The COVID-19 pandemic has impacted global morbidity and mortality without precedent in the past decades. To determine patients’ prognoses, many risk factors have been considered. In previous pandemics, such as influenza A(H1N1)pdm,1–3 an elevated body mass index (BMI) has been associated with worse outcomes, including a higher need for invasive mechanical ventilation (IMV) and/or longer ICU stay.

In patients with COVID-19, the association between obesity and worse clinical outcomes is still unclear.4–13 It has been previously recognised that adipokines and cytokines represent a marker of chronic low-grade inflammation characteristic of abdominal obesity, which may impair immune response and potentially have further effects on the lung of patients affected with SARS-CoV-2 infection.14,15 Despite the rational evidence, multiple studies and meta-analyses have been published and could not determine that obesity is associated with worse outcomes in patients with COVID-19 infection.16,17

Previous studies18 have reported that many patients with COVID-19 infection had an elevated BMI. Despite these elevated rates, most studies suggest that elevated BMI does not represent an independent risk factor for mortality. Moreover, the association between high BMI and bad outcomes has yet to be further explored in critically ill patients affected by COVID-19 infection.4–11 Interestingly, a recently published meta-analysis13 found that obesity was associated with an increased risk of death in patients with COVID-19. This study was one of the very few analysing BMI, and the results should be interpreted cautiously due to a high risk of bias and the certainty that their evidence needed to be better classified.

In this study, we aimed to determine the association of high BMI with patients’ outcomes considering the lack of adequate and definitive data, especially when considering critically ill patients. We hypothesise that patients with obesity are not at increased risk of death when admitted to the ICU. Therefore, the study aims to evaluate the impact of obesity on crude ICU mortality in critically ill patients admitted by COVID-19.

In the first step, a multilevel conditional logistic modelling and the intraclass correlation coefficient (ICC) were calculated (Table S1, Supplementary material) with patients nested in hospital to characterise hospital-level variation of ICU mortality and determine if a significant inter-hospital variation is present.

In the second step, we describe the differences between patients with and without obesity according to BMI categorisation. Subsequently, we present the characteristics of the patients according to their evolution at ICU discharge. All data, except those related to evolution, are shown concerning ICU admission values.

In the third step, the relationship between BMI and laboratory variables has been studied by Pearson correlation.

In the fourth step, to study the impact of obesity, BMI categories and BMI as a continuous variable on mortality, we used multiple logistic regression, including all variables with significance in the univariate mortality analysis. Because the subgroup of underweight patients is very small relative to the other subgroups, this causes the dataset to be class unbalanced. In these cases, it is difficult to create adequate test and training data sets, as most classifiers are built assuming that the test data come from the same distribution as the training data. Considering that unbalanced data will produce undesirable results, such as much lower performance in the test data than in the training data, we have decided to exclude this subgroup from the multivariate mortality analysis.

For all model internal validation, the database was randomly split into two subsets: (a) a “training set” (70%) and (b) a “validation set” (30%). Models’ performance was examined using accuracy test, sensitivity(S), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV) and area under the ROC curve (AUC). Multicollinearity was checked by calculating the variance inflation ratio (VIF).

In the fifth step, Smooth Restricted cubic splines (RCS) for Cox hazard regression were used to model possible non-linear relationships between BMI and outcome. Spline functions estimate exposure effects for specific intervals of the exposure variable of interest (BMI) through polynomial functions joined together at five knots. The estimated associations between the exposure and the outcome were plotted as a graphical depiction.

Finally, sensitivity analyses stratified by age, sex, diabetes, and hypertension were performed using adjusted logistic regression analysis. In the age-stratified analysis, we used a cut-off point of 63 years because 63 was the cut-off point with the lowest HR in the smooth RCS plot (Figure S1, Supplemental material). Results are shown as OR and 95% CI.

In a post hoc analysis, we evaluated the impact of class III obesity within the subgroup of obese patients to confirm whether class III obesity remains a condition associated with increased mortality.

Data analysis was performed using R software (cran.r-project.org).

Of the 1909 obese patients, 237 (12.4%) met the criteria for Class III obesity. Patient characteristics can be seen in Table S4 of the Supplementary material. The groups were very similar except for age (54 vs. 62 years, p < 0.001) and the frequency of male sex (54.9% vs. 68.2%, p < 0.001), which was lower in class III patients. ICU stay was more extended (16 vs. 15 days, p < 0.001) in class III patients, without differences in crude mortality at ICU discharge (30.4% vs. 28.7%, p = 0.64) but achieving significance in-hospital mortality (7.8% vs. 4.1%, p = 0.04).

