key: cord-1042134-9umi42i4
authors: Mohzari, Yahya A.; Alamer, Ahmad; Alattas, Majda; Alrashed, Ahmed A.; Alshehab, Nura A.; Alkhaldi, Turkiah K.; Alamer, Amnah A.; Asdaq, Syed M.B; Aljefri, Doaa; Alajami, Hamdan N.; Alsowaida, Yazed S.; Mathew, Maya; AlMusawa, Mohammed I.; Alomar, Mukhtar; Alharbi, Raghad Y.; Khuwaja, Malik; Bamogaddam, Reem F.; Alharthi, Ashwaq H.; Faqihi, Aisha Y.; Alrumayyan, Bashayer F.; Alshareef, Abeer; Alhassan, Batool M.; Damfu, Nader Y.; Alajmi, Ghada S.; Albujaidy, Asma; Alghalbi, Maram; Alajlan, Saleh A; Abraham, Ivo; Almulhim, Abdulaziz S.
title: Tocilizumab effectiveness in mechanically ventilated COVID-19 patients (T-MVC-19 Study): a multicenter real-world evidence
date: 2022-03-07
journal: Expert Rev Anti Infect Ther
DOI: 10.1080/14787210.2022.2046462
sha: 42b0cf14584685774861dc11f58cdeb1c3ff9b2c
doc_id: 1042134
cord_uid: 9umi42i4

BACKGROUND: This study aimed to evaluate the effectiveness of tocilizumab in mechanically ventilated patients with coronavirus disease 2019 (COVID-19). RESEARCH DESIGN AND METHODS: This retrospective multicenter study included adults (≥18 years) diagnosed with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) by real-time polymerase chain reaction (RT-PCR) from nasopharyngeal swab, and requiring invasive mechanical ventilation during admission. Survival analyses with inverse propensity score treatment weighting (IPTW) and propensity score matching (PSM) were conducted. To account for immortal bias, we used Cox proportional modeling with time-dependent covariance. Competing risk analysis was performed for the extubation endpoint. RESULTS: A total of 556 (tocilizumab = 193, control = 363) patients were included. Males constituted the majority of the participants (69.2% in tocilizumab arm,74.1% in control arm). Tocilizumab was not associated with a reduction in mortality with hazard ratio [(HR) = 0.82,95% confidence interval (95%CI): 0.62–1.10] in the Inverse propensity score weighting (IPTW) analysis and (HR = 0.86,95% CI: 0.64–1.16) in the PSM analysis. However, tocilizumab was associated with an increased rate of extubation (33.6%) compared to the control arm (11.9%); subdistributional hazards (SHR) = 3.1, 95% CI: 1.86–5.16). CONCLUSIONS: Although tocilizumab was not found to be effective in reducing mortality, extubation rate while on mechanical ventilation was higher among tocilizumabtreatedgroup.

In December 2019, a series of pneumonia cases secondary to severe acute respiratory syndrome coronavirus 2 (SARS- and what is currently known as the novel coronavirus disease June 2021, more than 180 million patients globally have been infected with SARS-CoV-2 and 3.9 million deaths have been reported [4] . In Saudi Arabia, 480,000 confirmed COVID-19 cases in 100 cities were reported, of which 7,744 patients have died [5] .

COVID-19 is characterized by fever, cough, fatigue, shortness of breath, pneumonia, and other respiratory tract symptoms [6, 7] . However, the severity of the disease can range from mild flu-like symptoms to a devastating course of disease requiring respiratory support and Intensive Care Unit (ICU) admission [8] . The binding of the virus to the airway epithelial cells in COVID-19 patients results in the activation and upregulation of the innate and adaptive immune response and releases a large number of cytokines, including Interleukin-6 (IL-6) [8, 9] . A correlation between COVID-19 and levels of IL-6 has been reported [10, 11] , suggesting the possibility of repurposing IL-6 inhibitors in the management of severe COVID-19.

