Journal of Hypertension:
Obesity paradox or inappropriate study designs? Time for life-course epidemiology
Ferreira, Isabela,b,c; Stehouwer, Coen D.A.b,c
aDepartment of Clinical Epidemiology and Medical Technology Assessment
bDepartment of Internal Medicine
cCARIM School for Cardiovascular Diseases, Maastricht University Medical Centre, Maastricht, The Netherlands
Correspondence to Isabel Ferreira, PhD, Department of Clinical Epidemiology and Medical Technology Assessment, Maastricht University Medical Centre, P. Debyelaan 25, 6202 AZ Maastricht, The Netherlands. Tel: +31 43 387 4386; fax: +31 43 387 4419; e-mail: email@example.com
Obesity is a major risk factor for cardiovascular disease (CVD) and all-cause mortality . On the basis of individuals’ level of BMI, this relationship is characterized by a J-shape with the nadir – that is, the point of the highest survival rate – around 22.5–25 kg/m2. Paradoxically, among patients with established CVD, this relationship seems to change in such a way that the nadir shifts to the right, with levels of BMI belonging to the overweight–obese categories apparently conferring the highest survival benefit (Fig. 1) the obesity paradox.
The paradox was first described in overweight and obese patients undergoing haemodialysis [2,3], and has since been confirmed in many other patient populations, such as those with congestive heart failure , stable coronary heart disease (CHD) [5,6] and acute coronary syndromes , peripheral artery disease  and stroke . The scientific literature on the topic has increased exponentially over the last decade, leading some to question the wisdom of targeting overweight and obesity in patients with prior CVD [7,10], as is currently recommended by European  and American  guidelines for (secondary) prevention and cardiovascular risk reduction therapy. Indeed, it seems that currently no one questions whether the existence of the paradox is real (e.g. it is listed in recent guidelines , though without a link to clinical practice recommendations, possibly because its meaning is unclear). Instead, research now seems to focus on finding the underlying (pathophysiological) explanations for it. In this line, many studies have focused on the extent to which BMI was an indicator of excessive body fatness in patients with CVD [13,14], with a recent meta-analysis confirming the paradox for BMI but not for waist circumference, when both anthropometric measures were adjusted for one another in the prediction of mortality in patients with CHD . Nevertheless, it is surprising that few efforts have been put in addressing the serious methodological flaws in the designs and analyses of the studies that ignited the whole discussion in the first place. Specifically, selection bias and ignored time-varying nature of exposure have been overwhelmingly (and repeatedly) overlooked, and seldom acknowledged, thus not yet enabling firm conclusions such as that implied by some of the paradox literature.
Selection bias. The potential impact of selection bias on the so-called paradoxes emerging from recurrent risk medical research has been recently illustrated by Dahabreh and Kent , who refer to this problem as index event bias. More generally, the epidemiologic literature describes this phenomenon as a form of selection bias  or, more specifically, as collider bias . Selection bias affecting recurrent risk research is that, when multiple risk factors contribute to the risk of an index disease (denoted by C, which can take values 0, absent or 1, present) and its progression/recurrence, conditioning on the index disease (C = 1) induces a spurious association between the different risk factors. The selection based on C = 1 thus influences the distribution of risk factors in the patients included in the study in such a way that it will affect the associations between the risk factors and thereby with a recurrent event in an unpredictable way.
