Hofer T P, Hayward R A
Department of Internal Medicine, University of Michigan, Ann Arbor.
Med Care. 1995 Mar;33(3):234-45. doi: 10.1097/00005650-199503000-00003.
There is widespread interest in using external quality measures, such as early re-admission rates (ERRs), to evaluate hospital quality. To evaluate the feasibility of using ERRs to identify poor-quality hospitals, a Monte Carlo simulation model was developed that describes the predictive power of ERRs for the 190 hospitals in Michigan using different assumptions concerning the distribution and variability of quality problems, the number of years of data aggregated, and unmeasured case-mix differences. The ability of ERRs to distinguish 171 average-quality hospitals from 19 poor-quality hospitals (assigned to have 5% vs. 15% premature discharges) was evaluated. First, the largest diagnosis-related groups (DRGs) were studied to determine if they included cardiac, gastrointestinal, pulmonary, and neurologic diseases. Despite using the highly optimistic assumptions that premature discharges are readmitted 50% more frequently than appropriately timed discharges and that no ERR variation was caused by unmeasured case-mix differences between hospitals, the results were poor. For example, for DRG 127 (heart failure), high ERR outlier status (using a .05 probability cutoff) had a positive predictive value of only 36%, meaning that approximately two thirds of hospitals labeled "poor-quality" (high ERR outliers) were false-positive results. Next, we repeated the simulation with sample sizes aggregated for all medical DRGs. The positive predictive value was 72%, but was very sensitive to ERR variability due to non-quality-related factors (e.g., unmeasured case mix). The positive predictive value decreases to 45% if unmeasured case mix accounts for even 10% of observed hospital ERR variation. The circumstances under which DRG-specific ERRs would be useful to detect poor-quality hospitals are unlikely to occur. Even collapsing to all medical DRGs, ERRs are likely to be accurate predictors only if quality differences are quite large and if unmeasured case-mix differences account for a small amount of interhospital variation in ERRs.
利用外部质量指标（如早期再入院率）来评估医院质量已引起广泛关注。为评估使用早期再入院率识别低质量医院的可行性，开发了一个蒙特卡洛模拟模型，该模型使用了关于质量问题的分布和变异性、汇总数据的年份以及未测量的病例组合差异的不同假设，来描述密歇根州190家医院早期再入院率的预测能力。评估了早期再入院率区分171家平均质量医院和19家低质量医院（分别设定有5%和15%的过早出院率）的能力。首先，研究了最大的诊断相关组，以确定它们是否包括心脏、胃肠、肺部和神经疾病。尽管使用了高度乐观的假设，即过早出院的再入院频率比适时出院高50%，且医院间未测量的病例组合差异不会导致早期再入院率的变化，但结果并不理想。例如，对于诊断相关组127（心力衰竭），高早期再入院率异常值状态（使用0.05的概率临界值）的阳性预测值仅为36%，这意味着大约三分之二被标记为“低质量”（高早期再入院率异常值）的医院是假阳性结果。接下来，我们对所有医疗诊断相关组汇总的样本量重复了模拟。阳性预测值为72%，但对非质量相关因素（如未测量的病例组合）导致的早期再入院率变异性非常敏感。如果未测量的病例组合即使占观察到的医院早期再入院率变化的10%，阳性预测值也会降至45%。特定诊断相关组的早期再入院率有助于检测低质量医院的情况不太可能出现。即使汇总到所有医疗诊断相关组，只有在质量差异相当大且未测量的病例组合差异占医院间早期再入院率变化的比例较小时，早期再入院率才可能是准确的预测指标。
