Introduction
Since the discovery in December 2019, the novel Coronavirus Disease 2019 (COVID-19)
has taken the world by storm with the declaration of a global health emergency on
30th January 2020 by the World Health Organization (WHO) [1]. At the latest count among 213 countries in early July 2020, 11 600 000 infections
and more than 537 000 deaths have been reported worldwide with many countries still
struggling to control the spread of this contagion [2].
The clinical course of COVID-19 ranges from asymptomatic or mild infection to severe,
life-threatening pneumonia with multi-organ failure. With no definitive curative treatment
found as yet, public health systems have sought to risk stratify and protect the more
vulnerable groups in the population. Several early studies have identified some host-related
risk factors, which may predispose to severe illness, such as older age and underlying
comorbidities such as chronic respiratory illness, cardiovascular disease (CVD), diabetes
mellitus and hypertension [3]
[4]
[5]
[6]. This has led to the international health authorities promoting shielding strategies
of vulnerable groups of individuals above 60–70 years, the immunocompromised, or those
with chronic comorbidities [7]
[8]
[9]
[10]. Although not studied during the initial period of COVID-19, increased body-mass
index (BMI) has been increasingly reported to carry an independent risk for severe
COVID-19 infection [11]
[12]
[13]
[14].
Studies before the COVID-19 pandemic have been conflicting in terms of the association
of obesity with poorer outcomes for acute respiratory illnesses. For severe outcomes
of influenza pneumonia including hospitalisation, ICU admissions and death, results
have been mixed regarding obesity as a risk factor [15]
[16]
[17]
[18]
[19]. However, a systematic review and meta-analysis of 234 studies and 610 782 participants
found that obesity is an independent significant risk factor for severe outcomes in
both the seasonal influenza as well as the pandemic H1N1 influenza [20].
With respect to the COVID-19 infection, we seek to elucidate whether obesity confers
a poorer prognosis as observed in previous pandemic pneumonias. Hence, we undertook
a systematic review and meta-analysis to assess the impact of obesity on unfavourable
outcomes of COVID-19 disease in hospitalised adult patients, as compared to non-obese
patients.
Materials and Methods
This systematic review was performed in accordance with the Cochrane Collaboration
guidelines [21]. The study followed the guidelines from the Preferred Reporting Items for Systematic
Reviews and Meta-Analyses (PRISMA) and the Meta-analyses of Observational Studies
in Epidemiology (MOOSE) [22]
[23].
Literature search
Three authors (CWSH, HK, and JHXL) undertook independent comprehensive search for
published articles from the MEDLINE, EMBASE, Web of Science, and Cochrane databases.
The articles were searched from 1st December 2019 until 28th June 2020. The search
terms included ‘COVID-19, ‘SARS-CoV2’, ‘coronavirus disease 2019’, ‘novel coronavirus’,
‘outcomes’, ‘obesity’, ‘body mass index’, and ‘BMI’. The titles and abstracts of studies
were screened, and only studies which appeared to match the pre-determined inclusion
and exclusion criteria were extracted. The full texts of the extracted studies were
read to determine relevance to the current study. To supplement the electronic searches,
we also examined the reference list of included studies.
Selection of studies
Two authors (KH and JHXL) selected the studies, and differences were resolved by discussion.
The inclusion criteria were: (1) studies which identified adult patients diagnosed
with COVID-19 based on a real-time reverse transcription PCR method who were hospitalised
in an acute hospital, (2) studies which clearly defined the obese and non-obese populations
based on BMI, (3) studies which appropriately defined obesity as BMI>30 or>28, in
non-Asians and Asians, respectively, (4) studies which used pre-specified criteria
for outcomes of severe COVID-19 illness or mortality, (5) cohort studies which may
be prospective or retrospective in nature, and (6) studies with an appropriate control
group allowing estimation of odds ratio of unfavourable COVID-19 outcomes between
the obese and non-obese groups. If there was suggestion of multiple publications from
the same or overlapping group of patients (e. g., studies arising from the same hospital
database), we decided to include the data only from the most comprehensive study.
We used the following exclusion criteria: (1) studies which were not in English, (2)
studies which focused only on a specific cohort of patients (e. g., only patients
in ICU, only patients with diabetes, only geriatric patients), (3) studies which included<10
patients, (4) studies whose population are not an inpatient cohort, or (5) studies
which focused on a paediatric or pregnant population.
