Correlation of patient profiles and biomarkers with outcomes in covid-19 icu patients: a retrospective analysis

The COVID-19 pandemic is placing an acute strain on resource-limited intensive care setups in many developing Asian nations, including India. This disease is characterised by protean manifestations ranging from asymptomatic illness to acute respiratory distress syndrome, which may be further complicated by pulmonary and systemic thrombosis, disseminated intravascular coagulation, direct myocardial injury, and multi-organ dysfunction. Experiences of the past few months suggest that the disease course is highly unpredictable with rapid clinical deterioration often seen in patients presenting with moderate to severe illness. There is a pressing need to effectively triage patients for intensive care. Prediction of disease severity and institution of early aggressive care for at-risk patients may promote judicious use of limited resources and improve outcomes.

Observations primarily from Western populations have identified numerous clinical and demographic factors associated with severe illness and poor prognosis, including advanced age, male sex, and co-morbid illnesses such as diabetes, hypertension and end-stage renal disease. [1, 2] Significant racial and ethnic variations in susceptibility to severe illness and mortality have been seen in reports from around the world. [3]

Numerous biomarkers expected to represent the underlying pathogenetic mechanisms of severe COVID-19 illness have been advanced as potential diagnostic and prognostic indicators. They include lymphopenia (increased neutrophil-lymphocyte ratio [NLR]), serum ferritin, serum lactate dehydrogenase (LDH), C-reactive protein (CRP), D-dimers, serum procalcitonin, cardiac troponins, and IL6 levels. [4] Alterations in these markers potentially represent the combination of direct pulmonary injury and dysregulation of inflammation and coagulation that is considered pathognomonic of severe illness. However, utility of these various biomarkers as predictors of disease severity, progression, recovery, and mortality is still being evaluated.

There is paucity of data of clinical and demographic profiles of Indian COVID-19 patients requiring intensive care, especially with regard to biomarker profiles and derangements. The aim of this observational study was to describe the demographic, clinical and biomarker profiles of patients with moderate to severe COVID-19 admitted to intensive care in northern India and to correlate these with clinical outcomes. The study results may help us in triaging critical patients needing aggressive management in the first 24 to 48 hours after admission.

Clinico-demographic and biomarker comparisons with respect to patient survival and need for ventilatory support are summarized in Tables 1 and 2, respectively. Ninety-nine patients with laboratory-confirmed COVID-19 who were admitted to intensive care with moderate to severe illness were recruited for this study. The median age was 57 years (IQR, 45–65) and 62 (62.6%) were men. 77 (77.7%) had one or more comorbidities including diabetes (44.4%), hypertension (48.5%), chronic kidney disease (15.2%), coronary heart disease (13.1%) and anaemia (24.2%). The overall mortality in ICU was 45.4%. 23 (23.2 %) patients required non-invasive ventilatory support with BiPAP and 37 (37.3%) patients required invasive mechanical ventilation. 39 (39.3%) patients were managed with oxygen therapy, including high-flow nasal oxygenation. Values of biomarkers of inflammation (serum ferritin, LDH, CRP, NLR, and IL6) and coagulation (D-dimers) were comparable across age groups. D-dimers were significantly lower in females (p=0.046) when compared to males, who showed higher mortality. Patients with diabetes, who also showed higher mortality, showed significantly higher NLR (0.023) and lower PFR (p=0.011). Correlation analysis revealed significant inverse correlation between all biomarkers and the baseline PFR. The strongest correlations observed were between CRP and PFR (Pearson’s R=0.421, p<0.01) and between CRP and IL-6 (Pearson’s R=0.313, <0.01). Male gender and diabetes mellitus showed significant association with mortality (p=0.044 and p=0.042 respectively) as well as with the need of non-invasive or invasive mechanical ventilatory support (p=0.021 and p=0.027, respectively). Coronary artery disease was significantly associated with invasive mechanical ventilation (p=0.015). Prevalence of other comorbidities was not significantly different across outcome groups. Elevations of all biomarkers showed significant association with mortality and the need of ventilatory support (Figure 1). Subgroup analysis (post-hoc Bonferroni correction) revealed that elevations in all biomarkers were significantly associated with incidence of invasive ventilation.

