Chinese diabetes datasets for data-driven machine learning

Abstract

Data of the diabetes mellitus patients is essential in the study of diabetes management, especially when employing the data-driven machine learning methods into the management. To promote and facilitate the research in diabetes management, we have developed the ShanghaiT1DM and ShanghaiT2DM Datasets and made them publicly available for research purposes. This paper describes the datasets, which was acquired on Type 1 (n = 12) and Type 2 (n = 100) diabetic patients in Shanghai, China. The acquisition has been made in real-life conditions. The datasets contain the clinical characteristics, laboratory measurements and medications of the patients. Moreover, the continuous glucose monitoring readings with 3 to 14 days as a period together with the daily dietary information are also provided. The datasets can contribute to the development of data-driven algorithms/models and diabetes monitoring/managing technologies.

Application of Machine Learning Algorithms to Predict Uncontrolled Diabetes Using the All of Us Research Program Data

“…In order to overcome these limitations, prediction of glycemic control using continuous glucose monitoring data might also help apart from the AoU dataset. More recent datasets such as the ShanghaiT1DM and Shanghai T2DM are design for data-driven machine learning to predict glycemic control in DM patients [ 62 ]. In addition, other studies have also reported the role of reinforcement learning in predicting blood glucose control; these complement the current study [ 63 ].…”

A generative pretrained transformer model for decoding individual glucose dynamics from continuous glucose monitoring data

“…T1D and T2D have different glucose dynamics. We next conduct diagnosing of T1D/T2D based on CGM data with CGMformer finetuned on this limited labeled data from Zhao et al 36 with 125 subjects including 109 T2Ds and 16 T1Ds. CGMformer consistently outperforms baseline methods including state-of-arts machine learning based diagnosis including LSTM 34 and MLP, as well as combining metrics-based predictors in T1D/T2D diagnosis (Fig.…”

“…Three inputs include individual embedding vector encoded from CGMformer 𝑣 𝑆 ∈ ℝ 𝑑 ; before-meal 1h glucose 𝐺 𝐵 ∈ ℝ 𝑡 , where 𝑡 is the number of CGM measurements in one hour, and 𝑡 = 4 for FGM used in Zhao et al36 which measure glucose every 15 minutes; and the information of the dietary intake, 𝐷 = (𝐻, 𝐶, 𝑃, 𝐹, 𝐵) ∈ ℝ 5 , containing the calories ( 𝐻inunitof𝑘𝑐𝑎𝑙 ), carbohydrates (𝐶inunitof𝑔 ), proteins (𝑃inunitof𝑔 ), fats (𝐹inunitof𝑔 ), and dietary fiber (𝐵inunitof𝑔 ) .CGMformer_Diet predicts post-meal 2h glucose 𝐺 𝑃 ∈ 𝑅 2𝑡 as output. The dietary information is encoded as a pulsed perturbation, 𝐷 ̂∈ ℝ 5 * 𝑡 , with 𝐷 ̂⋅.𝑡 = 𝐷 indicating a dietary intake at time 𝑡, and 𝐷 ̂⋅,𝑗 = 0, for 𝑗 ≠ 𝑡, indicating no dietary intake at other times.…”

“…36 , we established a standard balanced meal plan with fixed calories and an energy supply ratio of carbohydrate: protein: fat = 5:2:3. To assess the glucose response to different dietary intake plan, we designed three additional simulated meal plans with adjusted ratios-low carbohydrate-high protein (carbohydrate: protein: fat = 4:3:3), low carbohydrate-high fat (carbohydrate: protein: fat = 4:2:4), and high protein (carbohydrate: protein: fat = 5:3:2).…”

Publicly Available Data Set Including Continuous Glucose Monitoring Data

“…The presented Table 1 showcases the selected data sets. Notably, a recent publication called the ShanghaiT1DM and ShanghaiT2DM data sets 4 features a substantial sample size (n = 100) of patients with type 2 diabetes. This data set includes information on CGM, meals, medication, and blood test outcomes.…”