| [Download PDF]
|Year : 2010 | Volume
| Issue : 47 | Page : 77--87
The state of the art of predicting noise-induced sleep disturbance in field settings
Sanford Fidell1, Barbara Tabachnick2, Karl S Pearsons3,
1 Fidell Associates, Inc., 23139 Erwin Street, Woodland Hills, California, USA
2 California State University, Northridge (Emerita), USA
3 22689 Mulholland Drive, Woodland Hills, CA, USA
Fidell Associates,23139 Erwin Street, Woodland Hills, California 91367
Several relationships between intruding noises (largely aircraft) and sleep disturbance have been inferred from the findings of a handful of field studies. Comparisons of sleep disturbance rates predicted by the various relationships are complicated by inconsistent data collection methods and definitions of predictor variables and predicted quantities. None of the relationships is grounded in theory-based understanding, and some depend on questionable statistical assumptions and analysis procedures. The credibility, generalizability, and utility of sleep disturbance predictions are also limited by small and nonrepresentative samples of test participants, and by restricted (airport-specific and relatively short duration) circumstances of exposure. Although expedient relationships may be the best available, their predictions are of only limited utility for policy analysis and regulatory purposes, because they account for very little variance in the association between environmental noise and sleep disturbance, have characteristically shallow slopes, have not been well validated in field settings, are highly context-dependent, and do not squarely address the roles and relative importance of nonacoustic factors in sleep disturbance. Such relationships offer the appearance more than the substance of precision and objectivity. Truly useful, population-level prediction and genuine understanding of noise-induced sleep disturbance will remain beyond reach for the foreseeable future, until the findings of field studies of broader scope and more sophisticated design become available.
|How to cite this article:|
Fidell S, Tabachnick B, Pearsons KS. The state of the art of predicting noise-induced sleep disturbance in field settings.Noise Health 2010;12:77-87
|How to cite this URL:|
Fidell S, Tabachnick B, Pearsons KS. The state of the art of predicting noise-induced sleep disturbance in field settings. Noise Health [serial online] 2010 [cited 2022 Nov 27 ];12:77-87
Available from: https://www.noiseandhealth.org/text.asp?2010/12/47/77/63207
Despite decades of laboratory and field studies, linkages among environmental noise, sleep disturbance, and public health are not yet well-defined, nor appreciated in sufficient detail to support quantitative, policy-related decision making. Definitions of such disturbance, as well as levels of analysis and predictor variables considered appropriate by sundry parties, vary with their training and interests in research and standardization, and with their commercial, regulatory, and public health policy perspectives.
Some researchers view noise-induced sleep disturbance through a microscope, whereas others prefer a telescope. Basner et al,  for instance, regard electrophysiological measures of neural activity as a "gold standard for investigating noise effects on sleep". This perspective favors "medicalized" interpretations of findings, focused on potential acute health effects on individuals. Relatively little emphasis is placed from this perspective on issues of epidemiological rigor, societal cost/benefit balances, adaptation effects, and economic, regulatory, and political concerns. Others who have investigated effects of noise on sleep indicators such as bodily motility and/or behaviorally confirmed awakenings [2-4] tend to be less concerned with putative individual health effects, but more concerned with the rationale and implications of findings for operational and regulatory purposes.
Acute sleep deprivation has serious and indisputable adverse effects on health. The consequences of less severe, albeit chronic, noise-induced sleep disturbance are far less certain. Muzet,  for example, notes that "…it still seems necessary for some fundamental questions to be answered on whether environmental noise has long-term detrimental effects on health and quality of life and, if so, what these effects are for night-time, noise-exposed populations." Berry and Flindell  find that "...no quantitative link has yet been established between acute or transient sleep disturbance caused by noise and any long term adverse health effects." In a similar vein, Banks and Dinges  assert that "A causal role for reduced sleep duration in adverse health outcomes remains unclear...."
Even though the World Health Organization  considers sleep disturbance itself as a de facto health effect, it is far from certain that physiological responses to noise during sleep have meaningful health consequences. Regardless of whether they adapt to chronic noise intrusions, shifts from one sleep stage to another, as well as slight, transient elevations in heart rate and blood pressure, may be no more than signs of routine autonomic reactions to ever-changing environmental conditions.
Another difficulty in attributing adverse health consequences to noise-related sleep disturbance is that humans are among the most adaptable of species. Common experience indicates that nighttime urban noise environments that initially disturb the sleep of those adapted to sleeping in less densely populated areas may eventually be readily tolerated. By the same token, quiet rural nighttime noise environments may initially increase sleep latency or otherwise impair sleep quality of those adapted to sleeping in higher population density noise environments.
Recent reviews of the noise-induced sleep disturbance literature, such as that of Michaud et al., conclude that findings about noise-induced sleep disturbance differ considerably both with respect to measures of sleep disturbance and by study. They also indicate that nonaircraft related awakenings are more common than aircraft noise-induced awakenings in airport neighborhoods, and that only small percentages of habitually exposed people in familiar sleeping quarters are regularly awakened from sleep by aircraft noise intrusions. Likewise, Banks and Dinges  find that "Habitual sleep duration among adults shows considerable variance within and between individuals"; and "Interindividual variability in sleep and circadian parameters are substantial, and this is equally the case for neurobehavioral and physiological responses to sleep deprivation."
