Validity of the global physical activity questionnaire (GPAQ) in Bangladesh

Physical activity (PA) is a key behavioural factor for maintaining health and well-being at individual and population levels [1‐3]. It has been estimated that at least 9% of premature mortality globally could be avoided if everyone adhered to the WHO physical activity guidelines [1]. Furthermore, in 2013 the economic cost to health care systems worldwide related to non-adherence was estimated at $53 billion [4]. The World Health Organization (WHO) has therefore promoted the development of PA surveillance tools in order to evaluate public health interventions and policies [5, 6] aimed at reducing the burden of non-communicable diseases (NCDs) [7].

The Global Physical Activity Questionnaire (GPAQ) is one such instrument that was endorsed by the WHO for its STEPwise Approach to Chronic Disease Risk Factor Surveillance (STEPS) [8‐11]. The GPAQ was developed with special consideration of key physical activity domains in developing countries and of a length and complexity suitable for inclusion in STEPS [11].

Although GPAQ has been widely used for monitoring PA, the evidence base for its validity is limited. The most extensive study to date assessing the validity of the GPAQ was conducted in 2003–2005 in nine countries including Bangladesh [12]. However, since then an updated measure was released by the STEP wise program and evidence for the validity of the new version is still limited. Furthermore, six of these eight countries, including Bangladesh, used pedometers, a criterion measure which is not sensitive to activity intensity, and only four, of which Bangladesh was not one, included rural populations in their sample. The criterion validity and reliability of GPAQ for urban Bangladesh was found 0.06 and range 0.31–0.72, respectively [12]. Given there are substantial differences in patterns and frequency of PA between rural and urban populations [13‐16], it is yet to be determined if GPAQ is an appropriate instrument to assess the status of PA among rural populations [12].

As a part of the surveillance system, nationwide surveys of NCD risk factors following the WHO STEPS strategy are being conducting periodically in Bangladesh. Comparison of three STEPS surveys’ results (2006, 2010 & 2013) of Bangladesh clearly indicated that NCDs, particularly diabetes, are increasing [17]. A comparison of the prevalence rates indicates that diabetes is becoming as problematic in rural as in urban populations [17‐19]. Bangladesh is facing an escalating rise of NCDs and the validity of GPAQ for the WHO STEPS in Bangladesh needs to be established for the entire population. As the Bangladeshi sample in the validation of the previous version of the GPAQ was urban, validated against pedometers and showed very poor results, this study aimed to determine the criterion validity of the new version of GPAQ in both rural and urban populations using accelerometer as the criterion measure of physical activity.

This study was approved by Western Sydney University Human Research Ethics Committee (HREC # H11145) and Bangladesh University of Health Sciences Ethical Review Committee.

The characteristics of the 162 study participants are described in Table 1. Fifty-four percent were female, the overall mean age was 35 (SD = ±9) years and 19% of the participants had no schooling, which was higher in the rural compared to the urban population (21% vs 17%). There were no significant differences in the age by gender distribution. Valid accelerometer data (i.e., ≥10 h of wear times over ≥3 days) were received from 155 participants (urban = 61, rural = 94). The mean ± SD steps/day was 9998 ± 3936 (8658 ± 2788 and 12,063 ± 4534 for rural and urban respectively, p = 0.0001) and the mean daily MVPA was 58 min (51 for rural and 69 for the urban, p = 0.001). Based on GPAQ, rural residents reported significantly higher moderate work related PA (median MET-minutes/week: 600 vs. 360 p = 0.02) than did urban residents.

Table 2 shows the correlation between physical activity assessed by the GPAQ and measured by the accelerometer. Spearman correlation coefficients between GPAQ total MVPA and accelerometer MVPA, CPM or steps counts were low for the whole population (rho: 0.18, 0.24, and 0.28, respectively). However, stratification by place of residency indicated good correlations for urban residents (rho: 0.46, 0.55 and 0.63, respectively; p < 0.01) and very poor correlations for rural residents (rho: 0.0001, −0.01 and 0.05, respectively). The domain specific correlations across all indicators (i.e., MVPA, CPM or steps counts) among urban population were high for the work, travel and leisure domains (0.26 to 0.55) but among rural residents the coefficients were low for all domains. GPAQ occupational and leisure related PA showed significant fair to moderate correlation with light-intensity PA for urban population. Time in light intensity was inversely related to travel-related activity in rural area. A significant, low level of agreement between the GPAQ and accelerometer data for sitting was observed (rho: 0.23; p < 0.01). The agreement for categorization of participants into meeting sufficient physical activity level was fair for all participants (Kappa: 0.29; p < 0.0001) and good for urban participants (Kappa: 0.62; p < 0.0001) though low for rural participants (Kappa: 0.07) (data not shown).

Figure 1a and b present the Bland and Altman plots for the agreement between GPAQ MVPA min/day for all domains and accelerometer MVPA in min/day by place. Figure 1a and b showed that the differences between the two instruments were 64.75 and 88.88 min of MVPA per day, respectively. The limits of agreement were wide with the difference lying between −180.41 to 309.91 min/day for urban and −130.94 to 308.70 for rural. A clear pattern of increased error was detected with increased average of PA. Figure 1c and d showed the difference between the two instruments in urban and rural participants were −419.63 and −415 min of SB per day with wide difference (−659.55 to −179.71 min/day and −577.02 to −252.98 min/day) which indicates negative bias exists for the GPAQ.

