Dynamics of obesity paradox after stroke, related

to time from onset, age, and causes

of death
ABSTRACT
Objective: Paradoxical longevity in obese patients with established disease has
been documented in various conditions. We aimed to find whether such a
relationship exists in ischemic stroke patients, with stratified analyses according to
time of death after stroke, age, and cause of death.

Methods: The Korean Stroke Registry (KSR) is a nationwide, multicenter,

prospective registry of acute stroke. For 7.5 years, data on 34,132 patients with
acute ischemic stroke were collected through KSR, and their mortality information
was ascertained through a governmental statistical office. We assessed relative
hazard of mortality according to obesity status.

Results: Stroke survivors whose body mass index (BMI) values were lower than
the chosen reference level of 20–23 had increased risks of long-term mortality
(hazard ratio [HR] of 1.36 and 95% confidence interval [CI] of 1.25–1.48 for BMI
≤18.5; HR of 1.14 and 95% CI of 1.03–1.26 for BMI 18.5–20), whereas obese
stroke patients had decreased risks of mortality (HR of 0.83 and 95% CI of 0.74–
0.92 for BMI 27.5–30; HR of 0.77 and 95% CI of 0.63–0.93 for BMI 30–32.5).
Inverse association between obesity status and mortality was not evident until 90
days after stroke but became significant 1 year after onset of stroke. Such an
association was more prominent in stroke patients who were less than 65 years old,
but it remained constant in all age groups. The paradoxical relationship remained
significant, regardless of causes of death.

Conclusions: Our results documented obesity paradox in stroke survivors,

regardless of age and causes of death, and it became evident a sufficient time after
stroke onset. Neurology® 2012;79:856– 863
For neurologists and stroke physicians who are involved in the care of stroke
patients, prognostic indicators regarding recurrence and mortality are of the highest
interest because of the high mortality and morbidity associated with stroke. Obesity
is one of the well-known modifiable risk factors for vascular diseases, which
warrants stringent control and prevention for primary and secondary prevention of
stroke. Contrary to common knowledge of the deleterious effects of obesity, it has
been reported that obese stroke survivors are likely to have lower mortality than
their underweight counterparts. Recent post hoc analysis in a large randomized trial
also documented a reverse association between obesity and recurrence of stroke.
Such a paradoxical phenomenon of lower mortality or risk of recurrent vascular
disease in obese patients with established disease was coined the obesity paradox.
This paradox has repeatedly been reported to occur in various conditions, including
heart failure, coronary heart disease, hypertension, peripheral arterial disease,
diabetes, chronic kidney disease, percutaneous coronary intervention, and acute
kidney injury.
However, in spite of accumulating reports on populations with various diseases,
there are still doubts about the obesity paradox. Recent objections have included
lead time bias and index event bias. In this context, more stringent analyses to
exclude confounding effects of age and cause of death in a population of sufficiently
large size are warranted in investigations of obesity paradox. Therefore, the authors
analyzed 34,132 ischemic stroke patients, recruited from 30 stroke centers across
South Korea for 7.5 years, to test the hypothesis of paradoxical longevity in stroke
survivors; analyses were stratified according to age groups and causes of death.

METHODS Study population. The Korean Stroke Registry (KSR) is a

prospective nationwide stroke registry in which major academic and tertiary stroke
centers participate (figure e1A on the Neurology® Web site at
www.neurology.org). Between January 2002 and June 2009, 30 stroke centers with
nationwide coverage registered 43,723 acute stroke patients consecutively. From
this population, as described in figure e1B, we excluded 9,951 patients (21.9%) by
the pre-established selection criteria but included in the analysis 34,132 patients
with acute ischemic stroke who were admitted to the participating stroke centers
within 7 days after onset.

Standard protocol approvals, registrations, and patient consents. Participants

in this study gave their written informed consents voluntarily, and the study
protocol was approved by the institutional review boards of Seoul National
University Hospital (H-0911–065-301) and Hallym University Sacred Heart
Hospital (2008 –I008).

Data collection. KSR participating centers collected clinical, laboratory, imaging,

and treatment information on acute stroke patients and registered it in the
centralized Web-based database (www.strokedb.or.kr). Participating centers were
required to use a standardized case registration form to collect a predefined set of
data. Treating physicians or research nurses entered patients’ information into the
KSR database, and the consecutiveness and fidelity of data were confirmed by
experienced vascular neurologists in each stroke center.