Tocilizumab, a recombinant humanized anti-IL-6 receptor monoclonal antibody, has been approved for the treatment of severe chimeric antigen receptor T cell-induced cytokine release syndrome [12] . Several studies have demonstrated the clinical benefit of tocilizumab in severe COVID-19 cases [13] [14] [15] [16] [17] [18] , though this has not been consistent [15, 16] . In the RECOVERY trial, the allocation to tocilizumab was associated with a significant reduction in 28-day mortality compared with usual care alone in patients on non-mechanical ventilation [rate ratio (RR) = 0.85; 95% confidence interval (CI): 0.76-0.94; p = 0.003] but not in patients on invasive mechanical ventilation; [RR = 0.93. 95% CI, 0.74 − 1.18] [18] . More generally, studies were limited due to heterogeneity in study designs, small sample sizes, diversity of study populations, variations in disease severity spectrum, wide ranges of dose regimen, and its frequency, differences in timing of starting tocilizumab therapy, and limiting results to pooled crude unadjusted estimates.

A recent meta-analysis of 25 trials revealed an association of tocilizumab with a reduction in overall mortality [odds ratio (OR) = 0.70, 95% CI, 0.54-0.90, P = 0.007], and mechanical ventilation requirement [OR = 0.59, 95% CI, 0.37-0.93, P = 0.02] [19] . These studies were conducted in different countries including the United States, United Kingdom, Canada, Italy, and Spain, but none of these studies were conducted in Saudi Arabia or the Middle East. Additionally, the contradicting data challenge the decision-makers in determining optimal clinical practice for tocilizumab. In this study, we aimed to evaluate the effectiveness of tocilizumab in COVID-19 mechanically ventilated patients in daily clinical practice in Saudi Arabia at the height of the early pandemic.

This was a retrospective study conducted in six centers in three cities in Saudi Arabia: King Saud Medical City Hospital (KSMC) [20] . All methods were carried out in accordance with relevant guidelines and regulations.

We obtained lists of severely/critical ill COVID-19 patients or those who received tocilizumab treatment between March 2020 to January 2021. Through a random-selection process, we screened patients for eligibility. The random selection would minimize sampling bias and provide an equal opportunity for the patient's record to be selected and coded [21] . Inclusion criteria were as follows: age ≥18 years, diagnosis of SARS-CoV-2 by real-time polymerase chain reaction (RT-PCR) from nasopharyngeal swab, and requiring invasive mechanical ventilation at admission. Patients were excluded if they were outside the study period, did not require mechanical ventilation, or received tocilizumab treatment for an indication other than COVID-19.

Tocilizumab was dosed according to Saudi Ministry of health (MOH) protocols version 2.0 updated in June 2020. The adult dosing for tocilizumab was 4-8 mg/kg (usual dose 400 mg; maximum 800 mg) by IV infusion; repeated within 12 hours for a maximum of 2 doses [22] .

Data were manually extracted from electronic health records (EHRs) and then entered into the Research Electronic Data Capture (REDCap) system in a de-identified manner [23] . A trained team of data managers ensured the quality of data collection and resolved any discrepancies.

The aim was to assess tocilizumab effectiveness in mechanically ventilated COVID-19 patients as compared to a control group that did not receive tocilizumab. The primary endpoint was mortality after mechanical ventilation (referred to here as overall mortality). The secondary endpoint was the rate of extubation. Subgroup analyses for the mortality outcome based on baseline characteristics were explored. We also explored the association between time of tocilizumab administration and mortality.

Assuming a − 10% risk difference in the overall mortality between the tocilizumab and control arms, a sample size of 489 (166 patients who received tocilizumab and 332 who were unexposed to tocilizumab with an estimated ratio of 1:2) assured a power of 80% after applying continuity correction. A significance level (α, type 1 error rate) of 5% was applied.

We used R Core Team (2020) software (R Foundation for Statistical Computing, Version 4.0.1, Vienna, Austria. The following packages in the R interface were used to conduct the analyses: survival [24] , ggplot2 [25] , survminer [26] , survey [27], mice [28] , matchThem [29] , cobalt [30], crrSC [31] .

Descriptive statistics was used to present baseline characteristics. Continuous data were presented as means with standard deviations (±SD) and medians with interquartile ranges (IQRs). Student's t-test or Mann-Whitney U test was used for between-arm comparison.

