What Is a Control Chart?

Progress [in the control of quality] has been slow. … it requires the application of statistical methods, which up to the present time have been for the most part left undisturbed in the journals in which they appeared.

—WALTER SHEWHART (1931)

The first step in improvement is evaluating performance about defects, mistakes, errors, and turnaround times. It doesn’t matter if it’s medication errors or patient falls, defects, waste, and rework are consuming a third of most hospital budgets. You can use control charts to measure and monitor the performance of any process—medication errors or door-to-balloon times. Without control charts, you cannot sustain improvements. This is a key reason why so many hospital improvement projects fall back to previous levels of performance. See the video at www.qimacros.com/Moneybelt/what-is-a-control-chart.html.

A control chart (Figure 7.4) is made up of a few key parts:

Control charts originated at Western Electric in the 1920s. I recently got hold of a copy of the Western Electric Statistical Quality Control Handbook (1956). I love that the Preface begins by saying: “This book is written in non-technical language, and no attempt has been made to write for the professional statistician or mathematician.” The Western Electric program focused on “the control chart, and particularly the process capability study, as the foundation of the entire program.”

The authors go on to say that “Statistical Quality Control can be applied to anything that it is possible to express in the form of numbers.” Yep, anything—manufacturing or healthcare.

What Is Statistical Quality Control?

Western Electric summed it up quite nicely:

The Power of Control Charts

So that’s it in a nutshell: control charts help us study performance in ways that make it behave the way we want it to behave. If your hospital process isn’t behaving the way you want it to, maybe it’s because you haven’t subjected it to the scrutiny of a control chart. All processes vary: “Instead of being exactly the same from point to point or from time to time, the numbers vary.”

Using Control Charts to Detect Faint Signals in the Noise of Daily Operations

One of the characteristics of a high-reliability organization (HRO) is preoccupation with failure—detecting small failures before they become big. In the book, The Signal and the Noise, author Nate Silver examines why so many predictions fail, but some don’t. The trick is in separating the signal (what’s important) from the noise (what’s not important).

I realized while reading Silver’s book that control charts detect faint signals in the noise of daily operations. Unlike Silver’s book, which focuses on predicting what will happen, control charts look backward at the performance of any hospital process to detect “out-of-control” conditions. I have come to believe that without control charts, you cannot monitor and correct a hospital’s performance or sustain improvements. And failing to sustain improvements will kill Six Sigma.

The Costs of Confusing Signals with Noise

One of the by-products of trying to run a hospital by spreadsheets alone is the ongoing tampering and knee-jerk reactions to changes. This occurs because managers look at snapshots of performance (usually bar charts of monthly performance), not the ongoing movie of performance (control charts). One of my friends in corporate finance has found that almost weekly she’d be sent on a wild goose chase to figure out why some number moved up or down. In retrospect, she has come to believe that most of those movements were noise, not signals. If she’d had control charts of the financials, she could have easily separated the signals from the noise.

Signal or Noise

If we look at a spreadsheet of denied claims in healthcare (Figure 7.5), it’s difficult to identify the signals. Cell B5 looks high, but is it?

If we take performance data and chart it as a line graph (Figure 7.6), we still can’t really tell if there’s something going on. We can see a spike in May, but is it signal or noise?

But if we turn the same data into a control chart (Figure 7.7), the QI Macros will calculate upper and lower control limits (UCL/LCL) that encompass 99.7 percent of all the data. We might consider most of the points within these limits to be noise (i.e., common-cause variation). Using these limits, the software highlights points that are statistically unlikely, that is, signals or special-cause variation. We can see from the diamond-shaped points that the process is out of control.

So the control chart will identify the two points in May as a signal that the process is out of control, but it will also detect that the 11 points below the average are also a signal. Using the first signal, we would seek to determine the root cause of the problem and correct it. Using the second signal, we might want to figure out what we were doing to hold down the cost of denied insurance claims. Sometimes these signals can catch your doing something right, not wrong.

Control Chart Signals

The QI Macros statistical process control (SPC) software uses calculated control limits and control chart rules to separate signals from noise. Each of the rules should occur naturally only three times out of a thousand (3:1,000). So when they appear in 20 to 50 data points, it’s very likely that they are a signal, not noise.

Once the QI Macros software detects one of these signals and highlights it by turning the point red, the next step is to immediately ask “Why?” five times to determine the root cause and correct it.

Common-Cause Variation is a Form of Noise

A stable process has no signals. Variation from point to point is just the noise of daily operations. Of course, if we wanted to reduce days to pay insurance claims to a target of 30 days, we’d need an improvement team to investigate why it takes 33.3 days on average.

Overreaction to Noise Can Make Things Worse

Using a snapshot of the data can cause managers to tamper with a stable process. Tampering with a stable process can make it unstable. If, for example, a manager reacts to point 26 in “Days to Pay Insurance Claims,” it could cause overtime to clear the backlog or some other knee-jerk reaction. This might reduce days to pay for a short while but also might cause employee burnout and increased days to pay.

If a process is stable, don’t fiddle with it. Use Lean Six Sigma to reduce the cycle time and variation. Companies that don’t use control charts consistently perform around 3 sigma (a 6 percent error rate). At this level, it costs a quarter to a third of total expenses to find, fix, or replace the defective product or service. It causes overtime and crisis management. Wouldn’t you like to cut costs by 25 to 33 percent? You can, but as you make improvements and climb to 4 or 5 sigma, you will no longer be able to detect the signals that your process is out of control. Without control charts, you will not be able to detect the subtle changes that slowly drive performance back toward 3 sigma.

Using control charts in every aspect of your business will only take a little more effort than you are spending today, but it will save a ton of rework and waste, freeing up employees to satisfy customers without the firefighting.

Learn how to use control charts to detect the signals that your process is shifting. Use them to determine whether improvement is needed and initiate improvements as needed. Use them to monitor and sustain new and improved levels of performance. Your patients will be delighted. Employee morale will improve with the improved performance. Your CFO will be delighted with the reduced costs and increased profits. And everyone from the CEO to the nurse manager will get a better night’s sleep.

Isn’t it time to add control charts to your competitive toolbox?