Vesteghem Charles, Bøgsted Martin, Cronin-Fenton Deirdre, Poulsen Laurids Østergaard
Center for Clinical Data Science, Aalborg University and Aalborg University Hospital, Aalborg, Denmark.
Clinical Cancer Research Centre, Aalborg University Hospital, Aalborg, Denmark.
Clin Epidemiol. 2024 Mar 6;16:165-174. doi: 10.2147/CLEP.S442591. eCollection 2024.
Reconstructing patient treatment trajectories is important to generate real-world evidence for epidemiological studies. The Danish National Patient Registry (DNPR) contains information about drug prescriptions and could therefore be used to reconstruct treatment trajectories. We aimed to evaluate and enhance two existing methods to reconstruct systemic anticancer treatment trajectories.
This study was based on data from 8738 consecutive patients with solid tumors treated in the North Denmark Region between 2009 and 2019. Two approaches found in the literature as well as two new approaches were applied to the DNPR data. All methods relied on time intervals between two consecutive drug administrations to determine if they belonged to the same treatment line. MedOnc, a local dataset from the Department of Oncology, Aalborg University Hospital was used as a reference. To evaluate the performance of each method, F1-scores were calculated after matching the lines identified in both datasets. We used three different matching strategies: stringent matching, loose matching, and matching based on line numbers, controlling for overfitting.
Overall, the two new approaches outperformed the simpler and best performing of the two existing methods, with F1-scores of 0.47 and 0.45 vs 0.44 for stringent matching and 0.84 and 0.83 vs 0.82 for loose matching. Nevertheless, only one of the new methods outperformed the existing simpler method when matching on the number of lines (0.73 vs 0.72). Large differences were seen by cancer site, especially for the stringent and line number matchings. Performances were relatively stable by calendar year.
The high F1-scores for the new methods confirm that they should be generally preferred to reconstruct systemic anticancer treatment trajectories using the DNPR.
重建患者治疗轨迹对于生成流行病学研究的真实世界证据很重要。丹麦国家患者登记处（DNPR）包含有关药物处方的信息，因此可用于重建治疗轨迹。我们旨在评估和改进两种现有的重建全身抗癌治疗轨迹的方法。
本研究基于2009年至2019年在北丹麦地区连续接受治疗的8738例实体瘤患者的数据。将文献中发现的两种方法以及两种新方法应用于DNPR数据。所有方法都依赖于两次连续给药之间的时间间隔来确定它们是否属于同一条治疗线。奥尔堡大学医院肿瘤学系的本地数据集MedOnc用作参考。为了评估每种方法的性能，在匹配两个数据集中识别出的治疗线后计算F1分数。我们使用了三种不同的匹配策略：严格匹配、宽松匹配和基于治疗线编号的匹配，以控制过拟合。
总体而言，两种新方法的表现优于两种现有方法中较简单且表现最佳的方法，严格匹配的F1分数分别为0.47和0.45，而现有方法为0.44；宽松匹配的F1分数分别为0.84和0.83，而现有方法为0.82。然而，在按治疗线数量匹配时，只有一种新方法优于现有的较简单方法（0.73对0.72）。不同癌症部位存在较大差异，尤其是在严格匹配和按治疗线编号匹配时。各日历年的表现相对稳定。
新方法的高F1分数证实，在使用DNPR重建全身抗癌治疗轨迹时，通常应优先选择这些方法。
