Accelerometer-measured physical activity is not associated with two-year weight change in African-origin adults from five diverse populations

Sampling design and participant recruitment

Twenty-five hundred adults, ages 25–45, were enrolled in METS between January 2010 and December 2011. After the baseline exam, participants were asked to return to the clinic yearly for an additional two clinic exams: one-year follow-up and two-year follow-up. Participants were excluded if they self-reported that they were either HIV positive or pregnant at the time of recruitment. During the follow-up period, if a participant was pregnant or lactating, their weight data was not recorded during that year’s follow-up visit; only the health history was recorded. Participants were also asked about voluntary weight loss and other recreational drug use. One person was excluded from the USA cohort for intentionally losing 36 kg. For the current analysis, follow-up time varied between 1.32 years in Jamaican men to 1.98 years in Seychellois women (Table 1). Consequently, all analyses were adjusted for differing follow-up times and as described in the statistics section. A detailed description of the study protocol has been previously published (Luke et al., 2012). In brief, participants (approximately 50% of whom were female) were enrolled in each of five study sites: rural Ghana, urban South Africa, the Seychelles, urban Jamaica and metropolitan USA (Chicago). The participants were predominantly of African descent from each of the five countries. Study sites were selected to represent a broad range of social and economic development as defined by the UN HDI 2011: i.e., Ghana as a low-middle HDI country (HDI rank 135), South Africa as middle (123), Jamaica (80) and the Seychelles as high (52), and the US as a very high HDI country (4).

METS was approved by the Institutional Review Board (IRB) of Loyola University Chicago, IL, USA; the Committee on Human Research Publication and Ethics of Kwame Nkrumah University of Science and Technology, Kumasi, Ghana; the Health Sciences Faculty Research Ethics Committee of the University of Cape Town, South Africa; the Board for Ethics and Clinical Research of the University of Lausanne, Switzerland; the Research Ethics Committee of the Ministry of Health in Seychelles; the Ethics Committee of the University of the West Indies, Kingston, Jamaica; and the Health Sciences IRB of the University of Wisconsin, Madison, WI, USA. Written informed consent was obtained from all participants.

Measurements

Yearly measurements were made at outpatient clinics located in the respective communities. Weight and height measurements were made on all participants while wearing light clothing and no shoes and using the same model of equipment, which was calibrated using the same protocol at each locale. Weight was measured to the nearest 0.1 kg using the same standard balance at all five sites and during the entire study period (Seca 770; Seca, Hamburg, Germany). Height was measured to the nearest 0.1 cm using a stadiometer (e.g., Invicta Stadiometer; Invicta, London, UK). Waist circumference was measured to the nearest 0.1 cm at the umbilicus and hip at the point of maximum extension of the buttocks. Body mass index (BMI) was calculated as kg/m².

Body composition was estimated by bioelectrical impedance analysis (BIA) with the use of single-frequency (50 kHz) impedance analyzer (model BIA 101Q; RJL Systems, Clinton Township, MI). Fat-free mass (FFM) and fat mass (FM) were estimated from measured resistance by using an equation validated in the METS cohorts (Luke et al., 1997).

PA was assessed using Actical accelerometers (Phillips Respironics, Bend, OR, USA) as previously described (Dugas et al., 2014; Luke et al., 2014). Briefly, the monitor was worn at the waist, over an 8-day period encompassing the partial first and last days, including during sleep. We assessed activity conducted between the hours of 7 am–11 pm daily. Raw data downloaded from the accelerometers were first passed through a National Cancer Institute (2014) designed to infer non-wear time from 90 or more minutes of continuous zero activity counts. A valid day of PA monitoring was defined as one having 10 or more hours of wear time and participant files were included if they contained ≥4 or more valid days. Using the same protocol employed by the National Center for Health Statistics for the analysis of accelerometry data in the continuous National Health and Nutrition Examination Survey (Troiano et al., 2008), minutes defined as comprising sedentary, moderate, vigorous or moderate plus vigorous activity are presented as the total time in minutes (min) per day accumulated in either 1- or 10-min intervals and this was averaged for the number of days with valid accelerometry data.

Statistical analysis

Data management was centralized at Loyola University Chicago. All data forms and questionnaires were scanned at each study site and, along with electronic Actical data files, were sent via secure FTP (Bitvise Tunnelier, Version 4.40, http://www.bitvise.com/tunnelier) to the data manager at the coordinating center.

Descriptive statistics were used to summarize the characteristics of participants in each of the five study sites, including descriptive characteristics of the physical activity variables. Because the yearly follow up periods were not identical in the different sites or within sites, all analyses on weight change were adjusted for the different follow-up periods. Univariate analyses were conducted to determine partial correlation coefficients between parameters of PA and two-year weight change, after adjustment for age and follow up duration, and accounting for multiple observations for each individual (Hill, Peters & Blair, 2015). Partial correlation coefficients between weight change and PA are also presented separately among persons who gained weight and among persons who lost weight during follow up. For the purpose of our analysis, yearly weight change (kg/yr) was defined as the difference between the first and last recorded weight measure over the follow-up period, while weight loss and weight gain were defined simply by whether the final weight measurement either exceeded or was lower than the baseline starting weight (Table 2).

