Patients
The cohort consisted of 3,082 patients from 680 intensive care facilities in the United States (figure 1) table 1 shows the main characteristics of the patients, stratified into three groups according to the levels of anti-SARS-CoV-2 IgG antibodies (based on the signal-to-cut ratio). Overall, 61% of the patients were men, 23% were black, 37% were Hispanic, 69% were under 70 years of age and two-thirds had received transfusions before invasive mechanical ventilation. The average number of patients per site was 2 (interquartile range, 1 to 6). The maximum number of patients from any location was 59. As shown in table 1, the three groups (patients who received plasma transfusions with high, medium and low levels of IgG antibodies) were generally similar in terms of demographic characteristics, risk factors associated with severe Covid-19 and concomitant use of therapeutic agents for Covid-19 . The percentages of patients with hypoxemia and concomitant use of hydroxychloroquine (both variables included in the adjustment models) were lower in the high-titer group than in the other two groups.
Primary outcome
Models of association between levels of anti-SARS-CoV-2 antibodies in transfused plasma and risk of death.
Death within 30 days after plasma transfusion occurred in 26.9% of all patients (830 out of 3082 patients; 95% confidence interval [CI], 25.4 to 28.5). This primary outcome event occurred in 29.6% (166 of 561 patients) in the low titer group, 27.4% (549 of 2006 patients) in the medium titer group and 22.3% (115 of 515 patients) in the high titration group. Patients in the high-titer group had a lower relative risk of death at 30 days after transfusion than patients in the low-titer group (relative risk, 0.75; 95% CI, 0.61 to 0.93) (table 2) Additional analyzes with adjustment for the patient’s demographic characteristics (age, weight and race status) and clinical characteristics (receiving invasive mechanical ventilation, use of concomitant therapy and hypoxemia) were conducted to assess the general effect of anti-SARS-CoV – 2 Level of IgG antibodies on risk of death within 30 days after transfusion (Table S1 in the Supplementary Appendix). Adjusted models (as defined in table 2) generally showed a similar association – a lower relative risk of death among patients who received plasma transfusions with high levels of anti-SARS-CoV-2 IgG antibodies (model 2, relative risk, 0.79 [95% CI, 0.65 to 0.96]and model 3 [with additional adjustment], relative risk, 0.82 [95% CI, 0.67 to 1.00]) (table 2) The findings of the sensitivity analysis in which patients were excluded at discharge were qualitatively similar to each of these findings.
Subgroup Analysis
Characteristics of patients with Covid-19 who were not receiving mechanical ventilation and who received convalescent plasma, according to the level of anti-SARS-CoV-2 IgG.
In the cohort of 3,082 patients, 2014 patients did not receive mechanical ventilation before transfusion. Table 3 shows the main characteristics of the patient from the subgroup of patients who were not receiving mechanical ventilation, stratified according to the levels of anti-SARS-CoV-2 IgG antibodies. In the subgroup of patients who were not receiving mechanical ventilation, death within 30 days after plasma transfusion occurred in 81 of 365 patients (22.2%; 95% CI, 18.2 to 26.7) in the group of low titer, 251 of 1297 patients (19.4%; 95% CI, 17.3 to 21.6) in the medium titer group, and 50 of 352 patients (14.2%; 95% CI, 10, 9 to 18.2) in the high-tier group; Table S4 shows these results in the subgroup of patients who were receiving mechanical ventilation. In the subgroup of patients receiving mechanical ventilation, death within 30 days after plasma transfusion occurred in 80 out of 183 patients (43.7%; 95% CI, 36.7 to 51.0) in the low titer group , 277 out of 666 patients (41.6%; 95% CI, 37.9 to 45.4) in the medium titre group and 64 out of 158 patients (40.5; 95% CI, 33.2 to 48, 3) in the high title group. In both subgroups, patient characteristics were well balanced across the three groups of antibody titers.
In the fully adjusted relative risk regression model, the lower risk of death at 30 days after plasma transfusion in the high-titer group than in the low-titer group was observed among patients who were not receiving mechanical ventilation before the transfusion ( relative risk, 0.66; 95% CI, 0.48 to 0.91). No effect on mortality was observed among patients who received mechanical ventilation prior to transfusion (relative risk, 1.02; 95% CI, 0.78 to 1.32).
