INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The sleep apnea-hypopnea syndrome (SAHS) is characterized by reductions in ventilation resulting from total or partial collapse of the upper airway. As has been recently pointed out, accurate quantification of the magnitude of flow reductions is required because one of the most important indices used to determine the severity of the disorder, the apnea-hypopnea index, is based on the detection of breathing cycles with reduced flow (1, 2). Moreover, accurate flow assessment is helpful in detecting and classifying inspiratory flow-limited events by analyzing the time-course pattern of the flow curve (3, 4).

Although the pneumotachograph is the reference device for measuring flow (1), its use in routine diagnostic studies is not practical because it requires the application of a mask to the patient during sleep. The most common device used to assess flow in clinical practice is the thermistor/thermocouple. Although this device is useful to detect apneas, it is not suitable for accurately quantifying the degree of hypopnea or for tracking the flow time course to detect flow limitation (5). To find out a simple alternative device to assess flow during sleep, recent studies have demonstrated that pressure recorded at nasal prongs is a useful signal to detect flow events during sleep (6). Although this device exhibits an excellent time response for tracking the details of the flow time course, its nonlinear response could pose a problem for accurately measuring flow (7, 11). In fact, in a previous study in awake healthy subjects we showed that the relationship between the nasal prongs pressure and actual flow was quadratic (12). Moreover, we also showed that performing the square root of the nasal prongs signal could be a suitable procedure to linearize and, therefore, optimize the device.

To date there are no data evaluating the impact of nasal prongs nonlinearity on the quantification of hypopneas and flow limitation events in sleep studies in SAHS. There is also a lack of data demonstrating the potential improvement achieved by linearizing nasal prongs in this application. Accordingly, the main aim of this work was to evaluate the relevance of linearizing the nasal prongs signal for assessing flow in patients with SAHS. A secondary aim was to design and test a simple analog circuit to facilitate the real-time linearization of nasal prongs in routine use. We carried out a study based on flow signals recorded with a reference pneumotachograph in a polysomnography-controlled continuous positive airway pressure (CPAP) titration in six patients with severe SAHS. These flow signals, including hypopneas, flow limited events, and normal breathing cycles, were reproduced through a nasal analog and recorded by nasal prongs. The raw and linearized signals from nasal prongs were compared with actual flow recorded by a pneumotachograph. The analysis was focused on evaluating the relevance of linearizing nasal prongs to detect and quantify the reduction in flow amplitude during hypopneas and to characterize flow-limited events by assessing the curvature of the flow-time curve.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The study was carried out by means of an experimental setting specifically implemented to analyze the performance of nasal prongs in the laboratory (Figure 1). A flow generator consisting of a servocontrolled pump (5) was able to reproduce any breathing flow waveform stored in the computer. This flow generator was connected through a pneumotachograph to a nasal analog output. The nares were mimicked by two adjacent tubes with an internal diameter of 7.5 mm. Conventional nasal prongs were placed on the artificial nares and pressure at the nasal prongs (PNP) was measured with a pressure transducer (176PC14HG2; Honeywell, Freeport, IL). The nasal prongs device was linearized (NP) by on-line computation of the square root of PNP signal using a specially designed analog circuit which was previously described (13).

View larger version (16K): [in this window] [in a new window]   Figure 1.   Diagram of the experimental set-up to evaluate the relevance of linearizing the nasal prongs signal. PUMP = servocontrolled flow generator; PNT = pneumotachograph; NP = nasal prongs; PT = pressure transducer; SQRT = circuit to linearize a signal by computing its square root, PC = personal computer.

Nasal prongs linearization was analyzed by using a set of nasal flow signals recorded with a pneumotachograph in a patient study which was previously described in detail (14). Briefly, flow and esophageal pressure were recorded during a period of full-night, polysomnography-controlled stepwise CPAP titration in six patients with severe SAHS. These flow signals were exactly reproduced (Figure 1) and actual flow (), PNP, and flow estimated by the linearized nasal prongs signal (NP) were simultaneously sampled at 100 Hz and stored.

