Quantitative methodologies to assess sleep, wellbeing and physical health in dairy farm workplaces

Study 1: developing sleep assessment methodology

In Study 1, the sleep of a cohort of dairy farmers was assessed. We sought to understand the level of compliance (nights of data capture) that could be achieved with farmer participants in studies capturing physical sleep data, identify the most important sleep metrics for comparison among different farm workplace designs, and whether differences in sleep metrics among farm management approaches could be assessed.

Measurement of the sleep patterns of nine farmers was conducted from 11 July 2020 to 7 November 2020, representing a 119-day study over the spring calving period and leading up to mating. The participants worked on three commercial dairy farms, one farm was implementing the practice of three milkings every 2 days (3-in-2), whereas the two adjacent farms, within the same company and overseen by the same operations manager, used twice-a-day (TAD) milking. Each participant was provided with an Oura™ Gen 2 sensor, a ring placed on one of the user’s fingers (Oura Health Oy, Oulu, Finland, hereafter called the ‘sensor’). Oura™ sensors have been shown to be the best consumer wearable for accurate sleep assessment (Svensson et al. 2024; Willoughby et al. 2024). The sensor measures heart rate, sleep latency, sleep efficiency, heart rate variability, respiration, bedtime start and bedtime end, sleep duration, temperature, activity of the user, and the sleep phases of deep sleep, light sleep, REM (rapid eye movement) sleep, and awake (Penzel et al. 2021). Although the sensor provides a sleep score, readiness score, and sleep phases, we did not use these in our analyses because there is uncertainty around their accuracy and application (de Zambotti et al. 2019). Additionally, body temperature data were not relevant to this study and therefore were not used in the analysis. From the sensor data, a range of metrics was used in this study, including the following:

The sensor can capture and save data for up to 7 days, and data are uploaded to a cloud-based server by linking the sensor to a smartphone app via Bluetooth. Participants were inducted into the study on 10 July 2020 and were trained in the use of the sensor and the associated smartphone app. They were instructed to put on the sensor every night before they went to bed and take it off when they got out of bed, and to charge and synchronise the sensor each day. Participants had access to the Oura™ app which provided insights on their daily sleep data and statistics. As a duty of care measure, participants were contacted by the research team using SMS if their sleep was low (below 5.5 h in a night) to note that such low sleep can result in impairment.

They were asked to not wear the sensor during the day to limit potential injury risks associated with wearing it in the work environment. All participants also completed a Pittsburgh sleep quality index (PSQI) survey (Buysse et al. 1989), and although this information was not used in the current study, it served as a means of identifying any participants with pre-existing sleep issues. Timesheet data were collected from the farms to enable classification of ‘day type’ for participants as either a work day or rest day (day off). ‘Number of milkings’ in a day was calculated as 0 (before calving started), 1 (the one milking day of flexible milking schedules), or 2 (a twice-a-day milking day) from farm milking schedules. ‘Milking schedule’ data were captured by identifying which farms used TAD or 3-in-2 milking and for the 3-in-2 farms, which days involved two or one milkings. For some of the analyses, the data were broken into either the pre-calving (dry) and post-calving (milk) phase. Dry was defined as the period before planned start of calving, and Milk was the period from planned start of calving until the end of the study.

Cross-tabulation and chi-squared test of independence was used to compare compliance for different factors such as farm, milking schedule or period. Sleep metric differences between farm milking schedules and among participants were described using means and standard deviations (s.d.). To investigate which factors explained most of the variation in key sleep parameters, the daily data were subjected to a fully nested ANOVA. Variance components included milking schedule (3-in-2 or TAD), farm (within schedule), phase (dry or milk; within farm × schedule), number of milkings (within farm × schedule × phase), day type (work or rest; within farm × schedule × phase × milkings), participant (within farm × schedule × phase × milkings × day type), and week (within participant × farm × schedule × phase × milkings × day type). Week was included to represent the changing work routines when going from calving to milking to mating. Results are presented as percentage contribution of each variance component. Repeated measures ANOVA was used to test for effects of milking schedule, number of milkings, day type (rest or work), and day on sleep parameters. A posthoc power analysis was conducted to inform future studies of the likely number of participants that would be required to detect a statistical difference for time slept. Pearson correlation coefficients were used to describe linear relationships among the different sleep parameters. Principal component analysis (PCA) followed by biplots (Vu and Friendly 2024) were used to investigate whether combining the sleep metrics was a way to identify different groups of participants.

