I saw on the Olympics once when they were doing scoring for team events, that on some events the judges would take each members performance, eliminate the highest and lowest members performances, and average the remaining performances to give an overall score for the team. I don't know if that's changed or not but I wanted to know if it was a valid way of checking accuracy. Let me explain....
A friend was recently admitted to hospital and diagnosed with (extremely) high blood pressure (BP) or hypertensive shock. Readings were constantly in excess of 190/140 at the time and she was hooked up to a machine the whole night. Thankfully it was diagnosed because of a nose bleed that did not stop for over an hour. From what she was told, had the blood vessel been in the brain or heart she would have suffered a stroke or heart attack and possibly even death. Everyone was afraid to even let her walk to the bathroom. So it seemed pretty serious but yet, apart from the weakness caused from the loss of blood with the nosebleed, a slight headache and no chest pain at all, she felt relatively normal.
So I started to do research on BP machines and found out that their accuracy was sometimes, well, questionable. The doctor took a BP machine, calibrated it with readings he took manually and verified it was accurate. Told my friend to take it home and come back in two weeks with loads of readings (at least one per day, preferably 3). The thing is, we took a few readings one after the other, and they were all slightly different, sometimes up to 5.6% from an average (from our research, this is normal as BP is constantly changing anyway). So we also purchased another machine and decided to use the one calibrated by the doctor as the baseline that was most accurate.
Over the following days, we measured the blood pressure at 4 sittings (after waking before breakfast around 6am, before lunch around 1pm, before dinner around 6pm, before sleeping around 10.30pm). In each sitting we took 5 measurements with the doctors machine. Using excel, we eliminated the highest and lowest Systolic, Diastolic, and Heart Rate. Then averaged the other 3. Like in the Olympics above. These measurements varied by up to +/- 5.6% from the average.
We repeated the process with the other machine, taking in all, 10 measurements per sitting. The new machine was within +/- 5.5% of the average of its own readings which were within +/- 2.25% of the doctors machine so we accepted that was accurate also. Is this assumption correct?
We did this because we had to return the doctors machine after 2 weeks and wanted to be sure ours was accurate too. Given there was less than a 2.25% variance in readings between the two machines we taught it was ok. Finally we decided to average the two machines measurements for a single reading at each sitting to give 4 readings per day. Is this still accurate or have we introduced an error increase/decrease/variance in some way?
Now I have 3 questions:
- Was our methodology correct for determining accuracy?
- If we determine accuracy using the above method (best 3 out of 5)...then should we determine a +/- error range by using the average of those readings (giving us an error % from a slightly erroneous average) or should we get the error range by using the max value for the + side, and a min value for the - side. Or is there some other method to get the % error. Or does it really matter how this is reached given the % is relatively small, and there are variances of +/- 5.6% anyway and there may be no way to reduce this 11+% gap.
- Is there any way we can mathematically improve the accuracy before handing back the calibrated machine as we wish to ensure accurate readings over time as she changes her lifestyle and diet with a hope to move off medication (which we were told may never happen) and find more natural alternatives (with help from his doctor of course who is definitely in favour of this).
Our goal is to try various methods over the coming years from dietary changes, to exercise, to meditation, to natural alternatives, using medication only when necessary if possible. We want to have consistent readings before embarking on any changes, measure the changes throughout, and decide based on informed numbers, which changes are working if any and by what amount.
Any help with this problem would be appreciated.