A cross-sectional study of physical activity and sedentary behaviours in a Caribbean population: combining objective and questionnaire data to guide future interventions

Study population

The Health of the Nation (HotN) survey was a cross-sectional survey of the Barbadian population aged 25 years and over, which aimed to provide estimates of the prevalence and social distribution of NCD risk factors. Data were collected for 1234 participants on health conditions and behaviours (including questionnaire-assessed PA), as well as sociodemographic information. The survey has been described in detail elsewhere [18]. An age-restricted sub-sample (25–54 years; n = 527) of participants who completed HotN was selected for objective PA measurement. We decided to focus limited resources for PA measurement on this younger age group, because of the opportunity to intervene and change behaviour before the majority of NCDs develop.

Written informed consent was obtained from all participants. The study was approved by the Research Ethics Committee of the University of the West Indies, Cave Hill/Barbados Ministry of Health.

Assessment of physical activity

Perception of activity

Prior to the objective assessment, participants were asked to rate their own activity as either ‘extremely active’ , ‘moderately active’ or ‘not very active’. Perceived activity was compared to the objective measurements. We classified ‘overestimators’ as those who considered themselves extremely or moderately active, but were classified objectively as inactive according to WHO guidelines.

Assessment of other risk factor and sociodemographic variables

Details of the methods for assessing risk factors and sociodemographic characteristics in the HotN survey have been published elsewhere [18]. Body mass index (BMI; kg/m²) was derived from height and weight measured using standard protocols and categorised as normal weight (<25 kg/m²), overweight (25–30 kg/m²), and obese (>30 kg/m²).

Maximum level of educational attainment and occupational grade were used in this analysis as indicators of socioeconomic position (SEP). Education was grouped into four levels as follows: level 1 had not completed secondary school; level 2 completed secondary school; level 3 had technical, trade or teacher education; and level 4 had university education (undergraduate and postgraduate). Occupation was collected as free text and then coded using the Barbados Standard Occupational Classification (BARSOC-89) [31], which is based on the 1988 International Standard Classification of Occupations (ISCO-88) [32]. BARSOC-89 contains nine major groups, which were collapsed to create three broad occupational categories: group 1 (routine/manual) consisted of skilled agricultural, craft/elementary workers, and machine operators; group 2 (intermediate) comprised technical, clerical and service employees; and group 3 (professional) consisted of managers and professionals.

Statistical analyses

Descriptive statistics for the continuous PA variables (PAEE, sedentary time, time spent in LPA, time spent in MVPA) are presented as median and interquartile range (IQR) due to their skewed distribution.

The prevalence of physical inactivity (i.e. not meeting WHO minimum recommendations) and sedentary lifestyle (at least 8 h per day sedentary, excluding sleep) are presented as proportions and 95 %CIs. The difference in prevalence of physical inactivity between the self-report and the objective measure is presented with 95 % confidence intervals (CIs), and with the difference formally examined using McNemar’s test for paired data.

Associations between PA components and SEP indicators (education level and occupational grade) were examined using multivariable regression models. Due to the skewed distribution of the PA components, median regression was used, as recommended by McGreevy et al. [33].

Data were analysed using the Stata software package (version 13, StataCorp, College Station, Texas). All analyses were weighted to account for the sampling design and non-response, and to match the age and sex distribution of the Barbadian population according to the 2010 Barbados Population and Housing Census [14].

Participant characteristics

Demographic, socioeconomic, and anthropometric characteristics of the complete HotN sample (aged 25–54; n = 704) are shown in Additional file 3: Table S1. There were no significant differences (with all 95 % CIs overlapping) in these characteristics between the HotN sample and the sample that completed the PA assessment.

Objectively assessed physical activity and sedentary behaviour

The prevalence of objectively assessed inactivity and sedentary lifestyle are shown in Table 2. According to objective data, more women than men were inactive: 90.5 % (83.3,94.7) vs. 58.9 % (48.4,68.7). In contrast, there was not a significant difference in sedentary lifestyle between men and women, with the prevalence in both sexes combined being 50.7 % (95 % CI: 43.3,58.1).

