Table 6.
BMI associations for Fitbit versus IPAQ by overall activity and total active time in multivariable logistic regressionsa.
| Activity predictor | Unadjusted | Adjusted for demographics | Adjusted for demographics, health-related behaviors, and clinical characteristics | |||||||||||
|
|
Coefficient (95% CI) | P value | Coefficient (95% CI) | P value | Coefficient (95% CI) | P value | ||||||||
| Overall activity per 1 SD/wkb | ||||||||||||||
|
|
Separate predictor models | .001 |
|
.007 |
|
.01 | ||||||||
|
|
|
Fitbit | –1.65c (–2.15 to –1.16) |
|
–1.51c (–1.99 to –1.03) |
|
–1.37c (–1.84 to –0.90) |
|
||||||
|
|
|
IPAQd | –0.80e (–1.23 to –0.35) |
|
–0.82c (–1.24 to –0.39) |
|
–0.72e (–1.14 to –0.29) |
|
||||||
|
|
Combined predictor model | .001 |
|
.005 |
|
.01 | ||||||||
|
|
|
Fitbit | –1.57c (–2.09 to –1.05) |
|
–1.40c (–1.90 to –0.89) |
|
–1.28c (–1.78 to –0.77) |
|
||||||
|
|
|
IPAQ | –0.24 (–0.66 to 0.19) |
|
–0.30 (–0.72 to 0.12) |
|
–0.25 (–0.69 to 0.19) |
|
||||||
| Total active time (h/wk) | ||||||||||||||
|
|
Separate predictor models | <.001 |
|
<.001 |
|
.002 | ||||||||
|
|
|
Fitbit | –0.23c (–0.30 to –0.16) |
|
–0.22c (–0.28 to –0.15) |
|
–0.20c (–0.26 to –0.13) |
|
||||||
|
|
|
IPAQ | –0.10c (–0.15 to –0.04) |
|
–0.10c (–0.15 to –0.05) |
|
–0.09e (–0.14 to –0.04) |
|
||||||
|
|
Combined predictor model | .001 |
|
.002 |
|
.006 | ||||||||
|
|
|
Fitbit | –0.21c (–0.28 to –0.14) |
|
–0.20c (–0.27 to –0.13) |
|
–0.18c (–0.25 to –0.11) |
|
||||||
|
|
|
IPAQ | –0.04 (–0.10 to 0.01) |
|
–0.05 (–0.10 to 0.01) |
|
–0.04 (–0.09 to 0.01) |
|
||||||
aAll models were adjusted for Fitbit wear-time, data collection season, and Fitbit device wear location. Demographics include age, gender, education, income, race, and ethnicity. This table shows the BMI difference (kg/m2) per predictor increase (95% CI). Health-related behaviors include smoking and alcohol use. Clinical characteristics include coronary artery disease, diabetes, hyperlipidemia, and hypertension. Regressions accounted for multiple observations with robust standard errors clustered by participant. P values shown are for Wald tests comparing Fitbit versus IPAQ coefficients.
b1 SD for overall activity was 3663 steps/d for Fitbit and 37.2 MET-h/wk for International Physical Activity Questionnaire.
cThese values were significant at P<.001 for cluster robust association with BMI (586 participant clusters).
dIPAQ: International Physical Activity Questionnaire.
eThese values were significant at P<.01 for cluster robust association with BMI (586 participant clusters).