Extraction of data
The data from the included studies were independently extracted and collated on a
standardised form by three authors (CWSH, HK, and JHXL). CWSH examined the data for
any error. The following data were collated from each study: the first author, the
country and city, type of study, time period, location, definitions of obese and non-obese
categories, criteria for defining severe COVID-19 illness, clinical and demographic
characteristics, and outcomes of patients. Any reported results of univariate or multivariate
analysis of outcomes in obese versus non-obese group, as well as the confounders adjusted
for, were also collected.
Assessment of quality of the included studies
Two authors (JHXL and IH) were blinded to the study results and independently assessed
the quality of studies using the Newcastle-Ottawa scale [24]. Differences were resolved through discussion. This well-established scale assesses
each study across three categories: selection (4 items, maximum 4 stars), comparability
(1 item, maximum 2 stars) and exposure (3 items, maximum 3 stars). Each study receives
a total score ranging between 0 and 9. Scores of≥7, 5–6, and≤4 translate into ‘high’,
‘medium’, and ‘low’ quality scores respectively.
Aim of study
The primary outcome of interest was the pooled odds ratio (OR) in obese compared to
non-obese hospitalised patients with COVID-19, for the composite outcome of an unfavourable
clinical outcome, defined by mortality and, or severe disease, inclusive of:
-
Requiring intensive (ICU) or high-dependency care, or
-
Requiring mechanical ventilation,
-
Categorised as suffering severe pneumonia or acute respiratory distress syndrome (ARDS)
as defined by either the WHO, American Thoracic Society (ATS), or National Health
Commission of the People’s Republic of China criteria [25]
[26]
[27], or
-
Clinician-defined severe disease.
The secondary outcomes of interest were mortality, and severe disease. Among studies
which reported outcomes for normal BMI (defined as BMI<25 or<23.5 in non-Asian and
Asian studies, respectively), overweight and obese categories, we examined the effect
of increasing BMI categories on developing an unfavourable outcome, as well as the
OR of severe obesity, defined by BMI>35 or>40 on an unfavourable outcome.
Subgroup analyses were done to identify study characteristics that significantly influenced
the pooled OR estimate of unfavourable outcome and contributed to heterogeneity between
studies. Our pre-determined subgroup analyses of study characteristics compared unfavourable
outcomes in studies by:
-
Study region (Asian vs. non-Asian),
-
Quality of study (high, medium or low),
-
Type of study (prospective vs retrospective),
-
Multi-centre versus single-centre,
-
Sample size of obese population (above vs. below 100 participants), and
-
Prevalence*of possible confounding patient demographics and comorbidities including:
-
Population mean age above versus below 60 years,
-
Prevalence of diabetes mellitus above versus below 30%,
-
Prevalence of hypertension above versus below 60%,
-
Prevalence of CVD above versus below 20%,
-
Prevalence of chronic kidney disease (CKD) above versus below 15%, and
-
Prevalence of chronic pulmonary disease above versus below 20%
*These cut-offs were chosen based on the median of the prevalence reported in all
the studies.
Statistical analysis
For each study, the data for the unfavourable COVID-19 outcomes was reported as the
Odds Ratio. Given the expected heterogeneity in the effect sizes, we decided to perform
the meta-analysis using a random-effects model [28] based upon the method described by DerSimonian and Laird [29]. The heterogeneity among the studies was assessed using two methods [30]: (a) Cochran’s Q statistic which takes into account the overall variance of effect
sizes, with subsequent assessment of statistical significance of such heterogeneity
(as the tests for heterogeneity have low power, a p-value of<0.10 was taken as suggestive
of significant heterogeneity); (b) the inconsistency index (I2) which depicts the proportion of true heterogeneity among studies from the overall
heterogeneity (conventionally, the I2 values of<30%, 30–59%, 60–75% and>75% are considered to represent low, moderate,
substantial and considerable heterogeneity, respectively). Using the Remove-One index,
sensitivity analysis was performed to examine any disproportionate effect of a particular
study on the overall estimate of the pooled estimate. Publication bias was ascertained
quantitatively using Egger’s regression test and qualitatively using funnel plots
[31]
[32]. If publication bias was found, impact of publication bias was assessed using Duval
and Tweedie’s “Trim and Fill” method while the ‘fail-safe N’ test was utilised to
estimate the number of unpublished studies with non-significant results that would
be required to make the publication bias non-significant [33]
[34]. All analyses were performed using Comprehensive Meta-analysis (CMA) software, version
3 (Biostat Inc., Englewood, NJ, USA).
