Accelerometry-assessed physical activity and sedentary behavior patterns using single- and multi-component latent class analysis among postmenopausal women

Abstract

Background:

Patterns of physical activity and sedentary behavior among postmenopausal women are not well characterized.

Objectives:

To describe the patterns of accelerometer-assessed physical activity and sedentary behavior among postmenopausal women.

Design:

Cross-sectional study.

Methods:

Women 63–97 years (n = 6126) wore an ActiGraph GT3X + accelerometer on their hip for 1 week. Latent class analysis was used to classify women by patterns of percent of wake time in physical activity and sedentary behavior over the week.

Results:

On average, participants spent two-thirds of their day in sedentary behavior (62.3%), 21.1% in light low, 11.0% in light high, and 5.6% in moderate-to-vigorous physical activity. Five classes emerged for each single-component model for sedentary behavior and light low, light high, and moderate-to-vigorous physical activity. Six classes emerged for the multi-component model that simultaneously considered the four behaviors together.

Conclusion:

Unique profiles were identified in both single- and multi-component models that can provide new insights into habitual patterns of physical activity and sedentary behavior among postmenopausal women.

Implications:

The multi-component approach can contribute to refining public health guidelines that integrate recommendations for both enhancing age-appropriate physical activity levels and reducing time spent in sedentary behavior.

Keywords

accelerometer leisure-time physical activity older adults patterns postmenopausal women sedentary time

Introduction

Latent class analysis (LCA) has emerged as a powerful, data-driven tool that can separate individuals by multiple person-level characteristics into classes or groupings. In LCA, participants are assumed to belong to one of several mutually exclusive classes, but for which a priori class membership is not known.¹ This method helps identify homogeneous groups based on the data structure of specific variables under study.² It assumes conditional independence such that the observed variables are independent conditional on the latent class membership. Using a statistical model, participants are assigned to a category or class based on the associations among observed variables. One such case where this method is particularly useful is for accelerometer data; more specifically, for data exploration and reduction, identifying unique groups of individuals to target interventions, and combining other variables with the accelerometry data into a single metric.

Based on a recent review of the literature, studies that used LCA applied to accelerometry-assessed physical activity and sedentary behavior (e.g. physical behaviors) were identified.³ Three groups of adult studies applied LCA to accelerometry: the National Health and Nutrition Examination Survey (NHANES) among adults 18 years and older,^1,4
–7 the Physical Activity in Public Space Environments (PHASE) Study⁸ among adults 45 to 65 years, and the Western Australia Pregnancy Cohort (Raine) Study among expectant mothers.⁹ Two of the three study groups included an exploration into sedentary behavior patterns.^1,4,5,9 None of the studies utilized three axes of movement from the accelerometer, instead focusing on one axis using count data. Moreover, none of the studies focused specifically on older adults, for whom patterns of physical activity and sedentary behavior will likely differ as compared to younger- or middle-aged adults.

Latent classes can be developed using single components of physical activity or sedentary behavior, such as by focusing on a specific physical activity intensity level (e.g. light, moderate, vigorous). However, rather than considering the components in isolation, a multi-component approach considers each “physical behavior” in one model. This approach may be preferred, as it better reflects the finite time of day and the inter-relatedness among physical behavior characteristics during that time interval.¹⁰ Initial LCA studies used a single component applied to either physical activity^1,4
–7 or sedentary behavior.^1,4,5 More recently, two studies of adults used a multi-component approach to develop latent classes: Jansen et al.⁸ included both light and moderate-to-vigorous physical activity (MVPA) in the model, and Howie et al.⁹ included eight variables describing steps per day, MVPA, and sedentary behavior in the model.

The Women’s Health Initiative Objective Physical Activity and Cardiovascular Health (WHI OPACH) Study addressed the limitations of prior studies with a sample large enough to explore latent classes among ambulatory women 63 years and older in the community setting to assess physical behavior measured with an accelerometer that captured all three planes of movement.¹¹ The present cross-sectional study aimed to describe patterns of accelerometer-assessed physical activity and sedentary behavior among these postmenopausal women participating in WHI OPACH using both single- and multi-component LCA models to provide insight into the unique contributions both model types could provide. Understanding predominant patterns of physical behaviors among this cohort can aid development of appropriate interventions for postmenopausal women.

Methods

The WHI OPACH Study is an ancillary study to the WHI, a study in which postmenopausal women ages 50–79 years old were enrolled in the WHI Clinical Trials or the Observational Study from 40 clinical sites throughout the United States from 1993 to 1998, with more details elsewhere.^12,13 From March 2012 to May 2013, the ancillary Long Life Study was conducted among 7875 WHI women aged 63 and older on January 2012, with oversampling of African Americans and Hispanics. A subset of 7048 women agreed to participate in the WHI OPACH study, which involved wearing an accelerometer, keeping a sleep log, and completing additional questionnaires. More detail on the sample derivation and estimation is provided elsewhere.¹¹ The study included women from all 40 original US clinical centers, examining them in their homes. We followed the reporting guidelines from the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) for cross-sectional studies (Supplement A).

Physical activity measurement by accelerometry

Women who had consented to participate in the Long Life Study and agreed to wear an accelerometer were enrolled into the WHI OPACH Study.^11,14 Women were fitted with an ActiGraph GT3X + triaxial accelerometer (Pensacola, Florida) and asked to wear it for 1 week. The accelerometers were distributed by home visit examiners when an accelerometer was available at the time. If no accelerometer was available at the time of the home visit, or if the home visit could not be scheduled before the end of the Long Life Study, then the accelerometer was provided through express mail after a telephone call with the participant to alert them to the package arrival.

