Study farms and data sources

This study was conducted on five commercial farms, all located in Quebec, Canada, between 2022 and 2023. The farms were selected to cover a broad range of calf management practices but were required to have enough calves yearly to provide enough data for our analyses. The number of female Holstein calves on farm varied from 113 calves on Farm 2 to 880 on Farm 3 (Table 1). Furthermore, farms were required to feed their calves with an AMF with data recording capabilities and possessing a year’s worth of recorded information necessary to capture seasonal variations. All farms were equipped with one or more AMF (Förster-Technik, Engen, Germany), and data were accessible via the Calf Cloud (Förster-Technik, Engen, Germany; Farm 1 and Farm 5), a secure data card (SDC; Farm 2) or the Kalb Manager software (Förster-Technik, Engen, Germany; Farm 3 and Farm 4).

Table 1.

Characteristics of the 5 commercial farms participating in this study and descriptive statistics of the AMF recorded metrics observed on 1,995 preweaning dairy calves (from 7 to 70 days of age)

Table 1 Long description

Five commercial dairy farms are compared on herd size, calf enrollment, BRD treatment practices, feeding programs, automated milk feeder setup, staffing, and calf drinking metrics for preweaning calves. Lactating cow numbers range from 230 on Farm 2 to 1,400 on Farm 3, and study calves range from 113 on Farm 2 to 880 on Farm 3. BRD-treated calves range from 20 percent on Farms 2 and 5 to 41 percent on Farm 1; most farms used ZELERIS for BRD treatment, while Farm 2 used RESFLOR. Milk delivery differs by farm, including milk replacer or fresh or pasteurized milk, with AMF milk allowance typically 12 liters per day except Farm 3 (ad libitum) and Farm 4 (8 to 12 liters). Calves entered the AMF between 7 and 14 days of age, weaning occurred from about 60 to 70 days, and staffing ranged from 1 to 3 employees caring for calves. Average consumption is similar across farms at roughly 6.6 to 7.3 liters per day, while average drinking speed varies more, from about 569 milliliters per minute on Farm 1 to about 947 milliliters per minute on Farm 5. Visit patterns differ, with unrewarded visits highest on Farm 4 and rewarded visits highest on Farms 1 and 2. Some values are reported as ranges and some as averages with standard deviations, so direct comparisons should consider differences in reporting format and farm management context.

SDC, secure data card.

^a Composition: florfenicol and meloxicam.

^b Composition: florfenicol and flunixin.

^c Composition: bovine rhinotracheitis–parainfluenza 3–respiratory syncytial virus–Mannheimia haemolytica–Pasteurella multocida.

^d Composition: bovine rhinotracheitis–parainfluenza 3–respiratory syncytial virus.

In parallel, farms were also required to have a comprehensive recording of all treatments administered to calves differentiating BRD versus non-BRD treatments. The way treatments were recorded varied among the farms. For example, Farms 1 and 2 relied on traditional paper-based recording systems, Farm 3 utilized the software Uniform-Agri (Bridgwater, UK) and Farms 4 and 5 employed DairyComp (Valley Agricultural Software, Tulare, CA). Overall, our study gathered 90,088 daily calf records covering 1,995 unique Holstein calves.

Calf housing and feeding management

Regarding calf housing practices, they were all fed at least 3 L of colostrum within 6 h of birth (Table 1). The threshold to qualify the colostrum of sufficient quality varied between farms. Farm 1 required a minimal Brix level of 20%, while Farms 2 and 5 opted for 22%, and Farms 3 and 4 targeted a Brix level of up to 24%. Calves were initially housed in individual boxes where they were fed at least 6 L of milk per day, and water was ad libitum. Information on the milk composition provided to calves on each farm is presented in Table 1. Calves were subsequently transitioned to group pens and accessed the AMF for the first time between 7 and 14 days of age (Table 1). This management approach is consistent with previous research indicating that calves are typically housed individually until a mean age of 5.1 ± 3.9 days, with reported ranges from 0 to 14 days (Jorgensen et al., Reference Jorgensen, Adams-progar, De Passillé, Rushen, Godden, Chester-jones and Endres2017). Milk feeding practices during the study ranged from 8 L of milk or milk replacer per day to ad libitum, depending on individual farm practices and feeding plans. On all the farms, there was a consistent practice of introducing new calves into the existing groups. The farms in this study ensured that the age difference between calves in each group was no more than 2 weeks.

