Survey Data and Study Sample

The data for the present study were obtained from the NHANES 2003–2004 and 2005–2006. The NHANES provides cross-sectional data for a national representative sample of the US civilian non-institutionalized population selected by a complex multistage probability sampling scheme. The survey measured broad areas of health-related outcomes through household interviews and physical examinations at the mobile examination center (MEC). The original survey protocols were approved by the National Center for Health Statistics (NCHS) Research Ethics Review board, and informed consent was obtained from all participants.

Among participants who visited the MEC, all ambulatory participants (≥ 6 years) were eligible for accelerometer measures. The Actigraph accelerometer (model 7164) was used to obtain objective measures of PA. Participants were instructed to wear the accelerometer on their right hip during waking hours across 7 consecutive days, with the exception of when engaging in any water-based activities (e.g., bathing or swimming). Activity counts that represent integrated acceleration information of ambulatory movements were recorded in minute-by-minute intervals. A detailed description of the NHANES data can be found at www.cdc.gov/nchs/nhanes.htm.

A total of 9,151 adults (≥18 years) provided accelerometer data. After excluding those with missing values for one or more of the cardiovascular risk factors [with an exception of fasting sub-component, triglyceride (TG)] or covariates examined in this study, or with insufficient valid accelerometer data (refer to later section), the final sample consisted of 5,917 adults (2,941 male and 2,976 female) which included a fasting sub-sample of 2,663 who provided TG measure. The final sample was demographically different from the excluded sample (n = 3,234) by gender [unweighted x² (1) = 26.97, p <.001] and race/ethnicity [unweighted x² (3) = 25.70, p <.001] but similar with respect to family income levels [unweighted x² (3) = 0.84, p = .840].

Accelerometer-based SB Measures

An automated SAS macro provided by NCHS was used for quality control and to identify non-wear times. Non-wear time is defined as intervals of at least 60 minutes of 0 activity counts (i.e., no movement), with allowance for up to 2 consecutive minutes of activity counts between 0 and 100 [22]. After removing the non-wear times from accelerometer data, the following SB parameters were extracted for each measurement day using the traditional algorithms as previously described in the literature [15,23]:

1. sedentary time—a minute where activity counts are <100 cpm;
2. sedentary break—a transition point from a sedentary (<100 cpm) to active phase (≥100 cpm);
3. sedentary bout—a duration of continuous sedentary time (i.e., sedentary event);
4. mean intensity—the average activity counts within each sedentary bout.

In addition, because we aimed to explore accrued patterns of sedentary time and breaks, particularly focusing on sedentary bout durations, all SB measures were extracted within the respective bout durations of 1-min, 2–4 min, 5–9 min, 10–14 min, 15–19 min, 20–24 min, 25–29 min, and ≥30-min in addition to total accrued quantities.

All SB measures obtained for each measurement day were then averaged across only valid days (i.e., 10 or more hours of wear time), to represent the average measures of SB per day. Participants with 4 or more valid days are included in the analysis. In addition, because sedentary time or breaks are influenced by accelerometer wear times [23], the least-square adjustment for wear times was made for all SB measures using the residuals obtained from linear regression models where SB measures were regressed on wear times [12,24].

Cardiovascular Risk Factors and Other Covariates

The primary purpose of this study was to evaluate the impact of different measures of sedentary time on health outcomes after taking sedentary bout durations into account. Three measures of cardiovascular risk factors were examined in this study [TG (mg/dL), waist circumference (WC) (cm), and high-density lipoprotein cholesterol (HDL-C) (mg/dL)] since previous studies with NHANES data have reported significant associations with SB measures [12,19]. In addition, body mass index (BMI) (kg/m²) was also obtained as there is still controversy in regards to its relationship with SB measures in the NHANES data [14].

The average time spent in moderate-to-vigorous-intensity PA (MVPA), based on a modified 10-min bout condition (i.e., a minimum of 8 out of 10 consecutive minutes of MVPA), across valid days was obtained from accelerometer data using the threshold of ≥2020 cpm [22]. Demographic characteristics of the participants including age (years), sex, race/ethnicity (Non-Hispanic White, Non-Hispanic Black, Mexican American, and Other Hispanic/other races), and family income (<$15k, $15k-34.9k, $35k-64.9k, ≥$65k) were also used as covariates in the statistical model (See Table 1 for descriptive statistics).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive Statistics of Outcome Variables across Demographic Characteristics among US Adults.

https://doi.org/10.1371/journal.pone.0118078.t001

Statistical Analyses

To explore the accumulation patterns of SB measures, descriptive statistics as well as the proportions (%) of accrued sedentary time and breaks within each bout duration were estimated. Univariate normality of sedentary time and breaks within each bout duration were examined by the skewness and kurtosis statistics in order to assure the use of parameterized linear models for sequential steps.

