Introduction

As of May 2021, COVID-19 has infected more than 170 million people worldwide and caused the death of more than 3.5 million people [1]. Death is largely due to the inflammatory escalation or host cytokine storm secondary to SARS-CoV-2 infection [2]. This inflammatory response has clearly analytically identifiable components [3]. Clinically, most individual patients showed the changes of lymphocyte counts, D-dimer, interleukin-6 (IL-6), C-reactive protein (CRP), ferritin, LDH, etc. Why some patients become more inflamed than others is not known for certain but it is possible that there is a genetic background that facilitates this. Among those patients who present with this exaggerated inflammatory response, different degrees of inflammation have been described that predict the short-term future of these patients [4]. However, this inflammation is far from being homogeneous in all patients. From these studies, we know that the more parameters of analytical inflammation the worse the prognosis. The next step is to find out if some parameters have more prognostic value than others. In fact, in clinical practice, we see how some patients start their analytical inflammation in one way and others in another. This suggests different pathways and perhaps also different prognoses.

The present study aims to describe the different clusters or differentiated groups of inflammation in COVID-19 and to analyze their prognostic role.

Materials and methods

Study design, patient selection, and data collection

The present study is a retrospective observational study of consecutive patients included in the Spanish SEMI-COVID-19 Registry, created by the Spanish Society of Internal Medicine (SEMI). This is a multicenter, nationwide registry with over 150 hospitals registered so far. From March 1, 2020 to May 1, 2021, 21,962 hospitalized patients were included in the Registry. The characteristics of this registry have been detailed in previous reports [5]. In brief, all included patients were diagnosed by polymerase chain reaction (PCR) test or rapid antigenic test for SARS-CoV-2 taken from a nasopharyngeal sample, sputum, or bronchoalveolar lavage. The collection of data from each patient in terms of sociodemographic data, comorbidities, laboratory data, treatments, and outcomes was verified by the principal investigator of each center through the review of clinical records. All participating centers in the register received confirmation from the relevant Ethics Committees, including Bellvitge University Hospital (PR 128/20). Informed consent was obtained from the subjects.

The inclusion criteria were all patients in the registry with a community (non-nosocomial) SARS-CoV-2 infection and belonging to the high-risk category of inflammation according to our previous report and based on the lab test on admission [4]. This category of high risk is based on the decrease in the lymphocyte count 101.5 mg/L, lactate dehydrogenase (LDH) > 394U/L, ferritin > 1359.9mcg/L, and D-dimer > 1580 ng/mL.

The treatments received were in accordance with the medical guidelines available at the time of the pandemic [6,7,8,9,10,11]. In the absence of clinical evidence of any of the treatments at the initial time of the pandemic, their use was allowed off-label.

Outcomes definition

The primary outcome of the study was in-hospital mortality. Secondary outcomes were the requirement of high-flow nasal cannula (HFNC), non-invasive mechanical ventilation (NIMV), invasive mechanical ventilation (IMV), or ICU admission.

Statistical analysis

Multiple imputations of missing data were performed. Categorical variables were expressed as absolute numbers and percentages. Continuous variables are expressed as mean plus standard deviation (SD) in the case of parametric distribution or median [IQR] in the case of non-parametric distribution. Differences between groups were assessed using the Chi-square test for categorical variables and ANOVA or Kruskal–Wallis test as appropriate for continuous variables. p values < 0.05 indicated statistical significance.

The cluster analysis was performed by ascendant hierarchical clustering on the 5 laboratory variables previously selected using Ward’s minimum variance method with Euclidean squared distance [12]. Results are graphically depicted by a dendrogram. The number of clusters was estimated by the kmeans method. The cluster analysis model was included in a binary logistic regression, taking in-hospital mortality as the dependent variable. We introduced in the multivariate model those variables with a p value < 0.10 in the univariate model. In-hospital mortality between clusters was depicted by the Kaplan–Meier curves with their logarithmic range test (event: death; censored data: hospital discharge).

Statistical analysis was performed by IBM SPSS Statistics for Windows, Version 26.0. Armonk, NY, USA: IBM Corp.

Results

General baseline data between groups

From March 1, 2020 to May 1, 2021, 20,641 patients admitted for non-nosocomial COVID-19 were included in the SEMI-COVID-19 registry. Of these, 15,691 fell into the high-risk category for inflammation according to the laboratory tests at admission and were therefore included in the present study (Figure S1).

Hierarchical cluster analysis identified 7 clearly differentiated clusters (C1–C7) (Fig. 1). The baseline characteristics of the 7 clusters are shown in Table 1. It is noteworthy that C1 patients attend the hospital earlier from the onset. C2 and C3 patients have a lower Charlson index. C4 patients are older and predominantly male and have more comorbidity (mainly hypertension and dyslipidemia). C5 patients have more comorbidity (mainly hypertension, dyslipidemia and diabetes mellitus). Patients in groups C6 and C7 are older and have more dependency and comorbidity (mainly hypertension, dyslipidemia and diabetes mellitus).

