Proponents of the argument that there is an “obesity paradox” (OP) in cardiovascular disease (CVD) cite research which notes the following: when studying a group of people who have an obesity-related comorbidity, e.g., CVD, those who have a body mass index (BMI) in the obese range do better than those who have a BMI in the healthy weight range. There have been hundreds of studies reporting these findings, and there are now updates (1), systematic reviews (2), and meta-analyses (3). Since obesity is believed to be a cause of obesity-related comorbidities, the proponents of the OP say that it's a paradox. However, these results are not shocking or paradoxical. In fact, they are logical, predictable, and due to biases that are easy to demonstrate through causal diagrams (46). Obesity is not a cause of improved cardiovascular health, even in those with CVD.

Definitions

It has been said that “most controversies would soon be ended if those engaged in them would first accurately define their terms, and then rigidly adhere to their definitions” (7). Although defining terms may seem tedious, given the mass confusion despite numerous commentaries explaining that the conclusions are systematically flawed (4,5,810), proper definitions are needed. Terms defined in the next sections are obesity, cause, and paradox. The purpose of this manuscript is to explain that there is no OP. As a recent letter to the editor stated, it is “time to focus on effective treatments” (5).

Obesity

To understand the misnomer of the OP, it is necessary to start with the definition of obesity. Obesity is defined as excessive body fat (adiposity) (11), although the exact percentage of fat that is “too much” is unclear. Body mass index, defined as a person's weight (in kilograms) divided by height in meters squared (kg · m−2), was developed by Quetelet in the mid-1800s as a crude measure of percentage body fat (12). Clinical recommendations categorize obesity as a BMI over 30. The BMI cutoff is used as a surrogate measure of obesity because stadiometers and scales are readily available as opposed to equipment such as magnetic resonance imaging or underwater weighing systems. On a population basis, BMI does well as a crude measure of excess adiposity. As stated by Garrows in 1985, “Quetelet's formula is both a convenient and reliable indicator of obesity” (13). That is, BMI is not synonymous with obesity.

The distinction between BMI and adiposity is a common discussion point in articles about the OP (1,3,14,15). While body composition, as opposed to body weight, affects cardiovascular health, in terms of the discussion about the so-called OP, distinguishing between BMI and adiposity is largely a distraction. While there may be some mismeasurement bias, there are undoubtedly other pervasive systematic reasons for the results, such as selection bias (4,16) and reverse causation (17). Body fat, especially visceral body fat, is a cause (see below for definition of cause) of cardiometabolic dysfunction (18,19), and BMI is simply a crude measure of body fat percentage. I am unaware of anyone suggesting that a BMI over 30 is a cause of cardiovascular dysfunction without excessive adiposity, and a BMI in the healthy range is not healthy if too much of the weight is composed of adipose tissue.

Cause

As stated in a recent and provocatively titled commentary, The C-word: scientific euphemisms do not improve causal inference from observational data (20), “using the term causal is necessary to improve the quality of observational research.” The false interpretation of studies reporting that there is an OP is a prime example where observational research would advance if people began to distinguish causal from noncausal associations.

Unfortunately, articles often couch their claims by stating that obesity is “associated with. . . ,” “a risk factor for. . . ,” or “linked to” CVD. This vagueness opens the door to misinterpretations such as the OP. For example, in July 2018, JAMA Cardiology published a study of nearly 200,000 individual participants in the US over the past 7 decades in the “[CVD] lifetime risk pool” (21). Among the summary points were: “Obesity was associated with a significantly increased risk for CVD,” and, “Obesity was associated with shorter longevity and a greater proportion of life lived with CVD” (21). Among those with CVD, those who were “normal weight” lived the fewest years with their CVD (21). These statements could be interpreted as support of the OP. However, it is difficult to believe the findings are due to excess adiposity being somehow a cause of improved health in those with CVD.

