Associations between temperature–humidity index and rumination time in lactating and periparturient Holstein cows in a New England conventional dairy herd

1 Introduction

Global temperatures have been consistently rising, with 2024 documenting the highest temperatures since records began (Lindsey and Dahlman, 2025). Heat stress (HS), once considered a problem unique to tropical climates and regions close to the equator, has demonstrated its effect on animals in more Northern latitudes, where summers are short but there is little cooling occurring during the night hours (Polsky and von Keyserlingk, 2017; Müschner-Siemens et al., 2020). Heat stress poses a concern to animal welfare and production, wherein decreased dry matter intake (DMI; Gorniak et al., 2014), decreased reproductive success (Jordan, 2003), and reduced milk yield (MY; Ingraham et al., 1978) are among the most well-reported consequences. The hypothesized mechanism behind these adverse effects is thought to be a result of altered internal body temperature, likely leading to impaired physiological function (West, 2003).

As ambient temperatures rise, the efficiency of dairy cattle to undergo non-evaporative cooling measures such as convection, radiation, and conduction is decreased (West, 2003). When non-evaporative cooling techniques are limited, cows rely on evaporative cooling, such as sweating and panting, which are less efficient as relative humidity increases (West, 2003). The temperature–humidity index (THI) is a well-established form of quantifying the ambient temperature and humidity in order to understand critical temperatures for dairy cows (West et al., 2003).

Rumination time (RT) has been utilized to gain insight into the overall and digestive well-being of dairy cows, with dramatic deviations from an individual’s baseline often indicative of metabolic dysregulation. Healthy cows are reported to ruminate between 463 and 522 min/d; however, metabolic disease, environmental stressors, and pain have been shown to reduce daily RT (Soriani et al., 2012). The use of sensor-based RT monitoring has become a way to explore early disease monitoring, with much of the current technology on the market utilizing RT deviations to detect diseases (Simoni et al., 2024).

Recent studies looking at cross-generational impacts of HS have emphasized the lasting effect of in-utero HS on several generations (Laporta et al., 2020; Kipp et al., 2021; Larsen and Laporta, 2024). Though many studies have observed detrimental effects from HS, little research exists addressing the effects of HS on the early periparturient period, as well as differences in HS across lactation stages. The aim of this retrospective observational cohort study was to 1) explore differences in rumination time in differing HS levels, 2) explore changes in RT in the periparturient period of cows who calved during mild or high HS, and 3) analyze the effects of lactation stage on time to return to baseline RT after a heat wave. We hypothesized that increased THI would result in decreased RT, and that calving in during high heat stress would have negative implications on early lactation RT. Furthermore, we hypothesized that early lactation cows would take longer to return to their baseline RT after experiencing consecutive days with increased heat loads.

2 Materials and methods

2.1 Data collection

From May 1 to October 1, 2024, RT data were retrospectively collected from cows (n = 109) on a conventional dairy farm in Western New Hampshire milking approximately 400 Holstein and Jersey cows. Raw rumination was recorded by the ear tag loggers every 2 h, where data from individual tags were automatically transferred to system software (HR-LDn tags, SCR Engineers, Ltd., Netanya, Israel; Merenda et al., 2019) every 20 min. The 2-h records of RT were summed for 24-h periods to yield daily RT (min/d). The study population consisted of Holstein (HO) dairy cows, housed in a free stall greenhouse barn that utilized fans when ambient temperatures exceeded 21°C with a relative humidity of above 25%. Additionally, the side walls of the barn were rolled up to increase ventilation and airflow within the barn throughout the study period. Cows of similar production and lactation stages were housed together in pens with maximum stocking density of 100% and fed rations designed to meet or exceed their nutritional requirements for their respective stage of lactation. The farm has designated high and low cow groups, and heifers and first lactation animals are housed together. The ration for these cows consisted largely of corn and grass silage, concentrates, and supplements appropriate for the respective stages of lactation. All cows were provided ad libitum access to fresh water.

