Wellbuilt for wellbeing: Controlling relative humidity in the workplace matters for our health

Abstract This study offers a new perspective on the role of relative humidity in strategies to improve the health and wellbeing of office workers. A lack of studies of sufficient participant size and diversity relating relative humidity (RH) to measured health outcomes has been a driving factor in relaxing thermal comfort standards for RH and removing a lower limit for dry air. We examined the association between RH and objectively measured stress responses, physical activity (PA), and sleep quality. A diverse group of office workers (n = 134) from four well‐functioning federal buildings wore chest‐mounted heart rate variability monitors for three consecutive days, while at the same time, RH and temperature (T) were measured in their workplaces. Those who spent the majority of their time at the office in conditions of 30%‐60% RH experienced 25% less stress at the office than those who spent the majority of their time in drier conditions. Further, a correlational study of our stress response suggests optimal values for RH may exist within an even narrower range around 45%. Finally, we found an indirect effect of objectively measured poorer sleep quality, mediated by stress responses, for those outside this range.


| INTRODUC TI ON
The US General Services Administration's Wellbuilt for Wellbeing research project, led by the University of Arizona's Institute on Place, Wellbeing & Performance, was an exploratory study investigating how health-related metrics change as measured levels of indoor environmental quality (IEQ) factors change in real time.
Using a cross-sectional, observational study design, we saw a strong correlation between changes in relative humidity (RH) and changes in stress response. Specifically, we measured a 25% difference in stress response levels between those spending the majority of their time in the 30%-60% RH range set in ASHRAE  and those in drier conditions. The difference is of moderate effect size and may represent clinically significant levels cumulatively, over long-term exposure. 1,2 This is important because Americans spend more than 90% of their time indoors and over 50 million US office workers spend 20% of their time at the office. 3,4 Policymakers, standards developers, and building design and operations professionals are increasingly interested in indoor environment quality (IEQ) and its effect on occupant health and comfort. Despite several studies revealing the potential negative impact of both too high and too low relative humidity (RH) on health and comfort, 5-10 the importance of controlling RH has received less attention than other IEQ parameters and lower limits have been removed from thermal comfort standards.
There is an ongoing debate about the influence of RH on perceived indoor air quality and comfort and the impact RH has on health. The debate involves whether (a) RH has little or no direct effect on health or comfort, which are instead driven by the presence of indoor pollutants; (b) RH has a mediating effect caused by impacts to the precorneal tear film and mucosal membranes in the eyes, skin and airways, which make individuals more susceptible to the effects of pollutants; or (c) RH has a direct effect on perceived health and comfort. Several studies illustrate the extremes of this debate. A chamber study of 8 healthy, male, college-aged students found no impact of relative humidity, 11 while other studies concluded that indoor pollutants such as VOCs emitted from building materials and or particulates were the likely cause of perceived "dry air" and irritation of skin, eyes, and airways. [12][13][14] Furthermore, humans lack sensory receptors for humidity 5,[13][14][15][16] and perceive changes in RH indirectly through other sensory pathways which may affect the influence it has on perceived thermal comfort. Other literature finds plausible connections between RH levels and risk factors for eye irritation and sleep quality, 7 production of cortisol in the skin, 15 and the transmission, survivability, and virility of influenza viruses. 6,8 These contrasting findings indicate that the relationship between perceived comfort, health outcomes and humidity is complex and warrants further study. This complexity and lack of multidisciplinary, well-controlled, field-based study has affected how RH is treated in building codes and standards. Relative humidity levels in buildings are influenced by occupancy, ambient climate, and building design, construction, and operation. An earlier version of the most commonly used US standard for indoor air quality, ASHRAE Standard 55 (ASHRAE , required 30%-60% RH in office buildings for thermal comfort. 16 That standard was later relaxed by eliminating the lower RH limit, when ASHRAE began positioning its standards as a basis for building codes that focused on low energy consumption. ASHRAE found insufficient evidence based on real-world conditions to support negative impacts on wellbeing. 17 Earlier research by Sterling et al 18 suggested that a mid-range value of 40%-60% RH might be optimal for prevention of a number of health issues in buildings. However, while Derby et al 6 reiterated many of Sterling's findings, they caution against mandating a general, lower limit without additional research involving a larger number of subjects, a more diverse study population, and conducted in real-world settings. Additionally, prior research relating RH to health outcome metrics has considered just one outcome factor belying the complex nature of any potential relationship. [19][20][21] we found an indirect effect of objectively measured poorer sleep quality, mediated by stress responses, for those outside this range.