An exposure, such as obesity, is presumed to be a cause when “had the cause been altered, the effect would have been changed” (22). When researchers conduct an observational study of weight gain and subsequent cardiovascular morbidities, are they not asking whether obesity causes the morbidities? Given the associations, obesity could just be a marker for another cause of the morbidities. For example, there could be a gene and/or microbiome that causes people to become obese and also causes people to have CVD. If that was the sole explanation for the association of obesity and CVD, then weight loss would be meaningless, and all previous studies suggesting that obesity was harmful would need to be reinterpreted as due to confounding. However, there are a variety of mechanisms by which obesity/excess adiposity has direct harmful (causal) effects on cardiovascular health (2325). Weight loss decreases the risk of diabetes mellitus type 2 whether the weight loss was due to lifestyle changes, medications, or bariatric surgery (26,27).

Ambiguity about claims of causal effects can be minimized if researchers would make their assumptions about causes and effects explicit, e.g., with the use of causal directed acyclic graphs (DAGs) (28). A causal DAG joins variables via lines and single-headed arrows that travel from the causal variable to the variable that is thought to be changed by the cause (i.e., the effect). Causal DAGs are a valuable tool in explaining some of the logical flaws in the so-called OP. Figure 1 shows a causal DAG depicting how CVD could be caused by obesity and also caused by other things.

FIGURE 1.

Causal directed acyclic graph depicting causes of cardiovascular disease and mortality. “Other things” may be causes such as infections, cancer, or genetics. CVD = cardiovascular disease and mortality.

FIGURE 1.

Causal directed acyclic graph depicting causes of cardiovascular disease and mortality. “Other things” may be causes such as infections, cancer, or genetics. CVD = cardiovascular disease and mortality.

Close modal

A claim by proponents of the OP is that we should be cautious in recommending weight loss as weight loss is positively associated with mortality. Note that this is a different argument than the discussions about the ineffectiveness of weight loss promotion (29,30). A confounding factor in observational studies regarding body weight and mortality is that the dying process causes weight loss. This has been noted since the time of Hippocrates (31,32) and observed on a daily basis in homes and hospitals around the world. A more recent study suggests that the weight loss begins about 5–10 years prior to death (33). In a causal DAG, variables are defined in time. Obesity in the year 2010 can cause obesity in the year 2020, but not vice versa. A causal DAG such as that in Figure 2 shows that weight loss can be the result of either a healthy lifestyle (i.e., diet and exercise) or due to declining health. Note that the arrows signify that the variable has an effect on the outcome. The arrows do not specify whether the effect is an increase or a decrease in the outcome.

FIGURE 2.

Causal directed acyclic graph showing that “Body Weight” at time point 2 (T2) is the effect of several factors. CVD = Cardiovascular disease. T1 = Time point #1. T2 = Time point #2. Declining health at T2 is a confounding path when assessing the effect of body weight on death.

FIGURE 2.

Causal directed acyclic graph showing that “Body Weight” at time point 2 (T2) is the effect of several factors. CVD = Cardiovascular disease. T1 = Time point #1. T2 = Time point #2. Declining health at T2 is a confounding path when assessing the effect of body weight on death.

Close modal

As seen in Figure 2, declining health is a cause of both changes in body weight and mortality, i.e., it is a confounding variable in the assessment of the effect of measured body weight on mortality. This does not mean that intentional weight loss through a healthy diet and physical activity is a cause of increased mortality. Pack et al. reviewed studies of weight loss and mortality in those with coronary artery disease. Overall, weight loss was associated with an increased risk of death (risk ratio [RR] = 1.30, 95% CI = 1.00–1.69). However, when assessing only those studies that had an intervention component to encourage “intentional weight loss,” weight loss was associated (in this case, probably causally) with a reduced risk of death (RR = 0.67, 95% CI = 0.56–0.80).

Paradox

There is substantial evidence across various disciplines that excess adiposity causes CVD. It appears clear that, among those with CVD, the people who are obese do better than those who are not obese. However, there is no paradox. According to the Cambridge English dictionary, a paradox is “a statement or situation that may be true but seems impossible or difficult to understand because it contains 2 opposite facts or characteristics.” An understanding of the rules of causal DAGs show that the 2 facts stated above are not contradictory or difficult to understand. Rather, the results are explainable through biases.

