Armstrong 2011 Methodology Seminar
Transcript of Armstrong 2011 Methodology Seminar
-
8/3/2019 Armstrong 2011 Methodology Seminar
1/35
Reported frequency of physical activity ina large epidemiological study:
relationship to specific activities and
repeatability over timeMEG Armstrong, BJ Cairns, J Green, G K Reeves, and V Beral
for The Million Women Study Collaborators
Cancer Epidemiology UnitNuffield Department of Clinical Medicine
University of Oxford
www.millionwomenstudy.org
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
2/35
Overview
THE MILLION WOMEN STUDY
Background
Physical activity assessments in the Million Women Study
Frequency of physical activity at baseline and reportedspecific activities at follow-up
Repeatability over time of reported frequency of physical
activity at baseline
Summary
Methodological considerations
-
8/3/2019 Armstrong 2011 Methodology Seminar
3/35
Background Objective measures of physical activity (PA) are: Expensive
Impractical
Insensitive to seasons
Questionnaires usually used as:
Less expensive
Less likely to interfere with usual PA
Assessment of many variables with one instrument Simple to administer and score
However, questionnaires are:
Subjective
Prone to measurement error (memory, a portion of activity)
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
4/35
Background
THE MILLION WOMEN STUDY
In prospective studies PA assessed at baseline often used toassess influence of PA on health outcomes during follow-up
Yet, baseline assessments might not represent actual PAduring extended follow-up periods:
Measurement errors during assessment
Change in PA behaviours over time Could lead to underestimation of associations between
baseline PA and health outcomes
-
8/3/2019 Armstrong 2011 Methodology Seminar
5/35
THE MILLION WOMEN STUDY
Physical activity
assessments in the Million
Women Study
-
8/3/2019 Armstrong 2011 Methodology Seminar
6/35
Prospective cohort study of 1.3million UK women
Recruited through 66 breastscreening clinics in 1996-2001
Mean age of 56 years (range 50-64)
1 in 4 of UK women in age range
Self-administered questionnaire atbaseline and at follow-up
The cohort
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
7/35
Data collection At baseline women were asked:
How often do you do any strenuous exercise?
How often do you do any exercise? (91%) 6 frequency options
No distinction between summer/winter activity
18 655 repeat baseline questionnaires were completed 17 617 strenuous activity data repeats
12 748 any activity data repeats
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
8/35
Data collection 65% response rate between baseline and follow-up
~ 3 years after baseline over 600 000 women answered: About how many hours each week do you spend doing:
housework, gardening, walking, cycling, any work orexercise causing sweating or a fast heartbeat?
Reported separately on summer/winter except for housework
589 896 responded to the activity questionnaires on bothbaseline and follow-up
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
9/35
THE MILLION WOMEN STUDY
Frequency of physical activity at
baseline and reported specific
activities at follow-up
-
8/3/2019 Armstrong 2011 Methodology Seminar
10/35
Methods Mean estimated excess energy expenditure (EE) for
strenuous & specific activities reported at follow-up comparedacross baseline PA categories
Metabolic equivalents (METs) assigned to each activityaccording to Ainsworths Compendium1 of activities.
A MET = ratio of the metabolic rate required by a given worktask, to the standard resting metabolic rate obtained whilesitting quietly
Spearman correlation coefficients & P-values for trend
1Ainsworth BE, et al: Med Sci Sport Exer 2000, 32(9):S498-S516.
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
11/35
METs Housework (3 METs), gardening (4 METs), walking (3.5
METs), cycling (8 METs), strenuous (8 METs). Limitations?
Multiplying by these gives gross metabolic cost ie.
