Correlates of Agreement between Accelerometry and Self-reported Physical Activity

Abstract

Purpose

Understanding factors that influence accurate assessment of physical activity (PA) and sedentary behavior (SB) is important to measurement development, epidemiologic studies, and interventions. This study examined agreement between self-reported (International Physical Activity Questionnaire – Long Form, IPAQ-LF) and accelerometry-based estimates of PA and SB across six countries, and identified correlates of between-method agreement.

Methods

Self-report and objective (accelerometry-based) PA and SB data were collected in 2002-2011 from 3,865 adult participants in eight cities from six countries (Belgium, Czech Republic, Denmark, Spain, UK, and USA). Between-method relative agreement (correlation) and absolute disagreement (mean difference between conceptually- and intensity-matched IPAQ-LF and accelerometry-based PA and SB variables) were estimated. Also, socio-demographic characteristics and PA patterns were examined as correlates of between-method agreement.

Results

Observed relative agreement (relationships of IPAQ-LF with accelerometry-based PA and SB variables) was small to moderate (r=0.05-0.37) and was moderated by socio-demographic (age, sex, weight status, education) and behavioral (PA-types) factors. The absolute disagreement was large, with participants self-reporting higher PA intensity and total time in moderate-to-vigorous PA than accelerometry. Also, self-reported sitting time was lower than accelerometry-based sedentary behavior. After adjusting for socio-demographic and behavioral factors, the absolute disagreement between pairs of IPAQ-LF and accelerometry-based PA variables remained significantly different across cities/countries.

Conclusions

Present findings suggest systematic cultural and/or linguistic and socio-demographic differences in absolute agreement between the IPAQ-LF and accelerometry-based PA and SB variables. These results have implications for the interpretation of international PA and SB data and correlates/determinants studies. They call for further efforts to improve such measures.

Introduction

Prevalence data for physical activity (PA) and sedentary behavior (SB) are now available for dozens of countries based on self-report surveys (1, 2, 18). The validity and between-country comparability of these data relies on survey items having the same meaning across cultures and languages, and capturing relevant behaviors that are prevalent in specific populations (24). The International Physical Activity Questionnaire (IPAQ) was developed for international use and has evidence of reliability and validity (9). Although the IPAQ-Short Form has been most often used for international prevalence studies (1, 2), the IPAQ–Long Form (IPAQ-LF) has also been used in international research, especially in research focused on domain-specific PA and studies of environmental correlates of PA and SB (9, 6). The main difference is that IPAQ-LF provides more detailed estimates of PA for each domain of leisure, transportation, occupation, and household activities. IPAQ-LF has been extensively validated against more reliable objective measures of PA and SB derived from accelerometers (motion sensors) (5, 9, 15, 23). However, key questions remain about the sources of differences in agreement between accelerometry-derived and IPAQ-LF estimates of PA and SB (28, 34). This knowledge is critical for a robust comparison and interpretation of differences in findings across populations and measurement methods (self-report vs. accelerometry).

The first potential source of differences in agreement between accelerometry and IPAQ-LF (and any other self-report) pertains to the way PA and SB variables derived from accelerometers are computed. Accelerometry data are usually converted into time spent in SB and specific intensity levels of PA (light, moderate, and vigorous) using pre-established cut-points (15), and compared to their self-report counterparts. Although IPAQ-LF is meant to gauge moderate-to-vigorous PA (corresponding to ≥3 METs [multiples of resting energy expenditure]), better concordance of IPAQ-based total and moderate-intensity PA with corresponding accelerometry-based variables has been observed when the latter also included light-intensity activities (<3 METs) as defined by Freedson's cut-points (the cut-points that have been used in the original multi-country validation study of the IPAQ) (5, 9, 12, 23). This finding indicates that respondents might overestimate the intensity of their activities or/and that the accelerometer cut-points for adults developed by Freedson and colleagues may not be appropriate for certain subgroups. Yet, only one study has systematically examined this issue (5).

IPAQ-LF asks respondents to report activities undertaken for ≥10 minutes at a time (9). Although the respondents’ ability to accurately recall PA bout length has been questioned (39), only two studies have examined the agreement of IPAQ-LF estimates of PA with accelerometry-based PA accumulated throughout the monitoring period and PA accumulated through ≥10-minute bouts (5, 17). Identifying the accelerometry-based variables that yield the best agreement with the variables derived from IPAQ-LF administered in different countries would help to understand how participants self-report PA and identify PA profiles that accelerometers may underestimate. This information would help to establish the extent to which between-population differences in levels of PA/SB and their correlates are authentic phenomena or the results of measurement biases.

Studies have examined socio-demographic factors associated with the agreement between accelerometry- and IPAQ-based estimates of PA and SB. These include sex, age, education, weight status, and job status (3, 13, 22, 33). The types of PA participants engage in may also moderate the level of agreement. For example, as accelerometry underestimates PA accrued through bicycling (19), it is plausible to assume that the level of agreement between self-reports and accelerometry will be lower for those who regularly engage in bicycling. Respondents who accrue most of their PA through leisure-time activities may display higher levels of agreement because leisure-time PA is usually planned, structured, at least of moderate intensity, and, hence, potentially easier to recall. In contrast, those who accrue most of their PA via occupational activities may be more likely to overestimate their PA (3). Given that occupational activities are generally less structured and more variable in intensity, respondents may have difficulties recalling the actual amount of moderate-to-vigorous PA during working hours and instead report the hours spent working (3, 29, 35).