Results
Results of literature search
Initial search yielded 999 articles, and 997 articles remained after removal of duplicates.
A total of 819 articles were removed after screening the title and abstract for relevance.
A sum of 178 full-text articles were selected and evaluated in detail for potential
inclusion, with a further 158 excluded for reasons specified in [Fig. 1]. The remaining 20 studies were used in the qualitative and quantitative meta-analysis.
Fig. 1 PRISMA flow chart of study selection.
Characteristics of included studies
This meta-analysis included 20 studies consisting of a total of 28 355 patients who
were hospitalised with COVID-19 illness. The period of hospitalisation was between
January and May 2020 ([Table 1]). All included studies were observational cohort studies, of which 15 were retrospective
and 5 were prospective studies. Most studies (n=11) were from the United States, while
5 were from Europe, 3 from Asia (China), and 1 from South America (Mexico). Majority
of studies (n=12) were based on single centre while 8 were multicentre. Besides two
studies which defined obesity as BMI>28 kg/m2, all other studies used the criteria established by the WHO with respect to BMI.
Quality of studies
As assessed with the Newcastle Ottawa scale, overall quality of studies was moderate
to high. Six studies were of high quality, twelve were of moderate quality, while
only two were of low quality. Majority of the studies scored low in the ‘comparability’
domain of the scale (Supplementary Table 1S).
Population characteristics
All studies were conducted in adults. The patient characteristics in each study are
summarised in Supplementary Table 2S. The mean age of participants was 66.4 (SD 19.5) years, with a predominance of male
patients (59.5%). A total of 6361 (22.4%) of patients were obese, defined as a BMI
above 30 (or above 28 in Asians). Overall, 8097 (28.4%) patients had an unfavourable
outcome. 30.6% of obese patients had an unfavourable outcome, compared with 28.0%
of non-obese patients.
Risk of unfavourable outcome in obese patients
As the primary outcome of this study, OR of unfavourable outcome was estimated for
obese patients as compared to non-obese patients who were hospitalised with COVID-19
illness. The unadjusted pooled OR 1.25 (n=28 355 patients in 9 studies, 95% CI 1.07–1.45,
p=0.005; I2=65.8%) for an unfavourable outcome ([Fig. 2]). When adjusted for potential confounders, the pooled OR was 2.02 (n=17 861 patients
in 6 studies, 95% CI 1.41–2.89, p<0.001; I2=73.5%). For both ORs, there were significant and substantial heterogeneity among
the studies (discussed later).
Fig. 2 Forrest plot showing a significantly increased pooled odds ratio for obese compared
to non-obese patients for an unfavourable outcome, including studies reporting severity
and mortality. a Unadjusted odds ratio. b Adjusted odds ratio for age, sex and major co-morbidities.
Risk of mortality in obese patients
In the four studies (n=17 322 patients), the adjusted pooled OR for mortality was
significantly higher in the obese cohort (OR 1.51, 95% CI 1.13–2.21, p=0.006, I2=46.2%) (Supplementary Fig. 1S). The heterogeneity among studies (Cochran’s Q 5.48, p=0.134) and publication bias
(Egger’s test, p=0.362) were non-significant for this outcome.
Risk of severe COVID-19 illness in obese patients
Four studies consisting of 1547 patients reported severity outcomes with adjustment
for age, sex and other comorbidities. The pooled adjusted OR for risk of severe disease
in obese versus non-obese patients was 2.26 (95% CI 1.47–3.48, p<0.001; I2=67.2%) (Supplementary Fig. 1S). Publication bias was non-significant on Egger’s test (p=0.053).