Study 2: developing and testing wellbeing and health assessments

In Study 2, wellbeing and health measures were collected using three approaches, a paper-based wellbeing survey of 63 farmers, a shorter ‘pulse’ survey of 18 farmers, and a physical health assessment, as outlined below. The farmers, using either TAD or 3-in-2, were involved in the Flexible Milking project and opted into the wellbeing and health assessments. A range of farmer roles was included from dairy assistant through to operations managers, and a wide range of ethnicities were represented including NZ European, Māori, Asian and South American. These ethnicities represent the major ethnicities present in the NZ dairy sector (Kambuta et al. 2024). Through Study 2, we explored the potential of survey methodologies to identify differences in farmer wellbeing in farm systems research projects and whether basic physical health metrics were useful for comparing different farm system approaches.

Study 1: sleep study

Principal component analysis

Data for six sleep metrics (hours in bed, hours asleep, bedtime start, bedtime end, sleep efficiency, average HR) were summarised fortnightly, given the 3-in-2 milking schedule repeats on this cycle, and subjected to principal component analysis using standardised values (Fig. 1). Results showed that the first two components accounted for 69% of the variation, and that the two phases (dry and milk) were separated along the first component. The dry phase was characterised by greater number of hours in bed and hours asleep, mainly owing to a later bedtime end. Focusing on the milk phase, where the first two components accounted for 70% of the variation, showed a separation between participants working on 3-in-2 or TAD farms. Working on a 3-in-2 farm seems to result in earlier bedtime start and later bedtime end does working on a TAD farm. In addition, we found that the main difference between rest and work days was the inconsistency of sleep metrics preceding rest days, compared with the more consistent sleep patterns preceding a work day. Differences among participants appeared to be larger than any systematic differences between the groups, supporting the large proportion of variation being explained by participants and week and day within a participant (at least 53%).

Fig. 1.

Results of principal component analysis, showing (a) separation between dry and milk phases, (b) separation between 3-in-2 and TAD during the milk phase, (c) overlapping rest and work days during the milk phase, and (d) variation among individual participants during the milk phase.

Study 2: wellbeing and health study

Wellbeing pulse assessment during the season

The four-question wellbeing pulse survey was sent out to all participants fortnightly between July 2020 and February 2021 and received 141 responses. Although the survey was originally planned to run all season, a dropping engagement rate led to the research team deciding to cease the survey in February. By this stage, the busiest part of the season had been completed. Caution is needed when interpreting the results, because each survey period received a range of completed responses from 1 to 13 participants. Periods where there were fewer than two responses were removed. Response rate for the survey for the first 18 weeks was 50% or more, with at least 10 and up to 13 responses per period. However, from Week 20, the response rate dropped to 30%, then down to 20%, at which point the survey was discontinued.

The average response scores are shown in Fig. 2. There was a similar trend between the groups in their answers to the questions about being able to do things outside of work and getting enough sleep and rest. Before calving started, for the questions about ‘enjoyment of the job’ and an ‘overall score’ for the past fortnight, respondents on the TAD farms had higher mean scores than did those on flexible milking farms. Over spring, scores from people on the flexible milking-scheduled farms remained relatively stable compared with those from the TAD farms, where enjoyment and overall satisfaction declined.

Fig. 2.

Data from the fortnightly wellbeing pulse survey where as follows: 1, sad face; 2, straight face, and 3, smiley face (Week 0 = 11 July 2020).