To compare the performance of NP to estimate flow, we selected a total of 21 representative inspiratory patterns from the six patients and CPAP values. These flow patterns were hypopneas with different degree of flow limitation in the case of suboptimal CPAP and normal inspirations when optimal CPAP was applied to the patient. The selection of the respiratory events was made from the flow and esophageal pressure curves exclusively. The selection was blind to the signals obtained from the nasal prongs (PNP and NP). An event was classified as flow-limited when increased inspiratory effort was not associated with an increase in flow (3, 4). The degree of flow decrease during an event was quantified by computing an index accounting for the hypopnea flow amplitude (HFA). HFA was defined as the mean inspiratory flow during the event normalized by the mean inspiratory flow in the same patient when subjected to optimal CPAP; i.e., HFA = 0.30 indicated that the flow amplitude during the event was 30% of the value observed with optimal CPAP. For the hypopneas with a HFA > 0.30, we computed the flow limitation index (FLI) previously described by Teschler and coworkers (15). This index quantifies the curvature of the inspiratory flow signal. Specifically, FLI measures to what extent the central part of inspiratory flow exhibits a sinusoidal-like round contour (normal inspiration: FLI ~ 0.3), a square-like contour (moderately flow limited: FLI ~ 0), or a central part of inspiration with a magnitude below mean inspiratory flow (extremely flow limited: FLI < 0). For each event, HFA and FLI were computed from , PNP, and NP. The values obtained from PNP and NP were compared with the ones obtained from as proposed by Bland and Altman (16).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Figure 2 shows an example of measured by the pneumotachograph, PNP, and NP corresponding to the signals originally recorded in a patient with SAHS. In the figure, two first normal cycles are followed by flow limited cycles ending with a cycle of increased amplitude due to arousal. For the sake of comparison, in this figure the scales of both PNP and NP were modified in such a way that the amplitude of the first inspiratory cycle was the same for the three signals. In agreement with its nonlinear behavior, nasal prongs underestimated low flow values and overestimated high flow values. By contrast, linearizing the nasal prongs signal by computing its square root allowed us to accurately reproduce the actual pattern of flow measured by the pneumotachograph.

View larger version (25K): [in this window] [in a new window]   Figure 2.   Example of the actual flow simultaneously recorded with a pneumotachograph (top), flow estimated with nasal prongs (PNP), and linearized nasal prongs signal ( NP). The flow signal includes normal and flow-limited breathing cycles in a patient with SAHS. Positive flow corresponds to inspiration.

More detailed examples of the typical patterns of inspiratory flows found in patients with SAHS are shown in Figure 3. The left panels correspond to a markedly flow-limited inspiration recorded when applying suboptimal CPAP. As described by Teschler and coworkers (15), in this figure the time scale was shown as a fraction of inspiratory time and the flow scale was normalized in such a way that the unit value corresponded to the mean of the inspiratory flow. The figure clearly shows that, compared with the pneumotachograph, PNP over/under estimated the high/low flows. In addition to underestimating HFA (HFA = 0.26 instead of the actual HFA = 0.49), PNP resulted in an underestimation of the FLI from the actual FLI = 0.30 to FLI = 0.61. By contrast, the linearized nasal prongs signal accurately reproduced the actual flow pattern, resulting in a correct estimation of the degree of hypopnea (HFA = 0.48) and flow limitation (FLI = 0.32). Similarly, the right panels in Figure 3 show an inspiratory flow recorded when optimal CPAP was applied. Also in this case, high flow values were overestimated by nasal prongs with the result that the FLI was biased (FLI = 0.52 versus actual FLI = 0.34). The linearized signal accurately estimated the flow contour and the FLI (FLI = 0.36).

View larger version (35K): [in this window] [in a new window]   Figure 3.   Detailed example of inspiratory flows recorded with a reference pneumotachograph (PNT), with nasal prongs (NP) and linearized nasal prongs signal (Linearized NP) in the case of an extremely flow-limited inspiration (left) and of a normalized inspiration (right).

