Department of Epidemiology & Preventive Medicine, Centre for Research Excellence in Patient Safety, School of Public Health, Preventive Medicine Monash University, Melbourne, Australia.
BMC Health Serv Res. 2012 Aug 28;12:278. doi: 10.1186/1472-6963-12-278.
The need to improve patient safety has been identified as a major priority for health reform in developed countries, including Australia. We investigated the implementation and appropriateness of Variable Life Adjusted Displays as a quality control procedure to monitor "in-control" versus "out-of-control" processes in Victorian public hospitals.
Victorian Admitted Episode Data from Department of Human Services, Victoria for 2004-7 were used. The VLAD is a plot of a cumulative sum of the difference in expected outcome (range 0-1) and observed outcome (0 or 1) for sequential separations. Three outcomes were assessed: in-hospital mortality for acute myocardial infarction, stroke and heart failure. Logistic regression was used to obtain a realistic measure of expected mortality over the period 2004-5, adjusting for covariates and comorbidities, to estimate expected mortality risk for the separations between 2005-7. VLAD were plotted for the years 2005-7, by the 11 hospitals with the highest frequency of separations. Signalling limits for 30%, 50% and 75% risk decrease and risk increase were determined and plotted for each VLAD utilizing risk-adjusted cumulative sum techniques. This is a likelihood-ratio test statistic for signalling. If the VLAD signalled by intersecting with a limit, the limit was reset.
The three logit models displayed reasonable fit to the observed data. There were n = 2999 separations in the AMI model, n = 3598 in the HF model and n = 1922 in the stroke model. The number of separations plotted by VLAD ranged from n = 126 to n = 648. No signals were observed in 64%, 55% and 18% of VLAD for AMI, HF and stroke respectively. For AMI and HF 9% of hospitals signalled at least once for each of 30%, 50% and 75% risk increase, whereas this was 45% for stroke. Hospitals signalling at least once for risk decrease ranged from 18% to 36% across the levels of risk and outcomes. No VLAD signalled for both risk decrease and increase.
VLAD intersecting with limits to signal "out-of-control" states, may be an appropriate technique to help hospitals assess quality control. Preliminary work displays some between hospital differences. Relevant signals can be used to investigate why a system is potentially performing better than or worse than expected. Types and levels of investigation could depend on the type of signalling. Validation work, for example attempting to correlate signals with clinical notes, prior to VLAD distribution needs to be undertaken.
提高患者安全已被确定为发达国家（包括澳大利亚）医疗改革的主要重点。我们调查了可变生命调整显示（Variable Life Adjusted Displays，VLAD）作为质量控制程序的实施情况和适当性，以监测维多利亚公立医院中“在控”与“失控”过程。
使用维多利亚州人类服务部（Department of Human Services，Victoria） 2004-2007 年的住院病例数据。VLAD 是一个累积和图，用于显示预期结果（范围 0-1）与观察结果（0 或 1）之间差异的序贯分离。评估了三种结果：急性心肌梗死、中风和心力衰竭的院内死亡率。使用逻辑回归获得 2004-2005 年期间实际预期死亡率的估计值，调整了协变量和合并症，以估计 2005-2007 年期间分离的预期死亡率风险。根据分离次数最高的 11 家医院，绘制了 2005-2007 年的 VLAD 图。利用风险调整的累积和技术，为每个 VLAD 确定并绘制了 30%、50%和 75%风险降低和风险增加的信号限制。这是用于信号检测的似然比检验统计量。如果 VLAD 与限制相交，则重置限制。
三个逻辑回归模型对观察数据的拟合程度良好。AMI 模型的分离数为 n=2999，HF 模型为 n=3598，中风模型为 n=1922。VLAD 绘制的分离数范围为 n=126 至 n=648。AMI、HF 和中风的 VLAD 分别有 64%、55%和 18%没有观察到信号。对于 AMI 和 HF，9%的医院在 30%、50%和 75%的风险增加水平上至少发出过一次信号，而中风的这一比例为 45%。风险降低的医院信号至少一次的比例在风险和结果的各个水平上从 18%到 36%不等。没有 VLAD 同时发出风险降低和增加的信号。
VLAD 与限制相交以发出“失控”状态的信号，可能是一种帮助医院评估质量控制的合适技术。初步工作显示了一些医院之间的差异。相关信号可用于调查系统为何潜在地表现优于或劣于预期。调查的类型和级别可能取决于信号类型。在 VLAD 分发之前，需要进行验证工作，例如尝试将信号与临床记录相关联。
