Ethical approval to undertake this project was given by the University of Dundee School of Medicine Research Ethics Committee (20/55) prior to participant recruitment and data collection.

Convenience sampling occurred via social media, word of mouth, and email dispersal to recruit individuals throughout the United Kingdom. Involvement was completely voluntary, granting participants permission to withdraw themselves from the study at any point. Individuals’ participation in the study was excluded if individuals were outside the age range of 18 to 60 years, had long-term health conditions, took prescribed medication, were receiving medical treatment, or lived outside the United Kingdom. A Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist was completed during the preparation of this paper.

Participants were required to fill out a consent form and a preanonymized Personal Health Questionnaire (PHQ). Upon completion, they downloaded a third-party app, Health Auto Export to CSV. Through this app, the participants were able to sync data collected on their watch from the health app on their iPhone and export the data in the form of an Excel (Microsoft) file. The Excel file was then anonymized by the participant by renaming the file with the corresponding ID number found on their PHQ. All 3 documents were then uploaded onto a secure One Drive (Microsoft) folder. Each participant gave a month’s worth of data in hourly increments from a standardized time period (August 1-31, 2022). Choosing a time period in the past, before participants were recruited, allowed for the observation of free-living individuals unperturbed by observer bias. The study had 20 participants; this number is based on a previous study we had carried out [11].

Active energy (AE; kJ), average heart rate (HR [avg]; count per minute), step count (SC; count), and walking and running distance (WRD; km) were selected to investigate their effect on completing an SH. These 4 variables represent a range of physiological measurements, relating to the effect of activity (SC and WRD) and normal body measurements (HR and AE). HR (avg) was chosen as it was the only HR measurement that gave an indication of HR throughout the whole hour.

This study only considered Apple Watch–derived measurements. The iPhone health app automatically measures SC and WRD if an iPhone is present on a person’s body. Therefore, the data collected included iPhone measurements, as well as Apple Watch measurements. Watch measurements were distinguished from iPhone measurements as HR is continually recorded on a watch, but the iPhone has no technology to measure HR. Any hourly measurements that had absent HR data were removed from the data set. A total of 1901 (17%) data points were excluded from the study due to this.

Statistical analyses were performed and related graphs were created on R Studio (R Core Team) or Prism (GraphPad). Maximum and minimum adequate models were fitted as a generalized linear model with mixed effects (GLMM). The fixed effect model from these GLMMs report estimate (SE), followed by a P value and significance marking. The random effects model reports variance with SD, and a stepwise backward regression was performed to create the minimum adequate model from the maximal model. ANOVA test using type II Wald χ² test was used to confirm the findings of the GLMM [23].

Means and medians of the significant variables were calculated and graphically visualized on either bar charts or box plots and the differences were statistically tested for significance using Student t tests (mean) and Wilcoxon summed ranked tests (median) [24].

To investigate and understand the threshold at which the variables change the SH metric from 0 to 1, a graphical visualization of the fitted SH was created using marginal estimated means calculated from the data set for each variable to investigate them as separate entities. The values obtained were then applied back into the data set to see how accurate the estimated values were.

Twenty individuals participated in this study, with a total of 9236 hourly observations between participants. A total of 7301 recorded SHs and 1936 nonrecorded SHs were observed. Cohort demographics are shown in Table 1.

Tables 2 and 3 show the output from the most maximum and minimum adequate model fitted as a GLMM. As shown in Table 2, the maximum model was built considering all the fixed measurements that were chosen for further investigation. AE, SC, WRD, and HR (avg) were the measurements recorded on the watch while gender and age were the factors obtained from the PHQ. The minimum model removed any insignificant measurements from the SH. Table 2 shows that both AE (0.021, SE 0.002) and SC (0.014, SE 0.004) produced statistically significant coefficient estimates (P<.05). However, WRD (7.837, SE 5.46) did not have a significant P value (P=.15). Table 3 reports the random effects of the model and the variance shows the spread of the data between each participant. The value of variance increased in the minimum adequate model, from 0.194 to 0.245.

The results from Tables 2 and 3 were analyzed further by ANOVA using type II Wald χ² test, AE (W₁=169.549; P<.001) and SC (W₁=11.639; P=.001) and had a statistically significant effect on the recording of an SH. WRD had no statistical significance on the recording in the SH (W₁=2.026; P=.15).

Using the coefficient estimates from Table 2, odds ratios were calculated using logit transformations. A logit transformation was applied to the coefficient estimates as the SH metric was restricted to a finite interval (0 or 1) [25]. It is estimated that the Apple Watch recorded one SH 7% of the time when no added factors such as AE or SC were added within the participant cohort. This was calculated using the “intercept” value which is a value representative of the natural activity levels of the most average participant in the study before the addition of SC or AE. In this case, it was a female participant, aged 18-29 years. By increasing the AE value by 1 kJ, the likelihood of an SH being recorded is increased by 2% and for every additional 1 step taken, the likelihood of an SH being recorded increased by 1%.

Figure 1 shows the mean values of AE, SC, and WRD when an SH was not recorded (0) and when an SH was recorded (1). Student t tests were computed to see if there was any significant difference between the mean of recorded SH (1 SH) and non-recorded SH (0 SH) for each measurement. For every measurement, the mean value was statistically different between 0 SH and 1 SH. For AE, the mean value when there was no SH recorded was 39.63 kJ and 204.44 kJ when an SH was recorded (t_24.03=15.05; P<.001). For SC, the mean value when there was no SH recorded was 45.62 steps and 803.65 steps when an SH was recorded (t_19.42=14.89; P<.001), and for WRD the mean value for WRD when an SH was not recorded was 0.03 km and 0.62 when an SH was recorded (t_19.35=14.02; P<.001).