Although IPAQ-LF has been validated against accelerometry in many countries, these validation studies typically used small samples (3, 5, 9, 27 37) and/or examined only relative agreement (9, 37) rather than both relative agreement (correlation) and absolute disagreement (absolute difference) between measures (3, 5). Examining both is important because relative agreement does not capture systematic bias (i.e., overestimation or underestimation of self-reported PA and SB). Also, previous studies operationalized (i.e., scored) accelerometry-based PA in different ways and used socio-demographically diverse samples, making it difficult to compare findings. To determine the extent to which between-country differences in agreement between self-report and accelerometry PA/SB measures may be due to measurement and interpretational biases, analyses need to be adjusted for factors associated with agreement.

Using data from six countries, this study: 1) examined the relative agreement and absolute disagreement between IPAQ-LF and accelerometry-based estimates of PA and SB; 2) identified socio-demographic (sex, age, educational attainment, and weight status) and PA-type moderators (percentages of total PA time spent in occupation, leisure, and bicycling for transport) of agreement between IPAQ-LF and accelerometry-based estimates of PA and SB; and 3) estimated between-city differences in levels of agreement after adjustment for significant socio-demographic and PA-type moderators. This information is essential for a robust comparison and interpretation of levels of PA and SB across countries.

Methods

We used data collected in eight cities from six countries participating in the International Physical Activity and the Environment Network Adult study, an observational epidemiologic cross-sectional project (6, 20). The cities included were Ghent (Belgium), Olomouc and Hradec Kralove (Czech Republic), Aarhus (Denmark), Pamplona (Spain), Stoke-on-Trent (UK), and the Seattle and Baltimore regions (USA).

Participants and procedures

Participants (18-65 years) were recruited from neighborhoods stratified by walkability and socio-economic status (20). Study dates ranged from 2002 to 2011. Each participating country obtained ethical approval from their institutional review boards, and all participants provided written informed consent prior to data collection. Participants were invited to complete the IPAQ-LF and wear accelerometers to objectively measure PA. Denmark, Spain, and the UK asked only a subset of participants (20%-50%) to wear accelerometers. Denmark, Spain, the Czech Republic, and the USA employed self-administered methods (mail and online surveys) to collect survey data. The UK employed mailed surveys and Computer Assisted Telephone Interviews, while Belgium used in person interviews (20).

Of 6,251 participants, 1,787 did not wear an accelerometer because they did not consent or were not asked to wear it. When compared to participants who did not wear accelerometers those who consented to wear it (n=4,464) were more likely to be older (P<0.001), live with a partner (P<0.001), and hold a tertiary degree (P<0.001). Among those who consented to wear accelerometers, those with ≥4 days of valid accelerometer data (n=4,234) were more likely to be older (P<0.001), and self-report fewer minutes of total PA than those without ≥4 days of valid accelerometer data (n=230). Of 4,234 participants with valid accelerometry data, 176 were excluded due to missing data on IPAQ-LF or covariates, and 193 participants were excluded as they reported more than 12 h/day of total PA. Excluded participants were more likely to be employed (P<0.001) and live in lower income neighborhoods, and were less likely to have a tertiary degree, and to accumulate more of their total PA through leisure-time activities (P<0.001). Participants with missing (n=176) data were more likely to be older (P=0.014) and employed (P=0.024) than those with no missing data (n=4,058). They did not significantly differ in accelerometry-based estimates of PA and SB. The socio-demographic characteristics of the sample included in this study (N=3,865) are presented in Table 1.

Table 1.

Descriptive statistics: socio-demographics by city

Characteristics	Whole sample	Ghent (BEL)	Olomouc (CZ)	Hradec Kralove (CZ)	Aarhus (DK)	Pamplona (ESP)	Stoke on Trent (UK)	Seattle (USA)	Baltimore (USA)
n a	3865	957	195	103	241	303	98	1140	828
Age, yrs, mean (SD)	43.1 (12.4)	42.5 (12.5)	37.3 (14.1)	35.7 (13.9)	39.3 (13.7)	39.5 (13.5)	43.0 (13.3)	44.3 (10.9)	46.6 (10.8)
Sex, % men	48.1	48.8	34.9	36.9	38.2	38.0	48.0	54.6	49.4
Education, % tertiary degree	61.4	63.3	34.9	30.1	51.9	66.0	14.3	65.6	70.3
Work status, % working	79.8	81.4	75.4	82.5	73.0	75.6	63.3	80.9	82.6
BMI kg/m2, mean (SD)	25.6 (5.0)	24.2 (3.9)	24.6 (3.9)	24.2 (3.5)	24.2 (4.0)	23.8 (3.4)	27.1 (5.2)	26.6 (5.4)	27.1 (5.6)
Weight status, %
Normal	52.3	63.2	62.6	59.2	66.8	67.7	39.8	45.1	38.2
Overweight	32.8	28.4	28.7	35.0	25.7	27.1	33.7	34.6	40.1
Obese	14.9	8.4	8.7	5.8	7.5	5.3	26.5	20.4	21.7

^aIncluded in the analyses (with complete data).

BEL, Belgium; CZ, Czech Republic; DK, Denmark; ESP, Spain; SD, standard deviation; UK, United Kingdom; USA, United States of America; yrs = years

Measures

Physical activity and sedentary behavior

The International Physical Activity Questionnaire – Long Form (IPAQ-LF, last 7 days) prompts participants to report the frequency and duration of activities and sitting time in the last 7 days (9). Activities were classified into the domains of occupation, transportation, household, and leisure, for each category of walking, moderate-intensity PA (MPA), and vigorous-intensity PA (VPA). Overall sitting time on a typical weekday and weekend day were also reported. Average daily hours of sitting (including motorized transportation) and weekly minutes of total PA, MPA plus walking, and VPA were computed.

