Many epidemiological studies rely on self-report questionnaires to measure physical activity. Although these subjective instruments are an appropriate method for describing and quantifying behavioral patterns, they are a relatively imprecise way of characterizing the subdimensions of physical activity, such as energy expenditure or load-bearing (4). The development of more objective quantitative methods is desirable to allow the quantification of the relationship between the subdimensions of physical activity and particular health outcomes in order to inform preventive actions. It would also facilitate the monitoring of temporal trends, the demonstration of cross-cultural differences, and the evaluation of interventions aimed at increasing physical activity (27).
The possible objective methods for measuring physical activity are principally movement sensoring and heart rate monitoring, and more recently the combination of both methods (16). Movement sensors have several limitations. They cannot quantitatively estimate activities such as cycling and rowing or weight-bearing movement and are not usually waterproof and therefore cannot be worn during swimming. Heart rate monitoring has shown to be a feasible technique for measuring energy expenditure in field studies (25,26) when measuring minute-by-minute heart rate is interpreted in conjunction with individual-level data on the relationship between heart rate and energy expenditure. It is generally accepted that without that individual calibration, heart rate monitoring is less informative. The HRFlex method, which is one example of a heart rate monitoring approach using individual calibration, requires the measurement of the relationship between heart rate and energy expenditure on exertion together with resting energy expenditure. Due to the nonlinearity between heart rate and energy expenditure at the lower end of the spectrum of heart rate, an empirical heart rate “flex” point (HRFlex) is determined, which is an estimation of the heart rate at which the linear assumption does not hold. This HRFlex is taken as the mean of the highest heart rate at rest and the lowest on exercise (5). Below this point, energy expenditure is assumed to be equal to rest, but above this it is estimated from the slope and intercept of the line between energy expenditure and exercise heart rate. Once this relationship is known, an estimate of energy expenditure can be calculated using the heart rate data collected in the field for each subject. This method has previously been validated both against doubly labeled water and whole-body calorimetry with a correlation of 0.94 (5,13,21). There have only been a few validation studies of physical activity questionnaires against doubly labeled water, and these have, at best, displayed correlations of 0.58 (18).
Although it is now an accepted method for measuring energy expenditure, heart rate monitoring has not been used in large-scale epidemiological studies. Its use is limited not by the cost of the materials, which are relatively inexpensive compared to other methods, but rather by the time required for volunteer and researcher for the measurement and analysis. The calibration test to determine the oxygen consumption-heart rate relationship takes a minimum of 45 min per subject and subsequent processing of this data is also time-consuming. It has been suggested that a broader application of heart rate monitoring in the field could be achieved, particularly in developing countries, if no calibration test was required (11).
The study presented in this paper was undertaken to assess the feasibility of using heart rate monitoring without the need for the individual calibration test, in order to expand its use in epidemiological studies. The standard HRFlex method is based on the measurement of four individual parameters: resting energy expenditure, the slope and the intercept of the regression line between energy expenditure and heart rate during exercise, and the HRFlex. This study was designed to develop predictions for these four parameters from measurements that could be easily obtained in field studies. These estimated parameters, and the estimate of energy expenditure produced using them, were then tested in a separate, independent sample from a population-based cohort study.
The study sample used to derive the calibration parameter predictions was from a continuing population-based cohort study. This cohort was originally randomly selected from a large primary care practice in Ely, Cambridgeshire, in 1990 (30) and has been followed-up subsequently (24). The data presented here are taken from 789 Caucasian subjects who completed heart rate monitoring at the 4.5-yr follow-up (28).
The independent sample used in this study to test the validity of the parameter predictions was randomly selected from the European Prospective Investigation into Cancer (EPIC) cohort in Norfolk (7). This is a population-based cohort study involving the recruitment of 25,000 volunteers aged 40–70 in Norfolk, England, as part of a wider pan-European study that has recruited 450,000 individuals. This sample of 97 subjects was taking part in an energy intake validation study and underwent heart rate monitoring. The method outlined below was identical in both groups. All volunteers gave written informed consent before testing.
