Myers John V, Marks Charles, Kim Joanne, Fareed Naleef, Thomas Neena, Whitley Penn, Fernandez Soledad
Department of Biomedical Informatics, The Ohio State University, Columbus.
Millennium Health, San Diego, California.
JAMA Netw Open. 2025 Jun 2;8(6):e2514402. doi: 10.1001/jamanetworkopen.2025.14402.
Provisional estimates of fatal drug overdoses in the US are lagging by 6 months. Efforts to estimate the overdose burden for this 6-month lag window require up-to-date data, such as real-time urine drug test (UDT) data, capable of identifying sudden changes in the overdose trajectory, such as the increase in overdose deaths experienced at the beginning of the COVID-19 pandemic.
To evaluate the utility of using aggregated UDT data to estimate national-level drug overdose deaths for the 6-month lag window in which overdose data are unavailable.
DESIGN, SETTING, AND PARTICIPANTS: This cross-sectional study included 3 135 748 urine samples submitted for UDT by Millennium Health from patients aged 18 years or older in substance use disorder treatment health care facilities across the US between January 1, 2015, and January 31, 2025. Urine drug test results were aggregated to generate monthly positivity rates and mean creatinine-normalized levels of fentanyl and methamphetamine (among the sample testing positive for fentanyl). Monthly, national drug overdose mortality counts were obtained from the Centers for Disease Control and Prevention.
Urine drug testing.
Drug overdose death totals were estimated for every 6-month period from January to June 2019 through August 2024 to January 2025 by training generalized linear models with a negative binomial distribution on the preceding 4 years of data and using monthly UDT data to generate overdose estimates for the 6-month lag window of interest. Mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean squared error (RMSE) were calculated by comparing projected monthly estimates with observed overdose death totals.
A total of 3 135 748 UDT specimens (57.2% from men; mean [SD] age, 38.1 [12.4] years) were included in this study. From 2019 to August 2024, 537 104 people died of an overdose in the US, with a substantial increase in early 2020 at the onset of the COVID-19 pandemic. The UDT modeling strategy (MAPE, 7.1%; MAE, 540.9 deaths per month; RMSE, 659.4) outperformed the baseline autoregressive integrated moving average model (MAPE, 9.0%; MAE, 704.9 deaths per month; RMSE, 1075.8) across all metrics. Furthermore, the model detected the sudden increase in overdose deaths at the start of the COVID-19 pandemic.
In this cross-sectional study, findings suggested that aggregated UDT data may be used to estimate up-to-date overdose death trends. Model implementation can be improved by introducing additional exposure variables, such as those related to drug seizures and syndromic surveillance.
美国药物过量致死的临时估计滞后6个月。要估算这一滞后6个月期间的过量用药负担，需要最新数据，比如能够识别过量用药轨迹突然变化（如2019冠状病毒病大流行开始时过量用药死亡人数增加）的实时尿液药物检测（UDT）数据。
评估使用汇总的UDT数据估算在无法获取过量用药数据的6个月滞后期间全国层面药物过量致死人数的效用。
设计、设置和参与者：这项横断面研究纳入了2015年1月1日至2025年1月31日期间美国物质使用障碍治疗医疗机构中18岁及以上患者提交给千禧健康进行UDT检测的3135748份尿液样本。汇总尿液药物检测结果以生成每月阳性率以及芬太尼和甲基苯丙胺（在芬太尼检测呈阳性的样本中）的平均肌酐标准化水平。每月从疾病控制和预防中心获取全国药物过量致死人数。
尿液药物检测。
通过对前4年数据训练具有负二项分布的广义线性模型，并使用每月UDT数据生成感兴趣的6个月滞后期间的过量用药估计值，估算2019年1月至6月至2024年8月至2025年1月每个6个月期间的药物过量死亡总数。通过将预测的每月估计值与观察到的过量用药死亡总数进行比较，计算平均绝对误差（MAE）、平均绝对百分比误差（MAPE）和均方根误差（RMSE）。
本研究共纳入3135748份UDT样本（57.2%来自男性；平均[标准差]年龄为38.1[12.4]岁）。从2019年到2024年8月，美国有537104人死于药物过量，在2020年初2019冠状病毒病大流行开始时大幅增加。UDT建模策略（MAPE为7.1%；MAE为每月540.9例死亡；RMSE为659.4）在所有指标上均优于基线自回归积分滑动平均模型（MAPE为9.0%；MAE为每月704.9例死亡；RMSE为1075.8）。此外，该模型检测到了2019冠状病毒病大流行开始时过量用药死亡人数的突然增加。
在这项横断面研究中，结果表明汇总的UDT数据可用于估算最新的过量用药死亡趋势。通过引入额外的暴露变量（如与缉获毒品和症状监测相关的变量）可以改进模型的实施。