K E Y W O R D S
health, office workers, relative humidity, sleep quality, stress responses, wearable sensors

Practical Implications
• Thermal comfort standards for relative humidity (RH) in buildings have been relaxed over the past 30 years and since ASHRAE 55-1989 there has been no lower limit to RH in thermal comfort or ventilation standards.
• This project is the first to study associations between RH and three objectively measured health metrics including stress responses (heart rate variability), sleep quality, and physical activity in real time in a large observational study conducted in office buildings.
• We found that individuals who spent the majority of measured time within the range set in ASHRAE  (30%-60% RH) experienced lower stress responses at the office and better sleep quality than those who did not.
• Our results support further research to parse potential direct versus indirect effects, confounders, and to establish a mechanistic link between RH and health outcomes.

| Cardiac activity measurement (stress responses)
To record cardiac activity, we used a chest-worn sensor, EcgMove 3 (movisens GmbH, Karlsruhe, Germany). On the basis of participants' preference, the sensor was placed below the sternum with two standard electrocardiography patches attached to either the skin or on a chest belt. 24 To quantify stress responses, we quantified heart rate variability by two standard time-dependent measures including the mean of standard deviations (SDNN) and the square root of the mean squared differences (RMSSD) for all successive normalized-to-normalized heart rate intervals. 25 SDNN is a global index of HRV and reflects longer-term circulation differences and circadian rhythm. Lower SDNN values indicate higher stress responses, and higher SDNN values are linked to better wellbeing. 26 RMSSD is another method to quantify heart rate variability, which reflects vagus tone. Higher values equate to higher parasympathetic activities or more relaxation. 25 We calculated both SDNN and RMSSD in every 5-minute interval according to the guidelines of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. 25 We observed a average measure = 0.945 (0.929, 0.958)). We then took the average of these across the entire interval of interest (ie, at the office, outside the office, and during time in bed). To simplify the visual presentation of these data, we reverse the scale for SDNN and RMSSD in our figures presenting the data as 100 ms-any reported value.

| Physical activity measurement
The EcgMove 3 sensor includes a tri-axial accelerometer, which quantified physical activities using validated algorithms described in previous studies. [27][28][29] We assessed a number of physical activity metrics at the office including activity postures (sitting, standing, walking, lying), postural transitions (sit-to-stand, sit-to-lie, stand-to-walk), total number of taken steps, average of unbroken walking bouts in seconds (average duration of all recorded continuous walking bouts with a minimum of three consecutive steps), and average physical activity level (defined by the intensity of the acceleration magnitude [mG] per second). This last measure is our main method of capturing overall physical activity as it can be connected to clinical definitions for sedentary behavior and was utilized in previous work by our Wellbuilt for Wellbeing project. 30 To capture physical activity outside of the office, we estimated the percentage of activity behavior (eg, sedentary, light, and moderate-tovigorous activities) using a validated algorithm tailored for this study. 31

| Sleep quality measurement
Sleep quality was objectively measured from chest-worn sensor data relating to the motion, position, and posture of the participant's torso while in bed. We used a composite score called the sensor-based sleep quality index (SB-SQI). 32 SB-SQI is based on a validated algorithm estimating sleep-onset latency, total sleep time, and sleep efficiency from the chest-worn sensor data 27 and then applying the scoring method from the Pittsburg Sleep Quality 33 to present sleep quality. Although organized in similar fashion to the PSQI, the inputs for SB-SQI scoring are objective measures derived from sensor data.

