Study population

The study design and MANOS cohort have been previously described [33]. In brief, MANOS participants (n = 569 males) underwent extensive workplace exposure monitoring, including continuous Tc, HR, physical activity, and WBGT, over three (usually consecutive) days in January 2018–May 2018. Participants represented five industries: corn, plantain, brickmaking, construction, and sugar. We recruited workers from two sugar companies in El Salvador and three in Nicaragua, so the following codes are used for each: SUGAR-E1, SUGAR-E2, SUGAR-N1, SUGAR-N2, and SUGAR-N3.

Environmental monitoring

WBGT was measured every minute during work shifts using TSI (formerly 3 M) QUESTemp 46 Waterless Wet Bulb Globe Thermometers (TSI Incorporated, Shoreview, MN). Wind speed was also measured, using a TSI air velocity sensor attachment (TSI, Shoreview, MN). Thermometers were mounted on tripods one meter above the ground as close as possible to the participants. If participants moved locations during the work shift, the thermometers were moved to maintain proximity.

Personal monitoring

Tc during the work shift was assessed using wireless ingestible CorTemp® Disposable Temperature Sensors (HQ Inc., Palmetto, FL). Participants were randomly assigned to be monitored during work shifts on Days 1 and 3 or only on Day 2. The CorTemp Data Recorder was worn in a pouch strapped to the small of participants’ backs and recorded Tc readings every 10 s.

Physical activity was characterized using an ActiGraph wGT3X BT (ActiGraph, LLC, Pensacola, FL) accelerometer, which captures measured movement at 30 Hz or higher, worn on a belt around the participants’ hips during the work shift on all three days. Polar H7 heart rate monitors (Polar Electro Oy, Kempele, Finland), attached to a strap around the chest below the pectoral muscle, were worn during the work shift on all three days at baseline. Data were collected at a beat-to-beat resolution and transmitted via Bluetooth to the ActiGraph wGT3X BT devices.

Height and weight were measured with a Seca 769 column scale (Seca GmbH, Hamburg, Germany)—before and after each shift for weight, while only once for height. Weight was averaged across all six measurements to determine the participant’s average weight at baseline for Recommended Exposure Limit (REL) calculations and estimated energy expenditure calculations. Differences between pre- and post-shift weight measurements were not used to assess water loss via sweating as the protocol used at each measurement (e.g., clothing and equipment worn) varied considerably.

Biological samples

Blood samples were collected before and after the shift on the third day only, except for several brick workers (n = 29) for whom blood was collected on the first or second day due to unpredictable work schedules. Serum samples from Nicaragua were analyzed at the Ministry of Health’s National Laboratory in Nicaragua and samples from El Salvador were analyzed at Quest Diagnostics in Massachusetts, USA. All samples were analyzed for serum creatinine (IDMS-traceable) to estimate glomerular filtration rate (eGFR) using the CKD-EPI equation [34]. Subsequent serum testing of a random subset of baseline samples (n = 50 for each country), conducted at Quest Diagnostics in 2021, confirmed the minimal-to-no difference between laboratories.

Questionnaires

Questionnaires were administered to participants by trained field team members upon enrollment and at the end of the work shift on each day to capture characteristics of the workday (start and stop time, breaks, hydration practices, medications taken, personal protective equipment worn, and symptoms experienced). Workers reported the job tasks they performed each day, and the study team summarized these tasks into categories (e.g., sugarcane cutting was summarized under “harvesting”). It should be noted that some workers were assigned various tasks throughout the work shift and were summarized into more general job categories.

Statistical analyses

WBGT, Tc, HR, and physical activity data for each participant were cropped based on the start and stop times of their work shift. Implausible values for each device (e.g., <30 for HR, < 32 °C for Tc) were marked as missing and were not included in calculations. WBGT, Tc, HR, or physical activity data with more than 50% of any given work shift missing were excluded from relevant analyses (n = 123 person-days for HR; 40 for physical activity; 141 for WBGT; 60 for Tc).

All MANOS participants were outdoor workers, so the WBGT formula for outdoor settings was used. The first 10 min of each WBGT dataset was removed, prior to cropping at the work shift, to account for the stabilization period defined by the manufacturer [35]. When available, two thermometers were used simultaneously and the values from each device were averaged at each time point. Effective WBGT (WBGTeff) was calculated by adding a clothing adjustment factor of 0.5 °C for agrichemical applicators in Nicaragua, who wore polypropylene and plastic coveralls [9, 36]. Data were smoothed using a 20-minute rolling average. The REL—the recommended heat stress exposure threshold defined by the National Institute of Occupational Safety and Health (NIOSH)—was calculated for each participant using their average body weight and estimated average metabolic rate (kcalories/hour) derived from physical activity data (described below). Minutes above REL and percent of the work shift above REL were calculated for each participant on each shift. The heat index was derived from dry temperature and humidity using the National Weather Service formula through the weathermetrics package in R [37, 38].