The characteristics of the subgroup of obese patients according to their evolution in the ICU are shown in Table S5 of the Supplementary material. Significant variables (age, male, APACHE II, SOFA, serum lactate, C-RP, PCT, serum creatinine, WBC count, LDH, D dimer, arterial hypertension, Haematologic disease, chronic renal failure, COPD, chronic heart failure, chronic coronary artery disease, diabetes, invasive mechanical ventilation on ICU admission, PaO2/FiO2 at ICU admission, AKI, myocardial dysfunction, presence of shock on ICU, bacterial coinfection ventilator-associated pneumonia and obesity/class III were included in the mortality regression model. Class III obesity was maintained as a variable independently associated with mortality in ICU (Figure S6).

We found that overall obesity was not associated with worse clinical outcomes manifested by mortality, need for invasive mechanical ventilation at ICU admission, or higher incidence of VAP compared to non-obese patients. Interestingly, critically ill patients with COVID-19 patients characterised with Class III obesity were independently associated with a higher ICU mortality. Of note, patients with obesity spent more days under invasive mechanical ventilation and consequently had a more extended ICU stay than non-obesity patients.

The value of this manuscript is that previous studies were performed on non-critically ill patients with a limited sample size. Our results coincide with those reported by Petrelli CM et al.4 but disagree with those observed by other authors.5–11 Petrilli CM et al.4 observed in 990 critically ill patients in New York State that only a BMI > 40 was independently associated with mortality (HR 1.71, 95% CI 1.10–2.7). In contrast, Peñuelas O et al.,11 in a multicenter study that included 868 patients with COVID-19 who required invasive mechanical ventilation, did not observe that obesity was associated with increased mortality. However, the study only considered obesity as a dichotomous variable, and this analysis coincides with our results in evaluating obesity as a dichotomous variable and highlights the need to study obesity by differentiating those patients with Class III obesity criteria. Other studies present results that are difficult to interpret. Docherty AB et al.5 reported in more than 20,000 hospitalised patients with SARS-CoV-2 infection in the United Kingdom that obesity was a strong predictor of mortality (HR 1.33, 95% confidence interval [CI] 1.19–1.49) after adjusting for other comorbidities. However, only 17% of them were ICU patients and the impact of obesity in this subgroup of critically ill patients is not reported. Bello-Chavolla OY et al.6 reported that obese patients, compared to non-obese patients, had a higher mortality rate (13.5% vs. 9.4%). However, this study only included 3% of critically ill patients, so the results should be interpreted with great caution.

Contrary to expectations, we did not observe a greater need for mechanical ventilation in obese patients compared to non-obese patients on admission to the ICU. Neither did we observe a more significant application of non-invasive ventilation despite a more remarkable alteration in oxygenation on admission to the ICU. However, obese patients were ventilated at a higher rate than non-obese patients during their ICU stay. We hypothesise that this more significant need for MVI may be related to abdominal fat deposition, increasing the respiratory workload to lift the chest wall. In addition, obesity may increase intra-abdominal pressure and restrict the downward movement of the diaphragm.23 All these alterations may lead to a more significant and earlier ventilatory failure during COVID-19 infection, which justifies our findings.

Contrary to this observation, Bello-Chavolla et al.,6 in a cohort of over 50,000 patients hospitalised for SARS-CoV-2, reported a higher need for ventilatory support in obese patients than non-obese patients (5.2% vs. 3.3%). However, the study only included 3% of ICU patients (n = 1893) and did not analyse these patients specifically, so the results do not apply to the ICU population. Furthermore, data from France7 evidenced a higher need for mechanical ventilation (OR = 7.36 [1.63–33.14] in critically ill patients with COVID-19 with BMI > 35 vs. those with BMI < 25. Finally, a recent meta-analysis24 reported that obesity is a significant risk factor for invasive mechanical ventilation (IMV) in patients with COVID-19. This association with obesity was only observed when body mass index (BMI) was considered a categorical variable. In contrast, when exposure-effect risk was calculated using the robust error meta-regression method, BMI was associated with IMV in a nonlinear fashion, and these data are in the direction of our results.8 Differences in population characteristics, risk analysis methodology, and the need for mechanical ventilation may explain the differences between studies.

