Workers’ physical activity data contribute to estimating maximal oxygen consumption: a questionnaire study to concurrently assess workers’ sedentary behavior and cardiorespiratory fitness

Abstract

Background

Sedentary behavior (SB) and cardiorespiratory fitness (CRF) are important issues in occupational health. Developing a questionnaire to concurrently assess workers’ SB and CRF could fundamentally improve epidemiological research. The Worker’s Living Activity-time Questionnaire (WLAQ) was developed previously to assess workers’ sitting time. WLAQ can be modified to evaluate workers’ CRF if additional physical activity (PA) data such as PA frequency, duration, and intensity are collected.

Methods

A total of 198 working adults (93 women and 105 men; age, 30–60 years) completed anthropometric measurements, a treadmill exercise test for measuring maximal oxygen consumption (VO_2max), and modified WLAQ (m-WLAQ, which included questions about PA data additional to the original questions). Multiple regression analyses were performed to develop prediction equations for VO_2max. The generated models were cross-validated using the predicted residual error sum of squares method. Among the participants, the data of 97 participants who completed m-WLAQ twice after a 1-week interval were used to calculate intraclass correlation coefficient (ICC) for the test–retest reliability analyses.

Results

Age (r = − 0.29), sex (r = 0.48), body mass index (BMI, r = − 0.20), total sitting time (r = − 0.15), and PA score (total points for PA data, r = 0.47) were significantly correlated with VO_2max. The models that included age, sex, and BMI accounted for 43% of the variance in measured VO_2max [standard error of the estimate (SEE) = 5.04 ml·kg^− 1·min^− 1]. These percentages increased to 59% when the PA score was included in the models (SEE = 4.29 ml·kg^− 1·min^− 1). Cross-validation analyses demonstrated good stability of the VO_2max prediction models, while systematic underestimation and overestimation of VO_2max were observed in individuals with high and low fitness, respectively. The ICC of the PA score was 0.87 (0.82–0.91), indicating excellent reliability.

Conclusions

The PA score obtained using m-WLAQ, rather than sitting time, correlated well with measured VO_2max. The equation model that included the PA score as well as age, sex, and BMI had a favorable validity for estimating VO_2max. Thus, m-WLAQ can be a useful questionnaire to concurrently assess workers’ SB and CRF, which makes it a reasonable resource for future epidemiological surveys on occupational health.

Background

Many studies have shown that excessive sedentary behavior (SB) increases disease risk [1, 2]. In this mechanized society, workers are particularly likely to be placed in sedentary situations in the workplace [3, 4]. Therefore, workers’ SB is an important risk factor for occupational health. Similarly, cardiorespiratory fitness (CRF) is a conventional health issue because many studies have shown that low CRF level is strongly associated with increased disease and mortality risks [5]. Recent studies have indicated that midlife CRF plays a role in health-related incidents in later life such as the development of severe diseases [6], increased healthcare costs [7], and decreased longevity [8].

Thus, from the perspective of preventative medicine, both SB and CRF in working adults are key factors in occupational health. However, the relationship between workers’ SB and CRF as well as their interaction effects on disease risk has rarely been investigated in epidemiological surveys. One crucial reason would be inherent in the assessment methodology. The gold standard methods for SB and CRF assessment, such as thigh-worn inclinometer including the activPAL [9] for SB and maximal oxygen consumption (VO_2max) for CRF, require relatively high cost and considerable time; thus, they have a practical disadvantage for a population-based assessment. In epidemiological surveys, subjective measures such as questionnaires remain useful because they are more cost-effective and present a lower participant burden [10], although the key limitation of the questionnaire method is poor validity. To the best of our knowledge, validated questionnaires to simultaneously assess workers’ SB and CRF have not been proposed.

The Worker’s Living Activity-time Questionnaire (WLAQ) was primarily developed to assess workers’ sitting times in our previous studies [11, 12]. WLAQ can be used to measure a worker’s time spent sitting within four typical domains of a worker’s life: (a) working time, (b) commuting time, (c) nonworking time on a workday, and (d) time on a non-workday. Matsuo et al. [12] evaluated WLAQ and demonstrated favorable test–retest reliability values and criterion (vs. activPAL) validity values for the four sitting times.