Recognizing the Biases

Studies should specify the question of interest. In the case of the studies reporting an OP, the question seems to be, what is the effect of obesity on lifespan in those with an obesity-related comorbidity? If obesity is harmful, then a goal would be to reduce the obesity. If obesity is not harmful, then the goal would be to not affect or perhaps increase adiposity. A bias is a systematic error in determining this effect. In a previous paragraph, this manuscript described the confounding bias that occurs because the dying process results in weight loss. The studies reporting an OP also fall prey to a form of selection bias termed “collider stratification bias” (3436). Others describe the problem as “index event bias” (37), since the studies select only those with “index events” such as heart failure (HF).

Figure 3 shows a structural diagram with a box around CVD signifying that CVD is known to be present. In the case of the articles claiming an OP, this knowledge is the result of the studies restricting subjects to only those with CVD. Conditioning on CVD creates an inverse association between obesity and “other causes” (38) even though, according to our structural diagram, there is no other reason for these variables to be associated. According to Figure 3, if a person with CVD is not obese, then that person must have had another cause of the CVD. The inverse association that develops due to conditioning on CVD distorts the ability to assess the effect of any of the causes of CVD, including obesity.

FIGURE 3.

A structural diagram showing the causes of mortality from cardiovascular disease (CVD). The box around CVD diagnosis signifies that the analysis is restricted to only those with CVD. The dotted line signifies that this restriction induces a negative association between the causes of CVD. The red lines signify that this is the more dangerous path to CVD. “Other things” may be causes such as infections, cancer, or genetics. CVD = cardiovascular disease.

FIGURE 3.

A structural diagram showing the causes of mortality from cardiovascular disease (CVD). The box around CVD diagnosis signifies that the analysis is restricted to only those with CVD. The dotted line signifies that this restriction induces a negative association between the causes of CVD. The red lines signify that this is the more dangerous path to CVD. “Other things” may be causes such as infections, cancer, or genetics. CVD = cardiovascular disease.

Close modal

Structural selection bias is not simply a problem of external validity. It also affects the internal validity of studies. I was a co-author on a recent study which restricted the analysis to youth who had a test for diabetes mellitus (39). Our results were shocking in that obesity appeared to be a protective factor against diabetes mellitus. However, we explained that the shocking results were invalid due to collider stratification bias. In our study (39), as well as the more typical OP studies, this error makes a harmful exposure appear protective. In other cases, it may merely bias the results toward the null hypothesis.

The false claim that obesity is paradoxically beneficial in those with obesity-related comorbidities has similarities to the assertion about 50 years ago that smoking in pregnancy was beneficial for small neonates (40). Observational studies of low birth weight babies found that those born to mothers who smoked did better than those born to mothers who did not smoke (40). Researchers struggled to try and explain the benefits of smoking during pregnancy (40). Fortunately for the advancement of public health, Hernández-Díaz et al. in 2006 explained the bias that led to this so-called birth weight paradox (41). The cause of the phenomena was not that smoking was somehow beneficial. It was due to collider stratification bias. There are multiple reasons why a baby is born with low birth weight. Smoking is one reason. Other reasons include more serious problems, such as genetic defects and infections. Smoking can appear protective when the evaluation is restricted to low birth weight babies, as babies born with other causes of low birth weight fare worse.

Were someone to repeat studies restricted to low birth weight babies, I imagine that the results would continue to show that those born to smokers do better than those born to nonsmokers. Since the publication of the study outlining the causal reasons why smoking would falsely appear beneficial when studies are restricted to low birth weight babies (41), I am unaware of further analyses with similar restriction of subjects. However, with the so-called OP, studies are appearing weekly. In 2015, there was a meta-analysis of 89 studies in 1.3 million patients with established CHD which “confirmed a strong [OP]” (3,42). In a response to a letter stating that this is not a paradox, OP proponents admitted that “confounding factors may be involved” (42). Although meta-analyses are considered the highest quality of evidence, a systematic bias does not correct itself through aggregation of similar studies.

