Using Wearable Devices to Measure Personal Heat Exposure in a Preconception Cohort

Abstract

Background:

Heat is a risk factor for several adverse pregnancy outcomes. Most epidemiologic studies use outdoor air temperature as a proxy for personal heat exposure, but outdoor measures may not reflect personally experienced temperature.

Objective:

We conducted a personal heat monitoring sub-study within a preconception cohort to describe personal heat exposure, evaluate the correlation between personal and outdoor temperatures, and investigate factors that explain observed variation in correlations.

Methods:

Pregnancy Study Online (PRESTO) is a web-based preconception cohort study of pregnancy planners. We used wearable monitors to collect up to one month of personal temperature and humidity data every 10 minutes from 152 participants living in the contiguous U.S. in May-October of 2021 and 2023. We calculated personal daily heat metrics, including temperature and heat index, to measure the intensity, frequency, and duration of heat exposure. We compared individual metrics to residence-based outdoor temperatures and estimated Spearman correlation coefficients across and within participants. Finally, we stratified distributions of within-person correlations by time spent outdoors, home air conditioning (AC) use, neighborhood median income, and climate region.

Results:

Participants resided in 40 states and most (94%) had access to AC at home. The median daily personal mean temperature was 22.9°C (IQR: 21.7-24.0). Four percent of participants’ temperature readings exceeded a local extreme heat threshold (>90^th county percentile). The correlation between daily mean personal and outdoor temperatures was low in the full cohort (r=0.33), as was the median within-participant correlation (r=0.34; IQR: 0.08-0.57). Time spent outdoors and AC use only modestly explained variability in these correlations.

Significance:

In this U.S.-based pregnancy planning cohort, personal temperatures were generally moderate, although some participants experienced brief exposures to extreme heat. Personal and outdoor temperatures were not highly correlated, and variability in within-participant correlations was not appreciably explained by measured factors.

Introduction

Exposure to extreme heat contributes to more deaths annually than all other weather-related hazards combined.¹ Heat-related deaths have reached nearly 10,000 fatalities annually in the United States (US).² Extreme heat directly causes poor outcomes such as dehydration and sunstroke,³ and is also associated with increased risk of a range of adverse health outcomes (e.g., cardiovascular, renal, and respiratory disease, occupational injury and psychological distress).^3-7 Extreme heat exposure is expected to increase in the coming years due to continued climate change.⁸

Most epidemiologic studies assessing the health consequences of heat exposure use outdoor air temperature as a proxy for personal heat exposure. Estimates of outdoor air temperature are available globally at high temporal and spatial resolutions from dynamic modeling of satellite-derived data, weather stations, or complex space-time modeled datasets and can be easily merged with health data at almost any geographic resolution. However, measures of outdoor temperature at a fixed location may not accurately reflect the temperatures people actually experience,⁹ particularly in the US where most people spend the majority of their time indoors.¹⁰ Personal temperature depends on individual time-activity patterns (e.g., time spent outdoors or away from the home), characteristics of the home and work environments, built and natural features influencing microclimates, and seasonal and regional weather patterns.¹¹ The use of outdoor temperature measures in epidemiologic studies, particularly in study populations with access to air conditioning or other cooling resources, could potentially lead to substantial exposure measurement error and biased estimates of the health effects of extreme heat.

The use of wearable monitors for measuring personal temperature is an important advancement in research on health risks of extreme heat.¹¹ Several studies have implemented personal monitoring to measure personal temperature, providing unique insights.^12-26 For example, among 23 residents of South Boston, substantial differences in heat exposure were documented despite their geographic proximity to each other.¹² Among 180 women in Alabama, the correlation between personal and neighborhood outdoor heat index varied across urban vs. rural counties.¹³ Similarly, a study of 42 elderly residents of Baltimore found personal temperatures that were consistently lower than outdoor temperatures, with little to no correlation between them.¹⁴ Several studies have reported high personal heat exposures among outdoor workers and have generally found that correlations between individual and outdoor temperatures are higher during hours spent outside.^15-17 While these studies have provided essential preliminary data on how personal temperature correlate with outdoor temperatures in different contexts, they have been limited by small sample sizes (from 15-180), geographic homogeneity, and/or restrictions to populations at high risk of heat exposure.

