INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Obstructive sleep apnea (OSA) has been traditionally defined according to the apnea-hypopnea index (AHI), calculated as the sum of apneas plus hypopneas per hour of sleep. Although there is only one widely used definition of apneaa cessation of airflow for more than 10 sa variety of definitions of hypopnea exists. In a survey of American Sleep Disorders Association (ASDA)-accredited sleep centers, Moser and colleagues found that no two laboratories used the same definition of hypopnea (1). Criteria included various degrees of reduction in airflow and/or thoracoabdominal movement, and variably required an associated oxygen desaturation or arousal. Such differences might have important implications for both the diagnosis of OSA and the standardization of research results.

Few data have been published on the differential effects of scoring criteria on the AHI. Gould and associates showed that a 50% reduction in thoracoabdominal movement, as assessed by calibrated inductance plethysmography, correlated better with arousal and desaturation frequency than did reductions of 25% or 75% (2, 3). However, calibration of plethysmographs is cumbersome and unstable, and is rarely done in clinical practice (4). Additionally, if oxygen desaturation or arousal is considered to be the pathophysiologically significant consequence of hypopnea, one might question the rationale for scoring mechanical events at all.

Hoping to score only clinically relevant hypopneas, a number of sleep centers have incorporated associated arousals or oxygen desaturation as part of their definition of hypopnea. For example, in the Wisconsin Sleep Cohort Study, Young and coworkers defined hypopnea as any discernible reduction in amplitude of calibrated respiratory inductance plethysmography, accompanied by a  4% decrease in oxyhemoglobin saturation (5) (T. Young, personal communication).

In view of the lack of standard criteria for hypopnea, and the lack of a clinically based rationale for existing criteria, we conducted a study in which we compared the effects of four different arousal- and desaturation-based polysomnographic scoring criteria on the AHI and the measured prevalence of OSA.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patient Recruitment

All patients were recruited from the Alberta Lung Association Sleep Centre, which is the major sleep referal center for southern Alberta. All referred patients were considered eligible, providing they did not have any of the following exclusion criteria in their referral letter: age < 18 yr, history of alcohol abuse (> 5 oz. of hard liquor per night or equivalent), history of neuromuscular disease, severe lung disease (FEV₁ < 50% of predicted), severe daytime hypoxemia (Pa_O2 < 55 mm Hg), severe coronary artery disease (unstable angina, more than three anginal episodes per week, or myocardial infarction or coronary artery bypass surgery in the previous 6 mo), out-of-town residence, referral for chronic insomnia (without symptoms of snoring, gasping, or choking), restless leg syndrome, night terrors, history of psychiatric conditions requiring the use of tranquilizers, or language barrier. Prior to physician assessment, all eligible referred patients were consecutively assigned an identifier number, and were then randomly selected based on the basis of this number. A consecutive subset of 94 of these patients was selected for inclusion in the study, with the remainder being allocated to another study in which a slightly different polysomnographic protocol was used. Patients subsequently underwent assessment by a physician, followed by in-hospital polysomnography. The recruitment was done between January and October 1995, subsequent to the introduction of inductance plethysmography into our standard polysomnographic protocol.

Sleep Study

All polysomnographic data were recorded by a computerized polysomnographic system (Alice3; Healthdyne Technologies, Atlanta, GA). This included a standard montage consisting of two-channel electroencephalograms (EEGs; C4/A1, C3/A2), bilateral electrooculograms (EOGs), a submental electromyogram (EMG), bilateral leg EMGs, and an electrocardiogram (ECG). Airflow was measured with a thermistor (Healthdyne Technologies, Atlanta, GA). Respiratory effort was detected by inductance changes on plethysmography (Respitrace; Ambulatory Monitoring, Ardsley, NY). Oxygen saturation (Sa_O2) was recorded with a finger oximeter (Model 3700; Ohmeda, Denver, CO).

