Table 2.

Artificial intelligence (AI)–based vs standard Ozempic treatment in type 2 diabetes.

Clinical outcome	Standard fixed-dosing	AI-driven personalized dosing/insight	Improvement (difference)	Study references
HbA1ca reduction (%)	1.2% reduction on average in nonindividualized dosing	AI-assisted titration system (Warren) reduced HbA1c from 8.6% to 7.3%. EHRb-based AI models (Zhu) identified predictors of better response. Risk engine models (Shao) guided more precise treatment goals.	Up to +1.3% greater reduction in blood sugar	Warren et al (2024) [32]; Zhu et al (2024) [20]; Shao et al (2022) [36]; Giorda et al (2023) [24]
Weight loss (kg)	5-7 kg reduction from baseline	Imaging AI (Nelson) showed semaglutide targets visceral fat. MLc models (Zhu) predicted improved responders. Simulation models (Giorda) showed improved outcomes with earlier intervention.	+2-3 kg additional weight loss, with more targeted fat reduction	Nelson et al (2024) [13]; Zhu et al (2024) [20]; Giorda et al (2023) [24]; Yamada et al (2020) [5]
Cardiovascular risk reduction	Moderate benefit in high-risk patients	BRAVOd model (Shao, 2022) [36] simulated long-term heart disease outcomes. AI tools (Barkas, 2024) [33] enhanced ASCVDe prevention. GLP-1 RAf use (Foer, 2023) [25] reduced respiratory events. Quddos showed behavioral links to benefit.	Better risk targeting and long-term heart protection	Barkas et al (2024) [33]; Shao et al (2022) [36]; Foer et al (2023) [25]; Quddos et al (2023) [27]; Yamada et al (2020) [5]
Patient stratification	Manual, based on fixed clinical parameters (eg, BMI, HbA1c, and age)	AI/ML models (Zhu, Foer, Giorda) used patient-specific data (eg, age, gender, and history) to predict who benefits the most from semaglutide therapy.	Improved personalization and responder identification	Zhu et al (2024) [20]; Giorda et al (2023) [24]; Foer et al (2023) [25]; Quddos et al (2023) [27]; Shao et al (2022) [36]
Dose optimization	Physician-guided titration	AI-powered insulin adjustment system (Warren’s d-Nav) improved glucose control. Giorda’s model simulated better results with earlier, optimized dosing.	Better glycemic control and medication adherence	Warren et al (2024) [32]; Giorda et al (2023) [24]; Shao et al (2022) [36]
Adverse event risk (eg, GIg)	Gastrointestinal (GI) symptoms (20%-25% incidence in standard care)	Zhu: ML predicted patients prone to side effects. Foer: GLP-1 RAs reduced respiratory complications in comorbid patients. Quddos: behavioral factors linked to lower side effect risk.	Estimated 20%-30% reduction in adverse events via prediction	Zhu et al (2024) [20]; Foer et al (2023) [25]; Quddos et al (2023) [27]
Cost efficiency	Often inefficient when applied late or to nonresponders	Shao’s BRAVO model showed better cost-benefit in high-risk patients. Giorda’s simulations were intended to improve persistence. Warren’s AI model optimized insulin dose, reducing drug use.	Greater cost-effectiveness over time	Shao et al (2022) [36]; Giorda et al (2023) [24]; Warren et al (2024) [32]

^aHbA_1c: hemoglobin A_1c.

^bEHR: electronic health record.

^cML: machine learning.

^dBRAVO: Building, Relating, Assessing, and Validating Outcomes.

^eASCVD: atherosclerotic cardiovascular disease.

^fGLP-1 RA: glucagon-like peptide-1 receptor agonist.

^gGI: gastrointestinal.