PA was measured objectively with ActiGraph accelerometers (Pensacola, FL), which are widely used to objectively characterize intensity and duration of free living PA (16). The UK (Stoke-on-Trent) site used the GT1M, Denmark (Aarhus) used the GT3X, and the Czech Republic (Olomouc) and Spanish (Pamplona) sites used both the GT1M and GT3X models. The USA (Baltimore and Seattle) and Belgian (Ghent) sites used the older 7164/71256 models. Studies have consistently shown that the newer accelerometer models (GT1M, GT3X) produce comparable output (31) but findings are mixed when comparing the output of the older 7164/71256 models to the newer models (4).

Participants were asked to wear the accelerometer around their waist on a belt with the device oriented above the right hip for 7 days during waking hours when not engaged in water activities. Participants wore the accelerometers after completing the surveys in all cities except for Baltimore and Seattle where they completed the survey after wearing the accelerometer.

Accelerometer data were either collected with or aggregated to 1-minute epochs. Non-wear time was defined as 60+ minutes of consecutive ‘0’ activity counts. Days consisting of at least 10 wearing hours were coded as valid. Participants were included in the analyses if they had ≥4 valid wearing days. Accelerometer counts were converted to minute estimates of sedentary time and activity intensity, respectively, using the values established by Matthews et al. (25) and the Freedson adult cut points (15): 0-100 counts for sedentary time (1.0 METs), 101-1,951 counts for light PA (1.1-2.9 METs), 1,952-5,724 for moderate PA (3.0-6.0 METs), and >5,724 for vigorous PA (>6.0 METs). Total daily minutes and daily minutes accumulated in ≥10-minute-bouts of each activity intensity (with an allowance for up to 2-minutes of values to fall outside of the cut-point threshold) were summed across valid wearing days, divided by the number of valid days, and multiplied by seven to estimate weekly minutes of cumulative activity and bouts in each intensity. These variables were then adjusted for accelerometer wear time via regression methods. The following wear-time-adjusted variables were created: total weekly minutes and weekly minutes accumulated in ≥10-minute-bouts of moderate-to-vigorous PA (MVPA), light-to-moderate PA (LMPA), light-to-vigorous PA (LVPA), vigorous PA (VPA), and moderate PA (MPA). Also, we computed average daily hours of sedentary time.

Moderators of relationships and differences between IPAQ-LF and accelerometry-based PA and SB variables

In addition to the city of residence, the following socio-demographic characteristics were considered as moderators of relationships and differences in means between IPAQ-LF and accelerometry-based PA variables: age, sex, educational attainment (without vs. with tertiary degree), and weight status based on Body Mass Index (BMI; kg/m²). Height and weight were measured objectively in Stoke-on-Trent (UK) and self-reported elsewhere (10). Self-reported and objectively measured BMI are highly correlated, and BMI is a proxy measure for adiposity in large-scale studies (26). Data from IPAQ-LF were used to determine the percentages of total self-reported PA spent on leisure-time PA, transport-related bicycling, and occupational PA, which were examined as potential moderators.

Data analyses

Descriptive statistics were computed for the whole sample included in the analyses and by city.

Relationships of IPAQ-LF with accelerometry-based PA and SB variables

Relationships of corresponding pairs of accelerometry-based and IPAQ-LF variables were estimated using generalized additive mixed models (GAMMs) (40). IPAQ-LF variables were regressed onto corresponding accelerometry-based variables (Table 2), while adjusting for clustering at the administrative unit and city levels. The square root of the coefficient of determination was computed (hereafter, coefficient of correlation, r). As the coefficient of determination represents the proportion of shared variance, its square root is interpreted as a Pearson correlation coefficient typically used in validity studies of PA self-reports (34). Strength of relationships was determined using Cohen's cut points for correlation coefficients (r=0.10-0.30, small; r=0.31 to 0.50, moderate; r>0.50, large) (8). To estimate curvilinear relationships smooth terms (thin-plate splines) were used (40). Moderating effects of city, socio-demographic characteristics, and percentage time spent in specific PA domains were assessed by adding appropriate interaction terms to the previous GAMMs (separate GAMMs for each moderator). A difference of >5 units in Akaike Information Criterion values (similar to using a probability level of 10% for parametric tests of interaction effects) between the simpler vs. more complex GAMMs was used as evidence of curvilinearity and/or interaction effects (6). The Wald test was used to assess the significance of single interaction terms while adopting a two-tailed probability level of 0.05. Separate GAMMs were estimated for each stratum or pre-determined value of a significant moderator to obtain stratum- or value-specific correlation coefficients.

Table 2.

