Gan Fah Fatt, Tang Xu, Zhu Yexin, Lim Puay Weng
Department of Statistics and Applied Probability, National University of Singapore, 6 Science Drive 2, 117546, Republic of Singapore.
Int J Qual Health Care. 2017 Jun 1;29(3):427-432. doi: 10.1093/intqhc/mzx033.
The traditional variable life-adjusted display (VLAD) is a graphical display of the difference between expected and actual cumulative deaths. The VLAD assumes binary outcomes: death within 30 days of an operation or survival beyond 30 days. Full recovery and bedridden for life, for example, are considered the same outcome. This binary classification results in a great loss of information.
Although there are many grades of survival, the binary outcomes are commonly used to classify surgical outcomes. Consequently, quality monitoring procedures are developed based on binary outcomes. With a more refined set of outcomes, the sensitivities of these procedures can be expected to improve.
A likelihood ratio method is used to define a penalty-reward scoring system based on three or more surgical outcomes for the new VLAD. The likelihood ratio statistic W is based on testing the odds ratio of cumulative probabilities of recovery R. Two methods of implementing the new VLAD are proposed.
We accumulate the statistic W-W¯R to estimate the performance of a surgeon where W¯R is the average of the W's of a historical data set. The accumulated sum will be zero based on the historical data set. This ensures that if a new VLAD is plotted for a future surgeon of performance similar to this average performance, the plot will exhibit a horizontal trend.
For illustration of the new VLAD, we consider 3-outcome surgical results: death within 30 days, partial and full recoveries. In our first illustration, we show the effect of partial recoveries on surgical results of a surgeon. In our second and third illustrations, the surgical results of two surgeons are compared using both the traditional VLAD based on binary-outcome data and the new VLAD based on 3-outcome data. A reversal in relative performance of surgeons is observed when the new VLAD is used. In our final illustration, we display the surgical results of four surgeons using the new VLAD based completely on 3-outcome data.
Full recovery and bedridden for life are two completely different outcomes. There is a great loss of information when different grades of 'successful' operations are naively classified as survival. When surgical outcomes are classified more accurately into more than two categories, the resulting new VLAD will reveal more accurately and fairly the surgical results.
传统的可变寿命调整显示（VLAD）是预期累积死亡数与实际累积死亡数之差的图形化展示。VLAD假定为二元结果：手术30天内死亡或存活超过30天。例如，完全康复和终身卧床被视为相同结果。这种二元分类导致大量信息丢失。
尽管存在多种存活等级，但二元结果常用于对外科手术结果进行分类。因此，质量监测程序是基于二元结果制定的。若采用一组更精细的结果，这些程序的敏感性有望提高。
使用似然比方法为新的VLAD定义基于三个或更多手术结果的奖惩评分系统。似然比统计量W基于对恢复累积概率R的优势比进行检验。提出了两种实施新VLAD的方法。
我们累积统计量W - W¯R来评估外科医生的表现，其中W¯R是历史数据集W的平均值。基于历史数据集，累积和将为零。这确保了如果为未来表现与该平均表现相似的外科医生绘制新的VLAD，该图将呈现水平趋势。
为说明新的VLAD，我们考虑三种结果的手术结果：30天内死亡、部分恢复和完全恢复。在第一个示例中，我们展示了部分恢复对外科医生手术结果的影响。在第二个和第三个示例中，使用基于二元结果数据的传统VLAD和基于三种结果数据的新VLAD对两位外科医生的手术结果进行比较。使用新的VLAD时观察到外科医生相对表现的逆转。在最后一个示例中，我们完全基于三种结果数据使用新的VLAD展示四位外科医生的手术结果。
完全康复和终身卧床是两个完全不同的结果。当不同等级的“成功”手术被简单地归类为存活时，会有大量信息丢失。当手术结果被更准确地分类为两类以上时，由此产生的新VLAD将更准确、公平地揭示手术结果。