This phenomenon is partially illustrated in Table 1, which describes the data on the distributions of patients older than 60 years and with hypertension by categories of BMI as reported in a general population (from the National Health and Nutrition Examination Survey III ) and in a population selected on the basis of having CHD [from the European Action on Secondary and Primary Prevention by Intervention to Reduce Events (EUROASPIRE) III survey ]. In the selected population, the prevalence of overweight and obese patients was much higher than in the general population (46.5 vs. 32.5% and 35.3 vs. 22.0%, respectively). In the general population, age and BMI were positively associated with each other (and with incident CVD). However, in the study confined to patients with established CVD (i.e. conditioning on the index disease), a negative relationship between age and BMI was observed. Indeed, overweight and obese CVD patients are usually younger than normal weight patients [4,6–10,15]. The impact of index event bias on the associations between related factors can take shapes other than reversing the direction of the association, as in the example of the age–BMI association given above. For instance, in the example given in Table 1, the strength of the association between BMI and hypertension as observed in the general population, seemed to have been greatly attenuated in the selected population, and this beyond what apparently could be explained by the negative confounding effect of age. In this example, we could not ascertain the magnitude of this potential confounding effect because cross-tabulation of hypertension by age was not provided (to enable some form of sensitivity analysis).
Furthermore, this problem also occurs between (known or unknown) risk factors for a disease that are distributed in the general population independently of one another. To illustrate this principle to medical students during their training on how to review critically scientific (prognostic) literature, we often use the following example: consider the relationship between empathy (A), knowledge (B) and professional success as a physician (C), in which empathy and knowledge are not associated in the general population (i.e. empathy does not cause knowledge, knowledge does not cause empathy, and empathy and knowledge do not share a common cause). Suppose also that empathy and knowledge are separately sufficient for becoming a successful physician. Given these assumptions, professional success is defined as a collider variable. If one conditions on the collider, that is, by examining the relationship between empathy and knowledge only among successful physicians, this will result in the following observations:
1. Knowing that a nonempathic person is a successful physician implies that the person must be very knowledgeable.
2. Conversely, knowing that a nonknowledgeable person is a successful physician implies that the person is empathic.
3. Either way, by conditioning on the collider, we have created a negative association between empathy and knowledge among the successful physicians, even though empathy and knowledge were not related to each other in the first place.
In this example, we assumed A and B to be unrelated, but even in situations wherein A and B may be related to each other, the problem of selection bias remains (see examples of age and BMI and hypertension given above). Such relationships thus represent a mixture of the true association between A and B and the spurious association between them induced by conditioning on their common outcome. Because risk factors often have similar effects on the index and recurrent events, these biases will therefore taint any estimates on the effects of obesity on recurrent risk, unless one can account thoroughly for all shared risk factors ; still, unknown risk factors often operating in the causation of a disease will also suffer from the bias induced by conditioning on the index event, and therefore, residual bias will always be a concern .
Time-varying exposures. Even if one were to assume that no such thing as an index event bias has threatened the estimates provided in the obesity-recurrent risk literature thus far, another important methodological problem has remained overlooked: the temporally dynamic nature of the exposure (often BMI or other anthropometric measures). Indeed, practically, all cohort studies that have examined the association between BMI or waist circumference and recurrent disease or mortality risk among those who already had experienced a CVD event have measured exposure (i.e. the level of a patient's BMI or waist circumference) only once, at or around the time of the index event [baseline or time zero (t0) – Fig. 2].
Per definition, prospective cohort studies evaluate a possible association between exposure and outcome by following a group of individuals over a period of time, often years when in the context of a chronic disease. The incidence of disease in the exposed group is then compared with that in an unexposed group, and a relative risk (incidence risk or incidence rate ratio) is calculated to ascertain whether the exposure and disease are linked. Because a temporal relationship between exposure and outcome can be established, inferences of causality are often drawn. However, a key aspect to scrutinize in prospective cohort studies is how exposure was assessed. It is possible that the t0 measurement does not reflect the same burden of risk exposure if for some patients it has been present for a long (e.g. someone who has been obese throughout the whole adult age) or a short time (e.g. someone who has recently become obese). Furthermore, where exposure status can change during the course of follow-up, periodic assessments of exposure are necessary; otherwise, the estimates can be seriously compromised. This is due to the fact that person-years enter the denominators of the calculated incidence rates in the exposed and unexposed groups, and therefore, the changes in exposure status during follow-up need to be taken into account in these calculations. For instance, in the case of a 5-year follow-up study in which exposure is the patients’ level of BMI, with the normal weight category considered as referent (unexposed): if a patient is overweight for the first 2 years of the follow-up, but afterwards, due to the loss of body weight, switches to normal weight and maintains this exposure status until the end of follow-up, that patient would have only contributed 2 person-years of risk to the exposed group (and 3 to the unexposed), assuming he/she had not developed the outcome. Ignoring the possibility of change in the exposure status would mean that the person would have contributed 5 person-years to the exposed grouped (and 0 to the unexposed) instead! Likewise, possible modifying or confounding factors (e.g. smoking behaviour, dietary and physical activity habits, drug treatments) to the obesity–recurrent event relationship most likely also change over time, and thus need to be accounted for, because their cardiovascular protective effects may occur even before or in the absence of effective changes in BMI.
