Chang Hsien-Yen
Department of Health Policy & Management, Bloomberg School of Public Health, Johns Hopkins University, 624 N. Broadway, Baltimore, MD 21205, USA.
J Public Health (Oxf). 2014 Jun;36(2):300-7. doi: 10.1093/pubmed/fdt059. Epub 2013 Jun 5.
Incorporating longitudinal information into risk-adjustment models has been considered important. This study aimed to evaluate how morbidity trajectories impact risk-adjustment models in identifying high-cost cases.
Claims-based risk adjusters, with or without morbidity trajectories derived from 3-year claims from Taiwan's National Insurance System, were used to predict being a prospective high-cost user. A random sample of Taiwanese National Health Insurance enrollees continuously enrolled from 2002 to 2005 (n = 147,892) was the study sample. A logistic regression model was employed. The performance measures, based on the split analysis, included statistical indicators (c-statistics, sensitivity and predictive positive value), proportions of true cases identified by models and medical utilization of predicted cases.
As the comprehensiveness of risk adjustment models increased, the performance of the models generally increased. The effect of adding trajectories on the model performance decreased as the comprehensiveness of the model increased. Such impact was most apparent in statistical indicators and medical utilization of the predicted groups.
In identifying high-cost cases, adding morbidity trajectories might be necessary only for less comprehensive risk adjustment models, and its contributions came from higher c-statistics and increasing medical utilization of predicted groups.
将纵向信息纳入风险调整模型被认为很重要。本研究旨在评估发病轨迹如何影响风险调整模型识别高成本病例的能力。
使用基于索赔的风险调整器，有或没有从台湾国民保险系统3年索赔数据得出的发病轨迹，来预测成为未来高成本使用者的可能性。研究样本是2002年至2005年持续参保的台湾国民健康保险参保者的随机样本（n = 147,892）。采用逻辑回归模型。基于拆分分析的性能指标包括统计指标（c统计量、敏感性和预测阳性值）、模型识别的真实病例比例以及预测病例的医疗利用情况。
随着风险调整模型的全面性增加，模型的性能总体上有所提高。随着模型全面性的增加，添加轨迹对模型性能的影响减小。这种影响在预测组的统计指标和医疗利用方面最为明显。
在识别高成本病例时，仅对于不太全面的风险调整模型可能有必要添加发病轨迹，其贡献来自更高的c统计量和预测组医疗利用的增加。
