Introduction

Pneumonia and subsequent respiratory insufficiency caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV‑2) infection are leading reasons for hospitalization and high morbidity and mortality observed in coronavirus disease 2019 (COVID-19) patients [1, 2]. Due to strong inflammatory response mediated through interleukin (IL)-6, a subset of patients develop severe or critical symptoms and experience unfavorable clinical outcomes [3]. Severe or critical disease and hospitalization rates may vary depending on the dominant circulating virus strain, number of tested individuals and vaccination status of the population [4] and were reported to be 15–20% in the large cohort of patients at the start of the pandemic [2]. Clinical deterioration in these individuals developing due to hyperinflammatory response can potentially be modified using immunosuppressive drugs [5,6,7].

Among other parameters that are affected by strong inflammatory response to the SARS-CoV‑2 infection, C‑reactive protein (CRP) and albumin are easy to obtain and widely available. Both parameters are produced in the liver and regulated by IL‑6 promoted inflammation [8], although CRP is positively affected and albumin negatively affected by this process. These parameters are thus commonly deranged among other liver blood tests during the COVID-19 infection [9]. The CRP to albumin ratio (CAR) incorporates deviations of these parameters into a single direction and has been shown to be prognostic of more severe disease and higher mortality if elevated in COVID-19 patients [10,11,12,13]. These findings resemble those observed in the context of other chronic inflammatory diseases [14,15,16,17]. Nevertheless, clinical associations of CAR in COVID-19 patients outside respiratory status and death and its prognostic significance in relationship to established prognostic scores have not been investigated so far. This is the reason why we aimed to investigate the clinical and prognostic roles of CAR in a large cohort of hospitalized COVID-19 patients treated in our institution that we present in this paper.

Patients and methods

We retrospectively analyzed a total of 2309 consecutive COVID-19 patients hospitalized in our institution in the period from March 2020 to March 2021 with available data for CAR assessment on admission. We further validated our findings in an independent cohort of 1155 patients treated in our institution in period from March 2021 to June 2021.

Our institution is a clinical hospital that was repurposed as a regional tertiary COVID-19 center during the pandemic. All patients had a positive polymerase chain reaction or rapid antigen COVID-19 test prior to hospital admission. All patients were adults and of white race. Pregnant women were only occasionally treated as other institutions served as gynecologic referral centers. Patients were treated according to the contemporary guidelines. Only index hospitalizations with acute COVID-19 were investigated. Clinical and laboratory data used in this paper are a part of a hospital registry project and were obtained through analysis of electronic and written medical records. The study was approved by the Institutional Review Board.

The severity of COVID-19 on admission was graded as mild, moderate, severe and critical based on the World Health Organization (WHO) recommendations [18]. Comorbidities were assessed as individual entities and were summarized using the Charlson comorbidity index. The CRP (normal range < 5 mg/L) and albumin (normal range 40–48 g/L) concentrations were determined in addition to other hematological and clinical parameters. The CAR was calculated as CRP (mg/L) to albumin (g/L). The CAR values were stratified at median value (2.92) for the purpose of comparison with clinical characteristics and clinical outcomes presented in Tables 1 and 2. Mortality and other clinical outcomes were assessed from the start of hospitalization. The modified early warning score (MEWS) was used to quantify COVID-19 symptom severity. Confusion, urea, respiratory rate, blood pressure and 65 years of age (CURB-65) and Veterans Health Administration COVID-19 (VACO) index were used as prognostic risk scores and the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula was used for estimated glomerular filtration rate (eGFR) estimation.

