INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Obstructive sleep apnea hypopnea syndrome (OSAHS) is the health condition most commonly associated with disordered breathing during sleep and sleep fragmentation. A major component of the morbidity of the condition is thought to relate to short-term and possible long-term neurocognitive deficits caused by exposure to intermittent hypoxemia, hypercapnia, and arousal, acting independently or synergistically. Deficits have been examined using classic neuropsychological (NP) approaches with administration of test batteries that assess a range of functions (broadly considered to assess different structural areas of the brain). These include executive functions (processes involved in planning, initiation, or self-regulation of goal-oriented behavior), attention (including span of apprehension, sustained attention, vigilance, information processing efficiency, and response times), and learning and memory.

Interpretation of such tests requires careful consideration of the noncognitive influences that may confound study findings, including the sensitivity of the tests to the effects of age, alcohol consumption, education level, and motivation, as well as differences in baseline intelligence. Studies that have used these approaches to assess NP function in OSAHS include case series of patients with comparisons with published normative data (1), case-control and cross-sectional studies (2, 3), pre-poststudies (4, 5), small randomized control studies (6, 7), and a population-based study (8). A wide range of results that could be attributable to the effects of OSAHS (or to the effects of treatment aimed at reversing OSAHS) has been reported. In general, the largest and broadest range of NP deficits has been demonstrated in studies of patients with severe SDB, as evidenced by pathological levels of sleepiness, severe sleep-associated hypoxemia, and/or with RDIs > 40 (1, 2, 4, 9). Study results suggest that hypoxemia most closely predicts deficits in executive functions and psychomotor skills, whereas sleepiness predicts deficits in attention measures (3, 5). However, it is not clear from these studies if effects were due to unmeasured confounding (from studying subjects with underlying medical or psychological problems) and whether effects persisted across the entire spectrum of SDB.

This study is unique in that it includes a large number of subjects with low to intermediate levels of SDB, most of whom were not referred to a Sleep Center, and whom were also screened to be free of underlying neurological or medical problems that might impact NP function. This study assesses the extent to which variations in NP function in a nonreferred sample may be associated with variations in sleepiness, as described by both objective and subjective measures, and by direct measures of SDB, that is, indices of sleep-related hypoxemia, sleep fragmentation, and the RDI.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

Subjects were participants in a study aimed at determining the NP and functional deficits potentially attributable to SDB. Recruitment was designed to enroll subjects with a wide range of SDB who had no underlying neurological or medical conditions that could confound any association between SDB and the study outcomes. Seventy-six subjects were recruited who met the criteria for mild to moderate SDB, defined on the basis of an RDI between 5 and 30 and the absence of "pathological" sleepiness (did not fall asleep driving or in other potentially dangerous situations). An additional 24 subjects were recruited who met criteria for severe SDB (RDI > 30). To minimize the effects of confounding due to age or to underlying comorbidities, recruitment was restricted to subjects between the ages of 25 to 65 yr and to subjects who had no sleep disorder other than SDB (e.g., narcolepsy, insomnia defined as regularly sleeping less than 6 h per night, regular use of hypnotics, sleep insufficiency defined as sleeping > 2 h more on nonwork as compared with work days, or a history of periodic leg movements) and no underlying conditions that could interfere with NP test performance or with adherence to the study protocol. These included severe or unstable medical problems (myocardial infarction or congestive heart failure documented within the previous 3 mo, uncontrolled diabetes or thyroid disorder, cirrhosis, or recently diagnosed cancer), neurological disease (a history of stroke, seizure disorder, or head trauma with loss of consciousness for > 6 h or associated with memory impairment), alcohol abuse (a history of more than five alcoholic drinks/d for > 6 yr) or drug abuse (current use or heavy past use leading to tolerance of dependence), regular use of medications that impair sensorium (e.g., benzodiazepines), and < 8 yr of schooling, or education beyond a master's degree. Eligibility was initially ascertained by orally administered questionnaires and subsequently by a screening in-home sleep study (Edentech II Monitoring Device; Edentech, Eden Prairie, MN) (10).

Volunteers for this study were identified through fliers and letters distributed at local work sites, physician offices, and general outpatient clinics eliciting participation of "snorers" in a research protocol (n = 58), from rosters of participants in other research studies that included sleep apnea screening (n = 22), from physician identification of snorers for research (n = 17), and from identification through a Sleep Clinic (n = 3). Thirty-two of the subjects in the current report contributed to data for a previous preliminary report on neuropsychological function in SDB (11); data from subjects in this report also contributed to findings reported on changes in functional status after continuous positive airway pressure (CPAP) therapy (12).