The results derived from the vast majority of the obesity-recurrent studies suggesting the existence of an obesity paradox have thus relied on the false assumption that patients’ exposure status would have remained unchanged, a premise hard to accept in the context of secondary prevention of CVD. In fact, the same problem also applies in the context of primary prevention, but is thought to be of a smaller magnitude in that setting. Experiencing a myocardial infarction or a stroke or undergoing a coronary artery bypass graft surgery represents a major threat to a patient's life, which is likely to lead to greater subsequent intentional changes in one's lifestyle (and thus body weight) than when such an event has not yet occurred [22,23]. Such changes are likely to differ between patients who were overweight or obese at the time of the index event and those who were not, because excessive weight constitutes an indication for intensive lifestyle modification counselling and treatment . Indeed, the report from the EUROASPIRE III survey, the aim of which was to evaluate the Joint European Societies guidelines on cardiovascular disease prevention in everyday clinical practice, shows that patients who, at the time of the index event, were overweight and obese were more likely to engage in attempts to lose weight and adopt specific lifestyle (mainly dietary) changes to achieve that than those who had normal weight . In addition, although achievement of targets according to the guidelines remained suboptimal in all groups, weight loss was more often and weight gain was less often observed in those with overweight and obesity than in those with normal weight. Because estimates of incident risk across BMI categories are relative to the normal weight group, the fact that, during the course of follow-up, overweight and obese individuals display healthier trajectories of lifestyle and weight changes than those in the normal weight category, may account in part for the apparent paradoxical observations of less incident or recurrent risk in these groups (Fig. 3).
In this issue of the Journal of Hypertension, Dorresteijn et al. report on the associations of several measures of general and central adiposity with blood pressure (BP) in a well characterized cohort of men and women with symptomatic vascular disease enrolled in the Secondary Manifestations of Arterial Disease Study. In agreement with the phenomenon of redistribution of risk factors when conditioning on the index event, the prevalence of overweight (48.5%) and obesity (19.7%) was appreciably higher in this patient population than in the general Dutch population. In addition, and particularly in men, overweight and obese patients tended to be younger than those with normal weight. The levels of both SBP and DBP (and other cardiovascular risk factors) tended to be more adverse in those with overweight and obesity than in those with normal weight.
After adjustment for differences in age, sex, estimated glomerular filtration rate, type of prior event and carefully accounting for the effects of antihypertensive treatment on the measured levels of BP, patients with higher levels of general (BMI) or central adiposity (waist circumference and visceral adipose tissue mass) had significantly higher levels of both SBP and DBP levels. Although the authors acknowledge the potential effects of index event bias, and despite the careful nature of their data analyses, it remains that potential bias induced by conditioning on the event index on nonmeasured risk factors for both BMI and hypertension (e.g. dietary habits including alcohol intake, physical activity levels) may have remained present.
