Many coronavirus disease 2019 (COVID-19) risk factors, including obesity and diabetes, are associated with an abnormal basal metabolic rate (BMR). We aimed to evaluate whether BMR could impact the susceptibility to or severity of COVID-19. We performed genetic correlation and Mendelian randomization (MR) analyses to assess genetic correlations and potential causal associations between BMR (n = 448,348) and three COVID-19 outcomes: severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, COVID-19 hospitalization, and critical COVID-19 (n = 1,086,211–2,597,856). A multivariable MR (MVMR) analysis was used to estimate the direct effect of BMR on COVID-19 independent of BMI and type 2 diabetes. BMR has positive genetic correlations with the COVID-19 outcomes (genetic correlations 0.213–0.266). The MR analyses indicated that genetic liability to BMR confers causal effects on SARS-CoV-2 infection (odds ratio 1.14, 95% CI 1.09–1.20, P = 1.65E−07), hospitalized COVID-19 (1.31, 1.18–1.46, P = 8.69E−07), and critical COVID-19 (1.04, 1.19–1.64, P = 4.89E−05). Sensitivity analysis of MR showed no evidence of directional pleiotropy or heterogeneity, indicating the robustness of its results. The MVMR analysis showed that the causal effects of BMR on hospitalized COVID-19 and critical COVID-19 were dependent on BMI and type 2 diabetes but that BMR may affect the SARS-CoV-2 infection risk independently of BMI and type 2 diabetes (odds ratio 1.09, 95% CI 1.03–1.15, P = 4.82E−03). Our study indicates that a higher BMR contributes to amplifying the susceptibility to and severity of COVID-19. The causal effect of BMR on the severity of COVID-19 may be mediated by BMI and type 2 diabetes.
Introduction
Since the inception of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic, various risk factors have been reported for coronavirus disease 2019 (COVID-19), including obesity, diabetes, and smoking (1–5). Meanwhile, COVID-19 can cause a myriad of post–COVID-19 consequences, including diabetes and metabolic abnormalities (6–10). Both obesity and uncontrolled diabetes are characterized by an increased basal metabolic rate (BMR) (11), which is defined as the energy required for maintaining body functions at rest and represents ∼60–75% of the total daily energy expenditure (12).
In COVID-19, preexisting metabolic dysfunction aggravates the course of the disease (3,13). To date, it is not known whether BMR could impact the susceptibility to or severity of COVID-19. Exploring the link between BMR and COVID-19 may help improve the management of coronavirus infection. The Mendelian randomization (MR) framework can be used to infer a potential causative association between a phenotype (exposure) that can be genetically influenced and a disease outcome by using genetic variants as instrumental variables. We sought to test the potential causal effects of BMR on COVID-19.
Research Design and Methods
Genome-Wide Association Study Summary Data Sets
The study is based on publicly available genome-wide association study (GWAS) summary results. The summary statistics for the outcomes of COVID-19 were obtained from the COVID-19 Host Genetics Initiative GWAS meta-analysis round 7, including SARS-CoV-2 infection (122,616 case subjects and 2,475,240 control subjects), hospitalized COVID-19 (32,519 case subjects and 2,062,805 control subjects), and critical COVID-19 (13,769 case subjects and 1,072,442 control subjects) (14). The SARS-CoV-2 infection data set mainly reflects the overall susceptibility to the virus, whereas the hospitalized and critical COVID-19 data sets represent the severity of the disease. Therefore, we collectively called the latter two outcomes “severe COVID-19.” The BMR GWAS data set included 448,348 participants from the UK Biobank (15), which was obtained from YangLab (https://yanglab.westlake.edu.cn/) (16). All of the participants of the data sets were of European origin. The effects (values and directions) of single-nucleotide polymorphisms (SNPs) were harmonized between the GWAS summary data sets.
Genetic Correlation Analysis
The genetic correlations between the BMR and the COVID-19 outcomes were calculated using linkage disequilibrium (LD) score regression. LD score regression distinguishes confounding from polygenicity in GWAS. The 1000 Genomes Project phase 3 was used to estimate the LD structure for European populations. SNPs were filtered by 1.1 million variants, a subset of 1000 Genomes and HapMap3, with minor allele frequency >0.05. Significant genetic correlations were determined by the false discovery rate of <0.05.
