Kinetics of anti-SARS-CoV-2 antibodies and hematological parameters in hospitalized pre-vaccination COVID-19 patients in Peru

Study population and sample

This retrospective observational cohort study included a total of 157 serological samples were collected from 44 hospitalized COVID-19 patients at the Hospital Nacional Guillermo Almenara Irigoyen in Lima, Peru, between August and October 2020, during the first wave of infections and prior to the start of vaccination programs in Peru. The diagnosis was confirmed through antibody immunochromatography tests or qPCR. Samples were properly coded, labeled, and stored at −80 °C in the Cell Culture and Immunology Laboratory at the Universidad Científica del Sur in Lima, Peru, Pons et al. (2021).

Patients were classified as having critical disease according to the Clinical Management of COVID-19 guidelines by the World Health Organization (WHO). This classification includes individuals with acute respiratory distress syndrome (ARDS). Based on the degree of oxygenation impairment and ICU admission status, patients were further stratified into three groups: 10 moderate ARDS, 24 severe ARDS non-ICU, and 10 severe ARDS ICU. Moderate ARDS was defined by a PaO₂/FiO₂ ratio between 100 mmHg and 200 mmHg, while severe ARDS was characterized by a PaO₂/FiO₂ ratio of ≤100 mmHg (World Health Organization, 2023). These criteria were applied uniformly to ensure a standardized classification across patient groups.

Clinical and hematological data

Data were gathered from clinical records, including age, sex, hospitalization date, progression date to disease severity, and discharge or death date. Additionally, comorbidities such as hypertension, coronary disease, chronic respiratory disease, diabetes, and obesity were documented. hematological parameters were also collected, including levels of leukocytes, lymphocytes, hemoglobin, neutrophils, platelets, D-dimer, prothrombin time (PT), activated partial thromboplastin time (aPTT), C-reactive protein (CRP), ferritin, creatinine, neutrophil-to-lymphocyte ratio (NLR) and oxygen saturation to fraction of inspired oxygen ratio (SpO₂/FiO₂). The clinical-epidemiological characteristics, cytokine levels, and commercial anti-Spike antibody ratios of this cohort have been previously reported (Pons et al., 2021) at baseline. In the present study, we performed additional analyses not reported previously, including the assessment of antibody kinetics using in-house ELISA assays, temporal trajectories of hematological markers, correlation matrix construction, principal component analysis (PCA), survival analysis, receiver operating characteristic (ROC) curve analysis, and the development of a predictive nomogram.

IgG and IgM antibody levels by enzyme-linked immunosorbent assay

To analyze IgG and IgM antibody levels, serological samples stored at −80 °C from the 44 hospitalized patients were used. An in-house enzyme-linked immunosorbent assay (ELISA) kit was developed using recombinant SARS-CoV-2 Spike S1 and RBD proteins to detect anti-S1 and anti-RBD IgG and IgM antibodies. Following a protocol adapted from Stadlbauer et al. (2020), 96-well plates were coated with 50 µL of recombinant protein (one µg/mL) and incubated overnight at 4 °C, followed by PBS-Tween washing. Plates were then blocked with PBS-Tween-20 containing 5% skim milk for 2 h. Heat-inactivated serum samples (56 °C for 30 min, diluted 1:101 in PBS-Tween) were added to the plates for a 2-h incubation. Serum samples were heat-inactivated at 56 °C for 30 min, diluted 1:101 in PBS-Tween, and added to the plates for a 2-hour incubation. After washing, goat anti-human IgG or IgM conjugated with horseradish peroxidase (HRP) was added as the secondary antibody and incubated for 1 h. The plates were then washed again, the o-phenylenediamine dihydrochloride (OPD) substrate was added for a 10-minute dark incubation. The reaction was stopped with HCl, and absorbance was measured at 490 nm. This protocol was utilized to develop kits for detecting IgG and IgM antibodies against S1 and RBD proteins. The in-house ELISA assays were validated using a panel of negative sera (n = 56) and clinically confirmed SARS-CoV-2-positive samples (n = 161). Results were compared with those of an FDA-approved commercial anti-S1 IgG ELISA kit (2606-9601G; EUROIMMUN, Lübeck, Germany), demonstrating strong correlation. Intra-assay variability was assessed by testing replicates of the same samples within a single plate, while inter-assay variability was determined across plates on different days.