Pairs of IPAQ-LF and accelerometry-based physical activity and sedentary behavior variables compared in this study and rationale

Pair	IPAQ – LF PA variable	Accelerometry-based PA variable	Rationale for comparison of variables
1	Total PA (min/week)	Moderate-to-vigorous intensity PA accumulated in 10+ minute bouts (min/week)	Conceptually matched (intensity and bout-length matched). The IPAQ-LF measures moderate intensity PA, vigorous intensity PA and walking accumulated in 10+ minute bouts. Walking was considered a moderate intensity PA associated with 3.3 METs.
2	Moderate intensity PA, including walking (min/week)	Moderate intensity PA accumulated in 10+ minute bouts (min/week)
3	Vigorous intensity PA (min/week)	Vigorous intensity PA accumulated in 10+ minute bouts (min/week)
4	Total PA (min/week)	Total moderate-to-vigorous intensity PA (min/week)	Intensity matched. Respondents may have had difficulties reporting PA accumulated in 10+ minute bouts and may have instead reported total PA. Thus each minute of accelerometer data was counted that meets the criterion.
5	Moderate intensity PA, including walking (min/week)	Total moderate intensity PA (min/week)
6	Vigorous intensity PA (min/week)	Total vigorous intensity PA (min/week)
7	Total PA (min/week)	Light-to-vigorous intensity PA accumulated in 10+ minute bouts (min/week)	Respondents may have tended to overestimate the intensity level of their PA. Accelerometer cut-point to classify PA intensity may have been too high for mid-aged and older respondents as they were developed on a younger sample (mean age 24 years).
8	Moderate intensity PA, including walking (min/week)	Light-to-moderate intensity PA accumulated in 10+ minute bouts (min/week)
9	Vigorous intensity PA (min/week)	Moderate-to-vigorous intensity PA accumulated in 10+ minute bouts (min/week)
10	Total PA (min/week)	Total light-to-vigorous intensity PA (min/week)	Respondents may have tended to overestimate the intensity level of their PA and had difficulties reporting PA accumulated in 10+ minute bouts. Thus each minute of accelerometer data was counted that meets the criterion.
11	Moderate intensity PA, including walking (min/week)	Total light-to-moderate intensity PA (min/week)
12	Vigorous intensity PA (min/week)	Total moderate-to-vigorous intensity PA (min/week)
13	Sitting (including transportation) (average h/day)	Total sedentary time (average h/day)	Weakly conceptually matched. It is noted that accelerometers poorly differentiate standing (a non-sedentary behavior) from sitting and reclining (sedentary behaviors). They only provide a rough, possibly upward-biased estimate of sitting time.

h/day, hours per day; IPAQ-LF, International Physical Activity Questionnaire – Long Form; METs, multiples of resting energy expenditure; min/week, minutes per week; PA, physical activity.

Differences between IPAQ-LF with accelerometry-based PA and SB variables

Moderating effects of city, socio-demographic characteristics, and percentage of time spent in specific PA domains on computed differences between IPAQ-LF and accelerometry-based PA variables (Table 2) were examined using GAMMs by regressing the difference scores onto the potential moderators. All analyses were conducted in R (30) using the packages ‘mgcv’ (40) and ‘car’ (14).

Results

PA estimates based on IPAQ-LF were substantially higher than their conceptually-matched accelerometry-based counterparts (Table 3). For example, the average total PA as estimated by the IPAQ-LF was 1,185 min/week, while according to accelerometry, participants accumulated 256 min/week (counting every MVPA minute) and only 81 min/week of MVPA in ≥10-minute bouts. It seems that PA estimates based on IPAQ-LF were more variable across and within cities than those based on accelerometry (Table 3).

Table 3.

IPAQ-LF and accelerometry-based estimates of physical activity and sedentary behavior (means and standard deviations) by city

Variables	Whole sample	Ghent (BEL)	Olomouc (CZ)	Hradec Kralove (CZ)	Aarhus (DK)	Pamplona (ESP)	Stoke on Trent (UK)	Seattle (USA)	Baltimore (USA)
IPAQ-LF
Total PA [min/week]	1,185 (935)	924 (765)	1,522 (974)	1,472 (964)	1,604 (957)	1,364 (916)	1,348 (1203)	1,163 (943)	1,197 (985)
MPA, including walking [min/week]	965 (784)	767 (671)	1,283 (823)	1,186 (807)	1,362 (829)	1,179 (781)	1,077 (835)	916 (775)	953 (796)
VPA [min/week]	230 (358)	157 (299)	239 (335)	286 (330)	242 (237)	185 (307)	271 (378)	247 (367)	244 (333)
Sitting time [h/day]	6.3 (2.9)	7.1 (2.8)	5.0 (2.8)	4.7 (2.8)	6.3 (3.1)	6.1 (2.7)	4.9 (3.1)	6.2 (2.9)	6.3 (2.9)
% total PA time spent on leisure	23.8 (24.9)	23.2 (26.0)	22.5 (24.4)	24.6 (23.8)	25.6 (21.8)	27.5 (22.4)	14.4 (23.3)	25.7 (26.0)	22.2 (23.9)
% total PA time spent on bicycling for transport	3.9 (11.3)	9.7 (17.7)	1.9 (5.8)	3.9 (6.6)	10.3 (15.2)	1.6 (5.5)	<0.1 (0.2)	1.2 (6.6)	0.6 (3.0)
% total PA time spent on occupation	20.7 (29.2)	21.0 (31.1)	24.9 (29.4)	21.1 (26.3)	26.6 (28.1)	18.8 (29.2)	20.8 (32.1)	18.7 (28.5)	21.2 (27.6)
Accelerometrya
LVPA – bouts [min/week]	1,181 (609)	1,379 (675)	1,138 (572)	1,091 (532)	931 (456)	1,032 (909)	1,208 (570)	1,154 (571)	1,137 (592)
LVPA – total [min/week]	2,438 (642)	2,578 (682)	2,391 (569)	2,355 (559)	2,168 (527)	2,402 (603)	2,496 (683)	2,440 (629)	2,381 (643)
MVPA – bouts [min/week]	81 (106)	66 (88)	122 (118)	120 (127)	104 (97)	151 (164)	81 (113)	79 (102)	54 (83)
MVPA – total [min/week]	256 (173)	248 (162)	334 (186)	308 (178)	275 (154)	357 (209)	248 (187)	252 (169)	204 (152)
LMPA – bouts [min/week]	1,170 (610)	1,369 (676)	1,128 (576)	1,085 (532)	909 (456)	1,021 (573)	1,205 (570)	1,141 (573)	1,130 (592)
LMPA – total [min/week]	2,419 (641)	2,560 (682)	2,369 (572)	2,341 (562)	2,136 (522)	2,381 (598)	2,487 (680)	2,418 (628)	2,366 (640)
MPA – bouts [min/week]	70 (98)	55 (80)	113 (106)	113 (126)	82 (85)	140 (155)	78 (113)	66 (92)	47 (76)
MPA – total [min/week]	236 (157)	230 (147)	312 (161)	294 (172)	243 (132)	336 (190)	239 (180)	229 (151)	190 (137)
VPA – bouts [min/week]	11 (36)	10 (33)	10 (38)	6 (20)	22 (47)	11 (41)	3 (12)	13 (41)	7 (27)
VPA – total [min/week]	20 (45)	18 (42)	22 (54)	14 (26)	32 (55)	21 (52)	9 (20)	22 (49)	15 (37)
Sedentary time [h/day]	8.8 (1.7)	8.5 (1.8)	8.2 (1.6)	8.6 (1.5)	9.6 (1.5)	9.1 (1.5)	8.5 (1.6)	8.8 (1.7)	9.0 (1.7)