Age, sex, and body fat-related values have often been used in VO_2max estimation models [13,14,15]. Given that measurement accessibilities differ among different body fat-related values such as body mass index (BMI), waist girth (WG), and %fat, these previous studies [13,14,15] investigated the predictive power for each body fat item. Furthermore, previous studies [14,15,16,17,18] have shown that physical activity (PA) data, such as frequency, duration, and intensity, contribute to the estimation of VO_2max. However, the original WLAQ can help assess workers’ sitting times but not other PA data. Thus, WLAQ can be modified for evaluating workers’ CRF if additional PA data are collected, and it can be used along with age, sex, and body fat-related values to develop an equation model for VO_2max estimation.

Therefore, the purposes of this study were 1) to investigate associations between measured VO_2max and sitting times and other PA data collected using the modified WLAQ (m-WLAQ); 2) to investigate criterion validity of a developed equation model for estimating VO_2max; 3) to compare the accuracy of equation models that include BMI, WG, and %fat; and 4) to investigate the test–retest reliabilities of values derived using m-WLAQ.

Methods

Participants

The inclusion criteria were as follows: 1) aged 30–60 years, 2) living in the Tokyo area (Tokyo, Saitama, Chiba, and Kanagawa Prefectures) of Japan, 3) part-time or full-time worker for at least 3 days a week, and 4) no medical conditions that could affect VO_2max testing. Participants were recruited through a website advertisement. In total, 202 working adults (97 women and 105 men) participated in this study. The participants visited our laboratory to complete anthropometric measurements, a treadmill exercise test, and m-WLAQ. We excluded 4 participants due to insufficient data for the analyses. Consequently, 198 participants (93 women and 105 men) were included in the validity analysis. Furthermore, among the included participants, 97 participants (42 women and 55 men) visited our laboratory for a second time, with an interval of 1 week between the visits. At the second visit, they completed m-WLAQ again for test–retest reliability analyses.

This study was conducted in accordance with the guidelines proposed in the Declaration of Helsinki. The Ethical Committee of the National Institute of Occupational Safety and Health, Japan reviewed and approved the study protocol (ID H2810). Before obtaining written informed consent, the aims and design of this study were explained to each participant.

Measures

Anthropometric measurements

For each participant, height (to the nearest 0.1 cm) was measured once using a wall-mounted stadiometer (YG-200, Yagami, Nagoya, Japan). Body weight (to the nearest 0.1 kg) and %fat with bioelectrical impedance analysis were assessed using a body composition analyzer (InBody-3.2; Biospace, Seoul, Korea). WG (to the nearest 0.1 cm) was measured twice at the level of the umbilicus in the standing position by a skilled member of the research staff. BMI was calculated as weight (in kilograms) divided by the square of height (in meters).

m-WLAQ

m-WLAQ was used to evaluate participants’ sitting times during working time, commuting time, and leisure time on a workday and non-workday. A previous study [12] using the original WLAQ demonstrated favorable test–retest reliability values [intraclass correlation coefficient (ICC) = 0.72–0.98] and criterion (activPAL) validity values (Spearman’s ρ = 0.40–0.82) for the four sitting times. The original WLAQ was modified to add several questions for collecting PA data (frequency, duration, and intensity) for developing m-WLAQ. The PA score (0–44 points) was calculated as the sum of the points scored for PA data. For its calculation, PA intensity was weighted more heavily than PA duration and frequency on the PA score as per previous studies [17, 19]. An additional PDF file shows the m-WLAQ and calculation method for each value (See Additional file 1).

Maximal oxygen consumption

The participants underwent an electrocardiogram-monitored, exhaustion-limited, graded exercise test on a treadmill (EXCITE RUN, Technogym, Cesena, Italy) using the Bruce protocol to determine VO_2max. During the test, an open-circuit computerized indirect calorimeter (AE-310S, Minato Medical Science, Osaka, Japan) was used to measure ventilation and expired gases. The gas analyzer was calibrated before each trial. Heart rate (HR) was monitored using an electrocardiogram monitor (LifeScope, NIHON KOHDEN, Tokyo, Japan), and a rating of perceived exertion (RPE), using the 6–20 Borg RPE scale, was recorded during the exercise test. The highest 30-s average VO₂ value was defined as the VO_2max value. The exercise test was considered to achieve VO_2max when three of the following four criteria were satisfied: 1) respiratory exchange ratio > 1.10, 2) achievement of maximum HR within 10 bpm of the age-predicted maximal (220 − age), 3) RPE > 17, and 4) VO₂ plateau despite further increases in workload [20, 21].