Statistical analysis

Microsoft Excel version 16.84 was used for data collection and organization, while RStudio version 4.3.3 facilitated statistical analysis. Descriptive statistics were presented in tables, and inferential statistics included both bivariate and multivariate analyses. Data distribution was initially assessed using Kolmogorov–Smirnov normality tests. Linear regression analyses were then performed to examine IgG and IgM antibody trends over the first 35 days of hospitalization, with mean values used to detect significant changes over time. Box plots compared IgM and IgG levels and ratios (IgM/IgG) among patient groups (moderate ARDS, severe ARDS ICU and non-ICU) over time, with group differences analyzed using the Kruskal-Wallis test (for multiple groups) or the Mann–Whitney test (for two-group comparisons), with significance set at p < 0.05.

Dispersion analyses with trend lines were conducted to explore correlations between various variables, utilizing Pearson or Spearman correlation coefficients based on data distribution. A heatmap illustrated correlations among multiple variables using Pearson coefficients. Kaplan–Meier curves were generated to assess survival probabilities and mechanical ventilation needs based on clinical and hematological parameters, with log-rank tests identifying significant group differences (p < 0.05). Principal component analysis (PCA) was performed to visualize patient distributions across clinical variables. A nomogram was developed to predict clinical events based on antibody levels and clinical parameters, supporting individualized interpretation. Finally, receiver operating characteristics (ROC) curves assessed prediction model accuracy at 4, 7, and 10 days of hospitalization, with high areas under the curve (AUC) indicating strong predictive performance.

The sample size was determined based on the availability of hospitalized COVID-19 patients during the study period. Missing data were addressed through multiple imputation when appropriate. Cases with extensive missing values were excluded from specific analyses. Since this was a retrospective cohort study, there was no loss to follow-up; however, missing data due to incomplete medical records were accounted for as described above.

Bias and limitations

To minimize selection bias, all eligible hospitalized patients from the study period were included without preselection. Information bias was controlled by using standardized medical records and validated serological assays for antibody relative quantification. However, due to the retrospective nature of the study, missing data could not be entirely avoided, and confounding factors, such as individual treatment regimens, were not fully controlled.

Ethical aspects

This research study was approved by the Institutional Review Board at Universidad Científica del Sur (No. 343-CIEI-CIENTÍFICA-2020) and Universidad San Martín de Porres (0582-2024-CIEI-FMH-USMP). Due to the health emergency, the requirement for informed consent was waived by the COVID-19 Institutional Review Board at IETSI-EsSalud (42-IETSI-ESSALUD-2020).

Clinical characteristics and comorbidities

A total of 157 samples from 44 symptomatic COVID-19 patients hospitalized at Guillermo Almenara Irigoyen National Hospital (HNGAI) were analyzed. Patients with critical disease were divided into three groups according to their severity and ICU admission: Moderate ARDS (n = 10), Severe ARDS non-ICU (n = 24), and Severe ARDS ICU (n = 10). Table 1 details clinical characteristics, pre-existing conditions, treatments, and outcomes. Median age and gender distribution were similar across groups, and 91% received corticosteroids without significant differences between groups. Mortality was higher in ICU patients (90%) compared to others (p < 0.001). Obesity was significantly more frequent in ICU patients (p = 0.043), with no differences in diabetes, hypertension, or other comorbidities. Biomarker values, including hematological, inflammatory, and respiratory parameters, were assessed longitudinally, with medians used for central tendency due to variable measurement frequencies and data distribution.

Kinetics of anti-SARS-CoV-2 antibodies and hematological parameter evolution in hospitalized patients by disease severity

The in-house ELISA for IgG anti-S1 showed a sensitivity of 92.82% (95% CI [88.03–96.12]) and specificity of 95.65% (95% CI [85.16–99.47]). For IgG anti-RBD, sensitivity was 91.16% (95% CI [86.04–94.86]) and specificity 91.30% (95% CI [79.21–97.58]). The IgM anti-S1 assay had lower sensitivity at 47.20% (95% CI [39.30–55.22]) but high specificity at 96.43% (95% CI [87.69–99.56]). IgM anti-RBD showed a sensitivity of 61.03% (95% CI [53.51–69.04]) and specificity of 94.64% (95% CI [85.13–98.88]). The intra-assay and inter-assay coefficients of variation were lower than 5% and 10%, respectively, across all analytes. The IgG anti-S1 assay correlated strongly with the commercial ELISA kit (Spearman rho = 0.85). These results support the reliability of the in-house assay for detecting SARS-CoV-2-specific antibodies, particularly IgG responses.

