Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Aug 18.
Published in final edited form as: J Expo Sci Environ Epidemiol. 2021 Feb 18;31(4):641–653. doi: 10.1038/s41370-021-00303-x

Characterization of Heat Index Experienced by Individuals Residing in Urban and Rural Settings

Suwei Wang 1,2, Connor YH Wu 3, Molly B Richardson 4, Benjamin F Zaitchik 5, Julia M Gohlke 1
PMCID: PMC8273073  NIHMSID: NIHMS1667949  PMID: 33597724

Abstract

Heatwave warning systems rely on forecasts made for fixed-point weather stations (WS), which do not reflect variation in temperature and humidity experienced by individuals moving through indoor and outdoor locations. We examined whether neighborhood measurement improved the prediction of individually experienced heat index in addition to nearest WS in an urban and rural location. Participants (residents of Birmingham, Alabama [N=89] and Wilcox County, Alabama [N=88]) wore thermometers clipped to their shoe for 7 days. Shielded thermometers/hygrometers were placed outdoors within participant’s neighborhoods (N=43). Nearest WS and neighborhood thermometers were matched to participant’s home address. Heat index (HI) was estimated from participant thermometer temperature and WS humidity per person-hour (HI[individual]), or WS temperature and humidity, or neighborhood temperature and humidity. We found that neighborhood HI improved the prediction of individually experienced HI in addition to WS HI in the rural location, and neighborhood heat index alone served as a better predictor in the urban location, after accounting for individual-level factors. Overall, a 1°C increase in HI[neighborhood] was associated with 0.20°C [95%CI (0.19, 0.21)] increase in HI[individual]. After adjusting for ambient condition differences, we found higher HI[individual] in the rural location, and increased HI[individual] during non-rest time (5am-midnight) and on weekdays.

Keywords: exposure assessment, exposure sensors, environmental health policy

Introduction

Exposure to extreme heat is a significant public health problem. Extreme heat events increase adverse health outcomes in the United States (U.S.) and heat-related deaths are projected to increase in the absence of effective adaptation measures with continued warming [1–3]. According to the National Weather Service (NWS) hazard statistics report, there were more heat-related fatalities in the U.S. in 2018 than fatalities associated with other forms of weather [4]. Using death records between 2006 and 2010, 31% of weather-related deaths were attributed to excessive heat exposure in the U.S., and was second only to cold exposure as a coded cause of mortality [5]. Over 7,400 deaths reported from 1999 to 2010 in the U.S. were associated with exposure to excessive natural heat [6], and approximately 28,000 heat-related illness hospitalizations in 20 states were recorded from 2001 to 2010 [7].

Extreme heat warning systems rely on temperature and humidity forecasts made for fixed-point weather stations (WS) to trigger alerts and initiate response plans to reduce the human health consequences of heatwaves [8–10]. These systems typically do not take into account variations due to microclimates across a landscape, although an experimental HeatRisk forecast product developed by NWS provides finer scale heat risk guidelines for western U.S. states [11]. Additionally, these systems do not account for variations due to the activity patterns of individuals moving through indoor and outdoor environments and other factors including acclimatization, dehydration, medications, and health conditions [12]. For example, a study of U.S. football player deaths found that 100% of heat-related deaths occurred in conditions that did not trigger a National Weather Service alert [13]. As a first step to understand the importance of exposure variation in predicting risk at the population level, a more precise measurement of exposure at the individual level is needed.

Recent studies have characterized individually experienced temperature and humidity by using wearable sensors. The study by Bernhard et al. (2015) reported that it was feasible to measure individually experienced temperature through sensors clipped on the shoe in both urban and rural settings in Alabama (AL), USA [14]. Other studies used similar sensors to characterize exposure, with the sensor worn around the neck or on the waist [15, 16], or attached to a shirt pocket, belt or bag [17, 18]. Differences in exposure across a study area and differences between individually experienced temperature and temperature measured at regional WS have been reported [17, 19, 14, 16]. Neighborhood level heat index assessment, which provides a finer scale exposure estimation than WS, has not been used to explain and predict individually experienced temperature or humidity.

Studies estimating adverse health outcomes associated with extreme heat are usually undertaken in urban areas where more people are closer to a given weather station. Much less is known about heat-related health risks in rural areas. However, by reviewing studies on heat-related mortality in rural and urban settings published in 2000–2017, the study by Li et al. (2017) estimated that the relative risk of heat-related mortality in rural settings was about 3.3% larger compared to that in urban settings [20]. While the temperatures experienced in cities are heightened due to the urban heat island effect, persons living in rural areas typically have less access to air conditioning, longer travel time to medical help, and longer power outages during heatwave days [21, 22]. Studies characterizing heat index exposure simultaneously in urban and rural settings have been conducted only in outdoor workers, revealing significant outdoor worker variation in temperature exposures within and between the study locations [16]. Whether non-outdoor workers have different exposure in rural and urban settings is relatively unexplored.

This study investigated whether individually experienced Heat Index (HI, °C) was different in an urban vs. rural setting and evaluated how well WS HI or neighborhood HI predicted individually experienced HI in both settings. We recruited participants with comparable characteristics in urban Birmingham, AL and in rural west central Wilcox County, AL, settings that frequently experience daytime high temperatures of 29–32°C, which are considered at least in the Caution risk by the National Weather Service assuming a relative humidity of 50% [23, 10]. This work builds from prior studies that have been conducted to characterize and predict individually experienced temperature and humidity by combining with neighborhood microclimate differences in an urban vs. rural setting, together with individual-level factors that could potentially affect vulnerability to heat stress. This study aims to improve extreme heat event response by improving the characterization of who, when, and where people are at a risk of exposure to excess heat index.

Methods

Participant recruitment and individual monitor deployment

In the summer of 2017, residents in Birmingham, AL (N=90) and Wilcox County, AL (N=90) were screened and recruited in partnership with Friends of West End, Birmingham AL and West Central Alabama Community Health Improvement League, Camden AL. Sample size was determined to detect a mean difference between urban and rural exposure of 0.7°C, with a standard deviation of 1.7°C, if we assume that the average temperature in rural areas is 26.0°C based on our pilot study [14]. Eligibility criteria include women aged 19–66 and availability to participate in a one-week study between July 10–19, 2017. We recruited woman participants to reduce variability for primary exposure variables of interest and to improve the ability to recruit and follow-up with participants based on previous studies conducted by the community-academic partnership [14]. We excluded participants having medical conditions or taking medication that could prevent them from spending time outdoors or being out of town during the study period. Potential participants attended an informational enrollment session, provided written consent, and filled out demographic and employment questionnaires. We collected the height and weight of participants with a Befour Inc. Model #PS660 (Befour Inc., Wisconsin, USA) scale and a fold-up height stick. We collected body composition measurements of participants with a Tanita BC-553 (Tanita Corporation of America, Inc., Illinois, USA) portable body composition scale. Participants were asked to perform normal activity in the first two days and spend an additional 30 minutes outdoors beyond their normal activities on days 3–7 of participation. Participants kept a daily log of their time spent outdoors and pedometer readings, and they completed an exit survey. We made three follow-up phone calls to participants during the study to troubleshoot any challenges with compliances to wearing the monitors and filling out the daily logs. Participants received up to $150 to cover the time and travel expenses associated with participation in the study.

