Solutions Company Statit Training Home

Can Problems Lurk in Charts Without Control Violations?

Control charts are well-established tools for characterizing process variability. Variability is inherent in every process. These charts provide a way to monitor this variability. Given the available process data, control charts can identify points that are outside the expected limits. Control violations triggered by common rule sets identify data trends that are expected to occur by chance <.3% of the time. Identifying data points that are expected to occur so infrequently is a reasonable technique for triggering an alarm that something may be occurring in the process that is not expected from common cause variation. When a process is completely out-of-control, there may be multiple control violations since the process itself may be unpredictable. Such situations become readily apparent when examining the chart.

The information contained in a control chart describes the process as it existed during the period represented by the chart. This information makes no inferences about how the process should perform. It is possible that a chart appears to be in control, but the process is not performing in a desirable way. This is an insidious situation. If the behavior of the process is not consistent with internal goals or targets, it may be difficult to convey that a problem, in fact, exists since the control chart does not display control violations. It is important to add as much information as possible to a chart in order for the target audience to be able to discern the ramifications of the information being presented to them. If there are internal goals or targets, for example, add this information to the chart. This will give the viewer an opportunity to check the actual process performance against the goal.

In healthcare, there may be a level of performance that is expected or required. For example, if a facility strives to administer a treatment to every eligible patient with a certain condition, we can use a control chart to measure the goal against the actual performance.

Here is an example of a P-chart with sample data for the administration of ACEI or ARB to eligible LVSD patients. Based on this data, there are no control violations.

Java is not enabled in browser, data tips cannot work for this graph.

The calculated mean is 88.6%. Consider the possibility that a healthcare facility has a goal of 95% for this measure. The chart in its current form offers no information pertinent to the facility goal. If people viewing the chart know the goal, they can see that there is a discrepancy between the calculated average and the 95% goal. This method, however, is not a reliable way to convey the point that work needs to be done to reach the goal. If the goal is not common knowledge or if it has changed, the audience has no way to compare the current target to performance.

The next chart adds the target value to the chart.

Java is not enabled in browser, data tips cannot work for this graph.

This technique makes it clear to the viewer that the existing process of identifying and treating eligible patients is not adequate to achieve the facility’s goal for this measure. The target audience can see from the chart that only a few of the values meet or exceed the stated goal.

Once a discrepancy is identified, there may be many issues that need to be investigated in order to effect a change. Here are a few questions that can help to begin assessing a problem:

  • Does the data seem reasonable or is it what wewould expect?
  • Is the data complete—have all the data been collected and submitted?
  • Is there additional information we could get from the data if we grouped it differently or had the ability to drill down in order to see more detail?
  • Are the subgroup sizes consistent with the requirements for the type of data?
  • Should the sigma limits be altered to detect problems within a shorter time span?

Keep in mind that it is possible for problems to lurk under the surface of a control chart in the absence of statistically defined control violations. It is important to be aware of the data content and compare that to what is expected behavior from the process performance.

If you would like additional information, please send email to