Yan Yifang, Chen Qiushi, Nasir Rafay, Griffin Paul, Bone Curtis, Tuan Wen-Jan
The Harold and Inge Marcus Department of Industrial and Manufacturing Engineering, The Pennsylvania State University, University Park, PA, USA.
The Harold and Inge Marcus Department of Industrial and Manufacturing Engineering, The Pennsylvania State University, University Park, PA, USA.
Comput Biol Med. 2025 Apr;188:109870. doi: 10.1016/j.compbiomed.2025.109870. Epub 2025 Feb 20.
Given the significantly increased number of individuals prescribed stimulants in the past decade, there has been growing concern regarding the risk of cardiovascular events among adults on stimulant therapy. We aimed to quantify the added risk of cardiovascular events by prescription stimulant use and characterize patients who were adversely affected.
Using electronic health records of adults with Attention-Deficit/Hyperactivity Disorder from TriNetX Research Network in 2010-2020, we developed and compared different machine learning models to predict one-year cardiovascular risk based on individual's prescription stimulant use, demographics, and comorbidities for four separate age groups. With the trained risk prediction models, we estimated added risk of cardiovascular events and utilized association rule mining (ARM) to identify clinical characteristics of patients adversely affected by prescription stimulant use.
The study cohort consisted of 219,965 adults, including 102,138 (46.4 %) persons on stimulant therapy. All prediction models achieved high areas under receiver operating characteristic curve of 0.77-0.84 in predicting one-year cardiovascular risk across all age groups. Of patients with 25 % highest added risks, ARM identified critical features in major categories including common risk factors of cardiovascular events, prior cardiovascular events, substance use disorders, and psychological disorders. A watch list of comorbidities was constructed and validated for each age group to show added risk of prescribing stimulants to patients with these conditions.
We integrated predictive modeling and data mining to characterize patients adversely affected by prescription stimulant use. Future research is needed to externally validate identified features to guide safer stimulant prescribing.
鉴于在过去十年中使用兴奋剂的处方数量显著增加，人们越来越担心接受兴奋剂治疗的成年人发生心血管事件的风险。我们旨在量化使用处方兴奋剂导致心血管事件的额外风险，并描述受不利影响的患者特征。
利用2010 - 2020年TriNetX研究网络中患有注意力缺陷/多动障碍的成年人的电子健康记录，我们开发并比较了不同的机器学习模型，以根据个体的处方兴奋剂使用情况、人口统计学特征和共病情况，预测四个不同年龄组的一年心血管风险。借助训练好的风险预测模型，我们估计了心血管事件的额外风险，并利用关联规则挖掘（ARM）来确定受处方兴奋剂使用不利影响的患者的临床特征。
研究队列包括219,965名成年人，其中102,138人（46.4%）接受兴奋剂治疗。所有预测模型在预测所有年龄组的一年心血管风险时，受试者工作特征曲线下面积均达到0.77 - 0.84的高水平。在额外风险最高的25%的患者中，ARM确定了主要类别中的关键特征，包括心血管事件的常见风险因素、既往心血管事件、物质使用障碍和心理障碍。为每个年龄组构建并验证了一份共病观察清单，以显示给患有这些疾病的患者开兴奋剂处方的额外风险。
我们整合了预测建模和数据挖掘，以描述受处方兴奋剂使用不利影响的患者特征。未来需要进行外部研究以验证已确定的特征，从而指导更安全的兴奋剂处方开具。