Farmer sleep characteristics and methodological insights

In this study, we captured quantitative sleep data in a farming context for approximately 4 months. These results provide a valuable example of farmer sleep patterns and drivers, as well as proof of the methodology as a viable means to study farmers’ sleep. We found an average of 73% compliance in the sleep study, which we consider a satisfactory outcome considering the length of study. The participants were well inducted and supported during the study, with regular check-ins from the research team. There was a range in compliance among participants (36–94%) and a decline in compliance across the study period. Differences in compliance were not related to milking schedule, but may indicate a gradual participation fatigue, which highlights the need to actively engage participants throughout such studies. The research protocol asked participants to wear the sensor only in the evening, and this may have led to participants forgetting to wear it on some nights. It should also be noted that the study methodology could be considered a form of sleep behaviour intervention because of the active engagement with participants on the sleep topic, their access to daily data about their sleep patterns via the app and having the follow-up from the research team if their sleep was low. Therefore, this form of interaction may have influenced participant’s sleep patterns, and their sleep may have improved as a result of participation in the study.

In terms of sleep characteristics, participant’s ‘time in bed’ was 7.08 h, on average, and they were asleep for 5.75 h, on average. Sleep experts recommend between 7 and 9 h of sleep per night (Caldwell et al. 2019) and in terms of a single night <5.5 h is considered ‘impaired’ and 5.5–7 h can be considered ‘insufficient’. Therefore, many participants were likely impaired in terms of sleep over the study period. Peak workload periods in seasonal farming systems have been shown to be associated with less than optimum sleep among farmers (LaBrash et al. 2008; Hall et al. 2024). Consistent lack of sleep has been compared to alcohol impairment, particularly around speed and accuracy of movement and decision making (Williamson and Feyer 2000). Research in a range of agricultural contexts has highlighted the link between insufficient sleep and injury risk, particularly for those with less than 5 h sleep per night (Spengler et al. 2004; Lilley et al. 2012; Zhu et al. 2014). In a dairy farming context, fatigue-related impairment could have health and safety implications because of daily work with large animals, milking machinery, large machinery such as tractors, and other on-farm hazards (Summers et al. 2023). Insufficient sleep also negatively affects physical health factors such as cardiovascular health, diabetes, cancer, obesity and mental health (Medic et al. 2017). With participants being in bed only for around 7 h, on average, and having a bedtime of 22:35 hours, on average, this does not allow sufficient sleep opportunity before needing to wake for morning milking, especially for those who take longer to fall asleep. Improving farmer sleep requires that farmers both (a) understand the importance going to bed early enough to achieve sufficient sleep relative to the expected wake time, and (b) have workplaces that provide the opportunity for earlier bedtimes through end of work times that enable sufficient opportunity to eat, socialise, and wind down before bedtime.

Participants averaged 81% sleep efficiency (percentage of time spent asleep once in bed) in this study, where over 80% is considered good (Willoughby et al. 2024), and this was influenced most by milking schedule. However, people with insufficient total sleep time can still achieve good sleep efficiency. There was no difference in sleep efficiency when participants had days off work compared with working days; however, participants got 22 m more sleep on days off, possibly owing to a longer opportunity to sleep (e.g. later get-up times). Participants with a 3-in-2 milking schedule had more sleep across the study than did TAD participants, which appears linked to sleep start time (Table 3). Sleep efficiency was nominally (non-significantly) higher in 3-in-2 participants. The link between hours worked, and work finish time warrants further investigation via larger studies, particularly to understand the link between lower-stress workplaces and the ability of farmers to get better-quality sleep and better sleep efficiency. The study highlighted that hours in bed and bedtime end were moderately correlated with hours slept. Although this may seem logical, the data helped reinforce that there is an opportunity for dairy farmers to change morning milking times to be 30–60 min later, or when using flexible milking schedules, to enable people to have a later start time and therefore later wake time and more hours slept. Although not recorded in this study, work end time and work start time are likely to be simple indicators of the ‘sleep opportunity’ provided by farming workplaces, assuming that on working days, wake time is driven by start of work time. Workplace changes such as use of 3-in-2 milking can provide a larger window for farmer sleep opportunity.