The welter of findings from studies of noise-induced sleep disturbance (a fair amount of which is published only in nonpeer reviewed, gray literature technical reports and conference proceedings) is characterized by data sets and analyses that are difficult to interpret in comparable terms. For example, differences among definitions of arousals and awakenings are commonplace. Definitions of sleep disturbance range from entry from any other electrophysiologically defined state into single analysis epochs of Stage 1 or "waking" states;  to full (that is, behaviorally confirmed) waking consciousness;  to delayed self-reports of awakenings, including those occurring "at least once a month".  Some early studies  considered short duration shifts from any deeper to lighter sleep states as potentially useful indications of noise-induced sleep disturbance. [Endnote 1] More recent studies  analyze longer duration arousals that are nonetheless too short to lead to waking consciousness, cannot be behaviorally confirmed, and are not recalled the next morning. Conversions of actimetrically measured sleep motility into estimates of arousals and awakenings introduce yet more uncertainty into comparisons of findings.
Some ,, have attempted to establish associations between individual noise events and probabilities of noise-induced sleep disturbance, whereas others  have attempted analyses of probabilities of one or more instances of sleep disturbance associated with an entire night's noise intrusions. Others focus on indicators of sleep disturbance ranging from subjective self-report,  to cardiac function,  to biochemical markers of endocrine function. 
A handful of relatively short-term field studies of the sleep behavior of small numbers of paid, self-selected test participants living near a few airports can reasonably yield predictions that are fully generalizable neither to humanity at large, nor even to populations living near airports other than those at which the original observations were made. Even though predictions of sleep disturbance are never validated in field settings to which they are generalized, they are often erroneously treated for environmental assessment purposes as accurate and precise engineering estimates.
Notwithstanding the findings of some laboratory studies of effects of sleep restriction on metabolic processes, , the practical, population-level consequences of specific numbers of chronic noise-induced sleep disturbances remain quite uncertain. For noise environments that are not composed predominantly of discrete events, the uncertainty is greater yet. The complications and cautions noted above, however, have hardly discouraged efforts to develop dosage-effect relationships to link various measures of nighttime noise with sleep disturbance. The likelihood of near-term success in such endeavors, and implications of the various uncertainties noted above for well-informed policy analysis and regulatory action, are discussed below.
Analysis and Discussion
Relevance of dosage-effect relationships for policy analyses
The allure of dosage-effect relationships is that they offer the promise of a "scientific" basis for cost/benefit analyses and policy decisions. A dosage-effect relationship that accounts for the bulk of the variance in a large data set, while displaying an obvious inflection point and a steep slope, might indeed fulfill this promise. As is often the case for environmental pollutants, however, few - if any - such relationships exist between physical measures of noise and their population-level effects on people.
More commonly, purely acoustic measures account for only a small fraction of the variance in associations between noise metrics and noise effects; slopes of such relationships, although significantly different from zero, are often unhelpfully shallow, and unambiguous inflection points cannot be identified. In such cases, dosage-effect relationships add little clarity to policy debate, even when codified in "best practice" standards. In effect, the characteristically shallow slopes of predictive relationships and great variance of the underlying data limit the practical benefit that can be achieved by any level-related regulation.
Yet more fundamentally, no dosage-effect relationship can dictate appropriate regulatory goals. The basic purpose of regulation is to balance conflicting societal interests  as between the conveniences and annoyances of urban living, between demand for air transportation services and potential risks to individual and public health, between desires for habitable airport neighborhoods and commercial benefits of nighttime express air cargo, and so forth.
No quantity that can be measured with a sound level meter can determine the "correct" proportion of the population that should be at risk of any degree of sleep disturbance. Compromises among conflicting societal interests inevitably involve nontechnical value judgments, the charters of regulatory agencies, and reconciliations of political and economic motives. Some political jurisdictions may favor regulation that protects against minor risks, even to the most sensitive proportion of a population, of adverse effects of noise exposure. Others may decide that in return for the benefits of a regulated activity, it is acceptable for some small proportion of the population to be exposed to some degree of risk from that activity, at least for some proportion of the time. Explicit justifications for nontechnical judgments are rare, inherently nonobjective, and usually only notionally based on clear evidence of noise effects. Quantitative rationales for policy positions are thus also rare, and are not amenable to direct comparison when divorced from their political and economic milieux.
Chronology of efforts to develop dosage-effect relationships from field data
Early attempts to develop predictions of sleep disturbance based on acoustic quantities were confined to data produced almost entirely in small-scale laboratory studies. Pearsons et al.  identified 21 of 53 prior studies of noise-induced sleep disturbance, with a total of 167 observation nights, from which it was possible to infer relationships between single event noise metrics and sleep disturbance. They showed that efforts in prior decades to identify dosage-effect relationships , disagreed - by 15 dB - and that earlier laboratory findings overestimated the prevalence of noise-induced sleep disturbance in field settings. Pearsons et al.  also observed that the slopes of relationships intended to predict arousal or awakening in real-world settings were so shallow that they offered no practical guidance for regulatory purposes, and that the absolute prevalence of such sleep disturbance was quite small for noise events at levels of realistic concern in community settings.
