This study used a pilot non-randomised controlled trial design. One organisation was designated as the intervention site where participants received a workplace HIIT intervention, and another separate organisation was designated as a no-intervention control group. The designation of workplaces to groups was based on the feasibility for the participating workplaces for exercise sessions to be conducted during the designated timeframe (e.g., April-July 2018).

A pilot trial is a precursor to a definitive randomised controlled trial (RCT) aiming to determine intervention effectiveness, and is undertaken to test key intervention components such as conducting outcome assessments and interventions in the settings in which they will be implemented, as well as to explore intervention fidelity and data variability (Abbott, 2014). The reporting of this study conforms with the CONSORT 2010 statement: extension for pilot and feasibility trials (Eldridge et al., 2016). Teesside University's School of Health and Social Care Research Governance and Ethics Sub-committee approved the study (study number: 036/18) and participants gave written informed consent. The protocol was prospectively registered on clinicaltrials.gov (identifier: NCT03467594).

Two office-based organisations in Northeast England were invited to take part in the study. These worksites were invited based on their involvement in work undertaken specifically to design the BE@Work intervention, described elsewhere (Burn et al., 2020). In the 2 months preceding the intervention, using email distribution lists, recruitment was facilitated by the individual organisations with posters placed around the workplace and presentations during staff meetings. To allow potential participants to try the programme activities, and in-line with feedback collected during the development stage of BE@Work (Burn et al., 2020), trial HIIT sessions were conducted in the fortnight prior to baseline testing at the intervention site (Burn et al., 2020).

The BE@Work intervention was a pilot trial, of which one of the key outcomes is gaining information on baseline to follow-up random variability (Craig et al., 2008). Therefore, the target sample size which was deemed feasible for this pilot trial was 60 participants (~30 participants per study site). This target sample size reflects the fact that only one researcher was scheduled to facilitate the HIIT sessions and baseline and post-intervention testing (NB), coupled with the limited resources available for the project. Inclusion criteria were adult (aged ≥18 years) employees of the participating organisations with no health conditions that precluded them from exercise and on no medication. Exclusion criteria were symptoms of, or known presence of cardiovascular or metabolic disease, injury or co-morbidity affecting the ability to undertake exercise, early family history of sudden cardiac death, and pregnancy. All participants underwent pre-exercise screening using the Physical Activity Readiness Questionnaire for Everyone (PAR-Q+) (Bredin et al., 2013). Across the intervention and control sites, 54 participants were recruited. Thirty participants (10 males, mean [standard deviation (SD)] age: 46 ± 9 years) were recruited from the intervention site and 24 participants (10 males, mean age 46 ± 12 years) from the control site.

The BE@Work intervention consisted of three group-based HIIT sessions each week for 8 weeks (24 sessions in total). Exercise sessions were conducted in a meeting room or on a grassed area outside the workplace and delivered by the first author (NB, exercise science post-graduate research student with previous experience of delivering group-based exercise sessions). Based on opinions expressed by employees in focus groups during intervention development (Burn et al., 2020), multiple HIIT sessions were scheduled across the working week (19 sessions each week), with participants asked to attend any three sessions. A choice of exercise modes, which were selected by participants during intervention development, were offered both within and between sessions, based on stair climbing, stepping and boxing (Burn et al., 2020). In line with previous investigations showing a beneficial effect of HIIT (Weston K. S. et al., 2014) the target heart rate for high-intensity exercise was set at ≥85% HRmax (Weston K. S. et al., 2014) and the intensity of the prescribed exercise was quantified prior to intervention implementation. Here (Burn, 2020), participants (n = 15; mean ± SD age: 39 ± 11 years; body mass index 24.9 ± 3.4 kg·m²) conducted single sessions of HIIT based on boxing, stair climbing and stepping, with mean peak heart rates indicating high intensity work, i.e., ≥85% HR_max (Weston K. S. et al., 2014) (mean peak ± SD %HR_max: box 85 ± 5 %HR_max, step 86 ± 7 %HR_max, stair climbing 85 ± 8%HR_max). As there were limited differences in heart rate between modes, this indicated the modes could be used interchangeably within the intervention without compromising the exercise intensity. Example drills are detailed in Table 1.