Each participant wore an iButton® thermometer (model# DS1922L from Maxim Integrated, California, USA) clipped on the shoe and a pedometer, Yamax Digi-Walker (model# SW-200 from Yamax, Texas, USA), clipped at the waist. The iButton devices were clipped facing down to avoid direct sunlight. iButton is a computer chip enclosed in a 16mm thick stainless steel can, and it has a temperature resolution of ±0.5°C from −10°C to 65°C with an operating temperature range from −40°C to +85°C [24]. iButton thermometers were factory calibrated using NIST standards [24]. In previous applications we have confirmed that factory-calibrated iButtons return consistent values when tested in a common environment [25–27]. iButton thermometers were set to record temperature every five minutes. At turn-in sessions, we downloaded data from the iButton data loggers and gave a printout of individual results to each participant. We stored all data collected on password-protected computers for subsequent analysis. Informed consent was obtained from all participants. This study is registered at clinicaltrials.gov (NCT03614780) and was approved by Virginia Tech Institutional Review Board (15–761).

Nearby WS data collection and processing

We accessed meteorological data, including air temperature, relative humidity, wind speed, and location coordinates during the study period from all WSs in AL from the National Climate Data Center Surface Data, Hourly Global dataset (DS3505) (http://cdo.ncdc.noaa.gov). We matched a nearest WS to each participant’s home address. We matched six WSs to participant home addresses: Bessemer Airport WS, Birmingham International Airport WS, Craig Field WS, Demopolis Municipal Airport WS, Mac Crenshaw Memorial Airport WS, and Middleton Field Airport WS. We calculated hourly WS heat index (HI[WS]) (°C) from hourly temperatures and relative humidity from the WS by using “weathermetrics” packages in R based on the methods in Supplementary Figure 1 [28].

Neighborhood monitor deployment

A total of 43 iButton thermometers/hygrometers (model# DS1923 from Maxim Integrated, California, USA) were deployed in participants’ neighborhoods in Birmingham (N=29) and in Wilcox County (N=14). Each iButton was placed in a radiation shield [25] and affixed to a tree at various locations (e.g., in yards or along sidewalks). We measured the exact location coordinates of each neighborhood iButton using smartphone Global Positioning System. Neighborhood iButtons were set to measure air temperature and relative humidity hourly. We matched a nearest neighborhood monitor to each participant’s home address and calculated hourly neighborhood heat index (HI[neighborhood]) (°C) from the hourly temperatures and relative humidity from neighborhood iButton data loggers based on the methods in Supplementary Figure 2.

Participant iButton temperature measurement data processing

A total of 178 participant thermometers (89 in Birmingham and 89 in Wilcox County) had valid temperature measurements at turn-in. Thermometer temperature measurements outside each participant’s study period based on check-in/check-out session time were excluded from the analysis (decision tree presented in Supplementary Figure 3). We removed upper outliers of the participant thermometer temperature to remove potential artifacts of iButton measurements. We calculated hourly individually experienced heat index, HI[individual] (°C), from participant thermometer hourly averaged temperatures and matched WS relative humidity, based on the methods in Supplementary Figure 4. To test the effects of using different sources of relative humidity in estimating HI[individual], we calculated another set of HI[individual] using neighborhood monitor relative humidity in the place of WS relative humidity. For each person-hour of HI[individual], we matched a HI[WS] from the nearest WS and a HI[neighborhood] from the nearest neighborhood monitor to indicate the ambient conditions. This resulted in 27,470 person-hours of HI estimates in an Inclusive dataset for following analysis.

We performed sensitivity analysis using Complete dataset in which no outliers were removed. We additionally performed sensitivity analysis using Restrictive1 and Restrictive2 datasets that included the data from person-days on which the participant thermometer temperature variation at 5am-midnight was greater than 1°C or 2°C, respectively. In an attempt to reduce artifacts of shoe iButton temperatures, we applied a −0.5°C bias correction to all the participants thermometer temperatures taken at 8am-8pm in the Bias Corrected dataset when participants were more likely to wear shoes with a thermometer clipped on in the Inclusive dataset. A sensitivity analysis was performed in the Bias Corrected dataset. The methods of obtaining these additional datasets are presented in Supplementary Figure 4.

HI risk classification

We assigned a risk level to each person-hour based on the National Weather Service HI classification (https://www.weather.gov/ama/heatindex). A HI of 26.7°C or lower is classified as Safe, 26.7–32.2°C as Caution, 32.2–39.4°C as Extreme Caution, 39.4–51.1°C as Danger, >51.1°C as Extreme Danger. We performed a time-series analysis of the risk categories by using HI[participant], HI[neighborhood] and HI[WS].

Pedometer readings processing

Participants recorded their pedometer reading at night on each day without resetting the pedometer. We estimated daily steps based on the pedometer readings recorded on daily logs. Building from the previous decision tree [27], we modified it by removing invalid data based on daily log notes from participants and replacing extreme daily steps <1,000 or ≥25,000 with NA (Supplementary Figure 5) [29, 30] and used it to further process the pedometer data adjusting for missing or unrealistic recorded data. A sensitivity study evaluated the effects of pedometer data processing methods.

Data analysis

We examined the diurnal pattern of daily average and max HI[individual] in different groups. Only person-days with 24 hours data were included in the daily average analysis. We performed a sensitivity analysis on individually experienced temperature in place of HI.

We fitted linear mixed models to determine factors significantly associated with HI[individual] exposures of participants. Models included a random effect term, allowing us to account for multiple measurements from a single person. The dependent variable, HI[individual], was modeled using independent variables of HI[WS] or HI[neighborhood], and covariates including: age, income (≥ $20,000 vs. < $20,000 reported annual household income), education (≥ high-school diploma or equivalent vs. < high school diploma or equivalent), measured body fat (%), weekend, non-rest time (5am-midnight), hourly WS mean wind speed (m/s), intervention, log(mean steps), employment, and rural or urban setting. Whether HI[WS] or HI[neighborhood] explained more variance in models of HI[individual] was evaluated using the Akaike information criterion (AIC) for each model. The AIC of the three models (HI[WS] and HI[neighborhood], HI[neighborhood] only, HI[WS] only) were computed and the Δi = AICi − AICmin were calculated. The model best estimated has the Δi ≡ AICmin ≡ 0. When Δi ≤ 2, there is no substantial difference between the two models and a simpler model was preferred [31]. Models were stratified by urban/rural setting and across the occupationally and non-occupationally exposed groups. Suspecting indoor air-conditioning and human activity were potential reasons for daily fluctuations in individually experienced HI, we fitted the regression models separately for non-rest time (5am-midnight) and rest time (midnight-5am). Measured body mass index (BMI) and measured body fat (%) were highly correlated, so only measured body fat (%) was added in the models [14]. Seven participants (1,062 person-hour observations) were dropped from the analysis because of missing measured body fat (%), reported annual household income and education data. We used “lmer” function in “lme4” package in R to run the models [32]. We performed a sensitivity analysis on the Complete dataset, Restrictive1 and Restrictive2 datasets.