Participants were instructed to place the accelerometer on their right hip above the iliac crest, using the study-provided elastic belt worn around the waist, and to wear the accelerometer 24 hours/day for 7 days, except during bathing or swimming. Staff provided a phone number for troubleshooting and attempted telephone contact with each participant once during the first half of the week to encourage wearing the accelerometer and to answer any questions. After wearing the accelerometer for 7 days, participants mailed the accelerometer to the WHI coordinating center, where the data were downloaded and stored.

Women were asked to keep a sleep log (published as Additional File 1 by Rillamas-Sun et al.¹⁵) to record time in bed and out of bed throughout the 1 week when they wore the accelerometer. They were also asked to record any time they took off the accelerometer and the reason for non-wear time. We relied on the sleep logs to indicate accelerometer wear days and both in- and out-of-bed times. To maximize the use of accelerometer data when sleep logs were missing or suspected to have reporting errors, a computer-based automated algorithm was used to identify accelerometer wear.¹⁵ The algorithm was applied to all accelerometer tracings in alignment with the paper sleep log and visual inspection to identify the window of days with the maximum amount of wear over a consecutive 7-day period.

During data collection to output the data, the ActiGraph software (ActiLife) versions 6.0.0-6.10.1 were used. Both raw data (30 Hz) and counts data were downloaded, the latter in 15-s epochs with all three axes using the normal filter. Vector magnitude (VM) acceleration counts were derived by taking the square root of the sum of the counts squared from the vertical, anterior-posterior, and medial-lateral axes for a given 15-s epoch. Non-wear was defined by intervals with consecutive zero VM for at least 90 minutes, with an allowance of nonzero VM up to 2 minutes if no counts were detected during both the 30-min upstream and downstream from that interval; any non-zero VM counts (except the allowed short intervals) were considered wear time.^16,17 VM counts in the non-wear period were set to missing.

We characterized the average volume (e.g. total amount) of physical activity using average VM counts/15-s while awake for the wear day. Then we classified the intensity of each epoch using intensity-specific cutpoints calibrated by minimizing false positives and false negatives from calibration data derived among women ⩾60 years using the normal filter where the basal intensity was defined as 1 metabolic equivalent (MET) = 3.0 mL oxygen/kg/min.¹⁸ Using these criteria, the cutpoints were defined as follows: sedentary behavior 0–18 VM/15-s (1.0–<1.5 METs), light low 19–225 VM/15-s (1.5–2.0 METs), light high 226–518 VM/15-s (2.0–<3.0 METs), and MVPA ⩾ 519 VM/15-s (⩾3 METs). Light-intensity physical activity was separated since postmenopausal women tend to engage in a wide range of activities of lower intensity.¹⁸ These metrics, along with steps per day derived from ActiLife proprietary software algorithms, were used to describe behavior across the latent classes.

Other measures

Sociodemographics, including age, race/ethnicity per study design, and education, were collected at study enrollment. Height and weight were measured and general health status was self-reported at the Long Life Study home visit. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. We used categories promoted by the Centers for Disease Control and Prevention:¹⁹ underweight (<18.5 kg/m²), normal weight (18.5–24.9), overweight (25.0–29.9), or obese (⩾30.0).

Statistical analysis

Adherence definitions (days/week and h/day) were based on exploration of the accelerometry data and supported by commonly used definitions in other observational studies.^20,21 Overall, 7048 women enrolled in the WHI OPACH Study, of whom 6721 returned their accelerometers, and 6489 had data available for analysis. We required a minimum wear time of ⩾4 of 7 days and ⩾10 h/day, leaving a final sample of 6126 women.

Pearson correlation coefficients were used to describe correlations among accelerometer metrics. From the participant’s accelerometry, we used 4 to 7 adherent days in the LCA models to create natural groupings or classes of participants who tended to accumulate their physical activity or sedentary behavior in a similar pattern across (1) clock time (e.g. time of day) and (2) time since awake. Based on data exploration, we used the time since awake, rather than clock time, to better account for missing data. Based on hours worn, accelerometry data for each participant were averaged over the 1-h time window for each day, starting at the first hour with any wear time and ending at the last hour with any wear time, and then averaged for the week during the same time window. A missing 1-h time window was indicated only when the participant did not wear the accelerometer for the entire hour over all 7 days.

Single-component models were developed separately for continuous measures of sedentary behavior, light low, light high, and MVPA. Specifically, for each of these four measures mean percent for each hour of the day was calculated and then averaged over the week. Given that the majority (99.7%) of participants had recordings for ⩾13 h/day, LCA models used 13 continuous variables capturing this average, across the week, for each hour in a 13-hour time span. Multi-component models were then developed using these 13 continuous variables, representing the 13 h, for all four continuous metrics in one model. Several criteria were used to select the final number of classes including:^22,23

The Lo–Mendell–Rubin adjusted likelihood ratio test which compared the fit of k classes to (k – 1) classes.

The entropy, which estimated how distinct the identified classes were from one another and indicated how the model defines classes.

The Akaike information criterion (AIC) and Bayesian information criterion (BIC), both indicators of model fit.

The sample size of the classes, requiring each class to have at least 2% of the sample.

Substantive knowledge with a practical interpretation of what each class represented, along with visual inspection to ensure that the classes were sufficiently separated from each other.

The LCA was performed using MPlus (version 8.8) with 500 iterations and 20 random starts.²⁴ Full information maximum likelihood methods were used to account for missing data. LCA was applied to describe the relationship between the adherent days of accelerometry (averaged over the week by accelerometer hour worn) and the categorical latent variable using a set of linear regression equations. For each variable, a 3-class model was estimated first and continued up to 6 or 7 classes if necessary. Posterior probabilities were calculated for single- and multi-component models for each class. Each participant was assigned to the optimal class based on the highest posterior probability or modal allocation in each of the four single-component models and to one class for the multi-component models. SAS^® release 9.4 (Cary, North Carolina) was used for all descriptive statistics.