Figure 1

CRP levels of patients requiring either non-invasive or invasive ventilatory support were significantly higher than that of patients not requiring any ventilatory support (p<0.001).

D-dimers of patients requiring invasive ventilation were significantly higher than that of patients requiring non-invasive ventilatory support (p=0.013) or no ventilatory support (p<0.001).

A binominal logistic regression was performed separately to ascertain the effects of biomarkers (Ferritin, LDH, CRP, IL6, NL ratio, D-dimer) and PF ratio on the likelihood of need of ventilatory support and mortality of participants. The model explained 77% (Nagelkerke R²) of the variance in incidence of patients needing ventilation support and 64% (Nagelkerke R²) of the variance in mortality of patients. Similarly, the model correctly classified 87.9% of participants who received ventilatory support and 85.9% of participants who died from COVID. IL6 (p=0.024), D-dimer (p=0.025) and CRP (p<0.001) were significant predictors for the need of ventilatory support. Similarly, IL6 (p=0.036), CRP (p=0.041), D-dimer (p=0.006), and PFR (p=0.019) remained significant predictors for mortality.

Receiver operating characteristic curves of CRP, IL6, NLR, D-dimer and PFR were analysed to ascertain optimal cut-offs for predicting mortality or need of mechanical ventilatory support (Table 3). All the variables were significant (p<0.05) predictors, with AUC>0.7. CRP >40 mg/L predicted mortality with a sensitivity of 93.3% and a specificity of 88.9% (AUC 0.933, 95%CI 0.879-0.987). The same cut-off of 40 mg/L had a sensitivity of 76.7% and specificity of 94.9% for predicting need for mechanical ventilatory support (AUC 0.956, 95%CI 0.915-0.996) (Figure 2).

Figure 2

Worldwide, with over 94 million cases and over 2 million deaths, the pandemic is showing a worrisome resurgence in the West, putatively associated with emergence of newer, more infectious viral variants sowing apprehension of fresh waves of infections as the race to immunize populations and achieve herd immunity is underway. The unpredictability of COVID illness has fuelled searches for clinical and laboratory scoring systems for patient triage so that intensivists can predict which subset of patients are likely to deteriorate faster and need immediate attention, even at the time of admission. This present study aims to fill knowledge gaps regarding clinico-demographic and biomarker profiles of COVID-19 patients to improve accuracy of triage algorithms.

Available epidemiological profiles of COVID-19 patients from the Indian subcontinent have mostly evaluated cohorts with mild to moderate illness and biomarker profiles have not been studied. [5, 6]

The principal findings of this study are that male gender and diabetes mellitus are associated with severe illness and poorer outcomes in terms of need for mechanical ventilation and mortality. Derangements of almost all study biomarkers (CRP, IL6, D-dimers, NLR, ferritin and LDH) were associated with ventilatory support and mortality. Analysis of cohorts of COVID-19 patients from across the globe have yielded similar observations, including a meta-analysis of 32 studies from around the world by Malik et al. [1,7-9] On a lookout for a combination of biomarkers which accurately predicts patient outcomes, further data analysis was performed, and an interesting observation emerged, which may be helpful in resource-limited nations. In our study, CRP, IL6 and D-dimers emerged as independent predictors of mortality and need of mechanical ventilation, with a combined model predicting the outcomes correctly in over 85% of cases.

Figure 3

Contrary to hypothesis no association was found between age of patients and survival. This may be indicative of selection bias arising from inclusion of only moderate to severe cases under intensive care. Also, institutional triage policy during acute surge conditions kept a lower threshold for early referral of older patients or patients with comorbidities for intensive care in expectation of rapid clinical deterioration. Consequently, younger patients without comorbidities were likely to be referred to intensive care later in the course of their illness when their clinical condition had already worsened. This might have led to equivocal mortality figures. Complementing clinical criteria with biomarker data for triage decisions will aid more judicious allocation of intensive care resources.