The number of pens as well as the number of calves per group varied from one farm to another. On Farm 2, there were two pens with one AMF and two distribution stations: one distribution station per pen. The distribution station is where the milk is dispensed to the calves via teats. On Farms 1 and 4, there were four pens with two AMF and one distribution station per pen. Meanwhile, on Farm 3, there were 10 pens with 14 distribution stations and 7 AMF. The number of calves per pen varied from 10 to 13 calves on all the farms and depended on the month and the number of calves born at a given period. All the farms declared that the number of calves per group can be more than 14 calves on a given period, but they try to keep it lower than 14 calves per pen. The AMF fed the calves a combination of whole milk and milk replacer depending on the farm. Calves had unlimited access to forage, concentrate feeds and water during the entire pre-weaning and weaning periods.

Bovine respiratory disease identification and farm management practices

BRD complex was detected based on treatment given to calves directly by the producers or workers. A calf was deemed sick if it received at least one treatment for BRD, as indicated by the treatment records provided by the farm and considered healthy otherwise. The day of the first treatment was designated as the diagnostic day for the purposes of this study. All farms use a treatment consisting of an antibiotic and a nonsteroidal anti-inflammatory (Table 1). The number of employees responsible for detecting illnesses varied from one farm to another. For example, three employees were responsible for treatment on Farm 1, while two were responsible on Farm 2 and one on Farms 3, 4 and 5. This variation in personnel allocation underscores the diversity of operational structures and management practices among the participating farms. Although there were variations observed between farms concerning infrastructure and certain practices, all farms housed calves in groups starting at a specific age and fed them with AMF until weaning, as shown in Table 1. The management practices closely resembled those identified in previous Canadian studies (Vasseur et al., Reference Vasseur, Borderas, Cue, Lefebvre, Pellerin, Rushen, Wade and De Passillé2010), which suggests that the findings of the current study may offer broader insights into dairy commercial farming practices in Quebec.

Automated milk feeder data and statistical analyses

Daily feeding data recorded by the AMF included the date, calf identification, daily consumption in mL, average daily drinking speed in mL/min and number of daily rewarded or unrewarded visits. All retrieved feeding data were obtained either from the SDC, Calf Cloud or Kalb Manager and were standardized and assembled using R version 4.4.2 (R Core Team, 2024) into one data file. Similarly for the treatment data, they were transcribed manually in Excel for the paper recorded format and all imported and standardized in R. Descriptive statistics of the heifers and comprehensive statistical analyses were conducted using R. To ensure robustness and accuracy in our analyses, healthy calves and calves with BRD were uniquely paired to form a pair based on the month and days spent at the AMF, with a tolerance of ±2 days. In total, 87, 22, 210, 45 and 76 pairs of calves were used in the analyses on Farms 1–5, respectively. The statistical analysis only considered 5 days before the first treatment. One linear mixed model was constructed for each farm and metric (daily consumption, drinking speed, rewarded and unrewarded visits) using the packages ‘lme4’ (version 1.1-36) and ‘emmeans’ (version 1.10-7). Briefly, the mixed models included health status (healthy or BRD), days relative to treatment (from −5 to 0 days), month and interactions between day and health status as fixed effects. The calf is considered the experimental unit and was included as a random effect to account for individual variability within the data. Results are deemed significant with a P-value ≤0.05, and a tendency is reported for 0.05 < P-value ≤ 0.10. All models were validated using the check_model function from the ‘performance’ package (version 0.14), and met the assumptions of normality and homoscedasticity.

Between-farm variability in feeding behaviour

The average consumption of calves varied across farms, ranging from 6.6 L/day on Farm 5 to 7.3 L/day on Farm 4 (Table 1). Similarly, the average drinking speed exhibited notable differences, with calves on Farm 1 consuming at a rate of 569 mL/min, whereas those on Farm 5 exhibited a fastest rate of 947 mL/min. Rewarded visits also demonstrated variability, ranging from an average of five visits for Farms 4 and 5 to an average of nine for Farms 1 and 2. Conversely, unrewarded visits varied from 3 on Farm 1 to 13 on Farm 4.