Results

Description of the sample

Among 6126 participants, the mean age was 78.7 years (SD: 6.7, median 79.0, interqualtile range [IQR]: 73–84) and the mean BMI was 28.2 kg/m² (SD: 5.7, median 27.4, IQR: 24.2–31.3). Based on standard BMI categories, 1.3% were underweight, 30.4% were normal weight, 36.4% were overweight, and 31.8% were obese. Overall, 49.7% of the sample was Non-Hispanic White, 33.4% were Non-Hispanic Black, and 16.9% were Hispanic. Many women had at least some college experience (38.6% some college, 41.1% college graduate or more), and 20.3% had a high school education or less.

The average accelerometer (out-of-bed) wear time was 14.9 h/day (standard deviation (SD) 1.3; median 15.0; IQR: 14.1–15.8). There was a drop-off in accelerometer wear after 13 hours (Supplemental Figure 1). Therefore, we explored latent class patterns for up to 13 hours since awake. Overall, across 13 hours since awake, only 0.1% (100/(13*6126)) of the 1-h time windows were missing due to non-wear time. Overall, 98.7% (6046/6126) of the women had complete data across all 13 1-h time windows. Across all hours of adherent wear time, the sample averaged 62.3% of the day (555.6 min/day) in sedentary behavior, 21.1% of the day (188.9 min/day) in light low physical activity, 11.0% of the day (98.0 min/day) in light high physical activity, and 5.6% of the day (50.4 min/day) in MVPA.

Next, we explored whether there was variation in physical activity or sedentary behavior by day of the week to assess whether it was important to retain the day of the week in the latent class models. The patterns for both MVPA (Supplemental Figure 2; numeric values in Supplemental Table 1) and sedentary behavior (Supplemental Figure 3; numeric values in Supplemental Table 1) were similar by day of the week; therefore, we explored latent classes using weekly continuous averages of physical activity and sedentary behavior by the hour since awake.

Single-component models

Considering single-component models, women were optimally classified into five classes each for sedentary behavior, light low, light high, and MVPA. Each class was ordered based on average VM/15-seconds, with the lowest average VM in class 1 and the highest average VM in class 5. For each component, mean wear time was successively higher by class, but the variation around the mean was practically small across classes (mean wear range 14.5–15.3 hours/day; Supplemental Table 2).

The posterior probability of class assignment, considering five classes, would be 0.20 if based solely on chance. The median posterior probabilities indicated acceptable assignment, ranging from 0.87 to 1.00 (Supplemental Table 3). Considering the 10th percentile, posterior probabilities were 0.66–0.93, except two classes for percent of light low (class 3: 0.54 and class 4: 0.58). The corresponding entropy, AIC, BIC, and Lo–Mendell–Rubin tests for the single-component models are displayed in Supplemental Table 4 from 3 to up to 6 or 7 classes.

Sedentary behavior

For the mean percent of sedentary behavior out of total wearing time, five classes were identified from most to least sedentary: 11.5% of the sample (mean 682.7 minutes/day), 28.5% (614.8), 33.4% (547.3), 20.6% (473.5), and 5.9% (386.7) (Table 1). The mean percentages averaged across the week are plotted by the hour of the day in Figure 1 (numeric values in Supplemental Table 5). For the most sedentary two classes (lowest average VM/15-s for class 1 and class 2), the mean percent in sedentary behavior was lowest in the first hour of wear and then was higher for all subsequent hours of the day. For class 3, the mean percent of the day in sedentary behavior was successively lower from the first to the fourth hour of wear, and by the sixth hour, it exceeded the value in hour 1 of wear. For class 4, the mean percent of the hour in sedentary behavior was successively lower from the first to the third hour of wear, and it did not exceed the value at hour 1 until the nineth hour of wear. For class 5, the percent of the hour in sedentary behavior was successively lower from the first to the sixth hour, and it did not exceed the value at hour 1 until the 13th hour of wear.

Table 1.

Sample distribution for each latent class from single- and multi-component models; WHI OPACH Study 2012-2014 (n = 6126).

