An empirical evaluation of cusum control charts for use in robust quality control of surgical and binary outcomes

It is well-known that standard statistical process control tools (eg, Shewhart charts) are not robust to certain features of human-generated data typically seen in health care management. For example, the presence of positive serial correlation (the tendency for successive outcomes to cluster as opposed to being truly random) leads to increased "false alarms." Previous work has introduced potential work-arounds in the case of continuous data (eg, data that can take on many values). In this article we describe a different but related problem in the case of binary data (eg, "survived" vs "deceased"). We demonstrate the value of using the Cumulative Sum chart, which is shown to be relatively robust to serial correlation, and much more efficient and effective than existing control charts.