Comparison of objective and self-reported estimates of physical activity

The prevalence of PA overestimation was similar in men and women (Table 2). Overall, 60.1 % (95 % CI: 52.8,66.9) overestimated their activity.

For both genders, estimates of physical inactivity were underestimated by self-report compared with objective measures (Figs. 2 and 3; Additional file 3: Table S2). The prevalence (95 % CI) of self-reported inactivity was 64.9 % (56.3,72.7) in women and 27.7 % (17.6,40.6) in men. Physical inactivity was underestimated by the RPAQ by 25 percentage points (pp) in women (p value for difference: 0.001) and 31 pp (p value: 0.001) in men. The size of the difference between self-reported and objectively assessed inactivity tended to vary by sociodemographic characteristic, with statistically significant differences only in the older age groups, the more educated, those with non-manual occupations, and overweight/obese groups.

Social distribution of physical activity using objective and self-reported measures

Using self-reported data, total PAEE was not associated with education, but domain-specific variation of energy expenditure by education was observed (data not shown in table). Education was found to be directly associated with self-reported leisure-time PAEE and inversely associated with self-reported occupational PAEE. For every increase in educational level, reported leisure-time PAEE was higher by 3.85 (95 % CI: 1.1,6.5) kJ/kg/day, and reported occupational PA was lower by 1.7 (0.1,3.3) kJ/kg/day.

Self-reported measures: context and type of activity

The relative contribution of energy expenditure in different domains to total PAEE was examined using self-reported data (Additional file 3: Table S3), excluding unemployed participants (n = 78). Men and women had similar patterns of expenditure by domain. Occupational, leisure, home, and travel activity accounted for 67 %, 21 %, 18 %, and 4 % of total PAEE, respectively. Including unemployed participants did not change the pattern of domain-specific contributions to overall PAEE: occupational PAEE still made the largest contribution (45 % overall).

Strengths and limitations

The main strength of our study was the use of objective and self-report measures of physical activity to provide complementary information in a nationally representative sample of a developing country population, where little is known about these behaviours, nationally and regionally. In this paper we provide the most comprehensive assessment of physical activity patterns in an adult, Caribbean population. However, before we consider the interpretation and importance of the findings it is important to acknowledge the limitations of the study.

We assessed adherence to the WHO recommendations for physical activity [29], which we interpreted as at least 150 min of MVPA each week. However, the guidelines also specify that muscle-strengthening activities should be carried out on at least 2 days a week. We were unable to assess this with the objective measure, and therefore did not consider it for either measure to ensure comparability. Our estimates of the prevalence of inactivity are therefore likely to be conservative.

To assign energy costs to the activities reported in the questionnaire, we used a published physical activity compendium [19], which does not take into account individual variation and variation between populations. We also assigned assumed energy costs for types of occupation (sedentary, standing, manual, and heavy manual), and these assumptions were based on measurements made in European populations. It is not clear whether average occupational energy expenditure differs in our population.

We examined domain-specific contribution to overall energy expenditure using RPAQ-derived measures. However, it is not clear whether the bias with which activity is reported differs by domain. Our assessment assumes equal bias, and could therefore be inaccurate.

Finally, we compared the prevalence of inactivity derived from an objective measure to that derived from a questionnaire. These assessments were made at different points in time, with a median of 114 days between them. However, it is unlikely that this gap affected the conclusions drawn from this comparison, which was made to determine the public health implications of using self-reported measures of inactivity at a population level. Although individual changes in activity may have occurred in the period between measurements, it is doubtful that there would have been meaningful changes in population levels of activity.

