Effects of Varying Approaches for Identifying Respiratory Disturbances on Sleep Apnea Assessment

Transcription

1 Effects of Varying Approaches for Identifying Respiratory Disturbances on Sleep Apnea Assessment SUSAN REDLINE, VISHESH K. KAPUR, MARK H. SANDERS, STUART F. QUAN, DANIEL J. GOTTLIEB, DAVID M. RAPOPORT, WILLIAM H. BONEKAT, PHILIP L. SMITH, JAMES P. KILEY, and CONRAD IBER for the Sleep Heart Health Research Group Department of Pediatrics, Case Western Reserve University, Cleveland, Ohio; Department of Medicine, University of Washington, Seattle, Washington; Departments of Medicine and Anesthesiology, University of Pittsburgh, Pittsburgh, Pennsylvania; Department of Medicine and Respiratory Sciences Center, University of Arizona, Tucson, Arizona; Department of Medicine, Boston University, Boston, Massachusetts; Department of Medicine, New York University, New York, New York; Department of Medicine, University of California, Davis, California; Department of Medicine, Johns Hopkins University, Baltimore, Maryland; National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland; and Department of Medicine, University of Minnesota, Minneapolis, Minnesota Varying approaches to measuring the respiratory disturbance index (RDI) may lead to discrepant estimates of the severity of sleep-disordered breathing (SDB). In this study, we assessed the impact of varying the use of corroborative data (presence and degree of desaturation and/or arousal) to identify hypopneas and apneas. The relationships among 10 RDIs defined by various definitions of apneas and hypopneas were assessed in 5,046 participants in the Sleep Heart Health Study (SHHS) who underwent overnight unattended 12-channel polysomnography (PSG). The magnitude of the median RDI varied 10-fold (i.e., 29.3 when the RDI was based on events identified on the basis of flow or volume amplitude criteria alone to 2.0 for an RDI that required an associated 5% desaturation with events). The correlation between RDIs based on different definitions ranged from 0.99 to The highest correlations were among RDIs that required apneas and hypopneas to be associated with some level of desaturation. Lower correlations were observed between RDIs that required desaturation as compared with RDIs defined on the basis of amplitude criteria alone or associated arousal. These data suggest that different approaches for measuring the RDI may contribute to substantial variability in identification and classification of the disorder. Redline S, Kapur VK, Sanders MH, Quan SF, Gottlieb DJ, Rapoport DM, Bonekat WH, Smith PL, Kiley JP, Iber C. Effects of varying approaches for identifying respiratory disturbances on sleep apnea assessment. AM J RESPIR CRIT CARE MED 2000;161: (Received in original form April 7, 1999 and in revised form July 12, 1999) Supported by the NIH, National Heart, Lung, and Blood Institute U01HL53940 (University of Washington), U01HL53941 (Boston University), UO1HL53938 (University of Arizona), U01HL53916 (University of California, Davis), U (University of Minnesota), U01HL53931 (New York University), and U01HL53937 (Johns Hopkins University). Correspondence and requests for reprints should be addressed to Susan Redline, Division of Clinical Epidemiology, Rainbow Babies and Childrens Hospital, Euclid Avenue, Cleveland, OH sxr15@po.cwru.edu Am J Respir Crit Care Med Vol 161. pp , 2000 Internet address: The apnea hypopnea index, also referred to as the respiratory disturbance index (RDI), is the most common metric used to describe and quantify sleep-disordered breathing (SDB). This measure, reflecting the average number of apneas plus hypopneas observed per hour of sleep, is usually derived by identifying and manually counting each respiratory disturbance with subsequent division of the sum by the number of hours slept. Although identification of apneas is generally straightforward, classification of hypopneas frequently requires recognition of more subtle absolute and relative reductions in airflow and/ or ventilatory effort. The difficulty in scoring events is compounded by semiquantitative or qualitative recording techniques. There is great interlaboratory variation in methods used to identify hypopneas (1). This can be attributed to use of different sensors for characterizing breathing pattern, differences in the amplitude criteria (from discernible to 50%) applied to identify any given reductions in breathing signals as hypopneas, and different uses of corroborative data (associated desaturation and arousal) to discriminate normal from hypopneic breaths. The lack of standard definitions for identifying apneas and hypopneas may result in differences in the absolute magnitude of the RDI measured across laboratories. Thus, the magnitude of