The Predictive Potential of Elevated Serum Inflammatory Markers in Determining the Need for Intubation in CoVID-19 Patients

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), and its concomitant clinical disease coronavirus disease 2019 (CoVID-19), is an ongoing pandemic originating in China that has quickly spread across the world. As of late October, over 42 million cases and 1.1 million deaths have been reported worldwide [1]. Similar to previous coronavirus infections, patients infected with this novel coronavirus have a variable clinical course. The severity of the disease can vary widely amongst individuals: Fifty-two per cent to eighty-two per cent of patients requiring hospital admission will require supplemental oxygen, and 24-32% will require intubation [2, 3, 4, 5, 6]. Given the variable and possibly unpredictable clinical course of many CoVID-19 patients and the potential scarcity of intubation resources worldwide, more data are needed to identify those at greater risk for intubation.

Previous studies have found higher median C-reactive protein, lactate dehydrogenase, neutrophil-tolymphocyte ratio and D-dimer among hospitalised patients with severe CoVID-19 than less severe disease [7,8]. Other studies have developed predictive modelling for critical illness as a term encompassing ICU admission, need for intubation, and death on admission [9,10]. Regarding triaging admitted patients, one of the essential aspects of critical illness is the need for intubation.

The study aimed to determine the predictive potential of elevated serum inflammatory markers in determining the need for intubation in patients admitted for CoVID-19, based on admission data. The null hypothesis is that mean serum inflammatory marker levels are not equal for intubated and non-intubated patients.

A retrospective cohort study was undertaken using the records of the first 158 hospitalised patients aged 18 or older who tested positive for SARS-CoV-2 by nasopharyngeal polymerase chain reaction (PCR) at the University of Colorado Hospital in Aurora, Colorado, between March 19, 2020, and April 2, 2020. Data were collected from the electronic medical record and chart abstraction and recorded in REDCap. Patients were excluded if their records showed they had concomitant viral, bacterial, or fungal co-infections at the time of admission (n=0). The primary outcome was the need for intubation. The institution’s criteria for intubation was based on patients who could not maintain >90% oxygenation saturation on maximum settings of low-flow oxygen delivery devices. This assessment was carried out by attending staff of the intensive care unit. During the study period, positive pressure ventilation was not utilised due to initial concerns regarding aerosolisation of SARS-CoV-2. Additional outcomes included the need for vasopressor medication, prone positioning, hospital discharge, and mortality on or before April 6, 2020. The second cohort of one hundred and two patients consecutively admitted after April 2, 2020, with the same inclusion criteria, exclusion criteria, and chart data were used to validate the predictive model. The study was reviewed and approved by the Colorado Multiple Institutional Review Board as exempt research.

Categorical and continuous variables were first compared against intubation with Fisher’s Exact and Kruskal-Wallace tests, respectively. Then, D-dimer and ferritin were log-transformed. Intubation odds ratios (OR) for covariates of interest are presented as unadjusted and adjusted for a priori confounding variables: age, type 2 diabetes mellitus (DM), symptom duration, ethnicity, and body mass index (BMI). Nonlinear relationships of covariates with intubation were considered using generalised additive models via the mgcv R package [11]. (Supplemental online material - Figure A) Time to intubation by C-reactive protein was illustrated and tested via Kaplan-Meier curves and the log-rank test. Multiple imputations by chained equations via the mice R package [12] was used. In addition, missing data were modelled using random forests and imputed 20 times, and results from the primary analyses were computed and pooled across all imputed data sets.

The minimax concave penalty (MCP) regularised logistic regression [13] was employed to build a sparse predictive model using the covariates and confounders. This procedure fits a series of increasingly saturated models, where saturation is varied using MCP’s tuning parameter λ. A final MCP model indicated covariates most predictive of intubation. The MCP model was tuned via 10-fold cross-validation (CV) using the ncvreg R package [14]. The optimal model, selected by minimising CV deviance, is presented as a probability formula predicting intubation. The approximate formula was estimated with an MCP-regularized linear probability model. The receiver operating characteristic (ROC) curve and accompanying area under the curve, sensitivity, and specificity were calculated using CV predictions of our optimally tuned models (logistics and linear approximation). The model fitting for each imputed data set was performed to obtain coefficients and took the median coefficient estimate across all imputed data sets to account for missing data. Finally, single imputation was used to represent the figures and CV estimates. The data from the second cohort of 102 patients were plotted on a ROC curve using the predictive model for validating its accuracy. The Significance level was set at 0.05 for all relevant statistical analyses.

A retrospective cohort study characterised the demographics, comorbidities, presenting symptoms, and clinical findings of the first 158 patients admitted with positive SARS-CoV-2 at a large tertiary care institution. In addition, a variety of risk factors for intubation in these hospitalised patients, including diabetes mellitus, bilateral infiltrate on chest x-ray, and elevated inflammatory markers were identified.

