Prognostic Factors and Clinical Prediction Score for Progressive Respiratory Failure in Severe COVID-19 Pneumonia Patients Treated with Tocilizumab: A Multicenter Study

Introduction

Hypoxic respiratory failure in patients with COVID-19 is associated with an excessive systemic inflammatory response. Interleukin-6 (IL-6) is a key driver of this hyperinflammatory process, contributing to endothelial dysfunction, impaired oxygen exchange, and progression to respiratory failure. Tocilizumab, an IL-6 receptor blocker, is approved to treat several autoimmune inflammatory diseases. According to the RECOVERY trial,¹ COVID-19 patients with hypoxia and systemic inflammation treated with Tocilizumab were less likely to reach the composite endpoint of mechanical ventilation or death. A meta-analysis of randomized clinical trials (RCTs) confirmed that Tocilizumab decreased the risk of mechanical ventilation.² Current World Health Organization (WHO) guideline recommends treatment with IL-6 receptor blockers in patients with severe and critical COVID-19.³

Despite treatment with Tocilizumab, a portion of patients continue to have disease progression. The reported absolute risk of progression to invasive mechanical ventilation or death was 27% in RCTs.⁴ For endotracheal intubation, the reported risk was 6.2–15.4% in RCTs and 0.0–33.3% in cohort studies.² Several studies have attempted to identify prognostic markers among patients treated with tocilizumab; however, findings have been heterogeneous. De Biasi et al demonstrated that specific immune-cell signatures could differentiate responders from non-responders to IL-6 inhibition, while Masotti et al and Pagkratis et al reported clinical and biochemical predictors associated with poor outcomes in tocilizumab-treated patients.^5–7 Nevertheless, these studies primarily focused on immunologic or biochemical markers rather than simple bedside clinical parameters.

Medical early warning scores, such as the National Early Warning Score 2 (NEWS2), NEWS2 adding age as a variable, and COVID-19 severity index, are sensitive methods for predicting short-term deterioration of patients with COVID-19 in general.^8–10 However, their performance in patients treated with tocilizumab remains uncertain.^11,12 Recent studies in patients treated with tocilizumab have focused on immune biomarkers and early-warning constructs, but dedicated prognostic tools for this subgroup remain scarce.^13,14

We conducted this retrospective cohort study to identify prognostic factors associated with progressive respiratory failure requiring mechanical ventilation in severe COVID-19 pneumonia patients treated with Tocilizumab and to characterize their clinical outcomes. We also proposed a prediction model to stratify the risk of progressive respiratory failure in these patients.

Materials and Methods

Study Design and Participants

We conducted a multicenter retrospective study from June 2022 to February 2023. Recruited patients were admitted to Chakri Naruebodindra Medical Institute or Ramathibodi Hospital from April 2021 to May 2022. Both hospitals are affiliated with the Faculty of Medicine Ramathibodi Hospital, Mahidol University, Thailand, and share similar COVID-19 treatment protocols. To minimize selection bias, all consecutive eligible patients during the study period were included. Both hospitals followed identical institutional treatment protocols based on national and RECOVERY-aligned guidelines. Inclusion criteria included age > 18 years, diagnosis of COVID-19 based on reverse transcriptase polymerase chain reaction (RT-PCR), and severe COVID-19 pneumonia (new infiltration on chest radiograph without other explainable causes and oxygen saturation <92% on room air or signs of severe respiratory distress) treated with Tocilizumab in hospital ward or intensive care unit settings. Exclusion criteria were intubation before tocilizumab administration and terminal illness receiving palliative care.

Indication for tocilizumab administration was in accordance with the RECOVERY trial:¹ patients with hypoxia (oxygen saturation <92% on room air or requiring oxygen therapy) and evidence of systemic inflammation (elevated C-reactive protein or IL-6).

Data Collection and Study Endpoint

We reviewed electronic medical records and collected data for the recruited patients from admission day until discharge. On tocilizumab administration day, baseline data, including demographic, clinical, laboratory data, and treatment of COVID-19, were obtained. Patients’ clinicals were classified through the WHO ordinal clinical severity scale.¹⁵ The primary endpoint was progressive respiratory failure, defined as the need for invasive mechanical ventilation within 14 days after tocilizumab administration. Escalation to high-flow nasal cannula or noninvasive ventilation was not considered an endpoint. For patients with do-not-intubate (DNI) status, death attributable to respiratory failure within 14 days was classified as an event. The other clinical outcomes after tocilizumab administration, including acute kidney injury, venous thromboembolism, cardiovascular complications, infection, inotropic drug use, length of hospital stay, and death, were determined at discharge. Serious infections were defined as infections requiring intravenous antibiotics, prolonging hospitalization, or resulting in death. All secondary outcomes were adjudicated through detailed chart review by attending physicians and independently cross-checked by two investigators to ensure accuracy and consistency. Missing laboratory values were not imputed. Outliers were verified against original medical records to ensure accuracy.