Comparison with other studies

Our estimates of physical inactivity were high, but comparison with other populations is limited by different age profiles between populations and the application of different definitions of inactivity, as well as the use of different measurement techniques. In the US, the objective prevalence of inactivity was estimated as greater than 95 % overall [7], but this study used accelerometry and examined a different age range (20 years and over). Furthermore, the definition of inactivity used was based on a different implementation of the PA guidelines, with participants needing to accumulate 30 min of activity on at least 5 of 7 days of measurement [7]. As Thompson et al. [34] emphasize, even small changes in how inactivity is defined results in large variation in activity status. In the UK, 96 % of women and 94 % of men over the age of 16 years did not achieve the government’s recommended physical activity level, as assessed by accelerometry [35]. Again, this study examined a different age range, and used a different measurement technique. These important caveats aside, the level of inactivity observed in our population for women (90.5 %) was in a similar range to those reported for the US and UK, while the level of inactivity amongst men (58.5 %) was substantially lower. Estimates of population levels of inactivity in developing countries are based largely on questionnaires [6], and this limited our ability to draw meaningful comparisons in more similar settings. A study from Cameroon, also using combined heart rate and movement sensing, reported estimates of PAEE for rural and urban populations aged 25 to 55 years [36]. Mean PAEE in urban women and men was 37.9 and 51.5 kJ/kg/day, respectively, and 54.3 and 64.6 kJ/kg/day in rural women and men, respectively. Our estimates of PAEE in Barbadian women and men (36.4 and 47.2 kJ/kg/day, respectively) were similar to urban Cameroonian estimates, and less than rural Cameroonian estimates. In Kenya, similar methodology was used to measure PA in three rural populations [37]. The PAEE reported was substantially higher in all of these populations compared to our study, with the lowest (Luo population) being 58.9 and 74.4 kJ/kg/day in women and men, respectively.

In terms of age and gender patterns of PA, our data are consistent with a recent global review, which found that, on the whole, men are more physically active than women, and older people are less active than younger people [6]. Much less is known about the distribution of PA by SEP, even in developed countries. A recent European systematic review did not find consistent associations between total self-reported PA and SEP. [38] However, domain-specific gradients were reported: higher SEP was generally associated with more leisure-time activity and less occupational activity. The authors suggest that these findings demonstrate complex patterns of socioeconomic inequalities in physical activity, and that total activity may not be a suitable summary measure when investigating inequalities and how this affects morbidity and mortality. Our results imply that this complexity may be attributable to the use of questionnaires to measure activity, rather than a lack of social gradient for overall activity. Similar to European data, we observed a lack of association between overall self-reported total activity and education, whilst domain-specific associations exist in opposite directions for leisure-time and occupational activity. However, we also show that a clear social gradient in overall activity can be demonstrated when objective measures are used. The utility of questionnaires for describing social patterns of PAEE may therefore be limited, possibly due to social desirability bias.

Our finding that subjective methods substantially underestimate physical inactivity is consistent with results from the US [7] and the UK [8], and reinforces the need to interpret quantitative findings based on self-reported PA with care. A further note of caution comes from the difference in accuracy of PA reporting by age, SEP, and BMI category. Thus, while over-reporting of physical activity was apparent (based on the point estimates) in all the subgroups we examined (see Figs. 1 and 2) it was only statistically significant in in older age groups, the more educated, those with non-manual occupations, and overweight/obese groups. Over-reporting of PA by BMI category has been documented previously [39, 40]. Our results further underscore the need for objective measures to be used when investigating the relationship between physical activity and health outcomes that are related to BMI, as well as other demographic and socioeconomic characteristics. Another potential limitation of using questionnaires for PA surveillance is the possibility that objective and self-report measures differentially track trends over time, with people becoming more accustomed to giving socially desirable answers. If this is the case, trends in self-reported PA may represent bias in how the questionnaires are completed. Although this area remains largely unstudied, Cleland et al. [41] found that the Global Physical Activity Questionnaire (GPAQ) provided a valid measure of change in reported activity over time, compared with accelerometry. However, this study only repeated the measurements once, with a relatively short interval (3–6 months). It is therefore unlikely to represent a typical surveillance scenario, where large population-based surveys are repeated over years, with results publicised between rounds. Further studies that address this concern would help to fully determine the implications of using questionnaires for PA surveillance.