Categorical data were reported as frequencies and percentages and analyzed using either the Chi-square test for nxm tables or Fisher's exact test for 2×2-table group comparisons. Missing data were determined to be missing at random (MAR) and handled by Multiple Imputation by Chained Equations (MICE) procedure with Nelson Aalen estimator [28, 32] . We only considered variables that had <15% of their data missing. Fifty imputations were obtained for the missing value (5 imputed datasets with 10 iterations).

Due to the observational design and the presence of many confounders, we implemented propensity score-based methods to estimate the marginal treatment effect. First, we carried out the inverse propensity score treatment weighting (IPTW) procedure. In IPTW, propensity scores (probability of getting a treatment) were calculated from a multivariable logistic regression model that included all the covariates of interest. Correlation testing was conducted to avoid the inclusion of highly correlated variables in the models. Next, propensity scores were used to calculate weights based on Desai and Franklin formulas [33] using the average treatment effect among the treated (ATT) as the target estimand (i.e. the 'ideal' patients for tocilizumab treatment based on specific characteristics). Balance between arms was achieved by creating a pseudo-population in which treatment allocation was independent of the observed covariate. To handle extreme weights, we utilized weight trimming (truncation) and stabilization. Balance of the covariates was checked by standardized mean difference (SMD) and love plots. SMD values <0.2 indicated a good balance. Propensity scores distributions were illustrated by mirror diagrams. As weighting may inflate or deflate the sample size relative to the original population; therefore, a robust sandwich-type estimator for variance estimation was calculated using 'survey' package in R for the treatment effect estimates [34] .

After weighting and using this package, we fitted an adjusted (weighted) Kaplan-Meier (KM) model to estimate the probability of survival after mechanical ventilation (time 0) in both arms. Events were censored if patients were discharged alive or were still intubated at the time of data collection. A log rank test that accounted for the weighted data was applied to detect differences in survival curves.

To estimate the relative treatment effect, a Cox proportional model was fitted on the weighted data with the arm as a single covariate using the 'survey' package for robust variance estimation. The estimate was subsequently pooled across the imputed datasets. Proportional hazard assumption was checked by Schoenfeld residuals plots and statistical testing.

Propensity scores matching (PSM) was performed to confirm the results from the IPTW analysis. Propensity scores were generated from the same logistic regression model used in the IPTW model, which included all variables of interest. We then used the nearestneighbor matching within the imputed datasets approach (using Rubin's rules) [35] and 1:2 ratio with a caliper of 0.2 [36] and no replacement. Adequate balance of the data was checked by SMD (<0.2 desirable). Visual diagnostics such as love plots for covariate balance and mirror plots for propensity score distributions were generated. Next, we fitted a KM model to the matched data to compare survival probabilities after mechanical ventilation. To test for the statistical difference in the survival curves, we performed a stratified log rank test [37] . For relative treatment effect, a Cox proportional hazard model was fitted using the matched dataset. To account for the paired nature of the data, we stratified on the matched pairs and a robust variance was estimated using the 'survey' package.

The study design may entail a potential immortal time bias. In randomized control trials, the time of treatment assignment and time 0 is aligned. This was not possible in our case as patients may have started tocilizumab treatment after a few days of mechanical ventilation (time 0). During this period, the patient must be alive to receive the treatment later (i.e. immortal bias). The persondays during which patients in the tocilizumab arm did not receive treatment should be accounted toward the untreated group. A Cox proportional hazard model with time dependency covariance was used to evaluate a possible immortal time bias. We performed these analyses with the propensity score-based procedures (IPTW and PSM).

For the secondary outcome (extubation), we performed a competing risk analysis for death while on mechanical ventilation; specifically, patients who die while on mechanical ventilation cannot experience the extubation event. The cumulative incidence function (CIF) was estimated using a cluster Fine-Gray model to derive the subdistribution hazard ratio (SHR): the instantaneous risk of failure from an event in subjects who have not failed that type of an event [38] . In the case of the extubation outcome, a SHR>1 indicated a higher incidence probability of the event occurring in tocilizumab. Following the recommendation by Austin et al.,

we conducted the analysis on the matched dataset using the CrrSC package in R to account for the matched pairs [39] . The cause-specific hazard ratio (CSHR) was also estimated as part of the competing risk analysis as suggested by Latouche et al. by fitting Cox proportional models on the matched dataset [40] . Cause specific hazards refers to the instantaneous rate of the outcome of interest in subjects who are event-free. In the case of extubation outcome, a cause specific hazard ratio (CSHR) values of >1 meant a relative increase in the instantaneous rates of extubation rates favoring the tocilizumab arm. The CIF was plotted for each arm and competing risk.

