Prediction of Cytokine Storm and Mortality in Patients with COVID-19 Admitted to ICU: Do Markers Tell the Story?

COVID-19 has caused a million deaths as of November 24, 2020, worldwide [1]. COVID-19 (SARS CoV-2) spread rapidly via person to person contact and infect predominantly the pulmonary system [2]. Approximately 15% developed respiratory failure from ARDS, requiring mechanical ventilation and admission to ICU [3]. Progression to ARDS is unpredictable as the disease activity and intensity of inflammation is variable [4]. Therapy timing of anti-inflammatory agents and immune-suppressing medication is of utmost importance. This inflammation is dynamic in COVID-19 and most people get anti-inflammatory treatment, either from the beginning as a standard protocol or with the inflammatory markers monitoring [5]. Optimal administration may prevent cytokine storms and abate the development of ARDS [6]. Markers of disease activity like ferritin, C-reactive protein (CRP), and D-dimers are frequently monitored to detect the best opportunity for intensive treatment [7]. Less commonly levels of interleukin 6 are used to detect and target the most intensive part of the disease phase although this test is expensive and only rarely available [8]. It may not be cost-effective for the developing world. Hence, identification of intense activity of virus and inflammation is of utmost importance. Viral load monitoring is impractical. Inflammatory markers’ effectivity for disease monitoring and success of treatment and their cost-effectiveness is important and largely unknown.

Aims: We aim to measure the pattern of these markers’ usage, their relationship with the disease and with each other, and their prediction of disease activity and utility for treatment success and clinical outcomes. Moreover, we aim to detect the most active phase of the disease if one exists.

All patients with confirmed COVID-19 admitted to ICU of Dubai Hospital between January 1, 2020, and June 30, 2020, were included. We excluded patients <18 years of age. The primary variable of interest, ferritin, CRP, and D-dimers, on admission to ICU and for each patient throughout the hospitalization were also recorded. Primary clinical outcomes of length of stay in the intensive care unit (LOSICU) and survival were also recorded. Demographics: age, gender, BMI, and nationality. Comorbidities: diabetes, hypertension, coronary artery disease (CAD), renal failure, and outpatient dialysis were recorded. Immune status, smoking status, and alcohol use were also recorded. Inpatient clinical evaluation details including vital signs, fever, tachycardia, blood pressure, hypoxia, use of mechanical ventilation, use of pressers, or dialysis were also recorded. Laboratory parameters for evidence of infection: WBC and bacteremia were also recorded. Therapy on admission: chloroquine, antivirals, and steroids were recorded. The APACHE 2 scores calculated within 24 h of admission to ICU were recorded to assess the severity of illness. Ferritin, CRP, and D-dimers were recorded daily if available for the whole ICU stay, and all other variables were recorded on admission day to ICU.

All the variables were analyzed and found that they were not normally distributed; therefore, median with interquartile ratios was calculated. The average level of the marker was determined by adding all values available and dividing by the number of days values available. χ² test was performed to detect the difference between variables if they were categorical, and median test was performed if the variable was continuous. For survival analysis, multiple logistic regression was performed with survival (mortality) as dependent variables and all other variables as independent predictors.

Since LOSICU is very skewed and the variance of LOSICU across the levels of CRP was heterogeneous, data were transformed into log LOSICU and least square regression was performed for log LOSICU and markers (CRP) only for survivors. We used IBM SPSS Statistics for Windows, version 26 (IBM Corp., Armonk, NY, USA).

Characteristics of the sample N = 235 are described in Tables 1and2. More than 95% of patients have all markers on the day of admission to ICU. Average levels of markers were different from their admission day level which suggest levels were changing. Multiple logistic regression model showed only elevated average CRP during ICU stay-predicted survival (Table 3). Comparing the effects of marker levels on survival, a Kaplan-Meier plot (Fig. 1-6) was constructed for all markers (4th quartile vs. 1st quartile). Only higher level of average CRP level predicts worse survival (Fig. 6). Ferritin and D-dimer do not predict survival differences. For LOSICU prediction, linear regression on log-transformed LOSICU showed that the days that the swab takes to turn negative, bacterial infection, Cr, and ABG PH on the day of admission to ICU predicted LOSICU (Table 4). Specific regression for log LOSICU for CRP in the survivors’ group does not predict LOSICU (Fig. 7, 8).

Inflammatory markers are frequently checked. Ferritin and D-dimer on admission do not predict mortality. Only average CRP level for ICU stay predicts mortality. Using the 4th quartile against the first quartile as a categorical variable, CRP predicts mortality. Ferritin and D-dimers do not. Zhou et al. [9] found that ferritin is significantly elevated in nonsurvivors than survivors, but they did not consider other factors predicting mortality; therefore, their results show the only association of ferritin with mortality. Zhang et al. [10] showed D-dimer on admission greater than 2.0 μg/mL (4-fold increase) predict in-hospital mortality in patients with COVID-19. They also did not adjust for confounding factors by regression analysis. Liu et al. [11] also found that CRP can effectively assess disease severity and predict outcomes in patients with COVID-19. Tan et al. [12] documented the role of CRP as a reliable biomarker for disease activity (not mortality) with “area under the curve” in the receiver operating analysis of 0.87 (95% CI, 0.10–1.00) where values 83 and 91% represent sensitivity and specificity, respectively.

We did not find that comorbid conditions such as CAD, renal failure, or secondary bacterial infections predict mortality. Barman et al. [13] found in their sample that CAD is an independent predictor of mortality. Cheng et al. [14] also found that kidney disease is associated with in-hospital death of patients with COVID-19. Ruan et al. [15] found that secondary infection was associated with high mortality in their 150 patients from Wuhan, China.

Other than ABG PH no factor predicted LOSICU including ferritin, CRP, CAD, and secondary bacterial infection. Moratto et al. [16] found that high ferritin level on admission is associated with prolonged duration of hospitalization. To our knowledge, we are not aware of a study that recorded daily ferritin level and determines the impact of average ferritin level on the length of ICU stay. Moreover, we performed linear regression analysis considering the impact of >20 significant confounding factors. This suggests that finding association in other studies may be from different methodology or from not adjusting for other significant variables.

We identify the following limitations. Small sample size and single-center retrospective study may have provided results, not generalizable to other populations. Our extensive daily record of all markers of the whole sample and on each patient provided the dynamic changes during ICU stay within a sample and for each patient therefore it provided more reliable measurements. Extensive including of confounding factors allowed us to estimate the real and actual impact of markers on the outcome.

Inflammatory markers are elevated in COVID-19 infection. A single level of CRP on admission does not predict outcome although the average CRP level during the stay in ICU predicts survival. Other markers do not predict survival.

Ethical approval was provided by Emirates Institutional Review Board for COVID-19 Research, DSREC/2020/1324/approved on July 13, 2020.

None for all authors.

The authors did not receive any funding.

R.N.: conceived the research idea, proposal writing, data collection, data analysis, and manuscript writing. A.H.: conceived the idea, proposal writing, and review of the final manuscript. N.I.: idea conception and data collection. D.E.: data collection. Z.O., M.S., S.Z., R.A., S.E., C.S., N.A., W.A., M.A., and F.M.: idea conception and data collection. M.H.: data collection and manuscript writing.