Study design

The methods and results of the TICO trials of bamlanivimab (Ly-CoV555) [9], sotrovimab [5], amubarvimab/romlusevimab (BRII-196/BRII-198) [5], tixagevimab-cilgavimab [10], and ensovibep [11], as well as the ITAC trial of hIVIG [12] have been previously reported. For all trials, patients were eligible if they were 18 years or older, had a positive test for SARS-CoV-2 infection, and required acute hospitalization for COVID-19 without need for invasive mechanical ventilation or end-organ damage. Detailed inclusion and exclusion criteria for each trial are presented in their respective publications. Participants missing baseline neutralizing antibody and antigen measurements were excluded from the present study (4% of participants across TICO and ITAC).

Primary outcome

In the TICO trials, patients were followed for 90 days after randomization for the primary outcome of sustained recovery. Sustained recovery was defined as being discharged to home or the same level of care the patient required prior to COVID-19 diagnosis (e.g., long-term care hospital) and remaining there for 14 consecutive days. Patients in ITAC were followed for 28 days after randomization; the primary outcome was a 7-category ordinal endpoint measured at day 7 that considered pulmonary status and extrapulmonary complications. The 7 categories ranged from return to usual activities with no more than minimal symptoms due to COVID-19 to death and were based on similar outcomes in earlier influenza and COVID-19 studies. A key secondary outcome in the ITAC trial was time to discharge or the most favorable category on this ordinal scale [12]. In the present study, this secondary outcome was treated as the “time to sustained recovery” for patients in the ITAC trial [12].

Secondary outcomes

Key secondary outcomes in the TICO trials included 90-day mortality and a composite safety outcome of grade 3 and 4 adverse events, organ failure or serious co-infection, serious adverse events, or death. In the TICO trials, grade 3 and 4 adverse events were measured through day 28, whereas mortality was assessed through day 90. Thus, we also compared 90-day mortality and composite safety outcome measured through day 28 across treatment groups in this analysis. In the ITAC trial, grade 3 and 4 adverse events were measured only through day 7 and mortality through day 28. Due to the difference in follow-up times, participants in ITAC who did not experience the composite safety endpoint by day 7 were censored for this outcome, and those who did not experience mortality by day 28 were censored at day 28 for the mortality outcome. Additionally, due to TICO being a platform study, several of the included trials shared concurrent placebo groups. Specifically, when a participant was eligible for more than one trial on the platform, they were first randomized to a trial and then randomized to the active agent or its respective matched placebo. Placebo participants who were eligible for more than one trial at the time of enrollment were subsequently included in the primary analyses of each of these trials. In the present study, a trial was defined as including the patients assigned to the relevant active treatment in addition to patients assigned to their matched placebos, but not the concurrent shared placebo participants from other trials. Therefore, each participant was assigned to only one trial, and as a result no participant assigned to placebo contributed more than once to the present study [13].

Baseline antibody and antigen measurements

In each of the six included trials, blood samples were collected on the day of randomization prior to administration of the study intervention. SARS-CoV-2 plasma viral nucleocapsid protein levels were measured using a single-molecule immune bead assay (Quanterix, Billera, MA, USA) and a plasma surrogate neutralizing antibody assay specific for antibodies that disrupt the SARS-CoV-2 receptor binding domain (RBD)-ACE2 interaction (GenScript, Piscataway, NJ, USA).

Statistical methods

Baseline demographics were compared across trials, with continuous variables summarized by their means and standard deviations and categorical variables summarized by frequency and proportions. A two-step individual participant data meta-analysis approach was adopted, where patient-level outcomes were first analyzed in each trial separately. The trial-level results were then pooled in a fixed (common) effect meta-analysis [14]. These results were then compared to those from one-stage models fit directly on the pooled individual participant data.

Between-trial heterogeneity was assessed visually in forest plots in conjunction with the Cochran Q-test for heterogeneity and I² statistic.