Retrospective temperature data was obtained daily throughout the experimental period via the weather database, VisualCrossing (Visual Crossing Corporation, Reston, VA, USA), with a set location of Claremont, NH, USA. The location of the base station that recorded weather data was approximately 7 km from the farm. The maximum daily temperature and mean daily temperature was used to calculate the daily THI using the following equation from NOAA (1976):

THI=0.8×Temp(°C)+(%relativeHumidity100)×(Temp(°C)−14.4)+46.4

2.2 Statistical analysis

Prior to statistical analysis, the dataset was checked for completeness. Cows with more than 3 consecutive days of missed rumination data were excluded from analysis (n = 4), and five cows included in the final dataset were missing 2 d of RT data. Statistical analyses were conducted using R version 4.5.1 in RStudio (Posit Software, PBC, Boston, MA). Separate models were built to analyze multiple different outcomes. Fixed effects included parity (1, 2, ≥3), lactation stage [lactstage; early (EL;< 90 DIM; n = 25), mid (ML; 90–200 DIM; n = 32), and late (LL; > 200 DIM; n = 49) lactation, where applicable], and heat stress level which was differentially defined based on the given model. Biologically relevant interactions between the variable of interest and other fixed effects were tested, and cow ID was included as a random effect in all models with repeated measures where appropriate. Calendar date was not included as a random effect because the definition of the HS exposure was defined at the day level, making THI and calendar date highly collinear. Models were refined using backwards stepwise elimination, wherein any variables with P ≥ 0.10 were removed, assuming no change of more than 20% in the estimates of other variables. Significance was declared at P < 0.05, and tendencies if 0.05< P< 0.10. The AR(1) covariance structure was selected for all models (Models 1 and 2) due to having the lowest Akaike information criterion. Models 1 and 2 utilized Satterthwaite’s degrees of freedom method, as per the R lme4 package (Jensen et al., 2026). Model fit was tested for all models by examining normality of residuals and quantile-quantile plots. Means are reported as estimated marginal means and their associated 95% confidence intervals, wherein separate pairwise contrasts using the Tukey adjustment were run for variables with > 2 categories.

2.3 Model 1

The complete dataset [n = 105 cows; n = 13001 observations (obs)] was used to analyze differences in RT based on the level of HS by day. Using the maximum daily THI, days were retrospectively grouped into 4 HS categories; 1) no HS (NHS; THI< 68; n = 26 d), 2) mild HS (MHS; 68 ≤ THI< 72; n = 27 d), 3) moderate HS (MOHS; 72 ≤ THI< 80; n = 62 d), or 4) severe HS (SHS; THI ≥ 80; n = 39 d). Using the mean daily THI, days were retrospectively classified into the same 4 groups, wherein the number of days spent in NHS, MHS, MOHS, and SHS were 109, 21, 24, and 0 d, respectively. A sensitivity analysis between the groupings using the maximum and mean THI can be found in Table 1. These temperature groupings were loosely based on recommendations from Wellman (1973) and Armstrong (1994), where they reported that THI above 79 would be severe for livestock, and THI below 72 would be mild to no heat stress conditions. Fixed effects included parity, HS category, DIM, and subsequent interactions between the variable of interest and the other fixed variables. The final model included breed and HS category. The two models looking at mean versus maximum THI classifications were analyzed separately. A Tukey adjustment was used to compare pairwise comparisons between HS levels. Using the lme function of the lme4 package of R, the following linear mixed models were built for both mean and maximum THI, which included cow ID as a random effect and the correlation structure of Autoregressive Order 1 (AR(1)):

Table 1. Sensitivity analysis between use of the maximum temperature derived temperature–humidity index (THI) values versus the mean temperature derived THI values.

Retrospective data from a 154-d period (May 1 to October 1, 2024) was obtained and THI values based on both maximum and mean daily temperatures were classified into 1 of 4 heat stress (HS) levels: 1) no HS (NHS; THI< 68), 2) mild HS (MHS; 68 ≤ THI< 71.9), 3) moderate HS (MOHS; 72 ≤ THI< 79.9), or 4) severe HS (SHS; THI ≥ 80). During the 154-d period, only 26 days under conditions with NHS were agreed upon by both models.