Gross metabolic cost = resting METs cost + PA METs cost
Estimated net energy expenditure more appropriate for non-24 hour EE:
Middle aged to older - greater proportion of total PA composedof low intensity activities
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
12/35
Why estimated net energy expenditure? One example:
Person A:does no vigorous activity in a week (0)
Person B:does one hour of vigorous activity in a week (1)
Using Ainsworths multiplier of 8:
Person A:estimated at 0x8 MET hours (0)
Person B:estimated at 1x8 MET hours (8)
But, person A expended resting MET value of 1 duringsame time period a discrepancy. Use 7 instead of 8, to
account for resting EE
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
13/35
Why estimated net energy expenditure? Another example:
Person A:does one hour of housework in a week
Person B:does one hour of cycling in a week
Ainsworths multiplier is 3 for housework and 8 for cycling:
Person A:
proportion of EE attributed to resting metabolic rate = 33%
Person B:
proportion of EE attributed to resting metabolic rate = 12.5%
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
14/35
Excess MET-hours for activity at follow-up in relation to baseline frequency ofstrenuous activity
THE MILLION WOMEN STUDY
Hours and excess MET-hours per week spent doing various activitiesreported ~3 years after baseline
Strenuous activity Aggregate of various activities
Freq of strenuous PAreported at baseline
Number ofwomen
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
Never 261,857 0.8 5.6 (21.3) 21.7 52.4 (39.0)
< Once per week 79,962 1.1 7.8 (20.2) 22.7 57.3 (38.0)
Once per week 115,573 1.5 10.7 (20.5) 23.4 61.5 (38.5)
2 to 3 times per week 96,415 2.4 16.5 (25.0) 24.8 69.5 (42.5)
4 to 6 times per week 19,394 3.4 23.9 (34.1) 27.2 81.3 (50.6)
Daily 16,431 3.5 24.3 (50.8) 31.1 90.6 (69.1)
Correlation coefficient 0.37 0.12 0.22
P for trend< 0.0001 < 0.0001 < 0.0001
-
8/3/2019 Armstrong 2011 Methodology Seminar
15/35
THE MILLION WOMEN STUDY
Hours and excess MET-hours per week spent doing various activitiesreported ~3 years after baseline
Strenuous activity Aggregate of various activities
Freq of any PAreported at baseline
Number ofwomen
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
Never 107,346 0.8 5.8 (24.5) 21.0 50.3 (41.4)
< Once per week 49,073 0.9 6.6 (21.2) 20.6 50.9 (37.8)
Once per week 92,317 1.2 8.1 (19.5) 21.3 53.9 (37.0)
2 to 3 times per week 149,138 1.6 11.4 (20.6) 22.7 60.0 (37.5)
4 to 6 times per week 57,585 2.0 13.8 (24.0) 24.2 66.0 (39.6)
Daily 134,437 1.7 12.0 (30.0) 27.0 71.2 (49.0)
Correlation coefficient 0.17 0.17 0.22
P for trend < 0.0001 < 0.0001 < 0.0001
Excess MET-hours for activity at follow-up in relation to baseline frequency ofany activity
-
8/3/2019 Armstrong 2011 Methodology Seminar
16/35
THE MILLION WOMEN STUDY
Hours and excess MET-hours per week spent doing various activities reported ~3years after baseline
walking cycling gardening housework
Baseline reported freqof strenuous PA
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
n = 589,632 n = 589,632 n = 589,632 n = 589,632
Never 3.9 9.7 (12.8) 0.1 0.9 (5.7) 2.4 7.3 (9.9) 14.4 28.8 (23.0)
< Once per week 4.4 11.1 (12.1) 0.2 1.5 (6.8) 3.1 9.4 (10.8) 13.8 27.5 (21.2)
Once per week 4.8 12.0 (12.4) 0.3 1.9 (7.7) 3.3 9.8 (11.0) 13.5 27.0 (21.1)
2 to 3 times per week 5.2 13.1 (13.2) 0.4 2.8 (9.8) 3.5 10.5 (12.1) 13.3 26.6 (21.0)
4 to 6 times per week 5.9 14.8 (14.9) 0.6 4.4 (13.1) 3.8 11.4 (13.9) 13.4 26.7 (21.6)