Data analysis

Unpaired Student’s t-tests were performed to evaluate differences between groups. Chi-squared tests were used to analyze categorical variables. Pearson’s correlation coefficients were calculated to evaluate the relationship between the measured VO_2max and other measurement values. Multiple linear regression analysis was used to develop prediction equations for VO_2max. Changes in the squared multiple correlation coefficient (R²) and the standard error of the estimate (SEE) were used to assess the incremental gain in variance explained by the different variables added to the model. In the course of previous studies [13,14,15], R² and SEE were compared among some types of body fat evaluation models, i.e., BMI, WG, and %fat models, to investigate the influence of differences in methodology, because subjects’ body fat is assessed in several ways in epidemiological surveys. The generated models were cross-validated using the predicted residual error sum of squares (PRESS) statistical method [22]. This method calculates the error in prediction for each case when only that case is excluded from generating the estimation model and applying the model to the excluded case. PRESS adjusted R² (R²_p) is calculated as 1 − (PRESS statistic/SS _total), whereas PRESS SEE (SEE_p) is calculated using the following equation: SEE_p = $\sqrt{PRESS\ statistic/N}$. The generated models were further validated by comparing the constant errors (CEs) among the subgroups of sex, age, and VO_2max levels. CE is calculated as the mean difference between measured VO_2max and predicted VO_2max. The test–retest reliability was examined using ICC and 95% confidence interval (95% CI) with an ICC of < 0.40 indicating poor repeatability, 0.40–0.75 indicating fair-to-good repeatability, and > 0.75 indicating excellent repeatability [23]. P-value of < 0.05 was considered statistically significant. All analyses were conducted using SAS, version 9.4 (SAS Institute Japan, Tokyo, Japan).

Results

Table 1 summarizes the demographic characteristics of the participants. We observed higher anthropometric measurements in men than in women. Although sitting times during worktime and on a non-workday were significantly longer in men than in women, sitting time during leisure time on workday was significantly longer in women than in men. The percentage of clerical jobs was high in both sexes, but the percentages of sales and marketing and engineer/researcher were higher in men.

Table 1 Descriptive characteristics of the study participants (n = 198)

Table 2 shows the questions included in m-WLAQ, answer options (and their assigned points), and the results of the study participants. Although the answer trends were different between women and men for Q8 and Q10, the PA score showed no significant sex difference.

Table 2 Questions of m-WLAQ for CRF estimation and the results of this study participants (n = 198)

Table 3 shows Pearson’s correlation coefficients between the measured VO_2max and other variables. Participants who were older, women, and those with higher body fat-related values had lower VO_2max. Significant negative correlations were observed between VO_2max and sitting time during leisure time on workday (r = − 0.27, P < 0.01) and total sitting time (r = − 0.15, P = 0.03), although sitting times during commuting time, working time, and non-workday were not significantly correlated with VO_2max. Significant positive correlations were observed between PA data (frequency, duration, and intensity) and measured VO_2max with the largest association (r = 0.47, P < 0.01) revealed for the PA score.

Table 3 Pearson’s correlation coefficients between measured VO_2max and other values (n = 198)

The variables significantly related to the measured VO_2max were included as predictors in multiple regression models (Table 4). The model that included age, sex, and BMI accounted for 43% of the variance in measured VO_2max (adjusted R² = 0.43, SEE = 5.04 ml·kg^− 1·min^− 1 or 13.1%). When the PA score obtained using m-WLAQ was added to the model that included age, sex, and BMI, the adjusted R² significantly improved by 16% (adjusted R² = 0.59, SEE = 4.29 ml·kg^− 1·min^− 1 or 11.2%). Similarly, regarding the models using WG or %fat instead of BMI, addition of the PA score improved the variance of the models by 12% or 11%, respectively. The cross-validation results from the PRESS method are also shown in Table 4. The decrease in R² (approximately 0.01) and increase in SEE value (approximately 0.05 ml·kg^− 1·min^− 1) were small for all three models.

Table 4 Regression coefficients for prediction of VO_2max using selected independent values (n = 198)

Table 5 presents the results of other cross-validation analyses using CE values. The absolute CE values for the sex and age subgroups were < 1.00 for all three models. Regarding the VO_2max subgroups, CE values were negatively high (overestimation) for the low-fitness subgroup and positively high (underestimation) for the high-fitness subgroup, whereas lower CE values were observed for the mid-fitness subgroup in all three models.