Overview of Protocol

Subjects were screened for enrollment (to determine an initial RDI classification) with in-home sleep monitoring and screening questionnaires. If screening eligibility criteria were met and the RDI was  5, subjects underwent an in-laboratory polysomnogram. When possible, this was performed the night prior to the NP testing battery. However, if an in-laboratory study could not be scheduled the night immediately preceding the daytime evaluation, a polysomnogram was conducted on an alternative night preceding the daytime testing. These individuals then underwent an additional in-home study the night immediately prior to NP testing (to expose the subject to a preceding monitoring night and to document adequacy of sleep time).

Daytime Testing Battery

The testing day included assessment of objective sleepiness level with the Multiple Sleep Latency Test (MSLT) (13). The NP tests were administered primarily between the first and second and between the third and fourth naps. A Continuous Performance Test was administered in both the morning and afternoon (prior to naps I and III). The Pursuit Rotor Test was administered three times at equally spaced intervals over the course of the day. All other neurocognitive tests were divided into two blocks that were alternatively administered (in different subjects) in either the afternoon or morning. These different testing sequences were designed to minimize systematic time-of-day effects. Because the study also included a posttreatment evaluation (data not shown here), subjects were randomized to receive alternate forms of available tests (for the Complex Visual Design Test, Trails A and B, and the California Verbal Learning Test) to minimize practice effects.

Evaluation of Sleep-Disordered Breathing

SDB was assessed both with in-home and in-laboratory polysomnography (PSG). Overnight in-laboratory PSG consisted of measurement of airflow (by nasal/oral thermistry), chest effort (by piezo sensors) finger pulse oximetry, heart rate, right and left electrooculograms, central and occipital electroencephalograms, leg movements (piezo sensors), and body position with a Nihon Koden 4400 Polygraph or a Sensor Medics Somno Star 4100 Polygraph. In-home sleep studies, performed as an initial screening tool, were accomplished with measurement of nasal and oral thermistry, chest wall impedance, finger pulse oximetry, body position, and heart rate with a portable monitor (Edentech Model 2) as previously described (10).

Scoring of Sleep and Respiratory Data

The overnight in-laboratory PSGs provided data on the RDI and sleep fragmentation. Sleep was staged according to published criteria (14). Arousals were identified using the criteria established by the American Sleep Disorders Association (15). Respiratory disturbances (apnea or hypopnea) were defined as a discernible change in airflow or chest wall movement, lasting > 10 s, occurring in association with at least a 3% drop in oxygen saturation or an arousal.

The percentage of time with a saturation < 90% was obtained from the digitized oxygen saturation records from the in-home sleep studies. (Time in desaturation could not be adequately determined with the in-laboratory studies because of problems in the commercial in-laboratory software.) The overall comparability of SDB measures made using the in-home as compared with the in-laboratory studies is supported by a high correlation between the screening RDI and the RDI from the polysomongraph (r = 0.95).

Evaluation of Sleepiness

Objective sleepiness level was determined with the MSLT (13). On the day of NP testing, subjects underwent an MSLT consisting of four naps at 2-h intervals, starting at 0900-0930, allowing the subjects 20 min to fall asleep per nap, according to a standard protocol. Subjective sleepiness was assessed with the Epworth Sleepiness Scale (ESS) (16).

Subjects completed the ESS on the day of the testing battery, either during the morning or afternoon session, according to the testing sequence to which the subject was randomized. The ESS is a self-administered questionnaire that requires the subject to rate his or her chances of falling asleep in eight different situations, referring to his or her "usual way of life in recent times." Scores vary from 0 to 24, with increasing scores indicating greater sleepiness.

Evaluation of Neuropsychological Functions

General intelligence was assessed with four subtests (arithmetic, similarities, picture completion, and digit symbol) of the Wechsler Adult Intelligence ScaleRevised (WAIS-R) (17), and when summed, have the highest correlation with full-scale IQ (18) and are referred to in this study as "estimated IQ." The other NP tasks were selected to tap three general areas of functionattention, memory, and executive functionsshown to be impaired in previous studies of patients with severe SDB (2, 3, 5). Use of several measures for each cognitive construct was intended to minimize test-specific variance. Unless otherwise indicated, information about all tests may be found in Lezak (19).