To evaluate the measurement properties of accrued sedentary time and breaks across a set of bout durations, bivariate correlation analyses with total sedentary time and breaks were performed using a mean of uncorrected item-total correlation analysis. Item-total correlation analysis is a well-known statistical approach for evaluating the construct validity of measurement at the item level [25]. In this study, we assumed that the sedentary time and breaks within each bout duration were the sub-items that consisted of total sedentary time and breaks, respectively. A positive and relatively high correlation coefficient of ≥.40 [25] was expected for each of the correlation analyses. In addition, bivariate correlation analyses for pairs of sedentary time and breaks at each bout duration were conducted as the secondary analysis to aid in better understanding of the practical significance of sedentary breaks after taking bout durations into account.

Lastly, separate linear regression models were fitted for each bout duration in order to evaluate the measurement properties of sedentary time and breaks within each bout duration in relation to cardiovascular risk factors after controlling for covariates. In this analysis, total sedentary time or breaks were not adjusted in each regression model due to the fact that the separate independent associations of the sub-components of total measure with the dependent variable may be unreliably estimated when total measure is adjusted in the model [26,27].

All statistical analyses were performed using SURVEY procedures in SAS v9.3 (SAS Institute Inc, Cary, NC) to account for the complex sampling designs in the NHANES. Four-year sample weights were calculated using the 2-year sample weights of the NHANES 2003–2004 and 2005–2006 cycles. To account for a selection bias by inclusion criteria of this study, 4-year sample weights were recalculated based on sample weights in the NHANES 2003–2006 data after taking age, gender, and racial/ethnic groups into account. For the analyses of fasting sub-component measure (i.e., TG), four-year fasting sub-sample weights were used. A prior significance level was set at p <.05 for all statistical analyses.

How Long is the Minimum Sedentary Bout Duration to Define Prolonged Sedentary Time?

SB has been defined as any activity during waking hours that requires low energy expenditures of <1.5 MET, which typically involves prolonged sitting or a reclined posture such as watching TV, working on a computer, or driving a car [6,7]. However, there has been a lack of clear definition regarding the minimum duration of sedentary bout that could potentially be considered as prolonged sedentary time, requiring more efforts to explore the accrued patterns of sedentary time in relation to health biomarkers.

The descriptive analyses of the accrued patterns of sedentary time showed that the majority of sedentary time occurred within relatively short bout durations. Of an average of 93.02 sedentary bouts, approximately 70% were attributed to the sedentary time that occurred at bout durations of <5-min (≈85% for <10-min), which accounted for approximately 27% of total sedentary time (≈47% for <10-min). Furthermore, bivariate correlations analyses for pairs of sedentary time at bout durations of 1-min, 2–4 min, and 5–9 min with total sedentary time showed negative or relatively weak linear relationships (r’s <.40). These findings imply that the sedentary time that lasts over relatively short durations may not represent the same measurement construct as the sedentary time at relatively long bout durations.

To further examine the significance of bout durations in objectively measured sedentary time, separate linear regression analyses were performed in relation to the health outcomes. Pertaining to total sedentary time, our findings are consistent with previous studies [12,28], where significant and unfavorable associations were observed with WC, HDL-C, and TG. Moreover, our result showing the insignificant relationship of total sedentary time with BMI was also consistent with a recent study [14] that used the same study sample. However, the associations between the computed total sedentary time and health outcomes varied depending on how bouts were defined. As shown in Table 5, the accrued sedentary time computed using bout durations of less than 5 minutes yielded negative correlations for most risk factors (WC, HDL-C, and BMI). When sedentary time was computed as more prolonged periods of time (i.e. bout durations ≥10 minutes), however, the associations were reversed.

In evaluating these findings it is important to consider the capabilities of the Actigraph for detecting sedentary time. The Actigraph accelerometer (model 7164) used in NHANES is an energy classification device that records the accelerations of vertical movements of the waist in forms of activity counts [29]. There have been different thresholds proposed to identify SB (e.g., <150 cpm [8], or <50 cpm [30]) for waist-mounted Actigraph accelerometers, of which an activity threshold of <100 cpm has broadly been used for classifying sedentary time and this has been supported in a number of past studies [23,31]. Regardless of the threshold, however, it is important to acknowledge that the Actigraph accelerometer based on a single vertical axis is not able to distinguish changes in posture (e.g., sitting vs. standing) [8,29]. Thus, epochs that feature activity counts <100 cpm may also include some light-intensity PAs in a standing position, such as washing dishes or folding laundry [32], which can possibly produce positive physiological effects by the postural muscle activations [3]. The simple accumulation of minutes with activity counts <100 cpm would therefore lead to overestimations of sedentary time [33], which in turn influence the implication of the relationship between SB and health outcomes. The constraint of tracking minutes only if they were accumulated for 5 or longer consecutive minutes would help to avoid capturing intermittent light-intensity PAs. Bankoski et al. [19] suggested a duration of >5 minutes be used to distinguish bouts of sedentary time and these analyses support this threshold. However, the selection of any threshold of minimum bout duration would lead to tradeoffs between sensitivity and specificity in the ability to accurately identify periods of SB. The decision between a duration of 5 or longer minutes would depend on the relative importance of false positives or false negatives in the analyses and future study is warranted for further investigation on this issue.

