Background: This study assessed the performance of a Functional Approximation based on a Levenberg-Marquardt Algorithm (FA-LMA) to calculate airway dead space (VD(aw)) and the slope of phase III (S(III)) from capnograms.
Methods: We performed mathematical simulations to test the effect of noises on the calculation of VD(aw) and S(III). Data from ten mechanically ventilated patients at 0, 5 and 10 cmH(2)O of PEEP were also studied. FA-LMA was compared with the traditional Fowler's method (FM).
Results: Simulations showed that: (1) The FM determined VD(aw) with accuracy only if the capnogram approximated a symmetrical curve (S(III) = 0). When capnograms became asymmetrical (S(III) > 0), the FM underestimated VD(aw) (-3.1% to -0.9%). (2) When adding noises on 800 capnograms, VD(aw) was underestimated whenever the FM was used thereby creating a bias between -5.54 and -1.28 ml at standard deviations (SD) of 0.1-1.8 ml (P < 0.0001). FA-LMA calculations of VD(aw) were close to the simulated values with the bias ranging from -0.21 to 0.16 ml at SD from 0.1 to 0.4 ml. The FM overestimated S(III) and showed more bias (0.0041-0.0078 mmHg/ml, P < 0.0001) than the FA-LMA (0.0002-0.0030 mmHg/ml). When calculating VD(aw) from patients, variability was less with the FA-LMA leading to mean variation coefficients of 0.0102, 0.0111 and 0.0123 compared to the FM (0.0243, 0.0247 and 0.0262, P < 0.001) for 0, 5 and 10 cmH(2)O of PEEP, respectively. The FA-LMA also showed less variability in S(III) with mean variation coefficients of 0.0739, 0.0662 and 0.0730 compared to the FM (0.1379, 0.1208 and 0.1246, P < 0.001) for 0, 5 and 10 cmH(2)O of PEEP, respectively.
Conclusions: The Functional Approximation based on a Levenberg-Marquardt Algorithm showed less bias and dispersion compared to the traditional Fowler's method when calculating VD(aw) and S(III).