Seven commercial ventilation devices capable of delivering CPAP therapy were adopted: a flowmeter device [the StarVent2 from StarMed (FM)], two OSA devices [Isleep 20, Breas Medical (OSA1); AirSense 10, Resmed (OSA2)], and four mechanical ventilators configured to deliver CPAP [LUISA, Löwenstein Medical (V1); V60 Ventilator, Philips Respironics (V2); Hamilton G5, Hamilton Medical (V3); Evita V800, Draeger (V4)]. These devices are representative of those currently used in clinical settings to deliver CPAP therapy (3) and are designed by top companies manufacturing mechanical ventilators (22). The seven devices were tested using two different patient interfaces: the helmet (Castar R next, StarMed Intersurgical) and the full-face mask (MaxShield Mask, Pulmodyne, BiTrac Select). Each combination of components was tested setting three different CPAP levels, spanning the commonly used therapy range (23): 5 cmH₂O, 7.5 cmH₂O, and 10 cmH₂O. The therapy delivery performances were assessed in the canonical OC as well as in the CCs. Technically, in the OC (Figure 1A) all the tested devices were directly connected to a head phantom wearing the patient interface, with the patient's breathing delivered by a lung simulator (TestChest V3, Organis Gmbh). The interface was equipped with an expiratory valve. This valve takes the form of a whisper valve when using the full-face mask interface and of a PEEP valve when using the helmet interface.

The CC (Figure 1B) adopts the same components as in the OC, with the addition of two unidirectional valves (located upstream and downstream of the patient interface, respectively, to ensure airflow unidirectionality), two antiviral filters, and a CO₂ adsorber (specifically a soda lime canister designed to absorb the CO₂ generated by the patient during the expiratory phase). The need for additional components for the CC is dictated by the fact that while in the OC the air exhaled by the patient is released into the surrounding environment via the expiratory valve (4), in the CC the exhaled air is routed through the unidirectional valve (5), passes through the antiviral filter (6) and the CO₂ adsorber (7), and is then recirculated back into the circuit as purified air.

For the sake of clarity, it must be mentioned that regardless of the configuration, the tested ventilation devices retain their pressure regulation mechanisms: the FM lacks an integrated pressure sensor or feedback pressure control and therefore regulates pressure mechanically at the interface using the PEEP expiratory valve; in contrast, all other devices are equipped with pressure sensors enabling a feedback control to adjust the pressure accordingly. However, while OSA1, OSA2, and V1 measure pressure within the device's internal tubing, V2, V3, and V4 measure pressure at the patient connection port (i.e., at the patient interface).

In both OC and CCs, the air pressure generated at the patient interface (interface pressure) is recorded using a flow analyser (FlowAnalyser Pro, IMT Analytics), while the respiratory flows are recorded within the lung simulator.

Using the lung simulator, a male subject with a height of 180 cm was reproduced as a worst-case scenario. Male subjects generally have larger lung volumes and greater ventilatory demands, making them more challenging for evaluating ventilation devices mechanical performance. This subject was investigated under the healthy condition, a post-surgery hypoxemic condition, and the ARDS condition. The lung simulator settings adopted to replicate the healthy and the pathological conditions were selected based on typical clinical values reported in the literature for each condition (24–32) and are listed in Table 1 in terms of: (i) compliance of the respiratory system; (ii) functional residual capacity; (iii) respiratory rate and (iv) the maximum respiratory effort. The pressure waveforms corresponding to the set respiratory efforts prescribed to the lung simulator are presented in Figure 1.

All the devices were tested according to a multistep strategy. Firstly, the device under evaluation was connected to the selected configuration (Figure 1), and the CPAP operating mode was activated on the device following the manufacturer's instructions. The CPAP level was then selected while keeping the lung simulator deactivated. At this stage, the mean static pressure level (P_mean, Figure 2) reached at the patient's interface was measured. The P_mean value is an indicator of the device's capability to deliver the set therapeutic pressure at the patient interface in apnoeic condition. Subsequently, the lung simulator was activated replicating the selected clinical condition.

Once breathing stabilization was achieved, interface pressure and respiratory flow traces were recorded for a duration of 60 s. Figure 2 illustrates an example of the interface pressure curve, with the derived key parameters proposed to evaluate the pressure performance of all the devices. In detail, the amplitude of dynamic oscillations (ΔP) was calculated as the difference between the maximum (P_max) and minimum (P_min) values of the curve.

