Auto-adjusting positive airway pressure: the fine line between engineering and medicine

Abstract

Auto-adjusting positive airway pressure (APAP), unlike continuous PAP (CPAP), dynamically adjusts treatment pressure in response to events detected automatically from a derived flow signal. Introduced in the 90's, APAP quickly became a key tool in sleep clinics, initially serving as a faster alternative to manual titration for patients with obstructive sleep apnea (OSA), and later also as a long-term treatment option to expedite follow-up visits. APAP and CPAP are overall comparable in terms of adherence, efficacy and control of symptoms. However, concern remains that APAP offers less control of chronic health outcomes, such as blood pressure, kidney function and glycemic values. Other APAP-related challenges entail engineering aspects. A major issue is that APAP algorithms-which govern event detection/identification and pressure adjustments-are proprietary of and vary among manufacturers, making them poorly understood by clinicians. Furthermore, APAP algorithms do not always match-up well when compared to both manual titration or manually scored polysomnography, particularly in the presence of unintentional leak. Variability in event detection, leak compensation, and pressure adjustment algorithms among devices adds another layer of complexity to clinical decision-making. All this complicates the management of OSA patients, who could be left with substantial residual disordered breathing, high leak, and a wide pressure range.This review aims to bridge the gap between the clinical and engineering perspectives of APAP, providing an up-to-date overview of current knowledge and existing challenges that sleep clinicians should consider when managing OSA patients with PAP therapy.

Keywords: APAP; Algorithms; Automatic event detection; Continuous positive airway pressure; OSA therapy; Titration.

Fig. 1

A meta-analysis of 23 randomized controlled trials demonstrated no clinically significant difference in average hours of use in adults with obstructive sleep apnea treated with auto-titrating positive airway pressure (APAP) compared to continuous PAP (CPAP). In addition, a meta-analysis of 6 of these studies demonstrated no clinically significant difference in percent of nights PAP was used. Reproduced with permission from Patil et al. [18]

Fig. 2

Summary illustration comparing the effects of auto-adjusting positive airway pressure (APAP) and continuous positive airway pressure (CPAP) on key treatment and health outcomes. The central axis represents outcomes with no significant difference between the two treatments, supported by reasonable evidence. Advantages specific to APAP and CPAP are displayed on either side, with greater distance from the center indicating a stronger effect or (estimated) higher level of supporting evidence

Fig. 3

Stacked bar charts showing the rankings of four therapies for 24 h systolic blood pressure (SBP), 24 h diastolic blood pressure, daytime SBP, daytime SBP, nighttime SBP, and nighttime SBP according to a network meta-analysis. This ranking analysis was based on the surface of cumulative ranking curve area (SUCRA), which estimates the probability of a treatment being the “best” option among those compared. Treatments in Rank 1 are potentially more favorable than those in Rank 2, etc. Higher cumulative SUCRA values (percentages in brackets) indicate a stronger overall ranking by incorporating probabilities across all ranks. For example, even though OA and IC never rank first in the plot for daytime SBP, their probabilities of ranking in other positions contribute to their likelihood of being the best treatment in 13.63% and 23.73% of cases, respectively. Abbreviations: CPAP, continuous positive airway pressure; APAP, auto-adjusting positive airway pressure; IC, inactive control; OA, oral appliance. Reproduced with permission from Liu et al. [77]

Fig. 4

Schematic diagram illustrating the functional principles of a device for applying automatic continuous positive airway pressure (CPAP). V′ and P_ao are flow and pressure measured at the patient’s airway opening. Reproduced with permission from Farrè et al. [14]

Fig. 5

Real-world example traces from an APAP device. Flow and pressure signals were synchronized with the oxyhemoglobin saturation (SpO₂) signal using a specialized software. The left panel illustrates two hypopneas flagged by the device (red shading), yet neither is associated with a desaturation, making them unlikely to be true events. By contrast, in the right panel, a flagged hypopnea lines up correctly with a desaturation (i.e., a true event was detected). Integrating SpO₂ alongside the estimated flow signal from the device would enhance event detection accuracy

