There is increasing epidemiological research demonstrating the negative effects of transportation noise exposure on various chronic diseases, such as cardiovascular disease [1
], metabolic syndrome [5
], depression [10
], and cognitive functions [13
]. Several mechanism are implicated in these negative noise effects, such as activation of the hypothalamic–pituitary–adrenal axis (HPA), which leads to increased cortisol and glucose levels as well as increased blood pressure, with consequences for blood viscosity and blood coagulation [17
]. Also, chronic sleep deprivation is a stressor, which contributes to allostatic load and has been found to be connected to all these outcomes above. High allostatic load, characterized by repeated stress responses, affects the brain regions involved in memory consolidation and affective processing, with potential long-term effects on cognitive and mental health [18
]. In large population studies, impaired sleep quantity and quality has been associated with increased risk of developing coronary heart diseases [19
]. There is also solid evidence that short sleep duration, lack of slow-wave sleep, and circadian desynchronization of sleep increases sensitivity to food stimuli, contributing to adverse metabolic traits, in particular obesity and type 2 diabetes [20
]. Further, sleep deprivation reduces the motivation for physical activity and thus reduces the energy expenditure [21
], which further contributes to the risk of cardiometabolic syndromes.
Noise-induced sleep disturbance, as demonstrated in experimental human laboratory studies [22
], field trials [25
], and observational epidemiological studies [26
], are thus likely to be on the pathway for detrimental effects on the cardiometabolic system and mental health. Strikingly, evidence for noise effects on various sleep outcomes was only considered “very low” to “moderate” in the recent systematic review of the World Health Organization Environmental Noise Guidelines for the European region [28
]. Thus, many questions remain open.
Sound pressure level at the ear of the sleeper is often regarded as the only relevant entity for quantifying noise effects on sleep. It is possible to quantify reactions of sleepers to sound pressure level at the ear in field experiments using contrived exposure settings (i.e., reproducing noise in a controlled fashion with loudspeakers). This approach, however, does not adequately consider long-term habituation to a noise source, which may be relevant in a real-life situation at home. Observational studies would be appropriate to account for potential habituation effects, but they are usually based on noise exposure modeling for the outdoor façade and are thus rather imprecise regarding noise exposure at the ear.
Further, the average sound pressure level (Leq
and similar indicators) may not be the only relevant factor; other noise exposure characteristics not captured in energy-based exposure indicators may also be important. For instance, in experimental sleep studies of noise effects on sleep, different effects have been observed for road, rail, and aircraft noise [22
]. Thus, the effects of noise on sleep might be better predicted by the number of noise events [29
], the maximum sound pressure level [30
], the sound pressure level slope [24
], or the order of events [31
]. Confronted with the challenge of how to sum up and weight noise events, we developed an acoustical metric, the intermittency ratio (IR), to characterize short-term temporal variations of transportation noise exposure [32
]. In a large cohort study, we found some evidence that IR may have a modifying effect on the cardiovascular mortality risk [33
Timing of noise exposure is also considered to be relevant for sleep effects. For instance, an experimental sleep study observed that noise curfews at the end of the night were most beneficial for sleep because noise-induced sleep disturbances at the beginning of the night were at least partly compensated during the rest of the night [34
In order to quantify relevant factors that affect sleep quality through noise exposure in a real-life situation, we conducted an observational field study with volunteers wearing wrist actimeters to record their sleep–wake behavior during seven days with concurrent indoor and outdoor noise exposure measurements. The study explored (i) the relevance of indoor noise compared to outdoor noise, (ii) the predictive contribution of IR to sleep effects in addition to equivalent continuous sound pressure levels (Leq,night), and (iii) the effect of noise exposure at different times during the night. We also tested potential effect modification by noise annoyance, noise sensitivity, and sex.