Analyses of the data of the first of the large-scale field studies of aircraft noise-induced sleep disturbance  focused on actimetric measurements of sleep quality. Actimetric measurements of arm movements in bed, termed "arousals," were considered as the primary measure of sleep disturbance. Ollerhead et al. estimated from electroencephalographic measurements made on a subset of test participants that 40% of the actimetrically measured arousals during the entire night corresponded to EEG measures of awakening. They found no direct relation between arousals and awakenings, however.
Ollerhead et al. also measured outdoor sound exposure and maximum levels of outdoor noise events, but made no noise measurements inside sleeping quarters. Further, only one of the four UK airports studied had extensive nighttime aircraft operations. In subsequent efforts to compare the findings of Ollerhead et al. with those of other sleep studies in which noise levels inside sleeping quarters were measured, an outdoor-indoor noise reduction of 15 dB was found to yield a reasonable agreement.
Fidell et al. attempted to use indoor sound exposure levels to account for the behavioral awakening findings of six field studies described by Pearsons et al., those of Ollerhead et al.,  and those made in 45 households near two airports and control areas. Bivariate regression found a weak but reliable linear relationship between indoor sound exposure levels and awakenings occurring within 5 min of noise intrusions. The relationship accounted for less than a third of the variance in the awakening data. [Endnote 2]
Adding the observations of Fidell et al.  to those of prior behavioral awakening studies yielded a linear relationship between indoor sound exposure levels and the prevalence of awakening that accounted for only a fifth of the variance. Subsequent reworkings of these data  that have taken into consideration the findings of later field studies  have yet to produce relationships with markedly steeper slopes, or obvious inflection points, or that can account for much more variance. In an informative Annex, ANSI's recent  standard describes an alternate approach to predicting sleep disturbance,  discussed at length in the following sections.
Difficulties of direct comparisons among predictions of noise-induced sleep disturbance
[Table 1] summarizes half a dozen published relationships between indoor sound exposure levels and predicted awakening. [Endnote 3] Over the range of indoor sound exposure levels from 50 to 90 dB, the product-moment correlation between the earliest  and most recent  of these predicted awakening rates is 0.98. This near-perfect correlation is more reasonably viewed as an artifact of the limited amount of original data that has been repeatedly reanalyzed than any sort of confirmation of the reliability of the predictions.
Differences among definitions of sleep disturbance and of predictor and predicted variables complicate direct comparisons among the relationships shown in [Table 1] and those of Passchier-Vermeer et al., Basner et al., and the European Commission.  The relationship published by Passchier-Vermeer et al. (0.000532(SEL - 38) + 0.0000268 (SEL - 38)2) predicts motility, rather than awakening; Basner et al. base their relationship (1.894(.001Amax 2 ) + 4.008(.01Amax ) - 3.3243) on maximum A-weighted sound pressure levels of noise events rather than indoor sound exposure levels; and the European Commission's relationship (0.3504(10 (Lnight - 35.2)/10) ) predicts the maximum number of awakenings from nighttime equivalent levels rather than from any measure of individual noise event levels.
Additional factors further complicate comparisons among findings of individual studies. For example, behavioral definitions of sleep disturbance commonly include a "grace period" following a noise intrusion during which a response is considered to be attributable to noise. The duration of the grace period can vary from seconds to minutes from study to study. Thresholds of durations and other means of discriminating aircraft from nonaircraft noise events relied upon by noise event classification software likewise vary from study to study. Further, while most attempt to predict the prevalence of awakenings (in percent) from measures of indoor sound exposure levels of individual noise intrusions, others use maximum A-weighted sound levels  or even recommend use of cumulative nighttime noise exposure. 
Most predictions of awakening prevalence rates assume simple transformations from probabilities of awakening given Sound Exposure Level (SEL) values of individual events to percentages of the population awakened, based in part on a tacit assumption of independence of awakenings from one another. Brink et al., among others, question the assumption. Some of the relationships in [Table 1] are not dosage-effect relationships at all, but only estimates of maximum numbers or percentages of awakenings or arousals observed in combined data sets. Some relationships are based on field observations only, whereas others include laboratory data. Some attempt to predict "arousal or awakening," where arousal implies only shifts from deeper to lighter sleep states, rather than waking consciousness.
Notwithstanding these differences, predictions of noise-induced sleep disturbance generally indicate that noise intrusions only occasionally disturb sleep. Most of the dosage-effect relationships based on behavioral awakening field data, for example, predict that the prevalence of awakening does not exceed 5% until indoor sound exposure levels of intruding noises exceed 90 dB. [Endnote 4] (Corresponding outdoor levels of aircraft noise events may be as much as 2 orders of magnitude greater.)
Role of complexity of sleep behavior in interpretation of findings
Part of the difficulty in attributing sleep disturbance to noise is that sleep is a complex process whose quality can be influenced by many environmental and other factors. People routinely cycle multiple times per night through half a dozen electrophysiologically distinguishable sleep stages. Frequent shifts in sleep states throughout the night complicate unambiguous attribution of such shifts exclusively to temporally proximal measures of noise intrusions of outdoor origin into sleeping quarters.