MR Analysis
The main analyses were performed using the inverse-variance weighted (IVW) method and complemented with the weighted median and MR-Egger methods implemented in TwoSampleMR (17,18). The intercept from the MR-Egger regression was used to evaluate the average directional pleiotropy (19). The heterogeneity in the MR analysis was evaluated by both the Cochran Q test and I2 statistics (both P < 0.05 and I2 > 0.25) (20). The significant associations between BMR and COVID-19 were determined by IVW-based P < 0.017 (0.05/3). SNPs with genome-wide significance (P < 5 × 10−8) in the BMR data set were selected as IVs and further pruned using a clumping r2 cutoff of 0.01 within a 10-Mb window. For each MR analysis, we removed SNPs not present in the outcome data set and palindromic SNPs with intermediate allele frequencies. We harmonized each pair of the exposure and outcome data sets by aligning the effect allele for exposure and outcome and obtained variant effects and SEs of each data set. The statistical power of the MR analysis was estimated using mRnd (21).
Hypertension, coronary heart disease, obesity, and diabetes are known to be prominent risk factors for adverse COVID-19 outcomes (1). We extracted SNPs associated with hypertension, coronary artery disease (including coronary heart disease), BMI, and diabetes from the GWAS catalog (https://www.ebi.ac.uk/gwas/). To exclude the potential influence of these confounding factors, the MR analysis was rerun after the IVs that were associated with any of these conditions were excluded.
A multivariable MR (MVMR) analysis was used to estimate the direct effect of BMR on the COVID-19 outcomes independent of BMI and type 2 diabetes. The BMI data set contained 694,649 participants (22). The type 2 diabetes data set involved 74,124 subjects with type 2 diabetes and 824,006 control subjects (23). MVMR jointly evaluates the causal effects of all exposures on the outcome, which allows for estimating the direct effect of each exposure. We extracted IVs for each of the three exposures and combined them into a set of all instruments. The MVMR analysis was performed using the mv_multiple function from the TwoSampleMR package (17).
Data and Resource Availability
Summary-level data of COVID-19 GWAS are available at the COVID-19 Host Genetics Initiative website (https://www.covid19hg.org/results/r7/). Data of BMI and type 2 diabetes are available at the GWAS catalog (https://www.ebi.ac.uk/gwas/). Data of BMR are available at YangLab (https://yanglab.westlake.edu.cn/).
Results
Genetic Correlation Analysis
Genetic correlation analyses showed significant positive genetic correlations of BMR with SARS-CoV-2 infection (rg = 0.266 ± 0.034, P = 5.58E−15), hospitalized COVID-19 (rg = 0.222 ± 0.036, P = 6.33E−10), and critical COVID-19 (rg = 0.213 ± 0.043, P = 5.63E−07).
MR Analysis
Since each COVID-19 data set had different sets of SNPs and we removed SNPs not present in the outcome data set, different numbers of IVs were obtained for each MR analysis. In the MR analysis of the causal effects of BMR on the COVID-19 outcomes, a total of 580, 580, and 584 IVs were derived from the MR analysis of the causal effects of BMR on SARS-CoV-2 infection, hospitalized COVID-19, and critical COVID-19, respectively. Our MR analysis showed that genetic liability to BMR confers causal effects on SARS-CoV-2 infection (odds ratio [OR] 1.14, 95% CI 1.09–1.20, P = 1.65E−07), hospitalized COVID-19 (1.31, 1.18–1.46, P = 8.69E−07), and critical COVID-19 (1.40, 1.19–1.64, P = 4.89E−05) (Table 1 and Fig. 1A). The sensitivity analysis revealed that the directions of causal effect estimates across the methods were all the same (OR >1) (Fig. 1B). Notably, tests of MR-Egger regression did not support the directional pleiotropy of the genetic instrumental variables for the MR analysis (MR-Egger intercept < 0.01, P > 0.05). Although the output of the Cochran Q test supported the existence of heterogeneity in the MR estimates (P < 0.05), the I2 statistics indicated that the heterogeneity was small (I2 < 0.25). Therefore, no substantial heterogeneity in the MR analysis was detected. The proportion of variance explained by IVs was 0.023 for each MR analysis. The power analysis showed that our study had 100% power for each MR analysis (Type I error rate of 0.05).