Five measures were used to examine different aspects of attention. A Continuous Performance Test (CPT), a visual vigilance task, required the participant to monitor a computer screen for 10 min, responding whenever two successive complex geometric stimuli (each presented for 50 ms and replaced each second) are identical (20). Vigilance decrement (d'), that is, decline in perceptual sensitivity, was scored by subtracting signal discrimination accuracy during the last 2 min from accuracy during the first 2 min of the 10-min trial (i.e., the lower the number, the lower the falloff in signal discrimination over the testing period). Although this test was administered two times over the testing day, only data from the morning test were used in this report.

The Diller Letter Cancellation Test provided the opportunity to measure time to completion, initially posited as a measure of sustained attention. Decision time in a computer-administered four choice reaction time test developed for this project evaluated decision-making speed. WAIS-R digits backward (17) examined span of apprehension. Information processing efficiency was assessed with the Gilmore-Royer Symbol-Digit Substitution Test (21).

Verbal memory was assessed with two scores from the California Verbal Learning Test (CVLT), the total number of words recalled during the five learning trials and the number of words recalled after a 20-min delay interval. Nonverbal memory was examined by the percent recall of the copy score after a 45-min delay from the Mack (Department of Psychiatry, Case Western Reserve University, Cleveland, OH) or Rey-Osterreith Complex Visual Designs. (Scores for these two tests were standardized so they can be directly compared.)

Procedural or motor learning was evaluated with the Pursuit-Rotor Test. This test of eye-hand coordination was administered on three equally spaced times (trials) over the course of the testing day. Following a practice period, each trial consisted of six 20-s runs that required the subject to trace a target light with a light-sensitive wand (Lafayette Instruments, Indiana). The percentage of time "on target" for the last trial was divided by the percentage of time "on target" during the first trial.

Executive functions were assessed with four measures. Failure to maintain a sorting set and number of perseverative errors were derived from the Wisconsin Card Sorting Test (WCST). Verbal fluency was assessed as the sum of words generated for three stimulus letters presented during 1-min trials (Benton Multilingual Aphasia Examination). The fourth measure was the time to complete Trail Making B divided by time to complete Trail Making A. This ratio removed general speed variance (assessed by Trail Making A time) from the time requirement for shifting (from letters to numbers to letters).

Statistical Analysis

All analyses were conducted using Statistical Analysis System version 6.12. Variables were examined for skewness and kurtosis. To achieve approximate normality, the following variables were transformed: number of perseverative errors on the WCST (logarithm), Trail Making B/Trail Making A times (logarithm), decision time on Choice Reaction Time test (reciprocal), and response time for the LCT (logarithm).

The polysomnographic and sleepiness variables were considered to be independent variables. The RDI (as defined above) was used to characterize the frequency of breathing disturbances. Sleep hypoxemia was the percent of time/sleep period with < 90% oxygen saturation. Sleep fragmentation refers to the arousal index (arousals/sleep hour). The objective MSLT sleepiness measure and the self-report ESS were significantly correlated (r = 0.43, p < 0.001), so the measures were combined into a single composite index of the sleepiness construct, after reflecting the ESS to be scored in the same direction as the MSLT and standardizing each distribution (i.e., higher values indicate less sleepiness). This composite was derived to provide a better specification of the latent construct than use of a single measurement (22).The composite sleepiness index had an  coefficient of 0.60.

The large number of NP variables was reduced into clinically meaningful, empirically based dimensions using principal components and principal factors analysis (described below). To enrich the sample for the factor analysis, data from the 100 study subjects were combined with data from two other samples studied with the same NP protocol: 69 healthy controls and 25 other volunteers with SDB, who met initial study entry criteria, but who were later found to have exclusionary health conditions.

The associations between dependent and independent variables were assessed with simple and multiple regression analysis. Linearity assumptions were assessed by examination of plots and by testing for threshold effects (i.e., for hypoxemia or RDI). To examine threshold effects, we created variables of the form (X  c)⁺ where X is the variable of interest (RDI or percentage sleep time at saturation levels of < 90%), c is the proposed threshold, and (X  c)⁺ is defined as 0 if X < c and (X  c) if X > c. Inclusion of the term (X  c)⁺ in the model instead of X posits a relationship in which X has no effect on response when it is less than the cutpoint C, and has a linear effect for values larger than the cutpoint. Variables were created to examine four possible threshold levels of RDI severity (5, 10, 15, 30), and four threshold levels of sleep hypoxemia (percent sleep time at < 90% oxygen saturation of > 3%, 5%, 10%, 15%). A model without inclusion of a term for a threshold effect (including only the variable for RDI or hypoxemia, expressed as continuous variables, and age) was compared with the models that included the variables (X  c)⁺ for each threshold examined. The Akaike information criterion (AIC) (23), defined as the log likelihood minus the number of parameters, was used to compare models. The best threshold model was considered to be an improvement over the zero threshold model if it had a higher AIC value.

