I’m fat. I eat too much, I drink too much, I move too little, and my blood pressure is too high – well, it used to be. Some years ago I started myself on a daily dose of 20 mg enalapril. That worked well – at least on the blood pressure (BP).

Watchfull vigilance

According to the control chart in figure 1, which displays daily readings of my morning BP during a couple of weeks back in 2012, my BP is stable around 125 mm Hg and varies between 115 and 136.

Figure 1

Figure 1

Tinkered tampering

Now, let’s do a thought experiment. Imagine an engaged doctor who wanted to achieve even better control of my blood pressure squeezing it tightly around the mean of 125 and never reaching above 130.

To obtain that, one might be tempted to apply an “adaptive” dosing scheme that adjusts the enalapril dose to the daily BP readings so that a dose of 40 mg is given on days with BP above 125 and a dose of 10 mg is given when BP is below 125.

Sounds like a good idea? It isn’t! Do not try this at home.

Figure 2 shows what might happen. The red curve is from a computer simulation of random numbers having the same standard deviation as my BP readings. The mean, though, is continuously adjusted by an amount similar to the difference between the previous simulated BP value and the original mean of 125.

Figure 2

Figure 2

How can it be that the variation seems to explode in reaction to our efforts to achieve the exact opposite? We are only trying to mimic nature’s own physiological feed back mechanisms. Obviously, we are not doing a very good job.

Two types of variation

The answer to the question above is tampering.

When a system is stable, telling the worker about mistakes is only tampering.

– W Edwards Deming

Walther A. Shewhart and W. Edwards Deming taught us that there are two types of variation, common cause variation and special cause variation. Mistaking one for the other leads to problems.

Common cause variation

Special cause variation

Figure 1 is an example of a process showing common cause variation – stable and predictable. Leaving the process alone, we expect it to continue to produce BPs around 125 and always between the control limits 115-136. But if we start to react differently to individual random values based on whether they are above or below a certain target value, the process becomes unpredictable and the variation explodes.

Deming demonstrated common and special cause variation in his famous funnel experiment. I find it more convenient to have my computer do the simulation experiment that produces figure 2. That way I don’t have to move so much ☺

Tampering as management strategy

Figure 3 is part of a table that shows the weekly proportion of cancer patients who had acceptable waiting times to start treatment at a large teaching hospital. The number of interest is in the third column (71%, 68%, etc.). The coloured dot is green if at least 75% of the patients that week had acceptable waiting times. It is red if this number is 60% or less and yellow if the number is between 60% and 75%.

Each week management looks at the table and makes decisions based on the colour of the week. That is tampering.

Figure 3: red-yellow-green

Figure 3: red-yellow-green

Let’s have another look at data. The control chart in figure 4 is based on data from the red-yellow-green table but allows us to tell common cause variation from special cause variation.

Figure 4

Figure 4

It is clear that special cause variation is present in the process. Week number 32 represents something special that seems to begin in week 30 and end in week 34. These weeks are in the middle of the Danish summer vacation period. The reason for the drop in adherence to waiting times is not, what one might think, low staffing or reduced OR activity. The main reason for the drop is probably that some patients want their treatment start postponed a few days so they can spend vacation time with their family.

See what happens when we isolate weeks 30-34. Now we have three distinct periods each showing only common cause variation. That means that weeks 30-34 compared to the rest of the year represent something special and, most importantly, that the rest of the year represents a stable and predictable process.

Figure 5

Figure 5

If we want better results, we should study the common causes that produce the results rather than make different decisions based on results that are randomly above or below a certain target. We would need to develop different strategies for the vacation period and the rest of the year. None of these strategies should involve tampering.