Based on HIIT protocols previously shown to elicit adaptations in cardiorespiratory fitness and cardiometabolic health (Hood et al., 2011; Little et al., 2011; Weston et al., 2016), the HIIT protocol consisted of four to seven 60-s high-intensity bouts, interspersed with 75-s of rest. Progression was provided by increasing the number of high-intensity bouts by one 60-s repetition every fortnight. Each session began with a 5-min warm-up and concluded with a 2-min cool down involving heart rate raising exercises relevant to the exercise session. Exercise sessions lasted ~15 min in the earlier weeks of the intervention to ≤ 22 min in the later weeks.

Participants' heart rates were recorded using a validated (Rider et al., 2019) second-to-second wrist-worn monitors (Polar A360, Polar Electro, Kempele, Finland). For each participant, age predicted HR_max was calculated using the Tanaka equation (208-0.7^*age in years) (Tanaka et al., 2001). If a participant exceeded this predicted value during a HIIT session, their HR_max was recalibrated to the higher observed value (Weston et al., 2004). Following each exercise session, individual participant heart rate files were downloaded into the Polar Flow software (Polar Electro, Kempele, Finland). The data used in the analysis was the highest 1-s value from each high-intensity bout, expressed as a percentage of the individual participant's HR_max, across each attended HIIT session. Session ratings of perceived exertion (RPE) were recorded after the 2 min cool down, using the CR-10 scale (Borg, 1998). This scale ranges from “nothing at all” (0) to “absolute maximum” (Reed et al., 2017).

Outcome data were collected at baseline and post-intervention from intervention and control participants (at least 72 h after and within 7 days of the final HIIT session for intervention participants). All outcomes were assessed in the participants' workplaces.

Predicted maximal oxygen uptake (VO_2max) was estimated using the Chester step test (Sykes and Roberts, 2004) (Cartwright Fitness, Huntington, UK). Heart rate was recorded throughout the test using wrist-worn monitors (Polar A360, Polar Electro, Kempele, Finland) and was used to predict VO_2max using the Chester step test calculator (Cartwright Fitness, Huntington, UK). As outlined in our registered trial protocol, we had intended to conduct laboratory-based maximal aerobic exercise tests with a subsample of participants. However, management in the participating organisations highlighted that this would not be feasible, as tests needed to be conducted during work hours and participants could not be released from work to attend the university laboratory.

Leg extensor muscle power was assessed using the Nottingham leg extensor power rig (Medical Engineering Unit, University of Nottingham, Nottingham, UK). Ten maximal effort leg extensions were performed on each leg, each separated by 30 seconds of rest (Hurst et al., 2018a). The highest value was taken as the participants' peak power output for analysis. At baseline data collection, to reduce habituation effects associated with the Nottingham leg extensor power rig (Hurst et al., 2018a) it was initially intended that participants would undertake three repeated trials on the Nottingham leg rig. However, due to the significant time commitment required (~15 min per trial) the participating organisations were reluctant to allow staff extra time away from work. As a result, of the participants who consented to undertake this assessment, in the intervention group 14 and 16 participants undertook the familiarisation procedures twice and three times respectively, and in the control group 2, 15 and 5 participants undertook familiarsation procedures once, twice and three times, respectively.

Handgrip strength was measured using a hydraulic hand dynamometer (12-0240; Baseline Evaluation Instruments, Fabrication Enterprises Ltd, New York, United States). Participants performed three maximal efforts on each hand, alternating between hands each time (Perna et al., 2016). The highest value was retained for analysis.

Blood pressure was measured using an OMRON M6 AC (HEM-7322-E) monitor (Omron Healthcare UK, Milton Keynes, UK), following the European Society of Hypertension guidelines (O'Brien et al., 2010). Non-fasting blood lipids [cholesterol, triglyceride and high-density lipoprotein cholesterol (HDL-cholesterol)] and glucose were assessed using finger prick blood samples and a Cholestech LDX analyser (Cholestech Corporation, Hayward, CA, USA).

Body mass and height were measured to the nearest 0.1 kg and 0.1 cm, respectively, using the Seca 799 electronic column scale, fitted with a Seca 224 stadiometer rod (Seca, Hamburg, Germany). Body mass index (BMI) was calculated using the equation (kg/m²) = body mass (kg)/ height (m)². Waist circumference was measured using a non-elastic Gulick tape measure with a tension device. Following normal expiration, the circumference of the abdomen was measured at the narrowest point between the lower costal border and the top of the iliac crest to the nearest 0.1 cm (Marfell-Jones et al., 2012).