Results

Participant demographic characteristics

Characteristics of the final study population are presented in Table 1 and participant enrollment and follow-up are presented in the CONSORT flow diagram in Supplementary Figure 6. Participants were female and most of them identified as Black or African American. Thirty-two participants in Birmingham were outdoor workers (Urban OutWor). Urban OutWor participants were on average significantly younger than other Urban residents (p-value 0.03) and had higher mean daily steps than Urban residents (5,782 vs. 4,548 steps, p-value 0.02). The adopted pedometer processing method had a minimal effect on daily steps (Supplementary Figure 7). Urban OutWor participants on average had a lower measured body fat (%) (p-value 0.04) compared to other Urban residents. Rural residents on average had a higher measured body fat (%) (p-value = 0.04) compared to Urban residents. There was no significant difference in access to central air-conditioning at home, education level, annual household income level, BMI, and obesity prevalence between groups.

Table 1.

Participant demographics and characteristics.

Settings Urban setting Urban setting Rural setting
Group Urban OutWor p-value(1)a Urban residents p-value(2)a Rural residents
Participant number 32 NA 57 NA 88
Median age (range), years 39.5 (21–60) 0.03* 45 (20–69) 0.17 54 (19–67)
Gender-Female 32 (100%) NA 57 (100%) NA 88 (100%)
% Black or African American 30 (94%) NA 55 (96%) NA 88 (100%)
Employed 32 (100%) 8.84E–05* 34 (60%) 0.06 37 (42%)
Pedometer daily steps (Mean, range) 5782 (1713–10676) 0.02* 4548 (1080–7872) 1.00 4546 (1257–12740)
Central air conditioning at home
  Yes 12 (38%) 0.62b 36 (63%) 0.06b 21 (24%)
  No 6 (19%) 11 (19%) 17 (19%)
  Missing data 14 (44%) 10 (18%) 50 (57%)
Education
  < High School Diploma (or Equivalence) 14 (44%) 0.677b 29 (51%) 0.73b 40 (45%)
  ≥ High School Diploma (or Equivalence) 18 (56%) 28 (49%) 46 (52%)
  Missing data 0 (0%) 0 (0%) 2 (3%)
Annual household income
  < $20,000 22 (69%) 0.98b 37 (65%) 1b 57 (65%)
  ≥ $20,000 10 (31%) 19 (33%) 28 (32%)
  Missing data 0 (0%) 1 (2%) 3 (3%)
Body measurement
  BMI (Median, range) from check in session 34.31(19.3–52.3) 0.19 35.8 (24.7–60.3) 0.57 36.6 (19.5–64.8)
  Obese (BMI ≥30.0) from check in session 22 (69%) 0.42 45 (79%) 0.97 71 (81%)
  Body fat % (Median, range) from check in session 42.4 (22.9–52.8) 0.04 * 45.2 (25.7–54.7) 0.04 * 47.3 (25.3–70.6)
Open in a new tab

Note:

‘*’

denotes a statistically significant difference with p-values < 0.05. NA not applicable.

a

p-value(1) was obtained from comparison between Urban residents and Urban OutWor, p-value(2) was obtained from comparison between Urban residents and Rural residents in the same category.

b

Pearson’s Chi-squared test for available data only.

We compared the characteristics of the study participants in Birmingham and Wilcox County with the characteristics of populations residing in these two places, respectively. A higher proportion of the study participants were self-identified as African American or Black than Birmingham or Wilcox County census reports (95% vs 71% in Birmingham, 100% vs 71% in Wilcox County). The participants in this study were more highly educated (94% vs 86% high school graduates and above in Birmingham, 88% vs 77% high school graduates and above in Wilcox County), and less wealthy (59 [66%] of participants reported <$20,000 vs $35,346 median annual household income in Birmingham, 57 [65%] of participants reported <$20,000 vs $27,237 median household income in Wilcox) compared to US Census estimates [33].

Distances between participant residence and neighborhood iButtons/WS

The nearest neighborhood thermometer and the nearest WS were matched to each participant’s home address (Supplementary Figure 8), and the summarized distances are shown in Table 2. Neighborhood monitors were significantly closer to participants than WS in the three population groups. Rural residents were on average significantly farther away from nearby WS compared to Urban residents (50.12 kilometers [95%CI (48.61, 51.63)] vs. 12.37 kilometers [95%CI(11.28, 13.46)], p-value < 2.2E-16).

Table 2.

Distance between participant home and the nearest neighborhood monitor and the nearest WS.

Group Distance to participant home (kilometer) Mean and 95% confidence interval
Rural residents Neighborhood monitor 3.64 (2.59, 4.70)
WS 50.12 (4.86, 5.16)
Urban residents Neighborhood monitor 4.47 (2.89, 6.06)
WS 12.37 (11.28, 13.46)
Urban OutWor Neighborhood monitor 4.39 (2.52, 6.26)
WS 10.77 (8.81, 12.73)
Open in a new tab

Comparison of HI from participants, neighborhood and WS

We examined the diurnal patterns of HI[individual], HI[neighborhood] and HI[WS] over the 8 days of the study. The group average of individually experienced mean and maximum HI in the three population groups are shown in Figure 1 and Figure 2.

Figure 1.

Figure 1.

Open in a new tab

Diurnal pattern of maximum and mean HI[individual] (yellow triangle) compared to HI[neighborhood] (gray dot) and HI[WS] (blue square) in Rural residents (participant N=88, neighborhood iButton N=13, WS N=4), Urban residents (participant N=57, neighborhood iButton N=18, WS N=2), and Urban OutWor (participant N=32, neighborhood iButton N=11, WS N=2). The 95% confidence intervals were shown. HI[individual] was calculated from participant iButton temperatures and the closet WS relative humidity. Each participant wore the monitor for 7 days.

Figure 2.

Figure 2.

Open in a new tab

Date pattern of maximum and mean HI[individual] (yellow triangle) compared to the matched HI[neighborhood] (gray dot) and HI[WS] (blue square) in Rural residents (participant N=88, neighborhood iButton N=13, WS N=4), Urban residents (participant N=57, neighborhood iButton N=18, WS N=2), and Urban OutWor participants (OutWor) (participant N=32, neighborhood iButton N=11, WS N=2). The 95% confidence intervals were shown. HI[individual] was calculated from participant temperature and WS relative humidity. Each participant wore the monitor for 7 days. Only person-days with 24 hours of individually experienced HI data are included and 2,492 person-hours were excluded compared to Figure 1 dataset. July 15th and July 16th were weekends.

Among Rural residents and Urban residents in Figure 1, we found that the HI[WS] and the HI[neighborhood] underestimated the population-averaged HI[individual] at night (approximately 8pm-8am) for both mean and max HI. Max HI[WS] was similar to max HI[individual] during the day (8am-8pm) in Rural residents and Urban residents, respectively. HI[WS] deviated significantly less from HI[individual] than HI[neighborhood] between 10am-3pm among Rural residents in both mean and max HI. We examined the outdoor occupation effect by comparing HI[individual] in Urban residents and Urban OutWor participants. Urban OutWor participants on average had significantly higher max and mean HI[individual] during the day but similar HI[individual] at night compared to Urban residents. Both HI[WS] and HI[neighborhood] underestimated the population average max HI[individual] in Urban OutWor participants. Figure 2 shows that the daily mean HI[WS] and HI[neighborhood] overall overestimate the population average of HI[individual] in Rural residents and Urban residents. The daily max HI[neighborhood] estimated daily max HI[individual] better than daily max HI[WS] in all groups. Urban OutWor participants on average had a significantly lower daily mean and max HI[individual] on weekends. Analyses with temperature only, instead of estimated HI, from thermometers clipped on shoes, neighborhood monitors and WS were performed and presented in Supplementary Figure 9–10, with similar trends to those presented for HI.

