Study design and setting
This study, the HEAT intervention study, is a parallel, comparison, group randomized intervention study to evaluate the effectiveness of a multi-level HEAT intervention approach for agricultural workers and supervisors that includes: 1) worker education; and 2) a heat awareness mobile application (HEAT App) that informs supervisors of hot conditions during the coming week and provides recommendations to keep workers safe . The study took place in 2019 in agriculturally intensive areas of Central/Eastern WA, where tree fruit, cherries, and other crops such as grapes and hops are predominant . Eastern WA is characterized by warmer and drier summers than Western WA, with average summer high temperatures in the upper 80s to mid-90s°F (27–34 °C) . The study took place from May–September, as the majority of hot days in WA occur between May and September. Baseline survey data and initial rounds of weekly symptoms data collection began in May. Field data collection occurred from June to August, and the final round of weekly symptom data was collected in September. Agricultural workers in Central/Eastern WA are largely Latinx/e and include seasonal workers and US H-2A guest workers. Latinx or Latine are non-binary and neutral forms of Latinos, and they are used to acknowledge marginalized and excluded members of the diverse Latinx/e community [31,32,33]. The US H-2A program is a federal program that allows employers to hire workers on temporary work permits from other countries for agricultural jobs . The University of Washington Human Subjects Division (HSD) approved all study procedures, and participants provided written informed consent prior to study participation.
Study details and information about HEAT intervention development have been previously reported . In brief, the HEAT intervention was developed in collaboration with regional agricultural stakeholders and communities through long-standing partnerships with Pacific Northwest Agricultural Safety and Health (PNASH) Center researchers. Intervention development was grounded in the social-ecological model of prevention [28, 35, 36] and guided by two advisory groups: 1) a technical advisory group, which included agricultural industry, government, and community representatives; and 2) an expert working group, which included farmworkers and managers . Research staff included individuals who live and work in agricultural communities in WA. The HEAT intervention was designed to cover factors that affect HRI risk at multiple levels, including the individual, workplace, and community levels .
The first intervention component, HEAT education, was developed to be culturally and linguistically appropriate and tailored to agriculture and uses a relational and engaged approach in the language of preference of the target audience (Spanish or English) . HEAT education includes a Spanish/English train-the-trainer facilitator’s guide, uses poster visual displays, and covers: 1) types of HRI and treatments; 2) risk factors for HRI; 3) staying hydrated at work; 4) clothing for work in hot weather; 5) personal protective equipment and heat; and 6) keeping cool in the home and community . HEAT education was designed to comply with WA’s Outdoor Heat Rule for Agriculture worker training requirements . Feedback from advisory groups, results from focus groups and beta testing with promotores (community health workers) and agricultural workers, which involved providing early versions of the HEAT education and making adjustments based on feedback, and guidance from the University of Washington Center for Teaching and Learning were used to refine the HEAT education materials . The entire training guide takes approximately 60–90 minutes to complete but can also be broken down into 15-minute toolbox trainings for use in the field. Our prior study of HEAT education among WA farmworkers found greater improvement in worker heat knowledge scores across a summer season in the HEAT intervention group, compared to a comparison group that was offered non-HRI alternative training (p = 0.04) .
The second intervention component, the HEAT App, was developed in partnership with Washington State University’s AgWeatherNet (AWN) Program. AWN maintains a network of over 200 professional weather stations located mostly in agriculturally productive regions of Central/Eastern WA and is a trusted source of weather information for crop decision support in the WA agricultural community . The HEAT App links current and forecasted weather information with health and safety messages. HEAT App development was grounded in elements of the Technology Acceptance Model [28, 39], and the HEAT App was designed to notify agricultural supervisors about hot weather conditions and send messages through push notifications. Messages contain information about workers’ risk for adverse health effects from heat and strategies for prevention that are tailored to the agricultural industry (Fig. S1). As previously described , messages are sent one and 6 days before a forecasted Heat Index of 91 °F (33 °C) or higher at nearby weather stations selected by the user. Suggested actions for heat prevention are available for conditions between a Heat Index of 80–90 °F (27–32 °C), but push notifications are not sent out below 91 °F (99 °C) to avoid information overload.