the physiologic perturbation in a patient with a given RDI from one laboratory may not be comparable to that of a patient determined to have the same RDI in another laboratory. Furthermore, since cutoff values of the RDI (e.g., 5, 10, and 15 events per hour) often are used to identify disease, such differences would be expected to influence case finding in both routine clinical settings as well as in community surveys (and, thus, also influence population prevalence estimates). These factors may be anticipated to impact our perception of the health consequences of SDB. Variations in quantifying the RDI also would be anticipated to influence interpretation of the results of studies of pathogenesis, in which SDB is considered a potential exposure causally related to

2 370 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL hypertension or cardiovascular disease (2, 3). Although a main objective of the Sleep Heart Health Study is to identify outcome-based definitions for sleep-disordered breathing events, this will require longitudinal study. Until then, knowledge regarding the magnitude of the impact that varying definitions have on exposure will provide insight to clinicians and investigators as they attempt to resolve apparent conflicts in the conclusions of published data. In this study, we used data from the Sleep Heart Health Study (SHHS), a large multicenter study of SDB (4), to understand the magnitude with which nonuniformity of respiratory event definition impacts on estimates of SDB. Specifically, we examined the effect on event identification (and overall RDI) of changing the definition of apneas and hypopneas to include variable requirements for concomitant oxyhemoglobin desaturation and arousals in association with a breathing amplitude change. We hypothesized that such differences in respiratory event identification, as may be observed commonly across laboratories (1, 5), result in large differences in estimations of SDB, yielding differences in the distribution of the RDI and prevalence estimates based on given cutoff values. We also explored the extent to which various definitions may differentially classify symptom status. METHODS Study Sample The overall objectives and study design of the SHHS have been reported previously (6). Briefly, the SHHS is a prospective cohort study aimed at investigating the relationship between SDB and cardiovascular disease. Participants were recruited from eight existing epidemiological studies in which data on cardiovascular risk factors had been collected previously. From these parent cohorts, participants who met the inclusion criteria (age 40 yr or older; no history of treatment of sleep apnea; no tracheotomy; no current home oxygen therapy) were invited to participate in the baseline examination of the SHHS. Several cohorts oversampled snorers in order to increase the studywide prevalence of SDB. A total of 6,440 individuals was enrolled between November 1, 1995 and January 31, At the time of this report, 5,708 records was scored. Of these, 5,046 (88.4%) records were determined not to have technical problems with the electroencephalogram (EEG) channel that could prevent sleep staging and arousal identification. The present report is based on these 5,046 SHHS records, 99.8% of which had at least 4 h of total recording time and at least 2 h of time asleep. Of these records, 46.4% were judged in terms of quality to be outstanding to excellent; 41.9% were very good to good; and 11.6% were of fair quality (see Reference [6] for a description of quality codes). Records were screened for studies that had RDIs* 45. These received priority scoring (to identify high-risk subjects). Thus, this article contains data on all studies with RDIs 45 and an approximately 80% random sample of the total studies performed. Baseline SHHS Examination The baseline SHHS data collection session, conducted at the participant s home, included a brief health interview, assessment of current medication use (7), questionnaire administration, blood pressure and anthropometric measurements, and full unattended polysomnography (PSG). Questionnaires included a self-completed Sleep Habits Questionnaire that incorporated questions on snoring frequency and daytime sleepiness. The psychometric properties of these questions, used also in the Specialized Centers of Research in Sleep Disorders Questionnaire, were previously reported (8). * Based on an RDI defined if events were associated with 3% desaturation. Polysomnography Overnight PSG was performed with a portable PS-2 system (Compumedics, Abottsville, Australia). Sensors were placed and equipment was calibrated during an evening home visit by a certified technician. Data collection included C 3 /A 1 and C 4 /A 2 EEGs; right