Table 4.

Translational matrix of artificial intelligence (AI) methodologies in semaglutide-related type 2 diabetes mellitus research. Data statistics reflect the size of the underlying dataset (eg, claims records, EHR^a entries, and imaging scans) as reported in the original study, not necessarily the number of unique semaglutide-treated patients.

Author (year)	AI category	Algorithm	Input data type	Data statistics	Application domain	Deployment level
Beavers et al (2025) [28]	Meta-analysis	Bayesian modeling	DXAb RCTc data	5 trials (n=2348)	Musculoskeletal	Concept
Vitorino (2025) [14]	Omics analysis	Proteomic profiling	Biomarker data	112 proteins +68 metabolites	CVc modeling	Theoretical
Nelson et al (2024) [13]	Deep learning	3D U-Net	Abdominal CTe	1842 scans (614 patients)	Body composition	Pilot
Zhu et al (2024) [20]	Supervised MLf	LightGBMg	EHRh records	34,589 patients	Treatment response	Scalable
Crea (2024) [22]	Literature review	N/Ai	127 trials	3187-3191 citations	CV mechanisms	Concept
Cheng et al (2024) [31]	Policy analysis	N/A	52 health systems	Equity metrics	Health policy	Strategy
Warren et al (2024) [32]	Reinforcement Learning	d-Nav	Insulin data	12,345 titrations	Diabetes management	Deployed
Barkas et al (2024) [33]	Ensemble ML	Random Forest	Multimodal data	21,098 patients	CV risk	Clinical
Nguyen et al (2024) [34]	Cluster analysis	CDj markers	Immune data	459 samples	Phenotyping	Experimental
Mozaffarian (2024) [9]	Cost analysis	N/A	38 studies	Cost-benefit ratios	Therapy balance	N/A
Quddos et al (2023) [27]	Behavioral ML	K-means	Social media	4562 posts + 893 surveys	Addiction	Research
Ansari et al (2023) [35]	Systematic review	Multiple models	214 studies	Therapeutic trends	Innovation	Summary
Foer et al (2023) [25]	NLPk	ML pipeline	EHR text	89,432 records	COPDl	Validated
Giorda et al (2023) [24]	Logic ML	LLMm	Clinical DB	7812 patients	Treatment simulation	Simulation
Shao et al (2022) [36]	Risk modeling	BRAVOn	ACCORDo trial	10,251 subjects	Complications	Validated
Armandi and Schattenberg (2022) [37]	Pathogenesis	N/A	9 diagnostic tools	NAFLDp/NASHq	Liver disease	Background
Yamada et al (2020) [5]	DNNr	Deep learning	Claims data	145,678 claims	MIs risk	Experimental
Kovatchev (2019) [3]	CGMt algorithms	LBGIu/HBGIv	CGM data	2.1M measurements	Glycemic control	Implemented

^aEHR: electronic health record.

^bDXA: dual-energy X-ray absorptiometry.

^cRCT: randomized controlled trial.

^dCV: cardiovascular.

^eCT: computed tomography.

^fML: machine learning.

^gLightGBM: light gradient boosting machine.

^hEHR: electronic health record.

ⁱN/A: not applicable.

^jCD: cluster of differentiation.

^kNLP: natural language processing.

^lCOPD: chronic obstructive pulmonary disease.

^mLLM: logic learning machine.

ⁿBRAVO: Building, Relating, Assessing, and Validating Outcomes.

^oACCORD: Action to Control Cardiovascular Risk in Diabetes.

^pNAFLD: nonalcoholic fatty liver disease.

^qNASH: nonalcoholic steatohepatitis.

^rDNN: deep neural network.

^sMI: myocardial infarction.

^tCGM: continuous glucose monitoring.

^uLBGI: Low Blood Glucose Index.

^vHBGI: High Blood Glucose Index.