Candidate Predictors for Progressive Respiratory Failure

All variables collected on tocilizumab administration day were explored for their statistical significance and were considered candidate predictors for a clinical prediction score. However, these variables were prespecified based on clinical plausibility and prior literature, and univariate screening was performed only to identify candidates for multivariable modeling to reduce the risk of data-driven overfitting. For continuous data, the optimal cut-off values for predicting were determined by calculating the area under the receiver operative characteristic curve (AuROC) and the Youden index. Because these cut-offs were derived from the same dataset used for model development, there may be potential information leakage, and this limitation has been acknowledged.

Sample Size and Statistical Analysis

Because the outcome of interest (progressive respiratory failure) occurred infrequently, a formal sample size calculation was not feasible. Therefore, all consecutive eligible patients during the study period were included to maximize statistical power and minimize selection bias. Descriptive statistics were used to present the demographic, clinical, laboratory data, and treatment variables. Categorical data were compared using the Chi-Square test or Fischer’s exact test as appropriate. To compare normally distributed and non-normally distributed continuous data, the independent t-test and Mann–Whitney test were applied, respectively. Continuous variables were dichotomized before modeling to facilitate bedside clinical applicability, despite the acknowledged potential reduction in statistical power. We selected variables for a multivariable Cox proportional hazards model using forward stepwise selection. The proportional hazards assumption was tested using Schoenfeld residuals and was not violated. Then, we identified prognostic factors for progressive respiratory failure from significant clinical factors.

The clinical prediction score for progressive respiratory failure was developed using significant clinical factors from multivariable Cox regression analysis. With 32 events and three variables in the final model, the events-per-variable ratio remained above the recommended minimum of 10. Score performance was evaluated in terms of discrimination using AuROC and calibration using a calibration plot. Internal validation was performed using bootstrap resampling with 500 replicates. For each replicate, the model was refitted in the bootstrap sample and evaluated in the original dataset to estimate optimism. The apparent AUC of the final model was 0.715 (range 0.533, 0.846). The optimism-corrected AUC was calculated as the apparent AUC minus the average optimism across bootstrap samples. The final optimism-corrected AUC was 0.72. P value < 0.05 was considered statistically significant. Statistical analyses were performed using SPSS Statistics version 22 (IBM, Armonk, NY, USA) and Stata version 17.0 (StataCorp LP, College Station, TX, USA).

Ethical Considerations

This study was approved by the Human Research Ethics Committee, Faculty of Medicine Ramathibodi Hospital, Mahidol University, Thailand (COA. No. MURA2021/1062). The informed consent requirement was waived by the Ethics Committee because of the retrospective design of the study. This study complied with the Declaration of Helsinki, and all patient data were anonymized before analysis to ensure confidentiality.

Results

A total of 109 eligible patients were recruited into the study: 54 patients from Chakri Naruebodindra Medical Institute, 41 patients from Ramathibodi Hospital, and 14 patients received initial treatment from Ramathibodi Hospital and then transferred to Chakri Naruebodindra Medical Institute. A flow diagram summarizing patient screening and eligibility is shown in Figure 1. The mean ± standard deviation (SD) age of the participants was 60.3 ± 14.2 years. There was a slight male predominance (51.4%). Body mass index (BMI) was high among participants, with a median (interquartile range, IQR) of 26.8 (24.0–31.2) kg/m². The majority of patients (79.5%) did not receive COVID-19 immunization (Table 1). A single dose of Tocilizumab was administered intravenously to 86.2% of the patients; the remaining individuals received two infusions. Tocilizumab was given in doses ranging from 400 to 800 mg (7.6 ± 1.2 mg/kg/dose). The median COVID-19 illness duration (number of days from symptom onset to the administration of Tocilizumab) was nine days (IQR 7–11 days). Of the 109 patients, 32 (29.4%) progressed to respiratory failure requiring mechanical ventilation within 14 days after tocilizumab administration. The median time to respiratory failure was two days after tocilizumab administration (range 1–12 days) (Figure 2). The median length of hospital stay was 17 days (range 3–92 days).

Table 1 Demographic Data, Clinical Settings, Laboratory Features, and Previous Treatment of Severe COVID-19 Pneumonia Patients on Tocilizumab Administration Day

Figure 1 Flow diagram of patient selection for the study.

Figure 2 Probability of progressive respiratory failure after tocilizumab administration comparison of demographic data, clinical settings, laboratory data, and previous treatment.

