Laboratory findings predictive of critical illness in hospitalized COVID-19 patients in Tunisia

Study design

Inclusion and exclusion criteria :

Eligible patients were those admitted to the COVID-19 unit between October 2020 and June 2021 with moderate to severe COVID-19 pneumonia at admission, confirmed for SARS-CoV-2 via Reverse Transcription-Polymerase Chain Reaction (RT-PCR), Rapid Antigen Test (RAT), or positive SARS-CoV-2 serology (IgM or IgG) for vaccine-naive patients. All patients were at their first infection.

Patients were not included if they had a mild COVID-19 infection throughout their hospitalization and another reason for hospitalization, if they were patients admitted to the COVID-19 unit after an initial stay in the ICU, or if they remained in the emergency department for more than 7 days before being admitted to the COVID-19 unit.

We excluded patients who were admitted to the COVID-19 unit with hypoxemic pneumonia and suggestive findings of COVID-19 pneumonia on thoracic CT scan but with negative virological results (RT-PCR, RAT, and SARS-CoV-2 serologies), and an alternative cause was identified during the course of the disease. Additionally, patients were excluded if virological documentation data could not be located.

Outcome measure :

We defined the development of a critical state before or on the 7th day of hospitalization (D7) as the outcome measure. Patients were categorized into two groups:

Definition of a Critical State : We defined a critical state as:

Data collection

Data were retrospectively and anonymously collected between June 2023 and January 2024. All demographic, clinical, biological, radiological, therapeutic, and progression-related data were extracted from the medical records. The medical records followed a standardized format with predefined items consistently used in both departments of COVID-19 unit. The collected information included demographic data, clinical details recorded at admission and D7, and biological data obtained at admission. For the latter, all patients underwent a routine standard biological assessment at admission complete blood count (CBC) including leukocytes, lymphocytes, neutrophils, Neutrophil-to-Lymphocyte Ratio (NLR), platelets, hemostasis profile (Prothrombin Time, Activated Partial Thromboplastin Time, D-Dimers) as a hypercoagulability biomarker), renal profile (Urea, Creatinine, Blood Ionogram), liver profile (Aspartate Aminotransferase (AST), Alanine Aminotransferase (ALT)), markers of muscle injury (Creatine Kinase (CK), Lactate Dehydrogenase (LDH)), and C-Reactive Protein (CRP) as a marker of high inflammatory response. Some parameters were unavailable due to reagent shortages, and others were optional based on clinical presentation, such as Troponin (myocardial injury biomarker) and N-terminal prohormone of brain natriuretic peptide (NT-ProBNP); an indicator of cardiac stress or heart failure. Chest computed tomography (CT) data were recorded at admission and upon aggravation during evolution. Therapeutic data included oxygen therapy, antibiotic treatment, corticosteroids, anticoagulation, and other measures, adhering to a uniform protocol guided by INEAS recommendations. Evolvement data covered hospital discharge, transfers to ICU, COVID-19 unit deaths, and complication occurrences.

Statistical analysis

For the analysis of the association between two qualitative variables, we used Pearson’s Chi-squared test to compare two frequencies when the conditions for its application were met, and the Fisher’s test otherwise. For the analysis of the association between a qualitative variable and a quantitative variable, we employed the Student’s t-test to compare two means in the case of a normal distribution, and the non-parametric Mann Whitney test otherwise. Correlations between two quantitative variables were calculated using the Pearson correlation coefficient, with significance tested bilaterally. In the multivariate study, a binary logistic regression model was followed, and the risk was computed as the Odds Ratio (OR) with a retained 95% confidence interval (CI 95%). To assess performance, we utilized parameters related to the Receiver Operating Characteristic (ROC) Curve, as well as sensitivity and specificity. Pre- and post-test probabilities were determined through likelihood ratios represented on the Fagan nomogram. We set the significance threshold at p ≤ 5%.