and left electrooculograms (EOGs); a bipolar submental electromyogram (EMG); thoracic and abdominal excursions (inductive plethysmography bands); airflow (detected by a nasal oral thermocouple [Protec, Woodinville, WA]), oximetry (finger pulse oximetry [Nonin, Minneapolis, MN]), electrocardiogram (ECG), and heart rate (using a bipolar ECG lead); body position (using a mercury gauge sensor); and ambient light (on/off, by a light sensor secured to the recording garment). After equipment retrieval, the data, stored in real time on PCMCIA cards, were downloaded to the computers at each respective clinical site, locally reviewed, and forwarded to a central Reading Center (Case Western Reserve University [CWRU], Cleveland, OH) (described in detail in Reference [6]). Scoring of Polysomnograms Sleep stages were scored according to the guidelines developed by Rechtschaffen and Kales (9). Arousals were identified according to American Sleep Disorders Association criteria (10), modified to accommodate situations in which EMG artifact obscured the EEG signal (i.e., portions of an EEG that were totally obscured by EMG artifact were considered arousal if the period of obscuration lasted more than 3 s and began within an epoch of sleep). The common criterion to define an apnea (regardless of the nature of corroborating physiologic occurrences) was a complete or almost complete cessation of airflow (at least 75% of baseline), as measured by the amplitude of the thermocouple signal, lasting 10 s. The common criterion to define hypopnea (regardless of the nature of corroborating physiologic occurrences) was a clear reduction in the amplitude of an index of flow or volume (detected by the thermocouple or thorax or abdominal inductance band signals) to below 30% of the amplitude of baseline breathing for 10 s, but one that did not meet the criterion for apneas. Apneas were considered to be central if no effort was noted on either the thorax or abdominal effort channel. No attempt was made to classify hypopneas as obstructive or central in nature. After apneas and hypopneas were identified, software linked each event to data from the oxygen saturation and EEG channels. Each event was then characterized according to the degree of maximum associated desaturation (0 to 5%) measured from the end of the event until up to 25 s after the end of the event, and associated arousal (measured from the end of the event to 3 s after the event). Discrete RDIs were calculated on the basis of the airflow or volume criteria in conjunction with various combinations of associated desaturation and arousal: RDI-T: Hypopneas and apneas identified on the basis of amplitude changes, regardless of associated desaturation and arousal RDI-2D: Hypopneas and apneas identified on the basis of amplitude changes associated with a 2% desaturation RDI-2D,A: Hypopneas and apneas identified on the basis of amplitude changes associated with a 2% desaturation or arousal RDI-3D: Hypopneas and apneas identified on the basis of amplitude changes associated with a 3% desaturation RDI-3D,A: Hypopneas and apneas identified on the basis of amplitude changes associated with a 3% desaturation or arousal RDI-4D: Includes hypopneas and apneas, identified on the basis of amplitude changes associated with a 4% desaturation RDI-4D,A: Hypopneas and apneas identified on the basis of amplitude changes associated with a 4% desaturation or arousal RDI-5D: Hypopneas and apneas identified on the basis of amplitude changes associated with a 5% desaturation RDI-5D,A: Hypopneas and apneas identified on the basis of amplitude changes associated with a 5% desaturation or arousal RDI-A: Hypopneas and apneas identified on the basis of amplitude changes associated with an arousal regardless of the level of desaturation. RDI-4H: Hypopneas identified on the basis of amplitude changes associated with 4% desaturation; apnea identified on the basis of amplitude changes only. Overall summary measures of desaturation, heart rate variation, arousal frequencies, and sleep stages also were made but are not reported in this article.

