Considerations for an Optimal Electrical Activity of the Diaphragm Threshold for Automated Detection of Ineffective Efforts

To the Editor:

We have read with great interest the research letter authored by Jonkman and colleagues (1), and we agreed with the notion that suboptimal filtering of the electrical activity of the diaphragm (EAdi) signal together with a low threshold (>1 μV), could lead to incorrect interpretation of patient–ventilator interactions when detected by automated software.

In our reported validation investigation of Better Care software (2), the algorithm performance was made against five different experts’ opinions using 1,024 tracings of airway flow and airway pressure waveform from 16 different patients, with a reported sensitivity of 91.5% and specificity of 91.7%. Subsequently, as an additional confirmation, we used EAdi tracings with a threshold >1 μV in eight mechanically ventilated patients, obtaining a sensitivity of 65.2% and a specificity of 99.3%. This value was selected on an a priori basis, considering a midpoint between 0.1 μV and 2 μV and was intended to avoid inspiratory assistance during expiration in those cases when the EAdi peak is <1.5 μV and the cycling-off is at a 40% threshold from EAdi peak, instead of the usual 70% (3).

The drop in sensitivity of Better Care algorithm when EAdi was used could be due to, as the authors speculate, a mistaken overestimation of ineffective efforts by EAdi, leading to an increase in false-negative results in the Better Care algorithm. We have seriously considered this possibility in those tracings validated against EAdi, and we have reanalyzed tracings from that previously published cohort, searching for the best cutoff value of EAdi signal with the best performance. The new findings show that the best cutoff value of EAdi is 2.3 μV, with a sensitivity of 89.2%, a specificity of 96%, a positive predictive value of 72.5%, and a negative predictive value of 98.7%.

Overall, it seems that increasing the threshold of EAdi would decrease the false-negative rate, improving the sensitivity of any given automated detection software and keeping a good specificity. We believe that, according to our reassessed results, an EAdi >2 μV could be suitable for this purpose. In addition, as Jonkman and colleagues mentioned, the removal of cardiac electrical activity is technically challenging, particularly when the signal:noise ratio of the crural diaphragm electromyography signal is low. In this scenario, we hypothesized that the automatic detection of true ineffective efforts from EAdi will be improved by using a personalized adaptive threshold for each patient considering the signal:noise ratio of the diaphragm electromyography signal. Interestingly, nonlinear methods less sensitive to ECG interference based on sample entropy algorithms (4) could be used to reduce the delay on the neural onset when an ECG peak matches at the beginning of the breath.