Ethical considerations

This retrospective study received approval from the Ethics Committee of University Hospital Center Mongi Slim, under approval number 57/2023 on Friday, 22 December 2023.

We followed strict ethical committee guidelines that allowed for exemption of consent, due to the non-intrusive nature of the study and the use of non-identified data to ensure confidentiality and anonymity of participants. All personally identifiable data were anonymized prior to analysis to protect individuals’ privacy. We also carefully assessed the risks and benefits of our research, ensuring to minimize the former and maximize the latter for participants and the scientific community. No financial or other conflicts of interest were identified by the authors of this study. Policies regarding the future use of data and their potential sharing with other researchers were strictly established in accordance with ethical guidelines.

Univariate analysis

The univariate logistic regression indicated that laboratory findings between non-critical and critical groups showed numerous differences including C-reactive protein (p=0.047), D-Dimer (p=0.011) and creatinine (0.026) as well as, creatine kinase (p=0.039), lactate dehydrogenase (p=0.04), troponin (p=0.001) and NT-pro-BNP level (p=0.011) which were all higher among patients in critical condition. On the other hand, lymphocyte (p<0.001) and platelet (p<0.001) counts, natremia (p=0.011) and kalemia (p=0.029) were significantly lower among the critical group, whereas liver enzymes showed no significant differences. Table 3 summarizes the biological abnormalities on admission.

Multivariate analysis

Based on the multivariate logistic regression model, four variables were demonstrated as independent risk factors. As shown in Table 4, the results indicated that lymphopenia (OR=2.771, 95%CI=1.482-5.181, p=0.001), NLR (OR=2.286, 95%CI=1.461-3.578, p<0.001), thrombocytopenia (OR=1.944, 95%CI=1.092-3.459, p=0.024), and CRP>71.5 (OR=1.598, 95% CI=1.042-2.45, p=0.032) were associated with critical outcome.

The threshold values of 71.5mg/L for the CRP level and 5 for the RNL were defined according to the study of the ROC curve according to the critical and non-critical groups. The findings in Table 4 show that statistically significant AUCs were obtained for CRP (AUC=0.533, p=0.032) and NLR (AUC=0.589, P<0.001). Thus, the performance of the NLR was slightly superior to that of the CRP in predicting a critical condition on day 7 ( Figure 2). For a threshold equal to 71.5 mg/L, the sensitivity and specificity of CRP to predict critical illness were respectively equal to 66.1% and 38.1%, while for a threshold equal to 5, the sensitivity and specificity of the NLR to predict the critical illness were respectively equal to 68.7% and 51% ( Table 5). Figure 2 and Figure 3 show the area under the ROC curve of CRP and NLR ( Figure 2) and platelet count ( Figure 3). The platelet count was negatively correlated with the occurrence of a critical condition and it was quite efficient in predicting non-critical cases; (AUC=0.62; CI95%=0.56-0.679) ( Figure 2). The best threshold was estimated at 187.5*10³ elements/mm³ corresponds to a sensitivity and specificity equal to 72.4% and 48.7% respectively with a quite good accuracy ( Table 6).

Figure 2. ROC curve indicating the performance of CRP and NLR in predicting critical cases.

Figure 3. ROC curve indicating platelet count performance in predicting non-critical cases.

Hematological biomarkers

Our univariate analysis findings regarding the CBC parameters were different to previous findings.^17,18 Previous studies concluded that increased neutrophils count is characteristic of patients with severe cases.¹⁹ Evidence suggests that severe COVID-19 is associated with elevated neutrophil levels, which increase inflammation via cytokine storm, and hemorrhages especially in the lungs, which occur as a result of neutrophil-induced tissue damage.