Table 3 shows the correlation between GPAQ total MVPA MET-mins/day and accelerometer MVPA, CPM and steps across sociodemographic subgroups. Consistent, significant correlations were found between GPAQ total MVPA and accelerometer MVPA, CPM and steps among women (rho: 0.42, 0.46 & 0.49 respectively) and young adults (age ≤35 years) (rho: 0.31, 0.32 & 0.34 respectively). After stratification by place, significant fair-to-moderate correlation was found for females, whereas urban young adults showed a significantly higher correlation than young rural adults. For education subgroups, the patterns of the correlations were inconsistent and did not follow gradient. Overall the correlations with GPAQ MVPA and accelerometer MVPA, CPM and steps were moderate for the graduate group (rho: 0.48, 0.45 & 0.51 respectively) and fair for the illiterate (rho: 0.27, 0.35 & 0.22 respectively) and primary group (rho: 0.23, 0.20 & 0.38 respectively).

To the best of our knowledge this is the first validity study of the GPAQ in Bangladesh using accelerometer and also including rural population. The results demonstrated moderate evidence of criterion related validity for total GPAQ MVPA and all domains of MVPA for urban participants but poor criterion validity for rural participants. The GPAQ demonstrated fair-to-moderate criterion validity for women, young adults (≤35 years) and those with higher level of education.

Our results contradict some findings from the nine countries validity study by Bull et al. [12]. First, the rural samples (Ethiopia, Indonesia and India) had better coefficients (rho: 0.43) than their counterpart urban sample (rho: 0.23), albeit the criterion was pedometer steps counts. Second, in the nine countries’ samples the validity coefficients were better for men than women whereas in the current study it was the other way around. Third, China and South Africa used accelerometers as the criterion measure for urban samples and the coefficient between GPAQ total PA across all domains and accelerometer moderate-intensity counts per minute were 0.24 for China sample and very poor (−0.01) for South Africa sample, much lower than the coefficient for counts per minutes in our urban sample. In the nine countries study, criterion validity for urban Bangladesh was assessed by pedometer, which is a less sensitive objective measure than the accelerometer used in the current study. The overall correlation was 0.06, which was considerably lower than in our current study where accelerometer was used [12].

Our results are comparable to other studies where low-to-moderate validity (rho: 0.20–0.48) was demonstrated against objective measures [6, 27‐33]. Additionally, in the 12 countries study validation of the short International Physical Activity Questionnaire (IPAQ), the pooled validity correlations against accelerometers was found to be 0.30 (95% CI 0.23–0.36) [34]. In the Bland-Altman plot a clear pattern of increased error was detected with increased average of PA for rural and urban participants. Overestimation of GPAQ was observed in the US [27] and Singapore [28], whereas negative bias was seen in the Northern Ireland with the majority of points falling below zero [6].

Several reasons may explain the low validity of GPAQ for our rural population as opposed to the urban sample; firstly, the dominant work-related PA in rural area is farming, it is a hard work that involves digging, cutting crops, rice processing, carrying heavy loads etc. but the positioning of waist-worn accelerometers affects their ability to capture these upper body movement activities. Further, non-ambulatory activities such as cycling is also not captured by waist-worn accelerometers, and cycling is a very prevalent mode of transport in rural areas [35, 36], particularly among men. Additional explanation may be related to the pace of ambulation in the country-side which may result in accelerometer counts below the cut-off point for moderate activity [25]. For example, the Freedson determination of moderate and vigorous PA accelerometer cut points were based on walking and running on a treadmill [25] and are unlikely to capture the intensity associated with walking carrying heavy loads or on uneven surface as is common in rural areas of developing countries such as Bangladesh and thus accelerometer may underestimate total MVPA in these populations [37]. Support for this argument is the good correlation we found with time spent on light-intensity PA based on accelerometer and GPAQ occupation and travel-related physical activity, a correlation that was in the same range as for accelerometer MVPA. This may indicate that lowering the cut-point may improve the indicators for GPAQ validity.

Moreover, reference time of the GPAQ is ‘usual week’ which may create confusion in the participants to determine which particular week of a month would be best to address. The 12-countries study of validation on IPAQ discussed that the understanding of a ‘usual week’ was difficult for participants as they were not able to identify ‘what is usual?’ and participants recalled last 7 days instead of ‘usual week’ [34]. This could be more problematic if there is strong seasonal variation. In Bangladesh, there are 6 seasons and the main occupation in rural area is agriculture which follows the seasons [38]. A multi-site study of nine Asian rural areas including Bangladesh showed that PA was lower in the middle of the harvest season and increased during the more intensive harvest period [39]. Another reason for the low correlation of GPAQ MVPA in rural participants might be that accelerometer data were collected in autumn season when people are less active and GPAQ MVPA was the usual week. On the other hand urban residents had almost similar work patterns throughout the year because our urban participants were selected from one work site where occupation related PA is stable throughout the year.