Comparison of Demographic Data, Clinical Settings, Laboratory Data, and Previous Treatment

Older age (P = 0.034), receiving Tocilizumab in early of the disease course (P = 0.003), low pulse oximetry saturation to fraction of inspired oxygen ratio (SpO₂/FiO₂) (P = 0.009), high serum sodium (P = 0.036), and low serum potassium (P = 0.046) on tocilizumab administration day were associated with progression to respiratory failure requiring mechanical ventilation in severe COVID-19 pneumonia patients (Table S1). Vaccination status was evaluated in the univariate analysis but was not retained in the final multivariable model because it was not significantly associated with progressive respiratory failure. Complete univariate comparisons of baseline demographic, clinical, and laboratory variables between patients with and without progressive respiratory failure are presented in Supplementary Table 1.

Independent Predictors for Progressive Respiratory Failure

We performed multivariable Cox regression analyses to identify the potential poor prognostic factors for progressive respiratory failure. After adjusting for age and sex, the independent predictors of progressive respiratory failure were SpO₂/FiO₂ ratio ≤ 160 (HR 2.972, 95% CI 1.417–6.231, P=0.004), eGFR < 60 mL/min/1.73m² (HR 3.214, 95% CI 1.232–8.388, P= 0.017), and serum potassium ≤ 4.0 mmol/L (HR 2.824, 95% CI 1.375–5.797, P=0.005) (Table 2). Although older age and earlier administration of tocilizumab were significant in univariate analyses, these variables did not remain significant after multivariable adjustment.

Table 2 The Predictors of Progressive Respiratory Failure in Severe COVID-19 Pneumonia Patients Treated with Tocilizumab

Prediction Model for Progressive Respiratory Failure

The final prediction model for progressive respiratory failure in severe COVID-19 pneumonia treated with Tocilizumab was based on three independent prognostic factors, SpO2/FiO2 ratio ≤ 160, eGFR < 60 mL/min/1.73m², and serum potassium ≤ 4 mmol/L. Each independent predictor in the multivariable model was assigned a specific score derived from the multivariate Cox’s proportional hazard regression coefficient. The regression coefficient of each item was divided by the lowest coefficient and then rounded up to the nearest integer to develop a scoring system. The prediction score was defined as the number of prognostic factors for progressive respiratory failure (score of +1 for each). The scoring scheme had a total score ranging from 0 to 3. The AUC for the score was 0.72 (95% CI 0.63–0.80) (Figure 3), demonstrating good discriminative ability. Since we intended to use the score to screen patients at high risk of progressive respiratory failure, we chose the score cut-off point of 1 (sensitivity of 93.8% and specificity of 35.1%). The sensitivity and specificity for each cut-off point are depicted in the Table S2. All mortality events occurred among patients who developed progressive respiratory failure. Higher risk-score categories showed a clear increase in the likelihood of respiratory failure, supporting the score’s clinical relevance; however, mortality was not analyzed separately by score category due to small subgroup sizes.

Figure 3 Discriminative ability based on the area under the receiver operative characteristic curve (AuROC) for the prediction model (A) and model calibration (B).

Clinical Outcomes After Tocilizumab Administration

The overall mortality of severe COVID-19 pneumonia patients treated with Tocilizumab was 18.3% (20 patients). Causes of death were COVID-19 pneumonia with multi-organ failure (12 patients), secondary infection (7 patients), and pulmonary embolism (1 patient). Intubated patients were likely to have poor clinical outcomes, including acute kidney injury, venous thromboembolism, cardiovascular complications, infection, inotropic drug use during admission, and death (Table 3). A total of 48 secondary infections occurred among 32 patients (29.4%). Bacterial pneumonia was the most common infection, followed by urinary tract infection, primary bacteremia, cytomegalovirus infection, and invasive pulmonary aspergillosis. Serious infections occurred in 29 patients (26.6%).

Table 3 Clinical Outcomes

Discussion

Our retrospective cohort focuses on prognostic factors of progressive respiratory failure in severe COVID-19 pneumonia treated with Tocilizumab from April 2021 to May 2022. The patients received standard doses of Tocilizumab at the median duration of 9 days since COVID-19 symptom onset, similar to the individuals from the RECOVERY trial.¹ The rates of respiratory failure and mortality observed in our cohort were comparable to those reported in previous studies.^2,4 We found that SpO₂/FiO₂ ratio ≤ 160, eGFR < 60 mL/min/1.73m², and serum potassium ≤ 4.0 mmol/L were independent predictors of respiratory failure.