Causal associations between BMR and the COVID-19 outcomes. WM, weighted median. A: Causal effects of BMR on COVID-19 outcomes. The trait on the x-axis denotes exposure, the trait on the y-axis denotes outcome, and each cross point represents an instrumental variant. The lines denote the effect sizes (B) of an exposure on an outcome. B: Effect sizes of the causal associations between BMR and the COVID-19 outcomes. C: Effect sizes of the causal associations between BMR and the COVID-19 outcomes after excluding IVs associated with four confounding factors.
Causal associations between BMR and the COVID-19 outcomes. WM, weighted median. A: Causal effects of BMR on COVID-19 outcomes. The trait on the x-axis denotes exposure, the trait on the y-axis denotes outcome, and each cross point represents an instrumental variant. The lines denote the effect sizes (B) of an exposure on an outcome. B: Effect sizes of the causal associations between BMR and the COVID-19 outcomes. C: Effect sizes of the causal associations between BMR and the COVID-19 outcomes after excluding IVs associated with four confounding factors.
Causal effects of BMR on COVID-19 outcomes
Outcome . | Method . | β (SE) . | OR (95% CI) . | Instrumental variables (n) . | P . | Q_P . | I2 . | MR-Egger intercept . | MR-Egger intercept P . |
---|---|---|---|---|---|---|---|---|---|
SARS-CoV-2 infection | IVW | 0.133 (0.025) | 1.14 (1.09–1.20) | 580 | 1.65E−07 | 3.17E−05 | 0.202 | NA | NA |
SARS-CoV-2 infection | WM | 0.115 (0.034) | 1.12 (1.05–1.20) | 580 | 7.56E−04 | NA | NA | NA | NA |
SARS-CoV-2 infection | MR-Egger | 0.033 (0.082) | 1.03 (0.88–1.21) | 580 | 0.692 | 3.51E−05 | 0.199 | 0.001 | 0.200 |
Hospitalized COVID-19 | IVW | 0.272 (0.055) | 1.31 (1.18–1.46) | 584 | 8.69E−07 | 6.29E−06 | 0.217 | NA | NA |
Hospitalized COVID-19 | WM | 0.206 (0.074) | 1.23 (1.06–1.42) | 584 | 5.42E−03 | NA | NA | NA | NA |
Hospitalized COVID-19 | MR-Egger | 0.049 (0.180) | 1.05 (0.74–1.49) | 584 | 0.787 | 7.08E−06 | 0.214 | 0.003 | 0.193 |
Critical COVID-19 | IVW | 0.334 (0.082) | 1.40 (1.19–1.64) | 584 | 4.89E−05 | 6.86E−05 | 0.193 | NA | NA |
Critical COVID-19 | WM | 0.298 (0.117) | 1.35 (1.07–1.69) | 584 | 0.011 | NA | NA | NA | NA |
Critical COVID-19 | MR-Egger | 0.262 (0.269) | 1.30 (0.77–2.20) | 584 | 0.331 | 6.18E−05 | 0.193 | 0.001 | 0.779 |
Outcome . | Method . | β (SE) . | OR (95% CI) . | Instrumental variables (n) . | P . | Q_P . | I2 . | MR-Egger intercept . | MR-Egger intercept P . |
---|---|---|---|---|---|---|---|---|---|
SARS-CoV-2 infection | IVW | 0.133 (0.025) | 1.14 (1.09–1.20) | 580 | 1.65E−07 | 3.17E−05 | 0.202 | NA | NA |
SARS-CoV-2 infection | WM | 0.115 (0.034) | 1.12 (1.05–1.20) | 580 | 7.56E−04 | NA | NA | NA | NA |
SARS-CoV-2 infection | MR-Egger | 0.033 (0.082) | 1.03 (0.88–1.21) | 580 | 0.692 | 3.51E−05 | 0.199 | 0.001 | 0.200 |
Hospitalized COVID-19 | IVW | 0.272 (0.055) | 1.31 (1.18–1.46) | 584 | 8.69E−07 | 6.29E−06 | 0.217 | NA | NA |
Hospitalized COVID-19 | WM | 0.206 (0.074) | 1.23 (1.06–1.42) | 584 | 5.42E−03 | NA | NA | NA | NA |
Hospitalized COVID-19 | MR-Egger | 0.049 (0.180) | 1.05 (0.74–1.49) | 584 | 0.787 | 7.08E−06 | 0.214 | 0.003 | 0.193 |
Critical COVID-19 | IVW | 0.334 (0.082) | 1.40 (1.19–1.64) | 584 | 4.89E−05 | 6.86E−05 | 0.193 | NA | NA |
Critical COVID-19 | WM | 0.298 (0.117) | 1.35 (1.07–1.69) | 584 | 0.011 | NA | NA | NA | NA |
Critical COVID-19 | MR-Egger | 0.262 (0.269) | 1.30 (0.77–2.20) | 584 | 0.331 | 6.18E−05 | 0.193 | 0.001 | 0.779 |
NA, not applicable; Q_P, Cochran P value of heterogeneity analysis; WM, weighted median.
