Background

Digital health interventions based on tools for Computerized Decision Support (CDS) and Machine Learning (ML), which take advantage of new information, sensing and communication technologies, can play a key role in childhood obesity prevention and treatment.

Objectives

We present a systematic literature review of CDS and ML applications for the prevention and treatment of childhood obesity. The main characteristics and outcomes of studies using CDS and ML are demonstrated, to advance our understanding towards the development of smart and effective interventions for childhood obesity care.

Methods

A search in the bibliographic databases of PubMed and Scopus was performed to identify childhood obesity studies incorporating either CDS interventions, or advanced data analytics through ML algorithms. Ongoing, case, and qualitative studies, along with those not providing specific quantitative outcomes were excluded. The studies incorporating CDS were synthesized according to the intervention’s main technology (e.g., mobile app), design type (e.g., randomized controlled trial), number of enrolled participants, target age of children, participants’ follow-up duration, primary outcome (e.g., Body Mass Index (BMI)), and main CDS feature(s) and their outcomes (e.g., alerts for caregivers when BMI is high). The studies incorporating ML were synthesized according to the number of subjects included and their age, the ML algorithm(s) used (e.g., logistic regression), as well as their main outcome (e.g., prediction of obesity).

Results

The literature search identified 8 studies incorporating CDS interventions and 9 studies utilizing ML algorithms, which met our eligibility criteria. All studies reported statistically significant interventional or ML model outcomes (e.g., in terms of accuracy). More than half of the interventional studies (n = 5, 63 %) were designed as randomized controlled trials. Half of the interventional studies (n = 4, 50 %) utilized Electronic Health Records (EHRs) and alerts for BMI as means of CDS. From the 9 studies using ML, the highest percentage targeted at the prognosis of obesity (n = 4, 44 %). In the studies incorporating more than one ML algorithms and reporting accuracy, it was shown that decision trees and artificial neural networks can accurately predict childhood obesity.

Conclusions

This review has found that CDS tools can be useful for the self-management or remote medical management of childhood obesity, whereas ML algorithms such as decision trees and artificial neural networks can be helpful for prediction purposes. Further rigorous studies in the area of CDS and ML for childhood obesity care are needed, considering the low number of studies identified in this review, their methodological limitations, and the scarcity of interventional studies incorporating ML algorithms in CDS tools.

中文翻译：

---

用于预防和治疗儿童肥胖症的计算机化决策支持和机器学习应用：文献系统回顾。

背景

基于计算机化决策支持 (CDS) 和机器学习 (ML) 工具的数字健康干预措施利用新的信息、传感和通信技术，可以在儿童肥胖症的预防和治疗中发挥关键作用。

目标

我们对 CDS 和 ML 用于预防和治疗儿童肥胖症的应用进行了系统的文献综述。展示了使用 CDS 和 ML 研究的主要特征和结果，以促进我们对开发针对儿童肥胖症护理的智能有效干预措施的理解。

方法

在 PubMed 和 Scopus 的书目数据库中进行了搜索，以确定结合 CDS 干预或通过 ML 算法进行高级数据分析的儿童肥胖研究。正在进行的案例和定性研究，以及那些没有提供具体定量结果的研究被排除在外。纳入 CDS 的研究是根据干预的主要技术（例如，移动应用程序）、设计类型（例如，随机对照试验）、纳入的参与者数量、目标儿童年龄、参与者的随访时间、主要结果（例如， 、身体质量指数 (BMI)），以及主要的 CDS 特征及其结果（例如，当 BMI 高时，护理人员的警报）。纳入 ML 的研究是根据纳入的受试者数量和他们的年龄、使用的 ML 算法（例如，

结果

文献检索确定了 8 项纳入 CDS 干预的研究和 9 项利用 ML 算法的研究，它们符合我们的资格标准。所有研究都报告了具有统计学意义的干预或 ML 模型结果（例如，在准确性方面）。超过一半的干预研究（n = 5, 63 %）被设计为随机对照试验。一半的介入研究（n = 4, 50 %）使用电子健康记录 (EHR) 和 BMI 警报作为 CDS 的手段。在使用 ML 的 9 项研究中，针对肥胖预后的最高百分比 (n = 4, 44 %)。在包含不止一种机器学习算法和报告准确性的研究中，决策树和人工神经网络可以准确预测儿童肥胖。

结论

本综述发现 CDS 工具可用于儿童肥胖的自我管理或远程医疗管理，而决策树和人工神经网络等机器学习算法可用于预测目的。考虑到本综述中确定的研究数量少、方法学局限性以及在 CDS 工具中纳入 ML 算法的干预研究的稀缺性，需要在 CDS 和 ML 用于儿童肥胖护理领域进行进一步严格的研究。