We investigated the association between the time of tocilizumab administration and overall mortality. Using both propensity score approaches (IPTW and PSM), a logistic regression model was fitted that included time of administration as a covariate (prior mechanical ventilation, or within 48 hours and >48 hours of mechanical ventilation) versus control as a reference. We also conducted subgroup analyses to examine the baseline characteristics as treatment effect modifiers of overall mortality using the matched dataset.

A total of 899 patients were screened for inclusion eligibility ( Figure 1 ). We excluded 334 patients for various reasons, the most common one was not being on mechanical ventilation at admission (n = 231). We included 193 patients in the tocilizumab and 363 patients in the control arm.

Overall, there were no differences in the baseline characteristics of age, gender, weight, body mass index (BMI), kidney function and comorbid conditions such as hypertension, diabetes, cardiovascular disease. However, differences were observed in terms of ethnicity (P = 0.044); solid organ transplant (P= 0.025); medication used in the hospital prior to ICU admission such as favipiravir (P < 0.001), hydroxychloroquine (<0.001), therapeutic anticoagulation (<0.001), and convalescent plasma therapy (P < 0.001). More patients in the tocilizumab arm had a fever at the time of intubation (P = 0.030) and presented with a lower median partial oxygen pressure to fractional inspired oxygen (paO2/Fio2) ratio (71.1 versus 102.0 in the control arm). Arms also differed in terms of the following laboratory variables: albumin, aspartate aminotransferase, lactate dehydrogenase, ferritin, sodium, potassium, and magnesium. Variables with the highest missing data were alkaline phosphatase (35%) and erythrocyte sedimentation rate (41.2%) (see Table 1 footnote)

The IPTW and PSM mirror plots for propensity score distribution were presented in Figures S1 and S2 and the Love plots for covariate balance for both IPTW and PSM in Figure S3 and Table S1 , all showing adequate balance. Of the patients in the tocilizumab arm, 21.2% of received the treatment within (Table S2 ).

The overall death rate in the control arm (81.1%) was higher compared to the tocilizumab arm (62.6%). The IPTW analysis examining the two adjusted survival curves showed a difference in the absolute effect (log rank test, P = 0.033; Figure 2a) . As for the relative treatment effect, tocilizumab was associated with a lower incidence of mortality with HR = 0.73 (95%CI 0.55-0.96, P = 0.026). However, the PSM analyses did not confirm the IPTW results. No differences between the survival curves ( Figure 2b , stratified log rank test P = 0.900) and in the relative treatment effects was observed (HR = 0.80, 95% CI 0.57-1.12, P = 0.192). When accounting for immortal time bias with Cox time dependent covariance, both the IPTW (HR = 0.82, 95%CI 0.62-1.10, P = 0.190) and the PSM (HR = 0.86, 95%CI of 0.64-1.16, P = 0.349) analyses showed no difference in overall mortality between the two arms (Table 2; Figure 3 ).

A higher percentage of patients experienced extubation in the tocilizumab arm (33.6%) versus the control arm (11.9%). When compared to the control, the CSHR for the extubation outcome for tocilizumab was 2.72 (95% CI 1.56-4.76, P < 0.001) and the subdistribution hazard ratio (SHR) was 3.1 (95% CI 1.86-5.16, P < 0.001) from the Fine-Gray model. Whereas there was no significant result for death on the mechanical ventilation outcome with CSHR of 0.98 (95% CI 0.70-1.38, P = 0.927) but a significant result for the Fine-Gray model with a SHR of 0.68 (95% CI 0.511-0.901, P = 0.007). The results of the competing risk analysis were presented in Table 3 and the cumulative incidence function visualized in Figure 2c. 

