背景:医院服务通常使用病例组合工具进行报销,该工具根据诊断和程序对患者进行分组。我们最近开发了一种案例混合工具(即,Queralt系统)旨在支持临床医生进行患者管理。在这项研究中,我们比较了一种广泛使用的工具的性能(即,APR-DRG)与Queralt系统。
方法:对加泰罗尼亚卫生研究所的八家医院中的任何一家进行了所有入院的回顾性分析(即,大约,2019年加泰罗尼亚所有住院治疗的30%)。成本是从完全成本核算中检索的。使用电子健康记录来计算APR-DRG组和Queralt指数,及其用于诊断的不同子指标(主要诊断,入院时合并症,以及住院期间发生的并发症)和手术(主要和次要手术)。主要目标是工具的预测能力;我们还调查了效率和组内同质性。
结果:分析包括166,837次住院事件,平均成本为4935欧元(中位数2616;四分位数范围1011-5543)。Queralt系统的组件具有更高的效率(即,每个病例组合工具中增加的组百分比所涵盖的费用和住院百分比)和较低的异质性.在预先设定的阈值下预测成本的逻辑模型(即,80岁,第90,和第95百分位数)显示了Queralt系统的更好性能,特别是当结合诊断和程序(DP)时:80岁的接收器工作特性曲线下的面积,第90,APR-DRG的第95个成本百分位数分别为0.904、0.882和0.863,和0.958、0.945和0.928的QueraltDP;APR-DRG的精确召回曲线下面积的相应值分别为0.522、0.604和0.699,以及0.748、0.7966和0.834的QueraltDP。同样,预测实际成本的线性模型在Queralt系统的情况下拟合更好。
结论:Queralt系统,最初开发用于预测医院结果,对预测住院费用具有良好的性能和效率。
BACKGROUND: Hospital services are typically reimbursed using case-mix tools that group patients according to diagnoses and procedures. We recently developed a case-mix tool (i.e., the Queralt system) aimed at supporting clinicians in patient management. In this study, we compared the performance of a broadly used tool (i.e., the APR-DRG) with the Queralt system.
METHODS: Retrospective analysis of all admissions occurred in any of the eight hospitals of the Catalan Institute of Health (i.e., approximately, 30% of all hospitalizations in Catalonia) during 2019. Costs were retrieved from a full cost accounting. Electronic health records were used to calculate the APR-DRG group and the Queralt index, and its different sub-indices for diagnoses (main diagnosis, comorbidities on admission, andcomplications occurred during hospital stay) and procedures (main and secondary procedures). The primary objective was the predictive capacity of the tools; we also investigated efficiency and within-group homogeneity.
RESULTS: The analysis included 166,837 hospitalization episodes, with a mean cost of € 4,935 (median 2,616; interquartile range 1,011-5,543). The components of the Queralt system had higher efficiency (i.e., the percentage of costs and hospitalizations covered by increasing percentages of groups from each case-mix tool) and lower heterogeneity. The logistic model for predicting costs at pre-stablished thresholds (i.e., 80th, 90th, and 95th percentiles) showed better performance for the Queralt system, particularly when combining diagnoses and procedures (DP): the area under the receiver operating characteristics curve for the 80th, 90th, 95th cost percentiles were 0.904, 0.882, and 0.863 for the APR-DRG, and 0.958, 0.945, and 0.928 for the Queralt DP; the corresponding values of area under the precision-recall curve were 0.522, 0.604, and 0.699 for the APR-DRG, and 0.748, 0.7966, and 0.834 for the Queralt DP. Likewise, the linear model for predicting the actual cost fitted better in the case of the Queralt system.
CONCLUSIONS: The Queralt system, originally developed to predict hospital outcomes, has good performance and efficiency for predicting hospitalization costs.