A laboratory analyser is checked against manufactured control samples with known values, twice a day, and the result is plotted. Ours sits dead centre on every channel except potassium at three in the morning, when it goes half a millimole high and comes back.
I am a biomedical scientist in a district general laboratory. Nineteen years, mostly nights.
How quality control works, because the plot is the story.
You buy control material with assayed values and a known spread. You run it, you plot the result, and over time you get a chart with a mean and standard deviations marked on it.
Then you apply rules to that chart, which have names and which every laboratory in the country uses. One result outside three standard deviations is a reject. Two in a row outside two on the same side is a reject. Four in a row outside one on the same side is a reject, and so is ten in a row on the same side of the mean even if every one of them is well inside the limits, because a machine that is always a little bit high is a machine that has drifted.
If a control fails you stop reporting on that channel, you investigate, and you do not release a patient result until it passes.
Our main chemistry analyser has been in since 2019 and it is a good machine.
Potassium at three in the morning runs half a millimole high.
Potassium matters more than most of what we measure. The reporting range is narrow, the numbers that trigger a phone call to a ward are close together, and half a millimole is the difference between a result somebody acts on and one they do not.
At every other run of the day it is dead on the mean.
The three o'clock control is between two and three standard deviations high, most nights, and about one night in six it goes past three and rejects.
The obvious answer is the commonest error in this trade and I have to deal with it first.
Potassium leaks out of red cells. A sample that is old, or shaken, or taken through a small needle, comes back high, and a haemolysed sample is flagged by the machine and rejected by me.
That is about patient samples. Control material is not blood. It is a manufactured liquid in a vial with no cells in it at all, and there is nothing in it that can leak.
What has been checked, and the manufacturer has been in four times.
The reagent, the electrode, the calibrator, the sample probe and the tubing. All replaced, some of them twice.
The control material itself. Different lot, different manufacturer, and a lot kept in a separate fridge on a different circuit.
The fridge and the room. Temperature logged continuously, both within specification all night, and I have put a data logger on the analyser's own reagent compartment for a fortnight.
The timing. This is the one I was sure about for a year: our three o'clock run happens while the overnight maintenance wash is finishing, and a wash residue would do exactly this. So we moved the run. Two in the morning, four in the morning, and for one month a random time generated by a spreadsheet.
It is not the run. It is the hour. Whatever time we put the control on, the results that come off that machine between about ten to three and twenty past are high on potassium.
Which means the patient samples are too.
I have to be plain about that because it is the part that matters and it is the part I have acted on.
Anything run in that window reads about half a millimole high, and half a millimole high on potassium can turn a normal result into one that gets a doctor out of bed, and can turn a slightly low one into a normal.
Nobody has been harmed. I know that because we chased it: every result in that window for six months was re-run at four, and the repeat is the truth, and the repeats came back lower and the wards were rung and corrected where it mattered. Eleven results in six months were changed. None of them had been acted on yet, because at three in the morning a result sits in a queue for twenty minutes before anybody reads it, and that twenty minutes is the only reason this story has no harm in it.
What we do now.
Nothing is reported off that analyser between ten to three and twenty past. The samples wait, or they go on the back-up analyser in the next room, which does not do it and never has.
That is written into our standard operating procedure as a scheduled maintenance window, which is not what it is, and which is the only wording the quality manager and I could agree on that does not require a written explanation neither of us can give.
The external quality assessment scheme sends us blind samples four times a year and we score well. Our accreditation is intact. The chart is on the wall with a gap in it every night.
What I do.
I run the control, I plot it, I do the back-up run, and I do not report in the window.
The one thing that is mine is that I stopped putting the three o'clock control on by hand.
You load a control the same way you load a sample: you carry the vial to the machine, you put it in a rack, and you press a button, and it takes about ninety seconds and you stand at the analyser while it draws.
I programmed the autoloader to do it in 2022 so I can be at the bench in the other room, and my colleague thinks it is because I am efficient.
It is because on a night in March 2022 I was standing at the analyser at three with the rack in my hand, and the machine was between samples, and the sample probe went down into an empty position on the rack, and drew, and reported a result at 4.9, which is the value of a sample, out of an empty tube, in a room with nobody in it but me.