^aParticipants had on average 6.7 (SD=1.0) valid days of accelerometry data with 14.7 valid hours of data per day. Accelerometry-based estimates of physical activity and sedentary time were adjusted for accelerometer wear time.

– bouts, amount of PA accumulated in ≥10-minute bouts; BEL, Belgium; CZ, Czech Republic; DK, Denmark; ESP, Spain; IPAQ-LF, International Physical Activity Questionnaire – Long Form; LMPA, light-to-moderate intensity physical activity; LVPA, light-to-vigorous intensity physical activity; MPA, moderate intensity physical activity; MVPA, moderate-to-vigorous intensity physical activity; PA, physical activity; UK, United Kingdom; USA, United States of America; VPA, vigorous intensity PA.

Relationships of IPAQ-LF with accelerometry-based PA and SB variables (relative agreement)

Correlations between IPAQ-LF and accelerometry-based PA variables were small to moderate, ranging from 0.05 to 0.37 (Table 4). They were weaker for conceptually-matched PA variables, followed by intensity-matched variables. Stronger relationships were observed for IPAQ-LF PA variables with accelerometry-based PA encompassing lower- plus matched-intensity levels (e.g., total PA based on IPAQ-LF and total accelerometry-based LVPA including light through vigorous PA minutes; pairs 7-8 and 10-11 in Table 2). The strongest relationship (0.37) was observed between sitting time as measured by IPAQ-LF and accelerometry-based sedentary time. This relationship was curvilinear (see Figure, Supplemental Digital Content 1, Relationship between self-reported sitting time and accelerometry-based sedentary time), as was that between accelerometry-based total VPA and IPAQ-LF VPA (see Figure, Supplemental Digital Content 2, Relationship between self-reported and accelerometry-based vigorous-intensity physical activity).

Table 4.

Correlations (r) between IPAQ-LF and accelerometry-based estimates of physical activity and sedentary behavior for whole sample and significant moderator subgroups

IPAQ-LF	Accelerometrya	r	95% CI	Significant moderators	Correlations at specific values of moderators
					Values of moderators	r	95% CI
Total PA	MVPA – boutsb	0.12	0.08, 0.14	Weight status	Normal	0.12	0.07, 0.16
					Overweight	0.08	0.03, 0.14
					Obese	0.16	0.08, 0.24
	MVPA – totalc	0.20	0.17, 0.22	Sex	Men	0.25	0.20, 0.29
					Women	0.13	0.09, 0.17
				Weight status	Normal	0.18	0.13, 0.22
					Overweight	0.20	0.14, 0.25
					Obese	0.25	0.17, 0.33
	LVPA – bouts	0.30	0.27, 0.33	Sex	Men	0.37	0.33, 0.41
					Women	0.24	0.19, 0.28
				% total PA time spent on occupation	Average or below average	0.28	0.23, 0.32
					Above average	0.23	0.19, 0.26
				% total PA time spent on bicycling for transport	Average or below average	0.33	0.30, 0.36
					Above average	0.18	0.10, 0.25
	LVPA – total	0.31	0.28, 0.34	Sex	Men	0.38	0.34, 0.42
					Women	0.23	0.19, 0.27
				% total PA time spent on occupation	Average or below average	0.29	0.24, 0.34
					Above average	0.21	0.18, 0.25
				% total PA time spent on bicycling for transport	Average or below average	0.34	0.31, 0.37
					Above average	0.18	0.10, 0.25
MPA + walking	MPA – boutsb	0.13	0.10, 0.16	Weight status	Normal	0.13	0.09, 0.18
					Overweight	0.11	0.06, 0.17
					Obese	0.16	0.08, 0.24
	MPA – totalc	0.19	0.15, 0.21	Sex	Men	0.24	0.20, 0.29
					Women	0.13	0.09, 0.17
	LMPA – bouts	0.28	0.25, 0.30	Sex	Men	0.31	0.27, 0.35
					Women	0.23	0.19, 0.27
				% total PA time spent on occupation	Average or below average	0.27	0.22, 0.32
					Above average	0.21	0.15, 0.26
				% total PA time spent on bicycling for transport	Average or below average	0.30	0.26, 0.33
					Above average	0.18	0.10, 0.25
	LMPA – total	0.28	0.25, 0.31	% total PA time spent on occupation	Average or below average	0.28	0.23, 0.33
					Above average	0.22	0.18, 0.25
				% total PA time spent on bicycling for transport	Average or below average	0.30	0.27, 0.33
					Above average	0.17	0.10, 0.25
VPA	VPA – boutsb	0.09	0.06, 0.13	Sex	Men	0.05	0.01, 0.10
					Women	0.14	0.09, 0.18
	VPA – totalc	0.11	0.08, 0.14	Sex	Men	0.06	0.02, 0.11
		Curvilinear (Figure 1)			Women	0.14	0.10, 0.19
	MVPA – bouts	0.05	0.02, 0.09	Age	Average or below average	0.11	0.06, 0.16
					Above average	0.01	−0.03, 0.06
	MVPA – total	0.17	0.14, 0.20	Age	Average or below average	0.24	0.20, 0.28
					Above average	0.11	0.07, 0.16
Sitting	Sedentary timeb	0.37	0.35, 0.40	% total PA time spent on occupation	Average or below average	0.37	0.34, 0.41
		Curvilinear (Figure 2)			Above average	0.30	0.25, 0.35
				% total PA time spent on bicycling for transport	Average or below average	0.38	0.34, 0.42
					Above average	0.31	0.24, 0.37