Graded analysis between increasing BMI groups and risk of unfavourable outcome, compared
to a normal BMI
Nine studies of 3255 patients reported unfavourable outcomes in patients with normal
BMI, overweight and obese groups. Normal weight was defined as BMI<25 kg/m2, overweight was defined as BMI 25–29.9 kg/m2 and obese was defined as BMI>30 kg/m2. One study [11] defined normal weight as BMI 18.5–23.9 kg/m2, overweight as BMI 24–27.9 kg/m2 and obese as BMI>28 kg/m2. The pooled unadjusted OR comparing overweight and obese groups to normal weight
group was 1.81 (95% CI 1.10–2.98, p=0.019; I2=75%), and 2.11 (95% CI 1.17–3.82, p=0.014; I2=78.5%), respectively ([Fig. 3]).
Fig. 3 Pooled unadjusted OR of 9 studies reporting unfavourable outcome in overweight and
obese COVID-19 patients, as compared to normal weight patients. N=total number of
participants.
In patients with severe obesity (class II obesity, BMI>35) compared to a normal BMI,
the pooled unadjusted OR from 3 studies for an unfavourable outcome was significantly
increased at 2.54 (95% CI 1.20–5.37, p=0.015; I2=0%). The adjusted OR for an unfavourable outcome was similarly increased in 4 studies
of severely and morbidly obese patients, as compared to patients with a normal BMI
(Supplementary Table 3S).
Subgroup analysis
As mentioned earlier, significant and substantial heterogeneity were noted for the
primary outcomes of interest (ORs of unfavourable outcomes in obese hospitalised patients
with COVID-19 illness). For the unadjusted OR of unfavourable outcomes, pre-specified
subgroup analysis was performed to explore study-related cofactors which could have
influenced the outcomes (Supplementary Table 4S).
Reduction in the heterogeneity was noted when studies were separately analysed according
to whether they had large or small numbers of obese patients, with smaller studies
(<100 obese patients) reporting a significantly greater risk of unfavourable outcome
with obesity and larger studies (>100 obese patients) reporting no significant difference in unfavourable outcomes between
obese and non-obese participants. There was no difference observed between subgroups
when studies were separated into location, multi-centre versus single centre studies,
prospective versus retrospective studies, or study quality. Among patient characteristics,
a lower prevalence of CVD<20% and CKD<15% in studies were associated with a higher
pooled OR for unfavourable outcome in obese versus non-obese patients.
Sensitivity analysis and publication bias for primary outcomes
Using Remove-one index for sensitivity analysis, the pooled ORs were re-evaluated
after excluding one study at a time. However, no significant change in the heterogeneity
or pooled OR was noted, suggesting that our eligibility criteria for inclusion or
exclusion of studies did not have an inherent deficiency. Based on the visual inspection
of the funnel plot and Egger’s test, significant publication bias was found (Supplementary
Fig. 2S). For the primary outcome of unfavourable unadjusted OR, the publication bias was
re-assessed using the ‘Trim and Fill’ method of Duvel and Tweedie [33]. Despite trimming five of 20 studies (on the left of the mean), publication bias
still remained significant.
Discussion
To the best of our knowledge, this is the first systematic review and meta-analysis
studying an association between obesity and unfavourable outcomes in patients who
are hospitalised with COVID-19 illness. In this meta-analysis of more than 28 000
adult COVID-19 infected patients, the risk of an unfavourable outcome (of developing
severe disease or dying from COVID-19) was estimated to be approximately twice as
compared to the non-obese patients even after adjustment for other potential confounders.
Also, obese patients also had 50% greater odds of mortality compared to non-obese
patients. The association of an unfavourable outcome were seen even in patients with
BMI in the overweight category, as opposed to a normal BMI, and this was further increased
in the severely obese group. This strengthens the relationship between increasing
BMI and COVID-19 outcomes observed.
These findings are consistent with the increased risk observed in obese patients for
developing critical disease in acute respiratory and non-respiratory illnesses [20]
[50]
[51]. Increased adiposity has been proposed to exacerbate a proinflammatory phenotype
by increasing proinflammatory adipokines and reducing anti-inflammatory adipokines
[52], resulting in a chronic inflammatory process that involves tonic activation of the
innate immune system [53]. Obesity also results in mechanical compression of the thorax, including the chest
walls, lungs, and diaphragm [54], and has been shown to have detrimental effects to respiratory function in all age
groups [55].