Potential public health implications

This study demonstrates how combining objective and self-reported measures of PA and sedentary behaviours can provide useful information for guiding interventions in developing countries. The high prevalence of physical inactivity and having a sedentary lifestyle underscore the need for population-wide public health intervention. However, given the limited resources in this setting, a more pragmatic approach may be to target groups with particularly low activity. Our data highlight that women and those with a high SEP have particularly low levels of activity, and these needs should be considered when interventions are designed. A useful direction for future studies would be to utilise qualitative methodology to investigate how public health campaigns could effectively target the low-activity groups that we have identified. Alvarado et al. [42] have identified barriers to PA in young, overweight and obese Barbadian women, and have made recommendations on how activity could be facilitated in this group. Similar studies focusing on other low-activity groups are warranted.

In this population, 60 % of individuals overestimate their activity. This group is of particular public health importance, as people who believe they are sufficiently active are unlikely to see the need to increase their activity. The prevalence of PA overestimation in Barbados is higher than reported in a UK population, (46 %) [43], but similar to that reported in the Netherlands (61 %) [44]. Public education to improve awareness of PA levels should be considered an integral part of future efforts to increase activity.

Occupational PA is the main contributor to overall PAEE in this population, although this is to an extent driven by the assigned energy cost to the different occupations, combined perhaps with an underrepresentation of activities of daily living. Whether occupational PA is as beneficial for health as leisure PA is unclear [45], so this pattern may suggest that the energy is not being expended in a manner that optimises its impact on health. Active transport makes little to no contribution to overall PAEE (4 % for women and men). Anecdotally, a hot and humid climate, lack of changing facilities at work, an infra-structure of narrow roads with limited sidewalks, and a strong social preference for personal motorised transport are all cited as barriers to increasing levels of active transport in Barbados and similar Caribbean countries. Well designed, including qualitative, studies to properly investigate barriers to active transport and the feasibility of effectively promoting it are needed. A more promising focus for future interventions might be to encourage more leisure time activity. In other populations, higher socioeconomic groups tend to participate disproportionately more in leisure activities, compared with lower socioeconomic groups [38]. Encouraging leisure activities in these populations therefore has the potential to exacerbate social inequalities in physical activity. In the Barbadian population, however, we have demonstrated that physical inactivity is higher in those with a higher SEP, at least according to education and occupation. Population increases in leisure activity may therefore reduce PA inequality. Examining participation in different types of leisure activities highlights those that could be most effectively promoted. Walking was the most popular leisure activity, and interventions to increase population levels of walking have been successfully implemented in many populations [46]. Similar approaches could be adopted in our setting. An alternative strategy could be to identify gaps, and to promote currently uncommon activities. For example, on average, Barbadian women participate very little in team sports and racquet sports, with a mean time reported per week of less than 2 min. Efforts could be made to encourage participation of women in sports, perhaps starting with school-age girls. Further studies would be necessary to determine the how effective this approach is likely to be.

This study adds to a growing body of evidence that highlights the limitations of self-reported measures to assess inactivity prevalence. Despite this, questionnaires such as the GPAQ continue to be used in developing countries, and are widely integrated into national surveillance systems as the only instrument [47]. However, with objective methods becoming cheaper and more feasible to apply on scale, this may well change in the future. A stated target of WHO’s global action plan on non-communicable diseases is to reduce physical inactivity from a 2010 baseline by 10 % by 2025 [2]. We suggest that questionnaires on their own are not sufficiently valid for use in NCD risk factor surveillance systems, and their continued use in isolation impedes accurate evaluation of this important target. We emphasize the urgent need for cheaper and simpler objective methods for physical activity surveillance to be developed and implemented.