Shi Jia, Yin Chunyan, Wu Jinyi
Department of public health, Wuhan Fourth Hospital, Qiaokou, Wuhan, 430022, China.
Department of medical Records and Statistics, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Jianghan, Wuhan, 430022, China.
J Cancer. 2024 May 13;15(11):3625-3632. doi: 10.7150/jca.96052. eCollection 2024.
Due to inconsistent results in earlier investigations regarding the relationship between vitamin D and prostate-specific antigen (PSA), this study was conducted to gain a deeper understanding of the association between vitamin D and PSA. A total of 7174 male samples with 25(OH)D, PSA, and other variables were obtained from the National Health and Nutrition Examination Survey (NHANES) database. Three models, created through stepwise logistic regression, were employed to examine the dose-response association between PSA and 25(OH)D. Subsequently, restricted cubic spline analysis (RCS) was used to explore the nonlinear association between 25(OH)D and PSA. The study also compared the performance of four machine learning models in predicting PSA levels. The dose-response relationship indicated a negative impact of high 25(OH)D levels on PSA (p for trend 0.05). The odds ratio (OR) of Q4 (7.73 with 95% CI (0.26, 15.76)) was significantly higher than Q1 (6.23 with 95% CI (0.24, 12.57)). OR values in Q2 and Q3 were less than 1 (Q2= 0.57 with 95% CI (-6.37, 8.04) and Q3= 0.26 with 95% CI (-5.94, 6.86)), suggesting a potential protective effect of 25(OH)D on PSA. RCS analysis revealed a U-shaped relationship between blood 25(OH)D levels and PSA, with serum 25(OH)D in the range of 20-134 ng/ml showing a potential decrease in PSA levels. Above this range, an increase in 25(OH)D might elevate PSA levels. Age (2.67 with 95% CI 2.24 to 3.1) and BMI (17.52 with 95% CI 7.65 to 26.32), along with the OR of obesity (10.36 with 95% CI 0.68 to 20.18), were identified as potential PSA risk factors. Among the machine learning models, the random forest algorithm performed the best in predicting PSA levels. This study revealed a U-shaped relationship between 25(OH)D and PSA, with PSA potentially declining when 25(OH)D is between 20 and 134 ng/mL and possibly rising above this range. The random forest method proved effective in both predicting PSA levels and guiding vitamin D dosage.
由于早期关于维生素D与前列腺特异性抗原（PSA）之间关系的研究结果不一致，因此开展了本研究以更深入地了解维生素D与PSA之间的关联。从国家健康与营养检查调查（NHANES）数据库中获取了总共7174份包含25(OH)D、PSA及其他变量的男性样本。通过逐步逻辑回归创建了三个模型，用于检验PSA与25(OH)D之间的剂量反应关联。随后，使用受限立方样条分析（RCS）来探索25(OH)D与PSA之间的非线性关联。该研究还比较了四种机器学习模型在预测PSA水平方面的表现。剂量反应关系表明，高25(OH)D水平对PSA有负面影响（趋势p值为0.05）。第四四分位数的优势比（OR）（7.73，95%置信区间为(0.26, 15.76)）显著高于第一四分位数（6.23，95%置信区间为(0.24, 12.57)）。第二和第三四分位数的OR值小于1（第二四分位数 = 0.57，95%置信区间为(-6.37, 8.04)；第三四分位数 = 0.26，95%置信区间为(-5.94, 6.86)），表明25(OH)D对PSA有潜在的保护作用。RCS分析显示血液25(OH)D水平与PSA之间呈U形关系，血清25(OH)D在20 - 134 ng/ml范围内时PSA水平可能降低。高于此范围，25(OH)D的增加可能会使PSA水平升高。年龄（2.67，95%置信区间为2.24至3.1）、体重指数（17.52，95%置信区间为7.65至26.32）以及肥胖的OR值（10.36，95%置信区间为0.68至20.18）被确定为潜在 的PSA风险因素。在机器学习模型中，随机森林算法在预测PSA水平方面表现最佳。本研究揭示了25(OH)D与PSA之间的U形关系，当25(OH)D在20至134 ng/mL之间时PSA可能下降，高于此范围时可能上升。随机森林方法在预测PSA水平和指导维生素D剂量方面均被证明是有效的。
