Most hospital quality dashboards are designed to answer one question: Are we OK? Green means yes. Red means no. The trouble is, once the answer is green, nobody asks the more useful question: what did we do, and is it working?
I watched this happen recently at an academic medical center. On screen was a dashboard showing monthly counts for five targeted inpatient harms — C. diff, CAUTI, CLABSI, falls with injury, and HAPI. A bar chart with a linear trend line sloping gently downward. A banner in the corner read “24% Reduction.” The dashboard was green. The room seemed satisfied.
Later that day, I pulled the same data into a Process Behavior Chart. And a completely different story showed up — one the bar chart had been actively suppressing.
The Chart Was Hiding a Step Change
The linear trend line on the original bar chart slopes gently across 20 months, implying the improvement was gradual and continuous. Like someone slowly turning a dimmer switch. And because a line has to go somewhere, it also implies the improvement will automagically continue — just keep doing whatever you're doing (whatever that is) and the number will keep drifting down on its own. That's a comforting story. It's also not what's happening.
The data shows something different. The first 10 months cluster around 13-16 harms per month. Then something shifts around late in the fiscal year. The subsequent months drop to 7-10.
That's not a dimmer switch. It's a light switch. Someone did something specific, at a specific point in time, and it worked. The process moved to a new level and stayed there.
On the Process Behavior Chart, that pattern almost triggers a textbook signal — a run of eight or more consecutive points on one side of the average, which is one of the standard rules for detecting that a process has shifted. Without one unusually high month (a count of 21 that I'll come back to later), that's exactly what you'd see. The signal would be statistically unambiguous. With that spike interrupting the run, the chart is still pointing in the same direction — just less emphatically. The shift is there, hiding in plain sight.
Think about what the trend line costs you when it smooths over that step change. A step change is a story worth telling the board: here's what we did, here's when we did it, here's proof it held. A gradual decline is a shrug. The trend line was taking the first story and converting it into the second.
I see this in hospitals all the time. An improvement team does real work, gets real results, and then the chart makes it look like things just sort of… drifted in the right direction… and will continue to do so. Nobody gets to learn what actually worked. The improvement becomes anonymous, and anonymous improvements are hard to sustain and impossible to spread.
Two Averages Don't Tell You Much
The “24% reduction” was a comparison of two averages — a before number and an after number. That's it.
It's the equivalent of weighing yourself in January and again in June and saying “I lost weight.” Maybe. But when? Was it steady? Was it a crash in March that you've already reversed? Are you still losing, or gaining it back? The scale can't tell you. And neither can two averages.
With 20 months of data sitting right there, this hospital had the raw material to answer every one of those questions. Instead, they compressed it into a single percentage and a green checkmark.
I wrote about this in my book Measures of Success — cherry-picking two data points (or two averages) to show a before-and-after comparison can be misleading, even when the underlying improvement is real. In this case, the improvement was real. The metric just couldn't show how or when.
The Green Checkmark Is the Most Expensive Thing on That Dashboard
Here's the part that should genuinely concern the leaders in that room.
The target was 12.3 average monthly harms. Current performance was around 7. Dashboard: green. And the moment it turned green, the conversation stopped.
But think about what “green” actually means here. It means “above the floor.” That's it. It doesn't mean stable. It doesn't mean improving. It doesn't mean safe from regression. If this hospital's process crept back up to 13 next month — nearly double their current level — the target-based dashboard would still show green. Still “at target.” The dashboard would be smiling and nodding while the gains evaporated.
A Process Behavior Chart catches that immediately. A value of 13, plotted against limits calculated from the recent post-shift data, would show up as a signal. Something changed. Go investigate. Go learn.
I've written a lot about why red/green target-based dashboards are problematic, and this is the core issue. The target tells you whether to celebrate. The chart tells you whether to investigate. Only one of those protects your gains. And it's not the one most dashboards are built around.
That Spike Is Trying to Tell You Something
There was one month in the dataset with a count of 21 — far above everything else. On the bar chart, it was just a tall bar. Easy to glance at and move past.
On the XmR chart, that same data point nearly breaks through the Upper Natural Process Limit and produces the two largest moving ranges in the entire dataset. The chart is practically waving its arms. This month was different in kind, not degree.
Even if we assume there was a shift and that data point is an anomaly, the data point of 21 is a “near signal” (near the upper limit). That's worth investigating and understanding.
My hypothesis is that something specific happened at a single point in time that caused a downward shift in harms. But what was it??
Was it a reporting catch-up? A data correction? A genuinely terrible month? If so, why?? The answer matters, because it changes the baseline. If it's an artifact, removing it from the calculation tightens the limits and makes the improvement even more statistically clear. If it's real, it tells you something about where this process is still vulnerable.
On a bar chart, a spike is a visual curiosity. On a Process Behavior Chart, it's a hypothesis that demands testing.
Let's say that month had 12 harms… without the outlier, the chart would look like this and the shift would be more apparent:
What the Data Is Asking For
If your hospital or health system is tracking quality metrics on bar charts with trend lines — or on red/green dashboards with targets as the only reference point — the data is doing more work than your chart can show.
Plot the individual data points over time with a calculated center line and Natural Process Limits. Not targets. Not arbitrary goals. Limits derived from the voice of the process itself. As Don Wheeler puts it, let the data speak before you decide what it should say.
When you see a signal — a point outside the limits, a run of eight or more on one side of the mean — that's the process telling you something meaningfully changed. Go find out what. Everything else is noise, and reacting to noise is one of the most common and most expensive habits in management.
And when you have evidence that the process has genuinely shifted — as this hospital's data strongly suggests — split the chart and recalculate limits on the new baseline. Now you have a tool that will detect the next change, whether it's further improvement or early signs of regression. You've turned a rearview mirror into a windshield.
The PBC shows the average dropped from 14.1 to 10.5 (even with the spike to 21). That's a 25.5% reduction.
The Real Loss
This hospital's team reduced inpatient harms in what looks like a real, sustained way. That kind of result — a discrete step change to a lower level — is exactly what leadership should want to see, understand, protect, and replicate elsewhere.
But their dashboard compressed it into a percentage and a color. It took a story about specific people doing specific things at a specific time and turned it into “things are generally fine.” And in doing so, it didn't just undersell the result. It eliminated the organizational learning that should have come with it.
That's the real cost of a chart that doesn't distinguish signal from noise. Not that it's statistically imprecise — although it is. The cost is that the organization can't understand its own improvement, which means it can't protect what it gained or repeat it somewhere else.
If your dashboards are making you feel reassured but not any smarter, they're the most expensive screensaver in your building.
