Changes

Because optical heart rate monitoring relies on blood flow to the skin, air temperature is an important factor in accuracy. In practice, the relationship is a little more complex than you might expect. Warmer temperatures will add heat stress to a runner, and blood will flow to the skin to help control temperature. However, the relationship between air temperature and heat stress on the runner will depend on exercise intensity, as high intensity will produce more heat to be dispersed. Heat stress will also depend on humidity, as the evaporative cooling of sweat is less effective. Another factor is BMI, as higher BMI values represent less effective surface area to volume ratio. Finally, the amount of clothing you wear will have an effect on both heat stress and localized blood flow. You'll notice that the accuracy at 40f/4c is fractionally better than at 50f/10c, and this is because I wear an arm warmer with a hole cut in it at lower temperatures. (I have experimented with wearing an arm warmer in moderate conditions, but the discomfort makes this a little impractical for me.) All this means that the temperatures shown below should not be taken as absolute, but as a general indication that temperature is an important factor, and optical heart rate monitors are probably impractical in cooler conditions for most runners.

[[File:OHRM_ALL_ByTemp.jpg|center|thumb|500px|The accuracy of all optical heart rate monitoring systems combined by temperature.]]

==Accuracy and Rate Of Change Of Heart Rate==

Another patent in my data is that optical heart rate monitors tended to do quite well if my heart rate is steady, but don't track changes in heart rate very well, and can often become completely lost. This means that I can slow up and the optical heart rate monitor reading can jump wildly high, or I can start a high intensity interval and the reading can drop. I'm using ΔHR to mean the difference between the highest and lowest heart rate values recorded using the chest strap in the preceding 60 seconds. Naturally, ΔHR and air temperature interact, so let's look at how optical heart rate monitors cope with changes in heart rate under ideal (68f/20c to 85f/20c) conditions. You'll notice a fairly rapid decline in accuracy as heart rate becomes a changeable.

The graph below breaks down the accuracy by specific device. Some devices clearly do better than others, but I'd urge caution in interpreting this data, as different devices might fit different people better or worse. Looking at the [[Garmin Fenix 5X]], which does far worse than any other device, but I strongly suspect that the due to the weight of the device. This means the Fenix 5X is moving around rather more than a lighter device, increasing the number of bad readings. Even the best devices here only work reasonably well when my heart rate is fairly steady. Even moderate changes in heart rate cause a fairly significant drop in accuracy.

[[File:OHRM_Device_BydHR.jpg|center|thumb|500px| The accuracy of some optical heart rate monitoring systems by ΔHR.]]

==Verifying Chest Strap Gold Standard==

My first step was to verify that the chest strap based monitor is reasonably accurate. To do this, I ran with two different chest strap systems, a Garmin Ant+ (HRM4) and a Polar Bluetooth (H7). I compared over 10,000 readings and found that the two systems matched extremely well. I have uploaded the results below, using two different visualizations. The first graph shows the heart rate measured by the Garmin HRM4 on the horizontal against the Polar H7 on the vertical. If the two systems match exactly then the point will be on the green line of equality. You can see a few places where the two systems don't line up perfectly, but the vast majority of the readings align within a couple of heartbeats. I've used transparent points to give a better impression of the density of data, with black areas having at least 10 data points lining up. There is a blue regression line, which will be aligned with the green line if the system is accurate. I've also included two red lines that are 25 bpm away from the true value. The second graph shows the distribution of differences between the two systems, and again we can see that the vast majority of data points are within a couple of beats. For those with a statistical background, the standard deviation is 1.55 BPM with a bias of -0.06 BPM.

File:VerifyScatter.png| The distribution of readings between the Polar Bluetooth system vertically and the Garman Ant+ system horizontally. This is how the distribution should look if the system being tested is accurate.

File:VerifyDistribution.png| Here is another view of the same data, showing the most readings were exactly the same, or differing by just a couple of beats.

=Optical Heart Rate Monitors Compared With Pulse Oximeters=

''Main article: [[Pulse Oximeter]]s''