RT=HS Level+ϵ

2.4 Model 2

Using a subset of cows that calved during the experimental period (n = 39), a model was built to analyze the change in RT over the first 14 d postpartum. Cows with< 14 d worth of complete RT data postpartum were excluded from the analysis (n = 4), leaving a total of 35 cows for the repeated measures analysis. The variable of interest was calving in HS (CIHS), wherein maximum THI from -3 to -1 d prepartum were averaged for each cow. That average maximum THI value was then assigned to 1 of 4 HS categories, as described in Model 1: 1) NHS (n = 0), 2) MHS (n = 18), 3) MOHS (n = 20), or 4) SHS (n = 4). Due to unbalanced data, cows calving in during NHS or MHS were categorized as calving during mild HS (MHS; n = 18), and cows calving in during MOHS or SHS were classified as high HS (HHS; n = 24) calvings. Parity was removed from the model due to insignificance, and health and dystocia data was not available for these cows. Using the lme function of the lme4 package of R, the final linear mixed model included the fixed effects of CIHS and DIM, the interaction between CIHS × DIM, the random effect of cow ID, and utilized the correlation structure of AR(1).

RT=CIHS+DIM+CIHS×DIM+ϵ

2.5 Model 3

A Cox proportional hazards (CPH) model was built to analyze the hazard of a cow returning to their non-heat stressed, individual baseline RT after experiencing a heat wave, where the max THI was categorized as MOHS (72< THI< 80) using the mean THI classifications for 14 consecutive days (July 5 to July 18, 2024). We analyzed the time it took for cows to return to their non-heat stressed, baseline RT within 11-d post-heat wave. The 11-d post-heat wave included no days with MOHS or SHS (July 19 to July 29, 2024; THI< 72; NHS: n = 5 d; MHS: n = 6 d) based on mean THI values. Twenty four cows were excluded due to at least 1 d of missing RT data during the 11-day follow-up period, and therefore only 81 cows remained in the final analysis. Baseline RT was obtained by averaging RT on days where THI< 72 (based on mean THI values), and a recovery event was considered if that animal had at least 3 consecutive d of RT above their non-stressed baseline within the 11 d post-heat wave. Mean THI values were utilized to ensure that the effects of HS were evident, even using the lower daily (mean) THI values as opposed to the maximum. The recovery day was calculated using the first day in which the consecutive days above baseline RT began. The additional fixed variable of lactstage (EL:< 90 DIM, ML: 90–200 DIM, LL: > 200 DIM) entered the model as the variable of interest. Shoenfeld residuals were tested to ensure that the assumption of proportional hazards was met, and no evidence of violation was observed (global test, P = 0.58). Using the Coxph function in the survival package of R, the final CPH model included the fixed effect of lactstage at the day of return to baseline RT;

hi(t)=h0(t)exp(β1LactationStagei)

Wherein h_i(t) represents the risk of returning to baseline RT for cow i at time t, h₀(t) represents the risk of early lactation animals (referent) returning to baseline RT, B₁ represents the effect of lactation stage on the hazard of return, and lactation stage represents the categorical predictor variable for cow i.

3 Results and discussion

The average daily temperature using the mean daily temperature was 19.17 ± 3.95 °C (mean ± SD) throughout the duration of the study period, whereas the average daily maximum temperature was 25.51 ± 4.75 °C. Mean daily THI over the 154-d experimental period was 65.16 ± 6.04, where maximum daily THI was 74.74 ± 7.12. Figure 1 depicts daily mean and maximum THI over the study period as well as mean daily rumination time.

3.1 Model 1

The objective of this model was to evaluate the association between RT and the differing levels of HS over the entire study period using both the mean and maximum daily THI. Table 1 depicts the sensitivity analysis between the use of the mean THI versus maximum THI values. Table 2 depicts mean RT under each HS condition for mean and maximum daily THI.

Table 2. Estimated marginal mean [95% confidence interval] rumination time (RT; min/d) in 105 Holstein cows (13, 001 observations) by heat stress (HS) level.