Volunteers were invited to attend the clinic for an individual calibration test to calculate the heart rate/energy expenditure relationship. The volunteers were asked to refrain from smoking, drinking caffeinated drinks, and the use of a nicotine patch on the morning of the test. An assessment of resting and exercise oxygen consumption was undertaken, first at rest with the subject lying prone and then seated, using an oxygen analyzer (PK Morgan Ltd, Kent, UK) calibrated daily using 100% nitrogen and fresh air as standard gases. A one-way valve and mouthpiece was used to collect expired air. Volunteers spent a minimum of 15 min lying prone before the commencement of measurements until their heart rate had decreased to a resting level and remained constant for 5 min. The ambient room temperature and barometric pressure were also recorded daily. Subjects were then asked to pedal on a cycle ergometer in order to provide information on heart rate, minute volume of inspired air, expired air oxygen concentration, and breath rate. Each subject cycled at 50 revolutions per minute with the workload increased every 3 min from 0 W through 37.5 W, 75 W, and 125 W. However, the 125-W level was only undertaken in those subjects who had not reached a heart rate of 120 beats per min during the 75-W workload. For each stage during the resting and exercise protocol, measurements were taken each minute for 5 min and the final three readings taken to allow the heart rate and energy expenditure to reach a steady-state. The O2 concentration in the expired air and min volume data were used to calculate O2 consumption after correction for standard temperature and pressure. Energy expenditure (kJ·min-1) was calculated at each time point as O2 consumption (mL·min) × 20.35 (6). Body fat percentage was obtained using a standard impedance technique (Bodystat, Isle of Man). Height and weight were also measured in light clothing without shoes. Body mass index (BMI) was calculated as weight/height2. Ethical permission for the study was granted by the Cambridge Local Research Ethics Committee.
The information from the calibration test was used to calculate mean resting energy expenditure and heart rate flex, the average of the lowest heart rate on exercise and the highest at rest. The slope and intercept of the regression line between energy expenditure and heart rate above HRFlex were calculated. An example of a calibration graph is illustrated in Figure 1. Each subject then wore a heart rate monitor (Polar Sports Tester, Kempele, Finland) over the following 4 d after the visit, and minute-by-minute heart rate was recorded. The monitors were only worn during waking hours. Because the monitor has a memory capacity of just over 34 h, subjects were given two watches or arrangements were made for the data to be downloaded after the second day. Trained researchers explained how to start the heart rate monitors at the beginning of each day and were available for any queries about the functions of the monitors during the period of monitoring. The 4 d of monitoring were started on random days of the week.
Heart rate monitoring processing.
At the end of the four day period, heart rate data were downloaded into a computer through a serial interface. Data were processed using locally developed Windows software. The individual calibration data were used to calculate minute-by-minute energy expenditure for each subject. If heart rate was less or equal to HRFlex, then energy expenditure was assumed to be equal to mean resting energy expenditure. If heart rate was greater than HRFlex, then energy expenditure was calculated from the linear prediction (26). Sleeping energy expenditure was taken as 95% of basal metabolic rate (8), calculated from published prediction equations (10). Small periods of time at the beginning and end of each day were usually missed when the monitors were not worn, for example, during showering. For the purposes of calculating total energy expenditure, we assumed that this time was spent at rest. Physical activity level (PAL) was computed for each day from the minute-by-minute energy expenditure data and averaged over the four day period, where PAL is the ratio of total energy expenditure (TEE) to basal metabolic rate.
Estimating PAL without individual calibration using regression models.
