At each visit, participants met with study personnel in the FWAF field offices for clinical and non-clinical data collection; all data were saved in the REDCap platform²⁹. At the first visit, participants met on a non-workday to have their height, weight, and blood pressure recorded and a fingerstick to collect blood for analysis of hemoglobin A1C (A1CNOW^®, pts Diagnostics, Whitestown, IN); weight, A1C, and blood pressure were measured again at each visit. Study personnel orally queried participants in the participant’s primary language; information reported here included socio-demographics [age, sex, country of origin, marital status, and years of education], health history [frequency of non-steroidal anti-inflammatory drugs use (NSAID; sometimes/daily or rarely/never), history of gout or kidney stones, family history of gout or kidney stones, family history of diabetes, hypertension, or kidney disease], and general work characteristics [payment method (by-the-piece/hourly or salary/other), number of years in US agriculture, training in heat-related illness prevention, use of pesticides at work]. We also inquired if participants had relatives or household members that were also participating, using this information to adjust for possible clustering (non-independence) effects in statistical analyses.

Before and after their work shift on the observation day, blood and urine were collected. Blood was analyzed with the iSTAT Handheld Blood Analyzer Point of Care system with a Chem8+ cartridge (Abbott Laboratories, Abbott Park, IL), measuring sodium, potassium, chloride, BUN (blood urea nitrogen), and creatinine. Urine specific gravity was measured with an ATAGO meter (Atago PAL-10S digital refractometer, Bellevue, Washington). At visits 1 through 4, before work, study staff fitted participants with a monitor worn around the chest (Zephyr^™, Boulder, CO) that collected heart rate and physical activity measures every 30 seconds; the monitor was removed when participants returned to the study office after work. Average heart rate and physical activity were calculated for the period between each participant’s self-reported work start and stop times.

After work, a FWAF community health worker queried participants (one-to-one) by administering a structured survey that included questions on their work start and stop times, beverage consumption (type and quantities), and heat-related illness (HRI) symptoms (heavy sweating, headache, nausea, dizziness, confusion, muscle cramps, fainting, dysuria) experienced during that workday. The sum of sports drinks, energy drinks, juice, and soda were reported as “ounces of sugary beverages.”

The Florida Automated Weather Network (FAWN)¹¹ is managed by the University of Florida’s Institute of Food and Agricultural Sciences (UF/IFAS, which provides rural weather information to agricultural communities. FAWN has 35 monitoring stations which collect weather data every 15 minutes throughout Florida’s agricultural communities. FAWN data were obtained from the monitoring station located within each study area (Pierson and Apopka) in which the participants were working. Using the Lu and Romps extended heat index (HI) algorithm based on ambient temperature and relative humidity,³⁰ mean HI, max HI, and hours spent at HI >=100°F were calculated for each participant’s work hours.

The primary kidney function outcomes were pre- and post-work shift estimated glomerular filtration rate (eGFR) and acute kidney injury (AKI). We estimated the glomerular filtration rate using the 2021 CKD-EPI Creatinine Equation,³¹ as recommended for U.S. adults by the joint National Kidney Foundation/American Society of Nephrology Task Force³². Incident AKI was defined as an increase in creatinine from pre- to post-work shift of at least 50% or ≥0.3 mg/dL in accordance with KDIGO guidelines²⁵ and as recommended by the First International Research Workshop on Mesoamerican Nephropathy³³.

In addition to age, sex, work type, and time in study, risk factors examined for the primary outcomes of low eGFR and AKI incidence fall into three a priori categories: health factors, work conditions, and hydration practices and indicators. Health risk factors included body mass index (BMI), blood pressure (normal [<120 Systolic and <80 Diastolic], elevated 120–139 [Systolic or 80–89 Diastolic], high [≥140 Systolic or ≥90 Diastolic]), A1C ≥5.7 (i.e., prediabetes or diabetes), resting heart rate, sometimes/daily use of NSAIDs, and family history of kidney risk factors (any diabetes /hypertension /kidney disease). Work risk factors included training in HRI prevention (yes/no), occupation as a pesticide applicator (years), workday duration, shade breaks taken, heat exposure (mean HI, max HI, hours at HI ≥100°F), and average heart rate and physical activity (VMU units) during workday. Hydration practices and indicators included total self-reported beverage consumption and average beverage consumption per work hour, sugary beverage consumption, dehydration (urine specific gravity≥1.020 both before and after work), and serum osmolality (mOsm/kg).