3 Redline, Kapur, Sanders, et al.: RDI Measurements: Effects on SDB Estimates 371 Statistical Analyses The distributions of all RDIs (i.e., regardless of the definition of hypopnea or apnea) were highly skewed, and therefore were described according to their median and interquartile levels. The relationships among the RDIs were evaluated with Pearson correlation coefficients for the natural log-transformed RDIs and with Spearman correlation coefficients for the untransformed values. Differences in classification that might result from use of different RDIs were assessed by contingency table analyses, cross-tabulating the various quintiles (e.g., describing the extent to which groups identified in the highest quintile distribution of any RDI are similarly classified in the highest quintile of other RDIs, etc.). The ability of each RDI to predict sleep apnea symptoms is assessed by receiver operator curve analysis (11). In these analyses, the outcome was considered to be sleep apnea symptoms on the basis of responses to the written questionnaire. A sleep apnea symptom complex was considered present if the subject reported snoring that occurred frequently or more ( How often do you snore? ), and if s/he reported excessive daytime sleepiness, occurring often or more (How often do you feel excessively sleepy during the day?). A sleep apnea symptom complex was considered absent if either snoring or excessive daytime sleepiness was reported to occur less frequently than frequently or often, respectively. Other responses were categorized as indeterminate and were not used in the receiver operator curve analyses. True positives and false positives were plotted on y and x coordinates, respectively. The areas under the curve (similar to the c statistic from logistic regression) were calculated for models fit for each RDI with and without simultaneous adjustment for sex, body mass index, and age. RESULTS The characteristics of the study sample are provided in Table 1. Participants were, average, 63 yr of age (range, 40 to 100 yr). The sample included approximately 53% females and 24% minorities. Approximately 34% of participants were habitual snorers, as assessed by affirmative answers to snoring frequently or always. Pearson correlation coefficients among the log-transformed values of various RDIs were all highly significant (Table 2). Correlations were strongest (r 0.90) between RDIs that required similar levels of desaturation for event identification. Weaker correlations (r 0.7) generally were observed TABLE 1 SUBJECT CHARACTERISTICS: SLEEP HEART HEALTH COHORT* Parameter Value n 5,046 Age, yr (40 100) Female, % 52.6 Race White, % 76.0 African-American, % 8.0 Native American, % 9.3 Asian-American, % 1.5 Hispanic, % 5.2 BMI, kg/m ( ) Snoring history Never 10.0 Rarely 10.6 Sometimes 15.4 Frequently 13.8 Always 20.3 In past only 0.9 Snore, but don t know how often 15.6 Don t know if they snore 13.4 * Data represent percentages or means SD; values in parentheses are ranges. Missing values were observed for up to 1.2% of participants. Extreme values ( 15, 50) were excluded if a comparison with another weight on an individual indicated a 30% difference (n 10). between RDIs that did not require any associated desaturation (RDI-T or RDI-A) with RDIs that required some level of associated desaturation. Although correlations were strong between RDIs that required the same degree of desaturation for event identification but varied in their requirements for an associated arousal, they tended to be lower when the criterion for desaturation was higher (e.g., r 0.99 for the correlation between RDI-2D and RDI-2D,A and r 0.86 for the correlation between RDI-5D and RDI-5D,A). Nearly identical findings were observed for the Spearman correlation coefficients of the untransformed values. Although the RDIs were highly intercorrelated, their distributions differed substantially according to differences in RDI definitions (Table 3). Median values of RDI varied by approximately 10-fold for definitions that used the most liberal criteria for event identification (using amplitude changes without any requirement for associated desaturation or arousal) to the most conservative definition (requiring an associated 5% desaturation with amplitude changes). A comparison of two frequently used RDI definitions, RDI-3D and RDI-4D, revealed a twofold difference for median RDI level (9.1 versus 4.4, respectively). Requiring linked arousals to identify respiratory events (regardless of associated desaturation) reduced RDI estimates fourfold (from 29.3 to 5.9). Given the variation in overall median level of the various RDIs, it can be expected that the percentage of affected subjects would vary when disease is defined by using any single threshold level of RDI. This is demonstrated in Figures 1A and 1B, which show the variations in the percentage of subjects classified as having SDB on the basis of RDI threshold values of 5 and 15, respectively. Using an RDI cutoff value of 15 to identify subjects with SDB, prevalence estimates vary sixfold: from 82.8% when the RDI is derived without regard to linked desaturation or arousal, to 10.8% when the RDI is derived using a definition of apnea or hypopnea that requires 5% linked desaturation. Using an RDI cut-point value of 5, and an RDI derivation based on identifying events on the basis of amplitude criteria regardless of associated desaturation