Obesity has been associated with the secretion and expression of proinflammatory cytokines.14,15 Although we did not observe a higher level of inflammation or a significant correlation between controlled inflammatory biomarkers (CRP, PCT, D-dimer) and BMI, it is noteworthy that the presence of AKI and the need for CRRT was higher as BMI was higher than 30 kg/m2, which could be related to a higher level of inflammation that cannot be assessed with the usual biomarkers.

Finally, our study has shown discrepancies in the results depending on how BMI is analysed. It seems to be associated with mortality if BMI is analysed as a continuous variable (OR = 1.02). At the same time, this impact does not appear if it is analysed as a cut-off point to define obesity (BMI > 30) but reappears for Class III obesity when studying four cut-off points. According to the first analysis, mortality increases by 2% for each point increase in BMI. However, this relationship should be interpreted with caution since it is possible that the impact of the increase in BMI is not similar if it is still within the overweight range or if this increase conditions the fulfilment of the criteria for passing from obesity to Class III obesity. We consider that this relationship between BMI and mortality risk is best reflected in Fig. 3 (smooth HR), where it is observed that only high BMI values increase the risk of mortality.

Our study has important strengths. It is a multicentre study with a large number of critically ill patients and a robust statistical analysis, which confers homogeneity, internal validity and power to its results. However, it also has significant limitations that should be acknowledged. The first limitation is that it is a study on critically ill patients attended almost exclusively in Spain. Therefore, these findings can only be extrapolated to other types of patients or other healthcare systems with prior validation. Second, although our study confirmed the positive association between COVID-19 severity and class III obesity, we do not have data on the type of fat in our patients. Since visceral adiposity has been reported to be more predictive of the risk of death than subcutaneous fat, our results should be interpreted cautiously in the context of an observational study during the pandemic due to classifying obesity only by BMI, which may overlook the role of the distribution of fat and sarcopenia, which may increase the risk of death.

Third, due to the characteristics of the registry, no data on evolution at hospital discharge have been recorded. Although this data is essential to measure hidden mortality after hospital discharge, the in-hospital mortality of 7% suggests a high post-hospital survival rate. On the other hand, our focus has been on short-term mortality, most closely related to acute illness.

Fourth, the characteristics of pulmonary mechanics in ventilated patients may vary concerning the presence or absence of obesity. Due to the characteristics of the database developed during the pandemic, no data on aspects of pulmonary mechanics that could be related to patient outcomes were recorded, so the results of the present study should be assessed with this limitation in mind.

Finally, we have not found differences between markers of obesity-associated systemic inflammation. Considering that we did not have data on more specific biomarkers of inflammation, such as IL-1 or TNF, our findings should be interpreted with caution, as the inflammatory state secondary to obesity may be underestimated. However, these biomarkers are only sometimes used in routine practice.

Our results corroborate international recommendations, which strongly recommend that patients with a BMI of 40 kg/m2 or higher are at high risk of poor COVID-19 results and should, therefore, strongly consider the need for an effective vaccination plan and prolonged social distancing.

AR, IML, GM, ED, MB, CF, MS, JMC, AE, JSV, RZ, LS, MBS, and JG had substantial contributions to conception and design of the work.

ST, AR, JCPL, SSC, JCB, MVO, ACB and AAM had substantial contribution for data acquisition.

AR, IML, MB, ED, RZ, MS, JMC and JSV had substantial contribution for data analysis and interpretation of data for the work.

AR, IML, ED, GM, and MB drafting of the manuscript.

MIR, JJG, RZ, LFR, MBS and JG critically reviewed the draft manuscript.

The corresponding author (AR) had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors approved the final version of the manuscript. The views expressed in this article are those of the authors and not necessarily those of the SEMICYUC.

Declarations section

- Ethical Approval and Consent to participate

The study was approved by Institution’s Internal Review Board (Comitè Ètic d’Investigació amb Medicaments [CEIm] from Institut d’Investigació Sanitària Pere Virgili [IISPV] - IRB# CEIM/066/2020) and each participating site with a waiver of informed consent.

- Availability of supporting data

The data supporting the conclusions of this study are available from the Sociedad de Medicina Intensiva, Critica y Unidades Coronarias (SEMICYUC), but restrictions apply to the availability of these data, which were used under SEMICYUC authorization for the present study and are therefore not publicly available. However, the data can be obtained from the corresponding author (AR) upon reasonable request and with the permission of SEMICYUC.

This study was supported by the Spanish Intensive Care Society (SEMICYUC) and Ricardo Barri Casanovas Foundation (AR, MB). The study sponsors have no role in the study design, data collection, data analysis, data interpretation, or writing of the report.