Physical activity over the previous seven days was assessed using The International Physical Activity Questionnaire (IPAQ) short form (Craig et al., 2003). Scoring protocols outlined by the questionnaire developers (International Physical Activity Questionnaire, 2005) were followed to obtain a continuous physical activity score. The use of the IPAQ to assess habitual physical activity is a deviation from our registered trial protocol which stated that tri-axial accelerometers would be used. The latter were unavailable for the baseline testing period, so it was necessary to use an alternative assessment of physical activity.

Health-related quality of life (HR-QoL) was assessed using the Medical Outcomes 36-Item Short Form Health Survey 1.0 (SF-36) and scoring procedures outline by Ware et al. (2002). Possible scores range from 0 to 100 Arbitrary Units (AU) with higher scores indicating higher HR-QoL. Mental well-being was assessed using the Warwick Edinburgh Mental Well-being Scale (WEMWS) and scoring procedures outlined by Tennant et al. (2007). Scores range from 14 AU to 70 AU and higher scores indicate positive mental well-being. Perceived stress was measured using the Perceived Stress Scale (PSS) (Cohen et al., 1995). Scoring followed procedures outlined by Cohen et al. (1995) and possible scores range from 0 AU to 40 AU with lower scores indicating lower perceived stress.

During the post-intervention testing period, intervention participants were invited, via email, to attend a focus group. Focus groups provided an opportunity to elicit opinions via group discussion rather than individual reflection (Kitzinger, 1995) and also mirror the group-based nature of the BE@Work sessions. The focus groups were designed to explore participants' perceptions of the BE@Work intervention. Nine intervention participants (3 male, mean [± SD] age: 48 [± 7] years, session attendance 21 ± 2 sessions) volunteered to attend one of three focus groups. Focus groups consisted of two to four participants and were facilitated by one study author (KLW). Each focus group lasted between 48 and 59 min and were audio recorded and transcribed verbatim resulting in 93 pages of data (Arial, font size 12, 1.5 line spacing).

Heart rate data were used to explore intervention fidelity [i.e., whether the intervention was delivered as intended in a comparable manner to all participants (Dumas et al., 2001)]. The proportion of completed repetitions over the intervention in which the high-intensity exercise criterion was attained [≥85%HR_max (Weston K. S. et al., 2014)], was determined for each participant (Taylor et al., 2015). The median and interquartile range of these proportions was calculated. Then, to provide the correct overall between and within-subject variability (expressed as a standard deviation [SD]) in peak heart rate across the high-intensity bouts, a linear mixed model was applied, with exercise session specified as a fixed effect (SPSS v.25, Armonk, NY: IBM Corp). A similar process was used to analyse RPE data, which was included to further illustrate exercise intensity using a potentially more cost effective and scalable tool than heart rate monitors.

To inform the sample size for a future RCT, change variance was estimated for predicted VO_2max (Abbott, 2014). With the primary aim of piloting the data analysis that would be appropriate for a full RCT, we analysed our outcome data using an analysis of covariance (ANCOVA) model (SPSS v.25, Armonk, NY: IBM Corp) (Vickers, 2005). Plots of the model residuals vs. the predicted values were visually inspected to check they were correctly specified (uniform variance and normal distribution of residuals). The fixed effect was the group (intervention or control), and the dependent variable was the change from baseline to post-intervention. Model covariates were sex, age and baseline value of the outcome, to adjust for any imbalances between the intervention and control groups at baseline (Vickers and Altman, 2001). For the blood lipid and glucose measures, fasting status was included as an additional covariate. This variable was defined as the number of hours fasted post-intervention minus number of hours fasted at baseline. Data are expressed as mean ± SD, with uncertainty in all estimates expressed as 95% confidence intervals (CI). In keeping with guidance for the analysis of pilot trials (Lancaster et al., 2004), Cohens d effect sizes (Cohen, 1988) were derived and findings were interpreted as follows; trivial <0.2, small 0.2–0.3, moderate 0.4–0.8, and large >0.8. We did not undertake null hypothesis testing or report P-values (neither absolute or Bonferroni-corrected), in keeping with reporting guidance for exploratory pilot studies (Lancaster et al., 2004; Lee et al., 2014).