According to our data, lymphopenia occurred in 78% of cases. The lymphocyte count ranged between 500 and 1000/mm³ among 36.6% of our patients while 8.8% had deep lymphopenia (< 500/mm³). Our results also showed, using the multivariate logistic regression model, that lymphopenia was an independent risk factor for a severe illness. These findings are in line with the literature. Indeed, several studies have assessed the impact of lymphopenia on disease severity. For example, Zhang and al.²⁰ reported an average blood lymphocyte count of 1200/mm³, and Jiang and al.¹¹ showed that 41% of patients had a lymphocyte count of less than 1000/mm³. This count was significantly lower in patients with an unfavorable outcome. Additionally, Bellan and al.²¹ demonstrated that a low lymphocyte count at admission was correlated with the risk of mortality due to SARS-CoV-2 infection. Two Chinese meta-analyses^22,23 have concluded that patients with severe COVID-19 pneumonia had lower blood lymphocyte counts compared to non-severe forms, and lymphopenia was associated with a three-fold higher risk of severe infection, particularly ARDS. When viral particles resulting from the replication of SARS-CoV-2 spread through the respiratory mucosa, it leads to a cytokine storm, generating a cascade of immune responses primarily involving CD4 and CD8 T lymphocytes.²⁴

In our study, thrombocytopenia was present among 13% of cases and was correlated with the progression to a critical condition (p=0.023). We demonstrated that thrombocytopenia was associated with an odds ratio (OR) of 1.94 of developing a critical condition. Our findings are consistent with those of the literature. Indeed, a meta-analysis of 21 studies including 3,377 SARS-CoV-2 positive patients⁴ found that platelet count was negatively correlated with the severity of the infection as well as mortality. Another meta-analysis involving 1,779 COVID-19 patients²⁵ revealed that thrombocytopenia was associated with a threefold higher risk of severe disease, and mortality. Moreover, the platelet count was also correlated with disease severity scores and the risk of mortality in ICU.^26,27 Platelet count was considered as an independent risk factor for COVID-19 mortality. This count was significantly lower in COVID-19 patients who died compared to survivors.^21,28 Platelet kinetics could also help to identify patients at risk of an unfavorable outcome.²⁹ Similarly, Choi KW et al.³⁰ reported in a study evaluating the prognosis of SARS infection in 2003, that lymphopenia and thombopenia were found in 73% and 50% of cases respectively.

In COVID-19 patients, thrombocytopenia appears to be multifactorial. In the case of ARDS, there is a significant platelet consumption resulting from the combination of viral infection and mechanical ventilation, leading to endothelial damage and platelet activation, aggregation, and thrombus formation in the lungs.^25,29 Moreover, the lung can be a site for platelet release from mature megakaryocytes, and alteration of the pulmonary capillary bed can affect platelet degranulation and pulmonary megakaryocytopoiesis.^25,29

In our study, the mean NLR was 7.52. This ratio was higher in the critical group patients (p=0.073). As we previously mentioned, lymphopenia was more common among patients in a critical condition. It appears logical to observe a trend of increasing NLR in severe forms. This finding was consistent with studies of both Yang and al.³¹ and Yan and al.,³² where a higher NLR was correlated with severe forms and mortality, respectively (p<10⁻³). NLR is a biomarker that integrates two subtypes of leukocytes representing two inversely related immune pathways. It provides information about systemic inflammation and is a useful biomarker for predicting bacterial infection, including pneumonia, better than absolute leukocyte, lymphocyte, or neutrophil counts.³² A high NLR reflects a discrepancy in the inflammatory response.³² Numerous studies and meta-analyses^33–35 have proved that an increased NLR is associated with a poor prognosis in various diseases, such as cardiovascular diseases, solid cancers, and infections.

Increased NLR during COVID-19 could be a result of the expression of inflammatory cytokines, an unusual increase in pathological low-density neutrophils, and the upregulation of genes involved in the lymphocyte cell death pathway caused by SARS-CoV-2 infection.³⁶ Furthermore, two studies^37,38 have proposed integrating the NLR into nomograms to prove the prognostic value of this biomarker.