or arousal, resulted in classification of almost the entire SHHS cohort as having SDB. On the other hand, use of this threshold level (RDI 5) applied to an RDI defined by requiring events to be associated with 5% desaturation, resulted in classification of 31.6% of subjects as having SDB. Because the various RDIs are highly correlated, it is possible that they may classify SDB similarly after appropriate scaling for the definition being used. Thus, as an exposure measure, it could be asked whether an RDI-T of 29 is equivalent to an RDI-4D of 4. One way of assessing this is to determine the extent to which the uppermost and lower four quintiles of each RDI distribution classified subjects similarly. The level of agreement was compared between the upper quintile for RDI- 4D (a commonly used RDI definition) (12) with all other upper quintiles (Table 4). Conversely, we assessed the extent of disagreement by determining the proportion of subjects in the lowest four RDI-4D quintiles who were in the highest quintile of each of the other RDI definitions. The statistics, measuring the levels of agreement not explained by chance, for these relationships, also are provided. The lowest level of agreement was between RDI-4D and RDI-A (only 63.6% of subjects in the highest RDI-4D were also identified to be in the highest RDI-A quintile; 0.55). In contrast, approximately 90% of subjects in the highest RDI-4D quintile were also identified to be in the highest quintiles for RDI-3D ( 0.88) and RDI-5D ( 0.90). Between 2 and 9% of subjects in the lowest four RDI-4D quintiles would have been identified as having values in the highest quintiles for the other RDI definitions. For

4 372 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL TABLE 2 CORRELATION COEFFICIENTS AMONG VARIOUSLY DEFINED RESPIRATORY DISTURBANCE INDICES* RDI-2D RDI-3D,A RDI-3D RDI-4D,A RDI-4D RDI-5D,A RDI-5D RDI-A RDI-T RDI-4H RDI-2D,A RDI-2D RDI-3D,A RDI-3D RDI-4D,A RDI-4D RDI-5D,A RDI-5D RDI-A RDI-T * Based on the log-transformed RDI values. All correlations are significant at the 0.01 level (two tailed). RDIs defined by some degree of desaturation, almost all of the differential classification occurred between the two highest quintiles. However, of subjects identified in the highest RDI-4D quintile, 7.8% had RDI values that were in the lowest three quintiles for RDI-T and 14.2% were in the lowest three quintiles for RDI-A. For all RDI derivations other than RDI-4H, we applied the desaturation/arousal criteria used in any given RDI definition to both apneas and hypopneas. (For example, for RDI-4D, only apneas and hypopneas that were each associated with a 4% desaturation were used in the derivation of this RDI.) However, many laboratories identify apneas on the basis of amplitude criteria alone, while requiring desaturation and/or arousal to accompany hypopneas. We therefore examined the relationship of an RDI derived by identifying all apneas (regardless of associated desaturation) and all hypopneas that were accompanied by a 4% desaturation (RDI-4H). RDI- 4D and RDI-4H were highly correlated (r 0.976; Table 2); the median level of RDI-4H was modestly higher than RDI- 4D (4.4 versus 5.4; Table 3). The percentage agreement (based on the categorization within the highest quintile of each distribution) between these two measures was high (92.1% agreement; 0.89; Table 4). Because signal quality varied among the PSGs (6, 13), correlations between various RDIs also were examined in studies in which no problems with arousal identification or staging were noted (n 3,105). These analyses showed somewhat higher, but similar, correlations between the RDI-A values and other RDI definitions (i.e., ranging from 0.69 to 0.95). Given the finding of a moderate level of disagreement for classification based on using RDI definitions that required desaturation compared with classification that required arousal alone and classification that required neither arousal nor desaturation, we assessed whether there were differences in the abilities of the various RDIs to predict sleep apnea symptoms. TABLE 3 VARIATION IN RDI LEVELS ACCORDING TO APNEA/HYPOPNEA DEFINITIONS Median Quartile Q1 Q3 (range) RDI-T , 44.8 RDI-A , 10.9 RDI-2% D,A , 32.9 RDI-2% D , 30.2 RDI-3% D,A , 24.0 RDI-3% D , 19.2 RDI-4% D,A , 18.4 RDI-4% D , 11.5 RDI-5% D,A , 15.2 RDI-5% D , 6.8 RDI-4H , 13.8 Figure 1. The proportion of individuals labeled as affected (with SDB) according to RDIs on the basis of varying the use of corroborative data (desaturation and/or arousal) to identify apneas and hypopneas. SDB is identified on the basis of a threshold value of 15 (A; percentage of subjects with an RDI 5) or 5 (B; percentage of subjects with an RDI 15). See text for the definition of each RDI.