Depending on the definition adopted for "awakening," people may awaken for reasons having nothing to do with noise many times per night, at moments which may or may not closely coincide in time with the occurrence of noise events. According to Basner et al.,  people exhibit an average of 21 electrophysiologically detectable arousals per hour of sleep, or about 144 spontaneous arousals per night. Counting both shifts from deeper to lighter sleep states and momentary awakenings, Ollerhead et al.  reported about 45 "awakenings or arousals" per night, of which only 40% were thought to represent even momentary awakenings.  People commonly attain full waking consciousness two or three times per night for reasons having nothing to do with noise exposure. 
Noise intrusions into sleeping quarters that occur at random times during the night with respect to sleep states may coincide temporally with ongoing but noncausally related shifts from "deeper" to "lighter" sleep stages, or from "lighter" to "deeper" sleep stages. The latter shifts are rarely closely analyzed in studies of the adverse effects of noise intrusions on sleep quality, even though the ratio of numbers of shifts from deeper to lighter sleep stages to numbers of shifts from lighter to deeper sleep stages is clearly an important detail for making sense of supposedly noise-induced sleep disturbance.
Acoustic and other circumstantial factors that may affect sleep quality are generally site-dependent in field settings. Some acoustic factors that have been reported as affecting sleep include onset rates, maximum levels, and impulsiveness of outdoor noise events; numbers and detailed temporal distributions of noise intrusions throughout the night; and ambient noise levels. Other variables which are effectively uncontrollable in field settings, but which can potentially influence sleep quality either independently from or in conjunction with acoustic variables, include indoor noise environments, stable individual differences in gender and age, time of retiring, and total time in bed, as well as time-varying differences within and between subjects (and of others in their households) in health status, medication, fatigue levels, familiarity with intruding noises, and the like.
Limited utility of sound level-based dosage-effect relationships
No yet-developed dosage-effect relationship which relies solely on indoor sound exposure or maximum sound levels as predictors of sleep disturbance offers substantial useful guidance for policy or regulatory purposes. The underlying problems, as noted above, are that noise is only one among multiple determinants of sleep disturbance, sleep is too complex a process for noise level alone to suffice as a comprehensive predictor of disturbance, and the slopes and shapes of the relationships do not usefully inform policy discussions.
Perhaps the major obstacle to reliable predictions of awakenings from absolute noise level information alone is that levels of intruding noise events and awakenings simply are not strongly related to one another. The point biserial correlation between indoor SEL values and awakening responses in the Fidell et al. three-airport data set is only r = 0.048. This correlation is not merely small, but also smaller than the correlation between awakening responses and ambient noise levels immediately prior to aircraft noise events that elicited awakenings (r = 0.08), and even smaller than the correlation between a completely nonacoustic variable (time since retiring) and awakening responses (r = 0.11).
In the Fidell et al. data set, the average indoor sound exposure level that failed to awaken test participants was 70.5 dB, whereas the average indoor sound exposure level associated with awakenings was not even 3 dB greater (73.2 dB). [Endnote 5] Further, the 50-dB range of indoor sound exposure levels that failed to awaken sleepers at the three airports completely encompassed the less than 40 dB range of sound exposure levels that did awaken sleepers. [Figure 1] shows the overlap in distributions of indoor sound exposure levels that did and did not awaken sleepers. [Endnote 6] Factors other than absolute sound levels of noise intrusions must obviously affect sleep disturbance in familiar sleeping quarters more greatly than sound levels alone.
Complications for modeling of sleep disturbance
Data from studies of behavioral indications of sleep disturbance such as awakenings and motility do not lend themselves to simplified modeling and straightforward interpretation for several reasons. Both the within- and between-subjects variances of such data are typically large; nontrivial percentages of test participants are never behaviorally awakened by noise intrusions; bona fide, noise-induced awakenings are relatively rare events (at least with respect to numbers of spontaneous awakenings and of nighttime noise intrusions); associations between aircraft overflights and indoor noise levels in sleeping quarters are far from perfect (and can vary seasonally within and between homes); and awakenings are hardly independent of one another, nor of time since retiring.
The net effect of these complications is to reduce the overall credibility of simplified analyses. For example, some analyses may require strong (if not contrafactual) assumptions about the independence of observations from one another. It may also be necessary for some analyses (v.i.) to completely exclude data from nonresponsive test participants, thereby biasing eventual estimates of prevalence rates of awakening in the population at large.
Limitations of an alternative approach to deriving dosage-effect relationships
Recognizing some of the obstacles noted above to predicting sleep disturbance on a per-noise event basis, Anderson and Miller  have recently suggested that a pragmatically useful alternative is to predict the probability of occurrence of at least one awakening per night associated with an entire night's (aircraft) noise exposure. The logic and statistical assumptions of the analyses supporting the suggestion warrant examination in some detail, [Endnote 7] since the applicability of the approach and the utility of its predictions in specific settings require a careful understanding. Given that people routinely arouse and achieve waking consciousness several times per night for reasons unrelated to aircraft noise, some consideration of the practical implications of predicting at least one nightly noise-related awakening is also warranted.
