Automatic flow Curves analysis during mechanical ventilation (CURVEX) : application to intrinsic PEEP Detection
Résumé
RATIONALE:
Pressure and flow signal curves (flow and pressure) are routinely available on most recent mechanical ventilators. Clinicians’ visual screen monitoring may enable various patient-ventilator mismatching detection. The CURVEX detection process is developed based on a newly developed non-parametric hypothesis testing (SNT for Signal Norm Testing); the robustness of this test is that it does not require any prior information on the distribution of the signal and thus can run in very noisy conditions, with respect to a certain tolerance fixed by users. SNT can run both a single- breath and a sequential time-flow analysis (SNTseq) and can detect various types of signal abnormalities. We tested the hypothesis that this process may allow intrinsic PEEP (PEEPi) detection.
METHODS:
Simulated patients: thirteen different respiratory mechanic and ventilator settings conditions (n=7 with PEEPi/n=6 without PEEPi) were simulated using a respiratory system analog (ASL 5000 IngmarMed and a S1 Hamilton ventilator).
Real Patients: twelve different recordings were retrospectively analyzed (mean 1182±722 cycles). Automatic signal analysis was performed either using the single-breath and the continuous method, at 2 different tolerance levels (τ= 1.5±2 L/min). All curves were blindly analyzed by expert clinicians.
RESULTS:
In all simulated cases, SNT and SNTseq were in accordance with experts’ assessment. In most real patients cases (>90% cycles), SNT analysis was in accordance with experts. In all cases, SNTseq fitted to experts’ advice. τ=2 l/min was the best compromise between sensitivity and specificity.
CONCLUSION:
CURVEX integrates SNT and SNTseq, which are new detection technologies. These algorithms seem accurate to detect PEEPi on a single-breath analysis and to monitor it continuously. This signal-error processing platform may allow various other abnormalities detection during mechanical ventilation.