Within each trial, Fine-Gray models were used to compare time to sustained recovery across treatment groups while accounting for the competing risk of death. Cox proportional hazards models were used to compare time to the composite safety outcome and time to death across treatment groups. An interaction between treatment assignment and baseline serostatus was tested to assess whether the treatment effect differed based on whether the participant tested positive for SARS-CoV-2 neutralizing antibodies at baseline. We then conducted separate analyses for the baseline seropositive and seronegative groups testing whether the effect of treatment on the rates of sustained recovery and composite safety events differed across levels of continuous antigen measurements. These differences were assessed within each trial using an interaction between treatment assignment and transformed antigen values. Antigen measurements were first log₂ transformed. Natural cubic splines with internal knots at the 33rd and 66th percentiles of the log₂ transformed antigen measurements were then used to allow for non-linear relationships between log₂ transformed antigen and the relative rates of outcomes as well as the treatment effects; the boundary knots were set to coincide with the overall range of antigen measurements.

The resulting interaction coefficients from each of the above analyses were then pooled in a multivariate meta-analysis model along with the main effect of treatment to allow for the subsequent estimation of covariate-specific treatment effects. A fixed treatment effect was assumed in each analysis due to the small number of studies, similarity in the treatments assessed (all small molecules for use in passive immunotherapy) and similarity in the trial designs; the five TICO trials included were conducted under the same protocol.

Three sets of sensitivity analyses were conducted. The first removed the ensovibep (DARPin) and ITAC (hIVIG) trials to examine the results when only including trials assessing monoclonal antibody treatments. The second removed the tixagevimab-cilgavimab trial, by far the largest trial, to assess the degree to which the overall conclusions were driven by the result of this trial. The third assessed the risk of aggregation bias due to including the main effects of treatment in the multivariate meta-analysis models. Including the main effects can introduce between-study information into the estimation of the interaction coefficients if the association between an individual-level characteristic (e.g., baseline serostatus) and the treatment effect within trials differs from the association between the aggregated trial-level characteristic (e.g., proportion seropositive at baseline) and the trial-level treatment effects that would be observed in a meta-regression [15,16]. To assess this, separate models were fit pooling only the interaction terms, thus using only within-trial information; the point estimates for the interaction coefficients and the p-values for the test of the interaction were then compared between approaches. For the interaction between treatment and continuous log₂ transformed antigen levels, estimated sustained recovery rate ratio (RRR) curves were compared when pooling the main effect of treatment jointly with the interaction coefficients to when the main effect was assumed to be independent from the interaction terms. This independence assumption avoids incorporating between-study information into the estimated interaction coefficients [15].

All analyses were conducted using R version 4.1.2 [17]. The “cmprsk” [18] and “survival” [19] packages were used to fit the Fine-Gray and Cox proportional hazards models, respectively. The “mixmeta” package was used to fit all meta-analysis models [20]. No correction for multiple comparisons was performed, and p-values less than 0.05 were considered statistically significant for all analyses.

Ethical considerations

The study was conducted according to the Declaration of Helsinki in its current version; the requirements of Good Clinical Practice (GCP) as defined in Guidelines, EU Clinical Trials Directive (2001/20/EC), and EU GCP Directive (2005/28/EC); International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) Guidelines; Human Subject Protection and Data Protection Acts; the US Office for Human Research Protections (OHRP); or with the local law and regulation, whichever afforded greater protection of human subjects.

Characteristics of the population of each included trial

The six trials included in this meta-analysis enrolled a total of 3,079 patients with baseline antibody and antigen measurements with trial size ranging between 256 in the amubarvimab/romlusevimab trial and 1,302 in the tixagevimab-cilgavimab trial (Table 1). Enrollment in the trials took place during the second half of 2020 and early 2021 for bamlanivimab ([9], sotrovimab [5], amubarvimab/romlusevimab [5], and hIVIG [12] and February 2021 to the second half of 2021 for tixagevimab-cilgavimab and ensovibep [11]. The duration of symptoms was between seven and eight days prior to enrollment. Compared to patients in the other trials, those in the tixagevimab-cilgavimab and ensovibep trials tended to be younger and, based on the “baseline pulmonary ordinal scale category,” had more severe disease at enrollment [10]. They were also more likely to have been treated with corticosteroids or immune modulators at baseline. Reflecting the different enrollment periods, the percentage of patients who had received at least one dose of an anti-SARS-CoV-2 vaccine was minimal for bamlanivimab, sotrovimab, amubarvimab/romlusevimab, and hIVIG trials ranging from 1.9% to 8.4%. The corresponding percentages for tixagevimab-cilgavimab and ensovibep trials were 28% and 31.2%, respectively.