Anderson and Miller's approach to predicting at least one behaviorally confirmed, aircraft-induced awakening per night begins with a logistic regression based on three predictors of per-event awakenings: (1) the indoor sound exposure level of an overflight, (2) the time of occurrence of the noise event with respect to the time since the test participant retired, and (3) a post hoc index of subject "sensitivity."
By itself, the indoor SEL of a noise event is the least effective of the three predictors of behavioral awakening. [Endnote 8] [Table 2] shows the difference in SEL values of noise events associated with awakening responses ("Yes") and those not associated with awakenings ("No"). The mean difference of 2.7 dB is considerably smaller than the standard deviation around either mean level.
Logistic regression indicates that although awakening is significantly related to SEL (Wald z = 16.75, PR 2 = .012) reveals that the SEL of a noise event is almost completely ineffective as a predictor of awakening. (The statistical significance of the prediction is due to the large number of noise events in the data set.) Randomizing the SELs associated with noise events that elicited and failed to elicit awakening responses in the data set - or for that matter, omitting the SEL of noise events entirely from the prediction equation - has no meaningful effect on the variance accounted for by the prediction model.
The time at which an aircraft overflight occurs with respect to the time that test participants retired for the night is a more effective predictor of behavioral awakening than the SEL of an intruding noise event. Logistic regression shows that waking is significantly (albeit weakly) related to time since retiring (Wald z = 46.59, PR 2 = .033). [Table 3] shows the differences in time since retiring between events that elicited and failed to elicit awakening responses. Despite the large standard deviations, noise events that elicit awakenings occur, on average, about an hour and three-quarters later at night than events that do not elicit awakenings.
The distribution of aircraft noise events over the course of a night's sleep in the data set happens to be highly positively skewed. As seen in [Figure 2]Έ most nighttime noise events at the airports analyzed by Anderson and Miller occurred shortly after test subjects retired, with another, smaller, increase shortly before they arose. A square-root transform of time since retiring normalizes the distribution of time of occurrences of aircraft noise events somewhat, leading to stronger prediction. Logistic regression on the transformed time of occurrence of noise events yields a Wald z = 56.26, PR 2 = .044.
Time of occurrence of nighttime noise events can vary from airport to airport, however. Nighttime operations at large hub (primarily passenger) airports rarely exceed 10% of all operations. At express cargo and military airports, both the percentages and times of occurrence of nighttime operations can differ greatly. Applying predictions made on the basis of awakening responses observed at a small number of airports to airports with different nighttime operating conditions could lead to appreciable errors of estimate.
The sensitivity of predictions of aircraft noise-induced sleep disturbance to airport operating schedules implies that simple and universal prediction of sleep disturbance is unlikely. The alternative for environmental impact disclosure purposes (custom-tailored analyses, often based on assumptions about hypothetical operating schedules years in the future) is costly both in economic terms and in terms of the ratio of assumptions to predictions.
Anderson and Miller refer to the most effective predictor of awakening in the data set as "subject sensitivity," scaled in logit units derived from logistic regression in which each subject was coded to reflect individual differences. Logistic regression shows that awakening is significantly related to these subject-by-subject individual differences, categorized as 32 dummy variables (Wald z = 169.77, PR 2 = .126.) [Table 4] shows differences in "subject sensitivity" between those awakened by a given aircraft overflight vs. those not awakened. Not surprisingly, subjects who were classified as more "sensitive" were more likely to be awakened by any aircraft overflight, regardless of its sound exposure level. (In signal detection terms, this is akin to achieving a high hit rate as a byproduct of adopting a high false alarm rate.)
The very high correlation between the logit-scaled measure of sensitivity and the logarithm of button-pushing rate (r31 = .911, Pprior to observation of the responses of test subjects, post hoc definitions of individual differences among test subjects amounts to little more than using the data to predict the data, or to devising another name for error variance. (In analysis of variance terms, the strategy of dummy coding for "subject sensitivity" removes between-subject variance from a generalized error term simply by reclassifying it from error variance to an effect of interest.)
The logic of predicting sleep disturbance from unspecified individual differences, scaled in logit or any other units, scarcely differs from predicting sleep disturbance on any other nonexplanatory form of individual difference: height, weight, hair color, political preference, street address, etc. The statistical machinery "works" (in the sense of capitalizing on chance), but must inevitably be readjusted to reflect unexplainable differences in the next data set to which it is applied. Including individual differences as the strongest predictor variable in a regression equation is tantamount to asserting that the probability of awakening has (much) more to do with the particular set of self-selected subjects who happen to participate in a data collection exercise, than with any unique or measurable property of a set of aircraft operations.
Large unexplained differences are apparent among the mean sensitivities of test subjects at the three airports in the data set. [Figure 3] shows that the mean sensitivity of test subjects at Denver International Airport (DIA) (in logit-scaled units, M = -0.09, SE = 0.03) is far greater and significantly higher (F2, 7136 = 395.6. P h2 = .28, with 95% confidence limits from .22 to .34) than at Los Angeles International Airport (M = -0.97, SE = 0.02) and Castle Air Force Base (AFB) (M = -0.95, SE = 0.02). It is unclear why distributions of subject sensitivity differ so greatly from airport to airport, and whether relationships inferred from small and nonrepresentative samples at these three airports can be generalized to distributions of subject sensitivity at airports elsewhere.