	n	%	Mean % of day in each intensity (SD)	Mean minutes/day (SD)	Mean VM/15-seconds (SD)	Mean steps/day (SD)
Single-component
Percent of sedentary out of total wearing time per 1-h window per day
Class 1—Most sedentary	707	11.5	78.3	682.7	47.9 (15.5)	1386.9 (768.9)
Class 2	1747	28.5	69.2	614.8	75.8 (19.1)	2538.0 (1252.5)
Class 3	2047	33.4	60.8	547.3	104.1 (23.7)	3739.1 (1625.7)
Class 4	1262	20.6	52.2	473.5	138.5 (29.5)	5042.8 (2049.9)
Class 5 -Least sedentary	363	5.9	42.2	386.7	181.2 (40.0)	6748.4 (2658.2)
Percent of light low physical activity out of total wearing time 1-h window per day
Class 1—Least active	982	16.0	13.7	119.9	69.6 (32.1)	2325.5 (1628.7)
Class 2	2230	36.4	18.8	168.3	96.5 (39.5)	3471.1 (2117.6)
Class 3	786	12.8	23.9	212.0	105.2 (35.3)	3672.6 (1879.1)
Class 4	1441	23.5	23.9	216.4	116.1 (41.9)	4169.1 (2246.0)
Class 5—Most active	687	11.2	30.1	273.2	125.9 (39.7)	4317.6 (2099.6)
Percent of light high physical activity out of total wearing time 1-h window per day
Class 1—Least active	759	12.4	5.2	45.4	49.2 (18.4)	1442.7 (977.2)
Class 2	2122	34.6	8.8	78.5	82.5 (25.7)	2855.4 (1543.3)
Class 3	2049	33.5	12.1	108.9	112.5 (29.9)	4078.9 (1898.8)
Class 4	976	15.9	15.7	141.4	142.7 (35.4)	5158.7 (2212.7)
Class 5—Most active	220	3.6	19.9	180.0	171.8 (37.8)	6083.2 (2411.1)
Percent of MVPA out of total wearing time 1-h window per day
Class 1—Least active	2749	44.9	2.6	22.9	68.8 (20.5)	2132.4 (1014.1)
Class 2	2080	34.0	6.1	55.3	108.2 (20.7)	3838.0 (1317.1)
Class 3	166	2.7	9.9	91.3	145.6 (34.7)	7361.6 (2324.7)
Class 4	928	15.2	10.3	93.4	150.1 (22.8)	5546.8 (1800.5)
Class 5—Most active	203	3.3	16.1	147.5	208.3 (36.7)	8232.9 (2742.6)
Multi-component			n/a	n/a
Class 1—Highest sedentary and least active	805	13.1			49.3 (15.5)	1435.9 (779.4)
Class 2	1713	28.0			77.6 (17.9)	2621.4 (1193.0)
Class 3	1330	21.7			90.9 (16.1)	3036.0 (1170.9)
Class 4	1045	17.1			132.7 (21.5)	5105.3 (1871.8)
Class 5	858	14.0			134.0 (21.9)	4738.1 (1681.2)
Class 6—Lowest sedentary and most active	375	6.1			194.3 (33.2)	7460.2 (2527.7)

Abbreviations: MVPA, moderate-to-vigorous physical activity; n/a, not applicable; SD, standard deviation; VM, vector magnitude.

Figure 1.

Single-component latent class analysis results for percent of sedentary behavior graphed for averages by class and hours since awake; WHI OPACH Study 2012-2014 (n = 6126).

Light low physical activity

For the mean percent of light low physical activity out of total wearing time, five classes were identified from least to most active: 16.0% of the sample (mean 119.9 minutes/day), 36.4% (168.3), 12.8% (212.0), 23.5% (216.4), and 11.2% (273.2) (Table 1). The mean percentages across the week are plotted in Figure 2 (numeric values in Supplemental Table 5). For class 1 (lowest average VM/15-seconds), the mean percent of the day in light low remained flat from hour 1 until hour 6 when it declined for the rest of the day. The mean percent of the day in light low increased successively for classes 2, 3, 4, and 5 starting at hour 1 and then began mostly declining at hours 6, 6, 9, and 7, respectively. Class 3 dropped off more sharply than classes 2, 4, and 5. Class 3 also crossed over class 4 between hours 7–8, while class 5 remained highest.

Figure 2.

Single-component latent class analysis results for percent of light low physical activity graphed for averages by class and hours since awake; WHI OPACH Study 2012–2014 (n = 6126).

Light high physical activity

For the mean percent of light high physical activity out of total wearing time, five classes were identified from least to most active: 12.4% of the sample (mean 45.4 minutes/day), 34.6% (78.5), 33.5% (108.9), 15.9% (141.4), and 3.6% (180.0) (Table 1). The mean percentages across the week are plotted in Figure 3 (numeric values in Supplemental Table 5). The mean percent of the day in light high physical activity declined over the hours of the day from hour 1 for classes 1–3. For classes 4 and 5, light high physical activity successively increased from hour 1 until hour 5 and 7, respectively, when it started to decline.

Figure 3.

Single-component latent class analysis results for percent of light high physical activity graphed for averages by class and hours since awake; WHI OPACH Study 2012–2014 (n = 6126).

MVPA

For the mean percent of MVPA out of total wearing time, five classes were identified from least to most active: 44.9% of the sample (mean 22.9 minutes/day), 34.0% (55.3), 2.7% (91.3), 15.2% (93.4), and 3.3% (147.5) (Table 1). The mean percentages across the week are plotted in Figure 4 (numeric values in Supplemental Table 5). The mean percent of the day in MVPA was low and stable across all hours of the day for classes 1 and 2 (these classes had the lowest average VM/15-s). Class 3 had the highest of all measured MVPA in hours 1 and 2, potentially exercising during this period, and then dropped to low levels between hours 5–13. In contrast, class 4 and 5 successively increased their MVPA until hours 4 and 5, respectively, when MVPA started declining. Class 5 had a sharp peak at hours 1 and 2 but then were similar to class 2 by the fifth hour. To summarize assignments across percent of sedentary, light low, light high, and MVPA from single-component models, we report cross-tabulations (Supplemental Table 6).

Figure 4.

Single-component latent class analysis results for percent of moderate-to-vigorous physical activity graphed for averages by class and hours since awake; WHI OPACH Study 2012–2014 (n = 6126).

Multi-component models

Before conducting multi-component models we assessed bi-variate correlations, with no two components with high values that would preclude the modeling. For the sample (n = 6126), there was a positive correlation between light high with light low (0.62), light high with MVPA (0.59), and light low with MVPA (0.23). There was a negative correlation between sedentary behavior and all physical activity metrics: light high (−0.64), light low (−0.53), and MVPA (−0.49).

