Digital health technologies for diabetes-centered five-condition co-management in China: Theoretical foundations, practical experience, and technical challenges

Table 1 Evidence map of digital interventions in the management of diabetes complicated with multiple metabolic abnormalities in China and internationally

Research region	Study design	Sample size	Population characteristics	Intervention form	Follow-up duration	Primary outcome	Risk of bias/Limitations	Ref.
China	Quasi-experiment (pre-post control design)	189	T2DM patients complicated with hypertension	AI-powered remote monitoring system + personalized medical guidance	9 months	HbA1c decreased by 0.6% on average, blood pressure control rate increased by 36%, fasting plasma glucose reduced by 43.8 mg/dL	(1) No randomization (potential selection bias); (2) Loss to follow-up (15% attrition rate); and (3) Unblinded design (performance bias)	[118]
China	Quasi-experiment (single-arm intervention with historical control)	156	Primary T2DM patients (with DR risk)	Multimodal AI system (fundus images + clinical data) + referral management	6 months	DR screening accuracy reached the level of professional ophthalmologists, referral compliance was significantly improved, HbA1c decreased by 0.8% on average	(1) Historical control may introduce confounding bias; (2) Limited to primary care settings; and (3) No assessment of DR progression (only screening accuracy)	[119]
International	Parallel-group randomized clinical trial	246	T2DM combined with metabolic syndrome patients	APP-based self-management (blood glucose/diet/exercise recording) + healthy behavior rewards	12 months	HbA1c decreased by 0.4% (mean difference vs usual care), body weight decreased by 3.0 kg, LDL-C decreased by 0.38 mmol/L, intervention compliance increased by 32%	(1) Selection bias (strict inclusion/exclusion criteria); (2) Reward mechanism may introduce performance bias; and (3) Lack of subgroup analysis by age/gender	[120]
International	Quasi-experiment (non-randomized controlled trial)	112	T2DM patients complicated with proatherogenic dyslipidemia	Multimodal remote monitoring (blood glucose + lipid + postprandial glucose) + telehealth consultation	10 months	HbA1c decreased by 0.7% on average, LDL-C reduced by 0.45 mmol/L, postprandial glucose variability decreased by 52.6 mg/dL	(1) Small sample size (limited statistical power); (2) Single-center design (geographic bias); and (3) No blinding of outcome assessors	[121]
International	Real-world observational study	418	T2DM patients with multiple metabolic abnormalities	AI-integrated management platform (medication + diet + exercise + metabolic prediction)	15 months	HbA1c decreased by 1.3% on average, body weight reduced by 5.1 kg, metabolic index compliance rate increased by 48%, insulin sensitivity improved	(1) Selection bias (voluntary participation); (2) Lack of control group (cannot rule out temporal trends); and (3) Technical threshold for platform use (excludes elderly/illiterate patients)	[122]

T2DM: Type 2 diabetes mellitus; AI: Artificial intelligence; HbA1c: Glycated hemoglobin; DR: Diabetic retinopathy; LDL-C: Low-density lipoprotein cholesterol.