The data from the 789 volunteers from the Ely cohort was used to develop regression models to estimate PAL. The variables considered as potential predictors of the calibration parameters were sex, age, weight, body mass index (BMI), percentage body fat, and sitting pulse rate. The sitting heart rate had been recorded for 3 min during the individual calibration test in the analysis the first measure was used. The relationship between the calibration parameters and these easily measured variables was described using Pearson correlation. The confidence intervals around the correlation coefficient were calculated using Fisher’s Z transformation of r (20). Variables with significant univariate associations were entered into a stepwise multiple regression model, in which variables were rejected if they were not significant at 0.05. Weight and BMI were evaluated in separate models to avoid collinearity. The optimal models were those with the greatest adjusted R2. The final regression models for each of the parameters and the prediction of PAL were tested in the independent sample of individuals who had undertaken heart rate monitoring and the calibration test. As neither of the two methods were a gold standard method to measure energy expenditure, a Bland and Altman plot was used to assess the agreement between the estimated PAL using the estimated calibration parameters and the observed PAL using the measured calibration parameters (2). A plot of the difference between the two PAL scores against the mean of the two PAL scores was made. All analyses were carried out using SPSS 9.0 software (Statistical Package for the Social Sciences, SPSS Inc, Chicago, IL).
The baseline characteristics of the Ely cohort are shown in Table 1. All subjects completed 4 full days of heart-rate monitoring. The median estimate of PAL in this cohort was 1.93 (interquartile range 1.67, 2.21) in men and 1.72 (interquartile range 1.51, 1.93) in women. Medians are presented as the underlying data was skewed. The baseline characteristics of the independent EPIC sample are shown in Table 2. The median estimate of PAL in this sample was 1.60 (interquartile range 1.47, 2.06) in men and 1.95 (interquartile range 1.52, 2.40) in women.
The correlations between the possible prediction variables and the calibration parameters are illustrated in Table 3. Weight, BMI, body fat percentage, and sitting heart rate were strongly associated with resting energy expenditure in both sexes. BMI, body fat percentage, and sitting heart rate were associated with the HRFlex point in both men and women. In addition to these variables, weight was also significantly correlated to the HRFlex point in women. A strong negative relationship was found between the slope and intercept. Variables that displayed a significant positive association with the slope parameter were inversely associated with the intercept parameter.
We then proceeded to develop four separate multivariate models to predict the key parameters; mean resting energy expenditure, HRFlex, the intercept, and slope of the line (Table 4). Two models were considered to predict mean resting energy expenditure. One model included the characteristics sex, weight, sitting heart rate, and percentage body fat (adjusted R2 = 0.46), and the second model included sex, weight, and sitting heart rate (adjusted R2 = 0.45). The second model was chosen for its simplicity, because the amount of additional time and cost to obtain the percentage body fat measurement would only lead to a 0.382% increase in the variance explained. This model was then applied to the data from the 97 subjects in the independent sample. A significant correlation between the observed mean resting energy expenditure and the estimated mean resting energy expenditure from the model was found with a correlation coefficient of 0.73 (P < 0.001). From the regression analyses of HRFlex with the baseline characteristics a model containing the variables sex, BMI and sitting heart rate was chosen (adjusted R2 = 0.88). The model was then applied to the independent sample. A significant correlation was found between the observed HRFlex and the estimated HRFlex with a correlation coefficient of 0.93 (P < 0.001).
The two parameters of the line, the slope, and intercept used in the calibration test could not be independently analyzed as they were strongly correlated with each other. Therefore, two methods of analysis were employed and the model with the higher R2 value chosen. The first method involved the development of a regression model using the baseline characteristics to predict the slope parameter. This was compared to the second method, where a regression model was calculated to predict the intercept parameter using the baseline characteristics. The higher of the two-parameter models in terms of R2 value was used in a regression model to estimate the other parameter.