The sample size (n = 100) was sufficient size to provide, with 80% power, identification of predictor variables that increase the R² by 0.074 or more (i.e., that explained 7.4% of the variance in NP constructs).

The protocol was approved by the institutional Review Board of the Louis Stokes Cleveland Department of Veterans Affairs Medical Center, and written informed consent was obtained for all subjects.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The demographic characteristics and distributions of measures of sleepiness and polysomnographic measures of the research participants are shown in Table 1.

Participants included an equal proportion of men and women of generally average intelligence (Table 1). Of the sample 70% is white, 26% African-American, 2% Asian, and 2% Hispanic. Most subjects were overweight or obese.

The median RDI was 18.1, a level approximately two-thirds to a half lower than reported in previous studies of NP function of more severe sleep apnea (2, 8). Overall level of sleep-associated hypoxemia was modest. Mean levels of MSLT indicated that the sample, on average, was moderately sleepy. Of all subjects 23% had marked levels of sleepiness as demonstrated by MSLT levels of < 5.

The distribution of NP measures is detailed in Table E1 in the online data supplement. Factor analysis was employed to identify a smaller number of latent constructs assessed by these tests for use as dependent variables in subsequent analysis. The factor analysis was guided by a preliminary principal components analysis, which is used to estimate the number of sizable dimensions in the test correlation matrix. Evaluation of scree plots suggested either four (51% of total variance) or five (58% of total variance) component solutions. The four-component solution was more readily interpretable and so the factor extraction in a principal factors analysis was limited to four and the factors were rotated using varimax criteria.

The rotated factor structure is presented in Table 2. The four factors account for 51% of the common variance in the correlation matrix. Factor loadings greater than 0.35 were considered significant and are indicated in boldface.

The first factor, which accounts for 25% of the common variance, is defined by very high loadings on the learning and delayed recall CVLT scores (0.81) and by the Complex Visual Design standardized recall scores (0.40). The sum of these three variables constitutes a highly reliable index of "declarative memory" (coefficient  = 0.95). The Pursuit Rotor does not contribute to this factor, suggesting little relationship between this measure of procedural learning and declarative memory.

We refer to the second factor, accounting for 10% of the variance, as "signal discrimination," to capture the similarities among fast and accurate Letter Cancellation (0.45), short decision times on the Choice Reaction Time Test (0.48), good signal discrimination of the CPT (0.64), as well as Gilmore- Royer information-processing efficiency scores (0.68). Each of these reflects the ability to differentiate the important aspect of a stimulus from other stimuli. A composite of these four measures has good reliability (coefficient  = 0.70).

The third and fourth factors are smaller, accounting for 9% and 8% of the variance, respectively. WAIS-R digits backward and verbal fluency productivity define Factor Three; both tests are considered measures of "working memory" because they share the need to retrieve information, to hold that information in awareness, and to use that format to guide behavior. Two measures also define Factor Four, "shifting." These are slow performance on Trail Making Test part B/ Trails A, which we interpret to mean slowness in shifting sets, and large numbers of perseverative errors on the WCST, which reflects poor ability to shift attention from one stimulus-relevant concept to another. Because these factors are identified by only two items each, the reliability of the composites is only modest (0.52 and 0.42 respectively).

Measures that did not load substantially on any of the four factors, including the CPT response bias, were excluded from further consideration. Response bias is a measurement of caution exercised when responding to the target. One additional measure was included, vigilance decrement, defined by the difference between performance on the first 2 min and the last 2 min on the CPT. The drop in perceptual sensitivity on a detection task over 10 min was significantly associated with SDB in our prior work (11). This score did not contribute to the signal discrimination factor because of its colinearity with the total d' score.

The relationship of SDB and sleepiness measures to the NP constructs and vigilance, assessed with multiple regression analysis, for the 100 SDB subjects free of health or psychological problems is described in Table 3. Partial correlation coefficients are adjusted for age when age significantly predicted NP level (p < 0.05).

