Racial/ethnic disparities on inflammation and response to methylprednisolone in severe COVID-19 pneumonia

Real world data was collected from Hackensack Meridian Health (HMH), a NJ health network comprising of 13 hospitals on patients ≥ 18 years of age, and hospitalized for at least 2 days between March 1, 2020 and June 15, 2020 with severe COVID-19 Pneumonia [21]. These patients had a positive SARS-CoV-2 PCR and had SpO₂ < 94% on room air at sea level, a respiratory rate > 30 breaths/min, PaO₂/FiO₂ < 300 mm Hg, or lung infiltrates > 50%. We excluded patients who had different corticosteroid regimen other than methylprednisolone. Approval was obtained by the Hackensack Meridian Health Institutional Review Board (study #Pro2020-0485) and the study was also registered on ClinicalTrials.Gov as a prospective observational database (NCT04347993).

Demographic data such as age, gender, race, ethnicity, comorbidities, and sex were self-reported. Weight and height were measured. SARS-CoV-2 was detected in nasal swabs by RT-PCR. Routine blood tests included complete blood count (CBC), coagulation profile, complete metabolic profile (CMP), inflammatory markers [interleukin-6 (IL-6), C reactive protein (CRP), d-dimer, and ferritin], and arterial blood gas(ABG). Data was entered into Redcap and abstracted from June to December 2020.

The primary outcomes are the levels of the inflammatory markers of survivors and non-survivors and the association of methylprednisolone dose to in hospital survival for each racial/ethnic group.

The data in this study is primarily the propensity score matched cohort of comparison of in-hospital survival of COVID-19 admitted patients that were treated with methylprednisolone and patients not treated with methylprednisolone. [21] In the one-to-one propensity score matched design of the parent study, patients from the methylprednisolone administration groups were matched based on variables associated with mortality such as age (age ≥ 60 years vs. age < 60 years), obesity (BMI ≥ 30.0 kg/m² vs. BMI < 30.0 kg/m²), sex (M/F), diabetes (Yes/No), hypertension (Yes/No), cancer (Yes/No), respiratory Rate (respiratory rate > 22 bpm vs. ≤ 22 bpm), chronic kidney disease or chronic renal failure (Yes/No), low oxygen (oxygenation > 94% vs oxygenation ≥ 94%), CRP (CRP > 20 mg/dL vs. CRP ≤ 20 mg/dL), and qSOFA (score: 0,1,2,3) [21]. A nearest-neighbor method (greedy match) was employed using a caliper of 0.20 to obtain the propensity matched sample. A post-match assessment of the distribution of propensity scores (or logit of propensity scores) and balance in the adjusted variables between the no methylprednisolone (NMP) and methylprednisolone (MP) using standardized difference and variance ratio were performed. This analysis and graphical displays were obtained by the ASSESS statement of PROC MATCH in SAS 9.4 [21]. The aims of this subgroup analysis were pursued between and within the racial/ethnic groups that exhausted the span of racial/ethnic groups of the propensity score-matched cohort from primary study [21].

In this study descriptive statistics was used to summarize the data. Categorical variables were presented as frequencies and percentages. Continuous variables were presented as the median and interquartile range (IQR) for non-normally distributed data and mean and standard deviation for normally distributed data. The normality assumption of continuous variables was assessed using Shapiro–Wilk test. Comparison of matched continuous variables was performed using two-sided paired t-test or Wilcoxon signed-rank test, as appropriate. Comparison of paired categorical variables was performed using McNemar’s test or Bowker test of symmetry, as appropriate.

Comparison of levels of inflammatory markers, creatinine and age between racial/ethnic groups, which are independent, was performed using Kruskal–Wallis test followed by pairwise comparisons utilizing a two-sided Wilcoxon rank sum test.

Comparison of these continuous variables between survivors and non-survivors within each racial/ethnic group was performed using two-sided Wilcoxon rank sum test. Boxplots were used to illustrate the different distribution between survivor and non-survivors in the inflammatory markers. To assess the discriminative ability of inflammatory markers for in-hospital mortality, area under the receiver operating characteristics (ROC) curve of the inflammatory markers in each racial/ethnic group using ROC statements in PROC LOGISTIC SAS 9.4. Comparison of area under ROCs within each racial/ethnic group were evaluated by DeLong test, with ROC closest to 0.5 as the reference. Area under ROCs were reported as area (95% confidence interval). AUC closer to 1 with a 95% CI that excludes 1 was considered statistically significant.

Within each racial/ethnic group, this study also examined the association between in-hospital survival and inflammatory markers, methylprednisolone dose (none, low dose, high dose), older age obesity, sex, diabetes, hypertension cancer, respiratory rate (bpm), chronic kidney disease (or renal failure), supplemental oxygen support (None, non-mechanical ventilation, mechanical ventilation), CRP ≥ 20 mg/dL, qSOFA and creatinine ≥ 1.5 mg/dL. Methylprednisolone (MP) dose cutoff value of 1.36 mg/kg/day had been previously determined as optimal based on Youden index method such that low dose (LD) MP was defined as MP dose < 1.36 mg/kg/day and high dose (HD) MP was defined as MP dose ≥ 1.36 mg/kg/day. Dichotomized inflammatory marker we employed in all the time to in-hospital mortality event analysis as follows: IL-6 > 25 pg/mL, D-dimer ≥ 2 mcg/mL, CRP ≥ 20 mg/L and ferritin ≥ 1400 μg/L as high levels of each respective marker. Time to in-hospital mortality was estimated using Kaplan–Meier product limit method. Comparison of in-hospital survival between independent groups was estimated by a two-sided log-rank test and matched cohort were compared using stratified log rank. To examine the association of risk/benefit of in-hospital survival with the factor of interest such as methylprednisolone treatment, in a univariate model, the Cox proportional hazard regression analysis with robust covariance [22] (sandwich estimator) to account for paired observations was conducted. The proportional hazard (PH) assumption, critical in Cox regression modelling, was evaluated using a Kolmogorov-type supremum test [23] in the ASSESS statement of PROC PHREG. If the PH assumption was violated for any covariate, then a continuous variable which also violated the PH assumption and its interaction with time were included in the model to adjust for the significant interaction with time to the risk of mortality [24]. After conducting univariate PH Cox regression or NPH Cox regression, any covariate that reported P < 0.15 were entered into the initial full model of the multivariable Cox model to initiate backward elimination procedure. If any of covariates that satisfied the criterion of P < 0.15 above had been found to have violated PH assumption then a multivariate NPH Cox regression model was analysis, otherwise if all the variables satisfied the PH assumption then a standard multivariable PH Cox model was fit. Results of all univariate and multivariable Cox regression (PH/NPH) within each race/ethnic group were presented as hazard ratio (HR), 95% Cis and P-value. All data analysis was performed using SAS 9.4 (SAS Institute Inc., Cary, North Carolina, USA). Unless otherwise specified, any P < 0.05 was considered statistically significant.