Considering multi-component models, women were optimally classified into 6 classes (Figure 5; numeric values in Supplemental Table 5). The mean sample percent assigned to each class ranged from 6.1% (class 6) to 28.0% (class 2) (Table 1). Each class was ordered based on average VM/15-seconds, with the lowest average VM in class 1 and the highest average VM in class 6. This aligned with the ordering of mean steps/day as well. The average VM/15-seconds is shown in Table 2. The least-active class (class 1) had the highest mean percent of the day in sedentary behavior, and the lowest mean percent of the day in light low, light high, and MVPA. In contrast, the most active class (class 6) had the lowest average percent of the day in sedentary behavior, and the highest average percent of the day in light low, light high, and MVPA. The mean accelerometer wear time was successively higher by class, but the range variation was small (mean wear range 14.5–15.2 h/day from class 1–6; Supplemental Table 2).

Figure 5.

Multi-component latent class analysis results for percent of sedentary behavior (upper left) and light low (upper right), light high (low left), and moderate-to-vigorous physical activity (lower right) graphed for averages by class and hours since awake; WHI OPACH Study 2012–2014 (n = 6126).

Table 2.

Average VM/15-s for each multi-component latent class by hour of the day; WHI OPACH Study 2012–2014 (n = 6126).

	Average VM/15-sec by hours since awake
	1	2	3	4	5	6	7	8	9	10	11	12	13
Class 1—Highest sedentary and least active	309.7	252.5	239.0	236.5	226.3	217.6	201.9	188.3	188.0	193.3	175.2	167.4	148.7
Class 2	446.3	418.4	409.4	384.6	367.9	343.5	320.7	301.1	291.5	281.5	277.4	251.3	232.0
Class 3	455.3	452.7	444.1	431.8	409.8	393.7	385.8	379.0	368.4	353.8	338.2	319.0	293.0
Class 4	692.5	735.4	733.9	699.3	647.5	597.6	554.5	522.6	505.1	484.5	465.6	432.7	379.0
Class 5	615.2	664.4	666.7	641.2	607.7	584.5	567.3	546.1	546.1	535.9	522.2	491.0	447.8
Class 6—Lowest sedentary and most active	845.0	925.1	973.0	957.3	928.9	893.1	849.6	822.6	822.7	789.3	752.2	722.1	633.7

Abbreviations: MVPA, moderate-to-vigorous physical activity; WHI OPACH, Women’s Health Initiative Objective Physical Activity and Cardiovascular Health.

The posterior probability of class assignment, considering six classes, would be 0.17 if based solely on chance. The median posterior probabilities indicated highly acceptable assignment, with all medians at 1.00 (Supplemental Table 3). Considering the 10th percentile, posterior probabilities ranged from 0.77 to 0.96, while at the 90th percentile, posterior probabilities were all 1.00. The corresponding entropy, AIC, BIC, and Lo–Mendell–Rubin tests for the multi-component model are displayed in Supplemental Table 4 from 3 up to 7 classes.

The cross-tabulations comparing single- to multi-component classes indicated that membership was consistent across intensities, such that women with those assigned to classes with more MVPA tended to have lower sedentary behavior and light low physical activity (Table 3). Compared to class 5, class 6 had lower light low and similar light high, but the highest overall volume (based on average VM/15-s) due to their higher MVPA throughout the day. Similarly, women with more sedentary behavior tended to have lower light high and MVPA.

Table 3.

Cross-tabs of single-component compared to multi-component models; WHI OPACH Study 2012–2014 (n = 6126).

Single-component models	Multi-component Model
	Class 1		Class 2		Class 3		Class 4		Class 5		Class 6
	N	%	N	%	N	%	N	%	N	%	N	%
Percent of sedentary out of total wearing time per 1-h window per day
Class 1—Most sedentary	702	99.3	5	0.7	0	0	0	0	0	0	0	0
Class 2	103	5.9	1548	88.6	53	3	43	2.5	0	0	0	0
Class 3	0	0	160	7.8	1152	56.3	724	35.4	9	0.4	2	0.1
Class 4	0	0	0	0	125	9.9	278	22	668	52.9	191	15.1
Class 5 -Least sedentary	0	0	0	0	0	0	0	0	181	49.9	182	50.1
Percent of light low physical activity out of total wearing time 1-h window per day
Class 1—Least active	600	61.1	300	30.6	0	0	79	8	0	0	3	0.3
Class 2	197	8.8	1165	52.2	132	5.9	627	28.1	11	0.5	98	4.4
Class 3	1	0.1	107	13.6	379	48.2	115	14.6	142	18.1	42	5.3
Class 4	7	0.5	141	9.8	623	43.2	223	15.5	278	19.3	169	11.7
Class 5—Most active	0	0	0	0	196	28.5	1	0.2	427	62.2	63	9.1
Percent of light high physical activity out of total wearing time 1-h window per day
Class 1—Least active	567	74.7	166	21.9	22	2.9	4	0.5	0	0	0	0
Class 2	237	11.2	1233	58.1	446	21	190	8.9	8	0.4	8	0.4
Class 3	1	0.1	311	15.2	783	38.2	656	32	218	10.6	80	3.9
Class 4	0	0	2	0.2	79	8.1	195	19.9	500	51.2	200	20.5
Class 5—Most active	0	0	1	0.5	0	0	0	0	132	60	87	39.6
Percent of MVPA out of total wearing time 1-h window per day
Class 1—Least active	722	26.3	1077	39.2	824	29.9	1	0.04	125	4.6	0	0
Class 2	77	3.7	591	28.4	494	23.8	421	20.2	497	23.9	0	0
Class 3	4	2.4	22	13.2	12	7.2	96	57.8	22	13.2	10	6
Class 4	2	0.2	23	2.5	0	0	509	54.8	214	23.1	180	19.4
Class 5—Most active	0	0	0	0	0	0	18	8.9	0	0	185	91.1

Abbreviations: MVPA, moderate-to-vigorous physical activity; WHI OPACH, Women’s Health Initiative Objective Physical Activity and Cardiovascular Health.