After adjusting for age, "declarative memory" was significantly predicted by level of sleep-associated hypoxemia (r = 0.24, p = 0.02) or by RDI (r = 0.22, p = 0.03) (assessed in separate models). Similarly, after adjusting for age, "signal discrimination" was predicted by either sleep-associated hypoxemia (r = 0.31, p = 0.01) or RDI (r = 0.23, p = 0.02). Sleep hypoxemia and RDI were moderately correlated (r = 0.53); when they both were simultaneously considered in these regression models, only sleep hypoxemia was retained at a p value of < 0.05. "Working memory" also was significantly predicted by RDI (r = 0.22, p = 0.03); however, the relationship with sleep hypoxemia did not reach conventional levels of significance (r = 0.19, p = 0.07). "Shifting" was not significantly predicted by any of the SDB measures.

Sleepiness, assessed by the composite score (capturing objective and subjective sleepiness) significantly predicted vigilance (r = 0.21, p = 0.04). Specifically, less falloff in performance over the testing period was associated with less sleepiness. However, sleepiness did not contribute to the prediction of any of the other NP constructs. Conversely, none of the polysomnographic indices predicted vigilance.

We assessed whether the severity of SDB (as measured by the percentage time in desaturation or by RDI) predicted the NP outcomes in a linear dose-response manner, or whether these relationships were better explained by "threshold" effects. This was evaluated with the use of plots and by incorporating terms for threshold effects (see METHODS). The plots revealed some scatter at low levels of RDI and hypoxemia, but generally decreasing NP functions with increasing percentage time < 90% desaturation or with increasing RDI. None of the models that included threshold terms demonstrated an improved fit over models with no threshold effects.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

After summarizing the NP data into clinically meaningful and psychometrically sound constructs, we used regression analysis to develop models for each NP construct using age and SDB variables to determine which measures of SDB predicted each NP construct, whether such effects were demonstrable across the range of SDB, and whether they were explained by sleepiness. This study has identified a small, but significant association between polysomnographic indices of SDB and a broad range of neuropsychological functions in a large group of volunteers with SDB who were screened to be free of significant psychological conditions or of health conditions other than SDB. Specifically, ability to hold information in awareness, to internalize that structure, and to use that information to guide behavior ("working memory"), retrieval of acquired information after a delay ("declarative memory"), and ability to discriminate among rapidly changing perceptual stimuli ("signal discrimination") were reduced in a roughly dose-dependent fashion with increasing SDB severity (assessed by sleep-associated hypoxemia and/or RDI). Effects were independent of age and not explained by indices of sleepiness.

The only NP constructs (other than vigilance assessed by the CPT) unrelated to polysomnographic indices was the ability to shift from one mental construct to another in planning responses. Our inability to demonstrate an association between SDB and set "shifting" may have related to the low reliability ( coefficient 0.42) by which this construct was measured (in contrast to our measures of "declarative memory" and "signal discrimination," measured with  coefficients of 0.95 and 0.70, respectively). Such measurement error would be anticipated to reduce the ability to accurately model other effects. Improved methods for assessing "shifting" in a generally healthy individuals may help identify the extent to which this NP construct is impacted by SDB.

These latter findings, however, are consistent with those of Naegele and coworkers who found little difference in perseverative errors in 17 apneic patients compared with 17 control subjects (3). Larger differences in perseverative errors on the WCST and functions, which these investigators interpreted to indicate frontal lobe pathology, were demonstrated in subgroups defined by more severe levels of hypoxemia. Given the relatively modest levels of hypoxemia in our sample, we cannot exclude the possibility that greater deficits in "shifting" ability would have been demonstrated within more severely affected patients.

The overall pattern of deficits in our sample was generally consistent with the findings of Bedard and coworkers (2), who studied a smaller, more severe, referral sample. They reported significant deficits in severe apneics on tasks that, in our work, are captured by our more reliable factors, "declarative memory" and "signal discrimination."

There has been much controversy over the extent to which measured deficits in NP function are attributable to the sleepiness that often accompanies this condition rather than to effects more directly related to adverse overnight physiological exposures (e.g., to hypoxemia). Sleepiness, assessed both with subjective and objective sleepiness tests, was, on average, only moderately increased in this sample, and less prominent than in some previous studies of SDB (2, 5). Sleepiness also was measured with only moderate reliability ( coefficient 0.60). As measured and in this range, sleepiness did not significantly predict the NP constructs describing declarative memory, working memory, or signal discrimination. In contrast, indices of sleep-associated hypoxemia or frequency of breathing disturbances, although only modestly abnormal in this sample, accounted for approximately 4% to 6% of the variance in these NP constructs.