| Relative humidity (RH) and other IEQ measurement
We took continuous measurements of ambient levels of several IEQ factors at the workplace including RH, 34 T, carbon dioxide (CO2), and particulate matter (PM) as previously reported 35 (Figures S15-S18).
Measurements were made using an Aclima (Aclima Inc) measurement platform consisting of individual sensing nodes mounted on the walls near participants. The 1 Hz time-series RH data were transmitted via Wi-Fi or Ethernet to cloud-based servers where the data are processed, analyzed, and stored. To evaluate levels of RH exposure, we tracked participants' proximity to ambient RH measures using calendar information, recorded logs, and floor plan coding. 35 The RH sensor reports values that range from 10% to 90%, with a resolution of 0.3% and an accuracy of ±4%. Additional descriptions of sensors including type, range, resolution, and accuracy are included in supplemental materials S15-S18.
The participants' recorded IEQ data were averaged every 5 minutes, resulting in a time-series of RH exposures for each participant over the course of their time at the office ( Figure 1). We then classified participants based on whether more than 50% of their measured RH values were in or out of the 30%-60% comfort range established by ASHRAE 55-1989. 18 This resulted in 3 groups: • 1 = dry: majority of measurements < 30% RH, • 2 = comfort-humidity: majority of measurements ≥ 30% and ≤ 60% RH, and • 3 = humid: majority of measurements > 60% RH

| Statistical analysis
We performed all statistical analysis using SPSS statistics 23.0 (IBM), except the structural equation modeling (SEM), which was performed with R statistical package version 3.4.1 and the lavaan package. 36 We used independent t tests (proportional) or chi-square tests (bivariate), as appropriate, to compare the demographic information of groups.
Univariate analysis compared group differences for investigated variables with age, BMI, and seasonal effect as covariates. We measured effect size (ES) as Cohen's d with a range of small (~0.02), medium (~0.5), and large (~0.8). 37 We used several structural equation models (SEM) to estimate the direct and indirect effects of RH on outcome measures. In these models, we selected SDNN for representation of the stress responses, the SB-SQI score to represent sleep quality, and the general tri-axial accelerometer data (mG) to represent average physical activity at the office. We explored the idea that exposure to different levels of RH could affect one's stress responses as well as physical activity at the office, which in turn could lead to differences in sleep quality as anticipated by the reciprocal relationship between these health-related metrics. 22,23,35 Therefore, for structural equation modeling and ease of interpretation of the association between sleep and average physical activity and RH mediated by stress, SDNN values were reversed and multiplied by a constant value of 0.01. Global estimation with maximum-likelihood approach 38 F I G U R E 1 An illustration of recorded relative humidity (RH) over three days of recording for an individual participant. We matched the location of the participant, when available, to the relevant RH measurements from the environmental sensor in the proximity of the participant. 57% of this participant's data were recorded below 30% RH and they were therefore grouped in the "dry" category simultaneously estimated the path coefficients, while bootstrapping derived the standard error of estimates. We handled missing values in the model using the full maximum-likelihood method. 39 We tested the sensitivity of each variable using analysis of variance (ANOVA)/ linear regression as well as a comparison of modification indexes. 40

| Correlational analysis
We evaluated the relationship between noncategorized RH exposure and stress responses using a series of exploratory correlations.
This exploratory work is similar to previous analyses described in quantum correlations and synchronization measures. 41 The correlation was measured by the correlation coefficient, which represents the strength of the linear relationship between the stress responses and RH exposures at each comparison point as the following: where r is a correlation coefficient, x is a RH value converted by a RH comparison point, and y is the stress response defined by averaged SDNN of each participant. The RH comparison point ranged from 30% to 60% at 1% increments as the following: where K is the humidity comparison point and h is a measured relative humidity. For example, to get the correlation coefficient for the RH comparison point of 30%, the measured RH was subtracted by 30 then the absolute value was calculated so that a range of the converted RH value is from 0 to 70. Then, the converted RH values were applied to the above equation to get the correlation coefficient. This procedure was repeated up to the RH comparison point of 60.