Discussion

This study described patterns of accelerometer-assessed physical activity and sedentary behavior by time since awake among postmenopausal women using both single- and multi-component LCA models. Applying LCA to accelerometry utilizes a data-driven approach to discover underlying common patterns in the data that might not have been identified otherwise. In our approach, both the single- and multi-component approach provided new insights into physical activity and sedentary behavior patterns.

Patterns of physical activity

Participants spent almost 6% of their wear time engaging in MVPA. Definitions of MVPA differ widely across accelerometer studies for this age group, but this amount was within range of another study that explored varying cutpoints.²⁵ In our study, the highest MVPA occurred in the first 4 h that the accelerometer was worn across all classes. In general, if women participate in MVPA, it is often during the first few hours of awakening. Based on the single-component model, women accumulating the most MVPA (class 4 and 5) spread it out throughout the day, whereas women in class 3 had a period of MVPA in the morning and then were quite similar to class 4 for the remainder of the day. For the analysis, we divided light activity into lower and higher intensities since most physical activity occurred at this intensity level. In the multi-component models, for classes 4, 5, and 6, the pattern of accumulating more MVPA and less-sedentary behavior was concurrent with lower amounts of light low and moderate amounts of light high physical activity.

Patterns of sedentary behavior

A prior review found that adults 60 years and older spend on average 9.4 hours/day in accelerometer-assessed sedentary behavior.²⁶ Studies published since the review indicated an average sedentary behavior or sitting time between 9.1 to 10.1 hours/day.^27,28 When focused on those 80 years and older, the average is higher with a review reporting 10.6 hours/day in accelerometer-assessed sedentary behavior.²⁹ These reported values were close to the average of 9.3 hours/day of sedentary behavior found in this study. However, it should be acknowledged that across the range of studies, devices, wear location, and cutpoints varied. Based on both single- and multi-component results, sedentary behavior was higher later in the day, particularly after the sixth or seventh hour of wear time. Focusing on potential interventions, afternoons could be a time to focus on breaking up sedentary behavior with standing or ambulatory physical activity.

Physical behavior by day of the week and time since awake

We initially explored latent class patterns by each day of the week which was similar to other studies of adults that applied LCA to accelerometry.^1,6,7 The participants had low between-day variability among postmenopausal women; therefore, we simplified the model and explored weekly patterns. It should be noted that our findings are in contrast to other studies of older adults that found less physical activity on Sundays compared to Saturdays,^30
–33 and higher sedentary behavior on the weekends compared to the weekdays.³⁴

We also explored latent class patterns by both clock time and time since awake, with the latter being superior with regards to missing data. Our findings were similar to other studies of older adults in that physical activity generally peaked earlier in the morning and declined the rest of the day overall,^{31,32,35
–37} for men,³³ and for women.³⁸ This pattern is in contrast to another study of younger adults that accumulated more physical activity in the afternoon and sustained their physical activity levels longer.³⁷ Our findings also concurred with others that found sedentary behavior to be more common toward the end of the day,³¹ particularly with fewer sedentary breaks.³⁴

Single- and multi-component models

A unique contribution of this study is the contrast between single-component models that explored intensity levels separately and multi-component models that accounted for both sedentary behavior and physical activity in the same model. Notably, the single-component MVPA model identified the “morning exerciser pattern” (class 3) that was not evident in the multi-component model. Generally, the multi-component model had higher posterior probabilities and higher entropy than the single-component models indicating a superior approach statistically. The multi-component approach is also more in line with the concept of “exchangeability” of time spent in various awake behaviors, better representing the finite time of day and how behaviors occur. This approach is also useful to future public health recommendations that integrate all components of physical behavior into guidance.¹⁰ Our study focused on wake-time data. Future studies could explore the entire 24-hour clock by incorporating sleep into the models³⁹ and introducing more complex models which account for the multilevel nature of the data by avoiding the averaging of accelerometry within a time or across days.⁴⁰

Limitations and strengths

Several limitations of this study should be noted. First, accelerometry was measured for only 1 week and may not represent habits over the year, although times throughout the year were represented across the cohort. It may also not represent habits over broader lifespan transitions (e.g. middle age into older age). However, another study with women of similar ages as in WHI OPACH described consistent accelerometer-assessed physical activity over a 2- to 3-year period, with intraclass correlations ranging from 0.7 to 0.8.⁴¹ Moreover, women from the WHI Observational Study (n = 92,629), of which a subset participated in WHI OPACH, self-reported recreational physical activity that was stable over 8 years of follow-up based on up to five measurements.⁴² Second, these classes were based on a large cohort of postmenopausal women 63–97 years from across the United States, but we cannot be sure if these classes replicate elsewhere, such as in men of similar age or among women with specific comorbidities. Third, we did not account for the uncertainty of class assignment when we assigned each woman to the class in which she had the highest posterior probability. However, given the relatively high median values of posterior probabilities across all classes (Supplementary Table 3), we were above guidelines proposed by other studies which suggest a median probability of 0.90 to be ideal and values between 0.80 and 0.90 to be acceptable.²³ Other methods have been proposed as an alternative to address issues with uncertainty in class assignment.^43
–45 However, these methods are geared toward exploring the association between classes with observed covariates and outcomes and cannot be easily applied in cases such as ours where we seek to explore the differences in class membership between the derivation of several approaches to deriving latent classes using the same data—in our case, contrasting single-component to a multi-component approach. Fourth, the patterns may vary by other important factors, such as employment, education, season of the year, weather, and location. However, our intent was to create patterns that could then be explored with other potential correlates or outcomes, independent of accounting for these measures.