Levels of IgG against RBD and S1 progressively increased in all patient groups, reaching a plateau around 10 to 15 days of hospitalization. Patients with severe ARDS disease, particularly those admitted to the ICU, exhibited notably higher IgG levels compared to moderate ARDS and severe ARDS non-ICU patients, with significant differences between groups at various time points (p < 0.05). In contrast, IgM showed an early rise followed by a progressive decline from day 10 onward, with a faster reduction in moderate ARDS and severe ARDS non-ICU patients (Figs. 1 and 2).

Regarding the evolution of hematological parameters varied with disease severity, leukocytes, and the neutrophil-to-lymphocyte ratio (NLR) showed an early increase in severe ARDS patients, with persistently elevated values over time in ICU patients (p < 0.001). This pattern suggests a heightened immune and stress response in severe ARDS cases. In contrast, lymphocyte counts significantly declined in the early stages of hospitalization across all groups, followed by gradual recovery, particularly in moderate ARDS cases, which may indicate a lower inflammatory burden in these patients (p < 0.05). Inflammatory markers, including C-reactive protein (CRP), ferritin, and D-dimer, reached elevated peaks in the early days of hospitalization in severe ARDS cases, especially in ICU patients, reflecting an intense inflammatory and procoagulant response. Significant differences were observed between severity groups, particularly for CRP and ferritin during the first 10 days (p < 0.001), indicating a more aggressive activation of inflammation and coagulation pathways (Fig. 1) (Table S1).

Creatinine levels increased in severe ARDS patients, particularly in those in the ICU, suggesting renal dysfunction associated with disease severity. Although the differences did not exhibit statistical significance at certain points during hospitalization periods, an upward trend was evident, which could indicate a higher risk of renal failure in critical cases. Prothrombin time (PT) and activated partial thromboplastin time (aPTT) did not show significant changes between groups, suggesting that coagulation alterations are primarily related to inflammatory activation rather than intrinsic or extrinsic coagulation pathways. The SpO₂/FiO₂ ratio remained consistently lower in severe ARDS patients, especially those in the ICU, throughout the observation period, reflecting persistent respiratory dysfunction in these individuals (p < 0.05). This finding underscores the severity of pulmonary impairment in critical patients and the need for continuous respiratory monitoring (Fig. 1).

We further analyzed the IgM/IgG ratio for the RBD and S1 domains to assess the transition from the acute immune response, marked by IgM production, to a more stable, long-term IgG response (Fig. 2). In the early days of hospitalization, severe ARDS ICU and non-ICU patients exhibited a higher IgM/IgG ratio. However, after day 5, this ratio significantly decreased across all groups (p < 0.05), reflecting a rise in IgG and a decline in IgM. This trend was most pronounced in moderate ARDS patients, whose ratio reached the lowest values after 10 days, suggesting a faster resolution of the acute phase. In contrast, severe ARDS ICU patients maintained a higher ratio for longer, indicating extended immune activation.

Correlation analysis of antibody levels and clinical parameters in hospitalized COVID-19 patients

Correlation analysis revealed significant associations between IgM and IgG antibody levels (anti-S1 and anti-RBD) and key clinical parameters in hospitalized COVID-19 patients, stratified by disease severity (Fig. 3A).

Notably, an inverse correlation was observed between anti-S1 and anti-RBD IgG levels and the NLR (R = −0.39 and R = −0.40, respectively; p < 0.001), across all groups, particularly in severe ARDS cases. Additionally, IgG anti-S1 levels were inversely correlated with segmented neutrophil counts (R = −0.33, p < 0.001) and positively correlated with lymphocyte counts (R = 0.37, p < 0.001), suggesting an adaptive immune response influenced by disease severity (Fig. 3B, top panels).

Furthermore, a positive correlation between IgG anti-S1 levels and the SpO₂/FiO₂ ratio was observed in moderate ARDS (R = 0.41, p < 0.05) and severe ARDS non-ICU (R = 0.53, p < 0.01) patients, which indicates a link between antibody response and respiratory function (Fig. 3B, middle left panel); however, no such correlation was found in severe ARDS UCI patients. Similarly, IgM anti-RBD levels positively correlated with platelet counts (R = 0.37, p < 0.001), especially in moderate ARDS and severe ARDS ICU patients, suggesting potential associations with coagulation and inflammation pathways (Fig. 3B, middle right panel).

Additionally, CRP levels displayed a positive association with D-dimer levels (R = 0.42, p < 0.001), underscoring the interplay between systemic inflammation and coagulation activation across patient groups (Fig. 3B, bottom left panel). NLR also demonstrated robust correlations with CRP (R = 0.52, p < 0.001) and ferritin (R = 0.49, p < 0.001), particularly in moderate ARDS and severe ARDS non-ICU patients (Fig. 3B, bottom right panel), highlighting the relationship between systemic inflammation and hematological changes. Lastly, the SpO₂/FiO₂ ratio exhibited a significant inverse correlation with NLR (R = −0.39, p < 0.001), particularly in severe ARDS non-ICU patients, suggesting that elevated inflammatory responses may coincide with worsened respiratory function.