Still, the most interesting part of their report refers to the analyses of how changes in BMI (and other markers of adiposity) were related to concomitant changes in BP, because they account for the time-varying nature of the exposure (adiposity) in the context of secondary prevention. These analyses were conducted in a small subpopulation (n = 185) for whom measures of both (central) adiposity and BP were obtained at baseline and after 3.7 years of follow-up. Appreciation of the data at baseline and at follow-up (manuscript's online appendix table 2) reveals a nonsignificant overall decrease in patients’ weight (∼−0.7 kg) not appreciably affecting their changes in BMI, an increase in waist circumference (∼+2.2 cm) and a significant increase in the waist-to-hip ratio; abdominal subcutaneous but not visceral adipose tissue decreased significantly. Overall, patients’ antihypertensive treatment tended to be intensified at follow-up, with changes in the constellation of drugs used (more diuretics, angiotensin-converting enzyme inhibitors and angiotensin-2-receptor antagonists and less beta-blockers). Mainly the diastolic component of BP tended to decrease, and this, combined with a slight increase in SBP, resulted in a significant increase in pulse pressure (PP), a marker of arterial stiffening and a potent predictor of CVD risk. To some extent, these data suggest that increased PP and SBP remain suboptimally controlled in secondary prevention.
In line with the cross-sectional results, changes in body weight, BMI or waist circumference were positively associated with changes in BP. Likely due to the small numbers in the longitudinal subset, the reader is not informed about how many patients gained weight and how many lost weight, nor the extent to which patients who lost weight were those with higher levels of BMI at baseline and patients who gained weight were those with lower levels of BMI at baseline (the phenomenon illustrated in Fig. 3, excluding the extreme categories of BMI). Still, these findings are in agreement with observational and trial studies in the context of cardiac rehabilitation showing decreases in BP and several other risk factors accompanying decreases in body weight in these patients (reviewed in ). Taken together, these longitudinal findings, which per design are more robust, thus challenge the yet to be established (if any) pathophysiological mechanisms supporting the ‘obesity paradox’ theory.
However, the extent to which a favourable impact of weight loss in overweight or obese patients’ risk factor profile will ultimately result in reduced recurrence of CVD and mortality risk remains to be established. There is a remarkable paucity of information in this regard. Sierra-Johnson et al. have shown a more favourable prognosis (composite of mortality and acute cardiovascular events) among CHD patients who lost vs. those who did not loose weight after the index event, regardless of their initial BMI level. Lavie et al. showed that purposeful weight loss among overweight/obese patients who had been enrolled in a cardiac rehabilitation and exercise training programme resulted in improvements in exercise capacity, plasma lipids, levels of inflammation and quality of life, and a trend towards lower mortality, although the latter did not reach statistical significance. Myers et al., on the contrary, reported higher all-cause mortality among male patients who lost weight and lower mortality in patients who gained weight than among those whose weight remained stable. Similar but nonsignificant trends were observed when cardiovascular mortality was the study outcome instead, but nonvolitional weight loss related to occult disease could not be ruled out as an explanation for the paradoxical observations .
Clearly, these issues need to be better examined in larger longitudinal observational and randomized controlled studies, because accepting or dismissing the obesity paradox theory may result in completely different practices on how to ‘handle’ obesity in the context of secondary cardiovascular prevention.
Time for appropriate study designs – life-course epidemiology as a framework. The challenges of temporally varying exposures and outcomes, such as observed in the context of determinants of recurrent risk, can now be satisfactorily met using modern epidemiologic approaches for the design and analysis of cohort or trial studies. In order to meet these challenges, new biostatistical methods now make possible incorporation of temporally varying exposures and recurrent outcomes by means of longitudinal analyses of the data (see for examples [29–31]). Application of these analytic methods has been facilitated by the availability of software-making analyses that could not have been contemplated one or two decades ago, highly accessible if not a must now.
Life-course epidemiology conceptual models may prove very useful as a framework for the study of the determinants of recurrent risk. A life-course epidemiological approach to chronic disease uses a multidisciplinary framework to understand the importance of time and timing in associations between exposures and outcomes at the individual and population levels . Although this framework has been more often applied in the context of primary prevention (e.g. early life periods and trajectories), it can and should be extended to secondary prevention. After all, with the ageing of populations worldwide and the improvements in medical care, more and more years of a person's life will be spent in the context of secondary prevention than was usual some decades ago.