Table 1 Patient characteristics on admission and their relationship with CRP to albumin ratio (CAR) stratified at median
Full size table
Table 2 Clinical outcomes of index COVID-19 hospitalization/postdischarge in relationship to C‑reactive protein to albumin ratio (CAR)
Full size table

Statistical methods

Normality of distribution of numerical variables was tested using the Shapiro-Wilk test. All analyzed variables had non-normal distribution and were presented as median and interquartile range (IQR) and were compared between groups using the Mann-Whitney U-test or Kruskal-Wallis one-way analysis of variance (ANOVA) test where appropriate. Categorical variables were presented as frequencies and percentages and were compared between groups using the χ2-test or the Fisher test where appropriate. Survival analyses were based on the Kaplan-Meier method. Univariate survival analyses were performed using the Cox-Mantel version of the log-rank test [19, 20] and the Cox regression analysis. Multivariate survival analyses were performed using the Cox regression analysis while simultaneously controlling for all included parameters. P-values < 0.05 were considered statistically significant. All analyses were performed using the MedCalc statistical software version 20.008 (MedCalc Software Ltd, Ostend, Belgium).

Results

Patient characteristics and their associations with CRP to albumin ratio

A total of 2309 hospitalized COVID-19 patients treated in the period from 3/2020 to 3/2021 were analyzed. Median age was 73 years IQR (64–81 years). There were 1326 (57.4%) males. Median Charlson comorbidity index was 4 points IQR (3–6). Majority (1981, 85.8%) of patients presented with severe or critical severity of COVID-19 symptoms on admission. During hospitalization, a total of 678 (29.4%) patients required intensive care unit treatment, 512 (22.2%) required mechanical ventilation and 866 (37.4%) died. The 30-day survival rate was 63.5%.

Median CRP, albumin and CAR levels were 91 mg/L IQR (43.1–153.8), 32 g/L IQR (28–35) and 2.92 IQR (1.33–5.23), respectively. Both CRP (HR = 1.003; P < 0.001) and albumin (HR 0.91; P < 0.001), age (HR 1.04; P < 0.001) and male sex (HR 1.23; P = 0.004) showed significant mutually independent association with 30-day survival in our cohort of patients, providing a rationale for their analysis as a ratio in the context of COVID-19. Patient characteristics stratified according to the CAR median value are shown in Table 1.

Patients with higher CAR were statistically significantly more likely to be older, of male sex, to have more severe COVID-19 on admission and worse functional status, to present later during the disease course, have pneumonia, require oxygen supplementation and to have coinfection on admission. They were also more likely to receive specific treatment (corticosteroids, LMWH, remdesivir). As expected from the CAR definition, patients with a higher CAR were significantly more likely to have a higher CRP and lower albumin concentrations. Also, patients with higher CAR had higher IL‑6 concentrations (IL‑6 concentrations stratified over CAR prognostic categories defined below are shown in Fig. 1a). Higher CAR was also associated with a higher procalcitonin, higher white blood cells, lower absolute lymphocyte count, higher platelets, higher ferritin, higher D‑dimers, higher lactate dehydrogenase, higher aspartate aminotransferase, higher alanine aminotransferase, higher gamma-glutamyl transferase, higher alkaline phosphatase, lower prothrombin time, lower hemoglobin and lower eGFR (P < 0.05 for all analyses).

Fig. 1
figure 1

a Median IL‑6 concentrations and b 30 days from admission survival stratified according to the C‑reactive protein to albumin ratio (CAR) prognostic categories

Full size image

Charlson comorbidity index was similar regarding CAR level (P = 0.774); however, patients with higher CAR had a lower number of drugs in chronic treatment, were less likely to have hyperlipoproteinemia, congestive heart failure, coronary artery disease, previous myocardial infarction, chronic liver disease, liver cirrhosis and chronic obstructive lung disease but were more likely to have dementia (P < 0.05 for all analyses).

Clinical outcomes during and after hospitalization and their associations with CRP to albumin ratio

Associations of CAR with clinical outcomes are presented in Table 2. Median hospital stay did not significantly differ between patients with lower and higher CAR values; however, patients with higher CAR were significantly more likely to require intensive care unit treatment, high-flow oxygen therapy, and mechanical ventilation. Higher CAR was associated with prolonged immobilization (defined as ≥ 7 days without bathroom privileges) and development of deep venous thrombosis. Conversely, higher CAR patients were less likely to experience arterial thrombosis, driven by lower frequency of myocardial infarction and without significant difference in cerebrovascular insult rates. Patients with higher CAR were also more likely to experience bacterial sepsis during hospital stay. There was no significant difference in bleeding nor pulmonary embolism rates. Patients with higher CAR on admission stratified at median were significantly more likely to experience inferior 30-day survival rates (52% vs. 75%; P < 0.001) and lower 6‑month posthospital discharge survival rates (88.6% vs. 93.8%; P = 0.014).