5 Redline, Kapur, Sanders, et al.: RDI Measurements: Effects on SDB Estimates 373 TABLE 4 AGREEMENT WITH CLASSIFICATION BY THE RDI-4D DEFINITION In these exploratory analyses, sleep apnea symptoms (habitual snoring and excessive daytime sleepiness; see METHODS) were considered the outcome. This complex was considered present in 5.9% and indeterminate in 4.7% of the cohort. Receiver operator curves, relating the sensitivity and specificity of any given RDI (with and without consideration of effects associated with age, body mass index [BMI], and sex) were constructed to predict sleep apnea symptoms. The area under the curve, a measure of the predictive ability of the model (0.50 considered to represent no predictive ability and 1.0 representing perfect predictive ability) for each model varied from 0.61 (for RDI-A) to 0.65 (for RDI-3D, RDI-4D, and RDI- 5D). With inclusion of age, sex, and BMI in all models, all areas increased modestly (ranging from of 0.65 to 0.66). DISCUSSION Percent Agreement* Percent Disagreement RDI-T RDI-A RDI-2D RDI-2D,A RDI-3D RDI-3D,A RDI-4D,A RDI-5D RDI-5D,A RDI-4H * Percent agreement: Individuals identified in the highest RDI-4D quintile (14.1, n 1,009) also in highest quintiles of each of the other given RDIs. Percent disagreement: Individuals identified in the lowest four RDI-4D quintiles in highest quintiles of each of the other RDIs. Sleep-disordered breathing has been recognized increasingly as an important and treatable condition. During this time, PSG has been accepted as the gold standard for identifying individuals with SDB and for evaluating illness severity and responsiveness to treatment. Although many parameters are measured with PSG, the RDI is the measure that is used most commonly to gauge the presence or absence of disease, and to assess its severity. In addition, researchers have increasingly employed PSG to estimate the RDI in various populations. In these settings, the RDI is considered a potentially adverse physiological exposure, somewhat analogous to consideration of blood pressure or cholesterol levels as intrinsic risk factors for a variety of chronic conditions. Implicit is the assumption that when a threshold value is reached, the risk of adverse health sequelae is increased, and may increase further as the threshold is progressively exceeded (5). Unfortunately, the collection and interpretation of data on the RDI has not been standardized and there has been relatively little consideration of the sources of variability in these measurements and their impact on case finding, prevalence estimates, and pathogenesis research. Sources of laboratory variability include the magnitude of changes in breathing amplitude necessary to describe breathing as reduced (hypopneas; from discernible to 50%), variations in the use of sensors with different sensitivities to detect airflow/ventilation (i.e., thermocouples, thermistors, pressure transducers), and differential use of data on oxygen saturation and arousals to discriminate normal breathing from hypopneas (1). The extent of this problem was emphasized in a survey of 44 American Sleep Disorders Association (ASDA)-accredited sleep laboratories. This survey showed that no two laboratories used the same equipment and definition of hypopnea (1). In this study, we exploited the availability of a large PSG database to assess the extent to which the absolute level of RDI and classification of SDB may vary according to variations in requirements for associated arousal and desaturation in the RDI definitions. The data show that large differences in absolute RDI level result when the approach for identifying apneas and hypopneas changes. Likewise, prevalence estimates of SDB, based on any given cutoff values of the RDI, differ markedly according to RDI derivations. Such differences in absolute RDI level emphasize the problems associated with applying any single threshold value to identify SDB without considering the methods by which the RDI was derived and its underlying population distribution. These differences also emphasize the need to exercise considerable caution when comparing the results of research and clinical studies performed in different laboratories that do not adhere to comparable approaches for identifying respiratory events. Differences may be further amplified in the absence of standardized recording techniques. The large differences in population prevalence estimates for SDB (reviewed in Reference [5]) may be based in large part on differences in measurement approaches. When comparing studies, it is generally difficult to determine whether differences in estimated SDB prevalence are related to true biological differences or simply to differences in measurement techniques. Although all RDI derivations were highly correlated, the levels of agreement varied according to the criteria used to define events (i.e., flow/volume alone, associated desaturation, associated arousal). Because of the robustness of measurements of oxygenation, it was not surprising that the levels of agreement were greater between any pair of RDIs that required some level of desaturation than between any pair of RDIs that included an RDI that either was based on amplitude criteria alone or on associated arousal. For example, approximately 9% of subjects who would have been identified in the uppermost RDI-4D quintile would have been labeled as in the lowest four quintiles of the RDI-A (i.e., derived when events were identified on the basis of requiring an associated arousal but not requiring desaturation). The statistic, measuring the level of agreement beyond which would be expected by chance alone, showed only a fair level of agreement between these two measures. Such differences in classification suggest that the RDI, even when used as a continuous trait (or exposure measure), may provide different information regarding risk according to its derivation. It is possible that some of the heterogeneity among studies relating SDB to outcomes such as hypertension and cognitive impairment may be attributable to differences in approaches for quantifying the RDI as the exposure of interest. In this article, we explored the extent to which different definitions differentially predicted what is currently believed to be some fundamental sleep apnea symptoms and found few differences. However, future analyses of the SHHS data will allow assessment of the extent to which cardiovascular and other health outcomes may be differentially predicted by different RDI derivations. The differences in classification of individuals on the basis of RDIs that required oxygen desaturation for event identification as compared with RDI definitions that did not include this requirement, and also compared with definitions that considered associated arousal, suggest the need to better understand which markers of physiological stress during breathing during sleep are most closely linked to adverse outcomes. In the current analysis, we were able to vary the use of corroborative data only for apnea and hypopnea identification.

