CUHK develops an accurate machine learning model that uses big data to predict the risk of severe hypoglycemia in the next 12 months among older adults with diabetes

Press Releases

18 Apr 2024

CUHK develops an accurate machine learning model that uses big data to predict the risk of severe hypoglycemia in the next 12 months among older adults with diabetes

CU Medicine's research team developed a novel machine learning model that can predict the risk of severe hypoglycemia in the next 12 months among older adults with diabetes, using the big dataset from the Hospital Authority Data Collaboration Laboratory. With a high positive predictive value of 85%, the model has the potential to be integrated into electronic health record decision support systems for pre-emptive intervention in older adults at the highest risk.
Featured in the photo are the research team members. (From left) Professor Juliana Chan, Chair Professor of Medicine and Therapeutics at CU Medicine; Dr Jones Chan, Consultant, Department of Medicine and Therapeutics at Prince of Wales Hospital; Dr Elaine Chow, Associate Professor; Dr Yang Aimin, Research Assistant Professor; and Dr Shi Mai, Research Associate, from the Department of Medicine and Therapeutics at CU Medicine.

In Hong Kong, one in three adults over the age of 65 have diabetes. To improve the management of this group of older adults, The Chinese University of Hong Kong (CUHK)'s Faculty of Medicine (CU Medicine) conducted a territory-wide analysis of the big dataset from the Hospital Authority Data Collaboration Laboratory (HADCL).

Severe hypoglycemia is a common acute complication in patients with diabetes, which is associated with an increased risk of falls, cardiovascular disease, dementia and all-cause mortality. Using data from HADCL, CU Medicine developed a novel machine learning model that can predict the risk of severe hypoglycemia in the next 12 months among older adults with diabetes. With a high positive predictive value of 85%, the model has the potential to be integrated into electronic health record decision support systems for pre-emptive intervention in older adults at the highest risk. Details of two studies have been published in peer-reviewed international journals PLOS Medicine andDiabetes Research and Clinical Practice.

Study finds diabetes is associated with excess mortality risk in older adults

The HADCL provides anonymised data covering a broad range of information such as patients' medication and laboratory records, hospitalisation, residential area and comorbidity data.

Researchers from CU Medicine performed an analysis of data from HADCL, involving over 1.1 million older adults aged 65 or above from 2014 to 2018. They found that all-cause mortality decreased by 8% overall in older adults in the five-year period. However, those with diabetes remained at a 1.5 to two-fold higher risk of death from cardiovascular disease and other causes compared to those without diabetes.

Dr Yang Aimin, Research Assistant Professor in CU Medicine's Department of Medicine and Therapeutics, said, "Our findings indicate an excess mortality risk associated with diabetes. It is necessary to develop large-scale interventions to improve prevention and self-management of diabetes and its complications."

Severe hypoglycemia - a feared complication in older adults with diabetes

Severe hypoglycemia is another feared complication in the management of diabetes in older adults. Dr Jones Chan, Consultant, Department of Medicine and Therapeutics at Prince of Wales Hospital, stated, "Over 80% of older adults who attend emergency departments in Hong Kong with hypoglycemia are hospitalised. Apart from prolonged hospitalisation, the condition is associated with an increased risk of falls, cardiovascular disease, dementia and all-cause mortality."

To early identify high-risk adults with severe hypoglycemia, CU Medicine's research team collaborated with HADCL to analyse about 1.5 million records of over 360,000 older adults with diabetes from 2013 to 2018. They developed a severe hypoglycemia risk prediction model based on the machine learning algorithm XGBoost. In the novel model, 258 predictors including demographics, admissions, diagnoses, medications and routine laboratory tests in a one-year period were used to predict severe hypoglycemia events requiring hospitalisation in the following 12 months. Validation showed that the resulting model achieved a positive predictive value of 85%.

Potential integration into electronic health record system for pre-emptive intervention

Dr Shi Mai, Research Associate, Department of Medicine and Therapeutics at CU Medicine, said, "Our model outperforms conventional severe hypoglycemia event prediction models based on established risk factors. It is also the first long-term risk prediction model for severe hypoglycemia focusing on older adults with diabetes."

Dr Elaine Chow, Associate Professor, CU Medicine's Department of Medicine and Therapeutics and lead investigator for the research, added, "The newly developed model can be integrated into the local electronic health record system to identify high-risk adults for pre-emptive intervention - for example, by switching to diabetes medications with lower hypoglycemic potential or correcting the timing and dosage of insulin injections."

Professor Juliana Chan, Chair Professor of Medicine and Therapeutics at CU Medicine, concluded, "As the mortality gap between older adults with and without diabetes has not narrowed, we need to take action to address the excess mortality associated with diabetes by providing better and more precise management for patients. Our machine learning model offers a highly efficient and low-cost approach to identifying older adults with a very high risk of hospitalisation due to severe hypoglycemia, enabling corrective action for this group of patients without compromising the glycemic control in low-risk elderly patients."

The studies were supported by the CUHK Impact Research Fellowship Scheme and other funding. The original studies can be accessed at:

1. https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1004369
2. https://www.sciencedirect.com/science/article/pii/S0168822724001025?via%3Dihub

CU Medicine's research team developed a novel machine learning model that can predict the risk of severe hypoglycemia in the next 12 months among older adults with diabetes, using the big dataset from the Hospital Authority Data Collaboration Laboratory. With a high positive predictive value of 85%, the model has the potential to be integrated into electronic health record decision support systems for pre-emptive intervention in older adults at the highest risk.
Featured in the photo are the research team members. (From left) Professor Juliana Chan, Chair Professor of Medicine and Therapeutics at CU Medicine; Dr Jones Chan, Consultant, Department of Medicine and Therapeutics at Prince of Wales Hospital; Dr Elaine Chow, Associate Professor; Dr Yang Aimin, Research Assistant Professor; and Dr Shi Mai, Research Associate, from the Department of Medicine and Therapeutics at CU Medicine.

< Previous