Additionally, the effects of primary work type were examined. The two locations were selected because the nature of agricultural work in these two Florida locations is quite distinct. In Apopka, 95% of the recruited participants worked at a nursery while in Pierson, 90% worked in a fernery. Common tasks that nursery workers performed were planting; cutting; trimming; packing; loading and unloading; watering and pruning trees, shrubs, and plants; and moving potted plants and shrubs. Fernery workers mainly cut fern and bundled it into bunches, which they loaded onto a truck. While correspondence between location and primary work type were not absolute, for simplicity, workers from the Apopka cohort will henceforth be referred to as “nursery workers” and the Pierson cohort as “fernery workers.” Payment method was not examined as a separate risk factor as it was collinear with primary work type with fernery workers likely to be paid by piece work. This practice is not observed in nursery work.

Descriptive statistics were presented as % (n) or median (1^st quartile, 3^rd quartile). Simple differences in characteristics between primary work types at baseline were examined with methods that adjusted for the random effect of household cluster: the R package “lqmm” [linear quantile mixed models]^34,35 for median values of continuous variables, generalized linear mixed models for categorical variables.

Before attempting multivariate modelling of the longitudinal data, we used multiple imputation for risk factors having more than 10% missing values. Missing values were assumed missing at random as they were mainly due to machine malfunction. The imputation model included all risk factors, outcomes, and intended interaction terms that would be included in analysis models, plus demographic and clinical auxiliary variables. We conducted the imputation using the R package “mice”,³⁶ which uses fully conditional specification, and employed predictive mean matching (numeric data), logistic regression imputation (binary data, factor with 2 levels), or proportional odds model for (ordered, > 2 levels), as appropriate; 35 imputations and 15 iterations were specified. After examining convergence plots and imputed values, the originally missing outcome values were reset to missing.

Because workday heart rate and physical activity were not measured at the final visit, we examined how variable a participant’s measurements were over time by calculating the intraclass correlation coefficients (ICC). ICC for mean heart rate was 0.60 for winter and 0.62 for summer (0.55 overall); for mean physical activity it was 0.81 for winter and 0.86 for summer (0.75 overall), indicating low variability. We thus chose to use visit 1 values as the predictor rather than employing a less than optimal imputation of all final visit values. To account for the nested data structure, linear mixed modelling and generalized linear mixed modeling (GLMM) with a random intercept for ID were used for simple and multivariate inference. For predicting pre-work and post-work eGFR, a base model included time, primary work, and their interaction plus covariates age and sex. This simple model was run with all five visits, and again for summer and winter visits separately, dropping the interaction term if it was not significant in the base model. To the summer and winter models, risk factors were added one by one. Risk factors measured at visit 1 were also examined for interaction with time to examine whether they influenced the course of eGFR over time. Models that included imputed risk factors used SAS MIANALYZE³⁷ to calculate estimates and statistical significance. A similar analytical plan was executed for predicting AKI incidence, except that no winter model was run due to the limited number of cases of AKI. Additionally, the base model included only time, primary work, age, and sex as the interaction term of time by primary work was not significant. The assumptions of logit linearity for continuous predictors and absence of multicollinearity were confirmed.

Analyses were performed with SAS version 9.4 software (Cary, NC) and R 4.3.0. Statistical significance was evaluated using alpha = 0.05.

This study was conducted over 32 months from January 2020 to August 2022, with study visits and workday monitoring conducted in two winters and in three summers during that time. Sixty-seven participants in Apopka, FL, primarily working in nurseries, and forty-eight participants in Pierson, FL, primarily working in ferneries, were initially recruited. As previously reported,³⁸ more than three quarters of participants were retained over the first two years of the study, with retention decreasing for the final summer visit which occurred more than a year after the previous study visit. Retention at the last visit was higher among Pierson fernery workers (79%) compared with Apopka nursery workers (52%).

Several differences were identified in the nursery and fernery worker groups (Table 1). While more than two-thirds of the recruited participants working in either setting identified as female, there was a higher percentage (33%) of males among the fernery workers compared to the nursery workers (16%, p=0.03). There were no significant differences in age or in the age at which participants started working in agriculture, but there was a significant difference (p=0.003) in the number of years participants had been working in U.S. agriculture (median 6 in nurseries vs. median 16 in ferneries). Additionally, nearly all (94%) fernery workers claimed Mexican nationality with the remainder being from the U.S., while nursery workers were a less homogenous group (p<0.001) with roots in Mexico (57%), Guatemala (24%), the U.S. (3%), and other countries (16%). A lower percentage (21%) of fernery workers were also single compared to nursery workers (40%, p=0.04). Educational level was similarly low in the two cohorts (median of 8 years, p=0.8).

