Mnatzaganian G, Ryan P, Hiller J E
Dr George Mnatzaganian, Faculty of Health Sciences, Australian Catholic University, Room 8.70, Level 8, 250 Victoria Parade, East Melbourne, Victoria, Australia, E-mail:
Methods Inf Med. 2014;53(2):115-20. doi: 10.3414/ME13-01-0095. Epub 2014 Feb 11.
Using three risk-adjustment methods we evaluated whether co-morbidity derived from electronic hospital patient data provided significant improvement on age adjustment when predicting major outcomes following an elective total joint replacement (TJR) due to osteoarthritis.
Longitudinal data from 819 elderly men who had had a TJR were integrated with hospital morbidity data (HMD) and mortality records. For each participant, any morbidity or health-related outcome was retrieved from the linked data in the period 1970 through to 2007 and this enabled us to better account for patient co-morbidities. Co-morbidities recorded in the HMD in all admissions preceding the index TJR admission were used to construct three risk-adjustment methods, namely Charlson co-morbidity index (CCI), Elixhauser's adjustment method, and number of co-morbidities. Postoperative outcomes evaluated included length of hospital stay, 90-day readmission, and 1-year and 2-year mortality. These were modelled using Cox proportional hazards regression as a function of age for the baseline models, and as a function of age and each of the risk-adjustment methods. The difference in the statistical performance between the models that included age alone and those that also included the co-morbidity adjustment method was assessed by measuring the difference in the Harrell's C estimates between pairs of models applied to the same patient data using Bootstrap analysis with 1000 replications.
Number of co-morbidities did not provide any significant improvement in model discrimination when added to baseline models observed in all outcomes. CCI significantly improved model discrimination when predicting post-operative mortality but not when length of stay or readmission was modelled. For every one point increase in CCI, postoperative 1- and 2-year mortality increased by 37% and 30%, respectively. Elixhauser's method outperformed the other two providing significant improvement on age adjustment in all outcomes.
The predictive performance of co-morbidity derived from electronic hospital data is outcome and risk-adjustment method specific.
我们使用三种风险调整方法，评估了源自电子医院患者数据的合并症在预测因骨关节炎而行择期全关节置换术（TJR）后的主要结局时，对年龄调整是否有显著改善。
将819例接受TJR的老年男性的纵向数据与医院发病率数据（HMD）和死亡率记录相结合。对于每位参与者，从1970年至2007年期间的关联数据中检索任何发病率或与健康相关的结局，这使我们能够更好地考虑患者的合并症。在首次TJR入院前所有入院记录中HMD记录的合并症用于构建三种风险调整方法，即查尔森合并症指数（CCI）、埃利克斯豪泽调整方法和合并症数量。评估的术后结局包括住院时间、90天再入院率以及1年和2年死亡率。这些结局以Cox比例风险回归模型进行建模，在基线模型中作为年龄的函数，在其他模型中作为年龄以及每种风险调整方法的函数。通过使用1000次重复的自抽样分析测量应用于相同患者数据的成对模型之间的哈雷尔C估计值的差异，评估仅包含年龄的模型与同时包含合并症调整方法的模型之间统计性能的差异。
在所有结局中，将合并症数量添加到基线模型时，模型辨别能力没有显著改善。在预测术后死亡率时，CCI显著改善了模型辨别能力，但在对住院时间或再入院率进行建模时则不然。CCI每增加1分，术后1年和2年死亡率分别增加37%和30%。埃利克斯豪泽方法优于其他两种方法，在所有结局的年龄调整方面均有显著改善。
源自电子医院数据的合并症的预测性能因结局和风险调整方法而异。