Higher prevalence of comorbid illnesses among COVID-19 patients with moderate to severe illness requiring hospitalisation has been noted across the world, especially diabetes, cardiovascular disease including hypertension, chronic pulmonary disease, and chronic kidney disease. Individual comorbidities have been associated with adverse outcomes with varying significance across studies. [1, 10, 11] 77% of our intensive care patients had one or more co-morbidities.

In our study, coronary artery disease was seen to be associated with need of invasive ventilatory support. This likely reflects the role of direct myocardial injury and increased systemic thromboembolic events seen in moderate-severe COVID-19 which are particularly deleterious for this patient group. Similar observations have been reported in studies by Cummings MJ and Park BE. [10, 12]

Diabetes mellitus is known to induce immunodeficiency and immune dysregulation with increased susceptibility to infections. Guo et al. found diabetics in Wuhan to be predisposed to more severe COVID pneumonia with increased derangements in IL6, CRP, ferritin, and D-dimers but not increased mortality. [13] Our study findings were suggestive of increased mortality and need for mechanical ventilation in diabetics with severe COVID illness who also had significantly higher NLR. Asirvatham et al. in their study had also suggested that prevalence of diabetes was associated with higher adjusted case fatality in Indian patients. [14] Thus, diabetes should be factored in triaging Indian COVID-19 patients and it may be a factor in higher mortality in younger age groups of urban populations. It is notable that systemic steroids for moderate to severe COVID pneumonia precipitate further derangement of blood sugars in diabetics.

Dysregulated systemic inflammation and coagulation remain the enduring challenge in severe COVID-19. We have evaluated a full spectrum of inflammatory biomarkers previously associated with adverse outcomes: serum ferritin, LDH, IL6, hsCRP, and NLR. IL6 is one of the cytokines which has received intense interest as the possible driving agent of COVID-19 cytokine storm syndrome and is being evaluated as a therapeutic target. IL6 elevations have significant association with mechanical ventilation and mortality in COVID-19. [15] Herold et al. validated a IL6 cut-off of >80 pg/ml at any point in the course of illness, accurately predicting respiratory failure requiring invasive ventilation in 80% of cases. [16] Their study also suggested a cut-off of 35 pg/ml to be optimal for IL6 at initial assessment though with lower specificity. This correlates with our findings with respect to baseline IL6 values that suggested an optimal cut-off of 32.5 pg/ml to predict invasive mechanical ventilation and mortality. IL6 estimation has the additional value of guiding therapeutics. Studies evaluating monoclonal antibody therapies targeting IL6 have shown inconsistent benefits so far, though recent work has shown promising results with timely therapy in select critically ill patients. [17] Elevated IL6 with critical illness at presentation should prompt consideration of targeted therapies.

C-reactive protein elevations have been widely associated with severe COVID-19 [9]. It has special value in being an early predictor of deterioration. CRP production in the liver is driven by IL6, thus explaining the close correlation of these two markers. We found baseline hsCRP to be an independent predictor of mortality with a cut-off of 40 mg/L strongly predictive (AUC>0.9) of mechanical ventilation and mortality with 93% sensitivity and 88% specificity. Herold et al. have validated a cut-off of 32.5 mg/L for CRP at presentation to predict invasive ventilation, though they have suggested a cut-off of >97 mg/L at any point during the course of illness to predict impending respiratory failure with greater specificity. Thus, a baseline CRP is a valuable and relatively widely available biomarker which can be utilised independently for patient triage.

