Hubble Asteroid Hunter III. Physical properties of newly found asteroids
Bases talk for slideshare (atkinson)
1. Detecting and quantifying individual
response heterogeneity in sport and
exercise sciences
Greg Atkinson (Teesside University)
greg.atkinson@tees.ac.uk
2. The mean response of a sample “fails to recognize that
there are considerable inter-individual differences in
responses to any exercise program”
(Bouchard et al., 2014; p.2).
3. Let’s state this formally
There are true* and clinically-
relevant differences between
individuals in how much each of them
responds to exactly the same
exercise intervention
*True = Not explained by other sources of variability
5. Some more individual responses
- plotted the more common way
Church TS, et al. PLoS ONE 2009; 4: e4515.
6 months exercise @
8 kcal/kg/wk
6. And another of my own
Atkinson and Batterham, Exp Physiol 2015; 100:577-88
Observed DBP
baseline-follow-up
changes
7. But something tells me we
should be cautious
Church TS, et al. PLoS ONE 2009; 4: e4515.
8. There are actually no true individual
response differences here
Atkinson and Batterham, Exp Physiol 2015; 100:577-88
Observed DBP
baseline-follow-up
changes
9. Because we made it that way!
-5 mmHg for everyone! Individual differences
Atkinson and Batterham, Exp Physiol 2015; 100:577-88
10. Stated reasons for not
considering control data
• “We didn’t think we needed one anyway”
• “We just want to focus on the responses to the
intervention per se”
• “There was no mean change between baseline
and follow-up in the control group”
• “We had information anyway from a test-retest
reliability study”
12. Using a measurement error stat to
inform an MCID: Use the data!
• ±1TEM covers the central 52% of individual changes
• ≈ 24% of the people in the 3-day test-retest study itself would
be labelled “responders” (>1TEM)
• ≈ 24% “adverse responders” (< -1TEM) in the reliability st.
• ≈ 8% “adverse responders” (< -2TEM) in the reliability st.
• While TEM has “informed” the responder definition, the above
nature of TEM has been overlooked
TEM = Technical error of measurement ≈ standard error of measurement ≈
typical error ≈ CV
13. Prevalence of “adverse responders” (<-2TEM)
in various study intervention groups
Bouchard C, et al. PLoS ONE 7(5): e37887. doi:10.1371/journal.pone.0037887
Prevalence in the test-retest study itself ≈ 8% (± sampling error)
14. Using an error stat from an
incompatible test-retest period
T1 T2 Follow-up
1-7 days 6-24 weeks
Individual
baseline-follow-
up changes
Technical error
of measurement
(TEM or CV)
15. An error stat should not be used to
inform an MCID anyway!
BMJ 2018;363:k3750 http://dx.doi.org/10.1136/bmj.k3750
19. Little evidence for clinically-
relevant training response
heterogeneity
Williamson et al. Sports Med. 2017;47:1501-1513
20. Little evidence for weight loss
response heterogeneity
Pooled SD for true individual responses was ±0.63 kg (95%CI: -0.8 to 2.1)
21. Acute responses to exercise:
Replicated crossover is informative
Senn S. Stat Med 2015; DOI: 10.1002/sim.6739
22. Evidence for acute appetite
response heterogeneity
Goltz et al. Med Sci Sports Exerc. 2018; 50:758-768
23. Consistency of exercise responses
in replicated crossover
Goltz et al. Med Sci Sports Exerc. 2018; 50:758-768
24. Evidence for acute appetite
response heterogeneity
Goltz et al. Med Sci Sports Exerc. 2018; 50:758-768
25. Take home messages
• Looking at individual responses only in the treatment
group is not enough
• Comparing responder counts is not enough
• Basing “responder” definition on a short-term
measurement error stat (esp. TEM) is illogical
• Replicated crossover is informative
• Best that Response differences are known to be “true”
and clinically important BEFORE we do anything
else?
26. Hold on!
Why do we still seem to
identify associations
between some individual
factors and exercise
response?
27. Regression to the mean by conditioning on baseline
status (or anything associated with it)?
28. Focus on r2 in the context of
precision medicine?
Y = 0.59x + 9.96 (r2 = 17%,
P<0.001)
SD of residuals = ±3 units
Y = mean response = 10
SD of residuals = ±3 units
29. Reading response prediction/subgroup
studies
• Predictor x study arm interaction term?
• Covariate-adjustment for baseline status of
outcome?
• Individual prediction statistic?
– Standard error of prediction?
– Area under ROC curve?
– Absolute risk difference?
– Diagnostic test statistics (sensitivity, specificity, PPV)
30.
31. Thank you for listening
How much you listened
probably showed true
individual variation but I’ll
never know for sure unless I
repeat the talk
33. Crossover studies are better?
Atkinson et al. Diabetologia 2018; 61: 496–497
Acylatedghrelin(pg/ml)
34. Not necessarily: There are actually no true
response differences underneath here
Atkinson et al. Diabetologia 2018; 61: 496–497
Acylatedghrelin(pg/ml)
35. Focus on r2 in the context of
precision medicine?
"the heritability of VO2max among
sedentary adults could be as high as 50%
although this value is undoubtedly inflated
by non-genetic familial factors”
Bouchard C, et al. J Appl Physiol. 2000;88:551-9
36. Using the TEM to get responders and adverse
responders in the intervention group
±TEM
46. References
1. G Atkinson, AM Batterham. (2015). True and false interindividual differences in the physiological
response to an intervention. Experimental Physiology 100: 577-588
2. G Atkinson (2015). Individual differences in the exercise‐mediated blood pressure response: regression
to the mean in disguise? Clinical Physiology and Functional Imaging 35: 490-492.
3. PJ Williamson, G Atkinson, AM Batterham. (2017). Inter-Individual responses of maximal oxygen uptake
to exercise training: A critical review. Sports Medicine 47: 1501-1513
4. G Atkinson, AM Batterham (2017). The impact of random individual differences in weight change on the
measurable objectives of lifestyle weight management services. Sports Medicine 47: 1683-1688
5. FR Goltz, AE Thackray, JA King, JL Dorling, G Atkinson, DJ Stensel (2018). Interindividual responses of
appetite to acute exercise: A replicated crossover study. Medicine and Science in Sports and Exercise. In
Press.
6. G Atkinson, PJ Williamson, AM Batterham. (2018). Exercise training response heterogeneity: statistical
insights. Diabetologia, 61: 496-497
7. Taylor, C.E., et al., Blood pressure status and post-exercise hypotension: An example of a spurious
correlation in hypertension research. Journal of Human Hypertension, 2010. 24(9): p. 585-592.
47. Consistency of exercise responses
in replicated crossover
Goltz et al. Med Sci Sports Exerc. 2018; 50:758-768