In the subgroup analysis, females showed consistent correlations across all indicators of PA. In male overall correlation across indicators of PA was not seen, whereas other studies showed reverse result [6, 40, 41]. This may be due to the context specific nature of activities undertaken within both urban, and, particularly, rural Bangladesh which often requires considerable upper-body motion such as labor-intensive farming practices, as noted before, or construction jobs in the city. Moreover, males carry heavy loads such as crops, seeds, sacks etc. which limits their pace of walking. Both pedometers and accelerometers are likely to underestimate the intensity of these activities despite their being moderate-intensity efforts subjectively, as well as by energy expenditure measure (Ainsworth range 5METs to 8.5METS) [42]. In addition, swimming and cycling are common activities for rural people. Because accelerometers do not measure water-based and non-ambulatory activities, this may have contributed to the poor correlations found in males. In case of education, the patterns of the correlations were inconsistent and did not follow gradient, however, higher correlation was found in tertiary education group than other groups. This finding is similar to that of a study by Lee et al. who found that participants who had tertiary education performed better for IPAQ and over-reporting was almost double in those without tertiary education [40]. The nine countries study of GPAQ validation also showed higher correlation for those with higher education compared with those with less than 13 years of schooling [12], as in our study. Therefore it is possible that the overall low validity in rural sample was confounded by the lack of representation of participants with graduate degree in this sample. Yet, the coefficients for the illiterate groups, in both places, were better than those with primary and high-school education. The lack of gradient in coefficients by education levels suggest that factors other than cognitive errors may have contributed to the low validity such as the type of occupation they do (static, non-ambulatory).

The current study found the volume of sedentary behaviour (SB) was greater when measured by the accelerometer than by the GPAQ. This finding is similar to that of recent study where found that when SB was measured with a self-reported single item it significantly underestimates SB in comparison to accelerometer data [43]. However, a study conducted on a Chilean population found the single question from the GPAQ had fair validity for measuring SB, though poor ability for correctly classifying individuals into tertiles or quartiles of SB [30]. Our finding of a low correlation (rho: 0.23) between GPAQ measurement for minutes of sitting per day and accelerometer data agrees with previous studies [6, 12], nevertheless, this correlation coefficient increased for rural (rho: 0.38). Present findings demonstrate that GPAQ may not be appropriate when assessing minutes of SB for both urban and rural populations as it results with systematic under-estimation of amount of sitting by 7 h on average (range between 3 to 11 h) compared to accelerometer and this was true for both rural and urban population. More accurate measurement of SB may be provided by using a multiple item domain-specific questionnaire [24, 43].

The study had a number of strengths as it assessed validity of GPAQ-2 both in urban and rural population which is rare in Bangladesh and in general. Secondly, there was good compliance with accelerometer wear and adherence to the study protocol. Additionally, we followed WHO guidelines for administering the GPAQ, provided intensive training on data collection staff and close supervision during data collection to minimize avoidable sources of measurement error.

We used a triaxial Actigraph accelerometer as a reference measure for criterion validity. The gold standard measurement for assessing energy expenditure are indirect calorimetry, doubly labelled water or heart rate monitoring, however, these are expensive and require technical expertise for implementation. Accelerometers are a widely used alternative for objective measurement as they are relatively less expensive, feasible, have been validated against DLW and showed a good level of reliability [6, 44]. Nevertheless, accelerometers have their limitations. For example, in this study accelerometer data likely underestimated MVPA in the rural sample due to its inability to capture water-based, non-ambulatory and statics activities. Thus, using accelerometer as a criterion might be considered as concurrent or convergent instrument due to its pitfalls.

On the other hand, over-reporting with activity questionnaires is ubiquitous as they are prone to biases such as recall and social desirability [45, 46]. So, these could lead to overestimation of activity levels in some domains and underestimation in others. It could be better explained if we know the pattern of activities of rural Bangladesh where PA varies with seasonality. Moreover, GPAQ does not capture details of many activities culturally relevant to Bangladesh. These might be the reasons that low PA was found to be almost similar in both urban and rural (28.9 & 25.1 respectively) population in 2010 Bangladesh NCD Risk Factor Survey where GPAQ was used [47]. Hence, we suggest that in the introduction of the questionnaire the typical week should be referenced to the typical week of the season or asking about the past week, as is the case in many surveillance questionnaires. Another limitation could be that the urban sample was comprised of volunteers from a workplace setting, thus the results may not have complete reflection of the general urban population.

In conclusion, the present study adds important new data on the validity of the widely-used GPAQ for estimating PA and SB levels in a low income country. For the whole population, the GPAQ performed as well as other population PA surveillance tools. Its performance with regard to the urban population was at the highest range of most surveillance tools and better than for the rural population. The GPAQ seems to be an effective tool for measuring PA in females and people with high levels of education.

Given waist worn accelerometers do not capture the typical PA in rural context; further study using a physical activity diary and a combination of multiple sensors (e.g., wrist, ankle and waist worn accelerometer) to capture all movement would be informative. Such a study should include purposive sampling of all education levels to ascertain the extent to which education level is associated with better performance.