No existing studies, to our knowledge, have measured personal temperatures in a population of individuals who are pregnant or trying to conceive, despite known links between extreme heat and pregnancy health,^27-32 and potential differences in acceptable risk or behavior modification compared with the general population. To fill this gap, we conducted a large-scale warm-season personal temperature monitoring sub-study nested within a preconception cohort of pregnancy planners residing across the US. Our specific objectives were to 1) describe personal temperature exposure in this population, 2) evaluate the correlation between outdoor and personal temperatures, and 3) identify factors that explain variation in this correlation.

Methods

Study Population

Pregnancy Study Online (PRESTO) is a web-based preconception cohort study that has been recruiting North American pregnancy planners since 2013.³³ Recruitment occurs primarily online through social media advertising. Eligible participants are aged 21-45 years, residents of the US or Canada, assigned female at birth, and trying to conceive without the use of fertility treatment. Participants complete a screening questionnaire, online consent form, and baseline questionnaire that ascertains information on socio-demographics, lifestyle factors, and medical and reproductive histories. Participants then complete follow-up questionnaires every 8 weeks for up to 12 months and, among those who conceive, additional follow-up questionnaires (early pregnancy, late pregnancy, and postpartum). Participants are also given the option to invite their partners aged ≥21 years and assigned male at birth to complete a one-time baseline questionnaire. This study was reviewed and approved by the Institutional Review Board at the Boston University Medical Campus. All participants provided informed consent.

Personal Temperature Monitoring Sub-Study

During the warm season (May-October) of 2021 (pilot period) and 2023 (field period), we invited PRESTO participants to enroll in a personal temperature monitoring sub-study. Eligible participants included the subset of newly enrolled PRESTO participants who resided in the contiguous US and had been trying to conceive for <3 menstrual cycles at enrollment. We sent invitations to participants 1 hour after completion of the baseline questionnaire. We invited 107 participants to enroll during the pilot period and 217 participants to enroll during the field period. Of the 324 invited participants, 239 (74%) consented to participate (72% from the pilot period and 75% from the field period) and were mailed a Kestrel DROP D2 wireless environmental data logger (Boothwyn, PA, USA), a personal wearable monitor that records environmental exposure data including ambient temperature (F) (+/− 0.9°F) and heat index (F) (+/− 7.1°F) at ten-minute intervals. The data loggers did not record GPS or location information. Participants were instructed to keep the monitor in the same physical space as their body for one week (pilot period) or one month (field period). We encouraged participants to wear the monitor on a lanyard or to clip the monitor to their backpack or purse. We advised participants that measurements are most accurate when there is some airflow over the monitor sensor and indicated that the monitor should not be carried in the participant’s pocket, inside a backpack or purse, or up against the participant’s skin. We sent participants reminder emails every three days during the monitoring period to ensure continued proper use. At the end of the monitoring period, participants completed a brief end-of-study questionnaire to indicate the date and time at which they started and stopped using the monitor and how often they used the monitor (response options: almost all of the time, most of the time, more than half of the time, or less than half of the time), and mailed their monitors back to the study team.

Data Cleaning

We received monitors back from 162 participants (68%) and downloaded all 10-minute interval observations. We first converted each participant’s data into local hours according to their time zone and truncated the data according to the participant’s reported date and time that they started and stopped wearing the monitor. We excluded 5 participants who did not wear their monitors during the study period, 3 participants who started wearing their monitors in November, after our warm season study period, and 1 participant whose monitor malfunctioned and did not record dates or temperature metrics (n=9 participants excluded). We also excluded from analysis any days with fewer than 20 hours of data recorded (resulting in exclusion of n=1 more participant from the analytic dataset). We winsorized extreme outliers in the raw data to a maximum temperature of 55°C, which is the highest temperature reading the Kestrel monitors are capable of measuring. In the final analytic sample, we included 3,338 participant-days contributed by 152 participants (Web Figure 1).