The risk distribution patterns based on HI are shown in Figure 3. Participant thermometers, neighborhood monitors and WS reported similar population mean frequency of person-hours (%) in the Safe exposure category. When we looked at specific unsafe exposure categories, we found that neighborhood monitors and WS significantly underestimated the Extreme Danger category exposure in all groups and significantly underestimated the Caution category exposure in Rural residents only. Across all participants, we found that the HI[individual] estimated 1,420 out of 27,470 (5.17%) person-hours from 148 participants in Danger and Extreme Danger categories, respectively, which peaked between 1pm-3pm (Supplementary Figure 11). Of these 1,420 person-hours, HI[neighborhood] and HI[WS] estimated 406 (28.59%) and 453 (31.90%) person-hours in the Safe category, respectively. When extreme values for HI[individual] at night (N=142 person-hours) were removed in sensitivity analyses, the results were consistent with the main results, with an estimated 1,312 out of 27,328 (4.80%) person-hours from 148 participants in Danger and Extreme Danger categories (Supplementary Figures 12–15).

Figure 3.

Figure 3.

Open in a new tab

Mean frequency % of risk classification based on heat index in Rural residents (N=88), Urban residents (N=57), and Urban OutWor (N=32). Frequency % = Frequency of each risk classification/total person-hours of each participant. The frequency % in each risk classification was averaged in three population groups, respectively. The 95% confidence intervals were shown.

Predictive value of WS HI and neighborhood HI

A model including HI[neighborhood], but not including HI[WS], with the additional dependent variables remaining unchanged (and listed in Table 3), had the lowest AIC (Table 3). When models are stratified by group, Urban residents followed the full model, while for the models with only Rural residents and Urban OutWor participants, the inclusion of both HI[WS] and HI[neighborhood] improved the prediction of the individually experienced heat index, as measured by model AIC.

Table 3.

Linear mixed model fixed effect predictors of individually experienced HI in 24 hours.

Population All Rural residents Urban residents Urban OutWor
HI[neighborhood], HI[WS] AIC 159992.60 75521.80 51745.30 31346.60
HI[WS] only AIC 160200.10 75619.80 51809.00 31368.70
HI[neighborhood] only AIC 159990.60 75529.00 51743.40 31389.00
Model pick HI[neighborhood] only Both HI[neighborhood] only Both
Fixed effects β (95%CI) β (95%CI) β (95%CI) β (95%CI)
Intercept 17.66 (8.83, 26.49) 19.32 (6.25, 32.39) 15.39 (−0.14, 30.93) 12.07 (−5.03, 29.19)
HI[neighborhood] (°C) 0.20 (0.19, 0.21)* 0.17 (0.13, 0.20)* 0.20 (0.18, 0.23)* 0.20 (0.12, 0.28)*
HI[WS] (°C) NA −0.06 (−0.10, −0.02)* NA 0.33 (0.23, 0.42)*
Age 0.01 (−0.02, 0.04) 0.00 (−0.05, 0.04) 0.04 (−0.02, 0.09) −0.01 (−0.09, 0.08)
Education ≥high school −0.20 (−1.05, 0.66) −0.68 (−1.96, 0.59) −0.10 (−1.66, 1.46) 1.07 (−0.51, 2.65)
Annual household income ≥$20K −0.60 (−1.53, 0.34) −0.09 (−1.50, 1.33) −0.29 (−2.09, 1.50) −1.53 (−3.35, 0.29)
Weekend −0.50 (−0.65, −0.36)* −0.29 (−0.47, −0.10)* 0.08 (−0.17, 0.33) −1.63 (−2.03, −1.23)*
Body fat (%) −0.05 (−0.12, 0.02) 0.05 (−0.06, 0.16) −0.10 (−0.22, 0.03) −0.15 (−0.25, −0.04)*
Non-rest time (5am-midnight) 1.37 (1.21, 1.53)* 0.65 (0.42, 0.87)* 1.56 (1.29, 1.82)* 2.55 (2.12, 2.98)*
WS wind speed (m/s) 0.25 (0.20, 0.30)* 0.31 (0.23, 0.38)* 0.29 (0.21, 0.36)* 0.23 (0.11, 0.36)*
Intervention −0.26 (−0.40, −0.11)* −0.30 (−0.50, −0.11)* −0.11 (−0.35, 0.14) −0.70 (−1.09, −0.31)*
log(mean daily steps) 0.67 (−0.24, 1.58) 0.46 (−0.84, 1.77) 0.86 (−0.79, 2.51) 0.90 (−0.89, 2.70)
Employed −0.07 (−1.05, 0.90) −0.83 (−2.15, 0.49) 0.89 (−0.71, 2.48) NA
Rural residenta 0.43 (−0.56, 1.43) NA NA NA
Outdoor workera 1.22 (−0.05, 2.50) NA NA NA
Open in a new tab

Note:

a

Compared to Urban resident.

‘*’

denotes a β estimates with a 95% confidence interval did not contain 0.

NA not applicable. NA in HI[WS] indicated that HI[WS] was not included in model selection based on AIC.

Among all participants, results showed that HI[neighborhood] was significantly associated with HI[individual]. For a 1°C increase in HI[neighborhood], the mean of HI[individual] increases by 0.20°C [95%CI (0.19, 0.21)]. WS wind speed (m/s), as an additional environmental factor, was positively associated with HI[individual]. Participants had a 0.50°C [95%CI (0.36, 0.65)] lower mean HI[individual] during weekends and a 1.37°C [95%CI (1.21, 1.53)] higher mean HI[individual] in non-rest time (5am-midnight) in models accounting for changes in ambient conditions as measured at the nearest neighborhood monitors. This weekend and the non-rest time effect on HI[individual] was largely driven by relationships seen in data from Urban OutWor participants. Rural residents had a 0.43°C [95%CI (−0.56, 1.43)] higher mean HI[individual] compared to Urban residents after adjustment for differences in ambient outdoor temperatures captured by neighborhood monitors. Urban OutWor participants had a 1.22°C [95%CI (−0.05, 2.50)] higher mean HI[individual] compared to urban non-outdoor worker participants.

In models stratified by non-rest time (5am-midnight) and rest-time (midnight −5am), only HI[neighborhood] was included in the model selection based on AIC (Table 4-5). The increased HI experienced in the rural location was more pronounced at rest time (0.10°C [95%CI (−0.85, 1.04)] during non-rest time vs. 1.22°C [95%CI (−0.25, 2.69)] during rest-time). Urban OutWor participants on average had a 1.71°C [95%CI (0.49, 2.92)] higher mean HI[individual] than Urban residents during non-rest time but a 0.18°C [95%CI (−1.71, 2.07)] lower mean HI[individual] at rest-time (Table 4-5).

Table 4.

Linear mixed model fixed effect predictors of individually experienced HI in non-rest time (5am-midnight).