We Can Do Better

Some may assert that the studies claiming an OP are observational, and thus we can never know what causes what. However, it is incorrect to assume that causal effects can only be determined through randomized controlled trials (RCTs) (43), lest we think that smoking does not cause lung cancer. The key is that studies must have comparison groups, and the comparison groups should be as similar as possible except for the exposure of interest. This is termed exchangeability, meaning that, if both groups were given the same exposure, they would have had the same outcomes, on average (36,44). In studies reporting that there is an OP among people with a clinical diagnosis, the exposure is obesity, and the claim is that the obese do better. Compared to whom? The comparison group (those without obesity) is not similar to the obese group without obesity. Rather, those in the comparison group must have had another cause (i.e., other than obesity) for their CVD or other comorbidity.

Although there are no RCTs that randomize nonobese people to become obese, there are studies randomizing people who have obesity to strategies encouraging weight loss. A meta-analysis of 15 RCTs found that weight loss led to a reduction in mortality (45). What about studies of people with CVD? An RCT study published in JAMA in 2016 found that weight loss for patients with HF who were also obese led to fewer symptoms and objectively improved fitness (46). However, this RCT did not have large enough numbers or long enough follow up to determine improved long-term clinical outcomes. Due to the reports on an OP in observational studies, and because those studies are larger (they are observational), longer (again, observational), and with mortality as an outcome (ibid), the authors of the JAMA RCT tempered their conclusions by stating, “Because of the reported [HF-OP]. . . before diet can be recommended for obese patients [with HF], further studies likely are needed. . .” (46). Are we really not recommending changes in diet for people with HF who are also obese? Another review asked whether the OP was “an impetus to rethink clinical practice” (47). The authors framed the question as follows: “If I have an obese HF patient, should I strongly advocate weight loss, or should I soft-pedal—perhaps even withhold—such counsel?” (47). The authors did not take a stand, but rather offered the following statement: “A clinician who counsels an obese HF patient to lose weight is not directly inducing weight loss per se; rather the clinician is leading the (compliant) patient to initiate an attempt at weight loss” (47). The statement seems to be trying to separate counseling for weight loss and the goal of actual weight loss.

In summary, it is clear that obesity, defined as excess adiposity, is a cause of CVD and that, if one studies only people with CVD, then the people with obesity will do better than those without obesity. However, this is not due to the excess adiposity causing the better outcomes. Rather, it is due to biases. The people without obesity who have the obesity-related comorbidity must have had another cause for their morbidity (selection bias), and dying people lose weight (confounding bias). It is perfectly logical and not a paradox. If the medical research and clinical communities do not recognize these biases, then as noted in Judea Pearl's recent book, The Book of Why, research will just devolve into “paradoxes galore” (48).

This journal is titled, “The Journal of Clinical Exercise Physiology” (bold and underline added to emphasize the clinical focus). Noncausal associations are useful for predictions, but clinical treatment decisions should be based on causal effects. When seeing a person who has CVD and who is also obese, are we to say to the person, “Great news. Studies show that you will live longer than those unfortunate folks who have CVD and who are not obese,” or are we to say to the person, “You have CVD and, given your obesity, weight loss may be beneficial. Do I have your permission to discuss strategies that may result in weight loss?”

I am a clinician trying to practice evidence-based medicine. Evidence-based medicine calls on us to incorporate the best evidence with expert opinion to guide treatment decisions consistent with the patient's values and preferences. The best evidence is that which is unbiased, and the unbiased evidence shows that clinicians should encourage patients to maintain a healthy diet and an active lifestyle with a goal to be the avoidance of excessive adiposity. If patients contract a condition such as HF and they also have excessive adiposity, we should continue to encourage these same behaviors. If the proponents of the OP believe that their results are due to causal reasons with obesity being beneficial, then given the millions of patients already studied, why are they not calling for weight gain as part of the treatment plan (49)?