For accurate Tc readings, CorTemp® sensors need to pass through the digestive system to the small intestine, which requires swallowing the sensor several hours before monitoring and ideally eating/drinking something with the sensor. If the sensor is too high in the digestive tract, the Tc data can be influenced by the consumption of liquids and foods resulting in data reflecting a “bouncing ball” effect [39]. Despite study protocol stating that sensors should be swallowed the night before the monitoring workday, it was often difficult to put this into practice. Workers and investigators had concerns about bowel movements prior to the work shift and ingestion protocol compliance, therefore it was not uncommon for the sensors to be swallowed the morning of monitoring. The number of hours before the work shift that the sensor was swallowed varied widely, with distinct patterns by country, industry, and work site due to logistics. For this reason, the number of hours before the shift the sensor was swallowed was estimated to examine the effect of this variation. In addition to the sensor being higher in the digestive tract than desired, other issues can cause nonsensical Tc data. For instance, workers standing close to one another may cause interference in the transmission of Tc data to the correct data recorder and the presence of two sensors in the body (e.g., worker thinking they had excreted the first sensor) may produce unusable data. Tc data for each participant were carefully examined to identify such files. A script in R flagged Tc data afflicted by the “bouncing ball” effect—but otherwise deemed usable—and removed portions of the Tc data that were unrealistic based on the magnitude of the slope between neighboring points 1, 2, and 3 points away. Criteria for removal were as follows:

1.

The average slope between a given point and its neighboring points on either side was > 2*SD away from the mean of all slopes of that window size (i.e., 1 point away, 2 points away, etc.) and the value of the temperature at that point was > 2*SD away from the mean of all temperature values for that individual, or

2.

The absolute slope between a given point and its neighboring points was greater than the equivalent of a 2 °C change over 15 min.

All Tc data were then smoothed using local regression (LOESS) using a 25% smoothing span.

Vector magnitude (VM)—defined as the square root of the sum of the squares of the counts for each of the three axes measured by the accelerometers—was used to estimate energy expenditure in kilocalories at each minute interval using the 2011 Freedson VM3 equation [40] combined with the 1998 Williams Work-Energy Equation [41]:

$$ }}}}}}}\;}}}}}}}\, > \,}}}}}}}\;}}}}}}}\\ }}}}}}}/}}}}}}} = }}}}}}}.}}}}}}}\, \times }}}}}}} + }}}}}}}.}}}}}}}\, \times }}}}}}} - }}}}}}}.}}}}}}}} )\\ }}}}}}}\;}}}}}}}/}}}}}}} = }}}}}}}\, \times }}}}}}}.}}}}}}} \times }}}}}}}$$

where BM is body mass in kilograms and CPM is the counts per minute (i.e., vector magnitude at a minute interval). Vector magnitude was also used to determine when participants were on break or otherwise performing limited physical activity, using the threshold of VM < 150 CPM [42].

LOESS regression with a 10% smoothing span was used to smooth HR data. Maximum HR (HRmax) was calculated using the formula 220-age and percent of HRmax at each minute interval was calculated using the smoothed HR at that interval.

Multivariable linear regression models were used to examine the associations between job task, hydration practices, break duration, and baseline kidney function and maximum Tc experienced during the work shift, controlling for confounders which were selected using a literature review of relevant research and a directed acyclic graph. Mixed effects models with a random intercept and random slope for day were used for modeling the median percent of HRmax experienced during each shift. Pre-shift eGFR was used to assess kidney function at baseline using the following categories: < 60, 60–90, and >90 mL/min/1.73 m2. Participants with pre-shift eGFR <60 mL/min/1.73 m2 (n = 53) were considered to have impaired kidney function and were removed from all models, except for those examining the effects of kidney function on measures of heat strain. Data for overnight shifts (n = 48 person-days; 2.8% of all person-days) were removed from all models, as were any person-days for which the monitoring data captured < 50% of shift. Models examining electrolyte solution were restricted to Nicaraguan sugar workers, as they were the only workers reporting consumption of this beverage.

Analyses were performed using SAS Version 9.4 and R Version 3.6.1 (The R Foundation for Statistical Computing, www.r-project.org) [43].