At the initial study visit (January 2020), the median BMI among fernery workers (30.7) was higher than that of nursery workers (28.5, p=0.04), and this pattern remained consistent across study visits (Table S1). Fernery workers also had higher blood pressure (BP; p=0.049); at the initial study visit, 41% of nursery workers had BP readings classified as elevated or hypertensive (>120/80) compared to 59% of fernery workers (Table 1). By the final study visit, the percentage of participants with elevated or hypertensive BP readings had increased in both groups (48% nursery workers, 66% fernery workers), but the distinction between the groups remained (Table S1). No differences were found between the worker groups in diabetes status (based on A1C percent), family or personal history of kidney disease-related medical issues (gout, kidney stones, hypertension, or diabetes), NSAID use, or resting heart rate (Table 1). A seasonal pattern was observed for A1C percent, with slightly higher values in the summers in both worker cohorts (Table S1).

Nearly all (96%) nursery workers were paid by the hour and none by the piece, whereas most fernery workers (73%) were compensated per piece or by a combination of piece and hours (6%) (Table 1). This difference in compensation structure likely contributed to the differences in participants’ reports of their workday break practices (p=0.03); among fernery workers, 80% reported taking no breaks or just 1 break during workdays; 46% of nursery workers reported taking 2 or more breaks. Differences in work hours may also have contributed to these differences in breaks. Participants affiliated with nurseries worked longer hours (median 9.0) than did participants affiliated with ferneries (median 7.2) (p<0.001). There were no significant differences between the agricultural settings in participants’ reports of working with pesticides or recent training to prevent heat-related illness.

Participants’ physical activity also differed between the agricultural settings (Table 1). Fernery workers had a higher average heart rate (median 96.3 bpm) and activity level (median 0.20 VMU) compared with nursery workers (median 91.5 bpm, p=0.04; median 0.11 VMU, p<0.001). A median of .20 VMU is the activity equivalent of someone continuously walking during their whole work shift.³⁹ Fernery workers were also bending more during their work shifts (p<0.001). These activity differences remained consistent at all timepoints measured (Table 2).

Heat indices were calculated in each location for each study visit day (Figure 1). While there was greater variation in heat indices across study visit days for fernery workers, the mean heat index as well as the number of hours in which the heat index was over 100°F were comparable for both groups. The max heat index did not differ for the first three study visits but was slightly higher for fernery workers at the last two summer study visits.

While overall self-reported workday beverage consumption did not differ between fernery and nursery workers, choices of beverages did, with more fernery workers drinking sugary beverages than nursery workers (p=0.002) (Table 1). Sugary beverage intake during work was more common in both groups in the summers than in the winters, but the greater prevalence and higher intake among fernery workers persisted at all study visits (Table S2). Coffee consumption was higher among nursery than fernery workers, but overall, the numbers of participants reporting drinking coffee during work were lower than those reporting drinking sugary beverages or water (Table S2).

Over the course of the study, there was greater evidence of dehydration among fernery workers compared to nursery workers (Table 3). At all but one study timepoint (August 2020 pre-workday), a higher percentage of fernery workers than nursery workers had USG ≥ 1.020, and at all timepoints, more fernery than nursery workers were classified as dehydrated by USG ≥ 1.020 both before and after work shifts (OR 2.1, 95% CI 1.1, 3.8). In the summers, the percentage of workers with USG ≥ 1.020 was higher post-workday than it was pre-workday in both agricultural settings, but the increases in the percentage of dehydrated workers pre- to post-workday were higher among fernery workers than nursery workers. Furthermore, in the summers, markedly higher percentages of fernery workers compared to nursery workers were classified as severely dehydrated (USG ≥ 1.030) post-workday (25%, 11%, and 18% vs 2%, 4%, and 9%, respectively (OR 4.9, 95% CI 1.6, 14.8). At all but one timepoint, higher percentages of fernery workers compared to nursery workers also met the criteria for hyperosmolality post-workday.

Pre-workday eGFR values adjusted for sex and age of participants remained more consistent across the study than did post-workday eGFR values (Figure 3). There was no significant interaction between work season and primary work type, as there was with post-workday eGFR, but there were significant main effects, with pre-workday eGFR values slightly lower in the summers (beta −1.5, 95% CI −2.7, −0.3) and slightly higher in fernery compared with nursery workers (beta 4.1, 95% CI 1.4, 6.9). Pre-work eGFR < 90 was only seen in a single participant during the study (Table 3).

Patterns of AKI across the study differed between nursery and fernery workers (Figure 2, Table 3). At all but one timepoint (July 2021), about 10% of nursery workers met the criteria for AKI post-work shift. At all but one timepoint (Aug 2020), more than 20% of fernery workers met the criteria for AKI post-work shift, and this percentage increased progressively over the last three study timepoints from 21% in January 2021 to 43% in August 2022 (OR 8.1 for fernery workers in summer 2021 and 12.6 in summer 2022 relative to summer 2020 compared with 2.4 and 1.1 for nursery workers, respectively, Table S3). Besides agricultural setting, none of the demographic factors, health factors, work conditions, or hydration measures collected in this study were significantly associated with summer AKI(Table S3).

