Weight loss and mortality in the elderly: separating cause and effect

Abstract

Over 2000 years ago, Hippocrates recognized that ‘persons who are constitutionally very fat are more apt to die quickly than those who are thin’ (Aphorisms I, 44, qtd. in Katz et al.) [1]. Yet most observational studies in elderly populations report higher mortality among leaner persons and many report no association, or even an inverse association, between overweight and mortality [2-5]. These findings have led many to conclude that a heavier body weight is not detrimental in the elderly, and moreover, that weight loss may not be beneficial for overweight and obese elderly individuals. Such conclusions are paradoxical however, since low body mass index {BMI; weight (kg)/[height (m)]2} is protective for many incident diseases and few benefits of obesity have been documented. More likely, the apparently conflicting results do not reflect a causal relation between low BMI and mortality, but instead are driven by the high prevalence of chronic diseases in elderly populations that cause weight loss and also increase mortality risk. Studies of adiposity and mortality are particularly susceptible to methodological biases. These biases include confounding by smoking, over-control of biologic intermediates of adiposity (such as diabetes and hypertension), and most important in studies of the elderly, the potential for reverse causation – the effect of diagnosed or subclinical diseases that increase mortality risk and also cause weight loss or reduce weight gain that might have occurred in the absence of illness [6]. Physicians have known since the time of Hippocrates that many illnesses lead to weight loss or impaired weight gain and also increase the risk of death: ‘… following a long illness … the shoulders, clavicles, chest, and thighs melt away … this illness is fatal’ (Affections XXII, qtd. in Katz et al.) [1]. Lean people are a composite of smokers, ill persons and healthy persons, and thus they often appear to have a higher risk of death, especially among the elderly due to the higher prevalence of disease. Because of these methodological challenges, Dr. Frank Hu suggests healthy weight guidelines should be based on studies initiated during mid-life to mitigate the intractable bias from reverse causation in the elderly [5]. The study by Bamia et al. in this issue of the Journal of Internal Medicine makes a welcome and substantial contribution to resolving the controversy regarding the relation between adiposity and mortality in the elderly [7]. Particular strengths of the study include its efficient design and large sample size, as well as the authors’ thoughtful lagged analysis that illuminates the bias due to reverse causation. The authors’ initial analysis showed a strong positive association between weight loss and mortality, but most of this excess risk was observed in the year before death, reflecting reverse causation. Persons who lost more than 1 kg per year in the previous year were more than three-times as likely to die compared to those who maintained their body weight. In contrast, such weight loss measured 3 years earlier was associated with only a 27% increased risk of death [7]. As pointed out by Bamia et al. clinical or subclinical disease likely accounted for the strong association observed in the initial analysis and interpreting such results as causal (i.e. weight loss itself increased the risk of mortality) is inappropriate and misleading. Despite the authors’ careful study design and execution, the reported associations between weight loss and mortality and weight gain and mortality likely still reflect some degree of reverse causation. The authors appropriately excluded participants who had a diagnosis of chronic disease at the time of recruitment, but they did not exclude such individuals at the start of follow-up after weight re-assessment. Additionally, the longest lag period the authors could examine was 3 years [7]. As the authors acknowledge, 3 years is not sufficient to eliminate reverse causation since common conditions such as chronic obstructive pulmonary disease and congestive heart failure may remain undiagnosed for many years and cause weight loss or prevent typical weight gain. If individuals with such diseases are classified in the weight loss group, a spurious association between weight loss and increased mortality may be observed. If these individuals are classified in the maintained weight group, the adverse association between weight gain and mortality will be attenuated due to including ill persons with a high risk of death in the reference category (Fig. 1). Clinical or subclinical disease is associated with weight loss and also increases mortality risk independent of weight loss, resulting in a bias in the upward direction for estimates of the relation between weight loss and mortality. Similarly, subclinical disease is inversely related to weight gain, causing a bias in the downward direction for estimates of the relation between weight gain and mortality. The magnitude of these biases increases as the prevalence of illness approaches 50%, making this a particular concern in the elderly. Furthermore, weight change may not be the most informative metric in the elderly because changes in lean body mass and fat mass cannot be distinguished. As people age, they tend to lose muscle and bone mass and