Our study showed that for a threshold ≥ 5, the sensitivity and specificity of NLR in predicting critical status on Day 7 were 68.7% and 51%, respectively. This sensitivity was 56.52% in the study by Liu and al.³⁸ Therefore, NLR is an objective, simple, and inexpensive parameter that could be used in routine practice as a prognostic biomarker.

Thrombotic complications and coagulopathy represent a major event during SARS-CoV-2 infection. An increase in D-dimer and fibrinogen levels and a decrease in prothrombin time indicate a state of hypercoagulability. Many studies have shown that high levels of D-dimer are associated with severe^39,40 and critical¹⁰ forms, the need for intensive care management,^2,18 and in-hospital mortality due to COVID-19 pneumonia.^28,41 This hypercoagulability in COVID-19 patients could result from several mechanisms⁴²: in viral infections, there is often an unbalance between pro-inflammatory response and anti-inflammatory response; this can lead to endothelial cell dysfunction and excessive thrombin production; hypoxia contributes to thrombosis by increasing blood viscosity.

Increased D-dimer levels indicate that the fibrinolytic system is activated in COVID-19 patients. Furthermore, the increased release of cytokines during viral infections stimulates coagulation cascade.⁴³ A reduction in D-dimer levels was observed in recovered patients, independently of anticoagulant treatment, while a continuous increase in D-dimer levels was predictive of a higher risk of thrombosis and unfavorable outcomes.^42,44 Thus, D-dimers are an early and reliable marker for predicting a poor prognosis in COVID-19 hospitalized patients.⁴²

In our study, thromboembolic events occurred among 11 patients. The incidence of pulmonary embolism (PE) in COVID-19 patients is still unknown and likely underestimated.⁴⁵ Bilaloglu and al. reported a prevalence of PE of 3.2% in a study involving 3,334 COVID-19 patients in New York,⁴⁶ while according to a French study,⁴⁷ the incidence of PE was 23.7%. This difference could be explained by differences in disease severity, patient characteristics, and the limited use of CT angiography among patients.

Biochemical biomarkers

Biochemical tests findings, showed that the critical group of patients had elevated level of CRP, D-Dimer, creatinine, lactate dehydrogenase, creatine kinase, troponin and NT-Pro-BNP, which corroborates the conclusion made by Feng and al.⁴⁰

The CRP was elevated (>8 mg/L) in 93.5% of patients with an average of 110 ± 73 mg/L. We noted that the CRP on admission was significantly higher in the critical group (p=0.047). In multivariate analysis, an elevated CRP level was correlated with an increased risk of progressing to a critical condition on the 7th day of admission. Many studies have shown that elevated CRP levels are correlated with critical forms,^10,39 disease progression,⁴⁸ and mortality from COVID-19 pneumonia.²¹ As known, CRP is a non-specific marker of inflammation induced by interleukin-6 (Il-6) secretion. In clinical practice, it is used as a biomarker for various inflammatory and infectious conditions. High CRP levels have been directly correlated with the inflammation’s degree and disease severity.⁴⁹ Moreover, CRP levels among dead COVID-19 patients were decuple higher than survivors.⁵⁰

A meta-analysis of 20 studies including 4,843 COVID-19 patients and focusing on the clinical utility of CRP,⁵¹ emphasized that high CRP level was associated with a fourfold higher risk of an unfavorable outcome (p<10⁻³). In fact, at the early stages of COVID-19 infection, an increase in CRP was directly associated with the development of lung lesions, reflecting the severity of the disease.^51,52 Furthermore, Ali and al.⁵³ demonstrated that CRP level could predict disease worsening among non-severe cases, indicating 5% risk of progressing to a severe form for each unit increase in the CRP rate. Added to that, the CRP level has also been reported as a reliable biomarker for treatment responses in COVID-19 patients¹⁹; in fact, this marker could be used to select patients who would benefit from treatment with tocilizumab, another IL-6 receptor inhibitor similar to sarilumab.^54,55

