Ernsberger and Køster-Rasmussen highlight the subgroup analysis for mortality for the importance of a physical activity facility as part of the intervention (RR 0.84, 95% CI 0.72 to 0.97). Firstly, statistical testing for this group against the groups that only got physical activity advice, or the groups that were not reported as receiving physical activity advice, was unable to demonstrate statistical heterogeneity (Chi² = 1.22, df = 2, P = 0.54, I² = 0%). Secondly, even if that was the case, subgroup testing in meta-analysis should be regarded with great caution (1), particularly as there are many other differences between the groups of trials in subgroup testing. Individual patient data meta-analyses would be required for further exploration of these subgroup findings for physical activity.
Ernsberger highlights the influence of the Look AHEAD trial (2), which carried 55% of the weighting in the meta-analysis of mortality. That is the reason that we undertook a post hoc analysis without this trial, which showed that weight loss interventions were still associated with decreased all-cause mortality (n=33 trials, 309 events; RR 0.78, 95% CI: 0.63 to 0.96; I2=0%). Sensitivity analysis to take account of rare events, using a random effects Bayesian logistic regression model, found a similar result.
As we acknowledge, we may have failed to identify all trials with outcome data, which may have biased results if trialists did not present these outcomes, particularly for those who dropped out, although we could not see obvious funnel plot asymmetry for all-cause mortality for our included studies. We also examined the drop out rate closely, it averaged 16% across all intervention groups and 16% across all control groups. We agree that trials described patient outcomes poorly, but think it unlikely that the many smaller trials with few events would have consistently failed to describe deaths in control groups when they described them in the intervention groups.
Saripanidis states that we conclude that ‘weight reducing diets do not reduce cardiovascular events/cardiovascular specific mortality/cancer specific mortality’. This is not our conclusion. Our conclusion is that we have been unable to demonstrate an effect on these outcomes. As is clear from our report and appendices, many studies did not provide details on the causes of death (see particularly Appendix Table 1), or did not collect or were unable to provide data on cardiovascular events. Thus our statistical power for these outcomes was limited.
Koh states that there is robust evidence that a ‘subset of overweight and moderately obese individuals with a high level of fitness may be protected from obesity-related health outcomes’, quoting the review from their group, which focused on cardiovascular disease (3). We did not examine the effect of interventions on those who are overweight as defined by BMI in our review. We think caution is required in suggesting that people with obesity or subgroups of obese patients are protected from obesity-related health outcomes. Firstly, a narrow focus on cardiovascular disease is likely to obscure harms that are relevant to patients such as diabetes, several common cancers, osteoarthritis, non-alcoholic fatty liver disease and cirrhosis, infertility, stress incontinence, erectile dysfunction, and most importantly overall quality of life. Secondly, there is no consensus for Koh’s suggestion that there is a group of people who can be easily and prospectively identified and are not at long-term risk of meaningful obesity-related complications.
References
(1) Sun X, Ioannidis JP, Agoritsas T, Alba AC, Guyatt G. How to use a subgroup analysis. Users’ guides to the medical literature. JAMA 2014;311:405-11.
(2) Look AHEAD Research Group. Cardiovascular effects of intensive lifestyle intervention in type 2 diabetes. N Engl J Med 2013;369:145-54 [Erratum in: N Engl J Med 2014;370:1866.].
(3) Kim SH, Després JP, Koh KK. Obesity and cardiovascular disease: friend or foe? Eur Heart J 2016;37:3560-8.
Competing interests:
No competing interests
12 December 2017
Alison Avenell
Professor
Chenhan Ma, Mark Bolland (University of Auckland, New Zealand), Jemma Hudson, Fiona Stewart, Clare Robertson, Pawana Sharma, Cynthia Fraser, Graeme MacLennan
University of Aberdeen
Health Services Research Unit, University of Aberdeen, Foresterhill, Aberdeen AB25 2ZD, Scotland, UK
Rapid Response:
Authors' reply to Ernsberger, Koh and Saripanidis
We thank you for your interest in our research.
Ernsberger and Køster-Rasmussen highlight the subgroup analysis for mortality for the importance of a physical activity facility as part of the intervention (RR 0.84, 95% CI 0.72 to 0.97). Firstly, statistical testing for this group against the groups that only got physical activity advice, or the groups that were not reported as receiving physical activity advice, was unable to demonstrate statistical heterogeneity (Chi² = 1.22, df = 2, P = 0.54, I² = 0%). Secondly, even if that was the case, subgroup testing in meta-analysis should be regarded with great caution (1), particularly as there are many other differences between the groups of trials in subgroup testing. Individual patient data meta-analyses would be required for further exploration of these subgroup findings for physical activity.
Ernsberger highlights the influence of the Look AHEAD trial (2), which carried 55% of the weighting in the meta-analysis of mortality. That is the reason that we undertook a post hoc analysis without this trial, which showed that weight loss interventions were still associated with decreased all-cause mortality (n=33 trials, 309 events; RR 0.78, 95% CI: 0.63 to 0.96; I2=0%). Sensitivity analysis to take account of rare events, using a random effects Bayesian logistic regression model, found a similar result.
As we acknowledge, we may have failed to identify all trials with outcome data, which may have biased results if trialists did not present these outcomes, particularly for those who dropped out, although we could not see obvious funnel plot asymmetry for all-cause mortality for our included studies. We also examined the drop out rate closely, it averaged 16% across all intervention groups and 16% across all control groups. We agree that trials described patient outcomes poorly, but think it unlikely that the many smaller trials with few events would have consistently failed to describe deaths in control groups when they described them in the intervention groups.
Saripanidis states that we conclude that ‘weight reducing diets do not reduce cardiovascular events/cardiovascular specific mortality/cancer specific mortality’. This is not our conclusion. Our conclusion is that we have been unable to demonstrate an effect on these outcomes. As is clear from our report and appendices, many studies did not provide details on the causes of death (see particularly Appendix Table 1), or did not collect or were unable to provide data on cardiovascular events. Thus our statistical power for these outcomes was limited.
Koh states that there is robust evidence that a ‘subset of overweight and moderately obese individuals with a high level of fitness may be protected from obesity-related health outcomes’, quoting the review from their group, which focused on cardiovascular disease (3). We did not examine the effect of interventions on those who are overweight as defined by BMI in our review. We think caution is required in suggesting that people with obesity or subgroups of obese patients are protected from obesity-related health outcomes. Firstly, a narrow focus on cardiovascular disease is likely to obscure harms that are relevant to patients such as diabetes, several common cancers, osteoarthritis, non-alcoholic fatty liver disease and cirrhosis, infertility, stress incontinence, erectile dysfunction, and most importantly overall quality of life. Secondly, there is no consensus for Koh’s suggestion that there is a group of people who can be easily and prospectively identified and are not at long-term risk of meaningful obesity-related complications.
References
(1) Sun X, Ioannidis JP, Agoritsas T, Alba AC, Guyatt G. How to use a subgroup analysis. Users’ guides to the medical literature. JAMA 2014;311:405-11.
(2) Look AHEAD Research Group. Cardiovascular effects of intensive lifestyle intervention in type 2 diabetes. N Engl J Med 2013;369:145-54 [Erratum in: N Engl J Med 2014;370:1866.].
(3) Kim SH, Després JP, Koh KK. Obesity and cardiovascular disease: friend or foe? Eur Heart J 2016;37:3560-8.
Competing interests: No competing interests