Recorded information included the following: sex, age, height, and weight;
classification of stroke according to the Trial of Org 10 172 in Acute Stroke
Treatment (TOAST); history of stroke, hypertension, diabetes, cigarette smoking
(former or current smoker vs lifetime nonsmoker), reperfusion treatment for
hyperacute stroke (both IV and intra-arterial), and decompressive surgery; NIH
Stroke Scale (NIHSS) score at admission; modified Rankin Scale (mRS) score at
the time of discharge; systolic blood pressure; white blood cell count; and
hemoglobin, fasting blood glucose, hemoglobin A1c (HbA1c), and total cholesterol
values. Body mass index (BMI) was calculated as weight (in kilograms) divided by
height (in meters) squared.

Mortality data were acquired from the death certificates of Statistics Korea, a
governmental statistical office (current as of December 2009), with use of unique
personal identity numbers as retrieval keys, as described previously. Date and cause
of death were also ascertained, and causes of death were categorized as cancer
(ICD-10 code C00-D48), vascular (I00-I999; deaths due to cardiac diseases or
stroke), or other.

Statistical analysis. The distributions of demographic, clinical, laboratory, stroke,

and treatment data were analyzed against obesity status or mortality status with use
of the X2 test, t test, or one-way analysis of variance (ANOVA), as appropriate. The
trend of baseline data according to the obesity status was also calculated with the
X2 test for trends in proportion or one-way ANOVA with a contrast to linear trend,
as appropriate. The associations between BMI or obesity status and the risk of
mortality over the follow-up period were estimated by using Cox proportional
hazards regression models. Relative hazards of mortality were additionally assessed
at 30-day, 90-day, and 1-year time points after onset of stroke. Obesity status as a
categorical variable was established by dividing BMI into 8 levels, according to the
research criteria of obesity proposed by WHO expert consultation (<18.5, 18.5–
19.9, 20.0 –22.9, 23.0 –24.9, 25.0 –27.4, 27.5–29.9, 30.0 –32.4, and ≥32.5). A BMI
range of 20.0 –22.9 was used as a reference level because of the large number of
participants.

Because the KSR data were collected over a long period and from multiple stroke
centers, some information was missing from the dataset. Multiple imputation
methods were used to estimate the missing data on the basis of observed
information and to account for potential variance due to uncertainty. Missing data
were assumed to be missing at random, and 5 separate imputed datasets were
generated with IVEware version 0.1. The multiple sets of point estimates and
standard errors were combined to obtain final estimates by PROC MIANALYZE.
A sensitivity analysis was conducted between the imputed dataset and complete-
case dataset, and the results were comparable (table e1). The continuous measure
of BMI was used to fit a restricted cubic spline curve with 5 knots to obtain a smooth
representation of relative risk of mortality for various time points, with adjustment
for the effects of confounders.

Statistical analyses were performed by using SAS 9.2 (SAS Institute, Cary, NC)
and R version 2.10.1. Statistical significance was defined as a 2-tailed p value of
<0.05. Values are presented as frequency (percentage), mean ± SD, or median
(interquartile range), as appropriate.

RESULTS From a total of 43,723 patients in the KSR database, the authors
analyzed 34,132 patients with acute ischemic stroke who were admitted within 7
days after onset of stroke and were available to ascertain mortality data from a
governmental statistics office. Distributions of baseline characteristics of the
analyzed patients are presented in table 1. On the basis of the WHO obesity criteria
for the Asian Pacific population, 1,348 patients (4.7%) were classified as
underweight at the time of admission, 10,888 (37.7%) as normal weight, 7,666
(26.5%) as overweight, 8,056 (27.9%) as obesity I, and 924 (3.2%) as obesity II
(BMI was missing for 5,250 patients). The mean ± SD value for BMI in the study
population was 23.7 ± 3.2 kg/m2. During the follow-up period (mean, 32.6 ± 23.1
months), 9,073 patients (26.6%) were deceased as of December 2009. Mortality
rates were 4.1% (1,410 subjects) at 1 month, 7.0% (2,394) at 3 months, and 13.0%
(4,373) at 1 year after stroke. Among the deceased, 5,013 patients (55.2%) died of
vascular causes, 942 (10.4%) died of malignancy, and 3,131 (34.5%) died of other
causes.

