Being overweight in early adulthood is associated with increased mortality in middle age

Observational analyses of the association between body mass index (BMI) and all-cause mortality often suggest that overweight is neutral or beneficial, but such analyses are potentially confounded by smoking or by reverse causation. The use of BMI measured in early adulthood offers one means of reducing the latter problem. We used a cohort who were first measured while 16–24 year old students at Glasgow University in 1948–1968 and subsequently re-measured in 2000–2003, offering a rare opportunity to compare BMI measured at different ages as a predictor of mortality. Analysis of the later BMI measurements suggested that overweight was beneficial to survival, while analysis of BMI measured in early adulthood suggested that overweight was harmful and that the optimum BMI lay towards the lower end of the recommended range of 18.5–25 kg m−2. We interpret the association with later BMI as being probably distorted by reverse causality, although it remains possible instead that the optimum BMI increases with age. Differences when analyses were restricted to healthy non-smokers also suggested some residual confounding by smoking. These results suggest that analyses of BMI recorded in middle or old age probably over-estimate the optimum BMI for survival and should be treated with caution.


Supplementary
. Linear hazard ratios as in Table 2, but only adjusted for sex and date of birth (cubic splines). Supplementary Table S4. Linear hazard ratios as in Table 2, but for men only. Supplementary Table S5. Linear hazard ratios as in Table 2, but for women only. Supplementary Table S6. Linear hazard ratios as in Table 2, but excluding BMIs outside the 5 th -95 th percentiles. Supplementary Table S7. Linear hazard ratios as in Table 2, but with follow-up beginning after a ten-year lag. Supplementary Table S8. Hazard ratios by categories of BMI as in Table 3, but with follow-up beginning after a ten-year lag. Supplementary Table S9. Linear hazard ratios as in Table 2, but with additional adjustment for father's social class, height (cubic splines), birth order, number of siblings, pulse rate and age at menarche. Supplementary Table S10. Linear hazard ratios as in Table 2, but excluding those missing smoking data. Supplementary Table S11. Linear hazard ratios as in Table 2, showing effects of restriction to (i) healthy non-smokers, (ii) healthy subjects and (iii) non-smokers, at age 20.
Supplementary Figure S1. Hazard ratios relative to the first quartile, for quartiles of BMI at age 20. Supplementary Figure S2. Hazard ratios relative to the first quartile, for quartiles of BMI in 2001. Supplementary Figure S3. Fitted quadratic hazard ratios as in Fig. 2, but excluding BMIs outside the 5 th -95 th percentiles. Supplementary Figure S4. Fitted quadratic hazard ratios as in Fig. 2, but with follow-up beginning after a ten-year lag. Supplementary Figure S5. Fitted quadratic hazard ratios as in Fig. 2, but with additional adjustment for father's social class, height (cubic splines), birth order, number of siblings, pulse rate and age at menarche. Supplementary Figure S6. Fitted quadratic hazard ratios as in Fig. 2, but excluding those missing smoking data.
Supplementary Table S1. Conversion of International Statistical Classification of Diseases and Related Health Problems (ICD) codes into causes of death. ICD 9 was used exclusively for deaths occurring until 31 st December 2000 and ICD10 for deaths occurring from 1 st November 2000, with a ten-month transition period between.