Recruitment & eligibility
We used convenience sampling to recruit participants from agricultural companies from Central/Eastern WA in the late Spring 2019, as previously described . There were a total of four tree fruit and vineyard companies that agreed to participate. The research team provided information sessions about the study and recruited participants from participating employers’ crews. There were approximately two to six crews per participating company from which crews were recruited. Crews were already formed by the workplace, and researchers did not have the ability to assemble crews. As described in the Intervention allocation section below, crews within large and small companies were allocated to intervention and comparison groups separately, as large and small companies differ in their capacity for dedicated health and safety personnel and programs. Two of the four companies, hereafter referred to as ‘Large-1′ and ‘Large-2,’ were considered large companies, with more than 50 full-time employees during the growing season and dedicated health and safety personnel. We enrolled two crews from each large company for a total of four crews (Fig. S2). The other two companies had less than 50 full-time employees and were considered small companies. Since the two small companies were owned by brothers and had similar safety and health practices, the two small companies were considered one company, hereafter referred to as ‘Small,’ for the purposes of the analysis. We enrolled two crews from ‘Small’ for a total of two crews (Fig. S2). This recruitment strategy yielded ‘Large-1′, ‘Large-2′, and ‘Small’ enrolled companies and six enrolled crews (two per company) (Fig. S2), with eight to 17 participants per crew. Eligible participants included seasonal workers and US H-2A guest workers, workers aged 18 years or older, workers who planned to work in agriculture during the summer season, and workers who understood Spanish and/or English.
Research staff were trained to use simple randomization (coin flip) to randomly allocate crews of participating workers within each company to intervention and comparison groups. Workers and supervisors were not provided with information about which group they were allocated to, but researchers were aware of group allocation. One crew from each company was assigned to the intervention group (three crews total) and the other crew from each company to the comparison group (three crews total) (Fig. S2). Due to logistical constraints related to the timing of agricultural work, crews from ‘Small’ were not randomized; the first to enroll received the intervention, and the second was assigned to the comparison group. All participants were offered the intervention after data collection was complete.
Study procedures & flow
After obtaining informed consent, workers were asked to complete a baseline survey in Spanish or English. Work characteristics, including company, crew, and H-2A status, were noted by field staff on field observation sheets. Workers in the HEAT intervention group then received HEAT education from the same research staff member. Workers in the comparison group were offered education on another topic of interest to them (e.g., sexual harassment, pesticides). The HEAT App was provided in Spanish or English to intervention group supervisors who directly supervised each crew over the course of the season. Research staff assisted intervention group supervisors in downloading the application to their mobile device, selecting weather stations closest to their worksites, and viewing current heat indices and maximum daily heat indices forecasted over the following week. Approximately monthly, research staff conducted field monitoring, including field observations, surveys, and physiological monitoring at the farm (see Data collection and processing below). Participants were also asked to complete a weekly symptoms survey via a mobile phone application or phone call.
Details of the study flow are shown in Fig. 1. Overall, 87 participants were evaluated for eligibility. One participant was excluded because they were ineligible (age less than 18 years), and therefore 86 participants from six crews were enrolled. Three participants allocated to the intervention group did not receive the intervention and were excluded. Three and five participants did not have more than one field monitoring day or at least 2 hours of physiological heat strain data in the intervention and comparison groups, respectively, and were excluded from the primary analysis of the relationship between the HEAT intervention and heat strain. A total of 75 participants were available for the primary analysis of heat strain. Five participants did not have available weekly symptoms survey data and were additionally excluded from secondary analyses of the relationship between heat strain and symptoms and from descriptive analyses of factors associated with HRI symptoms reporting.
Data collection & processing
Participants completed the baseline survey on paper or a computer tablet in Spanish or English, depending on the participant’s preference (Fig. S3). Spanish/English bicultural/bilingual study staff members were available to read the questions and response choices to the participants, as needed. The baseline survey consisted of 42 questions covering years of experience working in agriculture, distance to toilet at work, previous HRI training, medical conditions, cooling practices, and demographic information (e.g., age, sex, country of origin, years in the US). The baseline survey and the weekly symptoms survey, discussed in the next section, were based on our previous survey, which has been evaluated for validity and reliability in a similar population, as previously described .
Weekly symptoms survey
A weekly Spanish/English check-in survey was administered to participants at the end of every week, on Thursday-Sunday, excluding holidays, throughout the study period (Fig. S4). The survey asked about the previous 7 days of work. Participants had the option to complete the survey using a smartphone application (LifeData, LLC; Marion, IN) that sent a notification to complete the survey on Thursday afternoon with subsequent reminders on Friday. Participants who did not complete the survey using the phone application, as well as those that did not feel comfortable filling out the survey using the application, were called every week on Friday by a bilingual/bicultural research team member and asked the survey questions. Participants who did not answer or did not have time to complete the survey by Friday were called on Saturday or Sunday. The weekly check-in survey was designed to take approximately 5 minutes and included questions about HRI symptoms, including: 1) skin rash or skin bumps, 2) painful muscle cramps or spasms, 3) dizziness or light-headedness, 4) fainting, 5) headache, 6) nausea or vomiting, 7) heavy sweating, 8) extreme weakness and fatigue, and 9) confusion.