Population All Rural residents Urban residents Urban OutWor
HI[neighborhood], HI[WS] AIC 121158.50 57253.60 39113.50 23785.20
HI[WS] only AIC 121332.30 57340.30 39161.00 23804.00
HI[neighborhood] only AIC 121156.80 57260.40 39111.60 23813.50
Model pick HI[neighborhood] only Both HI[neighborhood] only Both
Fixed effects β (95%CI) β (95%CI) β (95%CI) β (95%CI)
Intercept 17.91 (9.51, 26.30) 19.06 (6.98, 31.14) 16.75 (1.88, 31.63) 10.60 (−6.53, 27.78)
HI[neighborhood] (°C) 0.20 (0.19, 0.22)* 0.16 (0.13, 0.20)* 0.20 (0.17, 0.22)* 0.21 (0.12, 0.30)*
HI[WS] (°C) NA −0.06 (−0.10, −0.02)* NA 0.30 (0.20, 0.41)*
Age 0.01 (−0.02, 0.04) 0.00 (−0.04, 0.04) 0.03 (−0.02, 0.09) −0.01 (−0.10, 0.07)
Education ≥high school −0.16 (−0.97, 0.66) −0.47 (−1.65, 0.71) 0.01 (−1.49, 1.50) 0.93 (−0.65, 2.52)
Annual household income ≥$20K −0.74 (−1.62, 0.15) −0.16 (−1.46, 1.15) −0.78 (−2.50, 0.94) −1.09 (−2.91, 0.75)
Weekend −0.59 (−0.76, −0.41)* −0.28 (−0.50, −0.06)* 0.20 (−0.11, 0.51) −2.32 (−2.82, −1.81)*
Body fat (%) −0.06 (−0.13, 0.004) 0.04 (−0.06, 0.14) −0.10 (−0.23, 0.02) −0.16 (−0.27, −0.05)*
WS wind speed (m/s) 0.30 (0.24, 0.35)* 0.31 (0.24, 0.39)* 0.35 (0.26, 0.44)* 0.26 (0.12, 0.41)*
Intervention −0.42 (−0.59, −0.24)* −0.42 (−0.65, −0.20)* −0.25 (−0.55, 0.05) −0.89 (−1.38, −0.40)*
log(mean daily steps) 0.88 (0.02, 1.74)* 0.62 (−0.59, 1.82) 0.97 (−0.62, 2.55) 1.52 (−0.28, 3.31)
Employed 0.14 (−0.79, 1.06) −0.81 (−2.03, 0.41) 1.43 (−0.09, 2.96) NA
Rural residenta 0.10 (−0.85, 1.04) NA NA NA
Outdoor workera 1.71 (0.49, 2.92)* NA NA NA
Open in a new tab

Note:

a

Compared to Urban resident.

‘*’

denotes a β estimates with a 95% confidence interval did not contain 0.

NA not applicable. NA in HI[WS] indicated that HI[WS] was not included in model selection based on AIC.

Table 5.

Linear mixed model fixed effect predictors of individually experienced HI in rest time (midnight-5am).

Population All Rural residents Urban residents Urban OutWor
HI[neighborhood], HI[WS] AIC 34925.50 17118.30 11224.00 6456.10
HI[WS] only AIC 34979.10 17162.30 11228.20 6455.70
HI[neighborhood] only AIC 34925.20 17117.50 11222.10 6456.00
Model pick HI[neighborhood] only HI[neighborhood] only HI[neighborhood] only HI[WS] only
Fixed effects β (95%CI) β (95%CI) β (95%CI) β (95%CI)
Intercept 16.43 (3.22, 29.64) 10.78 (−9.14, 30.68) 15.38 (−8.12, 38.89) 30.26 (4.69, 55.84)
HI[neighborhood] (°C) 0.36 (0.28, 0.44)* 0.51 (0.37, 0.65)* 0.28 (0.15, 0.41)* NA
HI[WS] (°C) NA NA NA 0.30 (0.15, 0.46)*
Age 0.01 (−0.04, 0.06) −0.01 (−0.09, 0.06) 0.05 (−0.03, 0.14) 0.01 (−0.11, 0.14)
Education ≥high school −0.30 (−1.57, 0.97) −1.35 (−3.26, 0.57) −0.38 (−2.72, 1.96) 1.44 (−0.91, 3.78)
Annual household income ≥$20K −0.23 (−1.61, 1.15) 0.07 (−2.04, 2.19) 1.17 (−1.53, 3.86) −2.90 (−5.59, -
0.21)*
Weekend −0.20 (−0.38, −0.02)* −0.41 (−0.68, −0.13)* −0.23 (−0.52, 0.05) 0.16 (−0.22, 0.54)
Body fat (%) −0.01 (−0.11, 0.09) 0.09 (−0.07, 0.25) −0.08 (−0.28, 0.11) −0.11 (−0.27, 0.05)
WS wind speed (m/s) 0.05 (−0.05, 0.15) 0.18 (−0.05, 0.41) −0.07 (−0.20, 0.06) −0.06 (−0.24, 0.12)
Intervention 0.17 (−0.01, 0.36) −0.05 (−0.38, 0.27) 0.46 (0.18, 0.75)* 0.15 (−0.23, 0.54)
log(mean daily steps) 0.12 (−1.23, 1.46) 0.17 (−1.78, 2.13) 0.51 (−1.97, 2.99) −0.86 (−3.52, 1.80)
Employed −0.69 (−2.14, 0.75) −0.83 (−2.81, 1.15) −0.78 (−3.17, 1.61) NA
Rural residenta 1.22 (−0.25, 2.69) NA NA NA
Outdoor workera −0.18 (−2.07, 1.71) NA NA NA
Open in a new tab

Note:

a

Compared to Urban resident.

‘*’

denotes a β estimates with a 95% confidence interval did not contain 0.

NA not applicable. NA in HI[WS] or HI[neighborhood] indicated that HI[WS] or HI[neighborhood]was not included in model selection based on AIC.

Regression analyses with temperature, instead of HI, from thermometers clipped on shoes, neighborhood monitors and WS in mixed models were performed and presented in Supplementary Tables 1–3. The inclusion of both neighborhood temperature and WS temperature improved the characterization of the hourly mean participant thermometer temperature based on model AIC. Overall, factors associated with individually experienced temperature were consistent with HI results presented in Table 3-5.

We observed differences in relative humidity measurements from WS and neighborhood monitors (Supplementary Figure 16). The effect of using relative humidity from neighborhood to calculate HI[individual] was minimal compared to using relative humidity from WS (Supplementary Figure 17). Sensitivity results of individually experienced HI in the Bias-corrected dataset are presented in Supplementary Table 4–7 and Supplementary Figure 17. On average, HI[individual] was 0.44 °C [95%CI (−0.1°, 2.91)] lower after the correction, and there is no significant difference in the mean risk category frequency (%) of participants in the three population groups (Supplementary Table 4, Supplementary Figure 17). Mixed model regression results using corrected HI[individual] as the response showed minimal differences to the main results, and the same significant predictors were obtained (Supplementary Table 5–7). Additional sensitivity analyses performed examined outlier removal and noncompliance removal in characterizing individually experienced HI diurnally (Supplementary Figure 18), daily average (Supplementary Figure 19), risk classification (Supplementary Figure 20), and mixed models to identify significant predictors (Supplementary Table 8). The results were consistent with the results we have obtained from the Inclusive datasets. Results of the model including pedometer readings <1,000 or ≥25,000 (Supplementary Table 9) were similar compared to the main model results presented in Table 3 and did not alter our main findings on the association between HI[individual] and HI[neighborhood] and/or HI[WS].