Using shorter epochs or intervals in summarizing accelerometer data would provide better descriptions of the continuity of human movement in a free-living environment. However, the duration to define a bout of SB must be specified to ensure that the accumulated minutes are congruent with the conceptual definition (i.e., the time spent in prolonged SB). The major context of SB (e.g., watching TV, driving a car, etc.) is predominated by prolonged sitting with low energy expenditure [6]. Relying on a single minute or shorter bout to estimate the time spent in SB from accelerometer data may inappropriately include time spent in sporadic sedentary behavior but not necessarily prolonged sedentary behavior. For example, a person may take brief breaks in a sitting position during an exercise session but this time should not be considered as part of the accumulated time spent in prolonged sedentary behavior. This may be an issue not only for Actigraph accelerometers but also for other accelerometers such as the ActivPAL (PAL Technologies Ltd, Glasgow, UK) which directly evaluate posture. A recent study with the ActivPAL in adolescents [34] defined a 15-second interval as a minimum duration of sitting time and the results also highlighted that a large number of sitting events occurred in short bout durations of <5-min. Although it would be challenging to determine the minimum duration of sedentary time which may negatively impact the physiological responses in the human body, efforts are needed to distinguish the time spent in prolonged sedentary behavior from accelerometer data should be made when the implication of the public health message is placed on reducing prolonged sedentary time and not on total sedentary time.

Are We Measuring Sedentary Breaks or the Number of Sedentary Bouts?

Promoting breaks in sedentary behavior has emerged as a promising intervention strategy to promote metabolic fitness and help decrease the risk of cardiovascular diseases [6,15]. Dunstan et al. [35] highlighted that even short bout of interruptions in sedentary time with light- or moderate-intensity walking significantly reduced the levels of postprandial glucose and insulin. Despite its potential for improving public health, little is known as to how to operationalize the breaks in sedentary time from accelerometer data that may provide the outputs congruent with what has been defined in the conceptual level. Particularly, the absolute number of transitions from sedentary to active phase has been extensively examined assuming that sedentary time is considered to be interrupted when transition occurs [15]; however, the measurement properties of this approach has been questioned.

In this study, the accrued patterns of sedentary breaks using the algorithm used by Healy et al. [15] showed that approximately 70% of sedentary breaks occurred at sedentary bouts of <5-min. (85% for <10-min), which is almost identical to what we found with respect to the number of sedentary bouts. Furthermore, the accrued sedentary breaks at 1-min bout duration are identical to the accrued sedentary time at 1-min bout duration. The underlying reason for this finding is that the current algorithm to extract sedentary breaks from accelerometer data is, indeed, an alternative expression to extract the number of sedentary bouts. In other words, counting the number of transition points from a sedentary to active phase would produce the same or slightly smaller quantity as the number of sedentary bouts. Only small differences would exist depending on the existence of sedentary time at the end of a continuous measurement period (e.g., sedentary time at the end of wear time or day would be considered as a sedentary bout but not counted for sedentary breaks). This notion could also be supported by the perfect linear relationships between sedentary time and breaks at each bout duration which clearly imply that sedentary breaks obtained by the current algorithm are an alternative parameter that quantifies the amount of sedentary time at each bout duration.

Meanwhile, the correlation coefficient between total sedentary time and total sedentary breaks was -.23 (See Table 2), which may lead to the conclusion of a weak relationship between total sedentary time and breaks [12]. However, this is because of the expansion of range in total sedentary time that significantly attenuates the relationship with total sedentary breaks. For example, 1-unit increases in total sedentary breaks may indicate an increase in total sedentary time by a minimum of one to thirty minutes or more depending on the sedentary bout durations where the breaks occurred. As shown in Table 4, the correlation coefficients between sedentary time and breaks get close to 1.0 as the range of sedentary time narrows for 1-unit increases in sedentary breaks.

While acknowledging the operational limitations of breaks in sedentary time, the transitions from a sedentary to active phase may include standing from a sitting position or walking a step which could also be considered as an indicator of PA. [15]. As illustrated in Table 5, our results showed that favorable associations with health outcomes were consistently found when bouts were defined as being <5 min. However, unfavorable associations were observed when SB was defined using bout durations of ≥10-min. These findings are likely similar to the results we found with respect to sedentary time at each bout duration. Given that the sedentary breaks is an alternative score that represents the number of sedentary bouts as discussed above, the implications may be more related to the patterns of how sedentary time is accumulated rather than to PA which may matter for health (e.g., a higher number of sedentary events with long bout is negatively associated with health outcome). Therefore, sedentary breaks presented as absolute number of transitions from sedentary to active phase may not be considered as breaks in sedentary time or an indicator of PA, and caution is warranted when drawing conclusions about sedentary breaks in relation to health outcome.