Scoring Criteria

Sleep stages were scored according to the criteria of Rechtschaffen and Kales (6). Arousals were defined as episodes lasting 3 s or longer, in which there was a return of alpha activity associated with increased EMG activity. Apnea was defined as a cessation of oronasal airflow for a minimum of 10 s. Hypopnea was independently identified and defined as a discernible reduction in thoracoabdominal movement for  10 s, associated with any of the three criteria of: (1)  4% reduction in Sa_O2 (Type A); (2) either a  4% reduction in Sa_O2 or an arousal (Type B); (3) an arousal alone (Type C). Oxygen desaturation was defined as a  4% decrease in Sa_O2 (at nadir) with an increase to within 1% of the prehypopnea baseline value. From an operational standpoint, a discernible reduction in thoracoabdominal movement was a > 30% reduction in the signal on either the thoracic or abdominal Respitrace channel. The AHI was calculated as the total number of apneas plus hypopneas divided by total sleep time. The oxygen desaturation index (ODI) was calculated by dividing the total number of oxygen desaturations by the total sleep time, with desaturation defined as a  10-s reduction in Sa_O2 ( 4% of baseline at the nadir), independent of airflow or thoracoabdominal movement. The hypopnea index (HI) was calculated as the total number of hypopneas divided by total sleep time. Sleep stages and arousals were scored independently of mechanical events. However, for the purpose of scoring of hypopneas, both mechanical and EEG channels were viewed together. Deletion of scoring tags from previously scored records, and spacing of polysomnographic rescoring at intervals of several days, ensured independent scoring for each definition of hypopnea.

Rater Variability

A subset of 21 patient records was randomly selected for determination of intra- and interrater variability. These records were independently rescored by both the original scorer and a second polysomnography technician for comparison with the original polysomnographic results. Because polysomnography technicians from the same center would tend to develop convergent approaches to scoring, two raters trained at different centers were selected for this phase of the study. The differences in the raters' training background provided an estimation of interrater variability that may partly reflect the variation expected between different sleep centers.

Statistical Analysis

A descriptive analysis was performed. The AHI, ODI, arousal index, HI, apnea index, and delta-wave sleep were nonnormally distributed, and were logarithmically transformed to a gaussian distribution for the purposes of further analysis and presentation of summary statistics. Scatterplots were constructed for inspection of the association between each pair of scoring schemes. The strength of association (correlation) between polysomnographic indices based on different criteria was determined with Spearman's rank test.

Because a relatively high correlation does not necessarily indicate good agreement, Bland-Altman plots were constructed to assess the extent of agreement (7). Briefly, the plots were used to compare differences in values derived from the two scoring criteria with the mean of the values. The mean of the differences represents the bias between the scoring criteria, and the standard deviation (SD) represents the error (or variability) between the two scoring methods. The 2 SD value of the mean of the differences represents the limits of agreement. Agreement was reported as the bias between scoring criteria (mean of the differences) ± the limits of agreement (2 SD).

Analysis of variance (ANOVA) was used to evaluate differences in the AHI resulting from different hypopnea scoring criteria. Multiple pairwise comparisons were made if global hypothesis testing of AHI demonstrated significant differences among scoring criteria.

At a variety of AHI diagnostic cutoff values (5, 10, 15, 20/h), 2 × 2 frequency tables were constructed, comparing the prevalence of OSA according to the Type A and Type B hypopnea criteria. Differences in the frequency of diagnosis of OSA at each AHI cutoff value were compared through the use of Fisher's exact test. The number of patients tested per discordant result was calculated by taking the reciprocal of the change in point prevalence of OSA determined with the two criteria.

The characteristics of patients with a large number of arousal-based hypopneas (| AHI-B  AHI-A|  5/h) were compared with those of patients without many arousals through the use of unpaired t tests for continuous variables and the chi-square test for categorical variables. The significance level was set at p = 0.05.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patient Characteristics

During the study period, 94 patients underwent overnight polysomnography with inductance plethysmography. Tables 1 and 2 summarizes the patients' characteristics, anthropomorphic measurements, and polysomnography results.

Correlation and Agreement between Scoring Criteria

AHI-A was scored on the basis of hypopnea defined by an associated oxygen desaturation alone (Type A criteria), and AHI-B was scored on the basis of hypopnea defined by either an associated desaturation or arousal (Type B criteria). Table 3 summarizes the AHI values derived with three different scoring criteria and the ODI. Overall, no significant differences existed between mean AHI-A and AHI-B scores and the ODI (p = 0.29). The correlation between the AHI-A, AHI-B, and ODI scoring methods was high (r > 0.98 in all cases). The correlation between AHI-A and AHI-B was 0.99 (Figure 1). Both AHI-A and AHI-B correlated well with the ODI (r = 0.98 in both cases). Weaker correlations were observed between AHI values derived from the arousal (Type C) definition of hypopnea and AHI values derived from definitions that included a component of oxygen desaturation (r = 0.86 to 0.89).