Outdoor Air Temperature Data

We spatially joined the latitude and longitude of each participant’s residential address with measures of outdoor air temperature from the European Centre for Medium-Range Weather Forecast (ECMWF) Reanalysis v5 Land product (ERA5-Land), which provides global meteorological estimates at approximately 9-km gridded resolution.^34,35 We used residential address since we did not have GPS information to track where they spent time. ERA5-Land data is a land-only replay of the ERA5 reanalysis dataset, which dynamically models temperatures and other atmospheric variables from several data sources to produce continuous temperature estimates over land.³⁶ These data have been evaluated against ground truth points and have been found to perform particularly well in the US.³⁷ Additionally, ERA5-Land data are one of the most commonly used sources of temperature data given its global spatial extent, hourly temporal resolution, and availability of multiple temperature-related metrics making the results of this analysis more broadly generalizable compared to the use of other US-specific datasets, such as PRISM, GridMET, etc. that may exist as smaller scales but with limited extent.

We obtained the gridded hourly data for each participant’s residential address during the period of personal temperature monitoring. For analyses of daily temperatures, we aggregated hourly data to the daily mean and maximum. We also used a derived dataset originally sourced from the same ERA5-Land data to identify a locally relevant threshold for extreme heat in each participant’s community,³⁸ defined as the 90^th percentile of daily maximum temperatures in the ten years prior to our study period using population-weighted county-level estimates in each participant’s county of residence. The population-weighted county estimates are intended to represent the typical climate of where people live, work, and play, which tends to span relatively far distances from residential addresses. After cleaning and merging, there was no missingness in the final analysis dataset.

Statistical Analysis

We calculated personal daily temperature metrics in the population using four metrics designed to capture the intensity, frequency, and duration of exposure, informed by prior research.²¹ First, we calculated daily mean personal temperature (mean PT), defined as the mean of all 10-minute interval temperature observations measured throughout each person-day. We calculated daily maximum personal temperatures (maximum PT) after aggregating to the mean of temperature readings within each hour to avoid the influence of potentially spurious outliers within the raw temperature readings. Second, we similarly calculated mean and maximum daily personal heat index (PHI) – a measure which combines both ambient temperature and relative humidity³⁹ – from 10-minute or hourly aggregated monitor readings. Third, we calculated the percentage of daily personal temperature readings above the county-specific 90^th percentile of temperature (percentage of minutes above threshold; PMAT). Finally, we defined the longest exposure period (LEP) as the longest number of continuous minutes in each day that a participant’s personal temperatures were above the county-specific 90^th percentile. The mean and maximum PT and PHI were designed to capture the intensity of heat exposure; PMAT was designed to capture the frequency of exposure; and LEP was designed to capture the duration of exposure. We refer to calculated temperature metrics as “personal temperature” , whereas previous studies have used “individually experienced temperature”;²¹ both refer to the direct measurement of the ambient temperature that an individual is exposed to in their direct surroundings.

For each exposure metric, we presented the distribution of exposure in the population on the scale of person-days. We present distributions for all hours and stratified by time of day (day: 6am to 9pm; night: 9pm to 6am). We also present distributions restricted to hot days, defined as days with maximum outdoor temperature >90^th percentile for the participant’s residential county. Lastly, we calculated Spearman correlation coefficients between each of the daily heat metrics. All personal temperature metrics are presented here in degrees Celsius, but results in degrees Fahrenheit are shown in the supplement (Web Table 1), along with results of a secondary analysis assessing mean and maximum PHI (Web Tables 2).

Next, we compared personal temperatures with outdoor temperatures. First, we calculated Spearman correlation coefficients between daily personal temperatures and daily outdoor temperatures across all person-days for both mean and maximum temperature. We visually assessed these relationships using scatter plots. Although most person-days were recorded in July and August, because we were primarily interested in warm season exposures, we also stratified observations by sub-season: May/June, July/August, and September/October. We calculated the Spearman correlation coefficients within each participant (i.e. person-specific correlations) and plotted the distribution of person-specific correlations using histograms. We also generated Bland-Altman plots, which plot the difference between personal and outdoor temperatures on the y-axis and the average of the two measures on the x-axis, allowing us to explore the extent to which agreement varied systematically across the temperature spectrum.