Associations between feeding behaviours and BRD

Total milk intake (mL/day) differed significantly between healthy calves and those with BRD on three of the five farms (Figure 1). No significant differences were observed on Farms 1 and 2. However, on Farm 3, total milk intake was significantly associated with BRD up to 3 days before diagnosis (P = 0.008). On Farm 4, a significant reduction in intake was evident 2 days before the producer’s diagnosis (P = 0.005), while on Farm 5, the association was observed as early as 5 days before diagnosis (P = 0.008). These results suggest that BRD-affected calves consumed significantly less milk than healthy calves, although this pattern was only observed on specific farms and for each farm, at a specific number of days before producer intervention.

Figure 1.

Milk consumption (mL), consumption speed (mL/min) and the number of rewarded and unrewarded visits per calf per day for healthy calves and calves with BRD, paired on Farm 1 (n = 87), Farm 2 (n = 22), Farm 3 (n = 210), Farm 4 (n = 45) and Farm 5 (n = 76). Significance levels are indicated as ⁺P < 0.1, *P < 0.05, **P < 0.01 and ***P < 0.001 and represent differences according to BRD status by day. The earliest day with a significant difference before treatment (e.g., 0 day, −1 day) for each farm and metric is presented next to the corresponding n value.

Figure 1 Long description

Farm 1. Consumption (L). x-axis: Days before treatment, range negative 5 to 0. y-axis: Consumption (L), range 5 to 9. Text: n equals 87. Two line series with point markers and vertical error bars. Values stay near 6 to 7 across negative 5 to 0 for both series, with small day-to-day changes. Speed (mL/min). x-axis: Days before treatment, range negative 5 to 0. y-axis: Speed (mL/min), range 400 to 1000. Text: n equals 87, 0d. Two line series with point markers and vertical error bars. Values stay near 500 to 600 across negative 5 to 0 for both series. Number of rewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of rewarded visits, range 5.0 to 12.5. Text: n equals 87. Two line series with point markers and vertical error bars. Values stay near 7.5 to 10 across negative 5 to 0 for both series. Number of unrewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of unrewarded visits, range 0 to 5. Text: n equals 87. Two line series with point markers and vertical error bars. Values stay near 0 to 2 across negative 5 to 0 for both series. Farm 2. Consumption (L). x-axis: Days before treatment, range negative 5 to 0. y-axis: Consumption (L), range 5 to 9. Text: n equals 22. Two line series with point markers and vertical error bars. Values are near 6 to 7 across negative 5 to 0 for both series, with a dip near negative 3 to negative 1 in one series. Speed (mL/min). x-axis: Days before treatment, range negative 5 to 0. y-axis: Speed (mL/min), range 400 to 1000. Text: n equals 22, minus 2d. Two line series with point markers and vertical error bars. Values are near 550 to 700 across negative 5 to 0. Asterisk markers appear near negative 2 and 0. Number of rewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of rewarded visits, range 5.0 to 12.5. Text: n equals 22. Two line series with point markers and vertical error bars. Values are near 7.5 to 10 across negative 5 to 0. Number of unrewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of unrewarded visits, range 0 to 20. Text: n equals 22, minus 5d. Two line series with point markers and vertical error bars. Values are near 0 to 6 across negative 5 to 0. Asterisk markers appear near negative 5, negative 3 and 0. Farm 3. Consumption (L). x-axis: Days before treatment, range negative 5 to 0. y-axis: Consumption (L), range 5 to 9. Text: n equals 210, minus 3d. Two line series with point markers and vertical error bars. One series stays near 6.5 to 7.0 across negative 5 to 0. The other series stays near 7.0 to 7.5 across negative 