A total of 5,089, 4,156, 806, and 1,551 risk SNPs were obtained from the GWAS catalog for BMI, diabetes, hypertension, and coronary artery disease, respectively. These SNPs constituted a total of 11,167 unique SNPs. After excluding IVs overlapping with these SNPs, our MR analysis showed that BMR still exerts causal effects on SARS-CoV-2 infection (OR 1.12, 95% CI 1.07–1.18, P = 4.37E−06), hospitalized COVID-19 (1.26, 1.13–1.40, P = 3.83E−05), and critical COVID-19 (1.30, 1.10–1.53, P = 1.90E−03) (Table 2). The sensitivity analysis revealed that the directions of causal effect estimates across the methods were all the same (OR >1) (Fig. 1C). These results were very similar to those obtained in the original MR analysis.
Causal effects of BMR on COVID-19 outcomes after excluding instrumental variables associated with four confounding factors
Outcome . | Method . | β (SE) . | OR (95% CI) . | Instrumental variables (n) . | P . | Q_P . | I2 . | MR-Egger intercept . | MR-Egger intercept P . |
---|---|---|---|---|---|---|---|---|---|
SARS-CoV-2 infection | IVW | 0.117 (0.025) | 1.12 (1.07–1.18) | 567 | 4.37E−06 | 2.45E−04 | 0.181 | NA | NA |
SARS-CoV-2 infection | WM | 0.121 (0.035) | 1.13 (1.05–1.21) | 567 | 5.16E−04 | NA | NA | NA | NA |
SARS-CoV-2 infection | MR-Egger | 0.045 (0.084) | 1.05 (0.89–1.23) | 567 | 0.595 | 2.43E−04 | 0.179 | 0.001 | 0.368 |
Hospitalized COVID-19 | IVW | 0.229 (0.055) | 1.26 (1.13–1.40) | 569 | 3.83E−05 | 6.45E−05 | 0.196 | NA | NA |
Hospitalized COVID-19 | WM | 0.193 (0.073) | 1.21 (1.05–1.40) | 569 | 8.42E−03 | NA | NA | NA | NA |
Hospitalized COVID-19 | MR-Egger | 0.027 (0.184) | 1.03 (0.72–1.47) | 569 | 0.883 | 6.80E−05 | 0.194 | 0.002 | 0.252 |
Critical COVID-19 | IVW | 0.259 (0.083) | 1.30 (1.10–1.53) | 568 | 1.90E−03 | 1.19E−04 | 0.189 | NA | NA |
Critical COVID-19 | WM | 0.256 (0.113) | 1.29 (1.04–1.61) | 568 | 0.023 | NA | NA | NA | NA |
Critical COVID-19 | MR-Egger | 0.186 (0.279) | 1.20 (0.70–2.08) | 568 | 0.505 | 1.08E−04 | 0.189 | 0.001 | 0.785 |
Outcome . | Method . | β (SE) . | OR (95% CI) . | Instrumental variables (n) . | P . | Q_P . | I2 . | MR-Egger intercept . | MR-Egger intercept P . |
---|---|---|---|---|---|---|---|---|---|
SARS-CoV-2 infection | IVW | 0.117 (0.025) | 1.12 (1.07–1.18) | 567 | 4.37E−06 | 2.45E−04 | 0.181 | NA | NA |
SARS-CoV-2 infection | WM | 0.121 (0.035) | 1.13 (1.05–1.21) | 567 | 5.16E−04 | NA | NA | NA | NA |
SARS-CoV-2 infection | MR-Egger | 0.045 (0.084) | 1.05 (0.89–1.23) | 567 | 0.595 | 2.43E−04 | 0.179 | 0.001 | 0.368 |
Hospitalized COVID-19 | IVW | 0.229 (0.055) | 1.26 (1.13–1.40) | 569 | 3.83E−05 | 6.45E−05 | 0.196 | NA | NA |
Hospitalized COVID-19 | WM | 0.193 (0.073) | 1.21 (1.05–1.40) | 569 | 8.42E−03 | NA | NA | NA | NA |
Hospitalized COVID-19 | MR-Egger | 0.027 (0.184) | 1.03 (0.72–1.47) | 569 | 0.883 | 6.80E−05 | 0.194 | 0.002 | 0.252 |
Critical COVID-19 | IVW | 0.259 (0.083) | 1.30 (1.10–1.53) | 568 | 1.90E−03 | 1.19E−04 | 0.189 | NA | NA |
Critical COVID-19 | WM | 0.256 (0.113) | 1.29 (1.04–1.61) | 568 | 0.023 | NA | NA | NA | NA |
Critical COVID-19 | MR-Egger | 0.186 (0.279) | 1.20 (0.70–2.08) | 568 | 0.505 | 1.08E−04 | 0.189 | 0.001 | 0.785 |
NA, not applicable; Q_P, Cochran P value of heterogeneity analysis; WM, weighted median.
