Correspondence
Does midlife obesity really lower dementia risk?
See Online for appendix
See Online for appendix
498
Midlife obesity is thought to increase the risk of dementia,1 but Qizilbash and colleagues found that dementia risk reduced with increasing BMI.2 This surprising finding, obtained from a study of almost 2 million primary care patients, remained unchanged in all sensitivity analyses. Reverse causation bias could be the primary explanation of the result, as most dementias are known to have a long incubation period during which weight loss is possible. The sensitivity analyses by Qizilbash and colleagues—restricted to participants younger than age 55 years at baseline and excluding dementia cases during the first 15 years of follow-up—may not fully eliminate this bias. To explore this issue further, we analysed data from the original Whitehall study of 18 823 British men aged 40–69 at baseline. 3 A major advantage of our study is the long mortality follow-up (45 years), allowing better control for reverse causation by exclusion of dementia deaths up to the first 35 years after baseline. There was a marginally increased risk of dementia in underweight participants (BMI less than 20 kg/m²) and an inverse trend in those with BMI above 25 kg/m² (appendix). After the removal of deaths in the first 25 years, 30 years, and 35 years of follow-up, the inverse trend with dementia risk in participants with BMI above 25 kg/m² became, if anything, stronger (ageadjusted and smoking-adjusted hazard ratio per 1 kg/m² higher BMI was 0·87, 95% CI 0·76–1·00). The hazard ratio of dementia was 2·11 (95% CI 1·22–3·64) for underweight compared with normal weight participants; a similar pattern was observed when analyses were restricted to those aged 40–54 years at baseline. These data replicate the
surprising inverse BMI–dementia association reported by Qizilbash and colleagues.2 The findings from both studies should be interpreted cautiously. Although exclusion of cases at the beginning of follow-up reduces reverse causation, it might introduce other biases. Under-ascertainment of dementia cases and lack of information about preclinical and prodromal cases might further contribute to differential ascertainment bias in these studies. Qizilbash and colleagues did not adjust their estimates for education; in our study, any social gradient in dementia was small (data not shown). Further research on BMI and dementia would benefit from a mendelian randomisation approach, which uses genotype–disease associations to evaluate risk factors. Genetic studies are generally less prone to reverse causation and confounding than observational studies. Furthermore, there are almost 100 obesity-related or adiposityrelated genetic variants available for large-scale mendelian randomisation examinations of BMI, dementia, and endophenotypes, such as cognitive decline and neuroimaging biomarkers of neurodegeneration.4,5 We declare no competing interests.
*Mika Kivimäki, Archana Singh-Manoux, Martin J Shipley, Alexis Elbaz [email protected] Department of Epidemiology and Public Health, University College London, WC1E 6BT , UK (MK, AS-M, MJS); Centre for Research in Epidemiology and Population Health, INSERM, U1018, Villejuif, France (AS-M, AE) 1
2
3
Daviglus ML, Bell CC, Berrettini W et al. National Institutes of Health State-of-theScience Conference statement: preventing Alzheimer disease and cognitive decline. Ann Intern Med 2010; 153: 176–81. Qizilbash N, Gregson J, Johnson ME, et al. BMI and risk of dementia in two million people over two decades: a retrospective cohort study. Lancet Diabetes Endocrinol 2015; 3: 431–36. Reid DD, Hamilton PJS, McCartney P, et al. Cardiorespiratory disease and diabetes among middle-aged male civil servants. Lancet 1974; 303: 469–73.
4
5
Debette S, Wolf C, Lambert JC, et al. Abdominal obesity and lower gray matter volume: a Mendelian randomization study. Neurobiol Aging 2014; 35: 378–86. Keller L, Xu W, Wang HX, Winblad B, Fratiglioni L, Graff C. The obesity related gene, FTO, interacts with APOE, and is associated with Alzheimer’s disease risk: a prospective cohort study. J Alzheimers Dis 2011; 23: 461–19.
Qizilbash and colleagues1 report that underweight in midlife increases dementia risk, whereas overweight in midlife lowers the risk. This finding of a protective effect of overweight is controversial and contradicts the current consensus of increased risk,2,3 whereas the association between midlife underweight and increased dementia risk2,3 has received little attention—until now. Qizilbash and colleagues’ research has been questioned4 because they considered individuals between 40 and 80 years of age at BMI assessment as a single group, mixing midlife and late-life BMI. For this reason, we analysed separately the dementia risk for three age groups at BMI assessment: 35–49 years, 50–69 years, and 70–80 years. Our data includes more than 80 000 Norwegian participants followed up for a maximum of 38 years for dementiarelated mortality.5 Two aspects of our findings need to be considered (appendix). First, underweight was associated with increased dementia risk, both when BMI was measured in midlife and in old age. Second, the association between overweight and dementia depended on age at BMI assessment: when BMI was assessed in early midlife there was no decreased dementia risk, but when BMI was assessed in the older age groups, high BMI was associated with decreased dementia risk. Such contrasting findings with age were also reported in a large review, in which it was concluded, in accordance with our findings, that high BMI at old age was associated with reduced dementia risk, whereas high BMI in midlife
www.thelancet.com/diabetes-endocrinology Vol 3 July 2015
Does midlife obesity really lower dementia risk?