Despite the limitations, our study had several strengths. We addressed prior study limitations by including both physical activity and sedentary behavior, utilized three axes from the accelerometer, and focused on older adults. Our sample included a large racial and ethnically diverse cohort of women 63–97 years enrolled from the community across the United States without conditioning on health status broadly defined and without any targeted intervention on physical behavior during the device-wear interval. Thus, the LCA patterns in physical behavior reported herein likely provide a sense of what might be expected among women of similar sociodemographic characteristics nationally. Our analysis to develop latent classes explored using time of day versus clock time and using a day of the week versus the entire week. We were able to contrast the single- and multi-component approaches. Accelerometry wear time was high, physical activity and sedentary behavior were defined using cutpoints developed from a calibration study among similar-aged women,¹⁸ and there was minimal missing data across the 13 hours of the day. Based on a recent review,³ this is one of three parent studies,^1,4,5,9 to include sedentary behavior patterns and the first among those to include three axes of movement from the accelerometer which should improve the measurement precision. From the prior review,³ it was the only study to focus on older adults, specifically women.

Conclusion

Prior studies have used LCA for data reduction, to combine multiple variables into one metric, and to identify unique groups for intervention. We used LCA for descriptive purposes to explore patterns of accelerometer-assessed physical activity and sedentary behavior in an understudied population subgroup, community-dwelling women 63–97 years of age. Both the single- and multi-component models captured unique classes to describe habitual patterns and intensities of physical activity and sedentary behavior among participants. The multi-component approach can contribute to refining public health guidelines that integrate recommendations for both enhancing age-appropriate physical activity levels and reducing time spent in sedentary behavior with an emphasis on daily time intervals where breaks in sedentary behavior appear particularly relevant at older ages.

Supplemental Material

sj-docx-1-whe-10.1177_17455057241257361 – Supplemental material for Accelerometry-assessed physical activity and sedentary behavior patterns using single- and multi-component latent class analysis among postmenopausal women

Supplemental material, sj-docx-1-whe-10.1177_17455057241257361 for Accelerometry-assessed physical activity and sedentary behavior patterns using single- and multi-component latent class analysis among postmenopausal women by Kelly R Evenson, Fang Wen, Chongzhi Di, Michael Kebede, Michael J LaMonte, I-Min Lee, Lesley Fels Tinker, Andrea Z LaCroix and Annie Green Howard in Women’s Health

Footnotes

Acknowledgements

None.

Declarations

ORCID iDs

Kelly R Evenson

Fang Wen

Chongzhi Di

Michael Kebede

Michael J LaMonte

I-Min Lee

Annie Green Howard

Supplemental material

Supplemental material for this article is available online.

References

Evenson

Wen

Metzger

, et al. Physical activity and sedentary behavior patterns using accelerometry from a national sample of United States adults. Int J Behav Nutr Phys Act 2015; 12(1): 20.

Leech

McNaughton

Timperio

The clustering of diet, physical activity and sedentary behavior in children and adolescents: a review. Int J Behav Nutr Phys Act 2014; 11: 4.

Kebede

Howard

Ren

, et al. A systematic scoping review of latent class analysis applied to accelerometry-assessed physical activity and sedentary behavior. PLoS ONE 2024; 19(1): e0283884.

Evenson

Herring

Wen

Accelerometry-assessed latent class patterns of physical activity and sedentary behavior with mortality. Am J Prev Med 2017; 52(2): 135–143.

Jones

Wen

Herring

, et al. Correlates of US adult physical activity and sedentary behavior patterns. J Sci Med Sport 2016; 19(12): 1020–1027.

Metzger

Catellier

Evenson

, et al. Patterns of objectively measured physical activity in the United States. Med Sci Sports Exerc 2008; 40(4): 630–638.

Metzger

Catellier

Evenson

, et al. Associations between patterns of objectively measured physical activity and risk factors for the metabolic syndrome. Am J Health Promot 2010; 24(3): 161–169.

Jansen

van Kollenburg

Kamphuis

CBM

, et al. Hour-by-hour physical activity patterns of adults aged 45-65 years: a cross-sectional study. J Public Health 2018; 40(4): 787–796.

Howie

Smith

McVeigh

, et al. Accelerometer-derived activity phenotypes in young adults: a latent class analysis. Int J Behav Med 2018; 25(5): 558–568.

10.

Migueles

Lee

Sanchez

, et al. Revisiting the association of sedentary behavior and physical activity with all-cause mortality using a compositional approach: the Women’s Health Study. Int J Behav Nutr Phys Act 2021; 18(1): 104.

11.

LaCroix

Rillamas-Sun

Buchner

, et al. The objective physical activity and cardiovascular disease health in older women (OPACH) study. BMC Public Health 2017; 17(1): 192.

12.

Anderson

Manson

Wallace

, et al. Implementation of the women’s health initiative study design. Ann Epidemiol 2003; 13(9 Suppl.): S5–S17.

13.

Hays

Hunt

Hubbell

, et al. The women’ health initiative recruitment methods and results. Ann Epidemiol 2003; 13(9Suppl): S18–S77.

14.

Evenson

Bellettiere

Cuthbertson

, et al. Cohort profile: the women’s health accelerometry collaboration. BMJ Open 2021; 11(11): e052038.

15.

Rillamas -Sun

Buchner

, et al. Development and application of an automated algorithm to identify a window of consecutive days of accelerometer wear for large-scale studies. BMC Res Notes 2015; 8: 270.

16.

Choi

Liu

Matthews

, et al. Validation of accelerometer wear and nonwear time classification algorithm. Med Sci Sports Exerc 2011; 43(2): 357–364.