From the regression analyses of the slope with the baseline characteristics, a model including sex, age, and weight was chosen (adjusted R2 = 0.34). Regression analyses of the intercept with the baseline characteristics produced a model containing the variables sex, age, and weight and sitting heart rate (adjusted R2 = 0.32). Because the predicted slope produced a higher R2 than the predicted intercept, this model was chosen. Regression analysis was used to produce a regression model of the intercept on the slope computed separately for each sex. This model explained 75.6% of the variance in men and 86.7% of the variance in women (both P < 0.001). This model was then applied to the data from the independent sample. A significant correlation of 0.76 (P < 0.01) was found between the observed and predicted slopes and the correlation between the estimated intercept from this model and the observed intercept in the independent group was also significant (r = 0.59, P < 0.01). The final models chosen are shown in Table 4.
The PAL calculated using the four parameter estimates significantly correlated with the PAL score using the measured parameters in the independent sample with a correlation coefficient of 0.82 (95% confidence interval 0.74, 0.87, P < 0.01; men r = 0.83, P < 0.01; women r = 0.84, P < 0.01;Fig. 2). Figure 3 shows the Bland and Altman plot of the difference between the PAL scores derived from the observed and estimated parameters against the mean of these two PAL scores. This clearly demonstrates that although the estimated method provides an indication of the ranking of an individual, the absolute values are underestimated. When both the PAL scores, using the measured and estimated parameters, were split into quartiles, 58.8% of the scores were placed in the same quartile and 97.9% of the scores were placed in the same or adjacent quartile (Table 5).
The aim of this study was to assess the feasibility of using heart rate monitoring, without individual calibration, to estimate energy expenditure with the use of prediction parameters. Simplification of this technique would permit its use in medium-scale studies, for example, those involving more than 500 subjects. Although there are limitations to heart rate monitoring, the feasibility of using this method with individual calibration in studies of around 800 individuals has been reported (25,26,28). The estimate of PAL using the four estimated parameters significantly correlated with the measured PAL in the independent sample (r = 0.82, P < 0.01). With 98% of the sample placed in the same or adjacent quartile, this nonexercise regression approach may allow heart rate monitoring to be applied to larger epidemiological studies. However, it is still likely that questionnaires would be necessary in very large-scale studies.
Our study differs from previously published studies where group calibration methods to estimate energy expenditure were used. Luke et al. (14) reported a single group equation to predict oxygen consumption above a resting heart rate level based on regression analysis. However, this equation was derived from movement sensoring and heart rate data collected from a number of activities measured in a laboratory setting and not from easily measured variables. Rutgers et al. (17) derived a calibration curve from group mean values of heart rate and energy expenditure, which could then be applied to larger populations. However, this calibration curve was based on only thirteen elderly subjects. They concluded that the use of a group calibration curve was imprecise as mean TEE for three days measured with individual calibration was not significantly correlated to that computed using a group calibration (r = 0.37). In addition, there were large individual discrepancies between the TEE estimates using the two methods. Li et al. (12) also found poor levels of agreement when comparing energy expenditure estimates using heart rate/energy expenditure calibrations based on group and individual data. Again, this study was only based on a small sample (N = 40), and the calibration curve was derived from a number of activities where oxygen consumption was measured, thus increasing the burden of measurement. The level of agreement between the regression models and the observed parameters of the calibration graph in the current study, using only variables that can be easily measured in fieldwork, may allow this method to be applied to larger studies.
The variables selected to estimate mean resting energy expenditure, HRFlex and the slope and intercept of the line are easy to measure and have face validity. The concept of the HRFlex was introduced by Spurr et al. (21) in order to improve the estimate of the O2 consumption-heart rate relationship at low heart rates. No studies have been done to date to explore the causes of variation between individuals in HRFlex. However, Ceesay et al. (5) point out that finding one definitive HRFlex for an individual is a problem. In the current study, the variables sex, BMI, and sitting heart rate proved to be good predictors of HRFlex. Traditional estimates of resting metabolic rate (RMR) have been based on height and weight (10). This study also showed a strong relationship between weight and mean resting energy expenditure with 11.5% of the variance in mean resting energy expenditure being accounted for by weight alone. Many studies have reported a variation in RMR caused by exercise training (1,3,9,19,22,23,31). Individuals who participate regularly in exercise tend to have lower resting heart rates as a result of the aerobic training. In this study the sitting heart rate, which had a significant positive relationship with mean resting energy expenditure, may reflect this relationship and thus be a surrogate measure for fitness. The two parameters, the slope and intercept of the line, are highly correlated. The model for the slope parameter was a better predictor of the observed parameter than the model for the intercept. One reason for this maybe that the range of observed intercept values in the sample is far larger than the range of observed slope values, making estimation more difficult. Research is required to find further variables that are easy to measure, and which can explain more of the variance in the calibration parameters.