Principal component analysis of immunological and clinical parameters in hospitalized COVID-19 patients

Principal component analysis (PCA) revealed that the first two dimensions captured 41.5% of the total variance (26.8% for Dim1 and 14.7% for Dim2). In the two-dimensional space defined by these components, there was notable overlap among moderate ARDS, severe ARDS non-ICU, and severe ARDS ICU patient groups, suggesting limited differentiation based on these parameters alone (Fig. 4).

However, certain trends emerged, indicating that patients classified as Severe ARDS ICU tended to cluster at more positive values along Dim1, potentially reflecting heightened inflammatory burden and immune response. Key contributors to these dimensions included lymphocytes, anti-RBD IgG, segmented neutrophils, absolute neutrophil count, anti-S1 IgG, and the NLR, underscoring the role of immune and inflammatory responses in group separation.

Inflammatory markers such as CRP and D-dimer, along with immune response indicators (anti-RBD IgG and IgM), exhibited particularly strong associations with severe ARDS cases, aligning with established clinical patterns of COVID-19. These findings suggest that while certain biomarkers partially differentiate severity, the immune and inflammatory responses in COVID-19 remain complex, as evidenced by the overlap among clinical groups.

Survival and mechanical ventilation probability based on clinical and hematological parameters in COVID-19 patients

Kaplan–Meier curves reveal that advanced age (≥ 56 years), low hemoglobin levels (<14.1 g/dL), obesity, elevated CRP (≥ 67 mg/mL), low lymphocyte count (<9.3), high ferritin (≥877 µg/L), high absolute neutrophil counts (≥ 8.9), and elevated D-dimer levels (≥ 0.75 µg/mL) significantly impact survival and mechanical ventilation requirements in hospitalized COVID-19 patients (Fig. 5).

Patients aged 56 years or older (Fig. 5A; p = 0.0086), those with low hemoglobin (Fig. 5B; p = 0.0052), obesity (Fig. 5C; p < 0.0001), or elevated CRP levels (Fig. 5D; p < 0.0001) had significantly reduced survival and a higher likelihood of requiring ventilation. Also, low lymphocyte counts (Fig. 5E; p = 0.031), high ferritin (Fig. 5F; p = 0.0016), elevated neutrophil counts (Fig. 5G; p = 0.00056), and high D-dimer levels (Fig. 5H; p = 0.00024) were each linked to poorer survival rates and increased probability of ventilation, underscoring the impact of inflammation and coagulation abnormalities in severe ARDS COVID-19 cases. These findings highlight the prognostic value of these clinical and hematological markers, which could facilitate early risk stratification and help identify patients more likely to experience severe ARDS outcomes.

Nomogram for predicting adverse outcomes in hospitalized COVID-19 patients

The developed nomogram incorporates IgG anti-S1 levels, which indicate humoral immune response to SARS-CoV-2; obesity status (0 or 1, with 1 denoting obesity); ferritin, where high levels may suggest systemic inflammation or acute infection; and the NLR, an indicator of immune response dynamics. Each of these variables contributes a specific point value to a cumulative patient score, which is then translated into a linear predictor to estimate the probability of adverse outcomes over hospitalization.

For illustrative purposes, a patient presenting with obesity (assigned a score of 1; ∼80 points), elevated ferritin levels (e.g., 2,800 ng/mL; ∼30 points), a high neutrophil-to-lymphocyte ratio (e.g., 30; ∼60 points), and increased anti-S1 IgG titers (e.g., 2.5; ∼10 points) would yield a cumulative nomogram score of approximately 180 points. According to the model, this score corresponds to an estimated 90% probability of in-hospital mortality. As shown in Fig. 6, this example illustrates the clinical utility of the nomogram in synthesizing immunological and hematological markers into a quantitative tool for individualized risk stratification and early intervention during COVID-19 hospitalization.

ROC curves were analyzed at 4-, 7-, and 10-days post-hospitalization to assess the model’s predictive accuracy, producing AUC values of 0.97, 0.95, and 0.86, respectively. These consistently high AUC values highlight strong early predictive accuracy, although a slight reduction in AUC over time may indicate a decrease in predictive power as patients’ clinical trajectories evolve.