No association between the time of tocilizumab administration and overall mortality was found (Table S3 ). The subgroup analyses for effect modifiers based on baseline characteristics showed interaction effects only for chronic kidney disease stage, angiotensin converting enzyme inhibitor (ACEIs) or angiotensin receptor blocker use (ARBs), and Pao2/Fio2 ratio (>100 versus <100), but not for any other variables (see Figure S4 ).

Therapeutic management of COVID-19 is evolving and expanding, with several treatments having been approved, mainly under the emergency use authorization provision (EUA) [41] [42] [43] . Among them, onlydexamethasone and tocilizumab have shown survival benefits, yet not consistently so across disease spectrums nor across studies [6, [44] [45] [46] . The association between cytokine release syndrome and COVID-19 severity sparked an interest in the three available anti-IL-6 monoclonal antibodies (tocilizumab, sarilumab, and siltuximab) [47, 48] . Of these, tocilizumab has received the most interest due to its availability and the experience gained since its approval in 2008 [49] . Considering the accumulated evidence, tocilizumab was authorized for the treatment of COVID-19 in a special patient population and has been incorporated in the treatment guidelines, including for patients with rapid respiratory decompensation [22] . Our study differs from prior published studies in that it included only mechanically intubated and examined whether tocilizumab is effective in reducing mortality and increasing ventilator independency (extubation) compared to standard care. Although both the IPTW and PSM analyses showed that tocilizumab was effective in reducing mortality, this statistical significance was lost when we accounted for immortal time bias. These findings are in concordance with several recent studies [18, 46, [50] [51] [52] . Conversely, in a study in mechanically intubated COVID-19 All analyses were conducted on the matched dataset. † Competing risk analysis conducted using Fine-Gray model on the matched dataset using crrSC package to account for the matched pairs. ‡ Statistical test to compare cumulative incidence function curves MV: Mechanical Ventilation CSHR: Cause specific hazard ratio SHR: Subdistribution hazard ratio patients, Somers and colleagues found that exposure to tocilizumab was associated with a reduction in mortality by 45% [53] . However, more than 40% of patients in the tocilizumab arm were treated 24 hours after intubation, which may have introduced a misclassification bias [54] . Similar to our study, the authors balanced the tocilizumab and control groups using IPTW [53] . In other studies with analyses similar to ours, neither steroids nor PaO2/FiO2 were included in the propensity score model [55] ; or patients who died before receiving tocilizumab were excluded, thus possibly inducing selection bias [54] . Further, and although their study was not restricted to mechanically ventilated patients, Biran and colleagues [56] observed that tocilizumab was associated with a reduction in mortality by 29%. Here too, the mortality benefit was lost after adjusting for immortal time bias. A recent meta-analysis published by the World Health Organization (WHO) Rapid Evidence Appraisal for COVID-19 Therapies (REACT) Working Group aimed to synthesize the evidence on IL-6 inhibitors exposure and mortality [57] . While the overall analysis showed that IL-6 inhibitors were associated with reduction in mortality, the subgroup analysis of mechanically ventilated patients didn't favor IL-6 inhibitors when compared to control.

The absence of an unequivocal mortality benefit in our study may be explained by two factors. First, initiating an immunomodulator at a later stage of the cytokine release syndrome may be less effective than initiating it at an earlier stage. The mortality benefit observed with tocilizumab was observed mainly in patients with moderate to severe COVID-19 who did not require mechanical ventilation [56, [58] [59] [60] [61] . Russell and colleagues in a multicenter retrospective study comparing early (prior to or within 24 hours of intubation) versus late tocilizumab (more than 24 hours after intubation) administration, found that early administration was associated with a 85% reduction in mortality when compared to no tocilizumab [62] . In addition to the small sample size, it is worth noting that the matching model included only a few variables, which could have decreased the precision of the exposure effect [63, 64] . In our study, the majority (72%) received tocilizumab within 48 hours of intubation (Table S2) with no differences in mortality regardless of the timing of administration (Table S3) . Though subject to further investigation in a randomized controlled trial, because of its larger sample size and the rigorous differentiated statistical analyses, our study provides initial evidence of no mortality benefit with early tocilizumab administration. Second, despite the evidence of IL-6 as a proinflammatory cytokine involved in cytokine release syndrome, other proinflammatory cytokines may also be involved and might mediate or moderate outcomes in worsening COVID-19 patients [48] . The potential benefit of using other immunomodulators against these cytokines should be examined under controlled conditions as well as in the setting of daily clinical practice.