Discussion

This study aims to examine whether temperature and humidity measured at nearby WS are appropriate for predicting heat stress risk and whether neighborhood heat index improved the prediction of individually experienced heat index in an urban vs. rural setting. We found that neighborhood level temperature and humidity measurement, in addition to nearby WS measurements, improved the characterization and prediction of the heat index experienced by individuals. In Rural residents and Urban residents, neighborhood heat index measurements explained more variability in individually experienced heat index compared to WS measurements (Table 3). When both HI[neighborhood] and HI[WS] were included to explain the variance in HI[individual] in Rural residents, the effect size of HI[neighborhood] (0.17°C ) was greater compared to that of HI[WS] (−0.06°C), indicating a 1°C increase of HI[neighborhood] is associated with a greater change in HI[individual] compared to a 1°C increase of HI[WS]. It is interesting that a 1°C increase of HI[WS] was associated with a 0.06°C [95%CI (0.02, 0.10)] lower HI[individual] when adjusting for HI[neighborhood] and other individual level factors. A possible explanation is that as shown in Figure 1, HI[WS] was higher in the mornings compared to late afternoon and evenings while HI[individual] was a little higher in late afternoons and evenings in the rural location. This result may reflect adaptations by participants to avoid the hottest parts of the day by spending more time indoors with air conditioning between 10am and early afternoon and spending time outdoors or in non-air-conditioned indoor space when it was cooler.

We also found that there was a time window (approximate 8am-8pm) when HI[individual] was significantly overestimated by both WS and neighborhood monitors; outside this time window, however, HI[individual] was significantly underestimated (Figure 1). Dwelling characteristics including thermal mass [34], no or limited air conditioning use, insufficient night ventilation, human presences, and heat emitting objects such as a working burner/television/refrigerator within the indoor environment are likely reflected in the results showing underestimation of HI[individual] by neighborhood and regional monitors at night. Rural residents on average experienced a higher heat index exposure than Urban residents, even after controlling for differences in ambient conditions, particularly during rest time compared to non-rest time (1.22°C vs. 0.10°C). This suggests that relative to urban residents, rural residents may experience heightened risk to dangerous heat index exposure at a given ambient heat index measured at a nearby WS, particularly during the nighttime.

In urban outdoor worker participants, WS measurements explained more variance in HI[individual] compared to neighborhood monitors. This could be explained by outdoor work occurring away from residences and likely closer to downtown areas. Being an outdoor worker was associated with a 1.71°C [95%CI (0.49, 2.92)] higher mean individually experienced heat index in non-rest time, confirming that outdoor workers are likely at an increased risk of reaching dangerous exposure levels compared to non-outdoor workers in an urban setting. Interestingly, outdoor workers in this study had a 0.18°C [95%CI (−1.71, 2.07)] lower mean individually experienced heat index exposure in rest time, suggesting this population may preferentially seek cooler settings during non-work times to recuperate from daytime heat index exposure.

We also found that neighborhood level heat index measurements served as a better predictor of HI[individual] risk classification than WS data (Figure 3). The risk of overexposure to heat index would be underestimated if WS data alone were used. This result in residents was consistent with the results for outdoor workers presented in the study by Wang et al. (2019), which showed that WS data alone would recommend fewer person-hours into the most protective work-rest schedules compared to individually experienced temperatures [19]. It is not surprising to find neighborhood heat index was a better predictor of HI[individual], considering neighborhood monitors were much closer to participants’ homes and other studies have shown that microclimate conditions varies spatially on local neighborhood scales [35]. As continuously sampling individually experienced heat index of everyone who may be at risk in a large population is unrealistic, neighborhood microclimate measurement is a feasible step forward in downscaling heat index exposure measurement data generated at weather stations. Neighborhood heat index measurements could be used by public health professionals to better identify neighborhoods for interventions to minimize exposure during extreme heat events.

Our finding that WS HI was significantly associated with individually experienced HI was consistent with the findings revealed in the study by Bernhard et al. (2015) and Mac et al. (2018) [14, 15]. However, body fat (%) and income level, which were identified as significant factors associated with individually experienced temperatures in both urban and rural participants in the study by Bernhard et al. (2015), were only identified as significant factors among outdoor workers in this study. Considering the differences in results across these studies, we think it is worthwhile to include these factors in future studies on individually experienced heat index exposure.

Identification of who, when, and where people are truly exposed to extreme heat during heat wave days is critical for the effectiveness of heat warning systems. The effectiveness of heat warning systems depends on accurate weather forecasts of high temperature and humidity, appropriate heat-stress thresholds to trigger alerts, efficient mobilization of local agency actions, and communication and confirmation that affected residents change their behavior in response to alerts [8]. While regional WS monitoring is a reliable and accessible source of temperature and humidity measurement, we explored more factors influencing exposure experienced by individuals in an urban vs. rural setting while using WS measurements to account for ambient heat index. These factors included neighborhood-level temperature and humidity, human behaviors, and body measurements. We found a rural setting and outdoor worker occupation are factors associated with higher individually experienced heat index. Among outdoor workers, we found age and body fat (%) are negatively associated with HI[individual] while physical activity was positively associated with HI[individual], most likely through behavior modification effects. Our results support the integration of risk factors including body metrics and physical activity to customize personal WBGT exposure, such as the Heat Shield platform designed for outdoor workers [36]. We also found that weekdays, non-rest time (5am-midnight) and higher wind speed are associated with higher individually experienced heat index exposure. By incorporating these identified factors above and potentially more factors influencing heat index exposure at the individual level, we can better predict places and times dangerous heat index exposure may be experienced by individuals.

Our findings that neighborhood level heat indexes are better predictors of individually experienced heat indexes suggest additional measurement and more accurate accounting of variations due to microclimates in alert systems could help local agencies and neighborhood leadership better identify vulnerable residents and target neighborhood level mitigation strategies such as greening, which is likely a primary driver of neighborhood-level temperature differences in urban areas [26]. Additionally, neighborhood level measurements could help to further tailor warning messages of extreme heat health consequences and encourage resident engagement and behavior change in response to warnings [37, 9, 38, 39]. Findings from the present study suggest tailored messaging should focus on indoor environments as well as outdoor conditions, with elevation or reduction of risk messaging considering household risk factor status (e.g., elderly, dwelling without air conditioning, poor fitness, overexertion during work or leisure, minimal time spent outdoors).