5 to 0. Plus and asterisk markers appear from about negative 4 through 0, including* * and * *asterisk markers. Speed (mL/min). x-axis: Days before treatment, range negative 5 to 0. y-axis: Speed (mL/min), range 400 to 1000. Text: n equals 210, minus 2d. Two line series with point markers and vertical error bars. Values are near 550 to 650 across negative 5 to 0. Plus markers appear near negative 4, negative 3 and negative 2. Asterisk* markers appear near negative 2 and 0. Number of rewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of rewarded visits, range 5.0 to 12.5. Text: n equals 210, minus 2d. Two line series with point markers and vertical error bars. Values are near 7.0 to 10.5 across negative 5 to 0. Asterisk markers appear near negative 2, negative 1 and 0. Number of unrewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of unrewarded visits, range 0 to 20. Text: n equals 210, minus 3d. Two line series with point markers and vertical error bars. Values stay near 2 to 4 across negative 5 to 0. Plus markers appear near negative 3 and negative 2. Farm 4. Consumption (L). x-axis: Days before treatment, range negative 5 to 0. y-axis: Consumption (L), range 5 to 9. Text: n equals 45, minus 2d. Two line series with point markers and vertical error bars. Values are near 7.0 to 8.5 across negative 5 to 0. Plus and * * markers appear near negative 3 to 0. Speed (mL/min). x-axis: Days before treatment, range negative 5 to 0. y-axis: Speed (mL/min), range 400 to 1000. Text: n equals 45, minus 1d. Two line series with point markers and vertical error bars. Values are near 550 to 700 across negative 5 to 0. Plus markers appear near negative 2 and negative 1. Number of rewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of rewarded visits, range 2.5 to 7.5. Text: n equals 45. Two line series with point markers and vertical error bars. Values are near 4.0 to 6.0 across negative 5 to 0. Plus markers appear near negative 2 and negative 1. Number of unrewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of unrewarded visits, range 0 to 20. Text: n equals 45, minus 4d. Two line series with point markers and vertical error bars. Values are near 6 to 16 across negative 5 to 0. Asterisk and * * markers appear near negative 4 through 0. Farm 5. Consumption (L). x-axis: Days before treatment, range negative 5 to 0. y-axis: Consumption (L), range 5 to 9. Text: n equals 76, minus 5d. Two line series with point markers and vertical error bars. One series stays near 5.5 to 6.5 across negative 5 to 0. The other series stays near 6.5 to 7.5 across negative 5 to 0. Asterisk markers appear near negative 5 through 0, including * * and * *asterisk markers. Speed (mL/min). x-axis: Days before treatment, range negative 5 to 0. y-axis: Speed (mL/min), range 400 to 1000. Text: n equals 76, minus 5d. Two line series with point markers and vertical error bars. One series stays near 650 to 750 across negative 5 to 0. The other series stays near 800 to 950 across negative 5 to 0. Asterisk* * markers appear near negative 5, negative 4, negative 1 and 0. Asterisk markers appear near negative 3. Number of rewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of rewarded visits, range 2.5 to 12.5. Text: n equals 76, minus 4d. Two line series with point markers and vertical error bars. Values are near 4.0 to 7.0 across negative 5 to 0. Asterisk* * markers appear near negative 4 and negative 1. Asterisk  * markers appear near negative 3. Number of unrewarded visits. x-axis: Days before treatment, range negative 5 to 0. y-axis: number of unrewarded visits, range 0 to 20. Text: n equals 76, minus 4d. Two line series with point markers and vertical error bars. Values are near 2 to 5 across negative 5 to 0. Plus marker appears near negative 2. Legend text: Healthy, BRD, plus P less than 0.1, asterisk P less than 0.05, * * P less than 0.01, * * asterisk P less than 0.001.

