Distribution of demographics, LFN and HRV data
Among the 30 subjects recruited, 12 males and 13 females participated in the summer field campaign, while 9 males and 10 females participated in the winter field campaign, with 14 subjects (8 males, 6 females) participating in both seasons. Table 1 lists the demographic characteristics of the subjects. The age of the subjects ranged from 22 to 75 years. The average BMI of subjects was 23.9 ± 2.8 kg/m2 in summer and 24.3 ± 2.9 kg/m2 in winter.
As shown in Table 1, for the summer and winter measurements the mean SDNNs were 61.7 and 66.5 milliseconds, while the mean LF/HFs were 1.9 and 1.1, respectively. The mean levels of LFN exposure were 43.5 ± 2.9 at Site OD and 50.4 ± 0.8 dB (LAeq) at Site ID in summer and 49.1 ± 6.9 and 42.3 ± 1.8 dB (LAeq), respectively, in winter. While the winter data (Site OD > Site ID) support our hypothesis that close proximity to wind turbines results in higher LFN exposure, the summer findings (Site ID > Site OD) of a higher LFN recorded with the wind turbines distant away imply a greater contribution to LFN from other sources indoors. Reviewing the difference between the sites in the two seasons showed that the use of fans indoor in hot weather is a potential source of LFN at Site ID. Summer in Taiwan is hot and Site ID is equipped with fans for ventilation. Another possible source is human conversation. The voiced speech of a typical adult male will have a fundamental frequency from 85 to 180 Hz and that of a typical adult female will have a fundamental frequency from 165 to 255 Hz .
To evaluate the contribution of these two potential sources, a follow-up assessment of the LFN at Site ID was performed. First, 15-minute LFN measurements were taken for indoor background noise first without any fan or conversation, then they were taken with one turned-on household fan (medium-sized) roughly 3 m away from the NL-62, and finally they were taken with the ongoing conversation of five persons (three males and two females) at 2-5 m away from the NL-62. The LFN at a 1-minute resolution was assessed. The results show that the average indoor background LFN was 32.5 ± 1.4 dB (LAeq) without fan use and conversation, 32.5 ± 1.4 dB (LAeq) with the fan turned on, and 44.1 ± 2.2 dB (LAeq) with ongoing conversation. As can be seen, with the fan turned on the indoor LFN recorded was the same as the background LFN. In contrast, with ongoing conversation the indoor LFN recorded was 11.6 dB (LAeq) higher than the background LFN. In other words, the summer findings (Site ID > Site OD) of the LFN recorded indoors surpassing that of the LFN recorded outdoors with wind turbines nearby can be attributed to the ongoing conversation indoors at the activity center.
Impacts of LFN exposure on HRV
One of the main objectives of this study was to evaluate the potential impacts of LFN on HRV in terms of SDNN and LF/HF. Table 2 shows the 5 min percentage changes of HRV indicators per interquartile range (IQR) increase in LFN. GAMM analysis yielded 7.86 dB (LAeq) as the IQR of LFN. After adjusting for confounding factors, the SDNN was reduced by 3.39% (95% CI: 0.15-6.5%) per 7.86 dB (LAeq) of LFN with a statistical significance p < 0.05. In other words, with an increase of 1 dB (LAeq) in LFN, the SDNN decreases by 0.43%. In contrast, the reduction in LH/HF per IQR increase in LFN was much smaller and not statistically significant. These results revealed a significant association between LFN exposure and changes in HRV, especially in SDNN, indicating the potential health impacts of exposure to LFN. The present findings were consistent with previous observations of reduced HRV due to LFN exposure in the US . In that study on 10 healthy males, SDNN was reduced by 16% (95% CI: 6.1-26%) during 40 min LFN exposure as compared with no noise exposure. As stated above, few studies have assessed the impacts of LFN on HRV. With the increasing emphasis on renewable energy, a growing trend of more turbines being built for wind power can be expected. Hence, it is both timely and necessary to conduct more assessments on the potential health impacts of LFN generated by wind turbines.
Besides LFN, studies have also indicated the possible effect of environmental noise on HRV. For 110 German adults exposed to daytime noise, the SDNN reduced by 0.67% for a 5 dBA increase when Leq 65 dBA . Another study also showed that a decrease from 17.42 to 17.7 ms for a 10 dBA increase above the background noise (45 dBA) in forty college-going male volunteers exposed to traffic noise ; the SDNN reduction was roughly 0.2% per 1 dBA. Additionally, an SDNN reduction was observed for another important environment factor, PM2.5. A significant decrease in SDNN (0.51%; 95% CI: 0.01-1.01%) was associated with a 10 µg/m3 increase in PM2.5 for Japanese patients aged 20-90 years . A meta-analysis of 33 panel studies in North America, Asia, and Europe showed that a 0.92% reduction in SDNN was observed for a 10 μg/m3 increase in short-term PM2.5 exposure . Wang et al. showed that 0.39% (95% CI: -0.72%, -0.06%) and 0.92% (95% CI: -2.14%, 0.31%) reductions were observed for short-term and long-term exposures to a 10 μg/m3 increase in PM2.5, respectively, for adults aged above 55 years old . Besides this, SDNN was decreased from 54.7% to 39.6% with an increase in ambient temperatures from 17 to 38 oC (reduced by 0.72 % per 1 oC) in twenty-eight healthy young subjects . In comparison, our findings of SDNN reduction (0.43% per 1 dB) due to LFN from wind turbines were slightly higher than the reductions in traffic noise exposure > 65 dBA (0.13% per 1 dB), in a similar range or slightly lower than those with a 10 μg/m3 increase in PM2.5 exposure (0.39 to 0.92%) and slightly lower than those with ambient temperatures (0.72 % per 1 oC). In short, the impact of LFN from wind turbines on HRV was higher than that from traffic noise and lower than that from PM2.5 and ambient temperature.