17.

Choi

Ward

Schnelle

, et al. Assessment of wear/nonwear time classification algorithms for triaxial accelerometer. Med Sci Sports Exerc 2012; 44(10): 2009–2016.

18.

Evenson

Wen

Herring

, et al. Calibrating physical activity intensity for hip-worn accelerometry in women >=60 years. Prev Med Rep 2015; 2: 750–756.

19.

Centers for Disease Control and Prevention. Defining adult overweight and obesity: adult body mass index, 2024, https://www.cdc.gov/obesity/index.html

20.

Evenson

Scherer

Peter

, et al. Historical development of accelerometry measures and methods for physical activity and sedentary behavior research worldwide: a scoping review of observational studies of adults. PLoS ONE 2022; 17(11): e0276890.

21.

Migueles

Cadenas-Sanchez

Ekelund

, et al. Accelerometer data collection and processing criteria to assess physical activity and other outcomes: a systematic review and practical considerations. Sports Med 2017; 47(9): 1821–1845.

22.

Collins

Lanza

Latent class and latent transition analysis: with applications in the social behavioral and health sciences. Hoboken, NJ: John Wiley & Sons, 2009.

23.

Weller

Bowen

Faubert

Latent class analysis: a guide to best practice. J Black Psych 2020; 46(4): 287–311.

24.

Muthén

(ed.) Mplus User’s Guide—Fifth Edition (1998-2007). Los Angeles CA: Muthén and Muthén, 2007.

25.

Gorman

Hanson

Yang

, et al. Accelerometry analysis of physical activity and sedentary behavior in older adults: a systematic review and data analysis. Eur Rev Aging Phys Act 2014; 11(1): 35–49.

26.

Harvey

Chastin

Skelton

DA.

How sedentary are older people? A systematic review of the amount of sedentary behavior. J Aging Phys Act 2015; 23(3): 471–487.

27.

de Rooij

van der Berg

van der Kallen

, et al. Physical activity and sedentary behavior in metabolically healthy versus unhealthy obese and non—obese individuals—The Maastricht study. PLoS ONE 2016; 11(5): e0154358.

28.

Matthews

Kozey Keadle

Moore

, et al. Measurement of active and sedentary behavior in context of large epidemiologic studies. Med Sci Sports Exerc 2018; 50(2): 266–276.

29.

Webster

Zhou

Gallagher

, et al. Device-measured sedentary behavior in oldest old adults: a systematic review and meta-analysis. Prev Med Rep 2021; 23: 101405.

30.

Arnardottir

Koster

Van Domelen

, et al. Objective measurements of daily physical activity patterns and sedentary behaviour in older adults: age gene/environment susceptibility-reykjavik study. Age Ageing 2013; 42(2): 222–229.

31.

Dooley

Pompeii

Palta

, et al. Daily and hourly patterns of physical activity and sedentary behavior of older adults: atherosclerosis risk in communities (ARIC) study. Prev Med Rep 2022; 28: 101859.

32.

Davis

Fox

KR.

Physical activity patterns assessed by accelerometry in older people. Eur J Appl Physiol 2007; 100(5): 581–589.

33.

Sartini

Wannamethee

Iliffe

, et al. Diurnal patterns of objectively measured physical activity and sedentary behaviour in older men. BMC Public Health 2015; 15: 609.

34.

Yerrakalva

Cooper

Westgate

, et al. The descriptive epidemiology of the diurnal profile of bouts and breaks in sedentary time in older English adults. Int J Epidemiol 2017; 46(6): 1871–1881.

35.

Davis

Fox

Hillsdon

, et al. Objectively measured physical activity in a diverse sample of older urban UK adults. Med Sci Sports Exerc 2011; 43(4): 647–654.

36.

Doherty

Jackson

Hammerla

, et al. Large scale population assessment of physical activity using wrist worn accelerometers: the UK Biobank Study. PLoS ONE 2017; 12(2): e0169649.

37.

Wennman

Pietila

Rissanen

, et al. Gender, age and socioeconomic variation in 24-hour physical activity by wrist-worn accelerometers: the FinHealth 2017 Survey. Sci Rep 2019; 9(1): 6534.

38.

Chomistek

Shiroma

Lee

IM.

The relationship between time of day of physical activity and obesity in older women. J Phys Act Health 2016; 13(4): 416–418.

39.

Rosenberger

Fulton

Buman

, et al. The 24-hour activity cycle: a new paradigm for physical activity. Med Sci Sports Exerc 2019; 51(3): 454–464.

40.

Henry

Muthen

Multilevel latent class analysis: an application of adolescent smoking typologies with individual and contextual predictors. Struct Equ Modeling 2010; 17(2): 193–215.

41.

Keadle

Shiroma

Kamada

, et al. Reproducibility of accelerometer-assessed physical activity and sedentary time. Am J Prev Med 2017; 52(4): 541–548.

42.

Nguyen

Herting

Kohen

, et al. Recreational physical activity in postmenopausal women is stable over 8 years of follow-up. J Phys Act Health 2013; 10(5): 656–668.

43.

Elliott

Zhao

Mukherjee

, et al. Methods to account for uncertainty in latent class assignments when using latent classes as predictors in regression models, with application to acculturation strategy measures. Epidemiology 2020; 31(2): 194–204.

44.

Kamata

Kara

Patarapichayatham

, et al. Evaluation of analysis approaches for latent class analysis with auxiliary linear growth model. Front Psychol 2018; 9: 130.

45.

Vermunt

Latent class modeling wiht covariates: two improved three-step approaches. Political Analysis 2010; 18(4): 450–469.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.36 MB