Since post hoc definitions of sensitivity cannot reasonably serve as a priori predictor variables, event SEL and time since retiring are the only remaining predictors of behavioral awakening. In a new view of the Fidell et al. ) data, [Figure 4] and [Figure 5] show that little helpful information is available from either of these predictors. Probability of awakening increases from zero to only about 0.15 over a range of SEL values greater than 60 dB - a factor of more than 1,000,000:1 in sound energy. The probability of awakening likewise increases from zero to only about 0.15 over the bulk of time spent in bed. If more discriminating analyses of noise-induced sleep disturbance are desired for environmental impact predictions, cost/benefit analyses, policy decisions, and regulatory action, they will clearly have to take into consideration factors more closely linked to sleep disturbance than those investigated to date, and more substantial than individual differences among test subjects.
Implications of uncertainty about noise-induced sleep disturbance
Given the great and unresolved uncertainty about causal relationships between noise-induced sleep disturbance and its potential individual and public health consequences, claims that a medicalized perspective implies the need for stringent regulatory action are tenuous. Some have argued in similar situations that uncertainty justifies exigent action prior to the development of quantitative understanding of the threat, if any, of noise effects. For example, Berglund et al.  argue for "concerned action" to protect individuals and the public from low-frequency noise exposure, even without clear evidence of adverse health effects at common exposure levels.
Such nonevidentiary arguments are sometimes based on repeated anecdotal accounts of subclinical effects. They may also be rooted more in speculation and philosophy than in direct or rigorous epidemiological findings. Clark and Stansfeld,  for example, conclude that health effects of noise exposure "may be of importance given the number of people increasingly exposed to environmental noise and the chronic nature of exposure." Even more starkly, Ising and Kruppa  assert that "In the case of preventive health protection ... any reasonable assumption of a possible health hazard justifies protective measures." (emphasis added).
If, indeed, urgent action is believed to be essential to protect public health from potential adverse consequences of nighttime noise, it is preferable given the current poor understanding of functional relationships between nighttime noise, sleep disturbance, and health consequences to do so on expressly nontechnical grounds. Any rationale for such policy or regulatory action should also acknowledge the meaningless nature of artificially precise estimates of numbers of people awakened by aircraft noise, the absence of a compelling and generalizable technical rationale for such action, and the hidden costs of technically unjustifiable regulation.
Recommendations for next generation of field studies
Considering the variety of opinions about appropriate ways to characterize sleep disturbance, the limited stock of large-scale field studies of noise-induced sleep disturbance, difficulties of direct comparisons of results of the few that have been completed, the resultant absence of clearly interpretable dosage-effect relationships, and major uncertainties about putative risks of noise-induced sleep disturbance to public health and welfare, it is useful to consider ways in which future field studies of noise-induced sleep disturbance can improve upon studies conducted to date. Such improvements extend from more detailed and comprehensive measurement and processing of noise level information, to larger scale and longer duration studies, to more sophisticated study designs.
Although absolute levels of noises intruding into familiar sleeping quarters are by themselves inadequate to yield precise predictions of sleep disturbance, they do account for some variance in awakening data, and might account for more if information about noise intrusions were captured, processed, and analyzed in greater detail. For example, in the Fidell et al. data set, the differences between indoor SELs of sounds that did and did not awaken test participants was less than 3 dB, but the difference in A-weighted signal to noise ratios between noise events that did and did not disturb sleep [Endnote 8] was 5 dB greater.
This observation suggests that audibility (bandwidth-adjusted signal-to-noise ratio) might account for more variance as a predictor of sleep disturbance than A-weighted, absolute level alone. It also suggests the utility of more detailed measurements of noise environments in sleeping quarters, such as logging of continuous time series of rapidly sampled (say, one-half second), one-third octave band levels, time-synchronized with event-based measurements of outdoor noise levels.
Some other obvious improvements in field study designs include a wider range of noise sources and exposure conditions. These should include studies of spontaneous awakening and arousal rates in areas with few noise intrusions of outdoor origin (but perhaps more of indoor origin) to aid in the interpretation of observations made in noisy urban areas. Studies of longer durations would also be valuable, as would be studies with larger numbers of subjects, and more sophisticated study designs. The latter should include repeated observations of the behavior of the same subjects months apart in time, so that they can serve as their own controls and provide information about the replicability of findings and the time course of adaptation to naturally occurring changes in exposure conditions.
Large-scale field studies of noise-induced sleep disturbance to date have been wholly adventitious in character, in the sense that experimenters exerted no control over the level, timing, diversity, novelty, or meaning of noise intrusions in sleeping quarters. Given the meager associations observed to date between physical properties of naturally occurring noise intrusions and sleep disturbance, however, it could be more fruitful to consider study designs that introduce some degree of intentional control of exposure into home sleeping quarters.