Figure 2 shows the median values of AE, SC, and WRD when an SH was not recorded (0) and when an SH was recorded (1) and confirmed by Wilcoxon rank sum tests to determine the statistical significance of the medians. Again, for all measurements, the median value recorded for 1 SH was significantly higher than those recorded for 0 SH. For AE, the median value when there was an SH was not recorded was 18.96 kJ and 141.47 kJ an SH was recorded (z=–4.86; P<.001). For SC, the median value when there was no SH recorded was 15.00 steps and 521.59 steps when an SH was recorded (z=–5.40; P<.001) and for WRD, the median value when there was no SH recorded was 0.01 km and 0.39 km when an SH was recorded (z=–5.42; P<.001).

Figure 3 shows fitted models for the SH, created using estimated marginal means and analyzing each measurement as a separate entity. For each model, values were read at the point where the graph begins to curve sharply, which estimates a threshold above which it becomes much more likely that an SH will be recorded. The turning point where each graph turns sharply indicates where the threshold of values estimates the values are large enough to deem an SH has been sufficiently met. For AE, the graph turns at 0.999 giving an estimated reading of ~100 kJ. For SC, the graph turns at 0.961 to give an estimated value of 97.5 steps and for WRD, the graph turns at 0.999 to estimate a value of 0.468 km. These estimated values were then applied back into the original data set of gathered data to see their accuracy.

A total of 76% (n=5330) of the AE-recorded SH had a value larger than 100 kJ (Figure 3A). It was also shown that 4% (n=7) of the observations recorded as 0 SH had a value larger than 100 kJ. The fitted model for SC (Figure 3B) begins to turn at an earlier point than both the models of AE and WRD, turning at 0.961 rather than 0.999. A total of 92% (n=6682) of the SC-recorded SH had a value larger than 97.5 steps but it also showed that 10% (n=201) of the observations that were recorded as 0 SH had a value larger than 97.5. Only 40% (n=2920) of the WRD-recorded SH observed had a value larger than 0.486 km (Figure 3C). However, it had the least observed 0-recorded SH at 0.05% (n=1).

AE and SC were found to be significant measurements underpinning the recording of an SH. SC and AE are the most studied measurements across both brand and device types [26]. Apple watch–measured SC has been validated against hip-worn accelerometers in a laboratory-based setting [21] and free-living setting [27]. It is extremely accurate within the laboratory [21] but has been shown to overestimate SC in a free-living setting compared with hip-worn accelerometer [27]. Our finding of the 97.5 SC threshold above which recording of an SH becomes much more likely is consistent with previous evidence showing that 51-100 steps in 1 hour created an SH 95% (n=87) of the time [11]. The AE measurement has been investigated across multiple models of the Apple Watch. Series 6 Apple Watch underestimated the value of energy expenditure at multiple different physical activity intensities including walking (LPA) [17], while the Series 4 Apple Watch also had poor accuracy across multiple exercise intensities [18]. For both SC and AE, there is yet to be a study to compare the values between free-living and laboratory-based environments. It remains inconclusive from the results of our study and from previously documented evidence regarding AE and SC [26,28,29] if there is one measurement that is more effective or has more influence over the other regarding recording an SH. WRD did not prove to be a significant measurement used to record an SH; however, the WRD measurement was included in further analysis as it was the only other measurement that was expressed in the minimum adequate model of the GLMM.

In all three analyzed measurements (AE, SC, and WRD) it was shown that there was a significant difference between both the mean and median values that were needed to create a 1 SH recording and 0 SH recording. From the fitted models, SC proved to be the most accurate within our data set, showing 92% (n=6682) of the recorded SH had a value higher than 97.5 steps. For AE, 76% (n=5330) of the observed 1 SH recorded having a value larger than 100 kJ. The SC value of 97.5 steps is likely to be a feasible and realistic goal to achieve within an hour in many workplaces and other modern lifestyle settings. 100 kJ of consumed energy is 23.90 kcal (calories) [30]. It is estimated that it would take approximately 8-10 minutes of slow-to-moderate paced walking (1.9-3.0 mph) to burn 23.90 kcal [21] with an average male or female body weight (~80 kg and 70 kg, respectively). Given the SH measurement is intended to encourage 1 minute of standing every hour, it is possible that our value of 100 kJ for a change in recording is an overestimate; alternatively, our findings highlight that the SH is not measured by one entity alone, and there is more than one factor influencing the recording of an SH.

Convenience sampling was the most effective method for recruitment due to time constraints. Participants who already owned activity trackers are unlikely to be representative of the whole community. A larger participant cohort more representative of the community would have provided a greater range of natural activity levels, as many of the participants were naturally active. We did not include anybody with CVD, diabetes, or at-risk groups, so the external validity of our findings will need to be established in those groups. Although neither gender, age, nor BMI was found to play a role in affecting the outcome of an SH being recorded, a large majority of our participant cohort was younger females.

It is currently being investigated whether the detrimental health effects of long periods of SB may be reduced by switching SB for short, regular periods of LPA. This study quantifies the SH metric present within Apple Watch technology. The aim of this metric is that it records standing with movement, for 1 minute in any given hour. The findings of this study reveal the pivotal role of the AE and SC measurements present on the Apple watches, which underlie the algorithm behind the SH recording; however, our findings also suggest that a recording of an SH is influenced by more than one factor. The research from this study finds 100 kJ of energy or 97.5 steps define a threshold for activity above which recording of 1 SH becomes much more likely. Our findings demonstrate that any activity upwards of LPA will trigger an SH to record. Irrespective of the internal validity of the SH metric, the SH is representative of LPA and should, therefore, encourage individuals by any means (including notifications) to promote a reduction in SB by engaging with LPA or greater levels of activity.