1.
Lavie
CJ,
Sharma
A,
Alpert
MA,
De Schutter
A,
Lopez-Jimenez
F,
Milani
RV,
Ventura
HO.
Update on obesity and obesity paradox in heart failure
.
Prog Cardiovasc Dis
.
2015
;
58
:
393
400
.
doi:10.1016/j.pcad.2015.12.003.
2.
Oga
EA,
Eseyin
OR.
The obesity paradox and heart failure: a systematic review of a decade of evidence
.
J Obes
.
2016
;
2016
:
1
9
.
doi:10.1155/2016/9040248.
3.
Sharma
A,
Lavie
CJ,
Borer
JS,
Vallakati
A,
Goel
S,
Lopez-Jienez
F,
Arbab-Zadeh
A,
Mukherjee
D,
Lazar
JM.
Metaanalysis of the relation of body mass index to all-cause and cardiovascular mortality and hospitalization in patients with chronic heart failure
.
Am J Cardiol
.
2015
;
115
:
1428
34
.
doi:10.1016/j.amjcard.2015.02.024.
4.
Banack
HR,
Kaufman
JS.
Does selection bias explain the obesity paradox among individuals with cardiovascular disease?
Ann Epidemiol.
2015
;
25
:
342
9
.
doi:10.1016/j.annepidem.2015.02.008.
5.
Shrier
I.
The “obesity paradox” is not a paradox: time to focus on effective treatments
.
JACC Hear Fail
.
2016
;
4
:
234
.
doi: 10.1016/j.jchf.2015.11.002.
6.
Stovitz
SD,
Banack
HR,
Kaufman
JS.
Structural bias in studies of cardiovascular disease: let's not be fooled by the ‘obesity paradox’
.
Can J Cardiol
.
Published Online First:
2017
.
doi:10.1016/j.cjca.2017.10.025.
7.
Edwards
T.
The world's laconics: or, the best thoughts of the best authors
. 1st ed.
New York
:
M. W. Dodd
;
1853
.
8.
Banack
HR,
Stovitz
SD,
Kaufman
JS.
Letter by Banack et al. regarding article, ‘Body mass index and mortality among adults undergoing cardiac surgery: a nationwide study with a systematic review and meta-analysis’
.
Circulation
.
2017
;
136
:
507
8
.
doi:10.1161/CIRCULATIONAHA.117.027312.
9.
Stovitz
SD.
True and true, but not a paradox
.
Heart
.
2015
;
101
:
1850
.
doi:10.1136/heartjnl-2015-308477.
10.
Peeters
A.
Journals should no longer accept ‘obesity paradox’ articles
.
Int J Obes (Lond)
.
2018
;
42
:
584
5
.
doi:10.1038/ijo.2017.259.
11.
World Health Organization, Topics: Obesity
.
12.
Faerstein
E,
Winkelstein
W.
Adolphe Quetelet: statistician and more
.
Epidemiology
.
2012
;
23
:
762
3
.
doi:10.1097/EDE.0b013e318261c86f.
13.
Garrow
JS,
Webster
J.
Quetelet's index (W/H2) as a measure of fatness
.
Int J Obes
.
1985
;
9
:
147
53
.
14.
Kim
SH,
Després
J-P,
Koh
KK.
Obesity and cardiovascular disease: friend or foe?
Eur Heart J.
2016
;
37
:
3560
8
.
doi:10.1093/eurheartj/ehv509.
15.
Padwal
R,
Leslie
WD,
Lix
LM,
Majumdar
SR.
Relationship among body fat percentage, body mass index, and all-cause mortality
.
Ann Intern Med
.
2016
;
164
:
532
.
doi:10.7326/M15-1181.
16.
Banack
HR,
Kaufman
JS.
The obesity paradox: understanding the effect of obesity on mortality among individuals with cardiovascular disease
.
Prev Med (Baltim)
.
2014
;
62
:
96
102
.
doi:10.1016/j.ypmed.2014.02.003.
17.
Stokes
A,
Preston
SH.
Smoking and reverse causation create an obesity paradox in cardiovascular disease
.
Obesity (Silver Spring)
.
2015
;
23
:
2485
90
.
doi:10.1002/oby.21239.
18.
Gómez-Ambrosi
J,
Silva
C,
Galofré