The above results also imply the absence or minimal lag effect of LFN on SDNN. During field monitoring, the subjects were shuttled between Site OD (high LFN exposure) and Site ID (low LFN exposure). Should there be lag effect, their differences in LFN would be small and the impact of LFN on SDNN would be insignificant, resulting in similar SDNNs at both sites. Nevertheless, the field monitoring results indicated a significant reduction in SDNN at Site ID (55.8 milliseconds in summer and 64.2 milliseconds in winter) compared to those at Site OD (65.3 milliseconds in summer and 68.8 milliseconds in winter). Hence, there was either no or a negligible lag effect. This observation was consistent with the results of a study conducted in Germany that showed that during routine activities, a 5 dB increase in LAeq ≥ 65 dB (20-20k Hz) was not associated with lagged SDNN change .
Besides LFN, wind speed is the only environmental variable with a significant impact on SDNN and LF/HF though with opposing trends of changes (Table 2). In this study, wind speed and temperature are environmental variables adjusted in the GAMM of LFN on SDNN and LH/HF. The exact mechanism by which these variables impact the two HRV indicators remains to be explored.
The SDNN decreased with increasing age but did not reach statistical significance in our work. In the literature, studies have found a significant association of age with SDNN. For example, Kim and Woo found that SDNN significantly decreased with increasing age in 2748 males and 735 females in Korea . Voss et al. also indicated an inverse association of age with SDNN for 1906 Germans aged 25-74 years . Furthermore, population studies in the Netherlands found that SDNN decreases continuously from birth to old age for 28,827 participants with ages ranging from 11 days to 91 years . In addition, as seen in Table 2 female subjects had a higher SDNN than male subjects, again not reaching statistical significance. Previous studies have also indicated that female subjects had a significantly higher SDNN than male subjects [52, 54]. Our study results showed a trend of increased association of SDNN and age and higher SDNN in females, consistent with the results of previous studies. Nevertheless, the sample size in this work may be the reason for the statistical insignificance of these results.
LFN exposure in residential households
In order to assess whether the subjects’ daily LFN exposure was in a similar range to our“LFN and HRV monitoring”, we conducted household LFN exposure monitoring for seven recruited households. The households’ average LFN levels were 34.8 6.9 dB and 43.4 5.7 dB for indoors and outdoors, respectively. As shown in Table 3a, the indoor LFN exposure during the 24-hour period among different households ranged from 30.7 to 43.4 dB (LAeq). Moreover, the maximum indoor LFN exposure of residents in their households in the daytime ranged between 39.7 and 56.7 dB (LAeq), which was similar to the range recorded at Site ID (38.2 and 52.5 dB). Previous studies have reported wind turbine LFN of 15-45 dB indoors in residences in Australia situated 870-3100 m from wind turbines , and indoor LFN levels of 0-10 dB during wind turbine operational periods for residences situated 1500 m from wind turbines in Australia . In comparison, the maximum indoor LFN exposure during a 24-hour period (43.4 dB) in our monitoring was similar to that reported (45 dB) by Hansen et al.  and higher than that (10 dB) reported by Evans et al. .
As shown in Table 3b, the outdoor LFN measured during a 24-hour period among different households ranged from 38.2 to 50.0 dB (LAeq) in summer and 38.9 to 44.6 dB (LAeq) in winter. These household data were slightly lower than the field results at Site OD (38.3-53.5 dB, LAeq, in summer and 40.5-57.1 dB, LAeq, in winter), which was located much closer (20 m) to the turbines than the households. Our results confirmed that these residents indeed were exposed to similar LFN levels as in the field campaign at site OD. Therefore, HRV impacts from the LFN exposure evaluated in the field campaigns could be found in the daily lives of these residents. Hansen et al. and Evans et al. also reported that the LFN exposure levels from wind turbines for outdoor measurements were 25-40 dB and 21-25 dB, respectively [36, 55]. Our monitoring of outdoor LFN exposure (38.2-50.0 dB) was higher than their results. The seven households in this study were located closer (124-330 m) to the turbines than the households in Hansen et al. (870-3100 m)  and Evans et al. (1500 m) were .