Lastly, to examine the extent to which correlations between personal and outdoor temperatures varied by participant characteristics, we stratified the person-specific correlation coefficients by four variables selected to represent individual, household, neighborhood, and regional features. The four stratification variables included 1) average number of hours each participant spent outdoors during a typical summer day (ascertained from the PRESTO baseline questionnaire and categorized as ≤3, 4-7, and >7 hours per day based on the 25^th and 75^th percentiles); 2) presence and use of air conditioning (AC) at home (ascertained from the PRESTO baseline questionnaire and categorized as “AC present at home and used often or all of the time” vs. combined categories of “AC not present in the home” and “AC present at home but used occasionally or only sometimes”); 3) census tract median household income (obtained from the American Community Survey 2018–2022 5-year estimates and categorized into quartiles); and 4) climate region (defined by the National Climate Assessment [NCA] 5^th edition⁴⁰ and excluding regions with fewer than 10 participants). Additional stratifications by personal household income and positive pregnancy test after study enrollment (ascertained on follow-up and early pregnancy questionnaires) can be found in the supplement (Web Figure 2).

All analyses were conducted in R Studio using R Version 4.2.0 and the following packages: dplyr⁴¹, lubridate⁴², sf⁴³, lutz⁴⁴, tableone⁴⁵, corrplot⁴⁶, ggdist⁴⁷, and weathermetrics³⁹. We present an overview of the conducted analyses in the supplement (Web Table 3).

Results

We analyzed data from 152 participants who returned their monitors. These participants were more likely to be non-Hispanic White, had higher educational attainment and income, but were less likely to have ever been pregnant compared to those who did not return their monitors (Web Table 4). Eighty percent (80%) of the analysis cohort reported wearing their monitors most or all of the time, and only 2 participants reported wearing their monitors for less than half the time. Participants lived in all NCA climate regions of the contiguous US (spanning 40 states) and had a median age of 32 years (Table 1). Around 79% of participants identified as Non-Hispanic White and almost half reported having a graduate degree. The median household income in census tracts where participants resided was $61,500, though a little over half of participants reported a personal household income of $100,000 or greater. Nearly all participants (94.1%) reported access to AC at home, though slightly fewer participants (84.5%) reported using AC often. The median number of hours participants reported spending outdoors on a typical summer day was 2.2.

Table 1.

Baseline characteristics of the personal temperature monitoring sub-study sample, PRESTO, 2021-2023.

Characteristic n (%); median [IQR]	Participants N=152
Year data collected
2021 (pilot study period)	44 (28.9)
2023 (field study period)	108 (71.1)
Month monitoring started
May	10 (6.6)
June	35 (23.0)
July	49 (32.2)
August	15 (9.9)
September	28 (18.4)
October	15 (9.9)
Age, years	32.0 [28.0, 34.2]
Ever pregnant	88 (57.9)
Conceived after study enrollment (before or during monitoring period)	30 (19.7)
Race and ethnicity
Non-Hispanic White	121 (79.6)
Hispanic	13 (8.6)
Non-Hispanic Multiracial	8 (5.3)
Non-Hispanic Black	6 (3.9)
Non-Hispanic Asian	4 (2.6)
Education
Graduated high school or obtained GED	6 (3.9)
Some college	26 (17.1)
College degree	45 (29.6)
Graduate school	75 (49.3)
Personal annual household income (USD)
<$50,000	23 (15.1)
$50,000 - $99,999	42 (27.6)
$100,000 - $149,999	37 (24.3)
$150,000 - $199,999	17 (11.2)
≥200,000	33 (21.7)
Census tract median household income [quartiles]
$20,800 - $62,999	38 (25.0)
$63,000 - $91,499	38 (25.0)
$91,500 - $113,999	38 (25.0)
$114,000 - $204,599	38 (25.0)
NCA Climate Region**
Midwest	38 (25.0)
Northeast	25 (16.4)
Northern Great Plains	4 (2.6)
Northwest	4 (2.6)
Southeast	43 (28.3)
Southern Great Plains	15 (9.9)
Southwest	23 (15.1)
AC presence and summertime use at home
AC not present	9 (5.9)
AC present but used only sometimes	17 (11.1)
AC present and used often	131 (84.5)
Average hours per day spent outside in summer	2.2 [1.3-3.0]

^**Climate regions with fewer than 10 participants were excluded from our climate-region-stratified analysis due to concern about identifiability. Climate regions were defined using the National Climate Assessment (NCA), fifth edition.