As a known risk factor for diabetes and cardiovascular disease which are independently
associated with morbidity and mortality [56]
[57], obesity may also be associated with poorer overall health outcomes. However, studies
have not been consistent. The obesity paradox has been described in which obesity
seems to be a protective factor in patients with established CV disease and advanced
CKD [58]
[59]. This appears to be consistent with our subgroup analysis showing a lower OR of
obesity for an unfavourable outcome in populations with a higher prevalence of CV
disease and CKD.
There are several mentionable strengths in our study: systematic strategy for search
of the literature with well-defined criteria for inclusion and exclusion; appropriate
exclusion of redundant and non-informative studies, meticulous extraction of data
(both explicit as well as deduced from the studies), careful evaluation for quality
of studies, and appropriate quantitative statistical assessment. Based on the large
sample size in the included studies from different geographical locations of the globe,
our estimate appears to have enough power (i. e., low risk for type II error) for
a comfortable acceptance. Moreover, our estimates were carefully calculated with and
without adjustment for potential confounders, thereby providing a better understanding
of the impact of obesity on the unfavourable outcomes in patients with COVID-19 illness.
We also acknowledge several limitations in our study. First, majority of the studies
were retrospective in nature which might have introduced intrinsic biases (e. g.,
recall bias, reporting bias, etc.). Secondly, our population for the systematic review
was only hospitalised patients with COVID-19 illness. As the criteria for hospitalisation
may differ between hospitals, the population of included studies might not have been
uniform. To explore it further, we carefully performed a subgroup analysis for a variety
of study and patient related factors. The only partial explanation for heterogeneity
was that there was a significant difference in estimates in studies with larger and
smaller samples. Moreover, the sensitivity analysis did not alter the estimate of
pooled OR thereby providing weight to the robustness of our eligibility criteria.
We believe that most of the heterogeneity could have arisen from having different
types of hospitalisations (such as varying criteria for hospitalisation, escalation
of treatment or definition of severity).
Conclusion
Based on a large sample, our systematic review demonstrates that obesity significantly
and independently increases the odds of an unfavourable outcome in hospitalised patients
with COVID-19 illness. In addition, obesity is associated with increased mortality
and severe course of illness. Risk stratification and a review of resource allocation
for the obese patient with COVID-19 is pivotal to providing optimal care in this population.
Given the growing epidemic of COVID-19 and the high prevalence of obesity globally,
there is a pressing need for population strategies aimed to prevent and manage obesity,
and it remains to be seen if these can be helpful in reducing overall morbidity and
mortality in COVID-19.
Table 1 Characteristics of included studies.
|
Author, Year, [Ref]
|
Country, City
|
Study design
|
Setting
|
Number of participants
|
Time of study
|
Outcome studied
|
Participants with endpoint (%)
|
Quality
|
|
Docherty AB et al. 2020 [35]
|
UK
|
Prospective cohort
|
Multicentre, hospitalised patients
|
16 081
|
6 Feb–10 Apr 2020
|
Mortality
|
4185 (26)
|
High
|
|
Giacomelli A et al. 2020 [36]