Focus group data were analysed using directed content analysis (Hsieh and Shannon, 2005) where the structure of the analysis is informed by previous research or theory (Elo and Kyngäs, 2008) using NVivo 12. As the purpose of the focus groups was to explore participants' perspectives of the BE@Work intervention, a pre-defined categorisation matrix was developed focusing on important intervention elements (Table 2). Following transcription and re-reading of the transcripts, two authors (NB and MG) coded the transcripts based on the pre-determined categorisation matrix. These authors examined the data independently within the pre-defined categories and inductively coded to further explore participant perceptions of each intervention element. During this inductive coding, NB and MG discussed coding decisions, and iteratively developed sub-categories within the pre-defined categories. MG was blinded to BE@Work programme aims during qualitative data analysis.

The SD of baseline to follow-up change was 6.6 mL·kg⁻¹·min⁻¹ (95% CI: 5.5 to 8.3 mL·kg⁻¹·min⁻¹) across both study groups. This change variance can be incorporated into a sample size estimation for a RCT. Assuming a clinically important mean change of 3.5 mL·kg⁻¹·min⁻¹ (Lee et al., 2011) and a statistical power of 90%, it was estimated using GPower that 76 participants would be required in each study group for a non-clustered RCT. If, in a future trial, researchers recruit participants in a specific clustered fashion, then this should be considered in any future sample size estimation, depending on the specific cluster sizes and the intraclass correlation between clusters.

Baseline values and effect statistics for the pilot between-group comparisons for physical fitness, cardiometabolic health and mental well-being outcomes are presented in Tables 4–6. After adjusting for age, sex and baseline values, compared with control the intervention effect on VO_2max was a moderate effect of 3.9 mL·kg⁻¹·min⁻¹ (95%CI: −0.2 to 8.1 mL·kg⁻¹·min⁻¹, d = 0.47). When considering all other outcome measures, we observed moderate effects for two domains of HR-QoL (pain; −6.7 AU; −18.4–5.0 AU, d = −0.67 and vitality 8.7 AU; 1.7–15.8 AU, d = 0.51), and perceived stress (-2.6 AU; −5.7 to 0.4 AU, d = −0.4). We observed small effects for one domain of HR-QoL (general health perceptions 4.8 AU; −0.3 to 9.7 AU, d = 0.35), mental well-being (2.5 AU; −0.8 to 5.7, d = 0.34), physical activity (−310 MET.min.week⁻¹: −1,640 to 1,019 MET.min.week⁻¹, d = −0.25), non-dominant leg extensor muscle power (15.9 Watts; 1.6–30.1 Watts, d = 0.2) and HDL-cholesterol (−0.1 mmol.L⁻¹; −0.2 to 0.04 mmol.L⁻¹, d = −0.2).

The following section describes the findings of the post-intervention focus groups, with illustrative participant quotes.

Notwithstanding any potential adaptations induced by exercise training, to promote long term adherence and compliance, it is essential that interventions are practical and feasible for participants. In the BE@Work trial, mean HIIT session attendance was 83%, which is similar to reported attendance in previous workplace HIIT interventions [range 71% (Eather et al., 2020) to 99% (Allison et al., 2017)]. While high session attendance could indicate sampling bias as discussed subsequently, to promote session attendance in BE@Work, a flexible approach was taken to exercise session attendance; such that 19 group HIIT sessions were conducted each week, with participants asked to attend any three sessions per week. Although this may have promoted attendance, the scalability of providing this volume of exercise sessions across a week is questionable. Indeed, the cost of employing exercise facilitators may inhibit organisations from being able to provide such flexible services to employees.