Our findings indicated that NLR and CRP are good predictors of unfavorable outcome. The reported excellent accuracy of these parameters in the prediction of COVID-19 patients’ outcome corroborates findings of the studies of Yang and al.³¹ and. Liu and al.³⁸

We found that blood creatinine (p=0.026) and urea levels (p=0.061) were higher in the critical group. Furthermore, urea levels were significantly higher in elderly patients (> 70 years old) (p<10⁻³) and in patients with high blood pressure (p=0.012). Elevated blood urea (≥ 7 mmol/L) was noted in 46.6%. These findings align with the literature, where it has been demonstrated that elevated blood creatinine and urea values are associated with severe disease, unfavorable prognosis, and significant mortality.^41,56,57 A Tunisian study conducted between September and December 2020,⁵⁸ illustrated that acute renal failure was associated with poor outcome (OR:6.7) for patients hospitalized in ICU. The mechanism of renal involvement in COVID-19 is likely to be multifactorial.⁵⁶ It involves direct cytopathic effects on kidney tissue by the virus leading to renal cell necrosis as well as indirect damage by cytokines and metabolites induced by hypoxia, shock, or rhabdomyolysis.

Increased liver enzymes level was found in 39.5%, especially in male patients (p<10⁻³), while AST and ALT rates did not vary between both groups. Numerous studies have demonstrated the association between high transaminase levels and the severity of the disease,^39,40,57,59 transfer to the ICU ^2,18,60,61 and death^28,41 due to COVID-19. Jamoussi and al.⁶² have concluded that IL-6 ≥ 20 pg/ml, CK < 107 UI/L,AST < 30UI/L were independent risk factors for mortality among patients admitted in ICU between September and December 2020 in Tunisia. Moreover, the good accuracy of AST and ALT as a predictor of ICU admission have been clearly shown with AUC>0.7.⁶³ Added to that, Malik and al. demonstrated in their meta-analysis⁵¹ that high levels of AST and ALT (> 40 IU/L) were associated with a threefold higher risk of a poor prognosis. Some studies have shown that COVID-19 only transiently increases transaminases. Cytolysis is rather due to liver damage secondary to systemic inflammatory processes, hypoxia, or to the use of hepatotoxic drugs, especially antivirals such as Lopinavir and Ritonavir during patient management.^51,64 However, viral RNA has been detected in the liver at high titers, exceeding viremia, during autopsies, suggesting that SARS-CoV-2 hepatic infection can contribute to elevation of transaminase levels in patients with severe forms of COVID-19.^44,65

Regarding rhabdomyolysis markers, we found an elevation of LDH and CK levels among 98% and 24% of our patients, respectively. We also demonstrated that muscle lysis enzymes were significantly higher in critical group patients. Our data were consistent with the literature, where it has been shown that high levels of LDH and CK were associated with the severity^5,66 and progression⁶⁷ of the disease, transfer to ICU, and mortality.⁵⁷ Jamoussi A⁶² from COVID-19. In a meta-analysis assessing the prognostic value of LDH levels,⁴¹ it was found that elevated LDH levels were associated with a risk of mortality (OR: 16) and severe disease (OR: 6). Furthermore, a study conducted on COVID-19 patients⁶⁸ showed that increased LDH levels at the early stage of the disease can predict lung damage and severe cases of COVID-19. These high LDH levels may result from decreased tissue oxygenation leading to stimulation of the glycolytic pathway or from damage of multiple organs in case of multi-organ failure.⁵¹ Additionally, severe infections can induce cytokine-mediated tissue damage and LDH release.⁶⁹

Concerning cardiac injury markers, we found that elevated levels of cardiac troponins were correlated with critical conditions (p=0.001). Indeed, troponins have a prognostic value in sepsis⁷⁰ and have been proposed as severity markers¹⁸ of the disease and predictors of COVID-19 mortality.^28,41 These disturbances in cardiac enzymes can result from viral myocarditis, myocardial damage caused by cytokines or microangiopathies and coronary spasms secondary to hypoxia.⁷¹

