Invited editorialfor JAMA Oncology: revised 20th March 2016
The ‘obesity paradox’ and mortality after colorectal cancer: a causal conundrum
Andrew G Renehan,1Matthew Sperrin2
1 Institute of Cancer Sciences, University of Manchester, Manchester, UK
2 Health eResearch Centre, Farr Institute, Manchester Academic Health Science Centre, University of Manchester, Manchester, UK
Running title: The obesity paradox and colorectal cancer
Keywords: obesity; colorectal cancer, survival
Correspondence to:
Professor Andrew Renehan
Institute of Cancer Sciences
University of Manchester
Manchester Academic Health Science Centre
The Christie NHS Foundation Trust
Wilmslow Road
Manchester M20 4BX United Kingdom
Tel: +44 161 446 3157
E-mail:
Main text: 1534words (between 1200-1500); 15 references (max: 15); language: US English
Excess body adiposity, commonly approximated as elevated body mass index (BMI), is an established risk factor for increased incidence of several adult cancer types.1Worldwide, each year, almost half-million new cancers are attributed to elevated BMI, of which breast, colorectal and endometrial cancers account for two-thirds.2 By extension, elevated BMI at or after cancer diagnosis might be associated with a poor prognosis in obesity-related cancers, and indeed, many studies show such associations, especially for breast cancer.3 This is a key rationale for weight management strategies in cancer survivorship programs and endorsed by clinical guidelines,for example, those from the American Society for Clinical Oncology.4
However, it is becoming increasingly clear that the relationship between BMI and cancer survival is far more complex than a straightforward extrapolation of BMI-cancer incidence associations. Thus, in the example of endometrial cancer, where there is arguably the strongest association per incremental unit increase of baseline BMI and incidence risk, the association between peri-operative BMI and survival, evaluated as a secondary analysis within a large trial (to minimize treatment-related biases), is null.5 The World Cancer Research Fund report6on breast cancer survivors makes an additional important methodological contribution, pointing to the need to consider the timing of when BMI (or other anthropometric) measures are determined, namely pre-diagnosis (typically several years as a surrogate of steady-state BMI pre-cancer); peri-diagnosis; and 1 to 2 years post cancer treatment (as a surrogate of cancer survivorship BMI) – as different patterns of association emerge.
A further dimension to this relationship is the emerging topic of the ‘obesity paradox’. This is the observation of an apparent risk reductionin an outcome of interest (usually mortality) of elevated BMI (≥ 25 kg/m2) compared with normal weight (BMI: 18.5 to 25 kg/m2), where an increased risk is anticipated. The obesity paradox is well recognized in the cardiovascularand metabolic literatures, and is now being recognized in oncology; for example, after surgery for clear-cell renal carcinoma7 and in a population-level study of mortality after colorectal cancer diagnosis.8
In this issue of JAMA Oncology, Candyce Kroenke et al.9 add valuable new dimensions to this theme. These investigators evaluated the impact of BMI on mortality in 3408 patients with stages I to III colorectal cancer treated by surgery from the well-established Kaiser Permanente Northern California (KPNC) healthcare database. They uniquely had BMI values at pre-diagnosis, at cancer diagnosis, and at 15 months following diagnosis. Using low range of normal-weight at diagnosis (BMI: 18.5 to < 23.0 kg/m2) as their referent, they observed lower mortality risks for high-normal-weight (23.0 to < 25.0 kg/m2), low-overweight (25.0 to < 28.0 kg/m2), high-overweight (28.0 to < 30.0 kg/m2), no risk difference for class I obese (30.0 to < 35.0 kg/m2) and increased risk for classes II and III obese (≥ 35.0 kg/m2). They found similar patterns for cancer-specific survival. They interpreted these observations as consistent with an obesity paradox and set about to explain these relationships within a causal methodological framework, known as directed acyclic graphs (DAGs). They specifically sought to explore the role of the collider bias, caused through the presence of pre-diagnosis BMI and other confounders as mutual risk factors for colorectal incidence and mortality. In their multivariate analysis, afteradjusting for pre-diagnosis BMI, the obesity paradox remained. The authors concluded that the paradoxical association (“overweight patients had a lower mortality after colorectal cancer diagnosis”) was real and recommended that “(intentional) weight loss in the immediate post-diagnosis period may be unwarranted”.