See Online for appendix
See Online for appendix
498
Midlife obesity is thought to increase the risk of dementia,1 but Qizilbash and colleagues found that dementia risk reduced with increasing BMI.2 This surprising finding, obtained from a study of almost 2 million primary care patients, remained unchanged in all sensitivity analyses. Reverse causation bias could be the primary explanation of the result, as most dementias are known to have a long incubation period during which weight loss is possible. The sensitivity analyses by Qizilbash and colleagues—restricted to participants younger than age 55 years at baseline and excluding dementia cases during the first 15 years of follow-up—may not fully eliminate this bias. To explore this issue further, we analysed data from the original Whitehall study of 18 823 British men aged 40–69 at baseline. 3 A major advantage of our study is the long mortality follow-up (45 years), allowing better control for reverse causation by exclusion of dementia deaths up to the first 35 years after baseline. There was a marginally increased risk of dementia in underweight participants (BMI less than 20 kg/m²) and an inverse trend in those with BMI above 25 kg/m² (appendix). After the removal of deaths in the first 25 years, 30 years, and 35 years of follow-up, the inverse trend with dementia risk in participants with BMI above 25 kg/m² became, if anything, stronger (ageadjusted and smoking-adjusted hazard ratio per 1 kg/m² higher BMI was 0·87, 95% CI 0·76–1·00). The hazard ratio of dementia was 2·11 (95% CI 1·22–3·64) for underweight compared with normal weight participants; a similar pattern was observed when analyses were restricted to those aged 40–54 years at baseline. These data replicate the
surprising inverse BMI–dementia association reported by Qizilbash and colleagues.2 The findings from both studies should be interpreted cautiously. Although exclusion of cases at the beginning of follow-up reduces reverse causation, it might introduce other biases. Under-ascertainment of dementia cases and lack of information about preclinical and prodromal cases might further contribute to differential ascertainment bias in these studies. Qizilbash and colleagues did not adjust their estimates for education; in our study, any social gradient in dementia was small (data not shown). Further research on BMI and dementia would benefit from a mendelian randomisation approach, which uses genotype–disease associations to evaluate risk factors. Genetic studies are generally less prone to reverse causation and confounding than observational studies. Furthermore, there are almost 100 obesity-related or adiposityrelated genetic variants available for large-scale mendelian randomisation examinations of BMI, dementia, and endophenotypes, such as cognitive decline and neuroimaging biomarkers of neurodegeneration.4,5 We declare no competing interests.
*Mika Kivimäki, Archana Singh-Manoux, Martin J Shipley, Alexis Elbaz [email protected] Department of Epidemiology and Public Health, University College London, WC1E 6BT , UK (MK, AS-M, MJS); Centre for Research in Epidemiology and Population Health, INSERM, U1018, Villejuif, France (AS-M, AE) 1
2
3
Daviglus ML, Bell CC, Berrettini W et al. National Institutes of Health State-of-theScience Conference statement: preventing Alzheimer disease and cognitive decline. Ann Intern Med 2010; 153: 176–81. Qizilbash N, Gregson J, Johnson ME, et al. BMI and risk of dementia in two million people over two decades: a retrospective cohort study. Lancet Diabetes Endocrinol 2015; 3: 431–36. Reid DD, Hamilton PJS, McCartney P, et al. Cardiorespiratory disease and diabetes among middle-aged male civil servants. Lancet 1974; 303: 469–73.
4
5
Debette S, Wolf C, Lambert JC, et al. Abdominal obesity and lower gray matter volume: a Mendelian randomization study. Neurobiol Aging 2014; 35: 378–86. Keller L, Xu W, Wang HX, Winblad B, Fratiglioni L, Graff C. The obesity related gene, FTO, interacts with APOE, and is associated with Alzheimer’s disease risk: a prospective cohort study. J Alzheimers Dis 2011; 23: 461–19.
Qizilbash and colleagues1 report that underweight in midlife increases dementia risk, whereas overweight in midlife lowers the risk. This finding of a protective effect of overweight is controversial and contradicts the current consensus of increased risk,2,3 whereas the association between midlife underweight and increased dementia risk2,3 has received little attention—until now. Qizilbash and colleagues’ research has been questioned4 because they considered individuals between 40 and 80 years of age at BMI assessment as a single group, mixing midlife and late-life BMI. For this reason, we analysed separately the dementia risk for three age groups at BMI assessment: 35–49 years, 50–69 years, and 70–80 years. Our data includes more than 80 000 Norwegian participants followed up for a maximum of 38 years for dementiarelated mortality.5 Two aspects of our findings need to be considered (appendix). First, underweight was associated with increased dementia risk, both when BMI was measured in midlife and in old age. Second, the association between overweight and dementia depended on age at BMI assessment: when BMI was assessed in early midlife there was no decreased dementia risk, but when BMI was assessed in the older age groups, high BMI was associated with decreased dementia risk. Such contrasting findings with age were also reported in a large review, in which it was concluded, in accordance with our findings, that high BMI at old age was associated with reduced dementia risk, whereas high BMI in midlife
www.thelancet.com/diabetes-endocrinology Vol 3 July 2015
