Background Osteoporosis is a metabolic bone disorder characterised by decreased bone mineral density and mass. Due to its asymptomatic nature, it often remains undiagnosed and untreated until a fracture occurs. Traditionally, treatment decisions for osteoporosis are based on clinical appropriateness while balancing the treatment's risks and benefits. Machine learning (ML) is revolutionising healthcare domains through pattern recognition of previously “unseen” observations. Presently, its application in osteoporosis is limited to early diagnosis. More research is needed to examine its role in guiding osteoporosis treatment. This study aims to identify new predictive attributes for osteoporosis treatment using ML techniques on data from the Irish Hip Fracture Database (IHFD). Methods Datasets from January to March 2023 in University Hospital Waterford were sourced from the IHFD. Osteoporosis treatment decisions were obtained from discharge letters. Preliminary data cleaning was performed in Excel with zero-variance and near-zero predictor. Attributes excluded. The dataset was entered into the WEKA 3.8.6 environment for ML processing. Results The initial dataset containing 141 instances and 32 attributes was refined using the Correlation Feature Selection and Ranker Search Method, identifying key osteoporosis treatment predictors. The highest correlation attributes are pre-fracture total score, pre-fracture indoor score, and age. Moderately positive correlations are discharge destination, pre-fracture outdoor and shopping score, ASA grade, Length-of-stay, admission code, Admission 4AT score, Frailty scale, and fracture type. The implemented J48 Tree ML-trained model revealed Correctly Classified Instances and Incorrectly Classified Instances of 98.24% and 1.7%, respectively, indicating a high prediction accuracy rate. Conclusion This study demonstrates the potential of ML in enhancing osteoporosis treatment decision-making by leveraging datasets from the IHFD. Integrating ML algorithms with traditional approaches can provide a comprehensive, nuanced and personal approach to osteoporosis treatment and patient care. The study opens avenues for future research in applying big data and advanced analytics in healthcare, underscoring the evolving landscape of medical decision-making.

中文翻译：

---

真正的突破：机器学习利用爱尔兰髋部骨折数据库破解骨质疏松症治疗的密码


背景 骨质疏松症是一种代谢性骨病，其特征是骨矿物质密度和质量下降。由于其无症状的性质，通常直到发生骨折才得到诊断和治疗。传统上，骨质疏松症的治疗决策是基于临床适当性，同时平衡治疗的风险和益处。机器学习 (ML) 通过对以前“未见过的”观察结果进行模式识别，正在彻底改变医疗保健领域。目前，其在骨​​质疏松症中的应用仅限于早期诊断。需要更多的研究来检验其在指导骨质疏松症治疗中的作用。本研究旨在利用机器学习技术对爱尔兰髋部骨折数据库 (IHFD) 的数据进行分析，以确定骨质疏松症治疗的新预测属性。方法 沃特福德大学医院 2023 年 1 月至 3 月的数据集来源于 IHFD。骨质疏松症治疗决定是从出院信中获得的。在 Excel 中使用零方差和接近零预测器进行初步数据清理。排除属性。将数据集输入到WEKA 3.8.6环境中进行ML处理。结果 使用相关特征选择和排序搜索方法对包含 141 个实例和 32 个属性的初始数据集进行了细化，确定了关键的骨质疏松症治疗预测因子。相关性最高的属性是骨折前总分、骨折前室内评分和年龄。中度正相关包括出院目的地、骨折前户外和购物评分、ASA 等级、住院时间、入院代码、入院 4AT 评分、虚弱量表和骨折类型。实施的 J48 Tree ML 训练模型显示正确分类实例和错误分类实例分别为 98.24% 和 1。分别为7%，表明预测准确率较高。结论 这项研究证明了 ML 通过利用 IHFD 数据集来增强骨质疏松症治疗决策的潜力。将机器学习算法与传统方法相结合可以为骨质疏松症治疗和患者护理提供全面、细致和个性化的方法。该研究为未来在医疗保健中应用大数据和高级分析的研究开辟了途径，强调了医疗决策不断变化的格局。