Estimated GFR levels calculated pre- and post-work shift also differed across the study for workers in the two agricultural settings (Table 3, Figure 3). After adjustment for sex and age of the participants, post-workday eGFR values for fernery workers were higher than for nursery workers at the first three study timepoints (winter 2020, summer 2020, winter 2021), but, in alignment with AKI results, post-workday eGFR values for fernery workers dropped below nursery workers at the last two summer timepoints (Figure 3, Table 4). For fernery workers, comparing base summer 2020 to the two following summers showed substantial decline in eGFR (2021 beta −9.3, 95% CI −16.9, −1.7; 2022 beta −10.5, 95% CI −18.1, −2.9) that was not observed for nursery workers (2021 beta −2.6, 95% CI −7.5, 2.2; 2022 beta 2.8, 95% CI −2.3, 7.9) (Table 4). At these timepoints, we also observed higher percentages of fernery workers with post-workday eGFR < 90 than had been seen at previous timepoints or were ever observed among nursery workers in this study (Table 3).

Our primary predictor, heat exposure, significantly impacted summer post-workday eGFR; for every degree increase in average heat index, post-workday eGFR declined by 0.5 units (p=0.036), and controlling for it in the model revealed significant declines in eGFR from summer 2020 to summer 2021 in nursery workers (beta −6.0, 95% CI −11.7, −0.3) while deattenuating the effect in fernery workers for the same time period (beta −5.8, 95% CI −14, 2.3) (Table 4). Similar effects were observed when accounting for the number of work hours in which the heat index was above 100°F. A significant negative association (p=0.024) was identified between summer post-workday eGFR and the volume of beverage consumption during the workday independent of agricultural setting. Besides age, no other risk factors were significantly associated with this outcome, nor were they confounders of the relationship between work type and time.

Post-workday eGFR in the winters appeared to have a different relationship with predictors evaluated in this study (Table 5). There was not a clear effect of agricultural setting nor evidence of decline between winter 2020 and winter 2021. Other than age, the only predictor that was significantly associated with winter post-workday eGFR was serum osmolality, with eGFR declining 1 unit for every unit increase in serum osmolality (95% CI −1.5, −0.5).

There were no appreciable differences longitudinally across the summers or the winters (Tables 6 and 7, respectively). Resting heart rate, NSAID use, and work applying pesticides were significant predictors of summertime pre-work eGFR. Every beat per minute increase in resting heart rate was associated with an increase of 0.1 units in pre-workday eGFR (95% CI 0.1, 0.2, p=0.001). Participants who reported using NSAIDs more frequently had higher pre-workday eGFR (beta 2.2, 95% CI 0, 4.3, p=0.049). Each year of occupational pesticide use was associated with a decline of 0.4 units in pre-workday eGFR (95% CI −0.9, 0, p=0.042). Beyond age and sex, no other risk factors assessed were significantly associated with summertime eGFR, nor were they confounders of the relationship between pre-workday eGFR and work type (Table 6). Additionally, other than age, none of the risk factors assessed was significantly associated with wintertime pre-workday eGFR (Table 7).

A few workers from each location reported working in an agricultural environment other than a nursery or fernery. Furthermore, it is not uncommon for agricultural workers to assume different responsibilities and work in different environments over time. This is also a reflection of the different economic strategies that agricultural workers rely on over time to secure a livelihood under precarious job security situations. Understanding the dynamic nature of this occupation, we decided not to restrict this study to participants only working at a fernery or nursery, even though this necessitated an imperfect comparison of these agricultural settings.

Strengths of this study are that we recorded beverage consumption with a post-workday structured interview using props to demonstrate sizes; we followed workers for 32 months and had good retention both in summer and winter visits; and had a good range of heat indices that strengthened power to detect effect. Although we were able to examine many of the risk factors proposed in the literature, we do not know the types of pesticides or the level of exposure. There also was not much power to detect effects for AKI because of the relatively small number of events. It should also be noted that the classification of AKI was based on changes in serum creatinine, consistent with standard clinical practice, and not confirmed by additional clinical testing such as imaging or biopsies. Temperature and humidity are not from the work sites themselves and may not account for the micro-environment of the worker (inside greenhouse, under shade cloth, etc.), nor was core body temperature assessed in this study. This is a limitation since it is a combination of environmental heat exposure as well as physical exertion which results in hyperthermia and the physiological responses that increase the risk of AKI.