BACKGROUND Multimorbidity is highly relevant for both service commissioning and clinical decision-making. Optimization of variables assessing multimorbidity in order to enhance chronic care management is an unmet need. To this end, we have explored the contribution of multimorbidity to predict use of healthcare resources at community level by comparing the predictive power of four different multimorbidity measures. METHODS A population health study including all citizens ≥18 years (n = 6,102,595) living in Catalonia (ES) on 31 December 2014 was done using registry data. Primary care service utilization during 2015 was evaluated through four outcome variables: A) Frequent attendants, B) Home care users, C) Social worker users, and, D) Polypharmacy. Prediction of the four outcome variables (A to D) was carried out with and without multimorbidity assessment. We compared the contributions to model fitting of the following multimorbidity measures: i) Charlson index; ii) Number of chronic diseases; iii) Clinical Risk Groups (CRG); and iv) Adjusted Morbidity Groups (GMA). RESULTS The discrimination of the models (AUC) increased by including multimorbidity as covariate into the models, namely: A) Frequent attendants (0.771 vs 0.853), B) Home care users (0.862 vs 0.890), C) Social worker users (0.809 vs 0.872), and, D) Polypharmacy (0.835 vs 0.912). GMA showed the highest predictive power for all outcomes except for polypharmacy where it was slightly below than CRG. CONCLUSIONS We confirmed that multimorbidity assessment enhanced prediction of use of healthcare resources at community level. The Catalan population-based risk assessment tool based on GMA presented the best combination of predictive power and applicability.

中文翻译：

---

多发病率作为初级保健中卫生服务利用的预测指标：基于注册表的加泰罗尼亚人口研究。

背景技术多发病与服务调试和临床决策都息息相关。为增强慢性病管理水平而优化评估多发病率的变量是未满足的需求。为此，我们通过比较四种不同的多发病率指标的预测能力，探索了多发病率在社区一级预测医疗资源使用方面的贡献。方法采用注册表数据对2014年12月31日居住在加泰罗尼亚（ES）的所有≥18岁的公民（n = 6102595）进行了人口健康研究。通过四个结果变量评估了2015年期间的初级保健服务利用率：A）常客，B）家庭护理使用者，C）社会工作者使用者和D）多元药房。在有和没有多发病率评估的情况下，对四个结果变量（A到D）进行了预测。我们比较了以下多种发病率测度对模型拟合的贡献：i）查尔森指数；ii）慢性病的数量；iii）临床风险组（CRG）；iv）调整发病率组（GMA）。结果通过将多发病率作为协变量纳入模型，模型的歧视性（AUC）有所提高，即：A）经常服务人员（0.771 vs 0.853），B）家庭护理用户（0.862 vs 0.890），C）社会工作者用户（0.809 vs D）多元药店（0.835对0.912）；以及（0.872）。GMA对所有结局均显示出最高的预测能力，除了多元药店（略低于CRG）之外，其他药物店均如此。结论我们证实，多发病率评估增强了社区一级医疗资源使用的预测。