Exponential fitting of pressure-volume curves: confidence limits and sensitivity to noise.

It has recently become common to model the static pressure-volume curve of the lung as a monoexponential function of the form V = A - Be-KP, where V is volume and P is transpulmonary pressure. The parameters A, B, and particularly K have been employed as descriptors of the intrinsic mechanical properties of the lung. The purpose of the present study was to investigate the sensitivities of A, B, and K to noise in measurements of P and V and to incompleteness of data. Using Monte-Carlo simulation, we found that the presence of typical levels of noise in P led to biased estimates of K and that the 95% confidence intervals about A and K were large compared with the parameter values themselves. These effects were increased as points were systematically removed from either end of the data set. These findings show that values of K estimated from PV data are difficult to interpret without accompanying confidence intervals.