Table 5 Constant error for subgroups of sex, age, and measured VO_2max (n = 198)

The test–retest reliabilities were examined using the data of the 97 participants who completed m-WLAQ twice. Table 6 shows the ICC of sitting times and PA-related values obtained using m-WLAQ. The ICC values for sitting times were fair to good (commuting time and non-workday) or excellent (working time and leisure time on workday and total sitting time). All ICC values of PA-related values were excellent except those for exercise duration on non-workday and exercise intensity during leisure time on workday and non-workday, whose values were fair to good. The ICC value of the PA score was 0.87, indicating excellent reliability.

Table 6 Test–retest reliability of sitting times and PA-related values measured using data obtained with m-WLAQ at sitting times 1 and 2 (n = 97)

The relationships between measured VO_2max and estimated VO_2max are shown in Fig. 1. Estimated VO_2max correlated well with measured VO_2max for all three body fat models; Fig. 1 also shows overestimations in participants with low fitness and underestimations in participants with high fitness for all three models.

Fig. 1

Relationship between measured and estimated VO_2max in (A) BMI, (B) WG, and (C) %fat models. The regression line and correlation coefficient with P values are displayed. The dashed lines are the identity lines of the measured and estimated VO_2max.

Discussion

The study showed that 1) the PA score obtained using m-WLAQ, rather than sitting times, was associated with measured VO_2max; 2) the equation models that included age, sex, body fat-related values, and PA score obtained using m-WLAQ had favorable validity for estimating VO_2max; 3) no appreciable difference was observed in estimated VO_2max among the three models with regard to BMI, WG, and %fat; and 4) favorable reliability values were shown for sitting times and the PA score obtained using m-WLAQ.

Consistent with the findings of a previous study [24], significant negative correlations were observed between sitting times and measured VO_2max (Table 3). However, sitting times were not accepted as an effective explanatory variable for estimating VO_2max in our regression analyses. In contrast, questionnaire-based PA data, such as frequency, duration, and intensity, were significantly correlated with measured VO_2max (Table 3), and the regression model identified the PA score to be the principal explanatory value for the equation models. The PA score was calculated for precise VO_2max prediction in reference to the HUNT study by Nes et al. [17] and a previous exercise intervention study [19]. Nes at al [17]. used some question items regarding PA frequency, duration, and intensity for estimating VO_2max and relative weightings of different responses were set on the basis of their relation to VO_2max. In their estimation, PA intensity was weighted more heavily than PA duration and frequency on the PA score. Further, an exercise intervention study [19] emphasized the primacy of PA intensity rather than PA duration and volume in improving VO_2max. We followed these studies to develop the PA score, i.e., the questions regarding intensity, such as Q6, Q10, and Q15, were weighted more heavily than other questions (Table 2). The PA score was strongly correlated with VO_2max (Table 3) and functioned well for estimating VO_2max (Table 4). The results of previous studies and the present study suggest that PA intensity can have a potential role in estimating VO_2max.

Age, sex, and body fat-related values were significantly correlated with measured VO_2max (Table 3), and these three factors accounted for 43–51% of the variance in measured VO_2max (Table 4). These percentages increased by 11–16% following to addition of the PA score obtained using the questionnaire in the model (Table 4). Jackson et al. [14] suggested questionnaire-based VO_2max prediction models including age, sex, body fat-related values, and the PA score obtained using the questionnaire, demonstrating SEEs of 5.35–5.70 ml·kg^− 1·min^− 1. Similarly, Wier et al. [15] suggested questionnaire-based VO_2max prediction models including age, sex, body fat-related values, and the PA score obtained using the questionnaire, demonstrating SEEs of 4.72–4.90 ml·kg^− 1·min^− 1. Furthermore, Malek et al. [25] developed a VO_2max prediction equation including age, body weight, height, and questionnaire-based exercise values, which had an SEE value of 4.12 ml·kg^− 1·min^− 1. The present study showed results similar to these previous studies, with prediction model SEEs of 4.13–4.29 ml·kg^− 1·min^− 1. These SEE values can be replaced with %SEE values of 10.8–11.2%. Other types of VO_2max estimation studies reported %SEE values of 11.4% in the 20-m shuttle run test study [26] and 10–15% in wearable device studies [13, 27, 28]. The SEE values in the present study seem to be favorable when compared with those calculated in other VO_2max prediction studies.