| Participants
A total of 231 white-collar office workers (54.5% female, average age 43.4 years) participated in this study. Given participant proximity to ambient RH measurement sensors, 134 individuals had sufficient data for analysis. For our analyses, sufficient data were considered to be at least 40 minutes of measured RH data (this quantity exceeds minimum timeframe necessary to assess the effect of RH on heart rate). 21 Table 1 summarizes the demographic, clinical, and sensor-derived parameters, and the schema of recruitment is illustrated in Figure 2. Of the 34% of workers who were in the discomfort-humidity group, 67% (n = 31)

| Stress responses
Individuals in the dry and humid groups experienced 25% and 19% higher stress responses (lower SDNN), respectively, compared to those in the 30%-60% RH group (P < .050, Figure 3). These differences are of moderate effect size as shown in Table 2. Because participant responses in dry and humid conditions were similar, we then merged these groups into a single "discomfort" category (ie, outside 30%-60% RH). However, building operations typically attempt to control high humidity, and not low humidity, and the dry conditions have a stronger stress response.
On average, participants in the discomfort group experienced higher stress responses (lower SDNN) compared to the comfort group (22%, P = .006, ES = 0.54). This group also had higher stress responses outside of the office (13%, P = .055, ES = 0.39) and at night (15%, P = .054, ES = 0.39) though these findings were not statistically significant (Figure 4). Similar trends were observed when the metric indicator of relaxation (RMSSD) was considered ( Figure S2).

| Evidence of an optimum range for relative humidity and stress responses
The

| Physical activity and sleep quality
We performed group comparisons for sleep quality and physical activity health-related metrics. We did not find a significant difference in sleep quality between groups. Additionally, while we found significant differences in total step counts and duration of walking bouts, we did not find a significant difference in overall physical activity (mG). Overall physical activity was our primary health-related metric (Table 2 and Figure S5).

| Structural equation model for stress responses, sleep quality and physical activity
Despite the fact that we did not observe direct effects between one's RH grouping and sleep quality (SB-SQI), we observed a significant mediating effect for stress responses (SDNN) between the two RH groups with small effect size (B = −0.02 standard error = 0.01, 95% CI = −0.03, −0.01; Figure 6). In other words, inclusion in the discomfort humidity group was related to higher stress responses at the office,  Table S6 and Figure S7.

| D ISCUSS I ON
Our study measured a 25% difference in stress response levels between those spending the majority of their time in the 30%-60% RH range set in ASHRAE  and those in drier conditions.
The ASHRAE standard, which has since been superseded and the lower limit removed, may be beneficial for reasons more imperative than thermal comfort-that is, for both its direct impact on stress TA B L E 2 Between-group comparisons for sensor-derived parameters, including stress responses, physical activity, and sleep quality at the office, outside the office, and during time in bed Parameters Discomfort-humidity mean (SD) Our literature review provides some basis for the increased stress response. Previous studies found connections between acute conditions such as deterioration of precorneal tear film in the eyes and mucosal membranes in skin and airways, inflammation, and spread of influenza with complaints of dry or stuffy air. 8,9,42 Literature review studies have linked high or low RH with the potential development of illnesses such as asthma, and dry and irritated mucous membranes of the eyes and airways, via direct and indirect pathways of impact. 7,43,44 They refer to field studies that have illustrated the complexity of these relationships and suggested different mechanisms. For asthma in particular, a population control study found an association between VOC levels and reports of asthma in children 45 while a similar study in the UK found that asthma levels in children were significantly correlated with dampness but not with total VOCs. In the latter study, only the presence of formaldehyde in combination with damp conditions was associated with wheezing. 46