6 374 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL Thus, we do not know the implication of varying other parameters, such as the amplitude criteria applied for identifying hypopneas, or the choice of sensors used for characterizing breathing pattern. It may be anticipated that such changes may also substantially contribute to large differences in RDI values. In the current study, we used a rather liberal amplitude criterion for identifying hypopnea (approximately a 30% reduction). This definition is quite similar to ones used in other population-based research (12, 14, 15) and in many clinical laboratories and appeared to be reasonable given the inability to quantify airflow precisely. However, it is possible that stricter amplitude criteria would have minimized the effect of variations in the corroborative data used in event identification. In addition, it is possible that fixing the definition of apnea (e.g., on the basis of observing a virtual absence of airflow, regardless of desaturation or arousal) would have altered the interrelationships among RDIs. The large number of possible variations in different combinations of apneas and hypopneas, each defined differently, made it cumbersome to assess all possible interrelationships. We did, however, compare one definition in which apnea was identified on the basis of amplitude criteria alone whereas hypopnea was defined on the basis of desaturation (RDI-4H) with all other derivations of RDI. Overall, there was an extremely high level of agreement between this measure, and the RDI derived by requiring a 4% desaturation for both apneas and hypopneas. Further, the variation in the relationships between RDI-4H with other RDIs varied similarly as the relationship of RDI-4D with other RDIs. Given this, and because only a minority of events included in all RDIs were apneas (approximately 25%), it appears likely that significant differences in the distributions of the various RDIs would have persisted had the definition of apneas been based only on an amplitude criterion. In summary, variability in respiratory event identification and RDI derivation likely contributes to substantial variability in RDI estimates made across laboratories and across populations. Standardized approaches for characterizing the RDI are needed to compare data from different laboratories and populations, to assess accurately the public health impact of SDB, and for consistent case finding. Although our exploratory analyses of the ability of different RDIs to identify accurately individuals with symptoms of SDB (snoring and sleepiness) did not identify appreciable differences in the predictive ability of the various RDIs, other analyses are required to assess the extent to which the different derivations may predict individuals with differing levels of morbidity and risk for adverse health outcomes. Ultimately, association with such factors will determine the correct definition. Participating institutions and SHHS investigators: Framingham, MA: Boston University: George T. O Connor, Sanford H. Auerbach, Emelia J. Benjamin, Ralph B. D Agostino, Rachel J. Givelber, Daniel J. Gottlieb, and Philip A. Wolf; University of Wisconsin: Terry B. Young. Minneapolis, MN: University of Minnesota: Eyal Shahar, Conrad Iber, Mark W. Mahowald, Paul G. McGovern, and Lori L. Vitelli. New York, NY: New York University: David M. Rapoport and Joyce A. Walsleben; Cornell University: Thomas G. Pickering and Gary D. James; State University of New York, Stonybrook: Joseph E. Schwartz; Columbia University (Harlem Hospital): Velvie A. Pogue and Charles K. Francis. Sacramento, CA/Pittsburgh, PA: University of California, Davis: John A. Robbins and William H. Bonekat; University of Pittsburgh: Anne B. Newman and Mark H. Sanders. Tucson, AZ/Strong Heart Study: University of Arizona: Stuart F. Quan, Michael D. Lebowitz, Paul L. Enright, Richard R. Bootzin, Anthony E. Camilli, Bruce M. Coull, Russell R. Dodge, Gordon A. Ewy, Steven R. Knoper, and Linda S. Snyder; Medlantic Research Institute Phoenix Strong Heart: Barbara V. Howard; University of Oklahoma Oklahoma Strong Heart: Elisa T. Lee and J. L. Yeh; Missouri Breaks Research Institute Dakotas Strong Heart: Thomas K. Welty. Washington