This large-scale multicenter study on 9,638 hospitalized patients with severe COVID-19 was designed to evaluate lessons learned from empirical treatment modalities with five prevalent EUA-approved investigational drugs during the first two surges of the pandemic in 2020. The results show somewhat inconclusive data pertinent to the impact of these medications on patient outcomes during the “early” (March 1-June 30) and “late” (July 1 – December 31) surges of the 2020 pandemic. In contrast to the variable effect of these medications, the necessity for oxygen supply was associated with significantly increased odds of mortality beyond the effect of the five selected EUA-approved medications (Fig. 2). The requirement for any delivery modality for supplemental oxygen, reflective of patients who develop respiratory distress, was associated with a significantly increased odds of mortality by a factor of 4.67 in the “Early 2020” cohort, and 4.0 in the “Late 2020” cohort. Non-invasive ventilation using BiPAP was associated with a decreased risk of mortality in patients who required supplemental oxygen. In contrast, invasive mechanical ventilation was associated with the highest odds of mortality in both study time-periods, with dramatically increased odds of 8.34 in the “Early 2020” cohort, and 9.46 in the “Late 2020” cohort. These findings reflect on the early experience in the acute management of COVID-19 patients with acute respiratory failure which demonstrated that the early endotracheal intubation with mechanical ventilation was associated with increased mortality, leading to more judicial indications of invasive versus non-invasive ventilation strategies due to the detrimental lessons learned during the early phase of the pandemic, when patients in respiratory distress were more liberally intubated for mechanical ventilation (36, 37).
The five representative EUA-approved medications investigated in this study had less of an impact on mortality compared to the necessity for and modality of oxygen delivery (Fig. 2). Nevertheless, it was interesting to observe differences in mortality odds ratios between the individual medications and the two different time-windows of the pandemic surges in 2020. Remdesivir treatment did not show a significant association to mortality in surge 1, however, was linked to a significant decrease in mortality in surge 2. This may be due to the more widespread access to the antiviral drug, as suggested by the increase in the number of remdesivir administration from 834 in “Early 2020” to 3,290 in “Late 2020”. Azithromycin has seen a widespread anecdotal application during the early phase of the pandemic in spite of a lack of stringent favorable evidence in published systematic reviews and meta-analyses (23). Thus, its indication for the treatment of hospitalized patients with COVID-19 remains controversial and unsubstantiated. In the current study, about half of our patients received azithromycin which was associated with decreased mortality during both time-windows in 2020. The anti-malaria drug hydroxychloroquine has received early global attention as a potential medication in COVID-19 due to promising preliminary insights from small scale studies (26). However, the anecdotal “hype” around this medication was never supported by strong clinical evidence from better quality trials (26). In our study, the number of patients treated with hydroxychloroquine decreased dramatically between the first and second surges, from 36% down to only 1.28%. This observation likely reflects the lack of efficacy and risk of significant side effects by the drug which became more widely understood after the first wave of the pandemic (24, 38). In support of this notion, our study showed no relevant benefit in mortality between the two surges, with a large confidence interval for hydroxychloroquine in surge 2, likely attributed to the low sample size in the “Late 2020” cohort. Our data showed an overall increase in corticosteroid use. During the first surge, about 71% of patients admitted to the ICU were given corticosteroids, compared to 97% of hospitalized COVID-19 patients during the 2nd surge. The widespread administration of corticosteroids to nearly 100% of all patients makes trends on mortality difficult to interpret, as both survivors and non-survivors would have been treated with steroids. The controlled open-label RECOVERY trial investigated different modalities and dose regimens for dexamethasone administration in hospitalized COVID-19 patients, and demonstrated a decreased mortality in patients receiving invasive ventilation at 28 days (39). In this landmark trial on 6,425 patients, corticosteroids were thought to reduce inflammatory damage to lungs and reduce progression to respiratory failure and death (39). Subsequent to the publication of the RECOVERY trial, there has been widespread global use of corticosteroids for hospitalized COVID-19 patients with signs of respiratory distress. Our study was unable to determine a significant and sustained protective effect of corticosteroids during the two pandemic surges, likely attributed to the shortcoming discussed above. Similar to steroids, tocilizumab has strong anti-inflammatory properties as a neutralizing monoclonal antibody against the human IL-6 receptor. Early studies during the first surge of the pandemic showed a benefit of tocilizumab by reducing mortality in critically ill COVID-19 patients suffering from hyperinflammation and respiratory failure (40, 41). However, these beneficial effects postulated for tocilizumab were refuted in subsequent large-scale studies which failed to demonstrate the efficacy of tocilizumab in reducing mortality in critically ill patients with COVID-19 (42, 43). In our study, patients that required invasive mechanical ventilation had a higher likelihood of being treated with tocilizumab than those who did not require invasive ventilation. This phenomenon is likely due to the widespread notion during the first surge that tocilizumab was beneficial for critically ill COVID-19 patients with systemic hyperinflammation in terms of reducing ventilator-dependent days and improving patient outcomes. Later publications demonstrated that the lack of efficacy by tocilizumab is likely due to an underrecognized anti-inflammatory effect of IL-6 (44). Our current understanding of the immune pathophysiology suggests that the blockade of the IL-6 receptor will lead to a redundant massive increase of other pro-inflammatory cytokines which will subsequently exacerbate the “cytokine release syndrome” (CRS) and worsen the extent of harmful systemic hyperinflammation (44). Our current data demonstrate a decrease in the number of patients receiving tocilizumab between the first and second surge. This observation is likely reflective of the change in evidence and clinical recommendations in the peer-reviewed literature around the time of the second surge of the pandemic (45, 46).