D-dimer monitoring has become standard of care in COVID-19 due to high risk of venous thromboembolic events in moderate to severe illness and derangements have been widely associated with mortality risk. [10, 18] Systemic dysregulated coagulation associated with the inflammatory reaction results in hypercoagulability and may evolve into disseminated intravascular coagulation in critically ill patients. [19, 20] Zhang et al. described a cut-off of 2000ng/ml of D-dimers at admission to predict in-hospital mortality with 92% sensitivity and 83% specificity. [21] Our findings suggest a lower cut-off of 810ng/ ml can accurately predict all adverse outcomes, including non-invasive or invasive ventilation, apart from mortality, which would be valuable while triaging patients. We found progressive elevations of D-dimer to be associated with escalating need for ventilatory support, with significantly higher elevations in patients requiring invasive ventilation compared to those requiring non-invasive ventilation. Gao et al. further showed increased predictive value for severe illness when D-dimer evaluation was combined with IL6 measurement in tandem. [22] Also, D-dimers can vary significantly during the course of the illness and serial measurements to note the trend can be valuable in predicting outcomes, as noted by Naymagon et al. [23]

Profound lymphopenia is a nearly universal finding in severe COVID-19 infection. [9] The role of ACE-II receptor–mediated infection of subsets of lymphocytes by the SARS-CoV-2 virus is being evaluated. [24] Similarly, we found elevated NLR to be significantly associated with adverse outcomes and a cut-off of >8.5 was found to be a moderately good predictor of survival. However, baseline NLR was not found to be an independent predictor of outcomes. Considering the multiple factors likely to influence it during the course of the illness, we suggest that dynamic trends are likely to have better predictive value, much as observed by Ye et al. in their study. [25]

LDH is an established marker of lung injury in acute respiratory distress syndrome (ARDS) and predicts the extent of respiratory compromise. Thus, it is unsurprising to find significant inverse correlation with PFR in COVID ARDS. LDH elevations have thus been associated with severe

COVID ARDS, invasive mechanical ventilation and poor outcomes ^. [16, 26]. Our study results reflect the same, but regression analysis failed to show significant predictive value in predicting mechanical ventilation or mortality. Since it indicates ongoing lung inflammation and injury its efficacy as an early predictor of deterioration is limited. However, by the same principle a decreasing trend of LDH may be evaluated as an indicator of resolution of lung inflammation signalling recovery in times to come.

Serum ferritin is a known acute phase reactant and levels rise in various systemic inflammatory states and infections. Derangements in ferritin levels have been uniformly reported in COVID-19. [4] An association with possible hemophagocytic lymphohistiocytosis (HLH, macrophage activation syndrome) leading to clinical deterioration was explored. However, though a significant association with poor outcomes has been seen in several studies, ferritin has been seen to be a poor predictor of mechanical ventilation or mortality. The levels commonly encountered in COVID-19 are several orders of magnitude lower than that seen in HLH, refuting that theory of pathogenesis. [27] The elevations are likely representative of the ongoing systemic inflammatory response. Our data suggest a similar conclusion. Serum ferritin elevations were not predictive of outcomes though they showed significant association. Thus, ferritin has poor utility as a predictor of severe COVID-19. Elevations may persist for prolonged periods in spite of clinical recovery. [28]

Our study had some limitations. Selection bias would have been introduced since we were limited to recruiting patients with moderate to severe illness admitted to intensive care. Baseline biomarkers and PFR at the time of admission to the hospital could not be included in the analysis due to data not being available for all cases. Baseline laboratory assessment was done at the time of ICU admission which would have occurred at varied times in the course of illness in different patients. Also, analysis of serial measurements of biomarkers and PFR to understand trends could have improved the predictive value of some biomarkers and helped identify markers of recovery.

Among COVID-19 patients, male gender and diabetes mellitus are associated with need of ventilatory support and mortality. Baseline CRP >40 mg/L, IL6>32.5 pg/ml, and D-dimer >810 ng/ml are sensitive and specific independent predictors of need for ventilatory support and risk of mortality. hsCRP and IL6 elevations are highly correlated and represent the same pathogenetic mechanism and are early predictors of impending adverse outcomes when used in isolation or in conjunction. A baseline NLR >8.5 is sensitive and specific to predict adverse outcomes though its predictive value is likely to be greater if used as a dynamic measure. Ferritin and LDH are not accurate predictors of outcomes, though LDH may be a surrogate marker for the extent of lung injury. Further studies incorporating dynamic changes in biomarkers could identify algorithms to improve assessment of disease course, therapy, and prognosis.