Strategies to improve respiratory chemoreflex characterization by Duffin's rebreathing.

We assessed the test-retest variability of central and peripheral respiratory chemoreflex characterization by Duffin's modified rebreathing method and explored whether signal averaging of repeated trials improves confidence in parameter estimation. Over four laboratory visits, 13 participants (mean ± SD age, 25 ± 5 years) performed six repetitions of modified rebreathing in isoxic-hypoxic conditions [end-tidal ( )  = 50 mmHg] and isoxic-hyperoxic conditions ( = 150 mmHg). End-tidal ( ), and minute ventilation ( ) were measured breath-by-breath, by gas analyser and pneumotachograph. The versus relationships were fitted with a piecewise model to estimate the ventilatory recruitment threshold (VRT) and the slope above the VRT ( S). Breath-by-breath data from the three within- and between-day trials were averaged using two approaches [simple average (fit then average) and ensemble average (average then fit)] and compared with a single-trial fit. Variability was assessed by intraclass correlation (ICC) and coefficient of variance (CV), and the minimal detectable change was computed for each approach using two independent sets of three trials. Within days, the VRT and S exhibited excellent test-retest variability in both hyperoxic conditions (VRT: ICC = 0.965, CV = 2.3%; S: ICC = 0.932, CV = 15.5%) and hypoxic conditions (VRT: ICC = 0.970, CV = 2.9%; S: ICC = 0.891, CV = 17.2%). Between-day reproducibility was also excellent (hyperoxia, VRT: ICC = 0.930, CV = 2.2%; S: ICC = 0.918, CV = 14.2%; and hypoxia, VRT: ICC = 0.940, CV = 3.0%; S: ICC = 0.880, CV = 18.1%). Compared with a single-trial fit, there were no differences in VRT or S using the simple average or ensemble average approaches; however, ensemble averaging reduced the minimal detectable change for S from 2.95 to 1.39 L min  mmHg (hyperoxia) and from 3.64 to 1.82 L min  mmHg (hypoxia). Single trials of modified rebreathing are reproducible; however, signal averaging of repeated trials improves confidence in parameter estimation.