F I G U R E 3
Group comparison of individuals with majority of recorded exposure within dry, comfort, and humid conditions. Those in dry and humid conditions experienced 25% and 19% more stress, respectively, than those in the comfort condition. These differences are of moderate effect size. The results for the dry and humid conditions were similar enough to collapse into a single "discomfort-humidity" grouping for subsequent analyses. Results were adjusted by age, BMI, and season. These stress response values are presented as 100 ms -SDNN to simplify the visual representation of higher stress F I G U R E 4 Comparison between groups with majority exposure inside or outside of the 30%-60% relative humidity range established by ASHRAE 55-1989 while at the office. Stress responses were quantified by a comparison of heart rate variability (SDNN) between these groups. Those in the discomfort-humidity group at the office had lower SDNN (higher stress) in comparison with the comfort-humidity group at the office as well as outside the office. Results were adjusted by age, BMI and season. These stress response values are presented as 100 ms-SDNN to simplify the visual representation of higher stress We also found that dry conditions and humid conditions correlated with stress response to a similar degree. While physiological mechanisms are complex, and we cannot explore them given our exploratory study design, there is plausible explanation for a potential relationship in the literature. The association between dry conditions and stress responses could be explained by dry ambient air increasing the likelihood of water loss through the skin 47-49 leading to lower skin perfusion and consequently higher cardiac activity to accommodate the water loss. 39,50 Similarly, in humid conditions, the amount of water evaporation from the skin (the mechanism of action for cooling down the body) leads to an altered thermoregulatory mechanism and, thus, increased cardiac activity. 39,50 These studies align well with our correlational analysis which suggests the potential for an optimal range in the relationship between RH and stress response. We do not presume that a specific range can be inferred from this study, nor would we minimize the potential risks associated with controlling low RH in buildings. However, our data do support further exploration through a more rigorously controlled study to test whether the range established in ASHRAE 55-1989, or an even narrower range, would reduce physiological stress responses in individuals exposed to RH outside that range.
We did not observe a direct correlation between RH and sleep quality or a consistent relationship between RH and physical activity.
However, we used structural equation models (SEM) to investigate any indirect relationships between RH and these health metrics. Our SEM results revealed a relationship between RH at the office and sleep quality mediated by stress responses at the office. This finding supports previous research that stress during work hours may be associated with poorer sleep quality. 51 By influencing one's stress response, RH may magnify this association. Our SEM did not identify physical activity as a significant mediator of the influence of RH on stress responses or objectively measured sleep quality ( Figure S7).
Our findings also complement other research that suggests office workers in the US exhibit generally sedentary behavior. 52 Because RH varies with T and absolute humidity, we expected that T might have some influence on our findings and explored its effects in several ways. Interestingly, we found T had a limited impact on the association between RH and stress responses. This may be due to the well-controlled T in the study locations which resulted in low fluctuation of T in office settings ( Figure S9). However, when we added seasonally defined T ranges to our criteria for "discomfort" ( Figure S2), we found a 14% difference in stress response between discomfort and comfort groups (14%, P = .007, ES = 0.48). Thus, incorporating T into the criteria for discomfort reduces the difference between groups as compared to when they are based on RH alone (22%, P = .001, ES = 0.64). Additionally, we did not find any statistically significant relationship between temperature and participants' stress responses. This suggests that RH has a bigger influence on F I G U R E 5 Exploratory analysis of noncategorized RH exposure and stress responses. A, We studied the correlation between participants' average SDNN and the corresponding absolute distance between participants' average RH exposure and each comparison point between 30% and 60% RH. We then plotted that correlation on the y-axis at each comparison point. B, At the 30% RH comparison point, there was no correlation. C, All RH comparison points between 42% and 48% RH were statistically significant. At 45% RH, we saw the highest correlation. This suggests an optimal range in the relationship between RH and stress response, where RH values on either side of 45% (eg, drier or more humid conditions) are associated with higher stress F I G U R E 6 Structural equation modeling revealed a significant indirect effect of RH on sleep quality mediated by stress responses stress response in our data than T. T is the primary focus of thermal comfort standards today which is understandable since humans have no RH receptor and only perceive changes in RH indirectly.
Frequently, humans confuse the symptoms of changes in RH, especially to low RH, for other conditions like stuffy air, irritation of skin and eyes, or fatigue. 48 Previous studies also indicate that high RH at surfaces is associated with growth of various indoor contaminants including mold and mildew. 53 Our findings suggest that relative humidity may have a place in building standards beyond these traditional associations with thermal comfort and control of biological contaminants.
There is ongoing debate about the relationship between indoor RH and indoor pollutants like PM and VOCs, and the perceptions of indoor air quality and health outcomes of building occupants. [6][7][8] High or low RH in buildings can be related to high occupancy or is- We did measure levels for CO2 and PM and have used as proxies for ventilation effectiveness, high occupancy, and building ECS performance. We found them to be within expected ranges for a commercial facility (Figures S17, S18). While we do see variation in CO2 and PM levels between our discomfort and comfort RH groupings, they were not statistically significant (Figures S10, S14). As might be expected, average PM was progressively lower across the narrower RH groupings of dry, comfort, and humid ( Figure S10). However, the differences were not statistically significant, and, despite the differences in PM between them, the measured stress responses between dry and humid groupings were minimal. In the narrower "humid" RH grouping, average CO2 levels were significantly lower than in the dry or comfort RH grouping ( Figure S14). Based on these findings, we feel that crowding or poor ventilation is unlikely reasons for the differences in stress responses. Furthermore, we adjusted our structural equation models to test whether CO2 or PM has direct or moderating effects on stress responses. We found no significant direct effects ( Figure S7) or moderating effects ( Figures S8, S9) for PM, and we found no direct effects ( Figure S11) or moderating effects (Figures S12, S13) for CO2. We believe these findings make confounding effects associated with ventilation, crowding, or the presence of elevated PM unlikely.
We see some correlation between calendar season and the dry and humid conditions experienced by our study participants.
Seasons could potentially relate to changes in stress response due to factors such as dramatic transitions in thermal conditions between indoor and outdoor settings, physical inactivity during unpleasant weather, and effects of thermal conditioning operations on individual responses. While we were not able to address seasonal effects directly, we did explore the possible relationship indirectly. We compared stress response between participants grouped by participation during a heating or cooling season, as defined by the building operator ( Figure S4). We found the difference in stress response to be only 6% (6%, P = . understanding of how to control conditioning based on mean radiant temperature or air velocity. 57 We recommend direct testing of these strategies to measure their effectiveness in addressing the health outcomes mentioned in this paper. Sixth, because of the design of our study, we cannot conclude any causality between conditions and health outcomes during work hours to outcomes outside of work hours. However, because results from past studies have indicated the possibility of such carryover effects on physiological, diurnal variations, 58-60 our data are relevant to future investigation. In this study of four well-functioning office buildings, the indoor environment was generally were well controlled. This caused small variation in many IEQ factors, thus limiting the ability to observe significant links between some IEQ factors and physiological outcomes that might be found in other research. [5][6][7][8][9][10] Finally, Wellbuilt for Wellbeing is a cross-sectional, exploratory study that cannot demonstrate causation. It is possible that other characteristics of the built environment, job type, work relationships, social or economic conditions, family issues, or other individual differences may have mediating effects between relative humidity, stress, and sleep. Additionally, we are unable to fully account for individual confounders such as selection bias in the consent process and coping mechanisms that may alleviate certain effects. Using the collected data, we have explored the potentially contributing or confounding effects some of these factors including temperature, CO2, PM, physical activity, season, and basic demographics and found the relationship between RH and stress to hold. Using wearable technology, a diverse cohort of participants and a wide range of inputs and unique health outcomes, our study offers a novel and important perspective to understanding the humidity-wellbeing relationship.

