Revision as of 18:37, 4 September 2015

1 BSX Validity

I have a number of concerns with the validity of the results of the BSX.

• No independent validation. There is no independent validation of the results of BSX, which doesn't mean that the device it doesn't work, but it is a significant concern. BSX have pointed to three individuals who have reviewed their device, Tony Vienneau (Slowtwitch.com), Matt Fisher (triathletesdiary.com), and Ray Maker (dcrainmaker.com), as well as .
• Black Box. The BSX algorithm is proprietary and not clearly documented, which means it's hard to know if there is any scientific basis for their approach. BSX states that their algorithm is patented, but nowhere do they indicate what their patent number is. I believe it is patent US20130096403; see below for details.
• Hidden Data. BSX does not give any access to the underlying data that supports their measurement of Lactate Threshold. What little data is available online seems a little equivocal, though it's hard to judge. BSX told me that a second generation of the software will give access to the underlying data.
• BSX Validation. BSX has performed their own validation but the results are not published. There are various blogs online that perhaps describe the validation that BSX performed, but the details are unclear. These blog entries indicate that BSX did not use the gold standard of MLSS for Lactate Threshold in their validation. Instead they used an incremental treadmill test with lactate measurement from a blood draw, and relied on the rather poor Carmichael Training Systems determination of Lactate Threshold. I believe that this indicates that BSX has performed their internal validation against a seriously flawed baseline. (See Lactate Threshold for details.)
• 3 Minute Stages. The BSX protocol uses an incremental test with three minute stages, which does not appear to be a valid approach for measuring blood lactate directly. This is because blood lactate takes 20-30 minutes to stabilize for a given intensity. Of course, it is possible that muscle oxygenation would not be subject to this limitation as it is measuring a very different physiological parameter, but it is another concern.
• %O2 Display. In my testing, the BSX application showed a constant oxygen saturation level regardless of exercise intensity, typically 86 to 87%. This is in contrast to Moxy that showed a decline in oxygen saturation with increasing exercise intensity, roughly in line with the published literature. BSX has suggested that future versions of their software will present a more realistic estimate of oxygen saturation, and this problem should not indicate an issue with their validity. Personally, I find this a lack of even a relative change in oxygen saturation rather troubling.
• Invalid Results. In my personal testing, the results of the BSX did not seem valid, but seemed to be based on what I told the application was my 10K pace combined with the pace at which I terminated the test. The BSX indicated I had a Lactate Threshold pace of between 6:10 and 8:50 depending on what 10K pace I input and what pace I terminated the test. This of course doesn't mean that the BSX couldn't detect LT for other users; it may just be that if I don't have an obvious change in SmO2, the BSX resorts to guessing. However, it means that you wouldn't know if the LT pace shown is valid or a guess.
  • Tell BSX my 10K pace was 6:30 and terminating at 6:03 gives an LT pace of around 6:30
  • Tell BSX my 10K pace was 6:30 and terminating at 5:43 gives an LT pace of around 6:10
  • Tell BSX my 10K pace was 8:30 and terminating at 7:54 gives an LT pace of around 8:50
  • Tell BSX my 10K pace was 8:30 and terminating at 7:30 gives an LT pace of around 8:10
• Exercise Ignored. A bigger concern is I got the same indicated Lactate Threshold pace if I tell the BSX I'm performing a test, but don't actually do any exercise. Just sitting still and terminating at the same point gives almost exactly the same result.
• Sensor Location. A minor concern is the location of the sensor on the calf, where it may be more affected by changes in biomechanics than using the quads.

2 Patent US20130096403

Patent US20130096403 A1 (Apparatus and method for improving training threshold), which is a patent for determining the Lactate Threshold using an incremental stress test with an infrared sensor. I feel reasonably confident that this is the BSX patent, but I could not confirm this and BSX seem to carefully avoid any reference to the details of their patent. The patent describes an experiment using a 3 minute incremental treadmill run on 34 subjects, which looks like multiple aspects of muscle oxygenation (SmO2):

• • SmO2. The basic measure of SmO2 showed a good correlation between SmO2 and Lactate, but only in 19 of 34 subjects (56%). The example correlation uses Lactate ranges 0.6 to 2.5, suggesting that the correlation might only exist at relatively low intensities.
  • Rate of change of SmO2. This is the slope of the line drawn through the SmO during a stage. Here the correlation existed only for 11 of the 34 subjects (32%). Here the chart shoes a Lactate range of 1.2 to 6.8, which is a much higher intensity and crosses what BSX consider to be the marker for Lactate Threshold (even though there is a fairly linear increase up to the 6.8 mark.)
  • Magnitude of the reduction in SmO2. It looks like they do an exponential fit for the SmO2 and then find the rate of decay constant. This gives a correlation for 25 of the 34 subjects (73%). Here the charts are rather odd; the increment test chart shows Lactate in the range 0.6 to 2.0, but the regression chart only uses value of 0.6 to 1.3 (pretty much resting Lactate levels).
  • Rate of decay of SmO2. Like the magnitude, this uses the rate of decay from the exponential fit, but this only gives a correlation for 9 of the 34 subjects (26%). The charts for this have a Lactate range of 0.8 to 4.0, a reasonable range for a Lactate Threshold test.

Overall, 32 of the 34 had a correlation for one of the variables, but the patent does not suggest a way of determining which variable to use for a given subject. The patent proposes creating a model with a regression constant for each four variables, but the overall correlation for such a model is not shown. Given that the correlations are not shown for the full range of Lactate values seems odd to say the least.