For example, adaptive designs might be attempted in which temporal distributions of intruding noises are varied from week to week to permit closer examination of the effects on awakenings of times of occurrence of noise intrusions. On a longer time scale, longitudinal studies might be attempted in which the exterior-interior noise reduction of sleeping quarters was increased after collection of baseline data. It is conceivable, given the present state of knowledge, that the benefit of an additional 5 dB of noise reduction in sleeping quarters could well be negligible in terms of reduced sleep disturbance.
Similarly, experimenters might vary continuous background noise levels in sleeping quarters on some nights to determine the effects of audibility on probabilities of awakening attributed to noises of external origin. Additional noise events at levels dependent on the numbers of prior noise events of the same level could be introduced into sleeping quarters to permit careful study of the degree of independence of awakening responses from one another.
To estimate the magnitude of the effect of meaning of noise intrusions on probability of awakening, subjects could be penalized (offered a smaller bonus) for awakening responses following the occurrence of an experimenter-defined set of noise events, but rewarded (offered a larger bonus) for awakening responses following the occurrence of a different set of noise events. If the likelihood of awakenings associated with sounds considered relevant by sleepers demonstrably exceeds the likelihood of awakenings associated with higher level but irrelevant sounds, the role of purely acoustic analyses of noise-induced sleep disturbance in regulatory policy will clearly have to be reexamined.
Epidemiological evidence does not yet support either reliable prediction of noise-induced sleep disturbance, or well-informed policy debate, much less a plausible technical rationale for regulatory action. The practical, population-level implications of noise-induced sleep disturbance and its consequences remain poorly understood due to design and other limitations of field studies of noise-induced sleep disturbance already undertaken, and to limitations of the statistical analyses performed to date. Published relationships used to assess the probability or prevalence of noise-induced awakening remain highly uncertain and unhelpfully imprecise. Considerable caution must be exercised in extrapolating conclusions about sleep disturbance that have been inferred from the behavior of relatively small and purposive samples of people living near a few airports to wider populations.
Additional findings from large-scale, long duration field studies of the effects of a wide range of environmental noise exposure on behaviorally confirmed awakenings could improve understanding of relationships between noise and sleep disturbance. It is doubtful, however, that further analyses of the results of studies that are similar in design to those already conducted will meaningfully improve understanding of noise-induced sleep disturbance. New analytic approaches must systematically account for nonacoustic factors such as the source and meaning of noise intrusions and sleepers' familiarity with them, and must provide a context for distinguishing between incidence rates of spontaneous (nonnoise related) and prevalence rates of bona fide noise-induced sleep disturbance.
The authors are grateful to Dr. Grant Anderson and to Nicholas Miller for providing access to a spreadsheet containing the information used for the regression analysis in their 2007 publication, and to Drs. Paul Schomer and Kevin Shepherd for their comments on earlier drafts of this article.
Pearsons et al., (1974) also adopted a more stringent criterion for interpreting electroencephalographically defined awakenings, which yielded an estimate of an average awakening rate of only three per night.As noted by Fidell et al., bivariate regression tends to account for more variance than logistic regression in this application due to the greater variability inherent in dealing with individual events rather than with aggregations of events.Most of the relationships yield plausible predictions only within restricted ranges of sound exposure levels.Such agreement among various predictions is somewhat artificial. The bulk of the data on which the various predictions are based all comes from just a few field studies; few of the relationships are based on different compendia of original data; and some of the relationships are simply trivially variant curve fits, based on the authors' preferences for linear or curvilinear fitting functions.The difference in corresponding values of maximum indoor sound levels was smaller yet - only 1.8 dB.Note the great vertical exaggeration of the ordinate used to plot the distribution of sound exposure levels that awakened test participants (in the upper panel of [Figure 1]) with respect to the ordinate used to plot sound exposure levels of noise intrusions that did not awaken subjects (in the lower panel of [Figure 1]). A total of 7,139 noise intrusions identifiable from outdoor noise measurements intruded into sleeping quarters in this data set. Only 157 (2.2%) of these potential challenges to sleep could be associated with behavioral awakening responses.Anderson and Miller note that "In all, this paper's assumption of independence is not fully persuasive."Indoor ambient noise levels immediately preceding the occurrence of a noise event support better prediction of behavioral awakening than the SEL of the noise events themselves. Logistic regression on the difference between the SEL of a noise event and the SEL of the ambient noise preceding the event yields a statistically significant prediction (Wald z = 51.37, P
The average A-weighted indoor ambient level immediately preceding noise events that did not awaken test participants was 40.3 dB, whereas the corresponding ambient level preceding noise events that did awaken test participants was 34.4 dB. The difference in noise event-to-ambient levels was 30.2 dB for the sounds that did not awaken sleepers, but 38.8 dB for the sounds that did awaken them. Although the A-weighted measure of masking is crude, the potential benefit of an additional 8 dB of masking noise in protecting against sleep disturbance from noise intrusions merits further investigation.
