Analysis Of Glycemic Variability And Risk Prediction Of Nocturnal Hypoglycemia In Patients With Diabetes Based On CGM

Background: Glucose monitoring is an important part of diabetes management.Self-blood glucose monitoring(SMBG)is the basic form of glucose monitoring,and Hb A1 c is the gold standard to reflect the long-term glucose control.However,both SMBG and Hb A1 c have some limitations.Continuous glucose monitoring(CGM)system is a technique that indirectly reflects the level of blood glucose by monitoring the concentration of glucose in subcutaneous interstitial fluid through glucose sensor,which can sensitively detect occult high and low blood glucose,more importantly,reflect glycemic variability(GV).GV was reported to be closely related to diabetic microangiopathy and macroangiopathy.Therefore,accurate and comprehensive evaluation of GV is particularly important for short-term glucose control and long-term risk assessment of chronic complications in patients with diabetes.Previous studies had shown that GV of patients with type 1 diabetes mellitus(T1DM)was significantly higher than that of patients with type 2 diabetes mellitus(T2DM).However,GV characteristics of latent autoimmune diabetes in adults(LADA)and its differences with T1 DM and T2 DM were rarely reported as a special phenotype of diabetes.In addition,exploring the characteristics of GV in patients with different types of diabetes is expected to provide certain directions and clues for the prevention and treatment of chronic complications of diabetes.Objective: To investigate the pattern of GV and its influencing factors in patients with T1 DM,LADA and T2 DM by CGM.Methods: A total of 842 patients with diabetes(510 with T1 DM,105with LADA,227 with T2DM)who were in the outpatient or ward of the Second Xiangya Hospital of Central South University,the First Affiliated Hospital of Henan University of Science and Technology,and Heji Hospital Affiliated to Shanxi Changzhi Medical College from September2019 to January 2022 were selected and underwent 5-7 days of retrospective CGM to collect glucose data.The general data,laboratory test results and CGM data of the three groups were compared,and the differences between LADA-1 and LADA-2 subgroups were further compared according to GADA titer.Spearman correlation analysis and multiple linear regression were adopted to analyze GV related clinical indicators.Results: 1.There were no significant differences in sex,duration of diabetes and Hb A1 c among the three groups.Fasting C-peptide(FCP)and 2-hour postprandial C-peptide(2h CP)levels increased in turn in patients with T1 DM,LADA,and T2DM(all P<0.001).2.CGM results showed that: 1)The time above range(TAR)and time in range(TIR)levels were similar in T1 DM and LADA,and TAR of both groups was significantly higher than that of T2 DM group,while TIR was significantly lower(all P<0.001).2)hypoglycemia: hypoglycemia was the most frequent in the T1 DM group,with the highest time below range(TBR)(2.2%),followed by the LADA group(0.6%),and close to 0 in the T2 DM group.3)GV: the GV of T1 DM group was the highest among the three groups,and there was no significant difference in GV between LADA and T1 DM groups,but both were higher than that of T2 DM group(P >0.001).The GV of LADA-1 patients was close to that of T1 DM patients,while that of LADA-2 patients was close to that of T2 DM patients.3.Spearman correlation analysis showed that GV showed the strongest negative correlation with FCP and 2h CP.Taking the coefficient of variation(CV)of glucose as dependent variable,age,BMI,duration of diabetes,insulin dose,Hb A1 c,FCP and 2h CP as independent variables,multiple linear regression was used to find that age,insulin dose,FCP and2 h CP were independent predictors of CV.Conclusions: It was revealed for the first time that GV presented a continuously changing disease spectrum from T1 DM,LADA-1,LADA-2to T2 DM,and islet function marker FCP and 2h CP were the most important negative predictors of GV.There are 2 figures,4 tables.Background: Certain amounts of clinical factors were associated with glycemic variability(GV).The first part of our study had confirmed that fasting C-peptide(FCP)and 2-hour postprandial C-peptide(2h CP)were the strongest negative related factors of GV.As an autoimmune disease,T1 DM has severe insulin deficiency caused by immune destruction of islet beta-cell,also known as "fragile diabetes".The GV of T1 DM was the most violent and typical.However,although the duration of diabetes,Hb A1 c,islet function,insulin dose and other factors were comparable in some T1 DM patients,their GV still had a certain heterogeneity,whether there were other factors involved in GV was not known.Objective: This study aimed to compare the metabolomic profiles between higher GV(GV-H,CV≧36%)and lower GV(GV-L,CV<36%)groups,and to identify potential novel biomarkers of GV.Methods: Fifty-four patients with T1 DM wearing CGM were included and divided into GV-H(CV ≧ 36%)and GV-L groups(CV < 36%)according to CGM International Consensus recommendations to control the coefficient of variation(CV)of glucose within 36%.Of these,33(GV-H,n = 17;GV-L,n = 16)and 21(GV-H,n = 11;GV-L,n = 10)patients served as the exploratory and validation sets,respectively,and all underwent serum liquid chromatography-mass spectrometry(LC-MS)metabolomics