In a predictive model, a high C-reactive protein, lactate dehydrogenase and presence of diabetes mellitus predicted the need for intubation, with high sensitivity and moderate specificity. These findings were replicated in a separate validation cohort, revealing a similarly high sensitivity and moderate specificity, key characteristics for a screening test where a more cautious approach is desired. The initial C-reactive protein on admission was predictive of subsequent risk for intubation.

There is a need to quickly identify and triage high-risk patients with CoVID-19 who are likely to require intubation, both for appropriate patient triage and management of limited resources [15]. The described derived model is easy to apply and utilises commonly available laboratory values that have biologically plausible links to the disease process and clinical history previously shown to be a risk factor for intubation in pneumonia [16]. It provided good sensitivity for risk for intubation when used in a cohort with many similarities, although our validation cohort did require more prone positioning and had more deceased patients despite having more patients reach discharge in the specified time frame. Given the strain on resources CoVID-19 represents, it will be critically important to identify where resources will need to be used as soon as possible. The present findings that three simple measures, high C-reactive protein, lactate dehydrogenase, and presence of diabetes mellitus, can quickly identify those at highest risk for intubation can augment clinical findings in identifying patients that may necessitate intensive care unit admission.

Previous studies have examined the predictive qualities of inflammatory markers in CoVID-19, although to a different degree than this study. In a retrospective study, Barrett et al. (2021), a cohort of 1123 patients with CoVID-19 were analysed for initial laboratory values and severity of illness, including death, intubation, need for renal replacement therapy, and ICU admission [17]. Univariate analysis found statistically significant, but weak links between elevated C-reactive protein and lactate dehydrogenase and the need for intubation with Spearman’s p² reported as 0.090 and 0.108, respectively. A C-reactive protein of >8mg/L showed 100% sensitivity and negative predictive value for intubation when variables were dichotomised to normal or above normal. The specificity and positive predictive value were poor for C-reactive protein at 4% and 27%.

Compared to the dichotomised approach taken by Barret et al. (2021), the relationship between intubation and inflammatory markers as continuous and categorical variables in the analysis was tested. This allowed us to explore higher levels of inflammatory markers as predictors of severe disease. Ultimately, a cut-off for C-reactive protein of 100mg/L in our final predictive model, which was higher than the cut-off of 8mg/dl used in the Barrett et al. (2021) study, was used. The population studied by Barrett et al. was dissimilar to the population in the current study, with higher patients with chronic kidney disease, hypertension, coronary artery disease, and diabetes [17]. These differences in population, approach, and analysis may account for the differing predictive ability noted. Moving forward, the validity of this predictive model should be tested in large populations with different demographics and comorbidities.

In line with previous data regarding the severity of disease and gastrointestinal symptoms, gastrointestinal symptoms such as nausea, vomiting or diarrhoea were negatively correlated with the need for intubation [10]. Other studies have found no link between mortality, need for intubation, nor severe disease [18,19]. Thus, it is not clear whether potential gastrointestinal tract infection is protective for intubation or if this is due to reporting bias; it is an area that warrants more rigorous clinical and pre-clinical investigation.

In the current study, the ideal timing for intubation was not defined in CoVID-19, and individual providers could have different thresholds for intubation that would affect which patients experienced our primary outcome. This ideal timing has not been defined in CoVID-19, and individual providers could have different thresholds for intubation. To address this limitation, data on oxygenation status pre-intubation and 72 hours post-intubation was collected. Intubated patients, on average, required 12.9L of oxygen before intubation. Based on the updated and revised version of the “Berlin Definition” of the acute respiratory distress syndrome (ARDS), the intubated patients had, on average, moderate severity acute respiratory distress syndrome on intubation that eventually improved to mild after 72 hours [20]. The average length of intubation was 11.9 days. All of this data shows that despite different intubation patterns, the intubated patients included in this study showed a reasonable level of hypoxemia to warrant intubation.

As a single-centre study, the current findings may lack generalizability. The measurement of C-reactive protein was at a single point at admission, and the change in C-reactive protein over time may be more informative as to the need for intubation. However, the stated goal was to understand the utility of this initial assessment. It is accepted that symptoms were self-reported and may not be a reliable indicator of the full symptomology in the cohort, especially given the rapidity with which some patients were intubated. As reported in patients’ notes, laboratory data and imaging were obtained through routine clinical care, driven by clinical decision making, and have undoubtedly contributed to missing data. Moreover, the final evaluation of the primary endpoint was performed four days following the last admission and may have missed further intubations or clinical changes.

In patients hospitalised with CoVID-19, high initial C-reactive protein was independently associated with the need for intubation and, in conjunction with high lactate dehydrogenase and diabetes mellitus, was strongly predictive of intubation. However, further investigation using larger, multicentre cohort studies are needed to validate the relationship described in this study.