In addition, together with these commonly used pressure oscillations parameter (20), three quantities were introduced for the dynamic response assessment: the pressure deviation over time during inspiration (A_i) and expiration (A_e), and the persistence of the pressure curve above the P_mean during expiration (T%). The phases of inspiration and expiration (grey and white bands in Figure 2, respectively) correspond to the reversals of respiratory flow, and they were identified by detecting the time instants at which the zero-crossings of the flow curve occurred. Then, A_i and A_e were computed as the areas between the interface pressure curve and P_mean during the inspiratory and the expiratory phases, respectively. T% was quantified as the time during which the curve remains above P_mean relative to the entire expiration duration.

Contrary to the commonly used P_max or P_min parameters which are instantaneous measures, the integrated quantities A_e and A_i serve as additional descriptors of the entire breathing phases. Indeed, A_e and A_i discriminate between instantaneous high deviations from the P_mean and less pronounced but prolonged deviations. It is noteworthy that smaller expiratory and inspiratory areas, as well as a shorter T%, indicate better performance of the CPAP delivery device. In each performed test, all the values of the parameters were averaged seven breathing cycles.

The statistical analysis of the data was performed in MATLAB (Version R2023, MathWorks, Natick, MA, USA) environment. In detail, the normality of the statistical distribution of the quantities ΔP, A_i, A_e, T% was assessed using the Shapiro–Wilk test. Then, a multifactorial analysis of variance (Factorial ANOVA) was performed considering five independent variables (or factors): the device, the interface, the set CPAP level, the simulated clinical condition and the configuration (Table 2). The Factorial ANOVA was conducted on each dependent variable (ΔP, A_i, A_e, T%), to explore the existence of individual (i.e., main effect) or combined effects (i.e., interaction effect) of the factors on each dependent variable. The Factorial ANOVA evaluates the interaction effects exploring all possible factors combinations and statistically testing if the mean variation in the dependent variable (e.g., ΔP) differs for groups defined by the combination of two or more factors (e.g., device and interface). Significance levels were set to p < 0.05 for all tests.

The Partial Eta Squared (ηp2) was then computed as a measure of the effect size of single or combined factors. According to ηp2 value, the impact on the dependent variable was classified as small (ηp2<0.01), moderate (0.01≤ηp2<0.06) or large (ηp2≥0.06) (33).

Figure 3 provides the P_mean values and the ΔP values for each performed test at the three set CPAP levels investigated.

Notwithstanding the favorable apnoeic condition, in all the devices P_mean values deviate from the set CPAP level regardless of the configuration and the set CPAP level (Figure 3). Closing the ventilation circuit globally leads to a P_mean reduction, with major impact when the mask is used (−1.3 cmH₂O on average for the mask against −0.3 cmH₂O on average for the helmet). This reduction in P_mean values is scarcely affected by the set CPAP level.

The results highlight that closing the ventilation circuit invariably leads to an increase in ΔP (an indicator of the devices reactivity) regardless of the simulated clinical condition under test, of the device, and of the set CPAP pressure level, with an average rise of 50% compared to the corresponding OC. In some cases, increments up to three times emerge for ΔP values.

Conversely, increasing the patient interface volume (i.e., replacing the mask with the helmet) always reduces ΔP values, with decrements ranging from 10% to 64%. Therefore, the increase in ΔP induced by the ventilation circuit closure can be compensated for by the patient's interface. In this regard, Figure 4 highlights that adopting a larger interface confers to the CC a performance which is similar to (and often with smaller pressure oscillations) the canonical OC with the mask interface.

Focusing on the impact of the administered CPAP level, its influence on ΔP varies depending on the interface, configuration, or simulated clinical condition. Although a direct correlation between ΔP and the set CPAP levels does not robustly emerge, insights can be gained by examining the deviations in ΔP across the three CPAP levels, computed as the largest offset between the three ΔP values and their average. Employing the OC, ΔP values deviation is always above 10% in the healthy simulated condition, regardless of the interface adopted. However, in post-surgery and ARDS simulated conditions ΔP deviation never exceeds 8% with the mask, but always surpass 8% (max. 45% for post-surgery and 59% for ARDS) with the helmet almost in all combinations (7 out of 8). Conversely, the CC presents lower ΔP deviations across the three set CPAP pressure levels investigated, especially for OSA devices. Indeed, OSA1 and OSA2 exhibit ΔP deviations below 4% in pathological simulated conditions, while larger but <7% ΔP deviations are obtained using V2 and V1, the latter only for the post-surgery simulated condition. As for V1, ΔP deviations nearly reaching 20% emerge for the ARDS simulated condition.