County, MD: The Johns Hopkins University: F. Javier Nieto, Jonathan M. Samet, Joel G. Hill, Alan R. Schwartz, Philip L. Smith, and Moyses Szklo. Coordinating Center Seattle, WA: University of Washington: Patricia W. Wahl, Bonnie K. Lind, Vishesh K. Kapur, Coralyn W. Whitney, Richard A. Kronmal, Bruce M. Psaty, and David S. Siscovick. Sleep Reading Center Cleveland, OH: Case Western Reserve University: Susan Redline, Carl E. Rosenberg, and Kingman P. Strohl. NHLBI Project Office Bethesda, MD: James P. Kiley and Richard R. Fabsitz. Acknowledgment : The Sleep Heart Health Study (SHHS) acknowledges the Atherosclerosis Risk in Communities Study (ARIC), the Cardiovascular Health Study (CHS), the Framingham Heart Study (FHS), the Cornell Worksite and Hypertension Studies, the Strong Heart Study (SHS), the Tucson Epidemiology Study of Airways Obstructive Diseases (TES), and the Tucson Health and Environment (H&E) Study for allowing their cohort members to be part of the SHHS. The authors thank the many participants in the SHHS cohort who generously dedicated their time to study participation. They also thank the technical staffs of the clinical sites for their enthusiastic collection of data. They sincerely thank the Reading Center Staff (Jean Arnold, Michael Decker, Guang-Sheng Liao, Thomas Rosenbalm, Joanna Romaniuk, and Susan Surovec) for their many contributions to the development and implementation of the protocols for processing, coding, and scoring the sleep studies. They also are grateful for the many contributions of Kathleen Fisher, whose dedication to the study will always be remembered. The authors thank Compumedics for modifying equipment for use by the SHHS. References 1. Moser, N. J., D. T. R. Berry, and L. Harbison What is hypopnea, anyway? Chest 105: Partinen, M., and C. Guillenminault Daytime sleepiness and vascular orbidity at seven-year follow-up in obstructive sleep apnea patients. Chest 97: Young, T., L. Evans, L. Finn and M. Palta Estimation of the clinically diagnosed proportion of sleep apnea syndrome in middle-aged men and women. Sleep 20(9): Quan, S. F., B. V. Howard, C. Iber, J. P. Kiley, F. J. Nieto, G. T. O Connor, D. M. Rapoport, S. Redline, J. Robbins, J. M. Samet, and P. W. Wahl The Sleep Heart Health Study: design, rationale and methods. Sleep 20(12): Redline, S., and M. Sanders Hypopnea, a floating metric: implications for prevalence, morbidity estimates, and case finding. Sleep 20(12): Redline, S., M. Sanders, B. Lind, S. F. Quan, C. Iber, D. Gottlieb, W. Bonekat, D. M. Rapoport, P. L. Smith, and J. Kiley Methods for obtaining and analyzing polysomnography data for a multi-center study. Sleep 21: Psaty, B. M., M. Lee, P. J. Savage, G. H. Rutan, P. S. German, and M. Lyles Assessing the use of medications in the elderly: methods and initial experience in the Cardiovascular Health Study. J. Clin. Epidemiol. 45: Kump, K., C. Whalen, P. Tishler, I. Browner, V. Ferrette, K. Strohl, and S. Redline Assessment of the validity and utility of a sleep symptom questionnaire in a community sample. Am. J. Respir. Crit. Care Med. 150: Rechtschaffen, A., and A. Kales A Manual of Standardized Techniques and Scoring System for Sleep Stages of Human Subjects. Brain Information Service and Brain Research Institute, Los Angeles. 10. American Sleep Disorders Association EEG arousals: scoring rules and examples [ASDA Report]. Sleep 15: Centor, R. M A Visicalc program for estimating the area under a receiver operating characteristic (ROC) curve. Med. Decis. Making 5: Young, T., M. Palta, J. Dempsey, J. Skatrud, S. Weber, and S. Bader The occurrence of sleep disordered breathing in middle-aged adults. N. Engl. J. Med. 328: Whitney, C. W., D. J. Gottlieb, S. Redline, F. J. Nieto, R. G. Norman, R. R. Dodge, E. Shahar, and S. Surovec Reliability of scoring respiratory disturbance indices and sleep staging. Sleep 21: Redline, S., P. V. Tishler, M. Hans, T. D. Tosteson, K. P. Strohl, and K. Spry Racial differences in sleep disordered breathing in African-Americans and Caucasians. Am. J. Respir. Crit. Care Med. 155: Olson, L. G., M. J. Hensley, M. T. King, and N. A. Saunders A community study of snoring and sleep disordered breathing: health outcomes. Am. J. Respir. Crit. Care Med. 12: