You want nice equipment, so you go for the highest sampling rate you can find. You do that because everyone knows that more data is always best.
But is that true? Should you spend lots of money maximizing your sampling rate?
The answer is that sample rate is actually not as important as you might think. (Apologies to sales representatives, who may tell you the opposite.) Yes, more data is generally nicer to have than less data – but what you should actually buy is really a matter of 2 things, that you should consider carefully. One is what you want from your data. The other is how concerned you are about the many other variables that are involved in EKG measurement – some of them a lot more critical than sampling rate.
In this post, let’s talk about the first of these – the practical question of what you’re trying to do with the machine you buy.
Are you doing research? Are you looking at QT interval shortening and peak T wave? Or are you doing scary hospital work? Are you chasing chest pain and worrying about bundle branch block mimicking ischemia? Maybe you’re just running a family clinic, and your patients are generally okay, except sometimes they’re babies, and sometimes they wear pacemakers. These are all very different situations. If you really want to spend, you can get EKG equipment that samples 8,000 times a second and will work anywhere. The clinical standard is more like an eighth of that. And a lot of people are happy with even less. So go slowly here.
Let’s say you come from the old world of 128 Hz Holter recorders. Yes, there is some jitter due to sampling error, and that needs to be taken into account for risk stratification. But this might be fine for what you need. It’s common to see enormous data sets in research that were done at 1000 Hz, but then see disclaimers about overreading the fiducial point. Without a way to validate the input, even a fancy-looking data set may actually be worth less than you’d hoped. What rate is normal for clinical use? The European cardiology consensus is that 250-500 Hz will generally do.
Some people argue that this an outdated recommendation, and that problems of heart rate variation standardization would vanish if everyone sampled higher than 4000 Hz, which may be necessary anyway for patients who are old or sick or newly born. Much probably does need to be studied about how to optimize R peak detection. You need more than 1 sample per millisecond (that’s 1000 Hz) to resolve vagally imposed fluctuations, for example. And there are risks associated with R-R undersampling in obstetrics all the time, where you’re looking for a lot of high-frequency, short-duration events. On the other hand, there are good papers published with lower sampling rates than this. If you’re doing something academic and want to study only long-term time scale influences, you may settle for that. In clinical work you would be losing an interesting part of heart rate variation, and you are at risk of signal aliasing problems. But a compromise can work.
There is actually another school of thought about assessing vagal modulation, which says that 250 Hz is actually quite all right. That’s because the intervals rely on other parameters, that are more important. The quality and robustness of the QRS detector is one. EKG gain and quantization and artefact filtering are others. Sometimes you need manual filtering just to make sure there are no false positive or true negative beats. Aliasing is not a big problem, either, really, since there’s always a low pass analog filtering in any analog-to-digital conversion system (making sure the signal bandwidth is preserved). A 250 Hz sampling provides a 4 millisecond maximum error in time. This error is randomly distributed, too. Unless you’re interested in really deep study of the QRS wave, 250 Hz might therefore work fine even for vagal modulation. Don’t forget that there are also other physiological markers to look at nowadays anyway.
250 – 500 Hz can work perfectly well in clinic, in other words. 1000 Hz is better, admittedly. More than this, though, won’t add information for the classical assessment methods, linear or non-linear. Think about it mathematically. A normal EKG has a useful BW of 30-40 Hz, and let’s say a maximum frequency of 45 Hz. So 250 Hz sampling is more than 5 times the maximum signal frequency. (2 x max frequency is the minimum required.)
This very year, the first-ever study (amazingly) of optimal sampling frequency of R-R intervals in heart rate variation analysis came out. 83 real-life patients were assessed, with 1000 Hz sampling. These EKGs were down-sampled to 500, 250, 100, and 50 Hz frequencies with linear interpolation. Signals down-sampled to 100 Hz produced acceptable results for time-domain analysis and Poincaré plots. At 50 Hz the root-mean-squared successive differences and power of high frequency tended to have high values and random errors. For basic HRV analysis, 250 Hz really does work fine, in other words, and you can even get away with 100 Hz.
You do need to know if a pacemaker is controlling the heart beats. How can you know that at 1000-Hz sampling? Pacemaker stimuli are on the order of 0.2 ms, so you’d never catch them at that rate. The answer is that machines often have pacemaker detection features, involving specialized hardware and software that infers the presence of pacemakers by using band-pass filtering, an analog comparator circuit, and a high analog-to-digital conversion rate. This works. You may want to check that your machine has it.
In another post we’ll take a look at the other variables to consider while you’re looking at machines, beyond sampling rate. These include robustness to noise, wavelet parameter choice, and even simple things, like good electrode contact, and good amplifers and filters. Some algorithms are better than others, too, which is yet another point worth considering.
If you’re shopping for equipment, get the best you can, make it compact, and you should probably aim for the longest recording time. But fixating on sampling rate is, for most applications, silly.
If you’re shopping with us, have a look at the. It samples resting patients at 1000 Hz, and its ADC rate is 8000 Hz. (Yes, it can detect pacemakers.) It won’t let you print if you haven’t got the lead wires attached right, so you know you’re getting a good signal. There’s an interpretation package you can get, to use the machine for children and neonates. This machine is popular in cardiology clinics, and it’s one trusted by the Federal Aviation Administration for Class 1 medical check-ups. Or have a look at the . This one samples at 500 Hz. It’s designed for HRV analysis, and it comes with customized exercise protocols, for use with various treadmills and bicycles. It plugs neatly into your PC, and it measures and interprets with quite authoritative CSE databases.