The distributions of baseline demographic variables, stroke characteristics, and

vascular risk factors were skewed across obesity status (table e2). Underweight
stroke survivors were relatively older, had frequent cardioembolic stroke, and had
strokes of greater severity (NIHSS and mRS scores). On the contrary, obese stroke
patients tended to be younger, had frequent small-vessel occlusions, and had
prevalent hypertension and diabetes. The proportion of deaths up to December 2009
significantly decreased as obesity status worsened, and the long-term risk of
mortality also differed in relation to obesity status (log-rank test, p < 0.01). In the
Cox proportional hazards model, BMI was inversely associated with long-term
mortality (adjusted HR of 0.96 for 1-kg/m2 increase; 95% CI, 0.95– 0.97). Because
the association between BMI and mortality appeared to be nonlinear (p < 0.001),
the authors stratified BMI into a categorical variable with 8 strata and evaluated
restricted cubic spline curves. At 30 days and 90 days after ischemic stroke,
adjusted HRs of obesity status for mortality were not significant. NIHSS score
(adjusted HR, 1.035 for 1-point increase; 95% CI, 1.026 –1.043) and mRS score
(adjusted HR, 3.70 for 1-point increase; 95% CI, 3.33– 4.12) remained significant
for mortality at 30 days after stroke. However, when compared to the reference
range of 20 to 23 at 1 year after stroke and long term, stroke survivors in the lower
BMI group had a higher risk of mortality and stroke patients in the higher BMI
group had a lower risk of mortality (table 2 and figure 1). When BMI was treated
as a continuous variable, a BMI value of 27.6 ± 0.2 was associated with the lowest
relative hazard for long-term mortality.

To investigate potential modification of effects by age on the relative risk of

mortality, the authors performed stratified analysis by age group. The interaction
term of age and BMI was significant in the all-patients dataset ( p < 0.001).
Although confidence intervals became wider because of smaller sample sizes, the
parameter estimates of lower obesity strata were more elevated in younger age
groups (i.e., ages <55 years and 55– 64 years) than older age groups (figure 2).
Likewise, stroke patients in age group <55 years had a relatively lower risk for
longterm mortality in the higher obesity strata. The association between obesity
status and mortality in extremely older patients (≥85 years) was slightly less than
in younger age groups (table e3). BMI as a continuous variable remained significant
in multivariable analyses in all age groups.
Restricted cubic spline curves depicting associations between mortality at various
time points after stroke onset and body mass index as a continuous variable. Dashed
lines show 95% confidence intervals. Mortality for long term after stroke was
calculated with vital status data current as of December 2009, with average follow-
up duration of 32.6 ± 23.1 months.
As with the findings for death from any cause, inverse associations between obesity
status and mortality risk were preserved in the cause-specific hazard analyses (table
3). For patients with a vascular or other cause of death, such a relationship between
BMI and cause-specific mortality was significant. In spite of wider CIs, the global
trends of point estimates for relative risk of mortality due to cancer showed an
inverse relationship with increasing obesity. Additionally, the inverse association
between obesity status and risk for mortality was also documented after
consideration of recruiting center and
year of onset.

DISCUSSION In this study, the authors analyzed a nationally representative

prospective multicenter registry of patients with acute ischemic stroke and
documented an inverse association between obesity status and long-term mortality.
In comparison with a BMI reference range of 20 –23 kg/m2, the risk of longterm
mortality was higher for ischemic stroke survivors with a lower BMI: 36% higher
with a BMI ≤18.5 kg/m2 and 14% higher with a BMI of 18.5–20 kg/m2. Likewise,
those with a higher BMI had decreased risk: 23% lower with a BMI of 30 – 32.5
kg/m2 and 17% lower with a BMI of 27.5–30 kg/m2. A BMI value of 27.6 ± 0.2
was found to be associated with the lowest relative hazard of mortality. In
particular, such an inverse relationship was not evident until 90 days after stroke,
but it became significant beyond 1 year after onset of ischemic stroke. Such an
association was also documented among all the age-stratified groups and in the
cause-specific hazard analysis of mortality.

After the initial report on chronic heart failure, various theories have been suggested
to explain the biological mechanisms of this paradoxical phenomenon, including a
role for cytokines, an attenuated sympathetic nervous system, increased serum
lipoproteins with detoxification of bacterial toxins, modulation of systemic
response through adipocytokines, and preserved muscle protein metabolism in
obese patients. Such theories were limited by their accounting only partially for the
lifespan of elderly patients with a heavy disease burden. A series of clinical
investigations also yielded the proposed explanation that obesity in the aged
population may indicate an increased metabolic reservoir to overcome a higher
energy expenditure in catastrophic events or chronically debilitating conditions
after such events. However, such a proposal was not able to completely exclude
residual confounding from age or cause of death. Occult diseases at the time of
index stroke, such as a cancer, may underlie obesity paradox. Additionally, index
event bias, a common bias of recurrent risk analysis studies, was indicated to oppose
obesity paradox. To address these concerns, we stratified our population on the
basis of age and cause of death, and the results held true regardless of strata.
Because our index event was ischemic stroke, index event bias could be ruled out
from our analysis of mortality risk from cancer or other causes. Detailed post hoc
analyses were made possible by the design and organization of KSR, involving 30
participating stroke centers with nationwide coverage that collected information on
34,132 consecutive patients with acute ischemic stroke over 7.5 years.