Fig. 1
figure 1

Dendrogram. Clusters based on analytical inflammatory parameters upon admission. Cluster C1 is characterized by the presence of lymphopenia, cluster C2 by elevated ferritin, and C3 by elevated LDH. Cluster C4 is characterized by lymphopenia plus elevated CRP and LDH and frequently also elevated ferritin. Cluster C5 is defined by elevated CRP, and cluster C6 by elevated ferritin and D-dimer and frequently also elevated CRP and LDH. Finally, cluster C7 is characterized by elevated D-dimer

Table 1 Patient characteristics between clusters of inflammation

Symptoms at the time of hospital admission are listed in Table S1. In particular, group C1 presents more frequently with a headache. C2 and C3 present more frequently with headaches, arthromyalgia, sore throat, cough, diarrhea, anosmia, and ageusia. C4 patients present more frequently with cough, tachypnea, and less diarrhea. Cluster C5 most frequently presents with headache, arthromyalgia, cough, anosmia, and ageusia. Clusters C6 and C7 present less frequently with fever and cough.

Laboratory tests between groups

Table S2 shows the inflammatory analytical parameters presented by each of the clusters. According to the most relevant analytical feature in each group, we can define cluster C1 by the presence of lymphopenia, cluster C2 by elevated ferritin, and C3 by elevated LDH. Cluster C4 is characterized by lymphopenia plus elevated CRP and LDH and often also elevated ferritin. Cluster C5 is defined by elevated CRP, and cluster C6 by elevated ferritin and D-dimer and often also elevated CRP and LDH. Finally, group C7 is characterized by elevated D-dimer. Figure 2 shows the distribution of each of the inflammatory parameters in each cluster.

Fig. 2
figure 2

Distribution of the parameters of inflammation between clusters. Mann–Whitney U test (p < 0.001 for all parameters). Each box defines the median and the 25th and 75th percentile. The whiskers define the dispersion, the circles the outliers and the asterisks the extreme values

It should be noted that cluster C4 is the cluster with the lowest PaO2/FiO2 on admission. This fact possibly shows a greater inflammatory component and greater evolution to ARDS.

Treatments during admission

Table S3 shows the treatments received during admission. Logically, the clusters with a greater analytical inflammatory component (C4) received corticosteroids and tocilizumab more frequently. On the other hand, clusters with higher D-dimer elevation (C6 and C7) more frequently received high doses of LMWH.

Outcomes between groups

The clusters with the highest in-hospital mortality were C4, C6, and C7 (36.8% vs. 39.3% vs. 26.4%) (Table 2; Fig. 3). It is also noteworthy that the C4 cluster with a higher analytical inflammatory component is the group with the highest requirements for HFNC, NIMV, IMV, and ICU admission. On the other hand, the C6 and C7 clusters with greater prominence of D-dimer have high mortality but not so high requirements for HFNC, NIMV, IMV, and ICU admission.

Table 2 Outcomes between clusters of inflammation
Fig. 3
figure 3

Kaplan–Meier curves of in-hospital mortality between clusters of inflammation. C1: 17.4%, C2: 18%, C3: 15.6%, C4: 36.8%, C5: 17.5%, C6: 39.3%, C7: 26.4%; p < 0.001

Risk factors for in-hospital mortality

Table S4 shows the risk factors for in-hospital mortality. These included age (OR = 1.07, p < 0.001), female sex (OR = 0.71, p < 0.001), Black race (OR = 0.19, p = 0.032), higher BMI (OR = 1.02, p < 0.001), time from disease onset to admission (OR = 0.96, p < 0.001), higher degree of dependency (OR = 1.39, p < 0.001 for moderate and OR = 1.66, p < 0.001 for severe), ischemic heart disease (OR = 1.32, p < 0.001), higher Charlson index (OR = 1.14, p < 0.001), lower PaO2/FiO2 (OR = 0.99, p < 0.001), tachypnea on admission (OR = 2.72, p < 0.001), and inflammation clusters. In detail and, having cluster C1 as reference, the model revealed a worse prognosis for all other clusters: C2 (OR = 1.30, p = 0.001), C3 (OR = 1.14, p = 0.178), C4 (OR = 2.28, p < 0.001), C5 (OR = 1.07, p = 0.479), C6 (OR = 2.29, p < 0.001), and C7 (OR = 1.28, p = 0.001). The drugs received during admission (steroids, remdesivir, and tocilizumab) were included in the uni- and multivariate models due to their clinical importance and statistical influence on the rest of the variables but, despite reaching statistical significance in the multivariate model, their results must be interpreted with caution since this is not the type of study to assess their efficacy. Asthma was found to be a protective factor for in-hospital death.