Drinking speed (mL/min) was also investigated as an indicator of dairy calf health status, with significant findings observed in comparison with BRD on all farms. On Farm 1, drinking speed showed potential for BRD detection on the day of the diagnosis by the producer (P = 0.015). On Farms 2 and 3, early detection of BRD is possible 2 days before diagnosis (P = 0.048 and P = 0.002, respectively), while it is possible 1 day (P = 0.043) and 5 days (P < 0.001) before diagnosis on Farms 4 and 5. For the number of rewarded visits, BRD could be detected up to 2 days before diagnosis on Farm 3 (P = 0.016) and even 4 days on Farm 5 (P < 0.001), but no significant differences could be detected on Farms 1, 2 and 4. The findings suggest that calves with BRD have fewer rewarded visits per day prior to treatment by the producer, and this pattern was observed in only two out of the five farms included in the study. Unrewarded visits were found to be useful for detection on all farms except Farm 1, when it could be detected 5 days before on Farm 2 (P = 0.046), 3 days on Farm 3 (P = 0.016), 4 days on Farm 4 (P = 0.040) and 4 days on Farm 5 (P = 0.042). Calves with BRD have less unrewarded visits, but we can observe that the exact number is highly farm dependent (Figure 1).

Comparison with previous studies and biological interpretation

This study provides information on the associations between BRD and differences in feeding data via an AMF on five different commercial farms. The average prevalence of BRD in this study was 25.6% and varied from 20% on Farms 2 and 5 to 41% on Farm 1. Farm 1 had a severe BRD issue the year we obtained the data, but this problem has now been resolved. The average prevalence is in line with the prevalence reported in Conboy et al. (Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021), which reported a 22% BRD prevalence, and Cramer et al. (Reference Cramer, Ollivett and Stanton2016), which reported a 26% prevalence in Holstein group-housed dairy calves in Wisconsin, United States. When it comes to disease detection, the study of Cramer et al. (Reference Cramer, Ollivett and Stanton2016) relied on the treatment data recorded by producers and the calf was considered sick if it was treated with an antibiotic or a nonsteroidal anti-inflammatory drug in the 7 days before the farm visit or on the day of the visit. This method is comparable to what was used in our study. However, Conboy et al. (Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021) used a more objective method based on a scoring system where a calf is identified with BRD if it has a cumulative score for nasal discharge, ocular discharge, ear position, cough score and rectal temperature greater than 5.

Our study revealed that calves with BRD showed reduced milk intake compared to healthy calves only on three farms (Figure 1). These findings were not observed in Conboy et al. (Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021), where calves with BRD did not show a decrease in milk intake. Similarly, Knauer et al. (Reference Knauer, Godden, Dietrich and James2017) reported no significant change in milk intake in calves with BRD around the time of treatment. Previous studies have indicated that the relationship between calf health status and feeding behaviours varies with milk feeding strategies (Borderas et al., Reference Borderas, Rushen, Von Keyserlingk and De Passillé2009; Conboy et al., Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021). Specifically, in calves fed a low amount of milk, illness may have minimal impact on feeding behaviours because available quantities are too low (Jasper and Weary, Reference Jasper and Weary2002). Research suggests that calves fed 4–6 L/day do not reduce their milk intake, but show lower feeding motivation and fewer visits to the feeder (Svensson and Jensen, Reference Svensson and Jensen2007; Borderas et al., Reference Borderas, Rushen, Von Keyserlingk and De Passillé2009). This could potentially explain the lack of significant changes in milk consumption on Farm 1, where although daily milk allowances ranged up to 12 L/day, average consumption was 6.8 ± 2.52 L/day. The significant differences in milk intake for BRD-affected calves on some farms could be attributed to the varying milk allowances provided in this study as well as the calves’ average consumption.

Drinking speed was significantly different for calves with BRD compared to healthy calves on all farms. Farms had average drinking speeds ranging between 569 ± 202.3 and 947 ± 369.3 mL/min, quite similar to the results found by Cantor et al. (Reference Cantor, Pertuisel and Costa2020) and Conboy et al. (Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021), who reported 730 ± 500 and 744 ± 274 mL/min, respectively. Both studies found a link between calf health and drinking speed. Svensson and Jensen (Reference Svensson and Jensen2007) did not find differences in drinking speed between healthy and sick calves when they were fed up to 8.6 L of milk per day, which might explain why no significant differences were observed in some cases. Knauer et al. (Reference Knauer, Godden, Dietrich and James2017) found that calves fed an average of 9.4 L/day (range: 7–16 L/day) drank slower when they were sick. This slower drinking speed could be due to competition at the feeder. However, it was reported that more competition can lead to faster drinking speeds (Jensen, Reference Jensen2006). Another possibility is that the AMF may not be regularly calibrated. However, this does not apply to the farms in our study, as most calibrate the AMF on a weekly or monthly basis (Table 1). Another potential explanation is variation in the type of nipple used at the feeder. One other possibility is that the nipple used at the feeder is different. Although drinking speed alone is not a reliable disease detection tool, it can be useful when combined with other feeding behaviours to identify sick calves (Conboy et al., Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021).

We were only able to find associations between health status and rewarded visits on Farms 3 and 5, which aligns with previous research indicating that there is no strong association between the number of rewarded visits and diseases (Knauer et al., Reference Knauer, Godden, Dietrich and James2017; Conboy et al., Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021). This result could be explained by the fact that on most farms of our study, sick and healthy calves are equally motivated to consume enough milk to maintain their needs. Healthy calves, despite having the same number of rewarded visits, can consume more milk during these visits due to their faster drinking speed. However, on Farms 3 and 5, there were significant differences in rewarded visits between healthy and BRD-affected animals, with a decrease in visits when calves were sick. These findings are supported by other studies reporting connections between rewarded visits and calf health. Some studies found a decrease in feeder visits on the day the illness was detected (Knauer et al., Reference Knauer, Godden, Dietrich and James2017; Perttu et al., Reference Perttu, Peiter, Bresolin, Dórea and Endres2023), more than 2 days before signs of diarrhoea (Sutherland et al., Reference Sutherland, Lowe, Huddart, Waas and Stewart2018; Lowe et al., Reference Lowe, Sutherland, Waas, Schaefer, Cox and Stewart2019) and respiratory illness (Svensson and Jensen, Reference Svensson and Jensen2007). In our study, when there was an association between rewarded visits and sickness, a decrease in rewarded visits was reported, which can be explained as the sick calves could have less energy to go to the feeder, especially if there is a competition with the healthy calves. However, Perttu et al. (Reference Perttu, Peiter, Bresolin, Dórea and Endres2023) reported a contrasting situation where increased visits were seen as an indicator of a sick calf.