For the study, we included 107 individuals from 96 households, resulting in 720 nights with complete noise exposure data and recorded data on at least one sleep outcome (actimeter-derived or self-reported). Data from two individuals (14 nights) were excluded because their sleep was affected by their children. An additional 10 nights were excluded from the dataset because of sleeping out of home, two nights due to acute respiratory infection, and nine nights due to a recorded sleep duration of less than four hours. This left 694 nights from 105 individuals from 94 households for further analyses, although the number of observations were somewhat smaller for specific outcomes because actimetry data was not available for two individuals and self-reported sleep quality data was not available for two individuals.
The mean age of the study participants was 52.1 years (SD = 14.4 years, age range: 23 to 78 years). Fifty-three participants (51%) were female and 52 were male. Twenty-nine individuals (28%) had a university degree, 34 (32%) had a higher education, 39 (37%) had an apprenticeship, and 3 (3%) had compulsory education. Median noise annoyance was 6, with 23% being classified as highly annoyed (score ≥8). Forty-one subjects (39%) tended to agree (score 4–6) to the statement “I am noise-sensitive”.
Average sleep efficiency as recorded by actimetry was 88%, and average sleep duration was 7.0 h (Table 1
). Self-reported sleep quality scores ranged from 4 to 100, with a mean of 65. Average self-reported sleepiness score was 4.1. Sleep efficiency was strongly negatively correlated with sleep latency (−0.79), and self-reported sleep quality was negatively correlated with self-reported sleepiness (−0.49). Correlations of all other outcomes were low (see Supplementary Table S1
shows the summary estimates for the various noise exposure metrics. Mean measured nighttime noise level (Leq,night
) outside the window of the study participants’ bedroom was 47.0 dB(A), with a maximum of 62.7 dB(A). Estimated mean indoor nighttime noise level at the pillow of the study participants was 30.2 dB(A) (maximum: 55.3 dB(A)). Measured exposure was about 10 dB(A) lower during 01:00 to 05:00 compared to the beginning and end of the night. Correlation between estimated indoor and measured outdoor nighttime exposure was 0.46, reflecting variation in noise attenuation between study participants. Correlations were ≥0.63 for measured Leq
between different time periods of the night (see Supplementary Table S3
Measured nighttime noise (Leq,night
) tended to be negatively associated with sleep efficiency and positively associated with sleep latency, although they were not statistically significant (Table 3
). For instance, sleep efficiency decreased by 1.11% (95% confidence interval (CI): −2.44% to 0.21%) and sleep latency increased by 5.67 min (95% CI: −1.00 to 12.34) per 10 dB(A) increase in Leq,night
. No indications of an association (p
≥ 0.27) were seen for the other sleep outcomes, including self-reported sleep quality and sleepiness. This association pattern was also found in the raw data (Figure S1
Using estimated indoor noise instead of measured outdoor noise yielded similar results in terms of regression coefficients, but none of the associations was even close to significance (Table 4
). Similarly, we did not obtain any indications that IR was associated with any of the objectively recorded or self-reported sleep outcomes when also considering Leq,night
in the same model (see Supplementary Table S4
). A model with quartiles of IRnight
instead of a continuous variable suggested a nonlinear association for sleep efficiency and latency. For the second quartile of IRnight
(50%–63%), sleep efficiency was reduced by 1.24% (95% CI: −2.73 to 0.25), and sleep latency was increased by 3.48 min (95% CI: −2.54 to 9.50) compared to the first quartile (4%–50%).
As most indications for noise effects were found for sleep efficiency and sleep latency in relation to outdoor noise, we investigated the effect of noise exposure in potential critical time windows for these two outcomes in more detail (Figure 1
). Noise exposure in the evening (19:00–23:00) and in the early morning hours (05:00–06:00) was significantly associated with sleep efficiency, whereas the other time windows reached only borderline significance. For sleep latency, noise exposure until 01:00 was most relevant.
