Determination of gas exchange threshold by nonparametric regression

The gas exchange threshold (GET) has been used an an index of anaerobic threshold because it can be measured noninvasively. GET is estimated from a breakpoint in breath by breath values of carbon dioxide uptake (Vco2) and oxygen uptake (Vo2) obtained during a progressive exercise test. Three methods of estimating GET were evaluated: (1) the original V slope method (OVS) using two adjoining standard linear regressions, (2) the modified V slope method (MVS) where the breakpoint is detected by visual inspection, and (3) a new method that we developed with nonparametric regression (NPM) using cubic splines. Simulated data were used because the existence of a breakpoint is known with certainty. Detection accuracy for OVS and MVS never exceeded 63% because of a low specificity. The detection accuracy of NPM ranged between 50 and 89% depending on the amount of noise and abruptness of the threshold, and exceeded that of OVS and MVS at low levels of noise. NPM was significantly more accurate (p < 0.05) than OVS and MVS for detecting GET except with high levels of noise. Both NPM and OVS have similar degrees of numerical accuracy and are superior to the currently used MVS method in this respect. All three methods gave similar results on 20 exercise tests. We conclude from the simulated data that NPM is more accurate than OVS and MVS at detecting GET. NPM can be applied to human data and it provides results that are consistent with OVS and MVS.