The authors are to be complimented for highlighting the complexities in the analyses of peri-diagnosis anthropometric measurements in patients with obesity-related cancers and subsequent interpretations of impact on these exposure on outcomes. The authors are similarly to be acknowledged for emphasizing the utility of DAGs as methodological frameworks to interrogate causal links between an exposure and outcome. The principles of DAGs are an extension of logic, familiar to students of philosophy, butless familiar to many oncology readers. Appropriately used, DAGsare key to identifying confounding and conceptualizing hard to appreciate biases, such as collider bias.
So what are the causes of the obesity paradox? Have Kroenke and colleagues9completely uncovered these? And what are the clinical implications? Banack and Kaufman10categorizedthe many causes of the obesity paradox into three broad groups. The first two of these is that associations reflect methodological issues – in other words, the associations are spurious. Examples might include the following: BMI is a sub-optimal approximation of body adiposity; confoundingfrom cigarette smoking (as smoking is inversely related to BMI); and reverse causality. The latter is worth discussion. At presentation, colorectal cancer diagnosis is frequently associated with weight loss, even for low stage tumors. Importantly, the extent of such unintentional weight loss correlates with pre-diagnosis BMI such that patients with ‘bad prognosis’ cancers in elevated pre-diagnosis BMI categories migrate to lower BMI categories after cancer diagnosis. The net effect is that the prognostic discrimination of peri-diagnosis BMI on survival is attenuated to a null effect or even an inverse association for some elevated BMI categories compared with normal weight. Thus, one explanation of the findings from the KPNC analysis might be reverse causality.
There are additional sources of confounding that might contribute to the obesity paradox, but incompletely captured in the KPNC analysis. Compared with normal weight patients, among patients with elevated BMI, there may be differential allocation to and differential adherence to adjuvant chemotherapy; ceiling chemotherapy dosing in the presence of obesity remains common clinical practice; and there are complex inter-relationships between weight and emergency surgery for colon cancer and outcome.
To illustrate the obesity paradox, selection of the referent category is an additional critical point. In the KPNC analysis, the referent BMI range was 18.5 to 23 kg/m2, arguably less than the normal ‘BMI’ category in the age range for peak colorectal cancer presentation. A glance at the expanded BMI category analysis suggests that were the referent category BMI 23 to 25 kg/m2 chosen, the obesity paradox mighthave been less convincingly demonstrated.
The second group of explanations is a specific type of methodological issue, known as ‘collider bias’;a specific form of selection bias that occurs when common causes of a disease and an outcome affect inclusion into the analysis.11This bias occurs at the statistical analysis level – it is a mathematical twist - and persists even where traditional adjustments, like regression models, are used.12The selected population is then subject to new, or enhanced confounding compared with the unselected population.13Debate on the role of collider bias is prevalent in the methodological literature. However, this mechanism might be overstated. Using an equation derived approach within a counterfactual framework, we have shown that the confounding (between exposure and outcome) needs to be three times greater than the true causal effect to observe an association between the exposure and outcome in the opposite direction.14 Collider bias is best illustrated by contrasting the relationship between BMI and mortality in patients with and without the obesity-related cancer drawn from the total population. In the cancer sub-population, one observes the paradox, which is absent in the larger non-cancer sub-population. The KPNC analysis9correctly argues that the collider bias is accounted for through adjusting for pre-diagnosis BMI, but interpretation of the peri/post BMI coefficients is thereafter conditional on pre-diagnosis BMI remaining constant. Given the likely migration through BMI categories from pre- to post-diagnosis BMI for a substantial proportion of patients, the risk estimates before and after adjustment for pre-diagnosis BMI are unexpectedly similar.
In the relationship between BMI and colorectal cancer incidence, associations are gender-specific, being consistently stronger for men compared with women.1 In the KPNC data,9 the obesity paradox is strongly observed for men but barely apparent for women. While the test for gender-BMI interaction is non-significant, these gender patterns may reflect a greater selection bias for men than women, rather than a truly differential (protective) effect of excess adiposity.