JC,
Escalada
J,
Santos
S,
Gil
MJ,
Valenti
V,
Rotellar
F,
Ramirez
B,
Salvador
J,
Fruhbeck
G.
Body adiposity and type 2 diabetes: increased risk with a high body fat percentage even having a normal BMI
.
Obesity
.
2011
;
19
:
1439
44
.
doi:10.1038/oby.2011.36.
19.
Gómez-Ambrosi
J,
Silva
C,
Galofré
JC,
Escalada
J,
Santos
S,
Millan
D,
Vila
N,
Ibanez
P,
Gil
MJ,
Valenti
V,
Rotellar
F,
Ramirez
B,
Salvador
J,
Fruhbeck
G.
Body mass index classification misses subjects with increased cardiometabolic risk factors related to elevated adiposity
.
Int J Obes
.
2012
;
36
:
286
94
.
doi:10.1038/ijo.2011.100.
20.
Hernán
MA.
The C-word: scientific euphemisms do not improve causal inference from observational data
.
Am J Public Health
.
2018
;
108
:
616
9
.
doi:10.2105/AJPH.2018.304337.
21.
Khan
SS,
Ning
H,
Wilkins
JT,
Allen
N,
Carnethon
M,
Berry
JD,
Sweis
RN,
Lloyd-Jones
DM.
Association of body mass index with lifetime risk of cardiovascular disease and compression of morbidity
.
JAMA Cardiol
.
2018
;
3
:
280
.
doi:10.1001/jamacardio.2018.0022.
22.
MacMahon
B,
Pugh
TF.
Causes and entities of disease
.
In:
Clark
DW,
MacMahon
B,
eds.
Preventive medicine
.
Boston, MA
:
Little Brown
;
1967
.
p.
12
.
23.
Heymsfield
SB,
Wadden
TA.
Mechanisms, pathophysiology, and management of obesity
.
N Engl J Med
.
2017
;
376
:
254
66
.
doi:10.1056/NEJMra1514009.
24.
Ahima
RS.
Adipose tissue as an endocrine organ
.
Obes (Silver Spring)
.
2006
;
14
Suppl 5
:
242S
249S
.
doi:10.1038/oby.2006.317.
25.
Trayhurn
P,
Bing
C,
Wood
IS.
Adipose tissue and adipokines—energy regulation from the human perspective
.
J Nutr
.
2006
;
136
:
1935S
39S
.
26.
Grams
J,
Garvey
WT.
Weight loss and the prevention and treatment of type 2 diabetes using lifestyle therapy, pharmacotherapy, and bariatric surgery: mechanisms of action
.
Curr Obes Rep
.
2015
;
4
:
287
302
.
doi:10.1007/s13679-015-0155-x.
27.
Heneghan
HM,
Meron-Eldar
S,
Brethauer
SA,
Schauer
PR,
Young
JB.
Effect of bariatric surgery on cardiovascular risk profile
.
Am J Cardiol
.
2011
;
108
:
1499
507
.
doi:10.1016/j.amjcard.2011.06.076.
28.
Porta
M,
Vineis
P,
Bolúmar
F.
The current deconstruction of paradoxes: one sign of the ongoing methodological “revolution.”
Eur J Epidemiol
.
2015
;
30
:
1079
87
.
29.
Dansinger
ML,
Tatsioni
A,
Wong
JB,
Chung
M,
Balk
EM.
Meta-analysis: the effect of dietary counseling for weight loss
.
Ann Intern Med
.
2007
;
147
:
41
50
.
30.
Siahpush
M,
Tibbits
M,
Shaikh
RA,
Singh
GK,
Sikora Kessler
A,
Huang
TT.
Dieting increases the likelihood of subsequent obesity and BMI gain: results from a prospective study of an Australian national sample
.
Int J Behav Med
.
2015
;
22
:
662
71
.
doi:10.1007/s12529-015-9463-5.
31.
Katz
AM,
Katz
PB.
Diseases of the heart in the works of Hippocrates
.
Br Heart J
.
1962
;
24
:
257
64
.
32.
Doehner
W,
Anker
SD.
Cardiac cachexia in early literature: a review of research prior to Medline
.
Int J Cardiol
.
2002
;
85
:
7
14
.
33.
Alley
DE,
Metter
EJ,
Griswold
ME,
Harris
TB,
Simonsick
EM,
Longo
DL,
Ferrucci
L.