While useful, session attendance does not provide information regarding the extent to which the intervention was delivered as intended in a comparable manner to all participants; otherwise known as the intervention fidelity (Taylor et al., 2015). We used exercise intensity as an indicator of intervention fidelity. Across the intervention, the median proportion (interquartile range) of repetitions where the high-intensity exercise criterion (≥85% of HR_max) was attained was 70% (53–84%). Previous evaluations of intervention fidelity in HIIT trials with older adults (Hurst et al., 2018b) and adolescents (Taylor et al., 2015) have categorised high-intensity exercise criterion attainment proportions of between 58 and 62% of repetitions as “moderate” intervention fidelity. With this in mind and using proposed intervention fidelity thresholds of <50% “low”; 50–70% “moderate” and >70% “high” (Hurst et al., 2018b), the intervention implementation in BE@Work could be described as moderate to high. Across the high-intensity bouts performed across BE@Work, the mean peak heart rate was 87% HR_max. By adding and subtracting the within-subject variability to the mean peak heart rate (5.5 percentage points), this shows the intensity of the intervention ranged between 81.5% HR_max through to 92.5% HR_max in individual participants across the intervention. Furthermore, the relatively small between-subjects SD (3.7 percentage points of the mean peak heart rate) indicates that the intensity was similar between the participants (83.3% HR_max to 90.7% HR_max). These findings indicate the quality of the delivery of the BE@Work intervention, and its receipt and enactment by the participants (Taylor et al., 2015). The mean peak heart rate across BE@Work was however, lower than the mean peak heart rate reported in previous community [95 ±3% HR_max (Reljic et al., 2018)] and workplace [91 ±3% HR_max (Shepherd et al., 2015)] HIIT trials. Although between-study differences are likely multifaceted, one possible explanation could be differences in exercise modalities (Buchheit and Laursen, 2013). Previous trials (Shepherd et al., 2015; Reljic et al., 2018) exclusively used cycle ergometers as the HIIT mode which can be manipulated to control exercise intensity. The BE@Work intervention used a range of exercise modes, which may explain the different exercise intensity. In regard to rating of perceived exertion, the mean session RPE was 6 AU across the intervention, which is between the qualitative descriptors of “hard” and “very hard.” This is similar to that reported in a previous workplace HIIT trial (mean RPE 16 AU, “very hard” using the 6 to 20 RPE scale) (Allison et al., 2017). Similar to the heart rate data, the relatively small within and between-subjects SDs (1.3 AU and 1.4 AU, respectively) indicate that perceived exertion of the HIIT sessions was relatively uniform both across the intervention and between participants. The within and between subject variability demonstrate that RPE ranged from 4.7 AU (i.e., ~hard) to 7.3 AU (i.e., very hard) across the intervention, and between 4.6 AU and 7.4 AU between participants.

Post-intervention, we observed preliminary evidence of a moderate effect on VO_2max (3.9 mL·kg⁻¹·min⁻¹; −0.2 to 8.1 mL·kg⁻¹·min⁻¹, d = 0.47) in intervention participants compared to controls. This improvement is similar to the 2.8 to 3.5 mL·kg⁻¹·min⁻¹ increases in VO_2max reported following previous workplace HIIT trials (Shepherd et al., 2015; Allison et al., 2017; Cuddy et al., 2019). Although the effect we observed is higher than that reported in a previous meta-analysis of workplace exercise interventions (2.7 mL·kg⁻¹·min⁻¹) (Burn et al., 2019), it is lower than the increase in VO_2max reported following largely laboratory-based HIIT trials (5.5 mL·kg⁻¹·min⁻¹) (Milanovic et al., 2015). This could be because the magnitude of intervention effects are attenuated and more variable under less tightly controlled conditions outside of the laboratory (Courneya, 2010). Our somewhat larger mean treatment effect in this small pilot trial could also be due to sampling error, a phenomenon that is sometimes referred to as the “winners curse” (Tugwell and Knottnerus, 2018). Therefore, we also sought to use the SD of baseline to follow-up change in VO_2max (6.6 mL·kg⁻¹·min⁻¹) to estimate a sample size for a full RCT. Assuming a clinically important mean change of 3.5 mL·kg⁻¹·min⁻¹ (Lee et al., 2011), 76 participants would be required in each study group to detect this mean change. Of the previous workplace HIIT trials, sample sizes have ranged from 13 (Metcalfe et al., 2020) to 42 (Shepherd et al., 2015) participants in the HIIT group. This may highlight the difficulty in acquiring funding, conducting, or recruiting participants in the workplace to conduct a fully powered trial. It is important to note however, that our findings may have been influenced by sampling bias [e.g., when a sample does not accurately reflect the true target population (Lavrakas, 2008)]. In the case of the BE@Work intervention, it could be that individuals who volunteer for workplace HIIT studies differ from the working age population. Indeed, when comparing our participants' baseline fitness levels to normative reference values for apparently healthy middle-aged British adults (Ingle et al., 2020), the baseline VO_2max for males in our sample (42.7 mL·kg⁻¹·min⁻¹) was higher than the British median (36.9 mL·kg⁻¹·min⁻¹). Nonetheless, baseline VO_2max for women in our sample (33.5 mL·kg⁻¹·min⁻¹) was lower than the British median (36.5mL·kg⁻¹·min⁻¹).