Abbreviations:

AC = Air Conditioning

GED = General Educational Development

The median of daily mean personal temperatures across all participant-days was 22.9°C (IQR: 21.7, 24.0), or 73.2°F (IQR: 71.1, 75.3). The value was similar when restricted to outdoor hot days (23.4°C, IQR: 22.3, 24.6; 74.2°F, IQR: 72.2, 76.4). Daily maximum personal temperatures were also similar between all days, and the subset of hot days (Table 2). Overall, 18.3% of person-days recorded at least one hour of temperature exposure greater than the locally defined threshold for extreme heat. However, across the monitoring duration, only 4% of participants’ personal temperature readings exceeded the local threshold for extreme heat (PMAT), though this metric was 7% when looking only on hot outdoor days. The median LEP was 0 minutes (IQR: 0, 20, range: 0—1440) across all days and was highly left-skewed, with the majority (64.3%) of person-days measuring zero minutes. Participants may have been slightly more likely to experience longer durations of extreme heat on days when the outdoor temperature exceeded the local 90^th temperature percentile (median: 0 minutes, IQR: 0, 40, range: 0—1440).

Table 2:

Summary of daily heat metrics for full cohort for all days and for hot days (°C)

	Exposure metric	Mean	Standard deviation	Minimum	25th percentile	Median	75th percentile	Maximum
All participant-days (n=3,356)	Mean PT (°C)	22.9	2.1	11.9	21.7	22.9	24.0	37.1
	Maximum PT (°C)	26.8	4.0	17.2	24.1	26.3	28.7	54.7
	PMAT (%)	3.9	0.1	0.0	0.0	0.0	1.4	100
	LEP (minutes)	42.9	154.9	0.0	0.0	0.0	20.0	1440.0
Participant-days with outdoor extreme heat (n=912)	Mean PT (°C)	23.7	2.1	17.9	22.3	23.4	24.6	37.1
	Maximum PT (°C)	27.8	4.0	19.9	24.8	27.4	29.7	49.1
	PMAT (%)	7.3	0.2	0.0	0.0	0.0	3.7	100
	LEP (minutes)	87.1	257.9	0.0	0.0	0.0	40.0	1440.0

NOTE: All metrics are calculated as person-days. Outdoor extreme heat was defined as days with a maximum outdoor air temperature greater than the county-specific 90^th percentile. The personal heat metrics used were adapted from metrics proposed in Hondula et al. 2021.

PT: Personal Temperature

PMAT: Percentage Minutes Above Threshold

LEP: Longest Exposure Period

Comparing the different personal heat metrics for intensity, frequency, and duration of heat exposure with each other, daily mean temperature and daily mean heat index were highly correlated (r=0.97), as were daily maximum temperature and daily maximum heat index (r=0.97). PMAT and LEP were highly correlated (r=1.0), but only moderately correlated with daily mean and maximum metrics (Web Figure 3). Graphical distributions of personal heat metrics, stratified by day and night hours, are shown in the supplement (Web Figures 4, 5).

The median of daily outdoor temperatures on days during which participants wore their monitors was 23.1°C (IQR: 19.8, 26.0) and ranged from a minimum of 3.7°C and a maximum of 36.0°C. The correlation between personal and outdoor daily temperatures was modest using both mean and maximum estimates. The correlation between daily mean personal and outdoor temperatures was r=0.33 across participants (Figure 1); the correlation was slightly lower for daily maximums (r=0.27). When we stratified by sub-season, the correlation between daily mean personal and outdoor temperatures was r=0.35, r=0.14, and r=0.57 for May/June, July/August, and September/October, respectively. Personal temperatures tended to be warmer than outdoor temperatures at the lower range of outdoor temperatures and tended to be cooler than outdoor temperatures at the higher range of outdoor temperatures (Figure 2). Personal and outdoor temperatures also varied between day and night; hourly personal temperatures were on average 0.8°C cooler than outdoor temperatures during the daytime and 1.9°C warmer than outdoor temperatures at night (Web Figure 7).

Figure 1: Correlations between personal and outdoor temperatures (°C), across all participants using both mean and maximum temperatures.

