All adults who had sex, age and body mass index (BMI) data across two surveys conducted in the Dikgale Health and Demographic Surveillance System site (DHDSS)  in 1997  and 2005-7 , were included in the analysis (male: n=15; female: n=128, ≥30 years in 1997). Only the 2005-7 survey included an objective measure of PA. The methodology behind these cross-sectional survey data is described in detail elsewhere [14,15].
Given the same age ranges of the sample analysed (n=143), there were no significant associations between those included/not included in the analysis for age-, BMI- or educational status distribution (p≥0.1179), nor ambulation (p=0.7438). However, proportionally, significantly fewer males were used in the analysis, compared with those not used in the analysis (p≤0.0006).
Using body mass (kg) and stature (m), BMI (kg.m-2) was calculated and classified; under-weight (UW, <18.5 kg.m-2), normal weight (NW, 18.5-24.9 kg.m-2), over-weight (OW, 25-29.9 kg.m-2), obese (OB, 30-34.9 kg.m-2) and severely obese (≥35 kg.m-2) . In addition, three weight-change categories were constructed based on BMI changes over the approximately 10-year period (1997 to 2005-7); weight-loss, -gain or -stability. Due to sample size constraints, BMI change categories were grouped. UW (N=8) did not differ significantly from NW (N=37) for age or average steps.day-1 (p>0.9) and were collapsed into one group. The weight-change categories were defined as follows:
- Weight-loss: OW/OB → UW/NW (≥25 kg.m-2 → <25 kg.m-2)
- Weight-gain: UW/NW → OW/OB (<25 kg.m-2 → ≥25 kg.m-2)
- Weight stability: UW/NW → UW/NW (<25 kg.m-2) and OW/OB → OW/OB (≥25 kg.m-2)
With regard to the 2005-7 survey data, 7-day accelerometry-based pedometry data were collected using electronic pedometers (NL-2000, New Lifestyles Inc., Kansas City, MO, USA) . Step-based PA public health indices were defined as: sedentary: <5 000 steps.day-1, low-somewhat active: 5 000 - 9 999 steps.day-1, active: 10 000 – 12 499 steps.day-1, very active: ≥12 500 steps.day-1. A pedometry-based approach was used to estimate the degree to which participants met energy expenditure-based PA public health guidelines . Using daily (kcal.kg-1.day-1) and total weekly AEE (kcal.kg-1.wk-1) the following categories were determined:
- ≥7.5 kcal.kg-1.wk-1, ≥1.5 kcal.kg-1.day-1 for ≥5 days.wk-1
- ≥21 kcal.kg-1.wk-1, ≥3 kcal.kg-1.day-1 for 7 days.wk-1
For the purposes of this analysis a 150- and 420 min.wk-1 standard were used, which equates to ≥7.5 kcal.kg-1.wk-1 and ≥21 kcal.kg-1.wk-1, respectively [12,20].
Descriptive statistics comprised means (one standard deviation) and proportions.
Relationships between categorical variables and differences across multiple group proportions were examined through Fisher’s exact test and z tests with correction for multiple comparisons (Bonferroni).
For continuous data, independent and one sample t tests examined differences between the sexes and combined data, respectively. One-way Analysis of Variance examined differences across weight-change categories, with post hoc multiple comparison analyses (Sidak’s t test) assessing group differences.
To examine average daily step totals across weight-change categories, a Univariate General Linear Model was constructed, adjusting for 2005-7 survey age. Post hoc multiple comparison analyses (Sidak’s t test) assessed group differences.
Two linear regression models were examined for BMI delta (BMI 2005-7 survey minus BMI 1997 survey, kg.m-2) - Model 1: age, sex and average daily steps; Model 2: age, sex and average daily AEE. Age and PA variables were obtained from the 2005-7 survey.
Data were analysed using appropriate statistical software (IBM SPSS Statistics: Release 25 IBM Corporation, Armonk NY, 2017 and GraphPad Prism: version 8.12, GraphPad Software, La Jolla CA, 2019). Significance for all inferential statistics was set at p< 0.05.