Rosenman Marc, Madsen Kristine, Hui Siu, Breitfeld Philip P
Regenstrief Institute for Health Care and Department of Medicine, Indiana University School of Medicine, Indianapolis, Indiana, USA.
J Pediatr Hematol Oncol. 2002 May;24(4):263-8. doi: 10.1097/00043426-200205000-00009.
Administrative outcomes such as length of stay and charges are used to compare the quality of care across institutions and among individual providers. Clinical variables representing disease severity may explain some of the variability in these outcomes.
To determine the extent to which readily available clinical data can explain the variability in length of stay and charges for children with cancer admitted to the hospital for fever and neutropenia, and to assess the appropriateness of using a time-efficient electronic case-finding strategy for the development of administrative outcome models.
A retrospective cohort of 157 fever and neutropenia encounters in a single institution during 11 months in 1997 was identified using a largely automated case-finding strategy followed by independent, blinded review of the selected discharge summaries. Models of admission variables predicting log length of stay and log charges were developed using multiple linear regression. The "smearing" technique of Duan adjusted for logarithmic retransformation was used in calculating each subject's predicted length of stay and charges. R2 values were calculated. There were two secondary analyses. In one, the result of admission blood culture was entered as a potential covariate. In the second, to evaluate the appropriateness of basing models on automated case-finding without discharge summary review, the authors rederived the models using all of the encounters (n = 160) identified by the algorithm, which had included three false-positive cases.
Mean length of stay was 6.45 days. Mean charges were $11,967. Absolute monocyte count at admission was a significant, independent negative predictor of length of stay and charges. Underlying cancer diagnosis also was significant. Charges were highest for acute myeloid leukemia, followed by central nervous system tumors, other solid tumors, and acute lymphoblastic leukemia and lymphomas. Length of stay was highest for acute myeloid leukemia, followed by central nervous system tumors, acute lymphoblastic leukemia and lymphomas, and other solid tumors. Absolute monocyte count and tumor type were the major components of the model, but admission temperature (for both administrative outcomes) and the presence of localized infection (for length of stay) also were significant predictors. R2 values were 35.3% (charges) and 38.5% (length of stay), with validation R2 values of 26.6% and 29.2%, respectively. Entering bacteremia as a covariate improved the models. Inclusion of the three false-positive cases generated models with only a modest loss of accuracy; it introduced over-and underreporting of some of the less significant predictors but did not disrupt the ability to identify the major predictors, absolute monocyte count and tumor type.
The clinical variables that were significant in this study account, in validation R2 estimates, for more than 25% of the variability in administrative outcomes for encounters of fever and neutropenia. Adjusting length of stay and charges for these clinical variables would allow for a fairer comparison of institutions and individual providers. The electronic case-finding algorithm served as an efficient way to identify absolute monocyte count and tumor type as the major predictors and provided a conservative estimate of R2.
住院时间和费用等管理结果用于比较不同机构及个体医疗服务提供者之间的医疗质量。代表疾病严重程度的临床变量可能解释这些结果中的部分变异性。
确定现有的临床数据能在多大程度上解释因发热和中性粒细胞减少症入院的癌症患儿住院时间和费用的变异性，并评估使用省时的电子病例发现策略来开发管理结果模型的适用性。
采用主要基于自动化病例发现策略，随后对选定的出院小结进行独立、盲法审核，确定了1997年11个月期间某单一机构157例发热和中性粒细胞减少症病例的回顾性队列。使用多元线性回归建立预测住院时间对数和费用对数的入院变量模型。在计算每个受试者的预测住院时间和费用时，采用了Duan的“涂抹”技术对对数重新转换进行调整。计算R2值。进行了两项次要分析。其一，将入院血培养结果作为潜在协变量纳入。其二，为评估基于自动化病例发现而未进行出院小结审核构建模型的适用性，作者使用算法识别出的所有病例（n = 160）重新构建模型，其中包括3例假阳性病例。
平均住院时间为6.45天。平均费用为11,967美元。入院时的绝对单核细胞计数是住院时间和费用的显著独立负向预测因子。潜在癌症诊断也具有显著性。急性髓系白血病的费用最高，其次是中枢神经系统肿瘤、其他实体瘤、急性淋巴细胞白血病和淋巴瘤。急性髓系白血病的住院时间最长，其次是中枢神经系统肿瘤、急性淋巴细胞白血病和淋巴瘤以及其他实体瘤。绝对单核细胞计数和肿瘤类型是模型的主要组成部分，但入院体温（对于两种管理结果）和局部感染的存在（对于住院时间）也是显著预测因子。R2值分别为35.3%（费用）和38.5%（住院时间），验证R2值分别为26.6%和29.2%。将菌血症作为协变量纳入可改善模型。纳入3例假阳性病例生成的模型准确性仅有适度损失；它导致一些不太显著的预测因子出现高估和低估，但并未破坏识别主要预测因子（绝对单核细胞计数和肿瘤类型）的能力。
在本研究中具有显著性的临床变量，在验证R2估计中，占发热和中性粒细胞减少症病例管理结果变异性的25%以上。针对这些临床变量调整住院时间和费用，将使机构和个体医疗服务提供者之间的比较更加公平。电子病例发现算法是识别绝对单核细胞计数和肿瘤类型作为主要预测因子的有效方法，并提供了保守的R2估计值。