NOTE: These panels display scatterplots where daily personal temperatures are plotted against outdoor temperatures. The panel on the left contains information on daily mean temperatures whereas the panel on the right shows the daily maximum temperatures. The overall Spearman correlation coefficients between the daily personal and outdoor temperatures across all individuals are presented.

Figure 2:

Bland-Altman plots showing the difference between daily personal and outdoor temperatures (°C) by their average, using both mean and maximum daily metrics

Person-specific correlations between personal and outdoor temperatures ranged widely across participants (for mean temperatures: median=0.34, IQR=0.08, 0.57; for maximum temperature: median=0.23, IQR=0.12, 0.49; Figure 3, Web Figure 7). Person-specific correlations tended to be slightly higher among individuals who did not own or did not use AC and those who spent more time outside, particularly when looking at mean daily temperatures. For example, the median correlation was 0.51 (IQR=−0.02, 0.68) among those who did not use AC, and 0.32 (IQR=0.09, 0.52) among those who did (Figure 4). We also saw some differences between geographic areas, with individuals residing in the Southeast showing much lower correlations compared to those in the Southwest. However, we did not find substantive differences in person-specific correlations across strata of neighborhood median household income. Similarly, additional stratifications by personal household income and positive pregnancy test before or during the monitoring period did not show differences in person-specific correlations (Web Figure 2).

Figure 3: Distributions of person-specific correlations between daily personal and outdoor temperatures.

NOTE: These panels show the distribution of person-specific Spearman correlation coefficients between daily personal and outdoor temperatures. The panel on the left contains information on daily mean temperatures whereas the panel on the right shows the daily maximum temperatures.

Figure 4: Comparison of person-specific correlation between personal vs. outdoor temperatures by stratifying conditions for mean and maximum daily temperatures.

NOTE: Panel A shows the distribution of person-specific Spearman correlation coefficients between the daily mean personal and outdoor temperatures, while panel B shows the same for daily maximum personal and outdoor temperatures, by stratifying conditions.

Discussion

In this preconception cohort study of pregnancy planners in the US we found that personal temperature was only modestly associated with outdoor air temperatures, although there was substantial variability in person-specific correlations and by sub-season, with the lowest correlation found during the high summer months of July and August. Individual, household, neighborhood, and regional characteristics did not appreciably explain this variability. Overall, personal extreme heat exposure was relatively low in this cohort of pregnancy planners, both in frequency and duration.

On average, study participants did not experience significant heat exposure, despite a wide range of outdoor temperatures of up to 36.0°C, and with 27.3% of monitored days with outdoor temperatures exceeding the 90^th percentile of local temperatures. Most participants reported having AC at home and using it often in the summer, potentially explaining why the correlation between personal and outdoor temperatures was lowest in July and August. Only 4% of participants’ temperature readings exceeded their locally relevant threshold for extreme heat, and the median temperature experienced among participants across all days was similar to what was experienced on the hottest days. However, almost 20% of person-days were characterized by at least one hour of personal temperature of above the local threshold for extreme heat. This indicates that even among individuals with access to cooling resources, there may be short intervals of exposure to extreme heat. These exposures could stem from outdoor exercise or recreation, commuting via public transit or by walking, rolling, or other active transport, or doing chores or work outside or in hotter indoor environments. This is particularly relevant for pregnancy planners because some markers of reproductive potential – notably, sperm quality – can be affected by even short-term exposure to extreme heat.^48,49

When we compared personal temperatures to outdoor temperatures estimated concurrently at participants’ residential addresses, we found modest correlations using both mean and maximum daily temperature metrics. When mean outdoor temperatures were low, personal temperatures tended to be warmer than outdoor temperatures, whereas when outdoor temperatures were high, personal temperatures tended to be cooler than outdoor temperatures. These results suggest that members of this population are able to achieve adequate climate control much of the time. However, we did not see the same pattern when comparing daily maximum personal versus outdoor temperatures: average differences between maximum outdoor and personal temperatures were relatively similar across the temperature spectrum. This could reflect the fact that on days when the outdoor temperature is hot, participants can experience short durations of extreme heat, perhaps due to sun exposure or exertion, occasionally experiencing temperatures even hotter than outdoor air temperatures which are measured in the shade (e.g., the high outlier points in the top right of the maximum temperature Bland-Altman plot in Figure 2). This can occur even in a population that typically has the means to mitigate their exposure. This may suggest that access to AC and the means to pay for the electricity consumption can reduce but may not completely eliminate exposure to extreme heat.