A similar prediction approach using multiple regression models has previously been used to predict cardiorespiratory fitness without subjects having to undergo an exercise test. Variables such as age, gender, body weight, body composition, BMI, and activity level have been included in the models (15,29). Matthews et al. (15) classified the cardiorespiratory fitness status of 799 individuals using prediction equations to estimate fitness. When measured fitness levels were compared with the equation-derived estimates split into quintiles, it was reported that 36% were classified correctly into the same quintile and 83% were classified into the same or adjacent quintile. However, this testing of the prediction was carried out in the same population in which the equations had been derived. In this study, classification of PAL using the individual calibration was compared to the PAL using the prediction-equation derived parameters in a smaller, independent group that were representative of the larger population from which they were selected; 59% of the subjects were classified into the same quartile for both the measured and predicted PAL estimates, and 98% were classified into the same or adjacent quartile. Misclassification errors of two quartiles were only observed in two subjects. Though this method has its limitations, and further research is needed, we believe this is a good starting point for developing this method.
As in all V̇O2 testing, a number of assumptions about the heart rate response during the calibration test have to be made. It is assumed that maximal heart rate, resting heart rate, and V̇O2 at a given workload are constant for a given subject. Observer bias in the assessment of steady-state heart rate was limited as far as possible in this study. Two trained test administrators were used throughout the 2-yr testing period to limit error in the individual calibration measurements, especially during those taken at rest, where a decision on whether the subject has reached a steady-state heart rate has to be taken. The heart-rate monitor was worn for the 4 d after the clinic appointment, and therefore no allowance for differentiation between energy expenditure on weekdays and weekend days was made. However, it has previously been reported in a similar population that there is no evidence of systematic bias between weekday and weekend energy expenditure at a population level (26).
The regression models, predicting mean resting energy expenditure, HRFlex and the slope and intercept of the line, were developed and tested in a population of 40- to 70-yr-old Caucasian men and women of generally good health. Thus, these models are only generalizable to similar populations. It is certainly unclear how these models would perform in other ethnic groups, and we anticipate studies in such populations. In order to apply this method to other populations, a subsample of the chosen population should undergo heart rate monitoring with an individual calibration test in order to formulate regression equations for predicting energy expenditure in the rest of the population.
In conclusion, the results of this study support the concept that heart rate monitoring could be used in large population-based studies (N > 500) to estimate energy expenditure without the need for individual calibration. With data that can easily be obtained from a questionnaire or within a short appointment, an objective measure of group-level energy expenditure in epidemiological studies may be made. Categorization into quartiles in a small group of individuals displayed small extreme misclassification errors. However, further research is required to refine these techniques and increase the accuracy of the categorisation. Studies to compare these estimates with a gold standard measure of energy expenditure such as doubly labeled water are also necessary. Such studies would yield values for the accuracy of this method in comparison to the true energy expenditure of individuals in field studies. In addition, future research should examine how this method could be adapted and applied to other populations particularly in developing countries, where no estimates of population energy expenditure levels are available.
The authors are grateful to Ailsa Welch and Angela Mulligan for their collaboration in the fieldwork in the EPIC cohort. The Ely Project was funded by the Medical Research Council (MRC), National Health Service Research and Development and the British Diabetes Association. Dr. N. J. Wareham is an MRC Clinician Scientist Fellow. K. L. Rennie is in receipt of a MRC Ph.D. scholarship.