With regard to our results related to the extubation rate, we performed a competing risk analysis to account for events that might compete with this secondary study outcome (i.e. death on mechanical ventilation). The addition of tocilizumab reduced the hazard of COVID-19 patients to remain intubated by 172% (CSHR: 2.72). Our findings differ from those of a retrospective study by Fisher and colleagues in which the extubation rate in tocilizumab arm was similar to that in the control arm (OR 1.53; 95% CI, 0.71 − 3.30) [52] . An important difference in our study is that patients treated with tocilizumab received more concurrent medications (notably, favipiravir and convalescent plasma) compared to those in the study by Fisher and colleagues. Despite the equivocal results on the primary outcome of interest, our study underscores the importance of a number of methodological and analytical issues that need to be addressed, especially in critical care studies. One is the importance of sensitivity analyses. Our IPTW analysis on the primary outcome was subjected to a sensitivity analysis using PSM -which, in fact, failed to confirm the IPTW results. This discrepancy could be attributed to several factors. The IPTW analysis included more observations compared to the PSM analysis. It yielded a much narrower 95% CI and therefore greater precision of the estimate. It is also possible that the IPTW analysis may have been prone to model misspecification compared to PSM model. However, covariate balance diagnostics showed that all variables in IPTW analysis had adequate balance as indicated by the SMD, yet that the PSM model had less bias for the overall distance.

Further, we accounted for immortal time bias by counting the person-days where no treatment was received in the tocilizumab group toward the untreated group using Cox time dependent covariance. Here, neither the IPTW nor the PSM analyses found a difference in the overall mortality. Immortal bias is an analytical procedure common in observational studies [65] . It refers to a period of time in which patients must be alive to receive the treatment. If our study were a randomized control trial, the time of tocilizumab administration (i.e. treatment allocation) should be aligned with the mechanical ventilation (day 0) to avoid misclassification of the exposure. However, in observational studies, patients may start their treatment at different time points after day 0. Consider, for instance, the scenario where time 0 for the tocilizumab group is when treatment is started and time 0 for the control group is the time of starting mechanical ventilation. Excluding immortal time induces selection bias in that person days that should have been counted as untreated are excluded, shortening the person days for the untreated, and thus biasing the estimate. To avoid the ensuing misclassification of the exposure, treatment should be considered a time-dependent rather than a fixed-time covariate to ensure that all person days will be correctly counted toward the treated or untreated arms [54] . Our decision to control for immortal time bias is underscored by a study by Shinati et al. in mechanically ventilated ICU patients. Fixed time analysis yielded a strong association of ICU length stay and delirium but this association was not found when considering delirium as a time dependent covariate.

A competing risk analysis was conducted for our secondary extubation outcome. Koller and colleagues reviewed 50 publications in high-impact journals and reported that 70% of these publications were susceptible to competing risk problems [66] . Competing risks occur when the occurrence of an event may prevent the occurrence of the primary event of interest [38] . In critical care settings, the extubation endpoint, while an important marker of patient improvement, may also be affected by rapidly changing changes in status due to new events occurring that are unrelated to the treatments of interest. Further, there is a common misconception that hazard ratios derived from competing risk models convey the same information (Fine-Gray Model versus Cause-specific hazard (CSH) model); while, in fact, this is not the case. The CSH model evaluates a binary event of interest over time, considers all other events as non-relevant, therefore censors these events, and thus violates the assumption of noninformative censoring. When there are no competing risks, the survival function can be linked directly with the hazard function and the CIF can be calculated as 'one minus survival function.' However, in the case of competing risks, there is no direct one-to-one link between survival and hazard function. Therefore, a naïve KM estimator will be larger than an estimator that accounts for other competing risks. To estimate the impact of covariate of interest on CIF in the presence of competing risks, the use of Fine-Gray models is recommended [38, 67] . In our CSH analysis, the extubation rate was much higher in the tocilizumab arm. Likewise, the probability of event occurrence was also higher in our Fine-Gray model. However, the event of death on mechanical ventilation had a statistical non-significant CSHR result but a statistically significant SHR result. One interpretation, though an erroneous one, would be that patients treated with tocilizumab treatment have a lower probability of death while on mechanical ventilation. There was indeed a seemingly greater impact of extubation events on the CIF, which in turn had an indirect impact on the CIF for the competing risk outcome of death on mechanical ventilation. This made it appear that tocilizumab had an exclusive protective effect against death while on mechanical ventilation (MV) -when, in fact, extubated patients cannot experience death while on MV. That is why Latouche and colleagues recommend to report both CSH and Fine-Gray models side by side for competing risk analysis. This allows transparency and facilitates interpretation of the results [40] . One caveat for using Fine-Gray models is that statistical methods accounting for time varying covariance are not well established yet and need further research [68] .