Furthermore, according to our data, we observed an increase in NT-Pro-BNP levels among 25 patients (5.5%), indicating a left heart failure. Additionally, the mean NT-Pro-BNP value was higher in the critical group of patients (p=0.011). The mechanisms of heart failure can be attributed to an imbalance between increased cardiac output and reduced oxygen supply, with the possibility for type 2 myocardial infarction.⁷² According to Li and al.,³⁹ NT-Pro-BNP>500pg/L were significantly associated with severe forms of COVID-19, and Hong and al.,¹⁰ demonstrated that critically ill patients had higher NT-Pro-BNP levels (p=0.002). Therefore, cardiac biomarkers, including troponins and NT-Pro-BNP, can reflect cardiovascular involvement in COVID-19, since they are independent risk factors for poor prognosis and mortality.^72,73

Strength and limitations

Our study has shown that some biochemical and CBC tests are important in predicting COVID-19 patients’ need for ICU care. Specifically, laboratory tests that should be prioritized to determine patient risk of developing severe COVID-19 pneumonia, include lymphopenia, NLR, thrombocytopenia, CRP, D-Dimer, creatinine, LDH, CK, troponin and NT-Pro-BNP. NLR is most preferred as it was noted to be a very good test.

However, limitations should be considered in the interpretation of our findings. First, our study is retrospective and monocentric. Second, some biological parameters were missing in the medical records and other specialized tests are not commonly performed in our hospital such as Il-6 and Procalcitonin given their cost. Also, the lack of systematic data on blood gas analysis on Day 7, did not allow us to evaluate PaO₂/FiO₂ ratio, to define ARDS. Besides, the study did not consider the influence of pre-existing health conditions.

Despite its limitations, this study has provided insights into laboratory parameters that can be used to predict the severity of COVID-19 cases allowing prediction of severe illness at the time of admission. It is therefore recommended that healthcare providers consider these parameters in making evidence-based decisions regarding patient management especially where there are limited ICU facilities.

It is worth mentioning that most of reported studies were conducted in different geographical and temporal contexts, some studies were conducted during the same period as our study.^58,62 Indeed, COVID-19 pandemic waves showed significant time lags across the world due to several factors, including health control measures, vaccination rates, population density, and circulating viral strains. For example, while some countries experienced a severe first wave at the beginning of 2020, other regions saw a peak in cases later in the year or even in 2021 with the emergence of more transmissible variants like Delta.⁷⁴ In Tunisia, the first wave ranged from March to June 2020 and was characterized by small numbers of circulating lineages. The second wave ranged from July 2020 to January 2021 and was characterized by a higher genetic diversity with the circulation of at least 20 different lineages. The third wave ranged from February to May 2021 and was characterized by the emergence of variants of concern (VOCs) and variants of interest (VOI), such as Alpha who rapidly became the predominant variant. The fourth wave started on June/July 2021, the overwhelming majority of detected lineages was Delta.¹ Consequently, our study included several variants of the virus, which impacts the clinical presentation and the outcome of the infection. However, we believe that although the period of our study is lagged compared to the studies cited in the discussion, our results remain relevant for assessing the prognosis of patients with COVID-19, regardless of the variant.

Ethical approval and consent

This retrospective study received approval from the Ethics Committee of University Hospital Center Mongi Slim, under approval number 57/2023 on Friday, 22 December 2023.

We followed strict ethical committee guidelines that allowed for exemption of consent (institute policy), due to the non-intrusive nature of the study and the use of non-identified data to ensure confidentiality and anonymity of participants. All personally identifiable data were anonymized prior to analysis to protect individuals’ privacy. We also carefully assessed the risks and benefits of our research, ensuring to minimize the former and maximize the latter for participants and the scientific community Policies regarding the future use of data and their potential sharing with other researchers were strictly established in accordance with ethical guidelines.