|1||Basner M, Mόller U, Elmenhorst EM, Kluge G, Griefahn B. Aircraft noise effects on sleep: A systematic comparison of EEG awakenings and automatically detected cardiac activations. Physiol Measurement 2008;29:1089-103.|
|2||Fidell S, Pearsons K, Tabachnick BG, Howe R. Effects on sleep disturbance of changes in aircraft noise near three airports. J Acoust Soc Am 2000;107:2535-47.|
|3||Passchier-Vermeer W, Vos H, Steenbekkers J, van der Ploeg F, Groothuis-Oudshoorn K. Sleep disturbance and aircraft noise exposure: Exposure-effect relationships. TNO Inro Report 2002;27:1-245.|
|4||Cole RJ, Kripke DF, Gruen W, Mullaney DJ, Gillin JC. Automatic sleep/wake identification from wrist activity. Sleep 1992;15:461-9.|
|5||Muzet A. Environmental noise, sleep and health. Sleep Med Rev 2007;11:135-42.|
|6||Berry B, Flindell I. Estimating dose-response relationships between noise exposure and human health impacts in the UK. BEL Report 2009-002, Available from: http://www.defra.gov.uk/environment/quality/noise/igcb/documents/tech-report.pdf . [cited in 2009].|
|7||Banks S, Dinges D. Behavioral and physiological consequences of sleep restriction in humans. J Clin Sleep Med 2007;3:519-28.|
|8||World Health Organization. Night noise guidelines for Europe, WHO Regional Office for Europe, Copenhagen. Available from: ( http:///www.euro.who.int/InformationSources/Publications./Catalogue200904_12 ). [cited in 2009].|
|9||Michaud DS, Fidell S, Pearsons K, Campbell KC, Keith SE. Review of field studies of aircraft noise-induced sleep disturbance. J Acoust Soc Am 2007;121:32-41.|
|10||Rechtschaffen A, Kales A, Berger R, Dement W, Jacobsen A, Johnson L, et al. A manual of standardized terminology, techniques and scoring system for sleep stages of human subjects. Public Health Service. Washington, D.C.: U.S. Government, Printing Office; 1968.|
|11||Pearsons K, Barber D, Tabachnick BG, Fidell S. Predicting noise-induced sleep disturbance. J Acoust Soc Am 1995;97:331-8.|
|12||van den Berg F. Perspectives on wind turbine noise. Echoes (Acoust Soc Am) 2009;19:1,3.|
|13||Pearsons K, Fidell S, Globus G, Friedmann J, Cohen H. The effects of aircraft noise on sleep electrophysiology as recorded in the home. BBN Report 2422, NASA Contractor Report NAS1-9559-21, Langley, VA; 1973.|
|14||Basner M, Samel A, Isermann U. Aircraft noise effects on sleep: Application of the results of a large polysomnogrpahic field study. J Acoust Soc Am 2006;119:2772-84.|
|15||American National Standards Institute (ANSI). Quantities and procedures for description and measurement of environmental sound-Part 6: Methods for estimation of awakenings associated with outdoor noise events heard in homes. ANSI S12.9-2000/Part 6; 2008.|
|16||Anderson G, Miller N. Alternative analysis of sleep-awakening data. Noise Control Engg J 2007;55:224-45.|
|17||de Kluizenaar Y, Janssen SA, van Lenthe FJ, Miedema HM, Mackenbach JP. Long-term road traffic noise exposure is associated with an increase in morning tiredness. J Acoust Soc Am 2009;126:626-33.|
|18||Clark C, Stansfeld S. The effect of transportation noise on health and cognitive development: A review of recent evidence. Int J Comp Psychol 2007;20:145-58.|
|19||Ising H, Kruppa B. Health effects caused by noise: Evidence in the literature from the past 25 years. Noise Health 2004;6:5-13.|
|20||Fidell S. Assessment of the effectiveness of aircraft noise regulation. Noise and Health, 1999;3:17-25.|
|21||Lukas J. Measures of noise level: Their relative accuracy in predicting objective and subjective response to noise during sleep. EPA-600/1-77-010, Office of Health and Ecologic Effects, Office of Research and Development. Washington, D.C.: U.S. Environmental Protection Agency; 1977.|
|22||Griefahn B. Research on noise-disturbed sleep since 1973. Proceedings of the Third International Congress on Noise as a Public Health Problem. Freiburg: ASHA Report 10; 1980.|
|23||Ollerhead JB, Jones CJ, Cadoux RE, Woodley A, Atkinson BJ, Horne JA, et al. Report of a field study of aircraft noise and sleep disturbance. London: Department of Safety, Environment and Engineering, Civil Aviation Authority; 1992.|
|24||Passchier-Vermeer. Night-time noise events and awakening, TNO Inro Report 2003;32:1-61.|
|25||Cox P, Palou J. Directive 2002/49/EC of the European Parliament and of the Council of 25 June 2002 relating to the assessment and management of environmental noise. Annex I, OJ:189 18.7.2002; 2002. p. 12.|
|26||Brink M, Wirth K, Schierz C. Effects of early morning aircraft overflights on sleep and implications for policy making. Euronoise. Finland: Tampere; 2006.|
|27||Berglund B, Hassmιn P, Job RF. Sources and effects of low-frequency noise. J Acoust Soc Am 1996;99:2985-3002.|