Physiological strain index (PSI)
Our primary outcome was physiological heat strain (PSI). We measured tympanic temperatures using tympanic thermometers (Braun; Kronberg, Germany) at the beginning of the work-shift on field monitoring days. Baseline core temperature (T0) was estimated by adding 0.27 °C to the tympanic temperature to account for differences between tympanic temperature and core body temperature . Research staff assessed baseline heart rates (HR0) by asking participants to rest for approximately 10 minutes and taking participants’ radial pulses for 15 seconds, then multiplying by four, in the morning before work shifts. Workers’ heart rates were logged every 20 seconds throughout the work-shift using Polar® chest band monitors (Polar, Inc.; Lake Success, NY). Heart rate measurements below 40 beats per minute were removed, as these values were considered outside of the physiologically expected range. Only one participant had 39 minutes of nonzero heart rate measurements below 40 beats per minute on 1 day, and these values were excluded. No participants had heart rates above 200 beats per minute. One-minute average heart rates (HRx) were then computed. We employed a US Army Research Institute of Environmental Medicine (USARIEM) method , which uses an extended Kalman filter algorithm, to produce estimates of core body temperature every minute (Tx) from one-minute heart rate measurements (HRx) and baseline core body temperature (T0). This algorithm has been validated in military settings and evaluated among WA agricultural workers . We calculated PSI using the equation PSI = 5*[(Tx -T0)/(39.5-T0)] + 5*[(HRx-HR0)/(180-HR0)] . A higher PSI indicates higher heat strain.
Body mass index
Participant height and weight were measured on field observation days. Due to work demands, participants did not always have time to take off their work boots prior to measurements. If this was the case, shoes were accounted for by subtracting five pounds from the weight and one inch from the height measurements. Height and weight measurements were used to calculate body mass index (BMI) [kg/m2] . BMI was included in analyses because it may be associated with HRI risk .
For the primary heat strain analysis, research staff recorded work start and end times on field observation days. We obtained data on air temperature and relative humidity during the work shift from nearby AWN stations, which log data in 15-minute intervals . We selected the two closest weather stations on observation days from each known work area, resulting in the inclusion of stations within 8000 m of each known work area. We used Rothfusz’s modification of Steadman’s work to calculate the Heat Index from temperature and relative humidity [47, 48]. Values from included weather stations for each crew on each observation day were averaged. For each participant, we trimmed data to work start and end times. Data were then summarized per participant to generate maximum daily Heat Index (HImax) values on observation days.
Field research staff recorded participant task and crop observations on field data sheets. Based on field observations and review of crop and task combinations by study team members with training in occupational safety and health, we used the main observed task to generate the following effort categories: high = tree fruit harvest (there was no grape harvest during field observation days); medium-high = digging holes, fixing posts, installing wire (tree fruit), tying branches (tree fruit), uncovering trees, tree fruit pruning, tree fruit thinning; medium-low = weeding, grape thinning, irrigation, tying branches (grapes), installing wire (grapes); low = using tractor, driving car, welding. If more than one task was recorded as the main task, the task with the maximum effort level was used to determine the effort category. For the analysis, low and medium-low categories were combined together (low/medium-low).
We used descriptive univariate and bivariate statistics, box plots, and scatter plots to characterize participant baseline characteristics and time-varying characteristics of effort level, HImax, and PSI.
Association of HEAT intervention with PSI
The repeated or longitudinal assessments of participants requires an analysis method that accounts for correlation among these repeated measurements. We therefore assessed the association of maximum work shift PSI (PSImax) with group status (intervention versus comparison, with group assigned using intention-to-treat) using linear mixed effects models with random effects for workers. Although our power analysis  did not take into account covariates, as prior information on the effects of all covariates was not available, we report two models to demonstrate how the apparent intervention effect on PSImax is modified by two factors described extensively in the literature to be associated with heat strain (effort level and Heat Index) [49, 50], and then how all these effects are modified by adjustment for demographic factors. We present Model 1, which accounts for HImax centered around the mean (degrees Fahrenheit), effort level (low/medium-low [reference category], medium-high, and high), and the interactions of effort level with HImax and group. We hypothesized that the effect of the intervention may be greater among those with higher compared to lower effort levels. We also present Model 2, which accounts for the following potential confounders: 1) individual: age (years), sex (female [reference category], male), and BMI (kg/m2); 2) work: effort level, HImax, and company (small [reference category], large-1, and large-2); and 3) terms for the interaction of effort level with HImax and group. We do not report an interaction of group status with HImax as the modest sample size does not support meaningful (significant) estimation of possible variation of an intervention effect with heat exposure in addition to its variation with effort level. The nominal significance (p-values) for the 2-degrees of freedom terms involving the 3-level coding of effort were computed using the lmerTest package in R .
Relationship of PSI with HRI symptoms reported & factors associated with HRI symptoms reporting
We coded the symptoms variable as an ordinal variable: no symptoms reported (0), one symptom reported (1), and two or more symptoms reported (2+). We used box plots to visualize the relationship between PSImax and HRI symptoms reported. To describe the relationship of factors other than PSImax associated with HRI symptoms reporting (ordinal), we used bivariate descriptive statistics and mixed models with random effects for workers using the clmm2 function in the ordinal package in R.
All analyses were conducted using RStudio Server Version 1.4.1717 .