Exposure within fixed time periods (as shown above) may not match the individual time period a person is asleep or in bed. For instance, for somebody rising before 06:00, noise exposure between 06:00 and 07:00 is not relevant regarding sleep disturbance. We thus calculated hourly noise exposure levels that matched the individual sleeping pattern of each night for the first four and the last four hours of sleep. The corresponding analyses with all actimetry-derived sleep outcomes are shown in Figure 2
. This individualized noise exposure provided stronger associations between measured noise exposure and sleep efficiency. Sleep latency increased by five to seven minutes per 10 dB(A) increase in outdoor noise exposure during the first four hours after bedtime. For sleep efficiency, noise exposure during the last three hours prior to wake-up was most critical, with reductions of 2%–3% per 10 dB(A) increase in measured outdoor noise exposure (Leq,1h
). For sleep duration and moving time, no significant associations were found, although noise exposure two to four hours after bedtime tended to increase moving time during sleep, and noise exposure two hours before wake-up tended to be negatively associated with moving time.
Self-reported sleepiness in the morning was significantly associated with noise exposure in the last hour of sleep, whereas no associations were observed for noise exposure at the beginning of sleep (Figure 3
). For self-reported sleep quality, none of the noise exposure time windows were significantly associated, but noise exposure in the middle of the sleeping period (±4 h from bedtime and wake-up) and at the end of the sleeping period showed the strongest trends for a relation. For IR matched to individualized sleep patterns, no significant effects on actimetry-derived and self-reported sleep outcomes were observed.
Noise annoyance, noise sensitivity, and sex were not significant effect modifiers for any of the outcomes and noise exposure time windows. A slight, nonsignificant trend was seen for a stronger association of nighttime noise (Leq,night) with sleep efficiency for males (p = 0.11), people with a high (≥median) annoyance score (p = 0.23), and people not reporting to be noise-sensitive (p = 0.25).
Our study suggests that the timing of noise exposure within the night is a relevant factor for the deterioration of objective and self-reported sleep quality. Using individual noise exposure time windows, matched to the individual bed and rise time of each night, provided stronger associations compared to fixed time intervals (such as Leq between 19:00 and 23:00). Sleep latency, as expected, was most consistently associated with noise exposure at the beginning of the night, while noise exposure prior to wake-up was most relevant for sleep efficiency and self-rated sleepiness. This suggests that noise exposure in the middle of the night may be less relevant for sleep quantity and quality, whereas noise exposure toward the end of the night, when sleep pressure is reduced and noise levels tend to be higher, is most disturbing for sleep.
This is in line with a field study using contrived aircraft noise exposure observing that aircraft noise events in the early morning elicited stronger reactions, as measured with high-resolution actigraphy, than events at the beginning of the sleep period [31
]. Our findings are also in line with a large population-based survey conducted in 2009–2010 in Oslo with 13,019 participants. Road traffic nighttime noise was mostly associated with waking up too early and with difficulties falling asleep but barely associated with awakenings during night [27
]. In a Finnish study of 7019 public sector employees, people exposed to >55 dB road traffic nighttime noise were more likely to report nonrestorative sleep (odds ratio (OR) = 1.29, 95% CI: 1.01–1.65) and waking up too early (OR = 1.24, 95% CI: 0.96–1.61) compared to people exposed to 45 dB or less. No associations were observed for frequently waking up during the night, short sleep duration, and difficulties falling asleep, with the latter not in line with our results [43
]. In a smaller Swiss survey of 1375 adults, various self-reported sleep quality indices were not significantly related to road traffic nighttime noise [26
]. However, most indications for an exposure–response trend were found for “waking up too early in the morning”, while least indications were found for “agitated sleep” and “waking phases during the night” in that same study [26
Whether, or how, the observed pattern with stronger effects for evening and early morning noise translates into chronic health effects is unclear. Separating long-term noise effects regressed on average night noise exposure from time-specific effects in specific time periods is challenging for epidemiological studies, primarily because transportation noise in different time periods within the night is highly correlated, at least if no night curfews are in force. This is especially the case for modeled road traffic noise when traffic input data are based on traffic count samples, which are then extrapolated [2
]. In reality, as demonstrated in our study, diurnal and day-to-day variation in traffic flows leads to a lower correlation between environmental noise exposures at different time intervals than one would observe between such computed standard metrics [44
