Coste Joël, Mandereau-Bruno Laurence, Constantinou Panayotis, Olié Valérie, Thuret Anne, Bruyand Mathias, Makovski Tatjana T, Carcaillon-Bentata Laure
Department of Non-Communicable Diseases and Injuries, French Public Health Agency (Santé publique France), Saint-Maurice, France.
Data Science Division, French Public Health Agency (Santé publique France), Saint-Maurice, France.
Eur J Public Health. 2025 Aug 1;35(4):624-634. doi: 10.1093/eurpub/ckaf040.
Healthcare claims data are increasingly used to derive chronic condition (CC) surveillance indicators, although comparative evidence with self-reported data remains scarce. We explored the agreement and comparative validity (concurrent and predictive) of 20 CC prevalence indicators independently constructed using the French National Health Data System (SNDS) and Health, Health Care, and Insurance Survey (ESPS 2010-2014). Individual data from 5039 ESPS participants aged ≥25 years, representative of the 2010 French general population, were linked to the SNDS. Follow-up data included a 2014 health self-assessment and 5-year mortality. We considered 20 CCs with corresponding SNDS case-identifying algorithms and self-reported information from ESPS, including most cardiovascular diseases and frequent cancers. Kappa statistics assessed agreement between CC indicators across databases. Polytomous and dichotomous logistic regression assessed determinants of disagreement between sources and associations of indicators with health outcomes (concurrent and predictive validity). Prevalence values were much higher with survey data except for hypertension, diabetes, thyroid disorders, epilepsy, and most cancers for which they were closer (±20%) to claims data. Agreement between CC indicators varied from the strongest (hypertension, diabetes, thyroid disorders, most cancers) to the weakest (cardiac rhythm disorders, peptic ulcer, chronic liver diseases). Sex, age, and multimorbidity strongly influenced agreement. Most claims database indicators were more strongly associated with health outcomes. Health interview surveys and healthcare claims-derived indicators are not interchangeable given their specific determinants. Since no general rule applies to all CCs, the advantages and disadvantages of each data source should be closely considered in case-to-case analysis.
医疗保健理赔数据越来越多地用于得出慢性病（CC）监测指标，尽管与自我报告数据的比较证据仍然很少。我们探讨了使用法国国家卫生数据系统（SNDS）和健康、医疗保健与保险调查（ESPS 2010 - 2014）独立构建的20个CC患病率指标的一致性和比较有效性（同时性和预测性）。来自5039名年龄≥25岁的ESPS参与者的个体数据（代表2010年法国普通人群）与SNDS相关联。随访数据包括2014年的健康自我评估和5年死亡率。我们考虑了20种CC，以及相应的SNDS病例识别算法和来自ESPS的自我报告信息，包括大多数心血管疾病和常见癌症。卡方统计评估了不同数据库中CC指标之间的一致性。多分类和二分类逻辑回归评估了数据源之间不一致的决定因素以及指标与健康结果的关联（同时性和预测有效性）。除高血压、糖尿病、甲状腺疾病、癫痫以及大多数癌症（其患病率与理赔数据更接近（±20%））外，调查数据得出的患病率值要高得多。CC指标之间的一致性从最强（高血压、糖尿病、甲状腺疾病、大多数癌症）到最弱（心律失常、消化性溃疡、慢性肝病）各不相同。性别、年龄和多种疾病共存强烈影响一致性。大多数理赔数据库指标与健康结果的关联更强。鉴于健康访谈调查和医疗保健理赔衍生指标有其特定的决定因素，它们不可相互替代。由于没有适用于所有CC的通用规则，在个案分析中应仔细考虑每个数据源的优缺点。