The MVMR analysis shows that the causal effects of BMR on hospitalized COVID-19 (OR 0.99, 95% CI 0.87–1.13, P = 0.921) and critical COVID-19 (0.89, 0.74–1.07, P = 0.212) were dependent on BMI and type 2 diabetes. Notably, BMR may affect the SARS-CoV-2 infection risk independently of BMI and type 2 diabetes (1.09, 1.03–1.15, P = 4.82E−03) (Table 3). The P values of the Cochran Q test were 3.04E−05, 6.44E−08, and 1.71E−05 for SARS-CoV-2 infection, hospitalized COVID-19, and critical COVID-19, respectively. Therefore, the Cochran Q test supported the potential heterogeneity in the MR estimates.
MVMR analysis between BMR and the COVID-19 outcomes
Exposure . | Outcome . | β (SE) . | OR (95% CI) . | Instrumental variables (n) . | P . |
---|---|---|---|---|---|
BMI | SARS-CoV-2 infection | 0.088 (0.027) | 1.09 (1.03–1.15) | 198 | 1.32E−03 |
BMR | SARS-CoV-2 infection | 0.083 (0.029) | 1.09 (1.03–1.15) | 321 | 4.82E−03 |
Type 2 diabetes | SARS-CoV-2 infection | 0.005 (0.008) | 1.00 (0.99–1.02) | 77 | 0.567 |
BMI | Hospitalized COVID-19 | 0.409 (0.062) | 1.51 (1.33–1.70) | 198 | 3.41E−11 |
BMR | Hospitalized COVID-20 | −0.007 (0.066) | 0.99 (0.87–1.13) | 321 | 0.921 |
Type 2 diabetes | Hospitalized COVID-21 | 0.004 (0.019) | 1.00 (0.97–1.04) | 77 | 0.842 |
BMI | Critical COVID-19 | 0.538 (0.090) | 1.71 (1.44–2.04) | 200 | 2.22E−09 |
BMR | Critical COVID-20 | −0.120 (0.096) | 0.89 (0.74–1.07) | 322 | 0.212 |
Type 2 diabetes | Critical COVID-21 | −0.008 (0.028) | 0.99 (0.94–1.05) | 78 | 0.771 |
Exposure . | Outcome . | β (SE) . | OR (95% CI) . | Instrumental variables (n) . | P . |
---|---|---|---|---|---|
BMI | SARS-CoV-2 infection | 0.088 (0.027) | 1.09 (1.03–1.15) | 198 | 1.32E−03 |
BMR | SARS-CoV-2 infection | 0.083 (0.029) | 1.09 (1.03–1.15) | 321 | 4.82E−03 |
Type 2 diabetes | SARS-CoV-2 infection | 0.005 (0.008) | 1.00 (0.99–1.02) | 77 | 0.567 |
BMI | Hospitalized COVID-19 | 0.409 (0.062) | 1.51 (1.33–1.70) | 198 | 3.41E−11 |
BMR | Hospitalized COVID-20 | −0.007 (0.066) | 0.99 (0.87–1.13) | 321 | 0.921 |
Type 2 diabetes | Hospitalized COVID-21 | 0.004 (0.019) | 1.00 (0.97–1.04) | 77 | 0.842 |
BMI | Critical COVID-19 | 0.538 (0.090) | 1.71 (1.44–2.04) | 200 | 2.22E−09 |
BMR | Critical COVID-20 | −0.120 (0.096) | 0.89 (0.74–1.07) | 322 | 0.212 |
Type 2 diabetes | Critical COVID-21 | −0.008 (0.028) | 0.99 (0.94–1.05) | 78 | 0.771 |
Discussion
Our study provides convincing evidence for the potential influence of BMR on COVID-19 outcomes. Our results indicated that a 1-SD increase in BMR translates to a 14% increased risk for COVID-19 infection, a 28% increased risk for COVID-19 hospitalization, and a 34% increased risk for critical COVID-19. Thus, the associations of BMR with COVID-19 are severity dependent, with larger impacts displayed toward the more adverse outcomes of the illness. The MVMR analysis showed that the effects of BMR on the severity of COVID-19 can be accounted for by BMI and type 2 diabetes. However, BMR was associated with an increased risk for SARS-CoV-2 infection independent of the two confounding factors.