The SpO₂/FiO₂ ratio correlates well with PaO₂/FiO₂ and is practical for risk assessment, especially where arterial blood gas measurements are not readily available. A SpO₂/FiO₂ ratio of 235 corresponds with a PaO₂/FiO₂ of 200, utilized to define acute respiratory distress syndrome.¹⁶ In hospitalized COVID-19 patients, Kim et al¹⁷ found that the SpO₂/FiO₂ ratio predicted high-flow nasal cannula (HFNC) failure with subsequent respiratory failure. The median SpO₂/FiO₂ ratio was 160 (120–186) in the HFNC failure group, the same number as the cut-off from the ROC curve analysis in the current study. Lu et al¹⁸ reported a strong association between a sharply decreased SpO₂/FiO₂ ratio and mortality risk of severe and critically ill patients with COVID-19. The SpO₂/FiO₂ ratio threshold ≤ 160 determined in this study could be used to identify severe COVID-19 pneumonia patients treated with Tocilizumab at risk of respiratory failure.

Chen et al¹⁹ demonstrated that renal insufficiency (eGFR < 60 mL/min/1.73m²) on admission independently predicted poor prognosis (composite endpoint of intensive care unit admission, invasive ventilation, or death) among COVID-19 inpatients. Renal dysfunction has been consistently associated with worse outcomes in COVID-19.^20,21 Our study showed that eGFR < 60 mL/min/1.73m² was an independent predictor for respiratory failure in severe COVID-19 pneumonia treated with Tocilizumab, supporting the previous findings.

Hypokalemia is a frequent electrolyte abnormality in COVID-19 and has been associated with greater disease severity in previous studies.^22–24 Proposed mechanisms include RAAS dysregulation and physiologic stress responses triggered by SARS-CoV-2 infection.²⁵ In our cohort, serum potassium ≤4.0 mmol/L was independently associated with progressive respiratory failure. However, this association should not be interpreted as causal; hypokalemia may reflect underlying neurohormonal activation, systemic stress, or illness severity rather than directly contributing to respiratory deterioration.²⁶ Given its potential cardiac implications and its common occurrence in severe COVID-19, maintaining serum potassium above 4.0 mmol/L remains clinically prudent.²⁷

Masotti et al⁶ reported prognostic factors of poor outcomes (combination of mortality and/or intensive care unit admission with endotracheal intubation) in tocilizumab-treated patients with severe respiratory failure from COVID-19: age ≥ 65 years, procalcitonin ≥ 0.14 ng/mL, room-air pulse oximetry oxygen saturation ≤ 90% and chest computed tomography (CT) ground glass opacity ≥ 50% at admission. For prognostic factors specific to progressive respiratory failure, Pagkratis et al⁷ reported PaO₂/FiO₂ ratio at admission and lactate dehydrogenase on the day of tocilizumab treatment as predictors of respiratory failure. Overall, these prior reports align with our observations and reinforce the relevance of early physiologic and biochemical indicators in patients receiving tocilizumab.

Mussini et al²⁸ developed a score to estimate the risk of progressing to mechanical ventilation and death in tocilizumab-treated COVID-19 patients. Female gender, day-4 C-reactive protein value above the median, day-4 platelet value above the median, and day-4 PaO₂/FiO₂ ratio value above the median were used to generate the risk score. The accuracy of the score was very good (AUC 0.8). However, this score focused on composite outcomes and required median laboratory data values. In the current study, we proposed a new risk prediction model for respiratory failure in severe COVID-19 pneumonia treated with Tocilizumab. This score relies on readily available bedside parameters and may help flag patients who require intensified monitoring after tocilizumab. Higher scores can guide clinicians toward earlier reassessment or escalation of care, although the model is intended to complement clinical judgment.

This study has several limitations. First, the small sample size and retrospective design introduce the possibility of information bias and limit the robustness of the findings. Missing data on COVID-19 vaccination status and serum IL-6 levels, as well as the unavailability of PaO₂/FiO₂ measurements and chest CT severity scoring, constrained the comprehensiveness of our analyses. Second, selection bias may be present because only patients who received tocilizumab were included, and confounding by indication may have occurred if patients selected for tocilizumab differed systematically from those who did not receive it. Third, the prediction model has not undergone external validation, which restricts its generalizability. Finally, although the lower cutoff point enhances sensitivity, its limited specificity reduces the model’s usefulness as a standalone triage tool. Future studies with larger cohorts and external validation are needed to confirm these findings and refine the predictive score.

Conclusion

In this retrospective cohort of patients with severe COVID-19 pneumonia treated with tocilizumab, independent predictors of respiratory failure were a SpO₂/FiO₂ ratio ≤160, an eGFR <60 mL/min/1.73 m², and a serum potassium ≤4.0 mmol/L. These readily available parameters may help clinicians identify patients at higher risk of progressive respiratory failure and support decisions regarding monitoring intensity and ICU planning. We also proposed a simple, clinically applicable prediction score based on these variables; however, its current use should be considered preliminary until it has been validated in independent cohorts. Future research should focus on validating and refining this score, as well as on comparing its performance with that of existing early warning tools in patients receiving IL-6 receptor blockade.