Estimation of the gas exchange threshold in humans: a time series approach

An autoregressive moving average model is presented for detecting the gas exchange threshold (GET) from gas exchange response data taken during an incremental exercise test. The approach has the advantage of modelling the serial correlation in the noise between observations from the same subject. The model is fitted by maximum likelihood and a breakpoint or threshold in the mean function is the estimated GET. It is shown how an estimate of standard error can be obtained by bootstrapping the residuals from the fitted model. To evaluate the approach, gas exchange data on 14 healthy subjects during a ramp exercise test (20 W x min(-1)) to the limits of tolerance, were analysed. The results are compared to the v-slope method and to the method of maximum likelihood assuming independent noise.