The relationship between HFA computed from the actual flow () and the ones obtained from PNP and from the linearized signal NP in all the events corresponding to suboptimal CPAP is shown in Figure 4A. This plot clearly reflects that PNP is a signal that is nonlinearly related to flow. Owing to the parabolic relationship of the curve corresponding to PNP, PNP resulted in an underestimation of HFA. By contrast, this figure also shows that NP provided a suitable estimation of HFA. Linear regression of HFA values obtained from and NP was close to the identity line: HFA(estimated) = 1.098 · HFA(actual)  0.063 (r² = 0.99). The good accuracy of NP for assessing the degree of hypopnea is also demonstrated in Figure 4B. The difference between HFA values showed a slight positive dependence on the actual value. However, the magnitude of this difference could be neglected for clinical applications: the mean difference in HFA values was 0.014 and the limits of agreement, defined as mean ± 2 SD of differences (16), were 0.064 and 0.092.

View larger version (23K): [in this window] [in a new window]   Figure 4.   (A) Relationship between the actual values of HFA computed from the pneumotachograph and the values of HFA estimated from nasal prongs pressure (open circles) and from the linearized nasal prongs signal (closed circles). Dashed lines correspond to the linear and quadratic regressions of closed circles and open circles, respectively. Additional vertical (HFA = 0.50) and horizontal (HFA = 0.25 and 0.50) axes are shown in dotted lines. (B) Difference between HFA estimated from the linearized nasal prongs signal and actual value as a function of actual value. Dashed line is the mean of the differences, dotted lines are mean ± 2 SD of the differences, and solid line is the linear regression line.

Figure 5A compares the values of FLI estimated from PNP and NP with the actual ones obtained from . Estimation of FLI from PNP was biased: this signal resulted in an overestimation for normalized inspirations (FLI > 0) and in an underestimation for flow-limited events (FLI = 0). By contrast, NP provided virtually the same values of FLI as nasal flow measured with the pneumotachograph: FLI(estimated) = 1.044 · FLI(actual) + 0.007 (r² = 0.99). As shown in Figure 5B, the mean difference in FLI was 0.007 and the limits of agreement were 0.044 and 0.031. The dependence of this difference on the actual index value could be neglected for clinical applications.

View larger version (21K): [in this window] [in a new window]   Figure 5.   (A) Relationship between the actual values of FLI computed from the pneumotachograph and the values of FLI estimated from PNP (open circles) and from NP (closed circles). Solid lines correspond to the linear and quadratic regressions of closed circles and open circles, respectively. (B) Difference between FLI estimated from the linearized nasal prongs signal and actual value as a function of actual value. Dashed line is the mean of the differences, dotted lines are mean ± 2 · SD of the differences, and solid line is the linear regression line.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In this work we evaluated the relevance of linearizing the pressure measured with nasal prongs to quantify the hypopneas and flow limitation events typically found in sleep studies in patients with SAHS. We found that assessing flow by means of PNP resulted in a misestimation of the magnitude of respiratory events. We also found that computing the square root of nasal pressure linearizes the device and allows accurate quantification of both the magnitude of hypopneas and the degree of inspiratory flow limitation. This linearization was readily carried out by means of the analog circuit we developed.

The experimental design employed to study the improvement achieved in flow estimation by linearizing PNP consisted of two steps. First, in a patient study we collected a series of flow patterns that was representative of the ones found in SAHS: hypopneas of different magnitude and flow limitation events with a broad spectrum of curvature in the flow-time curve. In this patient study flow was recorded by means of a pneumotachograph, which is the reference device for flow measurements (1), and inspiratory effort was assessed by means of an esophageal balloon. Second, these representative flow signals were studied in an experimental setting aimed at exclusively analyzing both the impact of nasal prongs nonlinearity and the effects of its linearization on the quantitative assessment of hypopneas and flow limitation. Indeed, reproducing the breathing flow patterns recorded in patients through an analog of the nares and recording them by nasal prongs constituted a controlled procedure which avoided the potential practical problems encountered when using nasal prongs in patients (e.g., movement of the nasal prongs and mouth breathing) (12).