Cause of death ICD 9 ICD 10 All cause all all Cardiovascular disease 3900-4599 I000-I999; G450-G459 Respiratory disease 4600-5199 J000-J999 Smoking-related cancer 1400-1509; 1570-1579; 1600-1639; 1650-1659; 1880-1899 Supplementary Table S2. Baseline characteristics of participants by quartile of BMI in 2001. BMI: body mass index, CI: confidence interval, SEP: socioeconomic position, MD: mean difference, OR, odds ratio. a Measured age 16-25. b Interval-scale variables are summarised as means within each quartile of BMI and mean differences per 5 kg m -2 of BMI from unadjusted linear regression are presented. c Binary variables are summarised as percentages within each quartile of BMI and odds ratios per 5 kg m -2 of BMI from unadjusted logistic regression are presented. d Measured 2000-2002 and BMI standardised to age 63. e The analysis of smoking uptake was restricted to non-smokers at age 20 and the analysis of smoking cessation was restricted to smokers at age 20. Supplementary Table S5. Linear hazard ratios for all-cause and cause-specific mortality in women only, per 5 kg m -2 of BMI. BMI: body mass index, HR: hazard ratio, CI: confidence interval. BMI was measured at approximately age 20 and around 2001, when subjects were aged 49-78. Cox proportional hazards regression was used with age as the time axis. Models were adjusted for sex, date of birth (cubic splines) and smoking behaviour at the time of BMI reporting. Nonlinearity was assessed by adding BMI 2 to the model and assessing its coefficient's departure from the null (Pquadratic). BMIvertex was estimated as the BMI at which the tangent to the quadratic model was horizontal. In a convex quadratic curve (coefficient for BMI 2 >0), BMIvertex estimates the BMI at which mortality is minimised. BMIvertex values outside the observed range of BMI (11.8 to 44.7 kg m -2 ) indicate a monotonically increasing or decreasing association among the observed data and were abbreviated for ease of presentation. Supplementary Table S10. Linear hazard ratios for all-cause and cause-specific mortality per 5 kg m -2 of BMI, with those missing smoking data excluded instead of being included as a separate class. BMI: body mass index, HR: hazard ratio, CI: confidence interval. BMI was measured at approximately age 20 and around 2001, when subjects were aged 49-78. Cox proportional hazards regression was used with age as the time axis. Models were adjusted for sex, date of birth (cubic splines) and smoking behaviour at the time of BMI reporting. Nonlinearity was assessed by adding BMI 2 to the model and assessing its coefficient's departure from the null (Pquadratic). BMIvertex was estimated as the BMI at which the tangent to the quadratic model was horizontal. In a convex quadratic curve (coefficient for BMI 2 >0), BMIvertex estimates the BMI at which mortality is minimised. BMIvertex values outside the observed range of BMI (11.8 to 44.7 kg m -2 ) indicate a monotonically increasing or decreasing association among the observed data and were abbreviated for ease of presentation. Supplementary Figure S1. Hazard ratios for all-cause and cause-specific mortality, relative to the first quartile, for sex-specific quartiles of body mass index (BMI) at age 20. Quartiles are plotted at their median values of BMI. Analyses were adjusted for sex, date of birth and smoking behaviour at age 20.
Supplementary Figure S2. Hazard ratios for all-cause and cause-specific mortality, relative to the first quartile, for sex-specific quartiles of body mass index ( Supplementary Figure S3. Fitted hazard ratios for all-cause and cause-specific mortality from the quadratic model of BMI at age 20, including only those with BMI between the 5 th and 95 th percentiles. The midpoint of the recommended BMI range (18.5-25 kg m -2 ) is used as the reference point. Analyses were adjusted for sex, date of birth and smoking behaviour at age 20.
Supplementary Figure S4. Fitted hazard ratios for all-cause and cause-specific mortality from the quadratic model of BMI at age 20, with follow-up beginning after a 10-year lag. The midpoint of the recommended BMI range (18.5-25 kg m -2 ) is used as the reference point. Analyses were adjusted for sex, date of birth and smoking behaviour at age 20.  Figure S5. Fitted hazard ratios for all-cause and cause-specific mortality from the quadratic model of BMI at age 20, with additional adjustment. The midpoint of the recommended BMI range (18.5-25 kg m -2 ) is used as the reference point. Analyses were adjusted for sex, date of birth (cubic splines), smoking behaviour at age 20, father's social class, height (cubic spline), birth order, number of siblings, pulse rate and age at menarche.
Supplementary Figure S6. Fitted hazard ratios for all-cause and cause-specific mortality from the quadratic model of BMI at age 20, with those missing smoking data excluded instead of being included as a separate class. The midpoint of the recommended BMI range (18.5-25 kg m -2 ) is used as the reference point. Analyses were adjusted for date of birth and smoking behaviour at age 20.