At trial entry, the percentage of patients seropositive for endogenous neutralizing antibodies ranged from 40.6% to 59.4%, the highest in the ensovibep trial followed by the tixagevimab-cilgavimab trial (53.4%). The overall median plasma antigen level was 1,421 pg/mL (IQR: 231–4,568), ranging from 1,015 pg/mL (bamlanivimab trial) to 1,656 pg/mL (tixagevimab-cilgavimab trial). Additional characteristics of the patients included in each trial are summarized in Table 1, and further description can be found in the original publication of each trial [5,9–12].

Treatment effect by antibody status at trial entry

(a) Sustained recovery.

By the end of the 90-day follow-up period (28 days for hIVIG), sustained recovery was achieved by 1,494 of 1,710 patients (87%) assigned to receive active treatment, and by 1,170 of 1,369 patients (85%) assigned to placebo. Among participants seronegative at baseline for neutralizing antibodies, this was 702 of 815 (86%) assigned to active treatment, and 566 of 692 (82%) assigned to placebo; among seropositive participants, 792 of 895 (88%) assigned to active treatment and 604 of 677 (89%) assigned to placebo achieved sustained recovery. The forest plot in Fig 1A displays the treatment effect, as measured by the sustained RRR for treatment versus placebo. The overall summary estimate of the RRR was 1.06 (95% CI [0.99,1.14]). There was little evidence of between-trial heterogeneity in the treatment effect on the rate of sustained recovery (Q = 0.89 [p = 0.97]; I² = 0.00). Study-specific cumulative incidence curves for sustained recovery are shown in S1 Fig. The treatment effect varied significantly by baseline neutralizing antibody serostatus with greater benefit among seronegative patients; the summary estimate of the RRR was 1.16 (95% CI [1.04,1.29]) among seronegative patients and 0.97 (95% CI [0.88,1.07]) in seropositive patients (Fig 1A). The estimated summary RRR among seronegative patients was 20% greater than among seropositive patients (95% CI [4%, 38%]; pooled interaction p = 0.02) (Fig 1B). Similar to the overall treatment effect, there was little evidence of heterogeneity in the difference in the RRRs between baseline seronegative and seropositive participants across the trials (Q = 3.95 [p = 0.56], I² = 0.0%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Sustained recovery by baseline SARS-CoV-2 neutralizing antibody status.

(a) Forest plot of recovery rate ratio (RRR) comparing treatment arm versus matched placebo by neutralizing antibody status at study entry. (b) Forest plot of the relative recovery rate ratio (rRRR) comparing the estimated treatment effect in neutralizing antibody seropositive versus seronegative patients at study entry.

https://doi.org/10.1371/journal.pmed.1004616.g001

(b) Mortality and composite safety outcome.

There were 144 deaths among 1,710 patients (8.4%) randomized to passive immunotherapy (seronegative: 73/815 [9.0%]; seropositive: 71/895 [7.9%]) and 138 deaths among 1,369 patients (10.1%) randomized to placebo (seronegative: 86/692 [12.4%]; seropositive: 52/677 [7.7%]). The overall summary (fixed effect pooled estimate) of the hazard ratio (HR) for mortality comparing treatment versus placebo was 0.81 (95% CI [0.64,1.03]). As observed with sustained recovery, there was little evidence of between-trial heterogeneity in the treatment effect (Q = 3.13, p = 0.68, I² = 0.0%). Study-specific cumulative incidence curves for mortality are shown in S2 Fig. Consistent with the findings regarding the heterogeneity of the treatment effect on sustained recovery by baseline neutralizing antibodies serostatus, the estimated treatment effect on mortality was larger in seronegative than in seropositive patients, though this difference was not statistically significant (pooled interaction p = 0.18; Q = 4.54 [p = 0.47], I² = 0.0%). The summary HR was 0.69 (95% CI [0.50,0.95]) in seronegative patients and 0.96 (95% CI [0.66,1.39]) in seropositive patients (Fig 2A). The HR in seronegative patients was 28% lower than the HR among seropositive patients (95% CI [-56%, 17%]; pooled interaction p = 0.18) (Fig 2B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Mortality by baseline SARS-CoV-2 neutralizing antibody status.

