Lambert Judith, Leutenegger Anne-Louise, Baudot Anaïs, Jannot Anne-Sophie
Sorbonne Université, Université Paris Cité, INSERM, Centre de Recherche des Cordeliers, Paris, F-75006, France.
HeKA, Inria Paris, Paris, F-75015, France.
BMC Med Res Methodol. 2025 Mar 15;25(1):72. doi: 10.1186/s12874-025-02459-8.
Patient stratification is the cornerstone of numerous health investigations, serving to enhance the estimation of treatment efficacy and facilitating patient matching. To stratify patients, similarity measures between patients can be computed from clinical variables contained in medical health records. These variables have both values and labels structured in ontologies or other classification systems. The relevance of considering variable label relationships in the computation of patient similarity measures has been poorly studied.
We adapt and evaluate several weighted versions of the Cosine similarity in order to consider structured label relationships to compute patient similarities from a medico-administrative database.
As a use case, we clustered patients aged 60 years from their annual medicine reimbursements contained in the Échantillon Généraliste des Bénéficiaires, a random sample of a French medico-administrative database. We used four patient similarity measures: the standard Cosine similarity, a weighted Cosine similarity measure that includes variable frequencies and two weighted Cosine similarity measures that consider variable label relationships. We construct patient networks from each similarity measure and identify clusters of patients using the Markov Cluster algorithm. We evaluate the performance of the different similarity measures with enrichment tests based on patient diagnoses.
The weighted similarity measures that include structured variable label relationships perform better to identify similar patients. Indeed, using these weighted measures, we identify more clusters associated with different diagnose enrichment. Importantly, the enrichment tests provide clinically interpretable insights into these patient clusters.
Considering label relationships when computing patient similarities improves stratification of patients regarding their health status.
患者分层是众多健康调查的基石，有助于提高治疗效果评估并促进患者匹配。为了对患者进行分层，可以根据医疗健康记录中包含的临床变量计算患者之间的相似性度量。这些变量在本体或其他分类系统中具有值和标签结构。在计算患者相似性度量时考虑变量标签关系的相关性研究较少。
我们调整并评估了余弦相似性的几个加权版本，以便考虑结构化标签关系，从医疗管理数据库中计算患者相似度。
作为一个用例，我们根据法国医疗管理数据库的随机样本——受益人群通用样本中包含的年度药物报销情况，对60岁的患者进行聚类。我们使用了四种患者相似性度量：标准余弦相似性、一种包含变量频率的加权余弦相似性度量以及两种考虑变量标签关系的加权余弦相似性度量。我们从每个相似性度量构建患者网络，并使用马尔可夫聚类算法识别患者集群。我们基于患者诊断通过富集测试评估不同相似性度量的性能。
包含结构化变量标签关系的加权相似性度量在识别相似患者方面表现更好。事实上，使用这些加权度量，我们识别出更多与不同诊断富集相关的集群。重要的是，富集测试为这些患者集群提供了临床上可解释的见解。
在计算患者相似性时考虑标签关系可改善患者健康状况的分层。