View larger version (11K): [in this window] [in a new window]   Figure 1.   (A) Correlation between AHI with Type A scoring criteria and AHI with Type B scoring criteria (r = 0.99). (B) Correlation between AHI with Type A scoring criteria and the ODI (r = 0.98)

Figure 2 is a Bland-Altman plot of the agreement between AHI-A and AHI-B. The mean difference between AHI-A and AHI-B was only 2.04 ± 1.72/h (2 SD). When AHI (Type A or B) was compared with the ODI, the means of the differences were 1.04 ± 4.07/h and 3.07 ± 4.30/h, respectively. By contrast, the agreement between AHI-A and AHI-C was poor (12.88 ± 14.94/h).

View larger version (10K): [in this window] [in a new window]   Figure 2.   Bland-Altman plot illustrating the agreement in AHI scored with Type A and Type B criteria. The mean of the differences (bias) in AHI with the two scoring methodologies was 2.04 (thick solid line). The difference between results with the two types of AHI will fall within 1.72/h of this mean difference 95% of the time (thin lines, limits of agreement).

The hypopnea indices (HI) derived from Type A and Type B criteria were well correlated (r = 0.97, Figure 3). However, the increased dispersion of HI points outside the limits of agreement (Figure 4, 2 SD boundaries) indicated considerably more error in the agreement between HI scoring criteria than in the case of AHI criteria. Indeed, in contrast to the case with the AHI, the mean HI differed significantly among scoring criteria (p = 0.004). Moreover, a substantial difference (bias) between HI-A and HI-B was observed (5.12 ± 4.87/h [mean ± 2SD]).

View larger version (11K): [in this window] [in a new window]   Figure 3.   Correlation between Type A and Type B scoring criteria for HI (r = 0.97)

View larger version (9K): [in this window] [in a new window]   Figure 4.   Bland-Altman plot illustrating the agreement in HI scored with Type A and Type B criteria. The mean of the differences (bias) in AHI with the two scoring methodologies was 5.12 (thick solid line). The difference between the two test results will fall within 4.87/h of this mean difference 95% of the time (thin lines, limits of agreement).

Effects of Scoring Criteria on the Measured Prevalence of OSA

Despite the small differences between AHI-A and AHI-B, significant differences in the frequency of diagnosis of OSA (point prevalence) were observed with the Type A and Type B definitions of hypopnea (p < 0.0001). As summarized in Table 4, Type B scoring criteria resulted in a higher frequency of diagnosis of OSA regardless of the AHI cutoff value used. The prevalence of OSA with the Type A definition of hypopnea was 82%, 68%, 57%, and 48% with AHI cutoff values of 5, 10, 15, and 20/h, respectively. With Type A versus Type B definitions of hypopnea, the measured prevalence of OSA was greater by 5.3%, 7.4%, 3.2%, and 4.2%, respectively, with the latter. Therefore, when using Type B scoring criteria in this patient population, we would expect an additional patient to be diagnosed with OSA for approximately every 19, 14, 31, or 23 patients tested.

Effects of Large AHI Differences on Diagnostic Concordance

In the comparison of Type A and Type B scoring criteria, differences in the AHI of  5/h did not result in differences in classification of OSA according to the two criteria. Consequently, it can be inferred that large differences in the AHI (AHI  5/h) with different scoring methods occurred only for patients with high AHI values to begin with, and that the diagnostic frequency of OSA was therefore not altered.

Characteristics of Patients with Large Numbers of Arousal-based Hypopneas

Characteristics of patients with concordant (n = 64; both AHI-B and AHI-A produced a diagnosis of OSA [AHI > 10/h]) and discordant results (n = 7; AHI-B, but not AHI-A, produced a diagnosis of OSA) were compared. No significant differences were noted in age, sex, body mass index (BMI), neck circumference, Epworth Sleepiness Scale, time spent at an Sa_O2 below 90%, arousal index, or apnea index in patients with concordant and discordant AHI results. Similarly, the characteristics of patients with a relatively large number of arousal-based hypopneas (n = 5, | AHI-B  AHI-A| > 5/h) and those without (n = 89, | AHI-B  AHI-A| < 5/h) were compared. These groups showed no difference in age, sex, BMI, neck circumference, Epworth Sleepiness Scale, time spent at an Sa_O2 below 90%, arousal index, or apnea index.

