Elmotia Khadija, Abouyaala Oumaima, Bougrine Soukaina, Ouahidi Moulay Laarbi
Laboratory of Biology and Health, Department of Biology, Faculty of Science, Ibn Tofail University, BP 133, Kenitra, Morocco.
Prim Care Diabetes. 2025 Aug;19(4):345-354. doi: 10.1016/j.pcd.2025.05.004. Epub 2025 May 12.
This systematic review aims to assess the effectiveness of AI-Driven Decision Support Systems in improving glycemic control, measured by Time in Range (TIR) and HbA1c levels, in patients with diabetes. Included studies were randomized controlled trials (RCTs) that evaluated AI interventions in diabetes management. Exclusion criteria included non-English studies, non-peer-reviewed articles. Studies were identified by searching electronic databases including PubMed, EMBASE, and Cochrane Library up to December 2024. Risk of bias was assessed using the Cochrane Risk of Bias tool for RCTs. Results were synthesized using a random-effects meta-analysis model. The review included 17 RCTs with a total of 3381 participants in the intervention group and 3176 in the control group. AI interventions were found to significantly improve TIR and reduce HbA1c levels. The meta-analysis for TIR yielded a mean difference of 0.54 (95 % CI: 0.05-1.03), and for HbA1c a standardized mean difference of -0.91 (95 % CI: -1.23 to -0.58). Evidence was limited by high heterogeneity (I² > 90 % for both outcomes) and indications of publication bias, which may overestimate the effectiveness reported. Despite limitations, the results support the potential of AI interventions in enhancing diabetes management, though variability in effectiveness suggests the need for personalized approaches.
本系统评价旨在评估人工智能驱动的决策支持系统在改善糖尿病患者血糖控制方面的有效性，血糖控制通过血糖达标时间（TIR）和糖化血红蛋白（HbA1c）水平来衡量。纳入的研究为随机对照试验（RCT），这些试验评估了糖尿病管理中的人工智能干预措施。排除标准包括非英文研究、非同行评审文章。通过检索电子数据库（包括截至2024年12月的PubMed、EMBASE和Cochrane图书馆）来识别研究。使用Cochrane随机对照试验偏倚风险工具评估偏倚风险。结果采用随机效应荟萃分析模型进行综合。该评价纳入了17项随机对照试验，干预组共有3381名参与者，对照组有3176名参与者。发现人工智能干预措施能显著改善血糖达标时间并降低糖化血红蛋白水平。血糖达标时间的荟萃分析得出平均差值为0.54（95%CI：0.05 - 1.03），糖化血红蛋白的标准化平均差值为 -0.91（95%CI：-1.23至 -0.58）。由于高异质性（两个结果的I²均>90%）和发表偏倚迹象，证据受到限制，这可能高估了所报告的有效性。尽管存在局限性，但结果支持人工智能干预在加强糖尿病管理方面的潜力，不过有效性的差异表明需要个性化方法。