We further investigated prognostic properties of CAR for 30-day survival. When stratified at deciles, CAR showed a gradually worsening prognosis with each increasing decile (Supplementary Figure S1). By analyzing the course of survival curves and associated hazard ratios, we were able to categorize patients into 4 categories with a distinct prognosis, presented in Fig. 1b, and that corresponded to CAR values of < 1.0, 1.0–2.9, 3.0–5.9 and ≥ 6.0. Patients belonging to these groups experienced 30-day survival rates of 86%, 68%, 58% and 43%, respectively. There were mutually significant survival differences present between all four categories. Associated unadjusted hazard ratios in comparison to the first category were 2.47 (95% CI 1.88–3.25), 3.42 (95% CI 2.63–4.49) and 5.38 (95% CI 4.11–7.05) for 2nd, 3rd and 4th prognostic categories.

We could demonstrate robust independent prognostic properties of CAR prognostic categories in a series of multivariate Cox regression models. CAR prognostic categories remained significantly associated with 30-day survival after adjusting for age, sex, Charlson comorbidity index, COVID-19 severity and duration of symptoms on admission as shown in Supplementary Table S1. CAR prognostic categories also remained significantly associated with 30-day survival when compared to other relevant COVID-19 prognostic scores (WHO COVID-19 severity, CURB-65, VACO index) as shown in Supplementary Table S2. A similar model with prognostic indices stratified to their prognostic categories is shown in Supplementary Table S3. These analyses show that CAR prognostic properties are independent of disease severity and duration, age and comorbidities as well as of other prognostic scores.

Validation of CRP to albumin ratio prognostic properties in an independent cohort of patients

We further analyzed robustness of association of CAR prognostic categories with survival in an independent cohort of 1155 patients subsequently treated in our institution in the period from 3/2021 to 6/2021. In comparison to patients in the development cohort, patients in the validation cohort were significantly younger (median age 69 vs. 73 years; P < 0.001), had lower Charlson comorbidity index (median 3 vs. 4; P < 0.001), higher frequency of severe or critical COVID-19 on admission (91.5% vs. 85.8%; P < 0.001) but were of similar sex distribution (58.4% vs. 57.4% males; P = 0.602). The 30-day survival rate in the validation cohort was 68.3%.

Patients in the validation cohort had similar CAR levels on presentation as patients in the development cohort (median 27.8 vs. 29.2; P = 0.620). When stratified to CAR prognostic categories defined above, significantly different 30-day survival could be observed in the validation cohort as well (P < 0.001), with distinct survival course associated with each category as shown in Supplementary Figure S2. In the multivariate analyses, CAR prognostic categories retained statistical significance and demonstrated same prognostic properties in the validation cohort as observed in the development cohort as shown in Supplementary Table S1 and Supplementary Table S2.