Table 3a also shows that the indoor LFN exposure levels (dB, LAeq) at most households were higher in the daytime than in the evening and nighttime. The higher LFN exposure level in the daytime could be attributed to other sources of background noise in addition to turbine-generated noise, while the LFN at nighttime presumably came mainly from wind turbines. House 1 recorded the highest mean and maximum nighttime LFN exposure (40.8 and 48.5 dB (LAeq), respectively), attributed to its close proximity to wind turbines. According to the Guidelines for Community Noise , for a good night’s sleep the equivalent sound level should not exceed 30 dB (LAeq) for continuous background noise. However, the average indoor LFN levels at nighttime in Houses 1, 3, 4, and 7 were above 30 dB (LAeq) and 100% of the 5 min nighttime observations recorded in Houses 1 and 4 exceeded 30 dB (LAeq), implying that turbine-generated LFN may affect residents' quality of sleep at nighttime in these households.
The Taiwan EPA designates different noise standards in residential areas for different times of day: daytime (39 dB, LAeq, 7 am to 7 pm), evening (39 dB, LAeq, 7 pm to 10 pm), and nighttime (36 dB, LAeq, 10 pm to 7 am) (Taiwan EPA, 2017). According to these guidelines, House 1 (daytime, evening, and nighttime), House 2 (daytime), House 4 (nighttime), and House 7 (daytime and evening) had LFN exposure levels exceeding the respective standards designated by the Taiwan EPA (Table 3a). Among these residences monitored, residents at House 1 had higher LFN exposures from wind turbines round the clock, indicating that they were exceeding the LFN standards of the Taiwan EPA 99.6%, 89.1%, and 96.8% of the time for the daytime, evening, and nighttime, respectively. The impacts of distance, building materials, and having windows open/closed on the LFN could be illustrated by our cases, as briefly discussed in the following. The significant influence of distance from turbine on the indoor LFN exposure level is best illustrated by House 1. With the shortest distance from wind turbines, House 1 had the highest mean LFN exposure, both indoors and outdoors, during the 24-hour period, with its indoor LFN exposure ranging from 40.8 to 45.0 dB (LAeq) and the maximum level reaching 50 dB (LAeq) in the evening. However, House 6, the farthest (330 m) among the seven households monitored, did not record the lowest mean LFN exposure. Instead, the lowest LFN exposure of 30.7 dB (LAeq) in both seasons was recorded inside House 5, the only residence with airtight windows installed. Windows serve as sound attenuation . Although the resident of House 5 indicated a habit of keeping windows fully open in the summer (Table 3), the windows were actually closed during the monitoring period according to our observation. Moreover, House 5 had only one resident and the windows were kept closed when the house was empty. A single-member household with less ongoing conversation would imply the absence or negligible contribution of this indoor LFN source (unlike the situation at Site ID). Moreover, keeping airtight windows closed most of the time contributes to low sound transmission. The soundproofing effectiveness of airtight windows is evidenced by the largest indoor–outdoor difference in LFN of 13.7 dB (LAeq) recorded for House 5 in both seasons (Table 3c).
Houses 3 and 4 were located at equal distances (308 m) from the nearest turbine, but their average indoor LFN exposure differed by 4 dB (33.7 dB (LAeq) and 37.9 dB (LAeq), respectively) (Table 3a). The results indicated that though it had been built with CN, House 4 had a higher indoor LFN recorded, which can be attributed to the fully open windows, resulting in poorer sound insulation compared with House 3 with its fully closed windows. Therefore, the impacts of opening the windows on the indoor LFN were more significant than those of building materials in this case.
The impacts of building materials on LFN were not a focus of our study. However, the indoor–outdoor LFN difference in seven households presented certain indications of their impacts. As shown in Table 3c, the CN residences (Houses 1, 2, 6, and 7) had larger indoor–outdoor differences (range, 6.6-11.2 dB (LAeq)) than the CB residence (House 3; (range, 5.8-8.5 dB (LAeq)). The results indicated that CN had a higher LFN insulation compared with CB. In summary, distance from turbines, building materials used, types of windows installed, and whether they are open or closed all had impacts on the indoor LFN levels in our study.
Recommendations and study limitations
The present results show the adverse impact of LFN exposure on HRV. For public health protection, there should be regulations on the requisite distances of wind turbines from residential communities. In Taiwan, wind farms are owned by large corporations, and distance regulations would prevent these operators from reaping benefits at the expense of nearby residents suffering from long-term LFN disturbance and adverse health impacts. Residences in close proximity to wind turbines should be equipped with airtight windows for better sound insulation. To reduce LFN, windows should be kept closed as much as possible and especially at night to ensure a good quality of sleep.
This study has some limitations. Firstly, at Site ID for the households monitored, indoor LFN is caused by wind turbines but there may also exist other indoor LFN sources, such as indoor ventilation devices, ongoing conversations, or television, which have not been thoroughly explored. Secondly, our sample size is small due to difficulties in recruitment. Nevertheless, the findings obtained from the 30 subjects still demonstrated the impacts of LFN on HRV changes. Finally, the average analyzable wear time (86.2%) was lower than that (93.6%) found in previous research . Nevertheless, data loss occurred randomly and did not undermine the validity of the present findings.