It seems that an increased BMR may serve as a general indicator for disturbances within metabolic and/or endocrine systems, which are well known to be affected by COVID-19. Moreover, preexisting metabolic dysfunction (e.g., obesity, diabetes, and hypertension) may aggravate the COVID-19 symptoms and the overall course of the illness (24). BMR reflects the steady-state level of energy homeostasis. A higher BMR is detected when maintenance of the body requires extra effort, which may be interpreted as narrowed margins for the management of environmental stresses, including COVID-19, thus predisposing patients to more severe outcomes. Notably, a high BMR was identified as an independent and dose-dependent risk factor for all-cause mortality in otherwise healthy participants (25). A higher BMR may indicate endocrine abnormities or metabolic and vascular disorders, which may mediate the causal effect of BMR on COVID-19 outcomes. A higher BMR is also commonly associated with higher systemic inflammation, which is a known predisposing factor for severe outcomes of COVID-19 (26). It is also of note that COVID-19 may further aggravate preexisting metabolic problems. In fact, in the case series, the average resting energy expenditure of critical COVID-19 patients was much higher than the predicted value (27), indicating the need for changes in nutrition requirements and hypothermic treatments. A further collection of calorimetric measurements in COVID-19 patients traversing various courses of illness is warranted.
The main strength of the study is that MR analysis is less affected by causality pitfalls, which are common in traditionally designed observational studies due to confounding factors and reverse causation. The largest available GWAS summary data sets were used for tracing the causative association between severe COVID-19 and BMR. All of the participants in the GWAS data sets were of European ancestry, reducing the potential population heterogeneity.
Our study has several limitations. In particular, we assessed only genetic liability to BMR and outcomes of COVID-19, with no regard to the effects of the environment, which are critical for both BMR and COVID-19. We acknowledge that the causal associations identified in our study may be mediated by confounding factors other than hypertension, obesity, and diabetes, which could not be completely ruled out. Nonhomogenous data sets used in the MR analysis may result in heterogeneity in the causal effect estimates. To mitigate this risk, we tested the MR assumptions using various models, with no substantial evidence of heterogeneity detected.
Conclusions
In summary, our study suggested that a higher BMR may be a valuable indicator associated with an increased risk for COVID-19 susceptibility and severity.
This article is part of a special article collection available at diabetesjournals.org/journals/collection/43/Diabetes-and-COVID-19-Articles.
Article Information
Acknowledgments. The authors thank all investigators and participants from the COVID-19 Host Genetics Initiative and UK Biobank for sharing these data.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. A.B., Y.S., and H.C. interpreted the results and wrote the manuscript. F.Z. conceived the project, analyzed the data, interpreted the results, and wrote and revised the manuscript. F.Z. is the guarantor of this work and, as such, 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.