Blumberg M S
Health Serv Res. 1987 Feb;21(6):715-39.
In March 1986, the Health Care Financing Administration (HCFA) released ten lists of death-rate "outlier" hospitals, one for all 1984 Medicare discharges and nine for specific DRGs. Recent Medicare hospital discharge abstracts have substantially undercounted in-hospital deaths, with large variations by state. Apart from the proportion of a hospital's cases in 80 DRGs, the predictive models had no measures of case severity based on diagnosis or procedure. Having DRG 123 (all deaths from acute myocardial infarction) as an independent variable in the all-death regression model probably accounted for much of its high r2. Inclusion of an independent variable for average length of stay (ALOS) favored hospitals in higher ALOS states by higher predicted death rates. Model bias also favored lower-risk hospitals. Small numbers of predicted deaths for specific DRGs limited low-volume hospitals on these outlier lists to those with high ratios of actual to predicted deaths. On six of the nine DRG-specific outlier lists, a total 1,222 hospitals had unfavorable residuals, while only 8 were favorable. Ten recommendations are given to increase reliability of future outcome analyses.
1986年3月，医疗保健财务管理局（HCFA）公布了十份死亡率“异常”医院名单，一份涵盖1984年所有医疗保险出院病例，另外九份针对特定的诊断相关分组（DRG）。近期医疗保险医院出院摘要严重低估了住院死亡人数，且各州差异很大。除了医院80个诊断相关分组病例的比例外，预测模型没有基于诊断或手术的病例严重程度衡量指标。在全死亡回归模型中将诊断相关分组123（急性心肌梗死导致的所有死亡）作为自变量，可能是其高决定系数（r2）的主要原因。纳入平均住院时长（ALOS）自变量后，住院时长较长州的医院预测死亡率较高。模型偏差也有利于低风险医院。特定诊断相关分组的预测死亡人数较少，使得这些异常名单上的低流量医院仅限于实际死亡与预测死亡比例高的医院。在九份特定诊断相关分组异常名单中的六份上，总计1222家医院有不利残差，而有利残差的医院只有8家。文中给出了十条建议，以提高未来结果分析的可靠性。