A number of recent studies have pointed out that the use of nasal prongs is a simple and minimally disturbing method to assess flow during sleep (8, 9). When compared with the most widely used device, the thermistor/thermocouple, nasal prongs have the advantage of a fast response to track the time course of flow along the breathing cycle (12). Nevertheless, as flow detection is based on the pressure drop caused by turbulence in the nares, this device is essentially nonlinear, with a quadratic relationship between PNP and (12). Consequently, such a nonlinearity could result in a bias when detecting hypopneas. The results of the present study provide evidence that when applied to the flow-time patterns typical of SAHS, PNP actually overestimates the reduction in flow amplitude during hypopneas and could, therefore, result in overdetection of these events (Figure 4A). For instance, according to the criterion for defining hypopnea as a 50% reduction from normal breathing (1), 13 of the 21 events in Figure 4A would be classified as hypopnea using the actual flow measured by the pneumotachograph (HFA < 0.50). Nevertheless, if PNP was used to estimate flow events, four nonhypopneic events would be erroneously classified as hypopneas (approximately 30% overdetection). This bias is contrary to the one exhibited by thermistors/thermocouples (5). In addition to misclassification of hypopneas, the use of PNP would also affect the computation of FLI. However, this could not constitute a special problem because the index values we computed from the PNP signal were still linear with the reference values (Figure 5A). Consequently, any criterion for classifying flow limitation could be maintained by simply adapting the threshold index values. The main result of our work was that linearizing PNP provided us with a signal which almost perfectly reproduced actual nasal flow (Figure 2). Therefore, computation of the square root of PNP enabled us to accurately quantify the magnitude of flow reduction and, therefore, to improve the classification of hypopneas (Figure 4B).

The linearized nasal prongs signal provided an excellent reproduction of the time course of flow. Nevertheless, it should be emphasized that nasal prongs are not suitable for assessing patient ventilation. Indeed, similarly to PNP, NP is a signal with an uncalibrated gain because the value of NP depends on the PNP- relationship. This relationship is determined by the geometry of the nares and by the placement of the nasal prongs. The coefficients relating PNP and , which may vary after uncontrolled nasal prongs movement (12), are unknown in a given patient setting. However, the quadratic nature of the relationship between PNP and is not modified by changes in the nasal prongs position (12). Consequently, any change owing to nasal prongs movement would simply result in a variation of the gain in the NP. This would not be a practical problem given that hypopneas and flow limitation events are detected by relative comparison with the immediate normal breathing cycles and that possible changes resulting from nasal prongs movement are not reasonably expected within such a short time period.

Practical linearization of PNP could be carried out by a simple modification of the software when using a microprocessor-based polysomnographic recorder. Nevertheless, in routine practice such a procedure is not easily applicable because most of the available conventional polysomnographic systems do not allow arithmetic computations. Consequently, the use of an analog circuit similar to the one we developed would be most convenient because it provides a square root on-line voltage which can be directly fed to any conventional polysomnograph. As the analog circuit described in this work is small, simple, and inexpensive and does not require periodic calibration, it can be incorporated as an optional final stage into any pressure transducer system. In regard to the practical application of the linearization procedure proposed, it should be pointed out that computation of the square root of a signal is affected by any unbalanced offset in the nasal pressure signal. To avoid artifacts, the zero level of the pressure transducer connected to the nasal prongs should be verified during the routine polysomnographic calibration procedure. Given the excellent offset stability in conventional pressure transducers (17), the potential errors induced when computing the square root are expected to be negligible. Nevertheless, special attention should be paid in case of recording PNP by means of pressure transducers with internal AC compensation of the baseline because in this case computation of the square root of PNP may result in erroneous flow assessment. In this situation, it could be possible to indirectly linearize nasal prongs by modifying the threshold of PNP amplitude reduction in the hypopnea definition. Indeed, as shown in Figure 4, it would be equivalent to take a HFA threshold equal to 0.25 [i.e., (0.50)^1/2] for the raw nasal pressure signal or to take the conventional 0.50 for actual () or estimated (NP) flow.

In conclusion, comparison of PNP with measured with a pneumotachograph demonstrates that nasal prongs is a nonlinear fast-response device useful in assessing breathing flow (6, 12). We found that the nonlinearity of this device results in misestimation of the magnitude of hypopneas and flow limitation. Accordingly, using the criterion of 50% reduction in the raw PNP pressure does not coincide with the recommended definition of hypopnea as 50% reduction in actual flow (1). We also showed that the nonlinearity of nasal prongs is easily and effectively corrected by simply computing the square root of PNP. This procedure could be readily implemented in routine practice. Although further clinical studies should be focused on better evaluating the practical performance of nasal prongs within the framework of polysomnography, our findings suggest that nasal prongs linearization could improve the quantification of the magnitude of respiratory sleep disturbances in SAHS.