It is concerning that we observed a moderate increase (indicated by a decrease in HR-QoL score) in pain (−6.7 AU; −18.4 to 5.0 AU, d = −0.67. An increase in perceived pain has been reported following one previous community-based HIIT trial in inactive and overweight adults (Lunt et al., 2014). Given that critiques of HIIT have highlighted that perceived negative experiences, such as perceived pain during exercise may impact participation in the activity (Biddle and Batterham, 2015), this area is worthy of further investigation. Interestingly, while increases in pain were not explicitly reported by our focus group participants, they did report local muscular fatigue during HIIT. As the SF-36 does not differentiate between acute or chronic pain or type of pain, it could be that sensations of exertion influenced their perception of pain when completing the SF-36, however this requires further investigation.

We observed moderate improvements in vitality which is one domain of the SF-36 HR-QoL questionnaire (8.7 AU; 1.7 to 15.8 AU, d = 0.51). Vitality is the balance an individual feels between energy and fatigue (Ware et al., 2002). Both workplace HIIT specifically (Shepherd et al., 2015) and exercise training more generally, have been shown to improve markers of vitality (Puetz et al., 2006). However, because social interaction during exercise may improve vitality more than exercising alone (McNeil et al., 1991), while the increase in vitality observed in our study could be attributed to HIIT, it could also be attributed to the group-based nature of the intervention. We also observed a moderate decrease in perceived stress (−2.6 AU; −5.7 to 0.4 AU, d = −0.4) post-intervention. While perceived stress has not been assessed in previous workplace HIIT trials, improvements in this outcome could be particularly meaningful for employers because work-stress is associated with both absenteeism (Lauzier et al., 2017) and productivity (Hoboubi et al., 2017). Of the other well-being outcomes assessed, we observed a small effect on one domain of HR-QoL assessed using the SF-36 (general health perceptions: 4.8 AU; −0.3 to 9.7 AU, d = 0.35). This finding is supported by previous community-based HIIT trials, where improvements in quality of life have also been reported (Lunt et al., 2014; Shepherd et al., 2015). Furthermore, we observed small improvements in mental well-being assessed via the WEMWS (2.5 AU;−0.8 to 5.7, d= 0.34). This finding is supported by a previous systematic review indicating that workplace exercise interventions can improve well-being (Abdin et al., 2018).

Compared with controls, we observed a small reduction in self-reported physical activity (-310 MET.min.week⁻¹: −1,640 to 1,019 MET.min.week⁻¹, d = −0.25) in the intervention group. While within-group comparisons showed increases in physical activity in both groups, the increase was more pronounced in the control group and therefore the intervention effect was a reduction in physical activity. Given that previous HIIT interventions have observed increases in physical activity in inactive adults (Stavrinou et al., 2019) and the validity and reliability of self-report measures of physical activity have been questioned (Prince et al., 2008), future studies could seek to use device-based measures of physical activity to further explore the effect of workplace HIIT on habitual physical activity.

We observed evidence of a small increase in leg extensor muscle power (non-dominant leg extensor muscle power; 15.9 Watts: 1.6–30.1 Watts, d = 0.2), compared with controls. However, within-group comparisons showed decreases in muscle power in both groups, with the reduction more pronounced in the control group. The mechanism for a reduction in muscle power in both groups is unclear but could be related to habituation effects associated with the Nottingham leg extensor power rig (Hurst et al., 2018a). While contemporary guidance does not require familiarisation protocols to be conducted post-intervention (Hurst et al., 2018b), it is possible that these findings are a result of insufficient familiarisation at baseline as previously described or participants may have benefitted from further familiarisation, post-intervention. In comparison to previous work, while improvements in muscular power measured via standing long jump have been observed in one previous workplace HIIT study (Eather et al., 2020), a meta-analysis (Weston, M. et al., 2014) and HIIT trials in adults (Greenlee et al., 2017; Eather et al., 2019) have not observed changes in muscular power. Despite these conflicting findings, given that muscular power is an important determinant of effective physical functioning (Reid and Fielding, 2012), the effect of workplace HIIT on muscular power remains an interesting avenue for future research.