Finally, comparing the different simulated clinical conditions, pathological ones exhibit higher ΔP values. This is primarily due to an increase in the expiratory peak (P_max), while the inspiratory peak (P_min) remains relatively unchanged. The observed higher P_max values are ascribable to the pathologic simulated conditions' higher respiratory rate.

Elevated P_max values are particularly noticeable when using a mask interface, as the combination of the smaller interface volume and of the higher respiratory rate makes it harder for the controller of the devices the minimization of pressure oscillations. However, the use of the helmet interface proves effective in mitigating this issue. Furthermore, in the OC, as the severity of the pathology worsens (i.e., moving from post-surgery to ARDS), there is a general increase in ΔP values. Notably, in the ARDS simulated condition ΔP values are higher (up to +20%) compared to the post-surgery one. Interestingly, when the ventilation circuit is closed, in most combinations, larger ΔP oscillations (up to +20%) are observed for the post-surgery condition, compared to the ARDS one.

The comparison expiratory area A_e vs. inspiratory area A_i is presented in Figure 5. Like ΔP data, closing the ventilation circuit leads to an overall increase of the areas, with major impact on A_e. Indeed, on average, a 2.5 times increment is recorded on A_e with peaks reaching 13 times, while a 2 times increment is recorded on A_i with peaks reaching 11 times. This is explained by the positive correlation between the ΔP and the expiratory area (R² > 0.76). Moreover, both A_i and A_e are not affected by the administered CPAP set level.

Overall, A_i and A_e are larger in simulated pathological conditions, with a major increase for A_e (+38% on average) then for A_i (+27% on average) with respect to the simulated healthy condition.

Further considerations can be drawn from the A_e vs. T% relationships reported in Figure 6, where it emerges that moving from the top-right to the bottom-left corner, the combinations exhibit increasingly responsive behaviour.

For our purposes, this responsiveness is defined as the device's ability to minimize pressure oscillations while working within that specific combination of interface, set CPAP level, simulated clinical condition, and configuration.

Focusing on the configurations, OC is mainly located below 50th percentile of A_e, while CC is mainly located above, implying worse responsiveness. The introduction of the helmet interface shifts both configurations towards lower A_e. Moreover, in OC, it also reduces the time during which the curve remains above P_mean with respect to the entire expiration duration (i.e., T%); this doesn't occur in CC for pathological simulated conditions.

Finally, the multifactorial analysis of variance highlighted that all the investigated factors have a significant impact on ΔP, A_i, A_e, T%. Interface, device, and configuration variables showed a strong effect, as indicated by very low p-values (≪0.05), while the set CPAP level has a less pronounced effect. Specifically, no significant effects were observed in the following: (1) for the A_i parameter, in the interactions between set CPAP level and configuration (p = 0.47), set CPAP level and simulated clinical condition (p = 0.10), and set CPAP level and interface (p = 0.096); (2) for the A_e parameter, in the interaction between set CPAP level and simulated clinical condition (p = 0.26); and (3) for the T% parameter, with the set CPAP level factor alone (p = 0.16). The low variability observed in all the measured quantities confirms that the CC is stable and robust in delivering CPAP. Detailed factorial ANOVA results are provided in the Supplementary Tables S5–8.

The Partial Eta Squared analysis allowed to rank the factors according to their effect size on each dependent variable. For ΔP, the interface showed the largest effect (0.27), while the configuration had a smaller but still notable impact (0.19). For A_i and A_e, the configuration was the most dominant factor (0.14 and 0.28, respectively), while the interface showed a moderate effect (0.06 and 0.15, respectively). The configuration and the simulated clinical condition showed relevant effects on T% (0.19 and 0.12, respectively). Finally, the set CPAP level had minimal impact on all dependent variables, with ηp2 consistently below 0.01.