Background: Machine learning (ML) algorithms have been widely introduced to diabetes research including those for the identification of hypoglycemia. Objective: The objective of this meta-analysis is to assess the current ability of ML algorithms to detect hypoglycemia (ie, alert to hypoglycemia coinciding with its symptoms) or predict hypoglycemia (ie, alert to hypoglycemia before its symptoms have occurred). Methods: Electronic literature searches (from January 1, 1950, to September 14, 2020) were conducted using the Dialog platform that covers 96 databases of peer-reviewed literature. Included studies had to train the ML algorithm in order to build a model to detect or predict hypoglycemia and test its performance. The set of 2 × 2 data (ie, number of true positives, false positives, true negatives, and false negatives) was pooled with a hierarchical summary receiver operating characteristic model. Results: A total of 33 studies (14 studies for detecting hypoglycemia and 19 studies for predicting hypoglycemia) were eligible. For detection of hypoglycemia, pooled estimates (95% CI) of sensitivity, specificity, positive likelihood ratio (PLR), and negative likelihood ratio (NLR) were 0.79 (0.75-0.83), 0.80 (0.64-0.91), 8.05 (4.79-13.51), and 0.18 (0.12-0.27), respectively. For prediction of hypoglycemia, pooled estimates (95% CI) were 0.80 (0.72-0.86) for sensitivity, 0.92 (0.87-0.96) for specificity, 10.42 (5.82-18.65) for PLR, and 0.22 (0.15-0.31) for NLR. Conclusions: Current ML algorithms have insufficient ability to detect ongoing hypoglycemia and considerate ability to predict impeding hypoglycemia in patients with diabetes mellitus using hypoglycemic drugs with regard to diagnostic tests in accordance with the Users’ Guide to Medical Literature (PLR should be ≥5 and NLR should be ≤0.2 for moderate reliability). However, it should be emphasized that the clinical applicability of these ML algorithms should be evaluated according to patients’ risk profiles such as for hypoglycemia and its associated complications (eg, arrhythmia, neuroglycopenia) as well as the average ability of the ML algorithms. Continued research is required to develop more accurate ML algorithms than those that currently exist and to enhance the feasibility of applying ML in clinical settings. Trial Registration: PROSPERO International Prospective Register of Systematic Reviews CRD42020163682; http://www.crd.york.ac.uk/PROSPERO/display_record.php?ID=CRD42020163682

OBJECTIVE: To determine associations of systolic blood pressure (SBP) and diastolic blood pressure (DBP) with new-onset coronary artery disease (CAD) or cerebrovascular disease (CVD) according to glucose status. RESEARCH DESIGN AND METHODS: Examined was a nationwide claims database from 2008 to 2016 on 593,196 individuals. A Cox proportional hazards model identified risks of CAD and CVD events among five levels of SBP and DBP. RESULTS: During the study period 2,240 CAD and 3,207 CVD events occurred. Compared with SBP ≤119 mmHg, which was the lowest quintile of SBP, hazard ratios (95% CI) for CAD/CVD in the 4 higher quintiles (120–129, 130–139, 140–149, ≥150 mmHg) gradually increased from 2.10 (1.73–2.56)/1.46 (1.27–1.68) in quintile 2 to 3.21 (2.37–4.34)/4.76 (3.94–5.75) in quintile 5 for normoglycemia, from 1.39 (1.14–1.69)/1.70 (1.44–2.01) in quintile 2 to 2.52 (1.95–3.26)/4.12 (3.38–5.02) in quintile 5 for borderline glycemia, and from 1.50 (1.19–1.90)/1.72 (1.31–2.26) in quintile 2 to 2.52 (1.95–3.26)/3.54 (2.66–4.70) in quintile 5 for diabetes. A similar trend was observed for DBP across 4 quintiles (75–79, 80–84, 85–89, and ≥90 mmHg) compared with ≥74 mmHg, which was the lowest quintile. CONCLUSIONS: Results indicated that cardiovascular risks gradually increased with increases in SBP and DBP regardless of the presence of and degree of a glucose abnormality. Further interventional trials are required to apply findings from this cohort study to clinical practice.

Aims Although conventional interventions for people at high risk of developing type 2 diabetes are usually conducted face-to-face, such interventions are burdensome for health care providers. We developed a lifestyle intervention program combining lifestyle coaching via a smartphone application augmented by intermittently scanned continuous glucose monitoring without burdening health care providers. Its effectiveness for glycemic control and body weight reduction in people at risk of type 2 diabetes was investigated. Materials and Methods For this 12-week randomized unblinded trial with offline recruitment, participants with a hemoglobin A1c level of 5.6% to 6.4% or a fasting blood glucose of 110 to 125 mg/dL and body mass index (BMI) >23 kg/m² but <40 kg/m² were randomly assigned to the intervention group (App) and control group (C). The primary endpoint was the difference in time in range of blood glucose between 70 and 140 mg/dL (3.9-7.8 mmol/L) before and after the study period between the 2 groups. Results Among 168 patients (mean age, 48.1 years; mean BMI, 26.6 kg/m²; and male, 80.4%), 82 and 86 were assigned to the App group and C group, respectively. After 12 weeks, time in range of blood glucose at 70 to 140 mg/dL significantly improved in the App group compared with the C group (−2.6 minutes/day vs +31.5 minutes/day, P = .03). Changes in time above range did not differ, whereas time below range (blood glucose <70 mg/dL; +23.5 minutes/day vs −8.9 minutes/day, P = .02) improved in the App group. BMI (−0.26 vs −0.59, P = .017) was reduced in the App group compared with the C group. Conclusion Intervention with a smartphone app and intermittently scanned continuous glucose monitoring increased glycemic control accompanied by decreased carbohydrate intake and weight loss. Further trials are needed to confirm whether these interventions can reduce incident type 2 diabetes.