All days between May 1 through October 1, 2024 had maximum and mean THI values calculated based on the maximum and mean daily ambient temperature. The THI from each was then categorized into 4 groups; 1) no HS (NHS; THI< 68), 2) mild HS (MHS; 68 ≤ THI< 72), 3) moderate HS (MOHS; 72 ≤ THI< 80), or 4) severe HS (SHS; THI ≥ 80). Different superscripts denote a significant different between HS levels (all P< 0.001).

The average DIM was 167.6 ± 94.4 DIM. The parity breakdown for cows was parity 1(n = 24), parity 2 (n = 26 cows), and parity ≥3 (n = 55 cows). For Model 1 (n = 105 cows; n = 13, 001 obs), only the effect of HS level was retained in the final models. Using the mean THI, all HS levels were significantly different from one another, where cows experiencing NHS were significantly different from cows experiencing MHS and MOHS [NHS: 584 (95% CI: 574, 593) min/d vs MHS: 579 (570, 589) min/d; P = 0.005; NHS: 584 (574, 593) min/d vs MOHS: 562 (553, 572) min/d; P < 0.001; Table 1]. Cows under MHS also had significantly higher RT than cows experiencing MOHS (P< 0.001). Using daily maximum THI values, cows experiencing NHS or MHS had significantly higher RT compared to cows experiencing MOHS [NHS: 588 (578, 597) min/d vs MOHS: 581 (572, 591) min/d; P < 0.001; MHS: 585 (575, 594) min/d vs MOHS: 581 (572, 591) min/d; P = 0.05; Figure 2; Table 1]. Cows experiencing SHS had significantly lower RT compared to all other HS groups (SHS: 569 [560, 579] min/d vs NHS: 588 [578, 597] min/d; P < 0.001; SHS: 569 [560, 579] min/d vs MHS: 585 [575, 594] min/d; P < 0.001; and SHS: 569 [560, 579] min/d vs MOHS: 581 [572, 591] min/d; P < 0.001).

Sensitivity analysis between the mean and maximum THI was conducted and results are depicted in Table 1. Of the 154-d period, the two classifications only agreed for 26 of the days, of which all 26 were days classified as NHS by both the mean and maximum THI. Additionally, when using mean daily THI values, no days were considered to be SHS using the arbitrary classifications described above. Contrastingly, the maximum daily THI values yielded a total of 39 days with SHS over the experimental period. The poor agreement between the mean and maximum THI classifications suggest that the relative change in RT is dependent on the THI (mean or maximum) definition utilized during classification.

Heat stress has been reported to have a multitude of adverse effects on production outcomes in dairy cows (Ingraham et al., 1978; Jordan, 2003; Gorniak et al., 2014). A THI< 68 has been well reported in previous literature to declare days without HS (Mader and Davis, 2004), however research out of Germany has observed adverse changes in RT in Holstein Friesian cows at THI values as low as 52 (Müschner-Siemens et al., 2020). We categorized HS into 4 categories, and while arbitrary, they coincide well with other literature that details a THI of 77 to 82 being an adequate threshold to define SHS, whereby production and health are negatively impacted (Wellman, 1973; Mader and Davis, 2004; Dahl et al., 2020).

As we hypothesized, days with an increased heat load resulted in cows ruminating less, which could likely be due to the increase in heat production produced by fermentation and rumination (Brosh et al., 1998; Mader and Davis, 2004; Liu et al., 2019). Shiao et al. (2011) reported that Holstein cows housed in a barn equipped with fans and sprinklers (Fan + SP) ate significantly less DMI (18.5 kg/d) compared to cows housed within a tunnel-ventilated and water padded barn (TP; 18.8 kg/d) or TP barn with sprinklers (TP + SP; 18.8 kg/d), however they observed no significant differences between eating (4.8 to 5.7 h/d) and lying times (11.5 to 12.3 h/d) between treatment groups. For all cows in the Shiao et al. (2011) study, relative to comfortable conditions, cows spent more time eating and less time lying down during hot and humid conditions. Despite increased time eating, there was no significant difference observed in DMI during times of heat stress between the groups (Shiao et al., 2011), likely indicating more frequent visits to the feed bunk with less DMI during each visit. Although these authors found that days with higher heat and humidity increased eating time, other literature indicates that increased THI above 60 demonstrated a significant decrease on milk yield and DMI in mid-lactation Holstein cows (Gorniak et al., 2014). Our findings corroborate the findings of Gorniak et al. (2014), where we found increasing THI values resulted in overall decreased RT.