I.F. is supported by a senior postdoctoral research grant (#2006T050) from the Netherlands Heart Foundation.
Conflicts of interest
There are no conflicts of interest.
1. Whitlock G, Lewington S, Sherliker P, Clarke R, Emberson J, Halsey J, et al. Body-mass index and cause-specific mortality in 900 000 adults: collaborative analyses of 57 prospective studies. Lancet 2009; 373:1083–1096.
2. Fleischmann E, Teal N, Dudley J, May W, Bower JD, Salahudeen AK. Influence of excess weight on mortality and hospital stay in 1346 hemodialysis patients. Kidney Int 1999; 55:1560–1567.
3. Schmidt DS, Salahudeen AK. Obesity-survival paradox: still a controversy? Semin Dial 2007; 20:486–492.
4. Oreopoulos A, Padwal R, Kalantar-Zadeh K, Fonarow GC, Norris CM, McAlister FA. Body mass index and mortality in heart failure: a meta-analysis. Am Heart J 2008; 156:13–22.
5. Romero-Corral A, Montori VM, Somers VK, Korinek J, Thomas RJ, Allison TG, et al. Association of bodyweight with total mortality and with cardiovascular events in coronary artery disease: a systematic review of cohort studies. Lancet 2006; 368:666–678.
6. Oreopoulos A, Padwal R, Norris CM, Mullen JC, Pretorius V, Kalantar-Zadeh K. Effect of obesity on short- and long-term mortality postcoronary revascularization: a meta-analysis. Obesity (Silver Spring) 2008; 16:442–450.
7. Angeras O, Albertsson P, Karason K, Ramunddal T, Matejka G, James S, et al. Evidence for obesity paradox in patients with acute coronary syndromes: a report from the Swedish Coronary Angiography and Angioplasty Registry. Eur Heart J 2012 [Epub ahead of print] doi: 10.1093/eurheartj/ehs217.
8. Galal W, van Gestel YR, Hoeks SE, Sin DD, Winkel TA, Bax JJ, et al. The obesity paradox in patients with peripheral arterial disease. Chest 2008; 134:925–930.
9. Ovbiagele B, Bath PM, Cotton D, Vinisko R, Diener HC. Obesity and recurrent vascular risk after a recent ischemic stroke. Stroke 2011; 42:3397–3402.
10. Anker SD, von Haehling S. The obesity paradox in heart failure: accepting reality and making rational decisions. Clin Pharmacol Ther 2011; 90:188–190.
11. Perk J, De Backer G, Gohlke H, Graham I, Reiner Z, Verschuren M, et al. European Guidelines on cardiovascular disease prevention in clinical practice (version 2012). The Fifth Joint Task Force of the European Society of Cardiology and Other Societies on Cardiovascular Disease Prevention in Clinical Practice (constituted by representatives of nine societies and by invited experts). Developed with the special contribution of the European Association for Cardiovascular Prevention & Rehabilitation (EACPR). Eur Heart J 2012; 33:1635–1701.
12. Smith SC Jr, Benjamin EJ, Bonow RO, Braun LT, Creager MA, Franklin BA, et al. AHA/ACCF secondary prevention and risk reduction therapy for patients with coronary and other atherosclerotic vascular disease: 2011 update: a guideline from the American Heart Association and American College of Cardiology Foundation endorsed by the World Heart Federation and the Preventive Cardiovascular Nurses Association. J Am Coll Cardiol 2011; 58:2432–2446.
13. Romero-Corral A, Somers VK, Sierra-Johnson J, Jensen MD, Thomas RJ, Squires RW, et al. Diagnostic performance of body mass index to detect obesity in patients with coronary artery disease. Eur Heart J 2007; 28:2087–2093.