Multilevel models

Because the follow-up time was different between participants and sites, and some participants only had one follow-up measurement, we used the multi-level random slope and intercept model, which allowed us to use all available data points, to examine the associations between weight PA variables, including sedentary time. Briefly, for Level 1; an individual’s weight (kg) from each examination was regressed on follow-up time. Secondly, level 2 models the potential dependence of the individual baseline weight (random intercept from level 1) and rate of weight change (kg/yr, random slope of level 1) as a function of the other covariates. The multi-level model is as follows: (1)${\displaystyle \mathrm{Level1}:{Weight}_{ij}={\beta }_{0i}+{\beta }_{1i}{T}_{ij}+{\epsilon }_{ij}}$ (2)${\displaystyle \mathrm{Level2}:{\beta }_{0i}={\alpha }_0+{\alpha }_1{BaseWeight}_i+{\alpha }_2{Age}_i+{\alpha }_3{Gender}_i+{\alpha }_4{Site}_i+{\alpha }_5{PhysicalActivity}_i+{\xi }_{0i}}$ (3)${\displaystyle {\beta }_{1i}={\gamma }_0+{\gamma }_1{ghana}_i+{\gamma }_2{rsa}_i+{\gamma }_3{jamaica}_i+{\gamma }_4{seychelles}_i+{\gamma }_5{PhysicalActivity}_i+{\xi }_{1i},}$

where i = 1, 2…N and j = 1, 2. The effect of the PA variables on weight change is measured by the interaction coefficient γ₅, and the average effect of the PA variables on weight change estimate by α₅. We have previously used this multi-level approach (Luke et al., 2009), which utilizes all available data, including incomplete data, and the within- and between-person variability, while adjusting for potential confounders such as age and baseline weight. All analyses were conducted using SAS (SAS Institute, Cary, NC USA, version 9.4) and a conventional two-sided 5% alpha error was used to denote statistical significance.

Subject characteristics

Baseline measures where completed in 2010 and 2011 and follow-up yearly exams during 2011/2012 (follow-up 1) and 2012/2013 (follow-up 2). At baseline, a total of 2,506 participants were enrolled from the 5 sites over the enrollment period. Two thousand and sixty eight participants returned for a 1 year follow-up, Ghana (n = 417), South Africa (n = 414), Jamaica (n = 399), Seychelles (n = 481) and the USA (n = 357). While a total of 1,408 participants returned for a second follow-up visit, Ghana (n = 289), South Africa (n = 322), Jamaica (n = 136), Seychelles (n = 456) and finally, the USA (n = 205). The final sample therefore included a total of 888 men and 1,056 women with at least complete baseline measurements and 1 follow-up clinic visit (Table 1). We did exclude one participant from the USA who intentionally lost 36 kg after the baseline visit.

The mean age among the men and women were 34.9 ± 6.16 yr and ranged from 33.7 ± 5.6 yr in the South African men to 36.6 ± 5.1 yr in the Seychellois men. Among men and women, mean time between the baseline visit and 2year follow-up visit ranged from 1.32 ± 0.46 years among the Jamaican men and 1.98 ± 0.27 years among the Seychellois women. As a result, all of our statistical modeling was adjusted for both age and follow-up time differences.

The mean baseline weights were different among the 5 sites, with the lowest weights being captured among men from Ghana and South Africa (63.4 ± 9.5 and 64.9 ± 11.8 kg, respectively) and highest among the USA men (93.6 ± 25.9 kg). Among the women, Ghanaians weighed less than USA women (63.3 ± 12.7 vs. 91.7 ± 23.4 kg). Figures 1A and 1B present the weight change and PA data by site. After the two-year follow-up period, surprisingly USA men experienced the smallest yearly weight change 0.05 ± 3.27 kg/yr, while among the women, the Jamaicans actually lost on average −003 ± 2.79 kg/yr. In both men and women, yearly weight change was always greater among participants categorized as normal weight (BMI < 25 kg/m²) at baseline. For example, normal weight men from Seychellois and Ghana experienced a yearly weight gain between 0.56–0.58 kg/yr, respectively, compared to their obese counterparts who actually experienced a yearly weight loss between −0.28–2.06 kg/yr, respectively (Table 2). Similarly, among the women, yearly weight gain was greatest among the normal and overweight participants, than among the obese participants, ranging from 0.23 ± 1.95 kg/yr in the Jamaicans up to 1.37 ± 2.53 kg/yr among the Ghanaian women.