Lastly, we observed a small reduction in HDL-cholesterol, post-intervention (−0.1 mmol.L⁻¹; −0.2 to 0.04 mmol.L⁻¹, d = −0.2). Although a previous meta-analysis suggests aerobic exercise can increase HDL-cholesterol (Lin et al., 2015), the effect of HIIT on HDL-cholesterol in healthy populations is unclear and requires further investigation (Martland et al., 2020). Furthermore, workplace HIIT trials to date have reported no significant changes in HDL-cholesterol post-intervention (Shepherd et al., 2015; Allison et al., 2017). As HDL-cholesterol has a protective effect on coronary heart disease mortality (Cooney et al., 2009), this finding may be concerning and should be explored in future investigations. It is acknowledged that the post-prandial response for both glucose and triglycerides can be affected by a range of factors such as the content and timing of the previous meal and previous physical activity (Lopez-Miranda et al., 2007). While fasting status was included as a co-variate in the analysis, it is acknowledged that it was not possible to acutely control and replicate all of the factors that may have affected non-fasting blood lipid and glucose concentrations, and therefore this could be considered a limitation of this data.

We highlight the fact that we report here the results of an exploratory pilot study. As emphasised by multiple authors (Lee et al., 2014; Cipriani et al., 2015; Li et al., 2017) formal null hypothesis testing and the interpreting findings base on P-values is not consistent with our study approach. Rather, we have reported control-adjusted mean changes and associated 95% confidence intervals and effect sizes to explore the sensitivity of the outcomes. Therefore, it is important to quantify more precisely the mean treatment effects on these outcomes in a future definitive RCT. This future trial should be adequately powered and we have informed this issue via the reporting of change variance in predicted VO_2max. Our findings also indicate that some of these outcomes can be feasibly measured in the workplace.

The post-intervention focus groups provided a unique insight into participant perspectives of the acceptability of BE@Work. Here, participants generally reported that intervention logistics (e.g., frequency, length and location of HIIT sessions, exercise modes, group-based nature of the intervention and exercise facilitator) were satisfactory. This may be unsurprising as focus group participants attended an average of 21 out of 24 possible HIIT sessions. High session attendance could indicate that the participants were more likely to be satisfied with the intervention, as highlighted in a previous workplace HIIT intervention (Kinnafick et al., 2018), or it could be evidence of sampling bias. Regardless of views on intervention logistics, our participants considered flexibility in session attendance and exercise mode selection as instrumental in facilitating session attendance and enjoyment. While the scalability of providing a large volume of facilitated sessions across a week is questionable, computer assisted technology has shown promise for the delivery of workplace HIIT, thus eliminating the need for facilitated HIIT sessions (Metcalfe et al., 2020). While computer assisted HIIT may be regarded positively by some individuals, our participants reported that they enjoyed group-based facilitated HIIT, which highlights the need to cater to a wide range of exercise preferences in the workplace.

The BE@Work exercise modes (boxing, stair climbing and stair stepping) were selected by participants in focus groups conducted during intervention development (Burn et al., 2020). Post-intervention focus group participants regarded these exercise modes as acceptable and feasible, which highlights the importance of co-developing interventions with participants who are representative of potential study participants, in the settings in which interventions are to be delivered. Our participants highlighted the importance of having a choice in a variety of exercise modes within and between HIIT sessions, which has also been shown to enhance exercise adherence previously (Morgan et al., 2016). It is therefore surprising that most workplace HIIT interventions to date have used single exercise modalities throughout the intervention such as cycle ergometers (Shepherd et al., 2015; Metcalfe et al., 2020) or stair climbing (Allison et al., 2017). Our findings, along with recent work (Eather et al., 2020) highlight the utility of providing multi-modal HIIT in a workplace setting, which can cater to a wider range of exercise preferences.

When asked about their experiences during HIIT sessions, participants reported breathlessness, increased heart rate, local muscular fatigue and sweating; which were generally viewed as unpleasant or tiring. Previous laboratory-based work has highlighted that high-intensity exercise can result in unpleasant affective responses (Ekkekakis, 2003), which may in turn affect future exercise behaviour (Rhodes and Kates, 2015). Nonetheless, our participants also reported feeling alert during HIIT and energised afterwards. This positive post-exercise experience has been reported in previous laboratory-based work (Stork et al., 2020). In addition, social support and a positive group experience where individuals experience the same challenging situation has been reported to contribute to adherence in exercise interventions (Harden et al., 2015) and workplace HIIT interventions (Kinnafick et al., 2018). As our participants reported that the group-based nature of the intervention facilitated camaraderie and relationships with colleagues, it is possible that the group-based nature of the intervention engendered positive exercise experiences.