Intra- and Interrater Variability

Bland-Altman plots were constructed for both intra- and inter-rater assessments (figures not shown). The interrater difference for AHI-B was 1.72 ± 3.13/h (2 SD), with r = 0.96. The intrarater difference for AHI-B was 0.206 ± 3.51/h, with r = 0.95. Similarly, the inter- and intrarater differences for AHI-A were 1.80 ± 3.11/h and 1.09 ± 2.80/h, respectively. The inter- and intrarater differences for AHI-C were 2.45 ± 4.75/h and 2.37 ± 4.79/h, respectively.

Arousal Index

In general, the arousal index correlated moderately well with all three types of AHI examined in the study, as well as with the ODI (r = 0.70 to 0.75). The inclusion of arousal-based scoring criteria (Type B or C) did not produce a stronger correlation with the arousal index than with other measures of AHI.

Oxygen Desaturation Time

The mean Sa_O2 correlated poorly with all polysomnographic indices (r < 0.20). When the indices were correlated with the percentage of time spent at an Sa_O2 below a predefined level, a slightly better correlation was observed. When correlated with Sa_O2, all indices yielded values of r = 0.45 to 0.56, except the apnea index, which yielded values of r = 0.54 to 0.65.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The present study compared AHIs derived from polysomnograms according to three different definitions of hypopnea. Each definition required a reduction in thoracoabdominal movement for at least 10 s, in addition to one of the following associated events: (1) a  4% reduction in Sa_O2 (Type A); (2) either a  4% decrease in Sa_O2 or an arousal (Type B); (3) an arousal alone (Type C). The results reveal that AHI-A and AHI-B show a high level of correlation and agreement with one another. Consequently, the addition of an arousal requirement to the mechanical and desaturation requirements for hypopnea causes only a small change in AHI, even though it will increase the frequency of diagnosis of OSA if OSA is diagnosed according to an AHI cutoff value alone. By contrast, the correlation and agreement between AHI-C and AHI-A was poor, indicating that a definition of hypopnea requiring associated oxygen desaturation yields substantially different results than one based solely on associated arousal.

We found no significant differences in the mean AHI values for AHI-A and AHI-B and the ODI (p = 0.29). Furthermore, there was a high degree of correlation among AHI-A, AHI-B, and ODI (r > 0.98). Figure 2 shows that AHI-A and AHI-B differed by only 2.04 ± 1.72/h. When compared with the ODI, the mean differences for AHI Type A and Type B were 1.04 ± 4.07/h and 3.07 ± 4.30/h, respectively. Measures of agreement between AHI-A and AHI-B were similar to those seen for inter- and intrarater scoring variability. The use of arousal-based scoring criteria did not change the level of interrater agreement.

Despite the addition of arousal-based scoring criteria, good agreement between the AHI and ODI was still observed. This supports previous data demonstrating the effectiveness of portable monitors, based on continuous digital monitoring of Sa_O2 for the diagnosis of OSA. In one referral-based sleep center in which a polysomnographically derived AHI was used as the "gold standard," a simultaneously recorded respiratory distress index (RDI) derived with a portable monitor had a positive predictive value of 95 to 96% in identifying patients with OSA (8). A major criticism of portable monitors has been their inability to detect respiratory events associated with arousals alone. However, if the addition of arousal-based hypopnea criteria has little impact on the AHI, this becomes less of a concern.

Unlike the AHI, the HI values derived with different scoring methods demonstrated considerably less correlation with one another. The difference in HI values with HI-A and HI-B was also large (5.12 ± 4.87/h [mean ± 2 SD]). Indeed, a statistically significant difference (p = 0.004) was found in the mean HI values achieved with different scoring criteria. However, when combined into the AHI, this difference became insignificant in terms of mean AHI, correlation, and agreement.

We also examined the impact of adding arousal-based scoring criteria on the measured prevalence of OSA. Despite good correlation and only small differences between AHI-A and AHI-B, a significant difference in the frequency of diagnosis of OSA was observed. Table 3 shows that the inclusion of arousal-based scoring criteria will result in an increase in the measured prevalence of OSA. In the setting of our referral based sleep center, and depending on which AHI cutoff value is used, the addition of arousal-based hypopneas (independent of oxygen desaturation) results in OSA being diagnosed in an additional patient for every 13 to 31 patients tested, with a corresponding increase in the measured prevalence of OSA of 3.2 to 7.4%.