There were some limitations of this study. A few factors influencing the air temperature measured by the thermometer clipped to participant’s shoe could not be quantified. For example, solar radiation, radiation reflected by grounds/floors, type of shoe (sandals vs. leather shoes vs. cloth shoes), were not accounted for in this study. Solar radiation and body heat can increase the temperatures measured by thermometers on the shoe. Still, we placed the thermometer on the shoe to maximize comfort and minimize body heat impact. The position of the iButton facing down on the shoe reduced the solar radiation impact but may not be able to eliminate it. A radiation shield could be designed for use in future studies to reduce radiation impact on thermometer temperatures. The upper outlier removal and bias correction analyses were attempts to remove these artifacts. In sensitivity analyses, the outlier removal and bias correction effects were minimal in linear mixed models and the characterization of risk classifications (Supplementary Table 5–8 and Supplementary Figure 18–20). We did not use the wet bulb globe temperature (WBGT), a more comprehensive environmental heat measurement that incorporates temperature, humidity, wind speed, and solar radiation [40, 41], to estimate the individually experienced heat in this study because solar radiation exposure was difficult to estimate based on the limited data available to estimate indoor versus outdoor locations for each person-hour. We found two participants continuously experienced HI in Danger and Extreme Danger categories at night (Supplementary Figure 12–13), which may be due to lack of air-conditioning at home, insufficient indoor ventilation, or potential measurement error. The impact of these person-hours on the main results were minimal in risk classification in the sensitivity analysis (Supplementary Figure 15). Potential sources of artificial temperature spikes, such as proximity to warm objects (e.g., working electronics), need be examined in future studies examining nighttime heat index exposure. The influence of daily steps on the relationship between HI[individual] and HI[neighborhood] and/or HI[WS] were minimal in the present analysis (Table 3 and Supplemental Table 9), however we note limitations in the use of pedometers for estimating daily steps [42–45]. Use of accelerometers in future studies would provide a more accurate representation of the influence to steps on individually experienced heat index [46, 47].

In conclusion, neighborhood heat index measurement improved the prediction of individually experienced heat index in addition to WS measurement in the rural setting, and neighborhood heat index alone served as a better predictor in the urban setting. After adjusting for ambient environmental conditions, individually experienced heat indexes in the rural setting were on average 0.43°C [95%CI (−0.56, 1.43)] higher than that in the urban setting. WS and neighborhood heat index measurements significantly underestimated the Extreme Danger category exposure in all groups and significantly underestimated the Caution category exposure in rural participants only. Weekdays, non-rest time (5am-midnight), and higher wind speed were significantly associated with higher individually experienced heat indexes. This study is a novel approach to estimate individually experienced heat index exposure in an urban vs. rural setting considering regional weather, local microenvironments, and human behaviors. Understanding individually experienced heat index can help improve public health strategies to minimize adverse health outcomes associated with extreme heat. The incorporation of human behavior data beyond daily steps, more biomedical and health condition data, and indoor/outdoor time differentiation into the characterization of individually experienced heat index exposure is an important next step. This could reveal further factors influencing individually experienced heat index which can be used to improve the prediction of individually experienced heat index exposure.

Supplementary Material

1

Acknowledgements

This project was funded through a grant from the National Institute of Environmental Health Sciences (R01ES023029). We gratefully acknowledge collaboration with Sheila Tyson, Keisha Brown, and Nakeia Pullman (Friends of West End), and Sheryl Threadgill-Mathews and Ethel Johnson (West Central Alabama Community Health Improvement League), for their aid in recruitment and implementation of the research. Thanks to Mary Evans, Anna Scott, Michael Milazzo, Pranavi Ghugare, Kaya Bryant, and Claudiu Lungu for help with the data collection.

Footnotes

Supplementary information is available at Journal of Exposure Science & Environmental Epidemiology’ website.

Competing Financial Interests

The authors declare they have no actual or potential competing financial interests.

Conflicts of interests

The authors declare no conflict of interest.