3.2 Model 2

The objective of this exploratory model was to analyze associations between RT in the first 14 d postpartum with the level of HS cows (n = 39) experienced around the time of calving (MHS: THI< 72 or HHS: THI ≥ 72). When looking at RT in the first 14 DIM, cows calving in MHS tended to have higher RT than cows that calved in HHS [MHS: 586 (548, 624) min/d vs HHS: 543 (514, 572) min/d; P = 0.08]. Daily means and associated 95% CI for each respective CIHS group can be found in Table 3.

Table 3. Estimated marginal mean [95% confidence interval] rumination time (RT; min/d) through 14 d postpartum in 35 Holstein cows by calving in heat stress (CIHS) level.

All days between May 1 through October 1, 2024 had maximum THI values calculated based on the maximum and mean daily ambient temperature. The day of calving maximum THI value was then categorized into 1 of 2 CIHS groups: 1) mild HS (MHS; THI< 72) or 2) high heat stress (HHS; THI ≥ 72) by averaging the daily THI and assigning it to a corresponding CIHS group.

In general, regardless of CIHS, following calving RT increased across the first 7 d postpartum, as depicted in Figure 2. However, RT in the HHS group tended to be lower than their MHS counterparts throughout the entirety of the 14 d period (P = 0.08). A study looking at similar effects by Paudyal et al. (2016) found that when looking at a cohort of Holstein cows calving in hot (June 1 – September 30; n = 77) or cool (November 1 – April 30; n = 118) calving seasons, prepartum RT was significantly higher in the cool (447 ± 2.8 min/d) than hot (428 ± 3.6 min/d) season. However, contrary to our findings, Paudyal et al. (2016) identified no difference between daily average RT by calving season group from 0 to 14 d postpartum, but at 14 d postpartum cows calving in the hot season had reduced RT compared to those calving in the cool season (432 vs 487 min/d, respectively). It is of note that Paudyal et al. (2016) observed differences in RT later in the transition period (14 d postpartum), whereas our findings demonstrated a tendency for HHS cows to have lower RT during the first 14 d postpartum, however we did not follow cows for longer to determine if these effects persisted. Perhaps this discrepancy could be a result of the fact that Paudyal et al. (2016) calving season classification was determined based on the month of calving, while our CIHS groups were determined based on the THI from the last 3 days of the prepartum period.

Though only a statistical tendency, cows calving during MHS ruminated approximately 60 min more per day during the first 14 d postpartum compared to those calving during HHS. Lower RT around parturition have been associated with greater serum concentrations of non-esterified fatty acids, beta hydroxybutyrate, and haptoglobin (Santos et al., 2024), lower postpartum total calcium (Seely and McArt, 2024), and a greater frequency of postpartum disease (Soriani et al., 2012). The tendency for HHS cows to have lower RT could also be reflective of adverse effects on milk yield and production aspects (Souza et al., 2022). However, it is important to emphasize that this analysis was exploratory, and our sample size is limited for generalizing these results across different farms. We can hypothesize that cows calving in MHS experienced less environmental stress allowing them to increase the speed of their return to levels of prepartum RT. Knowing the hormonal and physical stress that calving entails, calving in during HHS likely exacerbates these difficulties, resulting in a more difficult return to homeostatic regulation.

3.3 Model 3

The objective of this model was to understand differences in time to return to baseline RT after an extreme heat wave across different lactation stages. The average daily temperature (mean temperature ± SD) was 25.60 ± 0.98 °C for the duration of the heat wave, and the average daily mean THI was 79.70 ± 1.61. During the 11-d post-heat stress period, the average daily temperature and THI were 21.37 ± 0.74 °C and 68.41 ± 1.36, respectively. Sixty-six cows returned to their non-heat stressed baseline RT, whereas 15 cows did not return within the follow-up period. The cows included in this analysis consisted of 24, 19, and 38 parity 1, 2 and >3 cows, and 17, 29, and 35 early, mid, and late lactation cows, respectively.

