more research questions appeared and more people were linked to the project. The research group Clinical Text Mining group consisted of two professors, five doctoral students, one university lecturer, and two doctors (one of the professors were doctors) when it was as largest. The project was also expanded with a Nordic network with another thirty people.

·        The main results of the project and publications and a discussion about them.
The main result of the Sabbatical project is a textbook in English with the title: "Clinical text mining: Secondary use of electronic patient records" to be published by Springer Verlag in April 2018 as Open Access. The book is an introduction to the research area analysing patient records with natural language processing methods.
The computerized patient records are now almost standard worldwide, the systems are centralized and large amounts of electronic patient records are produced describing the care of individual patients. This information is very valuable if it can be reused.

The book describes the background to patient records, electronic patient records, computerized patient record systems, the requirements of doctors and healthcare professionals on a patient record system and the language of patient records. The book also explains the different classification systems used in healthcare such as ICD-10 diagnostic codes, SNOMED CT and ATC drug codes, etc. The book further describes the building blocks of NLP (natural language processing) in computer science literature and how these are adapted to clinical text, and to the different classification systems. A large part of the problem in clinical text mining is to extract information from the unstructured free text and make it structured and use it together with existing structured data contained in the patient record, which encodes ICD-10 diagnosis, drug codes, times, blood values, etc.

The textbook continues by describing computer science methods as rule-based and machine-learning based methods (so-called artificial intelligence), including text mining.
The book describes the ethical problems of using patient records and how to solve them and how to avoid sensitive information about patients being distributed by identifying and pseudonymizing patient records and how to store the records safely.
The book finalises with describing a number of applications in clinical text mining. The vast majority of applications are at the prototypes and will be used in practice only in the future. All described applications use text (and data) in electronic patient records to facilitate for doctors, nurses to work on daily care and patient care. These applications are support for writing and reading the patient record, spell check and synonyms extraction. Further on are applications described to get a quick overview of the journal, so-called automatic text summarisation. Other applications are for medical researchers to be able to find new hypotheses to be proven, information retrieval and automatic clustering of texts in order to find drug side effects. For hospital management to be able to follow and analyse health care, eg. To detect and predict healthcare associated infections. The book also describes pathology reports, explaining methods for extracting textual information from pathology reports to insert them into the cancer registry database.
The book also describes the research front in the field, both for patient records written in English, Swedish and in several other languages.
The textbook is the first of its kind in the area of clinical text mining of electronic patient records written towards an audience of both health professionals and computer scientists.

· What the project has resulted in addition to the publications
The project has resulted in several new contacts, including the Center for Health Informatics, the Australian Institute of Health Innovation (AIHI), Macquarie University in Sydney, the Capital Markets Cooperative Research Center (CMCRC) at Sydney and Australian National University in Canberra. Several researchers and PhD students plan to visit my research group Clinical Text Mining group at DSV / Stockholm University and Sweden in the coming years.

· New research questions generated through the project
No new research issues that were known

New research questions generated through the project
No new research issues that were known before were generated, however, new knowledge and connections have been made to existing research and to reality, these skills are explained in the book.

· International project of the project
The Sabbatical project was conducted at CSIRO (Commonwealth Scientific and Industrial Research Organization), an Australian government research institute located in several locations in Australia with over 5,000 employees. I was stationed in Sydney.
I was also invited to Macquarie University in Sydney, located on the same campus in Marsfield, Sydney as CSIRO. Macquarie University has the same size as Stockholm University with 5000 employees and 50,000 students.

The research team at CSIRO and Macquarie University work together in Natural Language Processing and with health-related data. Australia is generally very successful in clinical text mining, with several groups working in the field. This is a natural step because Australian healthcare is very good and the requirements to improve and measure it are large.
I visited research groups in Australia at the CSIRO / Australian eHealth Research Center (AeHRC) in Brisbane, the Capital Markets Cooperative Research Center (CMCRC) at Sydney and Australian National University in Canberra and the company HLA Global in Sydney, Australian Institute of Health Innovation (AIHI), Macquarie University in Sydney,
all working with various aspects of health informatics and clinical text mining.
Another reason to be in Australia was to get a language bath in English while I wrote the book which facilitated the work with the book.

Other international contacts are obvious with researchers all over the world working with clinical text mining in the Nordic countries, England, Germany, France, Spain, Hungary, USA, Japan and China. Researchers I have contact with and refer to in my book.

· Publication list, as well as links to own web pages
In addition to the textbook, some scientific articles have also been written with my doctoral student Rebecka Weegar and postdoc Aron Henriksson.
https://people.dsv.su.se/~hercules/HDpublications.html

Weegar, R., J. F. Nygård and H. Dalianis. 2017. Efficient Encoding of Pathology Reports Using Natural Language Processing. In Proceedings of Recent Advances in Natural Language Processing, Recent Advances in Natural Language Processing, RANLP 2017, Varna, Bulgaria, pp. 778-783.

Henriksson, A., M. Kvist and H. Dalianis. 2017. Detecting Protected Health Information in Heterogeneous Clinical Notes. Presented at Medinfo. To appear in the Proceedings of Medinfo, Hangzhou, China.

Henriksson, A., M. Kvist and H. Dalianis. 2017. Prevalence Estimation of Protected Health Information in Swedish Clinical Text. In Proceedings of Informatics for Health, Manchester, U.K.

Weegar R. and H. Dalianis. 2016. Mining Norwegian pathology reports: A research proposal. Presented at the Australasian Language Technology Association Workshop (ALTA) 2016.

Some web pages
Riksbankens Jubileumsfond funds a sabbatical where Artificial Intelligence analyzes medical records, Press release, http://dsv.su.se/en/about/news/riksbankens-juileumsfond-funds-a-sabbatical-where-artificial-intelligence-analyses-medical-records

Sabbatical stay at CSIRO and Macquarie University, Sydney, Australia, blog
http://dash.dsv.su.se/2016/11/15/sabbatical-stay-at-csiro-and-macquarie-university/