Fig. 6

Signal processing methods. 180 and 181 refer to the start and the stop point detected by the algorithm for the flow curve analysis. 182, 184, 186 and 188 refer to the illustrated percentage in inspiratory volume. Flatness round baseline (FRB) is determined to find if there are any points between 5% and 95% of the flow curve that are below the 5% or the 95% of the flow value on the y axis. In practice, this is done to understand if there is any flow point in the middle of the flow curve that is below a line connecting the flow points at 5% and 95% of the curve. Roundness baseline (RB) is determined similarly, but using 20% and 80% of the flow curve. Flatness flat baseline (FFB) and Flatness baseline (FB) are calculated as the average of all flow points above FRB and RB, respectively. Hence, FB corresponds to weighted peak flow. The striped areas correspond to volume measures: the volume corresponding to FFB is A + B, while the volume corresponding to FB is B. Flow patterns 172, 174 and 176 are exemplified to illustrate that peak flow (Q_peak) does not always correspond to weighted peak flow (Q_wpeak; dashed lines). FRB, RB and FFB are used to control respiratory flow limitation. Modified and reproduced with permission from Freedman and Johnson [100]

Fig. 7

Example flow and pressure traces to illustrate the main principles of the titration algorithms of two manufacturers [103, 104]. Note that time is not represented in scale and the figure assumes all illustrated events are obstructive, with no concurrent leak. Titration pressure is indicated when stable breathing occurs. Different levels of titration pressure contribute to the determination of P90/P95. In the top panel, Resmed’s algorithm is described. A single respiratory event triggers a pressure increase. Pressure is raised more if events occur at low vs. high baseline PAPs. The rate of rise in pressure is limited to 12 cmH₂O per min. Pressure is decreased by 1 cmH₂O for each 40 min of apnea-free breathing (or 20 min, if no flow limitation is detected). Philips-Respironics’s algorithm is presented in the bottom panel. Every two respiratory events trigger a 1 cmH₂O increase in pressure in a “titration cycle”, with a ceiling at 11 cmH₂O, where pressure is held for 8 min. If events persist, pressure is first reduced by 2 cmH₂O and then further reduced to the pressure that prevents snoring (if such pressure was already determined by the algorithm previously) + 1 cmH₂O for 15 min. At the end of this period, all titration limits are cleared and a new “titration cycle” can start, if events keep occurring. If, during the 8 min hold, there is no events, the 11 cmH₂O ceiling is cleared and a new “titration cycle” can begin, if deemed necessary (above 11 cmH₂O, the maximum allowable pressure increase for the “titration cycle” is 3 cmH₂O), otherwise pressure can be reduced by 1 cmH₂O every 8 min. Notably, the starting pressure is 4 cmH₂O to avoid circuit rebreathing [2, 3]

Fig. 8

Exemplary sections from device data showing an obstructive apnea (A), and a central apnea (B), identified through forced oscillation technique (FOT). An obstructive hypopnea (C), and a central hypopnea (D) were detected through flow contour analysis (e.g., flow flattening examination). The arrow marks the point of pressure increase. Reproduced with permission from Herkenrath et al. [148]

Fig. 9

Pressure changes according to simulated obstructive apnea events with and without unintentional leak. The DreamStation device (soft algorithm) demonstrates smooth increases in pressure following onset of events, but minimal adjustments (ΔP = 0.5 cmH₂O) when leak is introduced along with events. By contrast, the AirSense 10 device (aggressive algorithm) shows a significant increase in pressure with events (maximal pressurization = 19.8 cmH₂O), but a modest response when leak is introduced (maximal pressurization = 6.5 cmH₂O, maximal ΔP = 13.3 cmH₂O, mean ΔP = 7.4 cmH₂O). The Prisma 20 A device (soft algorithm) exhibits higher pressure levels with unintentional leak (ΔP = 10.2 cmH₂O; a behavior possibly influenced by inaccurate estimation of unintentional leak and subsequent overcompensation) compared to without leak (ΔP = 8.0 cmH₂O), with the device starting to titrate only after 7th event (⁓540th second). All these devices did not raise pressure when a central event was simulated. Overall, aggressive algorithms result in higher mean pressure biases (ΔP, difference of pressure with and without leak) than soft algorithms. Similar results were obtained when hypopneas were simulated. Reproduced with permission from Fasquel et al. [120]