Taken together all above mentioned evidence, it is plausible to say that sedentary breaks represented by absolute number of transitions from sedentary to active phase is an incomplete measure of the patterns of sedentary time that does not take account for the respective bout durations which may matter more for health outcomes. Lyden et al. [9] proposed break rate calculated by total number of breaks divided by total sedentary time as a feasible metric specifically for detecting intervention effects. However, given that the breaks is an alternative measure of number of sedentary bouts, it also represents the average number of sedentary bouts to accumulate one sedentary hour. This may be more relevant to the accumulation patterns of sedentary time rather than to breaks in sedentary time, and more efforts to distinguish the breaks in sedentary time from the patterns of sedentary time accumulation should be made.

We believe that identifying the breaks or interruption in sedentary time in the observational study is a difficult task that cannot be comparable to examining the patterns of how sedentary time is accumulated. The noun break refers to the interruption of continuity or uniformity or a pause in work or during an activity or event [36]. This may imply that true breaks in sedentary time would possibly exist only within a continuous bout of SB pursuit. In other words, a strong assumption has to be made that being sedentary is a fundamental behavior of the participants during the measurement period if the transition from a sedentary to active phase is to be considered as a sedentary break. For example, in the laboratory experimental study conducted by Dunstan et al. [35], the participants were instructed to sit over 7 hours beginning with 2 hours to achieve steady state and then trial conditions (interruptions by 2-minute bouts of walking activities) were applied during the remaining 5 hours. The trial protocols were well-designed to fully reflect the conceptual definition of sedentary breaks (breaks or interruptions in sedentary time) because the participants were, again, forced to be sedentary during the measurement period. However, from an evolutionary perspective, humans are born to be active [37], and it may not be legitimate to consider a simple transition from a sedentary to active as a “break in sedentary time” in a free-living environment in which we do not know whether the observed transitions have purposely occurred within the continuity of SB pursuits or just simply at the true end of SB pursuit. One possible approach to identify objectively measured sedentary breaks in free-living environments could be a combination with subjective measures (e.g., PA log/diary) [38], from which one can obtain the time period information where the fundamental behavior of the participants were expected or assumed to be sedentary (e.g., office hours).

There have been a few attempts to operationally define the sedentary behavior bout that may possibly include true breaks within the bout. Carson and Janssen [39] defined a SB bout as a period of ≥30-minutes in which ≥80% of minutes are sedentary (i.e., <100 cpm) with no more than 5 consecutive minutes ≥100 cpm, from which the number of transitions from a sedentary to active phase was then extracted. Although there could be some practical issues such as restricting the break durations to <5 minutes, this could be one possible approach to overcome the limitation that may distinguish the operationalization of sedentary breaks from the patterns of sedentary time accumulation.

This study is not without limitations. First, the final analytic sample was demographically different from the sample excluded by selection criteria. Although the sampling weights were recalculated in order to produce unbiased estimates of population parameters of US adults, such differences may influence the precision of estimates due to the possible loss of non-ignorable missing data in the excluded sample. Second, the present study is data-driven research that relies on cross-sectional data that limits our ability to draw causal relationships of SB measures with health outcomes. Moreover, the implications of our findings are based on the practice of Actigraph accelerometer used in the NHANES 2003–2006 cycles, and caution is needed when interpreting the results for the study with different measurement protocols compared to the NHANES. Finally, the main focus of this study was limited to the data processing issues, particularly focusing on SB measures. There are several important issues in using accelerometer data for a large observational study, such as participants’ compliance, non-wear time, or defining non-wear time which could not be addressed in the present study. The readers who are interested in these particular issues should refer to previous research [20,40–42] for an extensive understanding of those issues and to potentially address them in future research.

SB is purposefully engaged activities in different contexts that are predominated by prolonged sedentary time with low energy expenditure and possibly include sedentary breaks [6]. However, the most commonly used algorithms to obtain SB parameters from accelerometer data may not perform well enough to fully reflect the conceptual definitions of respective parameters in free-living settings. The present study elucidated the necessity of determining the minimum duration of sedentary time that can potentially be considered as prolonged SB. Prior information on SB bouts would be required in order to identify true sedentary breaks that fully reflect the conceptual definition of sedentary breaks (i.e., interruptions in sedentary time). Future research should be aimed at developing a new algorithm/approach to discriminate SB bouts from accelerometer data in order to improve the measurement properties of objectively measured SB in health outcome research.