Some support for pathophysiological effects on NP functions due to hypoxemia may be derived from the chronic obstructive pulmonary disease (COPD) literature. In untreated COPD (where chronic daytime and nighttime hypoxemia may occur), a variety of NP deficits have been reported in patients with COPD as compared with control subjects (24). A dose-response relationship between level of hypoxemia and NP impairment has been demonstrated (26). However, even patients with "mild" daytime hypoxemia (mean daytime Pa_O2 66) have been reported to have deficits, particularly in set shifting (as measured by Trail Making B) and Digit Symbol substitution (26). There are many possible explanations for these observations, including other comorbidities in COPD. It has been speculated, however, that deficits could be based on cerebral vascular disease. However, chronic and/or intermittent exposure to hypoxemia has been postulated to alter cellular metabolism and gene regulation, contributing to neurodegeneration or other cell injury.

In contrast to the NP constructs above, vigilance, as assessed as ability to sustain attention over a 10-min testing period, was directly, negatively associated with sleepiness, but not explained by polysomnographic indices. Vigilance is generally thought important for maintaining accurate performance in tasks performed over time, especially more monotonous tasks such as driving. Vigilance deficits associated with daytime sleepiness have been described previously in experimental models of sleep deprivation (29). In SDB, the driving simulator tests have been specifically developed to assess this function in patients with SDB, and has been proposed as a clinical tool for assessment and follow-up of patients with SDB (30). The replication here, in a sample of subjects with only moderate levels of sleepiness, supports the sensitivity of this measure and attribute in individuals with a broad range of SDB severity.

Among the indices of SDB severity examined was the arousal index, which was interpreted as an index of sleep fragmentation. This measure, however, did not significantly predict NP functions. Arousals often are more difficult to score than respiratory events (31). Thus, it is not clear whether sleep fragmentation really impacts NP function less than exposures to respiratory events and/or hypoxemia, or if the decreased impact of sleep fragmentation as compared with hypoxemia is due to the poorer reliability of the former measure.

Comparison of the findings from this study to previous studies must recognize differences in the populations studied. A major strength of this study is that it is among the largest reported to date using volunteers who largely were healthy except for their complaints of snoring. Specifically, subjects were screened prior to entry into the study to exclude those with psychiatric or medical histories that might compromise cognitive functioning. The sample also included a high proportion of subjects with mild-moderate SDB, a group in whom relatively little is known about the morbidity of SDB. Some differences among prior studies may be explained by previous selection biases in which subjects most vulnerable to NP morbidity, such as those seeking care at a sleep referral center, are preferentially studied.

A controversial but important area is whether pathological effects or morbidity operate across the entire range of SDB severity, or whether threshold effects exist, below which function is unimpaired. Previous studies have demonstrated measurable negative effects of mild SDB, identified as levels of RDI < 30, on functional status or sleepiness, but with less consistent findings regarding NP functions (6, 7, 32). The present study included subjects with a wide range of SDB, who largely were unreferred for specialty evaluation. Statistical testing did not reveal significant threshold effects regarding level of hypoxemia or RDI. Although this finding might be due to limited power to detect such effects, the data are consistent with small, but measurable adverse effects operating across the range of SDB severity. The findings of measurable effects of SDB across a wide range of SDB severity are consistent with those of Kim and colleagues who demonstrated a linear relationship between the log transformed RDI and a factor interpreted as "psychomotor efficiency" in a population-based sample of middle-aged workers with a mean RDI of 1.2 (8).

These effects on NP function were relatively small and of unclear clinical significance. However, it is reasonable to postulate that SDB effects may vary in the population, with certain subgroups more vulnerable to the effects of hypoxemia or sleep disruption. It may be that knowledge of the full spectrum of SDB patients will allow further understanding of the underlying biological mechanisms that alter brain function and the extent to which interindividual differences in brain reserves, including brain structure and functional plasticity, modulate the response to potentially negative physiological exposures, as experienced with SDB.

Further work needs to be done to answer the question of whether NP tests are useful in identifying individuals who are more likely to experience cognitive deficits with SDB and whether such deficits significantly affect daily functioning. Answers might help in identifying who deserves treatment (as mild to moderate apneics are not covered by Medicare and some insurance plans), who can function better by learning compensation strategies, and under what circumstances individuals should be viewed as limited or disabled.