Discussion

In the present study, we demonstrate the presence of 7 clearly differentiated inflammation clusters. It is clear that COVID-19 elicits an exaggerated inflammatory response but it is also clear that patients do not all become inflamed in the same way. Recently, our group had highlighted risk categories based on the degree of analytical inflammation in patients with COVID-19 at the time of hospital admission [4]. We know from that research the additive importance of such analytical inflammatory criteria. We now also know from the present study that it is important not only how many criteria the patient meets but also which ones. And this differentiation in the type of inflammation can be predicted at the time of admission. Afterward, the cascade of inflammation will continue, but the onset of inflammation already translates into differential characteristics and a very important predictive power when it comes to directing our strategies as clinicians.

The fact that the clusters with the worst prognosis C4, C6, and C7 are also those that occur in the older population suggests that age alone is not only a risk factor, but also that this is a population that is more inflamed. Whether this is due to a failure in the autoregulation of inflammatory mechanisms inherent to age should be answered with further studies. They are also clusters with higher associated comorbidity (especially C4) and one would think that those associated diseases provide an inflammatory basis that could influence. However, the C5 cluster also has considerably associated comorbidity and yet does not have as much inflammation at entry.

Inflammation clusters add to the list of factors associated with increased in-hospital mortality in COVID-19 [13,14,15]. All the clusters have a worse prognosis compared to cluster C1, which we could consider the reference cluster. However, we should focus our efforts on detecting the C4, C6, and C7 clusters first, as they are really the ones with the worst prognosis. The C4 cluster presents a greater inflammatory component in the form of lymphopenia, elevated CRP and LDH, and very frequently, also elevated ferritin. It is a cluster that will evolve to acute respiratory distress syndrome (ARDS) and will require more HFNC, NIMV, IMV, and ICU admission. On the contrary, cluster C7 is a cluster with a prominence of D-dimer. It also has high mortality but clearly requires less HFNC, NIMV, IMV, and ICU admission. This could translate into higher mortality from thromboembolic disease and less progression to ARDS. Finally, cluster C6 is a mixture of clusters C4 and C7, i.e., a cluster with a prominence of D-dimer but also of ferritin and, very frequently, CRP, LDH, and lymphopenia.

Whether these 7 clusters respond to different pathways of inflammation and perhaps respond to genetic differences that condition a different host response to the same external aggression of SARS-CoV-2 is something that we cannot answer with the present study. As we have discussed previously, the inflammation cascade in patients with COVID-19 does not occur equally in all patients. In Fig. 4, we propose a logical sequence in the inflammation of these patients. It is based on the analytical characteristics of each of them and the additive possibilities. It also has a pathophysiological logic behind it as the initiation of the inflammatory cascade in COVID-19 is defined by macrophage activation. This includes elevation of LDH, IL6, CRP, ferritin, and cytopenias. Why some patients elevate some parameters earlier and other patients elevate others is unknown. After this first step, vascular endothelial damage occurs and there appears the elevation of D-dimer.

Fig. 4
figure 4

Proposed algorithm for analytic inflammatory pathways in patients with COVID-19. It is based on the analytical characteristics of each of them and the additive possibilities. The initiation of the inflammatory cascade in COVID-19 is defined by macrophage activation. This includes elevation of LDH, IL6, CRP, ferritin, and cytopenias. After this first step, vascular endothelial damage occurs and there appears the elevation of D-dimer. Cluster C7 suggests that there is a group of patients who present endothelial damage with subsequent micro- or macrothrombosis without going through a previous step of major inflammation

From such a model, there are several comments to consider. First, there are 3 clusters that should sound the alarm when detected, such as clusters C4, C6, and C7. They translate more advanced stages of the disease and early action should be taken to block such escalation. On the other hand, cluster C2 does not represent such an advanced stage of inflammation but possibly the transition to cluster C6 is just around the corner in these patients (compared to clusters C1, C3, and C5 where we would have more room for maneuver). Thus, ferritin and D-dimer should be the key indicators of increased risk in patients hospitalized for COVID-19. Finally, cluster C7 suggests that there is a group of patients who present endothelial damage with subsequent micro- or macrothrombosis without going through a previous step of major inflammation. This is a puzzling fact that should make us rethink the doses of LMWH in patients with little inflammation.

Of all the treatments tested in COVID-19, it is the immunosuppressants/immunomodulators (corticosteroids and tocilizumab) that have demonstrated the greatest effectiveness to date [16,17,18,19,20,21,22,23,24,25]. They are especially indicated in those patients with analytical parameters of inflammation such as the patients included in the present study. However, it is plausible to think that their effectiveness is not the same in all inflammation clusters. We believe that perhaps in C4 (with a greater inflammatory component) and even C6 they could clearly demonstrate their benefit but perhaps in C7, with a greater role of D-dimer, they might not be so useful.

In conclusion, the present study identifies 7 inflammation groups based on the presence of lymphopenia, elevated CRP, LDH, ferritin, and D-dimer at the time of hospital admission for COVID-19. Clusters C4 (lymphopenia + LDH + CRP), C6 (ferritin + D-dimer), and C7 (D-dimer) were the worst prognostic clusters in terms of in-hospital mortality.