In line with previous work in laboratory-based (Stork et al., 2020) and fitness centre HIIT trials (Burn and Niven, 2018) our participants regarded the rest breaks incorporated into the HIIT protocol prescribed during BE@Work as particularly important. Participants reported that rest breaks made the intensity of HIIT intervals achievable and maintainable throughout the session. In addition to rest breaks, the HIIT prescribed in BE@Work was relative in nature, meaning that the intensity was individually prescribed to each participant based on their individual maximal heart rate. The work required to achieve a given percentage (e.g., >85%) of this heart rate depends on the fitness level of the participant, meaning that HIIT may be enacted differently in each person. While critics have postulated that HIIT will be perceived as too difficult, leading to poor adoption of the activity (Hardcastle et al., 2014; Biddle and Batterham, 2015), our participants highlighted that because they were not required to maintain pace with another participant or instructor, they could self-monitor the exercise intensity which made HIIT manageable and achievable.

Post-intervention, we observed evidence of improvements in markers of physical fitness (e.g., VO_2max) and mental well-being (e.g., HR-QoL, perceived stress). Our focus group participants supported this finding, by reporting that they felt fitter and less stressed following the intervention. Given that participants of a previous workplace HIIT trial have also reported perceived changes in psycho-social outcomes post-intervention (Metcalfe et al., 2020), this suggests that a future trial could be designed to explore the effect of workplace HIIT with a psycho-social or mental health related primary outcome.

Although our study has produced some promising findings, it is not without limitations. Most notably, this was a pilot trial with exercise sessions delivered in a small sample of employees from one workplace. As such, the findings should be interpreted with caution. Despite high session attendance and promising preliminary implementation data, a range of generalisability biases may need to be considered ahead of a fully powered trial (Beets et al., 2020). Mostly notably, the scalability of delivering 19 HIIT sessions each week is clearly questionable and may lead to implementation support bias (Beets et al., 2020). The need for facilitated HIIT sessions could be overcome by the use of technology-based interventions as previously described (Metcalfe et al., 2020). Similarly, given that BE@Work was facilitated by a researcher experienced in the delivery of group-based exercise, the intervention may have been subject to intervention delivery agent bias. Future work could seek to train employees to deliver HIIT sessions, such as workplace wellness champions involved in previous workplace health promotion interventions (Ellis et al., 2021). Furthermore, participants in this study were healthy middle-aged adults working in office-based roles; therefore the findings cannot be extended to more diverse populations or workplaces. This could lead to target audience and setting bias (Beets et al., 2020). Critiques of HIIT commonly suggest that it will have limited appeal in unfit or inactive populations (Biddle and Batterham, 2015). Based on the baseline fitness of participants in the present study (VO_2max 33.5 mL·kg⁻¹·min⁻¹ in women and 42.7 mL·kg⁻¹·min⁻¹ in men), the effectiveness of HIIT in very deconditioned individuals cannot be addressed from the results of this study per se. However, the relatively large standard deviation of the baseline VO_2max (7.5 mL·kg⁻¹·min⁻¹) indicates that participants with a range of fitness levels were recruited to this study. Future studies could seek to implement workplace HIIT in a diverse range of workplace settings (e.g., manual labour settings) with a wider range of participants.

Randomisation to trial arms was not pragmatically possible in this study, because participating organisations were unable to accommodate exercise sessions at short notice. While this may have resulted in selection bias, allocation to groups was based on place of work rather than participant selection and baseline outcome values were included as a covariate in the statistical analysis, which may have partly mitigated this threat to validity. Another potential limitation relates to the use of a submaximal prediction of VO_2max. While a gold standard VO_2max assessment was not logistically possible as previously described, it is acknowledged that the validity of predictive VO_2max assessments have been questioned (Grant et al., 1999). The use of non-fasted finger prick blood samples could also be considered a limitation. Although non-fasted blood samples appear to better predict cardiovascular disease risk (Nordestgaard et al., 2016) it is acknowledged that the post-prandial response for both glucose and triglycerides can be affected by the content and timing of the previous meal (Lopez-Miranda et al., 2007). Although, efforts were made to ensure the content and timing of meals were the same at both data collection time-points, and fasting status was included as a co-variate in the analysis, it is acknowledged that it was not possible to acutely control and replicate all of the factors that may have affected non-fasting blood lipid and glucose concentrations.

Upon trial registration, we had intended to assess acute mood and enjoyment responses in the intervention participants at fortnightly intervals. However, due to difficulties in obtaining responses from participants to acute data collection attempts, we could not robustly assess acute psychological responses to workplace HIIT, and therefore we have elected not to present this data. Nevertheless, this remains an interesting avenue for future investigation and should be considered when designing workplace HIIT trials.