^aAccelerometry-based estimates of physical activity and sedentary time were adjusted for accelerometer wear time. Associations were adjusted for clustering at the administrative units and city levels.

^bConceptually-matched variables.

^cIntensity-matched variables.

– bouts, amount of physical activity accumulated in ≤10-minute bouts; CI, confidence intervals; IPAQ-LF, International Physical Activity Questionnaire – Long Form; LMPA, light-to-moderate intensity physical activity; LVPA, light-to-vigorous intensity physical activity; MPA, moderate intensity physical activity; MVPA, moderate-to-vigorous intensity physical activity; PA, physical activity; VPA, vigorous intensity physical activity.

Weight status, age, sex, and percentages of total self-reported PA time spent on occupation and/or bicycling for transport moderated the strength of relationships between some pairs of IPAQ-LF and accelerometry-based estimates of PA (Table 4). Stronger relationships were observed in obese, younger participants, and those with smaller percentages of total PA time spent on occupation and bicycling for transport. Men had stronger correlations of accelerometry-based PA variables with total PA and MPA + walking from IPAQ-LF, while women showed stronger correlations with respect to self-reported VPA.

Differences (absolute disagreement) between IPAQ-LF with accelerometry-based PA and SB variables

The mean differences between conceptually- and intensity-matched IPAQ-LF and accelerometry-based PA variables were large (Table 5 and Figure 1; also see Table, Supplemental Digital Content 3, Correlates of differences in IPAQ-LF estimates of moderate intensity physical activity plus walking, vigorous intensity physical activity, sitting time and corresponding accelerometry-based estimates of physical activity and sedentary time). For example, participants' self-reported total PA was higher by 1,222 and 1,040 min/week than the respective conceptually- and intensity-matched accelerometry-based variables (Table 5). In contrast, the means of total PA and MPA + walking as measured by IPAQ-LF did not significantly differ from those of their accelerometry-based ≥10-minute bouts counterparts including light-intensity PA. Also total PA as measured by the IPAQ-LF did not differ from accelerometry-based ≥10-minute bouts of LMPA in three English-speaking cities (Figure 1). Total MVPA was the accelerometry-based variable with the highest level of agreement with IPAQ-LF VPA (45 min/week difference, see Table, Supplemental Digital Content 3, Correlates of differences in IPAQ-LF estimates of moderate intensity physical activity plus walking, vigorous intensity physical activity, sitting time and corresponding accelerometry-based estimates of physical activity and sedentary time; pair 12 in Table 2). Participants reported 3 fewer h/day of sitting as compared to accelerometry-based sedentary time (see Table, Supplemental Digital Content 3, Correlates of differences in IPAQ-LF estimates of moderate intensity physical activity plus walking, vigorous intensity physical activity, sitting time and corresponding accelerometry-based estimates of physical activity and sedentary time).

Table 5.

Correlates of differences (Δ) in IPAQ-LF estimates of total physical activity and four accelerometry-based estimates of physical activity

