O040 Investigating the effects of airflow signal filtering on non-invasively quantified estimates of flow limitation

Abstract Introduction We previously reported a method to estimate flow limitation (FL, range 0-~1) from the airflow signal recorded within a polysomnogram. This method utilised high quality nasal pressure signals, without high-pass (HP) filtering. However, many clinical recording systems implement HP filtering on airflow signals. We hypothesised that FL estimates would be adversely affected when airflow signals have HP filtering present. Methods Within our existing framework for estimating FL, we manipulated the airflow signal by applying digital HP filtering (Butterworth, 4th order) at commonly used cutoff frequencies (0.005, 0.01, 0.03, 0.1Hz). At each HP filter cutoff, performance was assessed at the breath-level (pooled, all patients) and patient-level (median, sleep only) using coefficient of determination (R²) and root mean squared error (RMSE) comparing FL estimates with HP filtering against reference data without HP filtering. Results In N=96 individuals, R² decreased while RMSE increased as HP filter cutoff increased. For HP filter cutoffs at 0.005, 0.01, 0.03 and 0.1Hz, breath-level R² was 0.923, 0.879, 0.740, and 0.483; while RMSE was 0.062, 0.079, 0.129, and 0.219, respectively. Similarly, patient-level R² was 0.994, 0.977, 0.843, and 0.738; while RMSE was 0.015, 0.030, 0.100, and 0.170. Discussion Increasing HP cutoff frequency resulted in a reduction of consistency of FL estimates when compared to those made using an unfiltered airflow signal. Further work including reference to ‘gold standard’ is required, however, in line with current AASM recommendation (“Direct current (DC) or ≤0.03Hz”), this study suggests HP filter cutoff frequency for nasal pressure should be minimised, preferably avoided.