Sapporo Medical University and Fujitsu in Joint R&D for Diabetes Treatment that Uses AI-based Machine Learning

New Delhi, February 02, 2019: Sapporo Medical University, Fujitsu Limited, and Fujitsu Hokuriku Systems Limited today announced that this month they will begin development of an AI machine learning model to optimize prescription of oral hypoglycemic medicines in the treatment of diabetes, as part of joint research on AI use with clinical data.

This joint R&D project aims to support medical examination and treatment through AI prediction of the effects of medicines on diabetes patients. By applying machine learning to patient test values and the prescription data of oral hypoglycemic medicines, a type of diabetes treatment, the organizations aim to develop technology that can predict the effects of treatment. This is done in order to keep HbA1c levels(1) below 7.0%, which is a common target to prevent complications in diabetes. The target data consists of information stored in systems including electronic medical record systems and diagnostic data warehouses(2).

This technology is expected to make it possible to select and prescribe oral hypoglycemic medicines optimized for patients who do not need insulin preparation, taking combinations of medicines and other factors into account. Going forward, the three organizations will further improve the accuracy and general applicability of the results of this project, and will jointly conduct R&D applications of AI technology for other diseases.

Background

Expectations surrounding the use of AI in the medical field have recently been increasing, and there are hopes for the early implementation of AI in the treatment of lifestyle diseases, including diabetes, with regard to prevention and making treatment more efficient. Diabetes requires continual control to keep blood glucose levels within normal boundaries, with the treatment goal being maintaining HbA1c levels of less than 7.0%, a common target to prevent complications. To achieve this it is important to appropriately provide drug treatment using oral hypoglycemic medicines or insulin preparation, in addition to general treatment methods, including diet and exercise. In long-term treatment, however, coexisting illnesses can lead to complexities in a patient’s condition, and from among the many different oral hypoglycemic medicines there is currently no established method for determining prescriptions for oral hypoglycemic medicines in light of considerations like selection, combinations, sequencing, and side effects of medicines.

Summary of the Joint Research & Development

Based on the clinical insights, dataset creation technology, and AI technology of a research group centered on clinicians and on AI engineers, the three organizations of Sapporo Medical University Hospital, Fujitsu, and Fujitsu Hokuriku Systems will create an input dataset from the large volumes of clinical data stored in systems such as electronic medical record systems and clinical data warehouses, build an AI machine learning model primarily based on open source software(3), and evaluate the applicability of the system.

1. Trial details

Machine learning will be conducted using as training data a dataset extracted from clinical data warehouses and business intelligence tools(4) that hold information such as medical records, test values, and prescription data for about 5,000 diabetes patients examined at Sapporo Medical University Hospital, in a format that deletes personal information. This machine learning process will create trained models that will predict the effects of medicines. These trained models will be evaluated on measures such as the area under the curve (AUC) (5), accuracy, and replicability demonstrating their performance. They will also be evaluated for their ability to contribute to the optimization of the prescription of oral hypoglycemic medicines.

2. Technology under development

a. Technology to create a highly accurate dataset

A dataset will be created from prescription data for oral hypoglycemic medicines, as well as test values, for subject patients. As part of the creation of this dataset, the blood glucose control status of subject patients will be expressed using the fluctuation of HbA1c levels over time. 2.

b. Creation of trained models through AI technology

Using the dataset created with the previous technology, the organizations will create trained models to predict the effects of treatment, by using machine learning to train models on the relationships between factors such as test values, types and combinations of medicines, and treatment success or failure results based on patterns of changes in HbA1c levels. In addition, because it is essential for the basis of inferences made by AI technology to be easy to understand in computer systems that will form the foundations of society, such as hospital information systems, the organizations will take the selection of algorithms to be used into consideration. It is expected that this will reveal clinical insights hidden in prescription data and test values, making it possible to select treatments suited to individual patients.

http://www.acnnewswire.com/topimg/Low_OverviewDevelopmentProcess.jpg
Figure: Overview of the development process

3. Vision for the future

The organizations expect that, in the near future, the results obtained from this joint research and development will enable clinicians to select medicines suited to individual patients by displaying information predicted by AI technology, such as the probability that selecting certain oral hypoglycemic medicines will make it easier for the patient to safely and effectively control future blood sugar levels, based on individual characteristics. Moreover, by connecting this system with Fujitsu’s electronic medical record system, the organizations aims to contribute to even better medical services by providing greater efficiencies in the process of clinician’s prescription.

(1) HbA1c levels Short for hemoglobin A1c, this refers to the ratio of hemoglobin bound to glucose in blood, which is used as an indicator of blood glucose in diabetes treatment, as it reflects the average levels of blood sugar over the previous 1-2 months.
(2) Data warehouses Refers to systems for collecting and storing data from mission-critical systems for use in data analysis
(3) Open source software A general term for software whose source code can be freely used, changed, extended, redistributed, and so on.
(4) Business intelligence tool Refers to systems and techniques that collect, accumulate, analyze and report on organization data and make it useful for decision making.
(5) Area under the curve Refers to the area of the lower part of the curve expressing the change in true positive rate and false positive rate of discrimination of the trained model as a two-dimensional graph. It is expressed as a value from 0 to 1, indicating that the discrimination performance close to 1 is higher.

Corporate Comm India(CCI NewsWire)