profiling.In addition to matching for age,sex,and BMI,patients in the GV-H and GV-L groups were further matched for diabetes duration,daily insulin dosage,Hb A1 c,fasting and 2-hour postprandial C-peptide(FCP,2h CP)levels.The performance of the final selected GV biomarkers in predicting GV levels was tested with receiver operating characteristic curve(ROC)analysis and multiple linear regression.Results: In the exploration set,7 differential metabolic pathways were screened between GV-H and GV-L groups,and 25 differential metabolites were enriched.The roles of spermidine,L-methionine and trehalose in predicting GV level were confirmed in the validation set.ROC analysis showed that the area under curve(AUC)of above 3 selected markers combined was 0.989(Se/Sp: 94.0%/100.0%)in the exploratory set and was 0.918(Se/Sp: 90.9%/80.0%)in the validation set.Trehalose was most correlated with CV(r =-0.654,P<0.001),which was higher than spermidine and L-methionine(r = 0.517 and-0.459,respectively,P<0.001).Multiple linear regression analysis was performed using the stepwise method with CV as the dependent variable and age,BMI,diabetes duration,insulin dose,Hb A1 c,fasting and 2-hour postprandial blood glucose(FBG,2h BG),FCP,2h CP,spermidine,L-methionine and trehalose as independent variables.As a result,only trehalose(standardized β =-0.499,P=0.001)and L-methionine(standardized β =-0.360,P=0.013)were included in the final model(R2 =0.458,F=14.545,P<0.001),indicating that trehalose and L-methionine could be independent predictors of CV.Conclusions: GV-L and GL-H groups exhibited distinct metabolic profiles.Increased spermidine,decreased L-methionine and trehalose were associated with greater GV in patients with T1 DM.Moreover,L-methionine and trehalose were independent predictors of CV after adjustment for age,disease duration,daily insulin dosage,Hb A1 c and CP levels.There are 7 figures,6 tables.Background: Despite the rapid development of modern blood glucose monitoring technology,nocturnal hypoglycemia(NH)is still one of the main obstacles for patients with diabetes to achieve optimal glucose control.NH can lead to counter-regulatory response and cognitive impairment,even cause fatal cardiovascular events.Accurate risk prediction of NH may help to minimize the frequency and harm of NH in patients with diabetes.Although the short-term hypoglycemia prediction algorithm based on continuous glucose monitoring(CGM)data has been used in single-hormone and dual-hormone automatic insulin delivery systems of artificial pancreas,and has been proved to effectively reduce the frequency and duration of NH and improve glucose control,most patients can not afford such expensive equipment.In recent years,machine learning(ML)has been increasingly adopted in medical big data.As a result,ML has made great achievements in the field of NH event risk prediction,but most of them were short-term prediction with distinct prediction accuracy,and main input included in the prediction model was only CGM data and did not involve the clinical indicators closely related to hypoglycemia.Objective: To establish a risk prediction model for NH by combining CGM data and various clinical indicators in patients with diabetes.Methods: A total of 2065 patients with diabetes(554 with T1 DM,208with LADA,and 1303 with T2DM)who visited the Department of Endocrinology of Second Xiangya Hospital of Central South University and Shanghai Jiao Tong University Affiliated Sixth People’s Hospital from September 2019 to January 2022 were enrolled in this study.Glucose data were collected by retrospective CGM system for 5 to 7 days,and general data,biochemical indicators and glucose-lowering medications were also collected.NH events were extracted from CGM data,defined as at least 15 min with a CGM value below 3.9 mmol/L and occurring between 23:00 at night and 06:00 the following day.Random forest(RF)models were invoked to predict the risk of NH occurrence in patients with diabetes by adjusting the model inputs(different combinations such as daytime CGM data,CGM parameters,and clinical indicators).Seventy percent of the data was used for model training and the remaining 30% was used for model validation.Model performance was assessed by sensitivity(Se),specificity(Sp),and area under the curve(AUC).Results: When CGM data alone was used to predict NH events,the diagnostic efficiency significantly improved compared with clinical data(AUC: 0.869,Se: 82.96%,Sp: 76.16% for the former;AUC: 0.786,Se:75.78%,Sp: 72.56% for the latter).While the resulting AUC of RF model was 0.884 when CGM data and indicators were used in combination for prediction,Se and Sp were 85.20% and 77.23%,respectively.In addition,clinical indicators combined with LBGI predicted NH events better than combined with other CGM parameters(AUC = 0.838,Se: 81.17%,Sp:75.30%),followed by combined CV and TBR,resulting in AUCs of0.814 and 0.813,respectively.The RF model performed best when CGM data,CGM parameters and clinical indicators were used as inputs,with an AUC of 0.902(Se: 85.20%,Sp: 78.54%).Conclusions: We developed a RF model combined CGM data and clinical indicators that could be used to predict NH risk with high accuracy,which has an AUC of 0.902 and a sensitivity and specificity of85.20% and 78.54%,respectively.There are 3 figures,2 tables.