14. Lavie CJ, De Schutter A, Patel DA, Romero-Corral A, Artham SM, Milani RV. Body composition and survival in stable coronary heart disease: impact of lean mass index and body fat in the ‘obesity paradox’. J Am Coll Cardiol 2012; 60:1374–1380.
15. Coutinho T, Goel K, Correa de Sa D, Kragelund C, Kanaya AM, Zeller M, et al. Central obesity and survival in subjects with coronary artery disease: a systematic review of the literature and collaborative analysis with individual subject data. J Am Coll Cardiol 2011; 57:1877–1886.
16. Dahabreh IJ, Kent DM. Index event bias as an explanation for the paradoxes of recurrence risk research. JAMA 2011; 305:822–823.
17. Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias. Epidemiology 2004; 15:615–625.
18. Cole SR, Platt RW, Schisterman EF, Chu H, Westreich D, Richardson D, Poole C. Illustrating bias due to conditioning on a collider. Int J Epidemiol 2010; 39:417–420.
19. Brown CD, Higgins M, Donato KA, Rohde FC, Garrison R, Obarzanek E, et al. Body mass index and the prevalence of hypertension and dyslipidemia. Obes Res 2000; 8:605–619.
20. De Bacquer D, Dallongeville J, Heidrich J, Kotseva K, Reiner Z, Gaita D, et al. Management of overweight and obese patients with coronary heart disease across Europe. Eur J Cardiovasc Prev Rehabil 2010; 17:447–454.
21. VanderWeele TJ, Hernan MA, Robins JM. Causal directed acyclic graphs and the direction of unmeasured confounding bias. Epidemiology 2008; 19:720–728.
22. van Gool CH, Kempen GI, Penninx BW, Deeg DJ, van Eijk JT. Chronic disease and lifestyle transitions: results from the Longitudinal Aging Study Amsterdam. J Aging Health 2007; 19:416–438.
23. Keenan PS. Smoking and weight change after new health diagnoses in older adults. Arch Intern Med 2009; 169:237–242.
24. Dorresteijn JAN, Spiering W, Van Der Graaf Y, Visseren FLJ, on behalf of the SMART Study Group. Relation between adiposity and hypertension persists after onset of clinically manifest arterial disease. J Hypertens 2012; 30:2331–2337.
25. Ades PA, Savage PD, Harvey-Berino J. The treatment of obesity in cardiac rehabilitation. J Cardiopulm Rehabil Prev 2010; 30:289–298.
26. Sierra-Johnson J, Romero-Corral A, Somers VK, Lopez-Jimenez F, Thomas RJ, Squires RW, Allison TG. Prognostic importance of weight loss in patients with coronary heart disease regardless of initial body mass index. Eur J Cardiovasc Prev Rehabil 2008; 15:336–340.
27. Lavie CJ, Milani RV, Artham SM, Patel DA, Ventura HO. The obesity paradox, weight loss, and coronary disease. Am J Med 2009; 122:1106–1114.
28. Myers J, Lata K, Chowdhury S, McAuley P, Jain N, Froelicher V. The obesity paradox and weight loss. Am J Med 2011; 124:924–930.
29. Twisk JW, Smidt N, de Vente W. Applied analysis of recurrent events: a practical overview. J Epidemiol Community Health 2005; 59:706–710.
30. Tilling K, Sterne JA, Szklo M. Estimating the effect of cardiovascular risk factors on all-cause mortality and incidence of coronary heart disease using G-estimation: the Atherosclerosis Risk in Communities Study. Am J Epidemiol 2002; 155:710–718.
31. Hernan MA, Cole SR, Margolick J, Cohen M, Robins JM. Structural accelerated failure time models for survival analysis in studies with time-varying treatments. Pharmacoepidemiol Drug Saf 2005; 14:477–491.
32. Lynch J, Smith GD. A life course approach to chronic disease epidemiology. Annu Rev Public Health 2005; 26:1–35.
© 2012 Lippincott Williams & Wilkins, Inc.