(a) Forest plot of hazard ratio (HR) for mortality comparing treatment arm versus matched placebo by neutralizing antibody status at study entry. (b) Forest plot of the relative hazard ratio (rHR) comparing the estimated treatment effect in neutralizing antibody seropositive versus seronegative patients at study entry.

https://doi.org/10.1371/journal.pmed.1004616.g002

Similar results were found when the day 28 composite safety outcome was considered (S3 Fig). The overall summary HR comparing treatment group to placebo was 0.89 (95% CI [0.66,1.21]; Q = 3.47 [p = 0.63], I² = 0.0%), and it was 0.83 (95% CI [0.70,0.99]) and 1.04 (95% CI [0.86,1.26) in seronegative and seropositive patients, respectively. Study-specific cumulative incidence curves for composite safety are shown in S4 Fig. The summary HR was 20% lower among seronegative patients compared to seropositive patients (95% CI [-38%, 3%]; pooled interaction p = 0.09; Q = 3.99 [p = 0.55], I² = 0.0%), though this was again not statistically significant. Similar one-stage models fit directly on the individual participant data for each of the above analyses yielded similar results.

Treatment effect by baseline plasma antigen levels

(a) Sustained recovery.

The estimated summary RRR for sustained recovery, with a 95% confidence band, is plotted against baseline plasma antigen level in Fig 3A; the corresponding trial-specific curves are shown in S5 Fig. Overall, there was no evidence that the treatment effect on sustained recovery varied by baseline plasma antigen levels (pooled interaction p = 0.34) with the 95% confidence band of the estimated RRR curve including 1 across all levels of baseline antigen levels.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Pooled recovery rate ratio (RRR) for sustained recovery comparing treatment arm vs. matched placebo by baseline plasma antigen level.

(a) Overall Study, (b) By neutralizing antibody status at study entry. The points show the observed distribution of baseline antigen measurements in patients. The dashed lines represent the 33rd and 66th percentiles of antigen measurements across trials in the respective patient group (overall, seropositive, seronegative); these correspond to the locations of the internal knots for the restricted cubic splines. The dotted lines represent the 10th and 90th percentiles of antigen measurements in the respective group.

https://doi.org/10.1371/journal.pmed.1004616.g003

As the treatment effect on sustained recovery differed by baseline neutralizing antibodies serostatus, the association of the treatment effect on sustained recovery by baseline plasma antigen level was also considered separately by baseline neutralizing antibody serostatus. The estimated curves of RRR by antigen levels for the seropositive and seronegative subgroups of patients are shown, for each trial, in S6 and S7 Figs, respectively. The corresponding summary RRR curves, pooled across all trials, are shown in Fig 3B. In seropositive patients, there was no indication of any treatment effect on sustained recovery across all baseline antigen levels. The 95% confidence band of the estimated RRR curve included the value of 1 with the summary RRR being roughly 1 across all levels of baseline plasma antigen levels; the p-value for the interaction was equal to 0.58.

Similarly, in the seronegative sub-group, there was no evidence that treatment effect differed according to antigen level (p-value for interaction = 0.29; Fig 3B). Excluding the ensovibep and hIVIG trials (S8 Fig) or the tixagevimab-cilgavimab trial (S9 Fig) resulted in similar results as above for both the neutralizing antibody seropositive and seronegative groups. Using only the within-trial information when estimating the pooled estimated RRR curves also led to similar results within both the seropositive and seronegative groups (S10 Fig). The results were also similar when excluding patients with high-flow nasal cannula or noninvasive ventilation for both the seropositive and seronegative groups (S11 Fig). It should be noted though that none of the patients enrolled in the sotrovimab and the amubarvimab/romlusevimab trials were in this category (Table 1).