Wardenaar Floris C, Schott Kinta D, Seltzer Ryan G N, Gardner Christopher D
College of Health Solutions, Arizona State University, Phoenix, AZ, United States.
Department of Medicine, Stanford University, Palo Alto, CA, United States.
Front Nutr. 2024 May 15;11:1381731. doi: 10.3389/fnut.2024.1381731. eCollection 2024.
The aim of this cross-sectional study was to develop an algorithm to predict athletes use of third-party tested (TPT) supplements. Therefore, a nutritional supplement questionnaire was used with a section about self-reported TPT supplement use.
Outcomes were randomly assigned to a training dataset to identify predictors using logistic regression models, or a cross-validation dataset. Training data were used to develop an algorithm with a score from 0 to 100 predicting use or non-use of TPT nutritional supplements.
A total of = 410 NCAA Division I student-athletes (age: 21.4 ± 1.6 years, 53% female, from >20 sports) were included. Then = 320 were randomly selected, of which 34% ( = 109) of users consistently reported that all supplements they used were TPT. Analyses resulted in a 10-item algorithm associated with use or non-use of TPT. Risk quadrants provided the best fit for classifying low vs. high risk toward inconsistent TPT-use resulting in a cut-off ≥60% (χ(4) = 61.26, < 0.001), with reasonable AUC 0.78. There was a significant association for TPT use (yes/no) and risk behavior (low vs. high) defined from the algorithm (χ(1)=58.6, < 0.001). The algorithm had a high sensitivity, classifying 89% of non-TPT users correctly, while having a low specificity, classifying 49% of TPT-users correctly. This was confirmed by cross-validation ( = 34), reporting a high sensitivity (83%), despite a lower AUC (0.61).
The algorithm classifies high-risk inconsistent TPT-users with reasonable accuracy, but lacks the specificity to classify consistent users at low risk. This approach should be useful in identifying athletes that would benefit from additional counseling.
本横断面研究的目的是开发一种算法，以预测运动员对第三方检测（TPT）补充剂的使用情况。因此，使用了一份营养补充剂问卷，其中有一部分是关于自我报告的TPT补充剂使用情况。
将结果随机分配到一个训练数据集，使用逻辑回归模型识别预测因素，或分配到一个交叉验证数据集。训练数据用于开发一种算法，该算法的分数从0到100，用于预测TPT营养补充剂的使用或不使用情况。
共纳入了410名美国全国大学体育协会（NCAA）一级联盟的学生运动员（年龄：21.4±1.6岁，53%为女性，来自20多个运动项目）。然后随机选择了320名，其中34%（n = 109）的使用者一致报告他们使用的所有补充剂都是经过TPT检测的。分析得出了一个与TPT使用或不使用相关的10项算法。风险象限最适合对TPT使用不一致的低风险与高风险进行分类，截断值≥60%（χ(4)=61.26，P<0.001），曲线下面积（AUC）为0.78，具有合理性。根据该算法定义的TPT使用情况（是/否）与风险行为（低风险与高风险）之间存在显著关联（χ(1)=58.6，P<0.001）。该算法具有较高的敏感性，能正确分类89%的非TPT使用者，但特异性较低，只能正确分类49%的TPT使用者。交叉验证（n = 34）证实了这一点，尽管AUC较低（0.61），但敏感性较高（83%）。
该算法能以合理的准确性对高风险的TPT使用不一致者进行分类，但缺乏对低风险的持续使用者进行分类的特异性。这种方法在识别能从额外咨询中受益的运动员方面应该是有用的。