The average person-specific correlations between outdoor and personal temperatures were generally modest, but we observed wide variability across participants (range: −0.89 – 1.00, IQR: 0.08, 0.57). Identification of factors that explain this variability could be instrumental for improving exposure assessment in epidemiologic studies of the potential adverse health effects of heat. We explored several potential factors operating at multiple levels (personal, household, neighborhood, and regional) that could plausibly explain the variability in person-specific correlations. We found that individuals who did not have or use AC tended to have slightly higher correlations between personal and outdoor temperature measurements, on average. However, differences were moderate and varied across participants, which might indicate that the participants who do not have or use AC tend to be those that do not need it, at least in this population.

The largest difference in correlation was seen in the season stratified analysis, where we found the highest correlation for observations measured in September/October (r=0.57) and the lowest for those in July/August (r=0.14). This difference may be due to changes in daily activities during summer vs. fall months related to school or work, though they may also suggest some behavior modification happening in the hottest months to avoid heat exposure, such as A/C usage.

Similarly, there was only a slight trend towards higher personal and outdoor temperature correlations among those who spent more time outside in the summer. This is not entirely aligned with previous studies that have reported stronger correlations between personal and outdoor temperatures during hours when individuals are working outside.^16,17 There are several potential explanations for why we did not observe larger differences by time spent outdoors in the present study. First, the weak correlation in our study may reflect beneficial behavior modification: participants in our study who spend more time outdoors may have the choice to do so at times where indoor and outdoor temperatures are similar. This might be particularly applicable for a preconception cohort who may be more cognizant of potential health risks as they attempt to conceive. Second, it may be that the few hours that individuals reported spending outside each day (median=2.2 hours per day in the summer) are not enough to counteract the rest of their time spent indoors. Finally, our measure of time outdoors was asked on the baseline questionnaire and reflects an average over the course of the summer, rather than time spent outdoors during the monitoring period, and may be subject to recall bias.

The small differences seen between within-person correlation coefficients and neighborhood median household income or geographic region were also unexpected. We hypothesized that community investment might moderate the effect of outdoor heat given the urban heat island effects that are most commonly associated with lower-income communities.⁵⁰ Given that study participants tended to have higher income and education than the general population and lived in census tracts with relatively high median incomes, low-income neighborhoods were underrepresented in this analysis. There were some differences seen in correlations across climate regions, with those residing in the Southwest exhibiting higher correlations on average, and those residing in the Southeast exhibiting the lowest. The fact that larger differences were not shown could be further evidence of acclimatization in which people have adopted strategies that are effective to mitigate their personal heat exposure.

The multi-level factors measured in this study did not explain a large proportion in the variability in person-specific correlations between personal and outdoor temperatures. While we understand some of the risk and protective factors for heat exposure – such as the provision and use of AC – our results suggest that these measured factors do not sufficiently explain differences in experienced heat in this study population. This has significant implications for considering what interventions might be most impactful for reducing personal heat, particularly during preconception and pregnancy. Additional studies are needed to more comprehensively identify factors that are predictive of personal exposure, both to inform impactful interventions and to determine whether it is possible to predict personal temperature from outdoor measured temperature and personal characteristics for large-scale heat and health research without direct personal heat monitoring. While we believe that the results of this analysis are relevant for all studies on the health effects of extreme heat, future research should investigate whether personal and ambient outdoor temperatures are more highly correlated in the general population than among pregnancy planners. This is particularly relevant for our study population as those trying to conceive or those who recently conceived may have a lower level of acceptable risk for heat exposure and alter their behavior more intentionally than the general population.