	Pairs of PA variables: [IPAQ-LF] – [Accelerometry-based]a mean Δ min/week (95% CI)
	[Total PA] – [MVPA – bouts]b		[Total PA] – [MVPA – total]c		[Total PA] – [LVPA – bouts]		[Total PA] – [LVPA – total]
	1,222 (1,086, 1,357)d		1,040 (907, 1,172)d		182 (−40, 404)		−1,092 (−1,303, −881)d
Correlate	b	95% CI	b	95% CI	b	95% CI	b	95% CI
Intercept	952d	879, 1,026	743d	670, 815	−416d	−492, −339	−1,609d	−1,684, −1,535
City (ref. Ghent – Belgium)
Olomouc – Czech Republic	441d	316, 566	410d	285, 534	773d	625, 921	679d	533, 826
Hradec Kralove – Czech Republic	455d	290, 620	452d	289, 615	828d	635, 1,020	749d	558, 940
Aarhus – Denmark	556d	441, 671	576d	462, 691	1,103d	960, 1,246	1,097d	956, 1,237
Pamplona – Spain	413d	307, 519	395d	289, 500	857d	723, 992	711d	580, 843
Stoke-on-Trent (UK)	314d	146, 482	315d	150, 484	608d	416, 800	489d	299, 680
Seattle (USA)	289d	214, 364	290d	216, 365	496d	388, 604	397d	294, 501
Baltimore (USA)	292d	211, 373	310d	230, 391	429d	497, 641	466d	358, 374
Age (years)	7d	3, 7	7d	5, 9			3e	1, 6
Weight status (ref. Normal)
Overweight	−32	−90, 25	−14	−71, 43
Obese	−143d	−219, −67	−104e	−179, −28
Education (ref. no tertiary degree)
Tertiary degree	−193d	−246, −138	−192d	−246, −138
Sex (ref. Male)
Female	86d	36, 137	125d	74, 175	−104d	−157, −51	−100d	−154, −46
% total PA time spent on leisure	−3d	−5, −2	−4d	−5, −3			−1f	−3, 0
% total PA time spent on occupation	14d	13, 15	14d	13, 15	11d	10, 12	9d	8, 10

^aAccelerometry-based estimates of physical activity were adjusted for accelerometer wear time. Associations were adjusted for clustering at the administrative units and city levels. Continuous correlates are centered around their mean. Only significant correlates were included.

^bConceptually-matched variables

^cIntensity-matched variables

^dP<0.001

^eP<0.01

^fP<0.05

– bouts, amount of physical activity accumulated in ≥10-minute bouts; b, regression coefficient; CI, confidence interval; IPAQ-LF, International Physical Activity Questionnaire – Long Form; LVPA, light-to-vigorous intensity physical activity; min/week, minutes per week; PA, physical activity; ref., reference category.

Figure 1.

Differences between total physical activity (PA) as measured by International Physical Activity Questionnaire – Long Form (IPAQ-LF) and accelerometry-based estimates of PA adjusted for confounders. Solid lines are 95% confidence intervals.

Total PA (IPAQ-LF) minus Moderate-to-vigorous PA accumulated in ≥10-minunte bouts (accelerometry) (conceptually-matched variables)

Total PA (IPAQ-LF) minus Moderate-to-vigorous PA – total (accelerometry) (intensity-matched variables)

Total PA (IPAQ-LF) minus Light-to-vigorous PA accumulated in ≥10-minunte bouts (accelerometry)

Total PA (IPAQ-LF) minus Light-to-vigorous PA – total (accelerometry)

The mean differences between pairs of IPAQ-LF and accelerometry-based PA variables varied by city, with Ghent (Belgium) consistently showing the smallest differences for conceptually- and intensity-matched pairs of variables (Table 5 and Figure 1; see Table, Supplemental Digital Content 3, Correlates of differences in IPAQ-LF estimates of moderate intensity physical activity plus walking, vigorous intensity physical activity, sitting time and corresponding accelerometry-based estimates of physical activity and sedentary time). The differences between IPAQ-LF and accelerometry-based conceptually- and intensity-matched pairs of variables depended on a larger number of moderators than did those of other pairs of variables. Age, weight status, education, sex, and percentage of total PA time spent on leisure, occupation, and/or bicycling for transport were correlates of differences between IPAQ-LF and accelerometry-based estimates of PA (Table 5; see Table, Supplemental Digital Content 3, Correlates of differences in IPAQ-LF estimates of moderate intensity physical activity plus walking, vigorous intensity physical activity, sitting time and corresponding accelerometry-based estimates of physical activity and sedentary time).

Discussion

This study examined factors determining the agreement between IPAQ-LF and accelerometry-based estimates of PA and SB in a multi-country sample of adults (n=3,865). Self-reported total sitting, followed by total PA, showed the highest levels of agreement with accelerometry-based variables as in previous studies (9, 21). IPAQ-LF tended to show higher estimates of MVPA when compared to accelerometry, and less sitting time than accelerometry-based estimates of SB. The level of agreement was markedly influenced by the way accelerometry data were operationalized. Better agreement was observed when accelerometry-based variables included either total or ≥10-minute bout PA accrued at lower intensity levels (according to the cut-points by Freedson et al.) (15) than accelerometer scoring that was nominally the same intensity range as IPAQ-LF. This pattern of findings was observed across all study sites and for all IPAQ-LF PA variables, suggesting a generalizable and robust principle similar as in some previous studies (5, 12, 23).

Total self-reported PA and MPA including walking showed a higher level of absolute agreement with accelerometry-based LVPA (Figure 1) and LMPA accumulated in ≥10-minute rather than ≥1-minute bouts (see mean difference in Table 3). The opposite held true for self-reported VPA. Similar results were found in a validation study of the IPAQ-LF in older Chinese (5). Overall, respondents may be unable to provide valid estimates of PA intensity but may recall relatively accurately the time spent in sustained discrete bouts of activity. The fact that this was not the case for VPA may be due to the low prevalence of such intensity levels in the examined sample or to misclassification of VPA.

