Abstract：
OBJECTIVE: Metformin is clinically effective in treating polycystic ovary syndrome (PCOS) with insulin resistance (IR), while its efficacy varies among individuals. This study aims to develop a machine learning model to predict the efficacy of metformin in improving insulin sensitivity among women with PCOS and IR.
METHODS: This is a retrospective analysis of a multicenter, randomized controlled trial involving 114 women diagnosed with PCOS and IR. All women received metformin treatment for 4 months. We incorporated 27 baseline clinical variables of the women into the construction of our machine learning model. We firstly compared 4 commonly used feature selection methods to screen valuable clinical variables. Then we used the valuable variables as inputs to evaluate the performance of 5 machine learning models, including k-Nearest Neighbors, Support Vector Machine, Logistic Regression, Random Forest, and Extreme Gradient Boosting, in predicting the efficacy of metformin.
RESULTS: Among the 5 machine learning models, Support Vector Machine performed the best with an area under the receiver operating characteristic curve of 0.781 (95% confidence interval [CI]: 0.772-0.791). The key predictive variables identified were homeostasis model assessment of insulin resistance, body mass index, and low-density lipoprotein cholesterol.
CONCLUSIONS: The developed machine learning model could be applied to predict the efficacy of metformin in improving insulin sensitivity among women with PCOS and IR. The result could help doctors evaluate the efficacy of metformin in advance, optimize treatment plans, and thereby enhance overall clinical outcomes.
METHODS: This is a retrospective analysis of a multicenter, randomized controlled trial involving 114 women diagnosed with PCOS and IR. All women received metformin treatment for 4 months. We incorporated 27 baseline clinical variables of the women into the construction of our machine learning model. We firstly compared 4 commonly used feature selection methods to screen valuable clinical variables. Then we used the valuable variables as inputs to evaluate the performance of 5 machine learning models, including k-Nearest Neighbors, Support Vector Machine, Logistic Regression, Random Forest, and Extreme Gradient Boosting, in predicting the efficacy of metformin.
RESULTS: Among the 5 machine learning models, Support Vector Machine performed the best with an area under the receiver operating characteristic curve of 0.781 (95% confidence interval [CI]: 0.772-0.791). The key predictive variables identified were homeostasis model assessment of insulin resistance, body mass index, and low-density lipoprotein cholesterol.
CONCLUSIONS: The developed machine learning model could be applied to predict the efficacy of metformin in improving insulin sensitivity among women with PCOS and IR. The result could help doctors evaluate the efficacy of metformin in advance, optimize treatment plans, and thereby enhance overall clinical outcomes.
摘要：
目的：二甲双胍治疗多囊卵巢综合征（PCOS）伴胰岛素抵抗（IR）是临床有效的，而其功效因个体而异。这项研究旨在开发一种机器学习模型，以预测二甲双胍在改善PCOS和IR女性胰岛素敏感性方面的功效。
方法：这是对多中心的回顾性分析，纳入114名诊断为PCOS和IR的女性的随机对照试验。所有女性均接受二甲双胍治疗4个月。我们将女性的27个基线临床变量纳入我们的机器学习模型的构建中。我们首先比较了四种常用的特征选择方法，以筛选有价值的临床变量。然后，我们使用有价值的变量作为输入来评估五种机器学习模型的性能，包括k-最近邻居(KNN)，支持向量机(SVM)逻辑回归(LR)，随机森林(RF),和极端梯度提升(Xgboost)，预测二甲双胍的疗效。
结果：在五种机器学习模型中，SVM表现最好，受试者工作特征曲线下面积（AUC）为0.781（95％置信区间[CI]：0.772-0.791）。确定的关键预测变量是胰岛素抵抗的稳态模型评估(HOMA-IR)，体重指数（BMI），和低密度脂蛋白胆固醇（LDL-C）。
结论：开发的机器学习模型可用于预测二甲双胍改善PCOS和IR女性胰岛素敏感性的疗效。结果可以帮助医生提前评估二甲双胍的疗效，优化治疗计划，从而提高整体临床结果。
方法：这是对多中心的回顾性分析，纳入114名诊断为PCOS和IR的女性的随机对照试验。所有女性均接受二甲双胍治疗4个月。我们将女性的27个基线临床变量纳入我们的机器学习模型的构建中。我们首先比较了四种常用的特征选择方法，以筛选有价值的临床变量。然后，我们使用有价值的变量作为输入来评估五种机器学习模型的性能，包括k-最近邻居(KNN)，支持向量机(SVM)逻辑回归(LR)，随机森林(RF),和极端梯度提升(Xgboost)，预测二甲双胍的疗效。
结果：在五种机器学习模型中，SVM表现最好，受试者工作特征曲线下面积（AUC）为0.781（95％置信区间[CI]：0.772-0.791）。确定的关键预测变量是胰岛素抵抗的稳态模型评估(HOMA-IR)，体重指数（BMI），和低密度脂蛋白胆固醇（LDL-C）。
结论：开发的机器学习模型可用于预测二甲双胍改善PCOS和IR女性胰岛素敏感性的疗效。结果可以帮助医生提前评估二甲双胍的疗效，优化治疗计划，从而提高整体临床结果。