gain fat mass. Thus, a decrease in body weight in the elderly typically represents a loss of predominately lean mass (in contrast to middle-aged persons where changes in body weight generally reflect changes in fat mass) [5]. Likewise, elderly people who maintain their weight have typically lost lean mass and gained fat mass, which may have different implications for mortality risk than if they had maintained their body composition. Unfortunately, it is difficult to differentiate between lean body mass and fat mass in epidemiologic studies. Dual energy x-ray absorptiometry (DXA) is considered a gold standard for measuring fat-free mass, fat mass, and bone mass, but is expensive and not generally suitable for large epidemiologic studies. One potential alternative is change in waist circumference (WC). In the Health Professionals Follow-up Study, BMI predicted cardiovascular disease mortality among men younger than 65 years but not older men, while WC predicted cardiovascular disease mortality among both younger and older men [8]. Similarly, the Iowa Women’s Health Study reported that WC and waist-to-hip ratio were more strongly associated with mortality among women 55–69 years than BMI [9]. It is important to note that the use of sophisticated measures of body composition and adiposity (such as DXA) will not alone account for the biases present in studies of adiposity and mortality in the elderly. Misleading results, in which higher adiposity appears beneficial, may still be observed if smokers and persons with clinical or subclinical disease are not excluded from the study population [3]. Further exploration of the use of DXA and WC, as well as adipokines such as leptin and adiponectin, would be informative in well-designed, carefully conducted studies of adiposity in the elderly. Many studies have attempted to differentiate between intentional and unintentional weight loss, assuming intentional weight loss is less representative of weight loss due to disease and is mostly due to loss of fat mass. However, intentional weight loss is often accompanied by healthy behaviours (physical activity, dietary changes, etc.) that may independently reduce mortality risk and thus confounding by these factors is a serious concern in observational studies [5]. Moreover, some of the weight loss thought to be intentional may in fact be due to subclinical illness. Another strategy for identifying the true association between low BMI or weight loss in observational studies is to stratify by physical activity, since low BMI or weight loss among persons who engage in physical activity is less likely due to disease than among persons who are inactive. This was clearly demonstrated by Baik et al. who reported a two-fold increased risk of mortality among inactive men with a BMI <21.0 kg m−2 but no increase in mortality for active lean men [8]. There is still potential for reverse causation in such studies however, and thus randomized controlled trials of weight loss are needed to examine the relation between weight loss and mortality in the elderly. A recent secondary analysis of a randomized trial supports the beneficial effect of weight loss in overweight and obese elderly individuals on total mortality. The ADAPT trial was an 18-month randomized weight loss intervention among overweight and obese elderly persons (age: 69 ± 6 years; BMI: 34 ±5 kg m−2; 72% women) that found a 50% reduction (HR = 0.5; 95% CI = 0.3, 1.0; P = 0.03) in mortality among persons randomized to weight loss (mean weight loss = −4.8 kg) compared to the controls (mean weight loss = −1.4 kg) after 7 years [10]. Because of the randomization and exclusion of ill people at enrolment, bias due to reverse causation or confounding by lifestyle factors is minimized. Although the trial was small (n = 318), the results support the conclusion that intentional weight loss among overweight and obese elderly individuals is beneficial and clearly not detrimental to survival. These findings were indirectly supported by results from a weight loss trial that measured change in carotid vessel-wall volume, a marker of carotid atherosclerosis. Participants who lost even a modest amount of weight (−5.3 kg) were more likely to show regression of carotid vessel wall-volume after 2 years [11]. In conclusion, Bamia et al. masterfully demonstrated the significant bias due to reverse causation by clinical or subclinical disease that occurs in studies of adiposity in the elderly. Such bias may lead to erroneous conclusions that leanness or loss of fat mass in the elderly increases risk of mortality, but it is important to distinguish between a causal relation and a predictive relation. Low BMI certainly predicts higher mortality among older persons; it does not cause greater mortality – the clinical or subclinical disease that causes lower body weight causes a higher risk of mortality. Future studies in the elderly should follow the lead of Bamia et al. to examine sufficiently long lag periods to reduce bias due to disease, as well as utilize assessment methods that differentiate between changes in lean body mass and fat mass. In the meantime, prudent clinicians and public health officials will continue to emphasize the importance of physical activity and proper diet to promote maintenance of muscle and bone mass and prevent gain of fat mass in the elderly. No conflict of interest was declared.