An inverse association between obesity status and long-term mortality of stroke

survivors is parallel to previous reports involving ischemic stroke patients in Athens
or intracerebral hemorrhage (ICH). A major difference between our study and the
Greek study was the effect of obesity on mortality status in the early period after
stroke onset. In the Greek population, obesity paradox was found as early as the
first week after stroke and became more evident thereafter. However, in the KSR
population, an inverse association was not documented at 30 days after stroke onset
and was barely noted at 90 days after stroke. Such nonsignificance of obesity
paradox in the early period was comparable to the findings of a study involving
ICH, which showed that 30-day mortality after ICH was more influenced by stroke
severity than by obesity status. Because the authors hypothesized that internal
metabolic reservoir capacity would be a key mechanism for obesity paradox among
stroke survivors, the obesity paradox would not be evident until after a sufficient
duration of debilitation, and the severity of the index stroke would be a more
influential factor than obesity, as documented in the current study. Epidemiologic
differences between our study and the Greek study in terms of obesity burden or
study characteristics may also account for such differences. Additionally, although
a recently published analysis of 1,592 ischemic stroke cases revealed the occurrence
of obesity paradox in only the underweight population, our large nationwide dataset
provided sufficient power for detailed investigation of obesity paradox in stroke
survivors.
Although previous studies tried to adjust for the effect modification of age by
multivariable analyses, it would be difficult to completely exclude residual
confounding from age. Towfighi and Ovbiagele even indicated that there were
significant interactions between mortality and age, in that obesity correlated with
mortality among younger patients but was inversely associated with mortality
among older patients. Also, statistical modeling documented a robust pattern in the
association between obesity and mortality risk, similar to that of obesity paradox.
In our study, a stratified analysis based on age group showed that both BMI as a
continuous variable and obesity status as a categorical variable remained significant
in all age strata. We also detected a more significant association in younger patients
than in older stroke survivors, which may be explained by the fact that older stroke
survivors may be prone to other hazardous medical or environmental conditions,
and therefore the role of obesity paradox on long-term mortality may weaken more
for them than for younger stroke patients.

Reverse causation has been an important refutation against the results in this study.
We investigated associations between obesity and specific causes of mortality in
stroke survivors. Although a limited sample size prevented conclusive
interpretation, the trends in cause-specific analyses would be sufficient to assume
that obesity paradox was documented, regardless of causes of mortality.

A few points require further clarification. First, BMI was the only available
measurement of obesity in KSR. Although BMI is not an optimal measure of body
fat distribution, the use of BMI to predict overall mortality was recently advocated.
Second, temporal measurements of BMI in stroke survivors were not available.
Third, because the KSR population was collected largely from tertiary academic or
regional referral hospitals, the mortality rate may be slightly lower than previously
reported. Fourth, data on pharmacologic management beyond the acute period were
not collected in the registry. Fifth, stratified analyses with causes of death should
be interpreted with caution, because this information was retrieved from a
governmental archive of death certificates. Sixth, the small number of ischemic
stroke survivors at the extreme obesity level limited our ability to draw conclusions
about any association patterns in this range. Seventh, despite statistical adjustment
of baseline characteristics, discrepancy of age and stroke mechanisms between
obesity groups should be considered. Eighth, our study is a subanalysis of an
original ABBA study, and the results should be interpreted as such. Finally,
functional status after stroke was not measured beyond the time of discharge.
This analysis based on a nationwide prospective registry of patients with acute
ischemic stroke showed that obesity status was inversely associated with longterm
risk of mortality. Although the occurrence of obesity paradox in various conditions
has been documented repeatedly, no detailed analyses or studies of the underlying
mechanisms of the phenomenon have been published. In this context, it is
noteworthy that obesity paradox became evident a sufficient time after stroke onset
and that it was significant regardless of age group or causes of death. A recently
published cohort study involving 1 million Asians revealed a U-shaped relationship
between obesity and mortality risk, suggesting that a BMI interval of 22.6 –27.5
denoted the group with lowest mortality risk; these values fall within the obesity
range for Asians. In our study, we showed that a BMI value of 27.6 was associated
with the lowest relative risk of long-term mortality. In both studies, the BMI
associated with the lowest mortality risk fell into the “obesity” category for Asia–
Pacific residents. These findings may suggest that our weightmanagement
strategies and concept of optimal weight in secondary prevention may require
further consideration and individualization.

On the basis of our study findings, we are not going to insist that obesity provides
long-term protection against mortality from ischemic stroke. Such a radical opinion
would not be relevant unless it had been proven in randomized trial settings and its
biological mechanisms had been explained. However, our results do suggest that
we need further studies to learn more about the definition of obesity, how it should
be measured, what it means, and its relationship to primary and secondary stroke
prevention.