Although no significant differences in relative agreement (relationships) between IPAQ-LF and accelerometry-based estimates of PA and SB were observed across study sites, significant differences were observed in absolute disagreement (differences in means). Ghent (Belgium) had the highest level of agreement between conceptually- and intensity-matched pairs of PA and SB estimates. This might be due to Ghent being the only site that administered the IPAQ-LF via in-person interviews, whereby interviewers assisted participants in giving realistic estimates of PA. Interviewer administration of self-report measures is known to minimize errors associated with the misinterpretation of items (38), and interviewer-administered versions of the IPAQ have been reported to yield more valid estimates of PA and SB than self-completed forms (21).

After adjustment for potential moderators, differences in absolute agreement were similar across sites with a common language and culture. Seattle (USA), Baltimore (USA), and Stoke-on-Trent (UK) showed similar average absolute differences between self-report and accelerometry-based variables. The same was observed for the Czech sites (Olomouc and Hradec Kralove), while, in most cases, Aarhus (Denmark) showed the largest discrepancies between IPAQ-LF and accelerometry-based estimates of PA. This finding is unlikely to have arisen due to the higher prevalence of cycling in the Danish sample, as analyses were adjusted for self-reported proportion of PA time spent on cycling for transport, and the Belgian sample also reported higher proportions of PA time devoted to cycling for transport. Also, a recent study comparing PA estimates derived from the IPAQ – Short Form with objectively-assessed PA using accelerometry plus heart rate monitors found large between-method differences in PA estimates (33). These findings are suggestive of systematic cultural and/or linguistic differences in the interpretation of the IPAQ-LF items, which have implications for the interpretation of international PA and SB surveillance data as well as findings from multi-country studies.

Assuming uncorrelated measurement errors, the fact that relative agreement between IPAQ-LF and accelerometry-based PA and SB variables was not moderated by study city is reassuring as it supports the comparability of site-specific findings on correlates of self-reported and objectively-assessed PA and SB across various cultures and linguistic areas. However, the presence of potential culture- and linguistic-specific differences in absolute agreement raises concerns about the comparability of PA prevalence estimates.

As observed in other studies (22, 32), the relative and/or absolute agreement between self-reported and accelerometry-based PA and SB variables were age, sex, education, and weight-status specific. This may be due to differences in fitness level, ability to understand the questionnaire items, adequacy of the accelerometry cut-points used in this study, habitual patterns of PA and time use, and other factors. It is an issue that needs to be considered when interpreting socio-demographic differences in levels and correlates of PA and SB.

The types/domains of PA engaged in are another important factor that needs to be considered. Relative and absolute agreement depended on the percentage of total PA time spent on bicycling for transport, leisure-time PA, and occupational PA. As originally hypothesized, larger proportions of PA time spent on cycling for transport and occupational PA were associated with lower levels of agreement between self-report and accelerometry-based variables, while the opposite was true for percentage of PA time spent on leisure activities. The negative effect of time spent bicycling is likely due to accelerometers failing to pick up bicycling, while the effect of time spent on occupational activities may be a result of respondents likely reporting time spent at work rather than time spent being moderately or vigorously active at work (3). The latter problems may be in part overcome by interview administration of IPAQ-LF or by prompting the participants to avoid reporting light PA at work, and providing a better description of occupational MVPA.

The present study had several strengths. It was conducted in a large international community sample recruited from a balanced number of high/low socio-economic status and high/low walkable neighborhoods across six countries. This recruitment strategy enhances the generalizability of the findings. It is the first study to examine an extensive and theoretically justifiable number of measurement-related, socio-demographic, as well as behavioral factors affecting the agreement of self-report and accelerometry-based estimates of PA and SB. Participants were not aware the data would be used for a measurement study, probably enhancing generalizability of the findings. Validity studies may overestimate agreement between self-reports and accelerometry-based estimates of PA and SB because they request participants to wear the accelerometer for 7 days and keep a wear-time diary prior to completing the questionnaire. This may positively affect the retrieval of information on PA and SB in the past week. Assessing self-reported PA and SB before conducting an objective assessment may provide a more realistic measure of agreement between the two measurement methods and a conservative validity estimate of IPAQ-LF.

The study has also several limitations. Self-reports did not cover the same week of the accelerometer monitoring in all cities except two, so this could have underestimated agreement. We restricted our comparison of IPAQ-LF with accelerometry-based estimates of PA and SB based on the most-commonly-used Freedson's cut-points. The use of other cut-points would have resulted in different estimates of agreement (36). Waist-worn accelerometers poorly differentiate sitting and reclining from standing (7). Hence, accelerometry-based estimates of SB and self-report measures of sitting time are only weakly conceptually matched. Also, regardless of their placement these devices have substantial issues in capturing bicycling, water-based activities, upper-body exercise, or resistance activities such as carrying loads or stair climbing (11). Thus, accelerometers are not considered a ‘gold standard’ method of measurement of PA and SB. Comparisons between study sites using older vs. newer accelerometer models might have been somewhat confounded by differences in accelerometer performance (4, 31).

Conclusions

Our findings suggest that, when compared to accelerometry, IPAQ-LF tends to show lower estimates of time spent sitting and higher estimates of total time spent in MVPA. While relative agreement between methods did not differ across sites, absolute agreement did, even after adjusting for socio-demographic and behavioral moderators (percentage of total PA time spent in specific activities). These issues need to be carefully considered when interpreting PA and SB data collected in diverse cultures, language regions, and socio-demographic groups for the purpose of surveillance or investigation of determinants of PA and SB. The findings raise questions about the comparability of PA and SB data across countries and call for further efforts to improve such measures.