Daily 7.5 18.6 (20.4) 0.6 4.4 (16.5) 4.1 12.3 (15.7) 15.4 30.9 (25.7)
Correlation coefficient 0.18 0.15 0.15 -0.03
P for trend < 0.0001 < 0.0001 < 0.0001 < 0.0001
Various reported physical activities at follow-up in relation to frequency of
strenuous activity at baseline
-
8/3/2019 Armstrong 2011 Methodology Seminar
17/35
THE MILLION WOMEN STUDY
Hours and excess MET-hours per week spent doing various activities reported ~3years after baseline
walking cycling gardening housework
Baseline reported freqof any PA
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
hoursmean
MET-hoursmean (SD)
n = 589,896 n = 589,896 n = 589,896 n = 589,896
Never 3.2 7.9 (12.7) 0.1 0.6 (4.9) 2.2 6.7 (9.8) 14.7 29.3 (23.7)
< Once per week 3.3 8.2 (11.2) 0.1 0.9 (4.7) 2.6 7.9 (9.8) 13.7 27.3 (21.8)
Once per week 3.7 9.2 (11.3) 0.2 1.3 (6.1) 2.8 8.4 (10.0) 13.5 27.0 (21.6)
2 to 3 times per week 4.4 11.0 (11.5) 0.3 1.8 (7.6) 3.0 9.1 (10.6) 13.4 26.8 (20.8)
4 to 6 times per week 5.2 13.0 (12.1) 0.4 2.7 (9.1) 3.2 9.6 (11.2) 13.5 27.0 (20.5)
Daily 6.5 16.3 (15.9) 0.4 2.6 (10.9) 3.6 10.8 (13.0) 14.8 29.6 (23.3)
Correlation coefficient 0.31 0.11 0.14 0.01
P for trend < 0.0001 < 0.0001 < 0.0001 0.002
Various reported physical activities at follow-up in relation to frequency of any
activity at baseline
-
8/3/2019 Armstrong 2011 Methodology Seminar
18/35
Comments Older populations more sedentary might see
decreased between person variability
Measurement error may as proportion of PAcomprised of light intensity activities increases
Social desirability bias may influence accuracy of datafrom PA questionnaires
All may reduce power of a PA instrument to discriminate
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
19/35
Domestic activities = PA Do people consider housework when asked about PA?
Domestic activities = large proportion of PA in middle agedand older. Eg. British women 60-791:
Without domestic activities only 21% met PArecommendations
With domestic activities >2/3rds met recommendations
Physical impairments may report never active but stillactually do domestic activities?
THE MILLION WOMEN STUDY
1Lawlor DA, et al., J Epidemiol Community Health 2002, 56:473-478.
-
8/3/2019 Armstrong 2011 Methodology Seminar
20/35
THE MILLION WOMEN STUDY
Repeatability over time of
reported frequency of
physical activity at baseline
-
8/3/2019 Armstrong 2011 Methodology Seminar
21/35
Methods Distributions across PA categories by time between first and
repeat response
Distributions of change in PA by time between first andrepeat response
PA ordinal kappa coefficient with quadratic weighting usedto assess agreement (equivalent to intraclass correlation1)
Differences according to seasonality
1Fleiss JL, Cohen J. Educ psychol meas 1973,33:613-619.
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
22/35
THE MILLION WOMEN STUDY
Distribution of reported frequency of PA by time between first and repeat
baseline questionnaires
-
8/3/2019 Armstrong 2011 Methodology Seminar
23/35
THE MILLION WOMEN STUDY
Distributions of changes in reported PA frequency by time between first and
repeat baseline questionnaires
-
8/3/2019 Armstrong 2011 Methodology Seminar
24/35
Population level vs Individual PA
Distribution similar at population level BUTAgreementbetween individuals decreases eg.
Australia 1 (45-50 yrs): 14% PA, 12% PA, over 3 yrs
Netherlands 2 (20-59 yrs): 45% changed PA level over 10-yearperiod, equal distribution between decreasers and increasers