References

  • 1.Reidmiller DR, Avery CW, Easterling DR, Kunkel KE, Lewis KL, Maycock TK et al. Impacts, risks, and adaptation in the United States: Fourth national climate assessment, Volume II. Washington, DC: US Global Change Research Program. 2018. [Google Scholar]
  • 2.Gasparrini A, Guo Y, Hashizume M, Kinney PL, Petkova EP, Lavigne E et al. Temporal variation in heat–mortality associations: a multicountry study. Environmental health perspectives. 2015;123(11):1200–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Bobb JF, Peng RD, Bell ML, Dominici F. Heat-related mortality and adaptation to heat in the United States. Environmental health perspectives. 2014;122(8):811–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.NWS. National Weather Service Summary of Natural Hazard Statistics for 2018 in the United States. 2019. https://www.weather.gov/media/hazstat/sum18.pdf. Accessed September 21 2020.
  • 5.Berko J. Deaths attributed to heat, cold, and other weather events in the United States, 2006–2010. vol 76. US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Health Statistics; 2014. [PubMed] [Google Scholar]
  • 6.CDC. QuickStats: number of heat-related deaths, by sex—National Vital Statistics System, United States, 1999–2010. MMWR Morb Mortal Wkly Rep. 2012;61(36):729. [Google Scholar]
  • 7.Choudhary E, Vaidyanathan AJM. Heat stress illness hospitalizations—Environmental public health tracking program, 20 states, 2001–2010. J Morbidity Mortality Weekly Report: Surveillance Summaries. 2014;63(13):1–10. [PubMed] [Google Scholar]
  • 8.McGregor GR and Bessmoulin P and Ebi K and Menne B. (2015) ‘Heatwaves and health: guidance on warning-system development.’,. World Meteorological Organisation and World Health Organisation. https://www.who.int/globalchange/publications/heatwaves-health-guidance/en [Google Scholar]
  • 9.Lowe D, Ebi KL, Forsberg B. Heatwave early warning systems and adaptation advice to reduce human health consequences of heatwaves. International journal of environmental research and public health. 2011;8(12):4623–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.NWS. Heat. Naitional Weather Service. https://www.weather.gov/bgm/heat. Accessed November 17 2019.
  • 11.NWS. NWS Experimental HeatRisk: Identifying Potential Heat Risks in the Seven Day Forecast. https://www.wrh.noaa.gov/wrh/heatrisk/. Accessed September 7 2020.
  • 12.CDC. Heat Stress Risk Factors. 2017. https://www.cdc.gov/niosh/mining/UserFiles/works/pdfs/2017-125.pdf. Accessed November 21 2019.
  • 13.Grundstein AJ, Ramseyer C, Zhao F, Pesses JL, Akers P, Qureshi A et al. A retrospective analysis of American football hyperthermia deaths in the United States. International journal of biometeorology. 2012;56(1):11–20. [DOI] [PubMed] [Google Scholar]
  • 14.Bernhard MC, Kent ST, Sloan ME, Evans MB, McClure LA, Gohlke JM. Measuring personal heat exposure in an urban and rural environment. Environmental research. 2015;137:410–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Mac V, Hertzberg V, & McCauley LA (2019). Examining Agricultural Workplace Micro and Macroclimate Data Using Decision Tree Analysis to Determine Heat Illness Risk. Journal of occupational and environmental medicine, 61(2), 107–114. 10.1097/JOM.0000000000001484 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Sugg MM, Fuhrmann CM, Runkle JD. Temporal and spatial variation in personal ambient temperatures for outdoor working populations in the southeastern USA. International journal of biometeorology. 2018;62(8):1521–34. [DOI] [PubMed] [Google Scholar]
  • 17.Uejio CK, Morano LH, Jung J, Kintziger K, Jagger M, Chalmers J et al. Occupational heat exposure among municipal workers. International archives of occupational and environmental health. 2018;91(6):705–15. [DOI] [PubMed] [Google Scholar]
  • 18.Kuras E, Hondula D, Brown-Saracino J. Heterogeneity in individually experienced temperatures (IETs) within an urban neighborhood: insights from a new approach to measuring heat exposure. International journal of biometeorology. 2015;59(10):1363–72. [DOI] [PubMed] [Google Scholar]
  • 19.Wang S, Richardson MB, Wu CY, Cholewa CD, Lungu CT, Zaitchik BF et al. Estimating Occupational Heat Exposure From Personal Sampling of Public Works Employees in Birmingham, Alabama. Journal of occupational environmental medicine. 2019;61(6):518–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Li Y, Odame EA, Silver K, & Zheng S. (2017). Comparing urban and rural vulnerability to heat-related mortality: a systematic review and meta-analysis. J Glob Epidemiol Environ Health 1 (1): 9–15. DOI: 10.29199/GEEH, 101016, 3. [DOI] [Google Scholar]
  • 21.Martiello MA, Giacchi MV. High temperatures and health outcomes: a review of the literature. J Scandinavian journal of public health. 2010;38(8):826–37. [DOI] [PubMed] [Google Scholar]
  • 22.Gabriel KM, Endlicher WR. Urban and rural mortality rates during heat waves in Berlin and Brandenburg, Germany. J Environmental pollution. 2011;159(8–9):2044–50. [DOI] [PubMed] [Google Scholar]
  • 23.Runkle J Kunkel K, Stevens L, and Frankson R. 2017: Alabama State Climate Summary. NOAA, NOAA Technical Report NESDIS 149-AL. 2017. https://statesummaries.ncics.org/chapter/al/. Accessed December 2 2019.
  • 24.Maxim. DS1922L iButton Temperature Loggers with 8KB Data-log Memeory. 2018. https://www.maximintegrated.com/en/products/digital/data-loggers/DS1922L.html.
  • 25.Scott AA, Misiani H, Okoth J, Jordan A, Gohlke J, Ouma G et al. Temperature and heat in informal settlements in Nairobi. PLoS One. 2017;12(11):e0187300. doi: 10.1371/journal.pone.0187300. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Scott AA, Zaitchik B, Waugh DW, O’Meara K. Intraurban temperature variability in Baltimore. Journal of Applied Meteorology and Climatology. 2017;56(1):159–71. [Google Scholar]
  • 27.Richardson MB, Chmielewski C, Wu CYH, Evans MB, McClure LA, Hosig KW et al. The effect of time spent outdoors during summer on daily blood glucose and steps in women with type 2 diabetes. J Behav Med. 2020;43(5):783–90. doi: 10.1007/s10865-019-00113-5. [DOI] [PubMed] [Google Scholar]
  • 28.Anderson G, Peng R. weathermetrics: Functions to convert between weather metrics (R package). 2012. [Google Scholar]
  • 29.Tudor-Locke C, Bassett DR, Shipe MF, McClain JJ. Pedometry methods for assessing free-living adults. Journal of Physical Activity and Health. 2011;8(3):445–53. [DOI] [PubMed] [Google Scholar]
  • 30.Samara A, Sanna T, Algasser D, Alzahrani M, Aro AR. Pedometers as an effective tool for measuring physical activity in young females in Saudi Arabia. International Journal of Community & Family Medicine. 2017;2:133. [Google Scholar]
  • 31.Burnham KP, Anderson DR. Multimodel inference: understanding AIC and BIC in model selection. Sociological methods & research. 2004;33(2):261–304. [Google Scholar]
  • 32.Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. arXiv preprint arXiv:14065823. 2014. [Google Scholar]
  • 33.UnitedStatesCensusBureau. QuickFacts Birmingham city, Alabama; Wilcox County, Alabama. United States Census Bureau. https://www.census.gov/quickfacts/fact/table/birminghamcityalabama,wilcoxcountyalabama/PST045219. Accessed Feburary 13 2020. [Google Scholar]
  • 34.Kubota T, Chyee DTH, Ahmad S. The effects of night ventilation technique on indoor thermal environment for residential buildings in hot-humid climate of Malaysia. Energy and buildings. 2009;41(8):829–39. [Google Scholar]
  • 35.Reid CE, O’neill MS, Gronlund CJ, Brines SJ, Brown DG, Diez-Roux AV et al. Mapping community determinants of heat vulnerability. J Environmental health perspectives. 2009;117(11):1730–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Morabito M, Messeri A, Noti P, Casanueva A, Crisci A, Kotlarski S et al. An Occupational Heat–Health Warning System for Europe: The HEAT-SHIELD Platform. International journal of environmental research and public health. 2019;16(16):2890. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Toloo G, FitzGerald G, Aitken P, Verrall K, Tong S. Evaluating the effectiveness of heat warning systems: systematic review of epidemiological evidence. International journal of public health. 2013;58(5):667–81. [DOI] [PubMed] [Google Scholar]
  • 38.Kalkstein AJ, Sheridan SC. The social impacts of the heat–health watch/warning system in Phoenix, Arizona: assessing the perceived risk and response of the public. International journal of biometeorology. 2007;52(1):43–55. [DOI] [PubMed] [Google Scholar]
  • 39.Ebi KL, Schmier JK. A stitch in time: improving public health early warning systems for extreme weather events. Epidemiologic reviews. 2005;27(1):115–21. [DOI] [PubMed] [Google Scholar]
  • 40.NIOSH. NIOSH criteria for a recommended standard: occupational exposure to heat and hot environments. In: J B, W WJ, M K, C A, K J-H, T N, editors.: Cincinnati, OH: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Institute for Occupational Safety and Health, DHHS (NIOSH) Publication; 2016. [Google Scholar]
  • 41.Liljegren JC, Carhart RA, Lawday P, Tschopp S, Sharp R. Modeling the wet bulb globe temperature using standard meteorological measurements. J Occup Environ Hyg. 2008;5(10):645–55. doi: 10.1080/15459620802310770. [DOI] [PubMed] [Google Scholar]
  • 42.Bassett DR, Ainsworth BE, Leggett SR, Mathien CA, Main JA, Hunter DC et al. Accuracy of five electronic pedometers for measuring distance walked. Medicine and science in sports and exercise. 1996;28(8):1071–7. [DOI] [PubMed] [Google Scholar]
  • 43.Schneider PL, Crouter SE, Lukajic O, Bassett DR Jr. Accuracy and reliability of 10 pedometers for measuring steps over a 400-m walk. Medicine and science in sports and exercise. 2003;35(10):1779–84. [DOI] [PubMed] [Google Scholar]
  • 44.Melanson EL, Knoll JR, Bell ML, Donahoo WT, Hill J, Nysse LJ et al. Commercially available pedometers: considerations for accurate step counting. Preventive medicine. 2004;39(2):361–8. [DOI] [PubMed] [Google Scholar]
  • 45.Washburn R, Chin MK, Montoye HJ. Accuracy of pedometer in walking and running. Research Quarterly for Exercise and Sport. 1980;51(4):695–702. [DOI] [PubMed] [Google Scholar]
  • 46.Le Masurier GC, Tudor-Locke C. Comparison of pedometer and accelerometer accuracy under controlled conditions. Medicine and science in sports and exercise. 2003;35(5):867–71. [DOI] [PubMed] [Google Scholar]
  • 47.Foster RC, Lanningham-Foster LM, Manohar C, McCrady SK, Nysse LJ, Kaufman KR et al. Precision and accuracy of an ankle-worn accelerometer-based pedometer in step counting and energy expenditure. Preventive medicine. 2005;41(3–4):778–83. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1667949-supplement-1.pdf (3.9MB, pdf)

RESOURCES