The third explanation is that the obesity-paradox is a real association. Here, there are underlying assumptions that excess adiposity might confer some ‘risk reduction’ effect, for example, obesity-related tumors might have a less aggressive biology or might respond more favourable to treatment. An alternative hypothesis is that obesity represents an energy ‘reserve’ state better equipped (than normal weight) to combat the physiological stresses of oncological treatments. However, to arrive at the conclusion of that the obesity paradox is a real association, there is first a need to exhaustively exclude many causes of confounding or biases, listed above. This is central to clinical implications. In the cardiovascular literature, some commentatorshave rushed to (mis)interpret that the obesity paradox is a true causal relationship believing that“the optimum body weight is shifting to a higher BMI range, once a chronic disease of some kind is present” - a paradigm rather than a paradox.15We urge extreme caution, here. While we rapidly appreciate that the interpretation of the relationship between BMI and mortality in patients with cancer is considerably more challenging than that between BMI and incidence cancer risk, we have a duty to appropriately convey this complexity to our patients and populations.The observations of Kroenke et al.9might be paradoxical, but these are readily explained through several methodological mechanisms. These findings should not alter current clinical practice or recommendations.
Competing interests
No conflict of interest.
References
1.Renehan AG, Zwahlen M, Egger M. Adiposity and cancer risk: new mechanistic insights from epidemiology. Nature reviews. Cancer. Aug 2015;15(8):484-498.
2.Arnold M, Pandeya N, Byrnes G, et al. Global burden of cancer attributable to high body-mass index in 2012: a population-based study. The Lancet. Oncology. Jan 2015;16(1):36-46.
3.Chan DS, Vieira AR, Aune D, et al. Body mass index and survival in women with breast cancer-systematic literature review and meta-analysis of 82 follow-up studies. Ann Oncol. Oct 2014;25(10):1901-1914.
4.Ligibel JA, Alfano CM, Courneya KS, et al. American Society of Clinical Oncology position statement on obesity and cancer. J Clin Oncol. Nov 1 2014;32(31):3568-3574.
5.Crosbie EJ, Roberts C, Qian W, Swart AM, Kitchener HC, Renehan AG. Body mass index does not influence post-treatment survival in early stage endometrial cancer: results from the MRC ASTEC trial. Eur J Cancer. Apr 2012;48(6):853-864.
6.WCRF. World Cancer Research Fund International. Continuous Update Project Report: Diet, Nutrition, Physical Activity, and Breast Cancer Survivors. 2014. Available at:
[accessed 20th Dec 2014]. 2014.
7.Hakimi AA, Furberg H, Zabor EC, et al. An epidemiologic and genomic investigation into the obesity paradox in renal cell carcinoma. J Natl Cancer Inst. Dec 18 2013;105(24):1862-1870.
8.Schlesinger S, Siegert S, Koch M, et al. Postdiagnosis body mass index and risk of mortality in colorectal cancer survivors: a prospective study and meta-analysis. Cancer Causes Control. Oct 2014;25(10):1407-1418.
9.Kroenke CH, Neugebauer R, Meyerhardt J, et al. Analysis using causal diagrams to evaluate the obesity paradox in colorectal cancer patients from the Kaiser Permanente Northern California population [in press]. JAMA Oncology. 2016.
10.Banack HR, Kaufman JS. The obesity paradox: understanding the effect of obesity on mortality among individuals with cardiovascular disease. Prev Med. May 2014;62:96-102.
11.Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias. Epidemiology. Sep 2004;15(5):615-625.
12.Cole SR, Platt RW, Schisterman EF, et al. Illustrating bias due to conditioning on a collider. Int J Epidemiol. Apr 2010;39(2):417-420.
13.Banack HR, Kaufman JS. From bad to worse: collider stratification amplifies confounding bias in the "obesity paradox". Eur J Epidemiol. Oct 2015;30(10):1111-1114.
14.Sperrin M, Candlish J, Badrick E, Renehan A, Buchan I. Collider Bias Cannot Explain Most of the Obesity Paradox [in press]. Epidemiology. 2016.
15.Doehner W, von Haehling S, Anker SD. Protective overweight in cardiovascular disease: moving from 'paradox' to 'paradigm'. Eur Heart J. Oct 21 2015;36(40):2729-2732.
1