In this study, more of the variation in bedtime start and bedtime end was explained by factors of interest (e.g. dry or milk phase, milking schedule) than by the differences in individual participants (Table 2). A value of the bedtime start and bedtime end metrics is that they could be manually estimated by farmers, without the need for intensive studies of sleep parameters. Farmers could then determine whether they were getting sufficient sleep opportunity. The sleep opportunity in farming workplaces could be incorporated into broader workplace quality assessments of productivity, wellbeing and safety (Hogan et al. 2024). In terms of other potential sleep metrics explored in this study (Table 2), mean heart rate was highly dependent on individual participants and therefore would not make a good comparative metric; however, time to lowest heart rate and heart rate variability (HRV) have been shown to be good sleep metrics and would warrant exploration in future studies (Stein and Pu 2012). Therefore, we suggest that studies of farmer sleep focus on metrics such as hours slept, sleep efficiency, bedtime start and bedtime end.

A key aspect of this research was to understand how such sleep studies with farmers could be used to determine differences related to different farm workplace structures or farm management approaches. Even with the small sample size used in the current study, the results showed that differences between workplaces, in this case farms with different milking schedules, can be determined. Greater statistical power would be possible with larger treatment groups, and our analysis suggests that sample sizes of 22 participants would enable identification of meaningful differences in shorter (e.g. 6 weeks) studies, with slightly fewer participants (20) needed if the study length was double. This means that increasing the duration of such studies is less powerful statistically than is increasing the number of engaged participants. Limiting the length of studies may also help address issues observed with declining participant compliance over time.

Wellbeing results and process

In general, the wellbeing survey approach showed a good level of engagement, driven by the face-to-face nature of the wellbeing survey and health assessment and that the researchers waited onsite for completed surveys. Comprehension of the questions may have been an issue; although some participants were able to conduct the wellbeing survey in a few minutes, others took over 10 min and needed to ask for clarification from the researchers. This reflects the diversity of people present on these dairy farms in terms of English comprehension, with many migrants being among the study participants.

The wellbeing survey enabled the identification of some differences among times of the dairy season, or among people working on farms with different milking schedules (Table 1). The most valuable questions in the survey for this purpose related to having energy to give friends and family (Q.4), being able to take time off during working days for private business (Q.15), and overall stress levels (Q.16). There was limited difference across many of the questions, which could be an accurate representation of participant wellbeing or could indicate that the question wording needs to be more specific to dairy farming and highlights the importance of the time of the season, i.e. surveying during busy calving or mating periods could return different wellbeing results from those in the dry period. A dairy farmer-focussed wellbeing assessment tool was subsequently co-developed with farmers and sought to address some of the lessons we learnt in the current study (Knook et al. 2024).

Results from the wellbeing fortnightly pulse showed good engagement (>50%) for approximately 18 weeks before engagement tailed off. There were potential differences between people working on flexible milking and TAD farms, particularly through the early part of the season (peak of calving). The SMS-based pulse method could be a good way of longitudinally assessing broad level wellbeing in farm workplace studies, however the small number of participants in our study precludes any conclusions being made about farmer wellbeing. Use of this pulse approach may suit initiatives with a large (e.g. 50+) number of participants but there is a limitation around the number of questions that can be asked, statistical power may be limited and participation likely to be an issue.

Overall, the surveys indicated some wellbeing differences among individuals across the season, and among those working in different farm workplaces. The dairy sector needs to continue a focus on improving workplace practices to minimise stress and enhance mental health, such as through maintaining manageable workloads through the busy calving period (Hansen 2022; Edwards et al. 2024; Knook et al. 2024).

Physical health metric results and process

In terms of the use of basic physical health metrics in farm system studies, we found the measures easy to capture through use of accredited health experts, and there was a high level of engagement with farmers, in part because they got immediate feedback and advice from the experts. The results showed that more farmers were using stimulants (caffeine and sugar-based drinks) during the early part of the season; this may have been related to the winter period or upcoming busy period. On average, results for glucose were within healthy ranges, however total cholesterol and BMI were higher than the suggested healthy range. In the context of assessing different milking strategies, the capture of physical health metrics added little value to the study. Such metrics may have more value in longitudinal (multi-year) studies or specific interventions where physical health was a key focus. Overall, scaling up of wellbeing and physical health measures would be needed to explore meaningful differences among different farm workplace experiences. Additionally, capturing greater detail about individual demographics and lifestyle factors would help explain inter-personal differences in sleep and physical health. However, there is a trade-off with researcher and health expert time required, particularly with the physical health measures.