Changes in weight at the end of life: characterizing weight loss by time to death in a cohort study of older men
.
Am J Epidemiol
.
2010
;
172
:
558
65
.
doi:10.1093/aje/kwq168.
34.
Greenland
S.
Quantifying biases in causal models: classical confounding vs collider-stratification bias
.
Epidemiology
.
2003
;
14
:
300
6
.
35.
Hernán
MA,
Hernández-Díaz
S,
Robins
JM.
A structural approach to selection bias
.
Epidemiology
.
2004
;
15
:
615
25
.
doi:10.1097/01.ede.0000135174.63482.43.
36.
Hernán MA
RJ.
Causal inference
.
Boca Raton
:
Chapman & Hall/CRC
;
2018
.
37.
Dahabreh
IJ,
Kent
DM.
Index event bias as an explanation for the paradoxes of recurrence risk research
.
JAMA
.
2011
;
305
:
822
3
.
doi:10.1001/jama.2011.163.
38.
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
20
.
doi:10.1093/ije/dyp334.
39.
Stovitz
SD,
Banack
HR,
Kaufman
JS.
Paediatric obesity appears to lower the risk of diabetes if selection bias is ignored
.
J Epidemiol Community Health
.
2018
;
72
:
302
8
.
doi:10.1136/jech-2017-209985.
40.
Yerushalmy
J.
The relationship of parents' cigarette smoking to outcome of pregnancy—implications as the problem of inferring causation from observed associations
.
Am J Epidemiol
.
1971
;
93
:
443
56
.
doi:10.1093/oxfordjournals.aje.a121278.
41.
Hernández-Díaz
S,
Schisterman
EF,
Hernán
MA.
The birth weight ‘paradox’ uncovered?
Am J Epidemiol.
2006
;
164
:
1115
20
.
doi:10.1093/aje/kwj275.
42.
Lavie
CJ,
Ventura
HO.
Reply
.
JACC Hear Fail
.
2016
;
4
:
234
5
.
doi:10.1016/j.jchf.2015.11.006.
43.
Pearl
J.
Causal inference in statistics: an overview
.
Stat Surv
.
2009
;
3
:
96
146
.
doi:10.1214/09-SS057.
44.
Greenland
S,
Robins
JM,
Pearl
J.
Confounding and collapsibility in causal inference
.
Stat Sci
.
1999
;
14
:
29
46
.
doi:10.1214/ss/1009211805.
45.
Kritchevsky
SB,
Beavers
KM,
Miller
ME,
Shea
MK,
Houston
DK,
Kitzman
DW,
Nicklas
BJ.
Intentional weight loss and all-cause mortality: a meta-analysis of randomized clinical trials
.
PLoS One
.
2015
;
10
:
e0121993. doi:10.1371/journal.pone.0121993.
46.
Kitzman
DW,
Brubaker
P,
Morgan
T,
Haykowsky
M,
Hundley
G,
Kraus
WE,
Eggebeen
J,
Nicklas
BJ.
Effect of caloric restriction or aerobic exercise training on peak oxygen consumption and quality of life in obese older patients with heart failure with preserved ejection fraction
.
JAMA
.
2016
;
315
:
36
.
doi:10.1001/jama.2015.17346.
47.
Charnigo
R,
Guglin
M.
Obesity paradox in heart failure: statistical artifact, or impetus to rethink clinical practice?
Heart Fail Rev.
2017
;
22
:
13
23
.
doi:10.1007/s10741-016-9577-0.
48.
Pearl
J,
MacKenzie
D.
The book of why: the new science of cause and effect
.
New York
:
Hachette Book Group
;
2018
.
49.
Lajous
M,
Banack
HR,
Kaufman
JS,
Hernán
MA.
Should patients with chronic disease be told to gain weight? The obesity paradox and selection bias
.
Am J Med
.
2015
;
128
:
334
6
.

Conflicts of Interest and Source of Funding: None.

Author notes

1University of Minnesota, Department of Family Medicine and Community Health, Minneapolis, MN 55455 USA