Overall, we interpret the variability in beneficial effects of the five prevalent medications investigated in this study during the first and second surge by the notion that each individual medication likely had a lesser impact on the pathophysiology of COVID-19 than other confounding variables. Most importantly, the requirement for supplemental oxygen in patients with respiratory distress and the early intubation and mechanical ventilation in patients with respiratory failure appears to represent the highest independent predictor of mortality during both pandemic surges in 2020 (Fig. 2). This insight had changed the management of oxygen delivery strategies dramatically due to the lessons learned from the first surge of the pandemic, from early proactive airway management by endotracheal intubation and mechanical ventilation, towards leveraging safer non-invasive ventilatory strategies (36, 37, 47).
Another confounding factor in our current study is represented by the observation that most patients received at least two COVID-19 medications simultaneously, which decreases the ability to determine the impact of a single medication with a higher level of statistical certainty. The large sample size of 9,638 patients ensured the equal distribution of underlying demographic variables and risk factors for severe COVID-19 and adverse outcomes, including higher age, male gender, obesity, and ethnicity (Table 1). The patient population included in this study also showed a similar distribution of risk factors related to preexisting comorbidities, as determined by the Elixhauser Comorbidity Index (Tables 5&6). Finally, the selected patient cohort managed by temporary organ replacement support for salvage therapy by CRRT (for acute renal failure) or ECMO (for refractory respiratory failure in spite of maximized mechanical ventilation efforts) showed significantly higher odds of mortality during both surges of the pandemic in 2020 (Tables 5&6). Unequivocally, the observation that severely ill patients “in extremis” who require last-effort rescue strategies by CRRT or ECMO will have a higher predicted mortality than the cohort of COVID-19 patients without acute organ failure, represents a predictable platitude. Nevertheless, it is worth emphasizing that both sub-cohorts of CRRT and ECMO patients had significantly decreased odds of dying compared to patients with invasive mechanical ventilation in absence of these rescue modalities, with odds ratios of 3.65 and 2.53 for CRRT; 3.16 and 1.63 for ECMO; versus 8.34 and 9.46 for invasive ventilation, during the “Early 2020” and “Late 2020” study time windows, respectively (Tables 5&6). For example, the protocolized adherence to the guidelines by the “Extracorporeal Life Support Organization” (ELSO) for stringent patient selection for ECMO cannulation was associated with significantly improved outcomes from venous-venous ECMO salvage therapy in critically ill COVID-19 patients (48, 49). These findings imply a judicious and standardized patient selection process for COVID-19 rescue therapies across the 186 hospitals and thereby emphasize the benefits of protocolized care “at scale” across a large healthcare system in the United States.
There are several methodological shortcomings and limitations to this study. First, the definition applied to the “early” and “late” time-windows of the pandemic is arbitrary and prone to potential misinterpretation of the data. While the unequal time-windows of 4 months (“Early 2020”) vs. 6 months (“Late 2020”) had been intentionally selected to reflect on the approximate timing of the first two main COVID-19 surges in the United States, this pragmatic selection does not account for differences in geographical “tiers” as the pandemic evolved in the United States. For example, participating hospitals in Florida will have seen different patient volumes and acuity during the “early” time-window of this study compared to participating hospitals in California, Texas, or Colorado. Conversely, southeastern states, such as Florida, were more dramatically affected from surging COVID-19 patient volumes during the second wave of the 2020 pandemic. Finally, the unequal arbitrary stratification of the time-windows into 4 vs. 6 months does not allow for a direct comparison of respective patient volumes and treatment numbers. Furthermore, our data source from the 87,788 patients included in the multicenter data registry did not allow to retrieve specific information pertaining to convalescent SARS-CoV-2 plasma due to the incorporation of this treatment modality within the overall fresh frozen plasma transfusion numbers in the blood bank data. Thus, we were unable to attribute any potential survival benefit in this patient cohort to the transfusion of convalescent plasma. Last but not least, this study suffers from the classic flaws of retrospective analysis of “large data” registries. In general, large population-based databases provide high statistical power to determine effect estimates over smaller single center registries, however, at the shortcoming of being unable to retrieve granular level data that would allow to investigate a direct causative relation between treatment and outcomes. Although the methodology applied propensity matching to make the medication groups more comparable, we cannot rule out that differences in mortality between medication groups were due to other confounding variables, since patients were not randomly assigned to medication treatment groups. Nevertheless, this retrospective multicenter cohort study allows to provide a high-level “lesson learned” on the efficacy of investigational off-label medications and on the impact of oxygen delivery modalities on the mortality and adverse outcome during the early wave of the COVID-19 pandemic in the United States.