Statistics from Altmetric.com
We read with interest the recent paper published in Thorax by Plant et al1 which analysed predictors of the outcome of non-invasive ventilation (NIV) in patients with acute exacerbations of chronic obstructive pulmonary disease. Few data are available on this important question, and studies specifically designed to propose answers are either retrospective2 or have been performed in a small sample of patients.3 We recently published the results of a trial4 using a design similar to that of Plant et al. Our findings indicated that the greatest predictive power lies in variables reflecting the evolution of arterial blood gases (particularly pH) after a brief (1 hour) initial trial period of NIV. Our multivariate predictive model had an adequate power of discrimination, correctly classifying more than 95% of the patients with a sensitivity of 97% and a specificity of 90% in a subsequently analysed sample of new patients. Our model therefore performed quite differently from that of Plant et al which had poor discriminant value even though they studied a greater number of patients.
The explanation for the differences between our findings and those of Plant et al may arise from the fact that they used data from a previous study that had not been designed to predict the result of NIV.5 In fact, Plant et al did not analyse all possible variables. In particular, they did not look at those which we found to be of great predictive value—namely, the evolution of the arterial gases after a short session of ventilation. Instead, they chose to analyse the arterial blood gases after 4 hours of ventilation. From a clinical standpoint, 4 hours seems to be too long if the object is to detect a failure of NIV in order to choose another therapeutic option (invasive ventilation). It would have been interesting to calculate the power of prediction of our previously published model4 had it been applied to the large sample of patients analysed by Plant et al.
Finally, as the model they proposed was not validated in a subsequent study with a new sample of patients, the clinical usefulness of the predictors is poor, as noted by the authors.