In the Cox proportional hazards model looking at time to return to baseline RT, lactstage remained in the final model as the variable of interest. Though no statistically significant differences existed between lactation stages, the early lactation cows had a numerically lower hazard of returning to baseline RT within 11 d post-heat wave compared to their mid and late lactation counterparts [ML: Referent; EL: Hazard Ratio (HR) = 0.95 (0.48, 1.85); LL: HR = 1.10 (0.63, 1.89); P = 0.90].

It took cows in all stages of lactation approximately the same amount of time to return to their baseline RT after a series of days experiencing HS (EL: 4.95 ± 3.85 d to return; ML: 4.90 ± 3.93 d to return; LL: 4.94 ± 3.83 d to return; P = 0.90; Figure 3). Although there was variation in the sample sizes between the lactation groups (EL: n = 17; ML: n = 29; LL: n = 35), the hypothesized outcome was that a heat wave would have a greater negative effect on EL animals. During the first 21 d postpartum dairy cows face significant changes in physiology and behavior due to increased nutrient demands and the onset of lactation (Bradford et al., 2015). Additionally, approximately one third of dairy cows will experience a clinical disease and > 50% will experience some sort of subclinical disorder within the first 90 DIM (Caixeta and Omontese, 2021). The range of DIM categorized into the EL category (< 90 DIM) encapsulated the transition period, as well as the first 90 DIM, in an attempt to encompass the increased risks for disease, which would likely have an effect on a cow’s ability to return to their baseline RT.

Despite our initial hypothesis that lactation stage would impact a cow’s ability to return to baseline RT following a heatwave, we observed minimal differences between lactation stages following a heat event. Interestingly, early lactation cows took a similar amount of time to return to baseline RT after the heat wave compared to mid- and late-lactation cows. However, the limited cow numbers used in the analysis for Model 3 could have increased our chances of type II error. More work is needed in the future to determine whether or not these differences could exist.

A study by Leliveld et al. (2023) found that at a THI< 68, ML cows (150–220 DIM) ruminated more than EL cows (< 150 DIM). This association observed in the work from Leliveld et al. (2023) contrasts our data that numerically, where cows in all lactation stages took about the same time to return to their baseline RT. However, it is important to consider that all days post-heat-wave were spent with THI > 68, which has been reported to be associated with negative production effects in dairy cows (Mader and Davis, 2004).

We conducted a retrospective, observational cohort study and it is worth noting that our sample sizes varied between our different models. Due to the retrospective and exploratory nature of this study, power analyses were not performed for the outcomes of interest prior to the study, increasing the likelihood of committing type I and II errors when deciphering significance. Additionally, the use of a single farm limits the generalizability of this data and warrants multi-site and season validation in future studies. Additionally, the Cox Proportional Hazard model (Model 3) rests on a single iteration of a 14-d heatwave, so the generalization of these results may be limited. Perhaps conducting future studies with larger sample sizes to specifically address individual models reported herein would allow more insight into the variety of differences we observed within our analyses. Additionally, it could be interesting to compare the daily minimum THI values for this experimental period with the maximum THI values which we utilized in this study. It is highly possible that reclassification of THI based on the daily minimum temperatures, as opposed to the daily maximum, would adjust our results significantly.

4 Conclusions

Moderate and severe HS significantly decreased RT in Holstein dairy cows. Additionally, cows that calved during mild HS tended to have increased RT during the first 14 d postpartum when compared to cows calving during high HS. These findings incite questions regarding the physiological changes around parturition that HS may exacerbate and emphasize the importance of heat abatement methods on farm.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Charlotte Reed, DairyNZ, New Zealand

Reviewed by: Md. Morshedur Rahman, Bangabandhu Sheikh Mujibur Rahman Agricultural University, Bangladesh

Aarti Kamboj, National Dairy Research Institute (ICAR), India