Table 3 presents the absolute weight change data across the entire follow-up period, by site and sex, stratifying for either overall weight gain or weight loss. The number of men gaining weight (end weight > start weight) ranged from 40% among the Jamaicans to 54% among the Seychellois. The absolute weight gain varied from 3.2 ± 2.4 kg to 4.5 ± 3.5 kg. Among the women, the number of participants gaining weight varied widely from 44% of the Jamaicans, who gained on average 3.1 ± 1.9 kg up to 77% of the South African women, who gained 6.4 ± 5.3 kg.

Accelerometer PA at baseline, measured as either 10-min or 1-min bouts (Table 1), was greatest among the South African men, who accumulated almost 1 hr/day of MPVA in 1-min bouts and among the women, greatest in the Ghanaian women, who accumulated almost 30 min/day in 1-min MVPA bouts. Sedentary activity, also measured as either 10-min or 1-min bouts, was greatest among the Jamaican men (266.9 ± 60.2 min/day in 1-min bouts) and highest among the South African women (247.2 ± 44.7 min/day in 1-min bouts).

We categorized participants at each site by whether they met the USA Surgeon General guidelines for PA, i.e., 30 min/day on most days of the week, using the 1-min MVPA data. Among the men, over 76% of Ghanaians (N = 158) and South Africans (N = 183) met the guidelines, compared to only 44% of USA men (N = 109). Far fewer women met this guideline, with only 44% of Ghanaian women (N = 130) compared to about 20% of USA women (N = 50).

Surprisingly, the total weight gain at every site, were actually greater among participants who met the PA guidelines (Table 2) than among their site counterparts who did not meet the PA guidelines. For example among USA men who met the PA guidelines, yearly weight change was 0.25 ± 5.05 kg/y compared to a loss of 0.29 ± 4.09 kg/yr among those not meeting the guideline. Similarly, among the women, at each site, weight gain was less among the participants not meeting the PA guidelines, for example among South Africa women meeting the PA guidelines the yearly weight gain was 1.8 ± 3.1 kg/yr vs. 0.7 ± 3.5 kg/yr. USA women were the only exception, where women meeting the guideline lost 0.4 ± 5.1 kg/yr, compared to an average weight increase of 0.1 ± 4.1 kg/yr in those failing to meet the guideline.

At baseline, overweight Ghanaians and South Africans were 0.49 (95%CI [0.32–0.75], p = 0.001) and 0.48 (95% CI [0.29–0.77], p = 0.003) times as likely to meet the PA guidelines at baseline compared to normal weight participants. Among the obese, Ghanaians, South Africans and USA participants were 71%, 78% and 62% less likely to meet the PA guidelines (p < 0.001 for all) at baseline.

Relationship between weight change, and baseline physical activity and sedentary behavior

Tables 4A–4C and Tables 5A–5C present the partial correlation coefficients between weight change and MVPA (Tables 4A–4C), and sedentary behavior (Tables 5A–5C). Because overall 10-min bout MVPA was so low, we chose to model our exploratory statistics using 1-min bout MVPA only. We adjusted all analyses for age and length of follow-up and restricted the analyses to each site to account for differences in site weight change.

Among the men, overall weight change did not correlate with 1-min bout MVPA at any of the sites. Because this was not the expected relationship, we also performed this analysis on only the participants who either gained or lost weight, but this also did not produce any statistically significant associations. For example, among the participants who lost weight, only women from the Seychelles had a significant relationship for 1-min bout MVPA and weight loss (r = 0.326, p = 0.015), while among participants who gained weight, the only significant relationship between PA and weight gain was among the USA women who actually had a positive result (r = 0.218, p = 0.026), indicating higher PA levels resulted in greater weight gain.

Sedentary behavior, in recent years, has been thought to be a key component contributing to an increased risk for overall morbidity and mortality, independent of daily PA. Therefore, we also examined the relationship between sedentary time at baseline and prospective weight change during the two-year follow up period (Tables 5A–5C). Weight change was negatively associated with sedentary time at baseline, only among the Ghanaian women (r =  − 0.147, p = 0.029), i.e., weight gain was less among those with greater sedentary time at baseline. We did not find any significant relationships between either weight gain or weight loss and baseline sedentary time among any of the sites (Figs. 2A and 2B). Furthermore, we found that the nature of the relationship between the two variables was inconsistent, with some relationships being positive, while others negative.

Multi-level models

Table 6 presents the parameter estimates for weight change using baseline MVPA (model 1), baseline light activity (model 2) and baseline sedentary activity (model 3), all based on 1-min bouts, and with analyses adjusted for follow up time and age. In model 1, it can be seen that baseline MVPA is not significantly associated with prospective weight, only the interaction term between follow-up time and the South African site, where each year of follow-up time is associated with an increase in weight that is on average 0.96 kg greater than the USA (p = 0.02). This outcome was similarly found for model 2 (light activity and model 3, sedentary activity. In addition, among the 3 models; baseline weight, age and gender were all significantly associated with weight.