Address for correspondence: N. J. Wareham, Department of Public Health and Primary Care, Institute of Public Health, University of Cambridge, Robinson Way, Cambridge, CB2 2SR, UK; E-mail: [email protected]
1. Ballor, D. L., and E. T. Poehlman. A meta-analysis of the effects of exercise and/or dietary restriction on RMR. Eur. J. Appl. Physiol. 71: 535–542, 1995.
2. Bland, J. M., and D. G. Altman. Statistical methods for assessing agreement between two methods of clinical measurement: how do we know if 2 measures agree with each other? Lancet 1: 307–310, 1986.
3. Bullough, R. C., C. A. Gillette, M. A. Harris, and C. L. Melby. Interaction of acute changes in exercise energy expenditure and energy intake on resting metabolic rate. Am. J. Clin. Nutr. 61: 473–481, 1995.
4. Caspersen, C. J. Physical activity epidemiology
: concepts, methods, and applications to exercise science. Exerc. Sports Sci. Rev. 17: 423–473, 1989.
5. Ceesay, S. M., A. M. Prentice, K. C. Day, et al. The use of heart rate monitoring in the estimation of energy expenditure: a validation study using indirect whole-body calorimetry. Br. J. Nutr. 61: 175–186, 1989.
6. Elia, M., and G. Livesey. Energy expenditure and fuel selection in biological systems: theory and practice of calculations based on indirect calorimetry and tracer methods. In:Metabolic Control of Eating, Energy Expenditure and the Bioenergetics of Obesity
, A. P. Simopoulos (Ed.). World Rev. Nutr. Diet. Basel: Karger, 1992, 70:68–131.
7. Day, N., S. Oakes, R. Luben, et al. EPIC in Norfolk: study design and characteristics of the cohort. Br. J. Cancer 80: 95–103, 1999.
8. Goldberg, G. R., A. M. Prentice, H. L. Davies, and P. R. Murgatroyd. Overnight and basal metabolic rates in men and women. Eur. J. Clin. Nutr. 42: 137–144, 1988.
9. Goran, M. I., J. Calles-Escandon, E. T. Poehlman, M. O’Connell,and E. Danforth. Effects of increased energy intake and/or physical activity on energy expenditure in young healthy men. J. Appl. Physiol. 77: 366–372, 1994.
10. James, W. P. T., and E. C. Schofield. Human energy requirements. Oxford: Oxford Medical Publications, 1990, pp. 25.
11. Leonard, W. R., P. T. Katzmarzyk, M. A. Stephen, and A. G. Ross. Comparison of the heart rate-monitoring and factorial methods: assessment of energy expenditure in highland and coastal Ecuadoreans. Am. J. Clin. Nutr. 61: 1146–1152, 1995.
12. Li, R., P. Deurenberg, and J. G. Hautvast. A critical evaluation of heart rate monitoring to assess energy expenditure in individuals. Am. J. Clin. Nutr. 58: 602–607, 1993.
13. Livingstone, M. B. E., A. M. Prentice, W. A. Coward, et al. Simultaneous measurement of free-living energy expenditure by the doubly labeled water method and heart-rate monitoring. Am. J. Clin. Nutr. 52: 59–65, 1990.
14. Luke, A., K. Maki, N. Barkey, R. Cooper, and D. McGee. Simultaneous monitoring of heart rate and motion to assess energy expenditure. Med. Sci. Sports Exerc. 29: 144–148, 1997.
15. Matthews, C. E., D. P. Heil, P. S. Freedson, and H. Pastides. Classification of cardiorespiratory fitness without exercise testing. Med. Sci. Sports Exerc. 31: 486–493, 1999.