UK 3 (33 yrs): ~1/3 and ~1/3 their PA level over an 8-year
USA 4 : 16% PA, 12% PA, over 4 yrs
THE MILLION WOMEN STUDY
1 Guthrie J. Acta Obstet Gynecol Scand 2002, 81:595-602.2 Picavet HSJ et al., Med Sci Sport Exer 2010, 43(1):74-79.
3 Parsons TJ, et al., Eur J Clin Nutr 2005, 59:49-56.4 Eaton CB et al., Am J Prev Med 1993, 9(4):209-219.
-
8/3/2019 Armstrong 2011 Methodology Seminar
25/35
THE MILLION WOMEN STUDY
Number
of women
Absolute
Agreement
Kappa (95% CI)
with quadraticweighting
Time between completing identical questions onfreq of strenuous PA (avg time between the two)
6 months (0.3 years) 238 64% 0.71 (0.59 - 0.83)
> 6 months - 1 year (0.9 years) 1,224 59% 0.61 (0.55 - 0.67)
> 1 - 2 years (1.5 years) 7,002 54% 0.55 (0.53 - 0.57)
> 2 years (2.6 years) 9,153 52% 0.51 (0.49 - 0.53)
P for trend 0.03
Time between completing identical questions on
the freq of any PA (avg time between the two) 6 months (0.3 years) 221 57% 0.67 (0.53 - 0.81)
> 6 months - 1 year (0.9 years) 1,171 53% 0.67 (0.61 - 0.73)
> 1 - 2 years (1.5 years) 6,044 48% 0.60 (0.58 - 0.62)
> 2 years (2.6 years) 5,312 47% 0.58 (0.56 - 0.60)
P for trend 0.05
Agreement of reported PA, by time between completing identical questionnaires
-
8/3/2019 Armstrong 2011 Methodology Seminar
26/35
THE MILLION WOMEN STUDY
Numberofwomen
Kappa (95% CI)with quadraticweighting
Agreement according to season of first responseversus repeat response, strenuous PA
Winter first response vs winter repeat response 1,110 0.52
Winter first response vs summer repeat response 665 0.54
Summer first response vs summer repeat response 1,118 0.57
Summer first response vs winter repeat response 1,052 0.51
Agreement according to season of first response
versus repeat response, any PA
Winter first response vs winter repeat response 708 0.57
Winter first response vs summer repeat response 509 0.58
Summer first response vs summer repeat response 830 0.63
Summer first response vs winter repeat response 908 0.58
Agreement of reported physical activity, by season of reporting
-
8/3/2019 Armstrong 2011 Methodology Seminar
27/35
THE MILLION WOMEN STUDY
Summary
-
8/3/2019 Armstrong 2011 Methodology Seminar
28/35
Simple questions on frequency of PA at baselineassociated with hours spend on specific activities &estimated excess MET hours ~3 years later
Weakest associations with housework
Agreement for identical questions on PA frequencyasked on two occasions decreased over time
Agreement similar for different seasons
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
29/35
THE MILLION WOMEN STUDY
Methodologicalconsiderations
-
8/3/2019 Armstrong 2011 Methodology Seminar
30/35
THE MILLION WOMEN STUDY
Methodological considerations in PA
research
Combination of factors that may influence relationship
between PA and disease outcome
Past activity (over the life course)?
Present activity at baseline?
Changes in PA over follow-up time? Measurement error at baseline?
-
8/3/2019 Armstrong 2011 Methodology Seminar
31/35
Measurement error & PA assessment
Variability in answers when completing baseline questionnaire
Difficulty recalling past activity time periods, type Different interpretations housework, intensity
Social desirability bias
Random reporting errors
Real changes in PA over time
Lifestyle changes, retirement etc
Could be associated with morbidity reverse causality
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
32/35
Influences of measurement error
Measurement errors likely to result in attenuation of estimatesof association between baseline PA and disease risk
Degree of attenuation depends on:
Assessment instrument
Participant characteristics
Intensive assessment may sources of error but oftenimpractical, expensive & alter natural behaviour
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
33/35
Possible Approaches
Correct for regression dilution using objective measurements
Assess associations between PA at baseline and diseaseoutcome at follow-up split according to discrete time bands offollow-up
Report findings so readers may make their own conclusions
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
34/35
Acknowledgements Million Women Study steering committee, co-ordinating centre staff, and
participants
Supported by Cancer Research UK and the Medical Research Council
THE MILLION WOMEN STUDY
-
8/3/2019 Armstrong 2011 Methodology Seminar
35/35
QUESTIONS?
THE MILLION WOMEN STUDY