Measures of physical activity using cell phones: validation using criterion methods

Abstract

Background: Physical activity is associated with reduced risks of many chronic diseases. Data collected on physical activity in large epidemiological studies is often based on paper questionnaires. The validity of these questionnaires is debated, and more effective methods are needed.

Objective: This study evaluates repeated measures of physical activity level (PAL) and the feasibility of using a Java-based questionnaire downloaded onto cell phones for collection of such data. The data obtained were compared with reference estimates based on the doubly labeled water method and indirect calorimetry (PAL(ref)).

Method: Using a Java-based cell phone application, 22 women reported their physical activity based on two short questions answered daily over a 14-day period (PAL(cell)). Results were compared with reference data obtained from the doubly labeled water method and indirect calorimetry (PAL(ref)). Results were also compared against physical activity levels assessed by two regular paper questionnaires completed by women at the end of the 14-day period (PAL(quest1) and PAL(quest2)). PAL(cell), PAL(quest1), and PAL(quest2) were compared with PAL(ref) using the Bland and Altman procedure.

Results: The mean difference between PAL(cell) and PAL(ref) was small (0.014) with narrow limits of agreement (2SD = 0.30). Compared with PAL(ref), the mean difference was also small for PAL(quest1) and PAL(quest2) (0.004 and 0.07, respectively); however, the limits of agreement were wider (PAL(quest1), 2SD = 0.50 and PAL(quest2), 2SD = 0.90). The test for trend was statistically significant for PAL(quest1) (slope of regression line = 0.79, P = .04) as well as for PAL(quest2) (slope of regression line = 1.58, P < .001) when compared with PAL(ref).

Conclusion: A Java-based physical activity questionnaire administered daily using cell phones produced PAL estimates that agreed well with PAL reference values. Furthermore, the limits of agreement between PAL obtained using cell phones, and reference values were narrower than for corresponding estimates obtained using paper questionnaires. Java-based questionnaires downloaded onto cell phones may be a feasible and cost-effective method of data collection for large-scale prospective studies of physical activity.

Figure 1

Schematic figure of the data communication between the cell phone, web application server, study-specific webpage, and the database

Figure 2

Bland and Altman plot comparison of physical activity level obtained using cell phones during 14 days (PAL_cell) and physical activity level obtained using a combination of the doubly labeled water method and indirect calorimetry (PAL_ref)

Figure 3

Bland and Altman plot comparison of physical activity level obtained using the first questionnaire (PAL_quest1) and physical activity level obtained using a combination of the doubly labeled water method and indirect calorimetry (PAL_ref)

Figure 4

Bland and Altman plot comparison of physical activity level obtained using the second questionnaire (PAL_quest2) and physical activity level obtained using a combination of the doubly labeled water method and indirect calorimetry (PAL_ref)

Figure 5

Daily PAL values obtained using cell phones during day 1 through day 14 for twelve selected women compared with PAL_ref (covering the whole 14-day period and shown as a straight dotted line for each woman)