|
Italy, Milan
|
Prospective cohort
|
Single centre, hospitalised patients
|
233
|
21 Feb–19 Mar 2020
|
Mortality
|
48 (21)
|
Moderate
|
|
Petrilli CM et al. 2020 [37]
|
US, New York and Long Island
|
Prospective cohort
|
Single centre, hospitalised patients
|
2661
|
1 Mar–8 Apr 2020
|
Composite=ICU, IMV, hospice, or death
|
964 (36)
|
Moderate
|
|
Klang E et al. 2020 [38]
|
US, New York
|
Retrospective cohort
|
Multicentre, hospitalised patients
|
3406
|
1 Mar–17 May 2020
|
Mortality
|
1136 (33)
|
High
|
|
Buckner FS et al. 2020 [39]
|
US Seattle
|
Retrospective cohort
|
Multicentre, hospitalised patients
|
93
|
2 Mar–8 May 2020
|
Composite=ICU or death
|
43 (46)
|
Low
|
|
Hajifathalian K et al. 2020 [14]
|
US, New York
|
Retrospective cohort
|
Two centres, hospitalised patients
|
770
|
4 Mar–16 Apr 2020
|
Severe=ICU
Mortality
Composite=ICU or death
|
196 (25)
88 (11)
241 (31)
|
Moderate
|
|
Hur K et al. 2020 [40]
|
US, Chicago
|
Retrospective cohort
|
Multicentre, hospitalised patients
|
486
|
1 Mar–18 Apr 2020
|
Severe=IMV
|
138 (28)
|
Moderate
|
|
Hu L et al. 2020 [41]
|
China, Wuhan
|
Retrospective cohort
|
Single centre, hospitalised patients
|
294
|
8 Jan–10 Mar 2020
|
Severe=defined by WHO
|
164 (56)
|
Low
|
|
Dreher M et al. 2020 [42]
|
Germany, Heinsberg
|
Retrospective cohort
|
Single centre, hospitalised patients
|
50
|
1 Mar–31 Apr 2020
|
Severe=ARDS
|
24 (48)
|
Moderate
|
|
Kalligeros M et al. 2020 [43]
|
US, Rhode Island
|
Retrospective cohort
|
Muticentre, hospitalised patients
|
103
|
17 Feb–5 Apr 2020
|
Severe=ICU
|
44 (43)
|
High
|
|
Cai Q et al. 2020 [11]
|
China, Shenzhen
|
Prospective cohort
|
Single centre, hospitalised patients
|
383
|
11 Jan–16 Feb 2020
|
Severe=Any of 1) RR>30, 2) Resting SpO2<93%, 3) PaO2<FiO2<300mmHg
|
91 (24)
|
High
|
|
Huang R et al. 2020 [3]
|
China, Jiangsu province
|
Retrospective cohort
|
Multicentre, hospitalised patients
|
202
|
22 Jan–10 Feb 2020
|
Severe=Any of 1) RR>30, 2) Resting SpO2<93%, 3) PaO2<FiO2<300mmHg
|
18 (8.9)
|
Moderate
|
|
Busetto L et al. 2020 [12]
|
Italy, Veneto
|
Retrospective cohort
|
Single centre, hospitalised patients
|
92
|
23 Mar–11 Apr 2020
|
Severe=ICU+semi-intensive respiratory unit
Mortality
|
35 (38)
12 (13)
|
High
|
|
Lighter J et al. 2020 [13]
|
US, New York
|
Retrospective cohort
|
Single centre, hospitalised patients
|
1762
|
4 Mar–4 Apr 2020
|
Severe=ICU
|
431 (24)
|
Moderate
|
|
Argenziano MG et al. 2020 [44]
|
US, New York
|
Retrospective cohort
|
Single centre, hospitalised patients
|
781
|
1 Mar–5 Apr 2020
|
Severe=ICU
|
234 (30)
|
Moderate
|
|
Mani VR et al. 2020 [45]
|
US, New York
|
Retrospective cohort
|
Single centre, hospitalised patients
|
171
|
Mar–Apr 2020
|
Severe=ICU
Mortality
Composite=ICU or death
|
29 (17)
31 (18)
48 (28)
|
Moderate
|
|
Moriconi D et al. 2020 [46]
|
Italy, Pisa
|
Retrospective cohort
|
Single centre, hospitalised patients
|
100
|
16 Mar–15 Apr 2020
|
Mortality
|
18 (18)
|
Moderate
|
|
Ortiz-Brizuela E et al. 2020 [47]
|
Mexico, Mexico City
|
Prospective cohort
|
Single centre, hospitalised patients
|
140
|
26 Feb–11 Apr 2020
|
Severe=ICU
|
29 (21)
|
Moderate
|
|
Pettit N et al. 2020 [48]
|
US, Chicago
|
Retrospective cohort
|
Single centre, hospitalised patients
|
238
|
1 Mar–18 Apr 2020
|
Severe=hypoxaemia
Mortality
|
140 (59)
24 (10)
|
High
|
|
Suleyman G et al. 2020 [49]
|
US, Michigan
|
Retrospective cohort
|
Multicentre, hospitalised patients
|
355
|
9 Mar–27 Mar 2020
|
Severe=ICU
|
141 (40)
|
Moderate
|
ICU: Intensive care unit; IMV: Invasive mechanical ventilation; RR: Respiratory rate;
UK: United Kingdom; US: United States.