The significance of these findings depends on the population sampled. Although the AHI achieved with two different scoring techniques may be similar, a small difference in two AHIs, close to an AHI diagnostic cutoff value, may result in a diagnosis of OSA with one and not the other. In large-scale, community-based epidemiologic surveys, in which most AHI values are expected to lie close to OSA definition levels, small differences in two or more AHIs could produce moderate differences in the measured prevalence of OSA. Indeed, a recent multicenter, community-based longitudinal study reported up to a 16-fold difference in the prevalence of OSA according to the definition of hypopnea that was used (9).

However, in the setting of sleep centers, which are largely referral based, a greater proportion of the population is expected to have AHI values well above OSA diagnostic thresholds. More importantly, the decision to diagnose and treat OSA is not based on a specific AHI value alone. A host of clinical factors are also involved in the process of making this decision. Consequently, for clinical purposes, a quantitative measure of the difference in AHIs based on different scoring criteria (measurement bias), as presented in the present report, is far more important than differences in the measured prevalence of OSA, particularly since the latter would change according to the patient population studied.

In our study, no significant differences in patient characteristics existed between patients with a large number of arousal based hypopneas. Moreover, no significant differences existed in patients in whom OSA was diagnosed with either AHI-B or AHI-A. Consequently, we were unable to identify any patient characteristics that might predict which patients would have significant differences in their AHI according to whether Type A or Type B scoring criteria were employed. However, these findings must be taken in the context of small sample size.

When different scoring systems were related to the arousal index or oxygen desaturation profile, none was clearly superior to the others. In general, the arousal index correlated well with all polysomnographic indices, but no specific index provided a definitively higher correlation coefficient. In particular, this lack of advantage extended to indices that included arousals (Type B or C) as part of their definition.

The mean Sa_O2 did not correlate well with any of the polysomnographic indices (r < 0.20). The time spent with an Sa_O2 below a predefined level correlated moderately with all indices (r = 0.45 to 0.56), with the apnea index yielding the highest level of correlation (r = 0.54 to 0.65).

Because the measured prevalence of OSA depends on the chosen definition of hypopnea, it is important to clarify why we used specific definitions. An oxygen desaturation threshold of  4% was selected because of its widespread use in the research literature (2, 5, 10); the threshold should be above the 2% variability associated with oximetry, and the selection of a lower oxygen desaturation threshold would, if anything, further minimize the bias between AHI-B and AHI-A. If we reduced the desaturation threshold, we would expect an increase in desaturation-based hypopneas, whereas arousal-based hypopneas would remain unchanged. The net effect would be a "dilution" of arousal-based hypopneas by desaturation-based hypopneas, which would result either in minimal changes in AHI (| Type B  Type A| ), or in an actual decrease in the measurement bias. In summary, the use of a 4% desaturation threshold leads to a maximal estimate of measurement bias between AHI-A and AHI-B.

The definition of arousal used in our study differed from that of American Sleep Disorders Association (ASDA) (11) in that we required an increase in EMG amplitude during both rapid-eye movement (REM) and non-REM (NREM) sleep, whereas the ASDA definition requires an associated increase in EMG amplitude during REM sleep only. In our experience, the ASDA definition of arousal is associated with a higher level of interrater variability. Furthermore, because of a lack of information on the reliability and validity of the ASDA definition of arousal, it is not universally endorsed, and several publications require an increase in EMG activity during both REM and NREM sleep, in conjunction with traditional EEG criteria (2, 12, 13), for the definition of arousal.

In conclusion, the use of combined arousal- and desaturation-based scoring criteria has gained considerable popularity in the operational definition of hypopnea, despite the fact that neither approach has been well validated. Our results with a large sample of patients, show that the addition of arousal-based scoring criteria results in only small differences in the AHIs produced by the two scoring methods used in our study, and does not significantly alter measures of correlation. Although the various scoring criteria yielded significant differences in the HI, these disappeared when the HI was combined into the AHI. None of the scoring methods examined in our study showed any clear advantages over the others in terms of associated secondary outcome variables such as arousal frequency or Sa_O2 profile. Nevertheless, despite only small differences in the AHIs produced with different criteria, the addition of arousal-based scoring criteria will increase the measured prevalence of OSA if the latter is diagnosed solely with AHI cutoff values.