There was no difference in unrewarded visits between BRD and healthy calves on Farm 1. This finding aligns with those reported by Perttu et al. (Reference Perttu, Peiter, Bresolin, Dórea and Endres2023) that found no discernible relationship between unrewarded visits and illness. However, there was a significant association between unrewarded visits and animal health status on Farms 2, 3, 4 and 5. These findings are also comparable to the findings of Conboy et al. (Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021), which observed that calves affected by BRD exhibited a reduction of two unrewarded visits compared to healthy calves. Similarly, Knauer et al. (Reference Knauer, Godden, Dietrich and James2017) documented an average decrease of 2.3 unrewarded visits per day in diseased calves both before and after disease diagnosis. In a related study, Johnston et al. (Reference Johnston, Kenny, McGee, Waters, Kelly and Earley2016) reported a decrease in unrewarded visits among calves with BRD during the 3 days leading up to diagnosis.

Study limitations and implications for on-farm disease detection

The heterogeneity observed within this study underscores the complexity of identifying consistent parameters for disease detection across different commercial farms. It highlights the variability in farming practices and management strategies, which can influence the significance of specific parameters in relation to disease. For instance, a parameter that proves significant for one farm may not hold the same importance for another, depending on the unique practices implemented within each farm. Conboy et al. (Reference Conboy, Winder, Medrano-galarza, LeBlanc, Haley, Costa, Steele and Renaud2021) reported that it is crucial to acknowledge that while the feeding behaviours of calves at a single point in time may not be sufficient to definitively detect diseased individuals on their own, they can still play a valuable role as a preliminary tool in the disease detection process. In this context, the present findings suggest that AMF data should be interpreted cautiously when used as a standalone indicator for the specific detection of BRD. However, the results more strongly support the potential utility of AMF-derived metrics as a means of flagging calves that may benefit from closer, targeted health surveillance. This distinction is particularly relevant in large-scale production systems, where such an approach could assist producers in prioritizing monitoring efforts towards animals at higher risk of subsequent disease development.

When evaluating the results of this study, it is important to acknowledge certain limitations. Firstly, due to our inability to be consistently present on the farm, we relied on producers for identifying BRD. This approach introduces heterogeneity in the definition of BRD, as producers do not all use well-defined standard operating procedures and do not rely on the same number of employees. The study was also performed on larger farms since the number of calves required for our analyses was higher. The results presented here could thus only apply to larger farms. Furthermore, the number of calves diagnosed with BRD varied across farms, with some farms (e.g., Farm 2) having relatively few affected animals. This imbalance limited the statistical power to detect significant differences for certain feeding behaviour metrics and should be acknowledged as an inherent constraint of this observational study. Our study was also only performed on data obtained from Forster AMF, and it is possible that some of the findings presented here might be dependent on the way this type of AMF records the different metrics. Future studies might study metrics recorded using different AMFs. These limitations underscore the need for cautious interpretation of the study results in the light of the disease definition and specific setup of our study.

In conclusion, heterogeneity in detecting BRD on five commercial farms using AMF was demonstrated. It has been shown that automatic detection of BRD using AMF data would require algorithms to be farm-specific and using a combination of different metrics. Furthermore, the number of days prior to detection by the producer was also variable. One important question is related to the number of days before detection by the producer that could help further prevent disease complications of BRD. In other terms, what is the optimal number of days before detection by the producer that could be clinically relevant for the calves? This study underscores the necessity of considering the environmental and individual farm-specific factors when designing detection tools. Moving forward, it will be crucial to personalize detection models to ensure their effectiveness and relevance under varied conditions. Additionally, the continuous improvement of detection technologies and the integration of new data will refine these models, enhancing their accuracy and utility in managing the health of dairy calves.