Aims/Introduction: Recently, an increasing number of cohort studies have suggested using machine learning (ML) to predict type 2 diabetes mellitus. However, its predictive ability remains inconclusive. This meta‐analysis evaluated the current ability of ML algorithms for predicting incident type 2 diabetes mellitus.Materials and Methods: We systematically searched longitudinal studies published from 1 January 1950 to 17 May 2020 using MEDLINE and EMBASE. Included studies had to compare ML’s classification with the actual incidence of type 2 diabetes mellitus, and present data on the number of true positives, false positives, true negatives and false negatives. The dataset for these four values was pooled with a hierarchical summary receiver operating characteristic and a bivariate random effects model.Results: There were 12 eligible studies. The pooled sensitivity, specificity, positive likelihood ratio and negative likelihood ratio were 0.81 (95% confidence interval [CI] 0.67–0.90), 0.82 [95% CI 0.74–0.88], 4.55 [95% CI 3.07–6.75] and 0.23 [95% CI 0.13–0.42], respectively. The area under the summarized receiver operating characteristic curve was 0.88 (95% CI 0.85–0.91).Conclusions: Current ML algorithms have sufficient ability to help clinicians determine whether individuals will develop type 2 diabetes mellitus in the future. However, persons should be cautious before changing their attitude toward future diabetes risk after learning the result of the diabetes prediction test using ML algorithms.

Background: Evidence of the role of systolic blood pressure ( SBP ) in development of severe diabetic retinopathy is not strong, although the adverse effect of low diastolic blood pressure has been a partial explanation. We assessed the predictive ability of incident severe diabetic retinopathy between pulse pressure ( PP ) which considers both SBP and diastolic blood pressure, compared with SBP . Methods and Results: Eligible patients (12 242, 83% men) aged 19 to 72 years from a nationwide claims database were analyzed for a median observational 4.8‐year period. Severe diabetic retinopathy was defined as vision‐threatening treatment‐required diabetic eye diseases. Multivariate Cox regression analysis revealed that hazard ratios (95% CI ) of treatment‐required diabetic eye diseases for 1 increment of standard deviation and the top tertile compared with the bottom tertile were 1.39 (1.21–1.60) and 1.72 (1.17–2.51), respectively, for PP and 1.22 (1.05–1.41) and 1.43 (0.97–2.11), respectively, for SBP adjusted for age, sex, body mass index, hemoglobin A1c, fasting plasma glucose, lipids, and smoking status. In a model with SBP and PP simultaneously as covariates, the hazard ratios of only PP (hazard ratios [95% CI ], 1.57 [1.26–1.96]) but not SBP (0.85 [0.68–1.07]) were statistically significant. Delong test revealed a significant difference in the area under the receiver operating characteristic curve between PP and SBP (area under the receiver operating characteristic curve [95% CI ], 0.58 [0.54–0.63] versus 0.54 [0.50–0.59]; P =0.03). The strongest predictor remained as hemoglobin A1c (area under the receiver operating characteristic curve [95% CI ], 0.80 [0.77–0.84]). Conclusions: After excluding the significant impact of glycemic control, PP in comparison with SBP is a better predictor of severe diabetic retinopathy, suggesting a role of diastolic blood pressure and arterial stiffness in pathology.

Background: To determine the impact of metabolic syndrome (MetS) and/or metabolic dysfunction-associated fatty liver disease (MAFLD), which are pathophysiologically similar and include insulin resistance, on the development of new-onset cardiovascular disease with and without type 2 diabetes and according to sex. Methods: This study included 570,426 individuals without a history of cardiovascular disease who were enrolled in a nationwide claims database from 2008 to 2016 and were classified by the presence or absence of MetS and/or MAFLD stratified by the presence or absence of type 2 diabetes and sex. The fatty liver index was used to determine the presence or absence of fatty liver that required a diagnosis of MAFLD. Risks of developing coronary artery disease (CAD) and cerebrovascular disease (CVD) in each category were analyzed using a multivariate Cox proportional hazard model. Results: During a median follow-up of 5.2 years, 2252 CAD and 3128 CVD events occurred. Without type 2 diabetes the hazard ratio (HR) (95% CI) for CAD/CVD compared with neither MAFLD nor MetS was 1.32 (1.17–1.50)/1.41(1.28–1.57) for MAFLD only (without MetS), 1.78 (1.22–2.58)/1.66 (1.34–2.06) for MetS only (without MAFLD), and 2.10 (1.84–2.39)/1.73 (1.54–1.95) for MAFLD + MetS. For those with type 2 diabetes, the HR for CAD for MAFLD only (compared with neither MAFLD nor MetS) was 1.29 (1.06–1.58), for MetS only 1.34 (0.84–2.13), and for MAFLD + MetS 1.22 (1.02–1.47). For CVD, there was a significant increase in HR only in MAFLD + MetS [1.44 (1.18–1.76)]. The results of the analysis stratified by sex showed that MAFLD had a greater impact in men, and MetS had a greater impact in women regarding the development of CAD. Conclusions: Distinguishing between MetS and/or MAFLD in the presence or absence of type 2 diabetes and according to sex may aid in accurately identifying patients at high risk of cardiovascular disease.