| CON CLUS ION
We found a 25% lower stress response between individuals spending more than 50% of their time in 30%-60% RH compared with those spending most of their time in drier air at the office. This difference is of moderate effect size and may represent clinically significant levels cumulatively, over long-term exposure. An interpersonal study of all our noncategorized RH exposure and stress responses data suggests there may be an even narrower optimal range for RH and stress responses around 45% RH. We identified an indirect relationship between RH and objectively measured sleep quality. These findings lend new evidence to support the 30%-60% RH range included in the outdated ASHRAE 55-1989 thermal comfort standard. Further rigorously controlled studies are needed to provide evidence for causation and to understand the mechanism driving this relationship. The experimental manipulation of relative humidity and other conditions may be able to address some of the limitations of the current study. Cross-sectional, exploratory studies cannot fully address the role that characteristics of the built and ambient environment or individual differences play. Future interventions may also be able to investigate the effect that different types of cooling and heating systems, such as microclimate control, have on stress and sleep.
In the near term, these findings support a range of strategies that may improve health and comfort. Such strategies include behavioral interventions such as periodic restorative breaks during the workday as well as using personal comfort devices and radiant heating and cooling strategies to address extreme low RH while simultaneously improving thermal comfort. 50 million white-collar workers in the US spend over 20% of their time at the office. 3,4 Over many years of work, narrowly controlling RH values to avoid both overly dry and humid conditions at the office could have a potentially large impact on the health and wellbeing of those workers.