Discussion

CRP and albumin seem to depict different pathophysiologic processes and different aspects of inflammation. In contrast to CRP where elevation reflects the severity of acute phase inflammatory response, decrease in albumin can reflect different processes including inflammation, nephrotic range proteinuria, damaged liver function and worse nutritional status [21]. As we demonstrate, both CRP and albumin are independent of age, sex and each other associated with adverse outcome, providing rationale to investigate their ratio as a separate parameter. In indirect comparison to CRP alone, CAR seems to have better prognostic properties achieving higher AUC values in the multivariate models presented in Supplementary Table S1 (AUC for CAR 0.768 vs. 0.764; data not shown). CAR is associated with a number of unfavorable features in hospitalized COVID-19 patients. Among them, older age, male sex, more severe COVID-19 and worse functional impairment are well known negative prognostic factors that reflect worse clinical outcomes. CAR is also associated with features of stronger inflammation (elevated IL‑6, D‑dimers, WBC …) and other liver blood test derangements. It is notable, however, that higher CAR values are observed among patients with longer duration of symptoms at the time of hospital admission and no significant association with bilirubin levels are present. These findings imply that more developed inflammatory response but not functional liver injury is underlying this parameter. Thus, it seems that higher CAR patients have specific, inflammatory phenotype with more severe COVID-19 on presentation. Duration of symptoms did not affect prognostic associations of CAR as shown in the multivariate models where both factors were independently associated with worse survival (higher CAR and shorter duration of symptoms). Higher CAR patients were also more likely to receive specific therapies, probably owing to more severe disease presentation; however, after including specific drugs into multivariate analyses they did not change associations of CAR with worse clinical outcome (data not shown). Also, there was a similar comorbidity burden as assessed by Charlson comorbidity index between patients with lower and higher CAR values and this parameter seems to be less affected by comorbidities and more by acute inflammatory state induced by SARS-CoV‑2 infection. Specific metabolic comorbidities associated with unfavorable clinical course of COVID-19 did not group with higher CAR but were even more frequent in patients with lower CAR values (hyperlipoproteinemia, congestive heart disease, coronary artery disease, history of myocardial infarction, chronic obstructive lung disease, chronic liver disease and liver cirrhosis). Only comorbidity associated with higher CAR was dementia, probably due to the older age of these patients.

Worse clinical course including more rapid respiratory deterioration and higher need for intensive level of care as well as higher mortality, have been previously reported to be associated with higher CAR in smaller cohorts of COVID-19 patients [10,11,12,13]. Similar observations regarding disease severity were established in non-COVID community-acquired pneumonia patients as well [22,23,24]. Our paper validates these findings but also extends these observations to demonstrate that survival is gradually improved with lower CAR values. We were able to define four novel CAR prognostic categories that add up to prognostication of COVID-19 patients. These categories predict survival independently of established prognostic scores CURB-65 and VACO index as well as independently of age, comorbidity burden and COVID-19 severity on admission. We also show that CAR reflects on postdischarge mortality as well. Substantial advantage of laboratory dependent parameters in comparison to complex prognostic indices is less cumbersome procedure of calculation and ability to quickly classify patients when blinded for information regarding specific comorbidities and other included parameters. Its use in adjunction to established prognostic scores can help improve prognostication of COVID-19 patients. CAR was also shown to be predictive of worse functional impairment and higher occurrence of deep venous thrombosis and bacterial sepsis. Considering these associations, CAR could reflect more severe COVID-19 which was shown to be associated with higher occurrence of venous thrombosis [25, 26]. Conversely, probably due to a baseline association with a more favorable cardiovascular risk profile, acute arterial thromboses were less frequent in patients with higher CAR. This finding could reflect the fact that patients presenting with acute arterial thrombotic events might not have had the time to develop full inflammatory profile of their disease and presented earlier during the disease course.

Main limitations of our work are retrospective design, single-center experience and lack of longitudinal assessment of CAR. No P-value adjustments for multiple testing were performed when comparing characteristics and outcomes between two group of patients (higher and lower CAR values). Main strengths of our work are a large sample size and validation of findings in an independent cohort of patients treated in the same institution. This was possible due to high volume of patients and represents largest single-center cohort reporting CAR-related clinical and prognostic aspects of COVID-19 so far. Our findings are representative of tertiary center experience with high proportion of patients with severe or critical disease at the time of admission and might not be generalized to other clinical contexts.

In conclusion, among hospitalized COVID-19 patients, higher CAR at the time of hospital admission is associated with older age, male sex, parameters reflecting stronger systemic inflammation and more severe COVID-19. Higher CAR is associated with higher 30-day and postdischarge mortality. We defined and validated four CAR prognostic categories with robust prognostic properties for 30-day survival independent of disease severity, age and comorbidities. CAR seems to have a good potential to improve prognostication of hospitalized COVID-19 patients.