Regarding the method to validate a regression equation, although the data-splitting method is well known, in which the entire data are divided into a fitting group and validation group, the PRESS method [22] is particularly recommended for studies with a small sample size. This method can provide useful diagnostics while avoiding the disadvantages of the data-splitting method such as lack of equation stability due to diluted sample size. In fact, studies with a large sample size, such as those of Jackson et al. (1999 participants) [14] and Nes at al. (4637 participants) [17] used the data-splitting method. However, the PRESS method has not only been used in studies with a large sample size such as in that of Matthews et al. (799 participants) [29] and Wier at al. (2801 participants) [15] but also in studies with a small sample size such as those of Malek et al. (115 participants) [25] and Cao et al. (148 participants) [13]. The PRESS method appeared to be appropriate for the present study on 198 participants.

Jackson et al. [14] recommended questionnaire-based VO_2max prediction models including age, sex, the PA score obtained using the questionnaire, and body fat-related values such as %fat (skinfold method) and BMI, and they demonstrated SEE values of 5.35 ml·kg^− 1·min^− 1 for the %fat model and 5.70 ml·kg^− 1·min^− 1 for the BMI model. Wier et al. [15] also recommended questionnaire-based VO_2max prediction models including age, sex, the PA score obtained using the questionnaire, and body fat-related values such as %fat (skinfold method), WG, and BMI and they showed no considerable differences in accuracy among the three models using WG (SEE value of 4.80 ml·kg^− 1·min^− 1), %fat (SEE value of 4.72 ml·kg^− 1·min^− 1), or BMI (SEE value of 4.90 ml·kg^− 1·min^− 1). The present study obtained results similar to those of previous studies, i.e., no considerable difference was observed in accuracy among the three body fat-related variables, i.e., BMI, WG, and %fat (bioelectrical impedance analysis). Although the SEE value of the BMI model (4.29 ml·kg^− 1·min^− 1 or 11.2%) was relatively higher than those of the WG (4.17 ml·kg^− 1·min^− 1 or 10.9%) and %fat (4.13 ml·kg^− 1·min^− 1 or 10.8%) models, which are consistent with the findings reported by Wier et al. [15], the BMI model could be more convenient than the other models because BMI is a basic and less burdensome assessment item in adult health checkups. Therefore, the following equation model is suggested for VO_2max estimation in the present study (using sex = 0 for women and 1 for men): VO_2max = 59.96 + (− 0.23 × age) + (7.39 × sex) + (− 0.79 × BMI) + (0.33 × PA score).

There are some limitations to the present study. First, response bias may have occurred because the participants had advance knowledge of the experimental procedure, i.e., they could decide to participate in this study after viewing our research advertisement, which may have led to greater inclusion of participants preferring PA or exercise. Second, CE analyses (Table 5) and scatter graphs (Fig. 1) showed that the CRF evaluation model derived in the present study significantly underestimated VO_2max in participants with high fitness and overestimated VO_2max in those with low fitness. This systematic error usually occurs in VO_2max estimation studies [15, 17]. As pointed out by other researchers [15, 17], while underestimation in individuals with high fitness may not be a pressing problem because high fitness relates to low disease and mortality risks, overestimation in individuals with low fitness may be more problematic because low fitness relates to increasing disease risks. A correction method such as a compensation formula or including a convenient stress test should be considered to correct the error, particularly in individuals with low fitness. Third, we could not include approximately 50% of the participants in test–retest reliability analyses because they did not participate in the second round of m-WLAQ. Participant selection bias could have occurred because the selection was not conducted at random but in accordance with participant convenience. Fourth, in recent public health research, moderate-to-vigorous intensity PA (MVPA) has been treated as an important terminology separately from SB [30, 31]. MVPA and SB are defined as accelerometry-measured PA of ≥3.0 metabolic equivalents (METs) and PA of ≤1.5 METs, respectively [32]. m-WLAQ can assess SB but not MVPA.

Conclusion

The PA score obtained using m-WLAQ, rather than sitting time, correlated well with measured VO_2max and had a favorable test–retest reliability. The equation model that included PA score with age, sex, and BMI had a favorable validity for estimating VO_2max. Thus, the study suggests that m-WLAQ can be a useful questionnaire to assess workers’ CRF and SB.

The findings of this study could help to advance the quality of future epidemiological surveys in occupational health research fields. On the other hand, given the CRF classification based only on age, sex, BMI, and questionnaire results could impose a limitation for detecting differences among individuals, further research is necessary to advance CRF assessment. Use of sophisticated wearable sensors could improve classification [33].