Our study has some limitations. It was observational and may have included potential confounders. Hence our decision to apply two propensity score methods to control for bias reduced unmeasured confounding [69] . Although we included many variables in our propensity score models, variables that predict ICU mortality such as the Sequential Organ Failure Assessment (SOFA) or the Acute Physiologic Assessment and Chronic Health Evaluation II (APACHEII) scores were not available in or calculable from the medical records. Having these aggregate metrics available, as opposed to using the individual components of these scores (in as far as available, as we did), places the propensity score models at risk of overfitting. The competing risk analysis did not consider immortal time bias as the inclusion of time-dependent variables in Fine-Gray models is less established and is an area of ongoing statistical research. The recent evidence suggests that clinical improvement or reduction in mortality following exposure to tocilizumab could be seen in patients with elevated IL-6 levels [70, 71] . In our institutions, IL-6 levels were not readily available; however, the following biomarkers were measured: ferritin, c-reactive protein (CRP), and D-Dimer (Table 1) . Importantly, CRP levels are predictive of IL-6 mediated disease severity [72, 73] , hence absence of IL-6 levels should not adversely impact the findings of our study. As shown in figure S4 subgroup analysis, we found no interaction in patients with CRP (mg/dL) ≥75 or <75. Additionally, and per MOH COVID-19 treatment guidelines, one or more of the aforementioned biomarkers could be used to predict cytokine syndrome and thus indication for tocilizumab [74] .

This current study suggests that tocilizumab was not effective in reducing mortality. After accounting for immortal time bias, there were no differences between IPTW and PSM in terms of the absolute effect. However, extubation rate while on mechanical ventilation was higher among tocilizumab-treated patients, indicating a possible benefit in this patient population. A randomized controlled trial evaluating tocilizumab against standard of care in critically ill mechanically ventilated patients is needed.

Clinical characteristics of coronavirus disease 2019 in China

COVID-19) situation report -66 [Internet]. World Health Organization

World Health Organization: WHO coronavirus disease (COVID-19) dashboard

World Health Organization

COVID-19 fatality rates in hospitalized patients: systematic review and meta-analysis

Clinical course and outcomes of critically ill patients with SARS-CoV-2 pneumonia in Wuhan, China: a single-centered, retrospective, observational study

Systematic review and meta-analysis of tocilizumab in persons

Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan, China

COVID-19): a global pandemic

Pathogenic human coronavirus infections: causes and consequences of cytokine storm and immunopathology

Complex immune dysregulation in COVID-19 patients with severe respiratory failure

The many faces of the anti-COVID immune response

A cluster of health care workers with COVID-19 pneumonia caused by SARS-CoV-2

Effective treatment of severe COVID-19 patients with tocilizumab

Tocilizumab for severe COVID-19: a systematic review and meta-analysis

Efficacy and safety of tocilizumab in COVID-19 patients: a living systematic review and meta-analysis

Efficacy of tocilizumab in COVID-19: a systematic review and meta-analysis

Decreased mortality in coronavirus disease 2019 patients treated with Tocilizumab: a rapid systematic review and meta-analysis of observational studies

Tocilizumab in patients admitted to hospital with COVID-19 (RECOVERY): a randomised, controlled, open-label, platform trial

Tocilizumab treatment for COVID-19 patients: a systematic review and meta-analysis. Infect Dis Poverty

The strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies

The retrospective chart review: important methodological considerations

Saudi Minestry of Health: Saudi MoH Protocol for supportive care and antiviral treatment of suspected or confirmed COVID-19 infection (Version 3.1)

Saudi Arabia: Saudi Minestry of Health

The REDCap consortium: building an international community of software platform partners

Modeling Survival Data: Extending the Cox Model

Springer link: ggplot2: elegant graphics for data analysis

Survminer: drawing Survival Curves using 'ggplot2

mice: multivariate imputation by chained equations in R

MatchThem: matching and weighting multiply imputed datasets

CrrSC: competing risks regression for stratified and clustered data

Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls

Alternative approaches for confounding adjustment in observational studies using weighting based on the propensity score: a primer for practitioners

Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies

Wiley online library: multiple imputation for nonresponse in surveys

Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies

Survival analysis: techniques for censored and truncated data

Introduction to the analysis of survival data in the presence of competing risks

Propensity-score matching with competing risks in survival analysis

A competing risks analysis should report results on all cause-specific hazards and cumulative incidence functions

Repurposed antiviral drugs for Covid-19 -interim WHO solidarity trial results

Dexamethasone in hospitalized patients with Covid-19

COVID-19) Update: FDA Authorizes Monoclonal Antibodies for Treatment of COVID-19 [Internet]. Food and Drug Administration

Effect of corticosteroids on mortality in hospitalized COVID-19 patients not receiving invasive mechanical ventilation

Tocilizumab in patients hospitalized with Covid-19 Pneumonia

Tocilizumab in hospitalized patients with severe Covid-19 Pneumonia

COVID-19 treatment guidelines: interleukin-6 inhibitors [Internet] National Institutes of Health

Cytokine release syndrome in severe COVID-19. Science (80)

Food and Drug Administration

Impact of tocilizumab administration on mortality in severe COVID-19. Sci Rep

Tocilizumab plus standard care versus standard care in patients in India with moderate to severe COVID-19-associated cytokine release syndrome (COVINTOC): an open-label, multicentre, randomised, controlled, phase 3 trial

Tocilizumab in the treatment of critical COVID-19 pneumonia: a retrospective cohort study of mechanically ventilated patients

Tocilizumab for treatment of mechanically ventilated patients with COVID-19. medRxiv

Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes

Tocilizumab among patients with COVID-19 in the intensive care unit: a multicentre observational study

Tocilizumab therapy in SARS-CoV-2 pneumonia: a matched retrospective cohort analysis. Med Clin (Barc)

Association between administration of IL-6 antagonists and mortality among patients hospitalized for COVID-19: a meta-analysis

Effect of Tocilizumab in hospitalized patients with severe COVID-19 Pneumonia: a case-control cohort study

Impact of low dose tocilizumab on mortality rate in patients with COVID-19 related pneumonia

Tocilizumab in patients with severe COVID-19: a retrospective cohort study

Efficacy of tocilizumab in patients with COVID-19 ARDS undergoing noninvasive ventilation. Crit Care

Early Tocilizumab dosing is associated with improved survival in critically Ill patients infected with severe acute respiratory syndrome coronavirus-2

Propensity score-based methods in comparative effectiveness research on coronary artery disease

Variable selection for propensity score models

Time-dependent bias was common in survival analyses published in leading clinical journals

Competing risks and the clinical community: irrelevance or ignorance? Stat Med

Survival analysis in the presence of competing risks

Simulation shows undesirable results for competing risks analysis with time-dependent covariates for clinical outcomes

The central role of the propensity score in observational studies for causal effects

Tocilizumab improves the prognosis of COVID-19 in patients with high IL-6

Pulmonary vascular improvement in severe COVID-19 patients treated with tocilizumab

Prognostic value of interleukin-6, C-reactive protein, and procalcitonin in patients with COVID-19

Elevated levels of IL-6 and CRP predict the need for mechanical ventilation in COVID-19

Saudi Minestry of Health: Saudi MoH protocol for patients suspected of/confirmed with COVID-19; supportive care and antiviral treatment of suspected or confirmed COVID-19 infection

Saudi Arabia: Saudi Minestry of Health

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.