]. For large-scale epidemiological studies on long-term risks, individual noise measurements, as done here, are not feasible. A Swiss cohort study on cardiovascular mortality evaluated the effects of diurnal noise variation in their analysis by combining road, rail, and aircraft noise. Although such a combination introduces additional diurnal variability due to different pattern and night curfews for aircraft noise, the correlation between different exposure time windows remained high (≥0.94), precluding any firm conclusion. There was a trend that, for all cardiovascular causes combined, exposure during 01:00 to 05:00 was most relevant. For hypertensive-related causes of death, early morning noise (05:00–06:00) was most relevant, and for ischemic stroke and heart failure, early evening and early morning noise were more detrimental than the rest of the night, although strongest association were seen with daytime noise [45
We hypothesized that we would find stronger associations for estimated indoor noise than measured outdoor noise, but we could not confirm this with our data. This is in line with results from a survey on self-reported sleep disturbance in the same Special Issue of this journal [46
]. However, our findings do not match the results of an epidemiological study done on indoor noise, which found stronger associations for hypertension with indoor noise compared to modeled noise at the most exposed façade [47
We did not measure indoor noise but rather applied an indirect procedure to estimate indoor noise levels from measured outdoor noise because our main interest was outdoor noise penetrating into the building. Direct indoor noise measurements would be heavily affected by the behavior of the participants. For instance, a sleepless person may produce some sound, which would yield a biased correlation between sleeplessness and noise exposure. There is no obvious explanation why indoor noise was less good a predictor than outdoor noise in our study. One may speculate that people who feel disturbed by outdoor noise close their bedroom window and thus have a lower indoor noise estimate, as this is the strongest predictor of indoor noise. This would mean that we deal with reverse causation in the sense that the sleep quality affects the noise exposure and not the other way round. Alternatively, estimated indoor noise levels may be subject to higher exposure misclassification as some levels were low. Estimated indoor noise levels below 20 dB(A) were censored and replaced with 20 dB(A) for the analysis; this was the case for 79 nights in 21 individuals. Finally, we obtained window opening habits for each season from the baseline questionnaire but did not specifically ask about the window position for each night, which may also add to exposure misclassification. Exposure misclassification may have resulted in reduced statistical power, which would explain the observed similar regression coefficients for indoor and outdoor noise but higher p-values for indoor noise.
We also hypothesized that sleep effects depend on the exposure characteristics. In particular, exposure situations with individual noise events clearly standing out from average (background) noise, as quantified with the IR [32
], were considered to be more detrimental for sleep than exposure to a steady sound level. However, we could not confirm that sleep disturbances are increasing with increasing IR. There was a nonsignificant, nonlinear pattern with lowest sleep efficiency for moderate (50%–63%) levels of IR, which is in line with findings of the effect of IR in a cohort study on cardiovascular mortality [33
]. However, in a cross-sectional survey on arterial stiffness, number of events during night was relevant [48
], and in a cohort study on diabetes, IR was not associated with the diabetes risk [7
]. Thus, it remains open whether IR is contributing to Leq
-based metrics for predicting long-term health effects of noise.
The strengths of this study include the prospective and detailed data collection with acquisition of objective sleep data (actimetry) and measurement of noise exposure. By measuring outdoor noise exposure instead of calculating it, we could adequately record the diurnal variability of noise acting upon study participants. Our measurements allowed us to estimate outdoor noise passing into the sleeping rooms for individual bedroom characteristics. We could also match exposure time windows to the individual bed and rise times. The relatively small sample is a limitation, and thus the power of the study is rather limited. The p-values of the analyses are not adjusted for multiple comparisons because we were interested in the pattern of the effect estimates rather than hypothesis testing. Thus, some significant coefficients may, in fact, be chance findings. Note also that some of the exposure and outcome measures were correlated, and thus analysis should not be considered as mutually independent. For instance, sleep latency was found to be significantly associated with noise exposure two hours prior to wake-up. This seemingly paradoxical result is likely explained by the fact that noise exposure in the early morning is correlated with noise exposure noise late at night. Only the latter is causally related to sleep latency.