Table S2 shows the regression of relative risk with or without complete adjustment for the patient’s demographic characteristics, levels of IgG anti-SARS-CoV-2 antibodies, clinical characteristics and study period, including all three models (the basic model, model 2 and model 3), for the subgroup of patients who were not receiving mechanical ventilation. Table S3 shows the regression of the relative risk for the subgroup of patients who were receiving mechanical ventilation.
Relative risk of death within 30 days after convalescent plasma transfusion.
Forest graphs of the relative risks of death associated with medium versus low antibody levels (Panel A) and high versus low antibody levels (Panel B) are shown. The subgroups are 12 mutually exclusive categories from the study’s time period in 2020, patient age and ventilatory support in patients who received convalescent plasma transfusions. The estimated relative risks of death are shown among patients who received convalescent plasma with signal rates for cutting IgG in the range of 4.62 to 18.45 (medium titer) or more than 18.45 (high titer), in comparison with the relative risks among those who received plasma with an IgG signal-cut ratio below 4.62 (low titer). The combined estimates for all subgroups are based on the Mantel-Haenszel estimator. Table S5 provides the sample sizes and the number of deaths in each subgroup. 𝙸 bars indicate 95% confidence intervals.
These findings were further supported by an analytical approach of stratified data that provided direct analytical control for the main variables associated with the risk of death (age, receipt of invasive mechanical ventilation and study period) (Figure 2) The pooled (or common) relative risk of death among all patients within 30 days after plasma transfusion in the high-titer group, compared to the low-titer group, was 0.80 (95% CI, 0.65 to 0.97) (Figure 2) Figure S1 shows the risk of death within 7 days after the convalescent plasma transfusion, as determined with this analytical approach to stratified data.
Exploratory Analysis
Among patients who received mechanical ventilation before transfusion, the mean number (± SD) of days between Covid-19 diagnosis and convalescent plasma transfusion was 10.0 ± 7.7; this was almost double the average number of days among patients who were not receiving mechanical ventilation (5.4 ± 4.8). Unadjusted mortality at 30 days after transfusion was lower among patients who received a transfusion at 3 days after receiving the diagnosis of Covid-19 (point estimate, 22.2%; 95% CI, 19.9 to 24, 8) than among those who received a transfusion 4 or more days after receiving a diagnosis of Covid-19 (point estimate, 29.5%; 95% CI, 27.6 to 31.6). In model 3, replacing the ventilation status with a binary classification of days for transfusion resulted in a relative risk of death of 1.18 (95% CI, 1.04 to 1.35) among patients who received a transfusion 4 or more days after receiving the diagnosis. This effect size was smaller than that observed in patients who had previously received mechanical ventilation in model 3 (relative risk, 2.16; 95% CI, 1.90 to 2.46).
The trained gradient augmentation machine was used to estimate the relationship between the main variables associated with the risk of death in 30 days after plasma transfusion and mortality in 30 days. Two methods were used to explore how this machine learning technique linked the key variables to mortality predictions.
In the first method, a graph of variable importance was generated for each variable included in the model (Fig. S2). The “importance” of the variable is the relative amount by which it improves the forecast, both in terms of location in the decision trees (where more observations are classified at the top of the decision tree) and the number of times it is used in the collection of trees. The primary variables associated with the risk of death at 30 days were age; evidence of advanced clinical evolution of Covid-19, such as receiving invasive mechanical ventilation and admission to the intensive care unit (ICU); and the level of anti-SARS-CoV-2 antibody, in order of varying importance.
The second method used to explore the association between a given variable and the mortality forecast was through a partial dependency graph. The partial dependency graph shows that, after adjusting for all other variables included in the model, the levels of anti-SARS-CoV-2 IgG antibodies maintained an inverse relationship with the risk of death. Figure S3 shows similar partial dependency plots for the primary analysis model in which antibody levels were treated as a continuous variable using a natural spline with four evenly spaced nodes. In this model, the partial dependency graph for the general sample is closely aligned with the pattern observed in the gradient reinforcement machine model. The inverse relationship with antibody levels was again seen in patients who were not receiving mechanical ventilation, and there was a general lack of a clear association in these patients.