16. Rennie, K., T. Rowsell, S. A. Jebb, D. Holburn, and N. J. Wareham. A combined heart rate and movement sensor: proof of concept and preliminary testing study. Eur. J. Clin. Nutr. 54: 409–414, 2000.
17. Rutgers, C. J., M. J. C. Klijn, and P. Deurenberg. The assessment of 24-hour energy expenditure in elderly women by minute-by-minute heart rate monitoring. Ann. Nutr. Metab. 41: 83–88, 1997.
18. Schuit, A. J., E. G. Schouten, K. R. Westerup, and W. H. M. Saris. Validity of the Physical Scale for the Elderly (PASE): according to energy expenditure assessed by the doubly labeled water method. J. Clin. Epidemiol. 50: 541–546, 1997.
19. Sjodin, A. M., A. H. Forslund, K. R. Westerterp, A. B. Andersson, J. M. Forslund, and L. M. Hambraeus. The influence of physical activity on BMR. Med. Sci. Sports Exerc. 28: 85–91, 1996.
20. Snedecor, G. W., and W. G. Cochran. Statistical Methods. Ames, IA: Iowa State University Press, 1980, pp. 498.
21. Spurr, G. B., A. M. Prentice, P. R. Murgatroyd, G. R. Goldberg, J. C. Reina, and N. T. Christman. Energy expenditure from minute-by-minute heart-rate recording: comparison with indirect calorimetry. Am. J. Clin. Nutr. 48: 552–559, 1988.
22. Thompson, J. L., M. M. Manore, and J. R. Thomas. Effects of diet and diet-plus-exercise programs on resting metabolic rate: a meta-analysis. Int. J. Sport Nutr. 6: 41–61, 1996.
23. Visser, M., P. Deurenberg, W. A. van Staveren,and J. G. Hautvast. Resting metabolic rate and diet-induced thermogenesis in young and elderly subjects: relationship with body composition, fat distribution, and physical activity level. Am. J. Clin. Nutr. 61: 772–8, 1995.
24. Wareham, N. J., C. D. Bryne, R. Williams, N. E. Day, and C. N. Hales. Fasting proinsulin concentrations predict the development of Type 2 diabetes. Diabetes Care 22: 262–270, 1999.
25. Wareham, N. J., S. J. Hennings, C. D. Byrne, C. N. Hales, A. M. Prentice, and N. E. Day. A quantitative analysis of the relationship between habitual energy expenditure, fitness and the metabolic cardiovascular syndrome. Br. J. Nutr. 80: 235–241, 1998.
26. Wareham, N. J., S. J. Hennings, A. M. Prentice, and N. E. Day. Feasibility of heart rate monitoring to estimate total level and pattern of energy expenditure in a population-based epidemiological study. Br. J. Nutr. 78: 889–900, 1997.
27. Wareham, N. J., and K. L. Rennie. The assessment of physical activity in individuals and populations: why try to be more precise about how physical activity is assessed? Int. J. Obes. 22 (S2)S30–38, 1998.
28. Wareham, N. J., M.-Y. Wong, and N. E. Day. Glucose intolerance and physical inactivity: the relative importance of low habitual energy expenditure and cardiorespiratory fitness. Am. J. Epidemiol. 152: 132–139, 2000
29. Whaley, M. H., L. A. Kaminsky, G. B. Dwyer, and L. H. Getchell. Failure of predicted VO2peak to discriminate physical fitness in epidemiological studies. Med. Sci. Sports Exerc. 27 (1): 85–91, 1995.
30. Williams, D. R. R., N. J. Wareham, D. C. Brown, et al. Glucose intolerance in the community; the Isle of Ely Diabetes Project. Diabetic Med. 12: 30–35, 1995.
31. Wilmore, J. H., P. R. Stanforth, J. Gagnon, et al. Endurance exercise training has a minimal effect on resting heart rate: the HERITAGE study. Med. Sci. Sports Exerc. 28: 829–835, 1996.