This study is notable for its restriction to individuals trying to conceive. In existing epidemiologic studies, heat exposure (primarily estimated using outdoor air temperature) has been consistently associated with a range of adverse pregnancy and birth outcomes.^31,32 A key consideration when measuring the extent to which outdoor air temperature is an accurate reflection of personal temperature is the relevant critical window of susceptibility for a given outcome. For example, long-term heat exposure (i.e., over the course of pregnancy) has been associated with fetal growth restriction.^30,31 On the other hand, even short-term exposures to high temperatures can adversely affect semen quality (hours) or miscarriage risk (days).^27,48,49 Personal heat monitoring might be particularly important if short-term exposures are of interest, as it would allow us to identify even short-term extreme heat exposures within a population of individuals who, on average, have low exposures.

Our study has several limitations. First, of the 238 monitors mailed to consenting participants, only 162 were mailed back to us, and only 152 provided usable data. Although this is a larger cohort than many personal monitoring studies, it is a small proportion and may not be representative of the overall PRESTO sample. However, the protocol was acceptable to participants, with over 80% reporting that they wore their monitors most of the time. Overall, this sample was a low-risk exposure population, which may limit our ability to generalize these results to the broader US population of pregnancy planners.

Additionally, we did not collect behavioral, time-activity, or GPS information to link to the personal temperature readings. Information on time-activity patterns during the monitoring period may have informed what outlier measures were true exposures (e.g., time spent over a stove or in a hot car) as opposed to inflated estimates (e.g., monitor in direct sunlight, which may affect the air temperature reading even though both monitoring devices and humans have higher surface temperatures in direct sunlight) or a result of non-compliance. Time-activity information might also be useful for developing models to predict personal heat exposure from outdoor temperatures given that the time-invariant factors we considered did not appreciably explain variation in the correlations between personal and outdoor temperatures. The use of relatively coarse weather data undoubtedly introduces some exposure measurement error into our analysis. However, we anticipate this error to be randomly distributed across the participants and hence our results are likely to be conservative. Moreover, the lack of real-time precise location data for the participants most likely was a greater source of error that would not be mitigated by a higher-resolution weather dataset. Future researchers with additional detailed location data might consider using an outdoor temperature data source with a finer spatial resolution to assess how correlations differ.

Lastly, even among pregnancy planners, we may not have measured heat exposure during critical windows that are relevant to specific reproductive and perinatal outcomes. Researchers planning future studies may benefit from prioritizing data collection during etiologically relevant risk periods for their specific outcome of interest.

In this US-based preconception cohort monitored during the warm season, on average, personal heat was not strongly correlated with outdoor temperatures, suggesting that participants were able to achieve adequate climate control much of the time. Interestingly, even in a population characterized by typically having the resources to mitigate heat exposure, some individuals still experienced short-duration exposures to extreme high temperatures. We saw modest differences in the average correlations between personal and outdoor temperatures by factors that could plausibly affect how heat is experienced, such as air conditioning and time spent outdoors.

This study within a preconception cohort adds evidence that outdoor air temperatures are not always well correlated with personal exposure, particularly in this cohort which may have higher incomes than the general public and who have a high access to AC. Nonetheless, for most applications, outdoor air temperature estimates will likely remain the best proxies for personal temperature until well-validated prediction models integrating personal exposures and other characteristics are developed. Outdoor temperatures represent a shared exposure metric across large populations. Individual misclassification may vary around this ambient estimate, essentially “averaging out,” leaving a robust signal in population-level studies – a phenomenon known as Berkson error.⁵¹ Additionally, even if personal temperature averages don’t align with ambient measures, brief periods of elevated heat (e.g., during outdoor activity or commuting) may be sufficient to trigger physiological stress responses. Further research is needed to better understand factors that might affect personal experiences of temperature. In the meantime, both outdoor temperature estimates as well as personal heat monitoring studies will remain important tools for assessing individual heat exposure, particularly for short duration or episodic exposures that may be relevant for reproductive and perinatal outcomes.

Impact Statement:

In this US-based preconception cohort, metrics of personal heat exposure were not highly correlated with heat exposure metrics derived from outdoor temperature at the residence. Although some participants experienced brief exposures to extreme heat, the results suggest that most participants achieved adequate climate control much of the time. Variation in the correlation between personal and outdoor temperatures was not appreciably explained by hypothesized factors. Even in a population with resources to mitigate heat exposure, some individuals still had short duration exposures to extreme high temperatures.

Data Availability Statement:

The data used in this analysis are not available for sharing due to concerns about participant privacy.