Our population of interest included individuals at potential risk of TB, HBV, or HCV complications, who were receiving TCZ based on any clinical indication.

We searched the global WHO database of Individual Case Safety Reports (ICSRs) / adverse drug reactions (ADRs) (“VigiBase”) on 21st February 2021 (46); search terms are presented in Supplementary Table 1. We set out to identify studies reporting the risk of TB, HBV, or HCV complications in individuals taking TCZ as compared to those not receiving TCZ. We undertook a systematic literature review (on 25 February 2021), searching Medline, Embase, and Web of Science databases using search terms presented in Supplementary Table 2, in accordance with PRISMA guidelines (Supplementary Table 3). We searched all databases from 1 January 2002 to 25 February 2021. We did not apply any further restrictions (specifically, we did not impose any restrictions based on study design, as we intended to produce a comprehensive summary of the evidence surrounding this research question). We included any article (including abstracts) reporting/investigating the risk of adverse HBV, TB and or hepatitis C virus (HCV) outcome(s) associated with TCZ use. Results from each database were combined and deduplicated prior to eligibility screening. Reference lists of relevant systematic reviews/meta-analyses were also searched to identify studies for inclusion.

Studies were screened and were considered eligible for inclusion if they met pre-stipulated criteria, namely that they were observational cohort or post-marketing surveillance studies reporting both (i) the number of patients in the study sample receiving TCZ and (ii) the number of TCZ-treated patients experiencing an outcome of interest (TB or HBV reactivation, or complications of HCV infection) (Figure 3). In the case of ambiguity, a second author reviewed the study and articles were discussed thoroughly until agreement was reached. For each study, we recorded country, publication year, study design, study population, follow-up period, number of participants receiving TCZ, number of TB/HBV reactivation cases, age at baseline (mean/median), and sex. We also collated information pertaining to screening of participants for TB/HBV at baseline, as this is essential to provide context about the population treated, and is a potentially important source of bias.

Our original analysis plan was to undertake meta-analysis to calculate inverse-variance weighted pooled risk estimates for complications pertaining to TB, HBV and HCV, using individual risk estimates from identified studies. However, having collated our dataset, we did not perform meta-analysis, as studies were inconsistent in reporting the number of patients with chronic HBV and/or latent TB infection. This is important as the included studies were undertaken across a diverse range of geographical regions, and therefore HCV, HBV and TB prevalences in the underlying populations from which study samples were drawn varied substantially, with an impact on the estimated risk of reactivation. For example, a large study from a population with low prevalence of HBV and TB infections might report with high confidence that TCZ is not associated with reactivation, as it is unlikely that anyone with those infections was included in the study. Additionally, some studies offered prophylactic treatment to patients who screened positive for TB or HBV before administering TCZ, whilst others did not. Definitions of HBV/TB reactivation endpoints were also inconsistent, with varying lengths of follow-up. Instead, we calculated mean cumulative incidence estimates for HBV and TB reactivations from observational cohort studies. Individual and mean cumulative incidence estimates were displayed in forest plots constructed in R (version 4.0.2) using the “metaphor” package (version 2.4-0 [https://cran.r-project.org/web/packages/metafor/metafor.pdf]). Upper 95% confidence intervals were calculated for studies reporting zero events of interest using the rule of three, whereby the upper limit is approximated using 3/(n+1), as previously described (47). For HCV, data were insufficient for this purpose, and we have summarised data in a narrative form.

The VigiBase database (46) holds 47,205 records for complications of TCZ, of which 14,147 (30%) pertain to “infections and infestations,” with tuberculosis mentioned in 135 cases (0.9% of infections), and HBV in 42 (0.3% of infections) (Supplementary Table 1). However, further interpretation and risk quantification is not possible based on this source, as the database does not contain any denominator data (total number of individuals treated with TCZ therapy), no details of clinical or demographic history, nor the indication/dose/duration of TCZ therapy.

We identified 19 observational cohort studies in which the incidence of TB reactivation was reported, with a follow up duration of between 24 weeks and 11 years (Figure 4A; Supplementary Table 4). Studies were typically small, with a median size of 49 participants, although six studies reported on data for >1,000 individuals. The incidence of TB reactivation varied from no reported cases (in six studies; Supplementary Table 4), up to 29% (but with wide confidence intervals, 95% CI 5.4, 66.1) in a small Indian study (48). This variation is likely to reflect significant differences in risk between settings according to the population prevalence of latent TB, as well as varied approaches to screening and prophylaxis. Although the Indian study reports risk assessment and prophylaxis, they nevertheless identified reactivation events (48). In contrast, a study in South Africa—another high prevalence setting—recorded zero cases of TB reactivation (49), which is likely to reflect successful chemoprophylaxis. Several studies did not report their approach to screening and prophylaxis, nor did they provide specific information regarding TCZ dosage (Supplementary Table 4), thus making direct comparisons between studies difficult.

A mean cumulative incidence estimate from the available data puts the estimated risk of TB reactivation at 4.3% at a population level, but this figure should be interpreted with caution on the basis of the heterogeneity described above.

We identified 8 observational cohort studies in which the risk of HBV reactivation was reported, with a median of 46 study participants (range 18–159); (Figure 4B; Supplementary Table 5). Follow-up was undertaken over periods ranging from 1 month up to 9 years, with no cases of HBV reactivation reported in three cohorts, but up to as high as 13% in a Japanese cohort (50). Baseline prevalence of HBsAg and anti-HBc varied between cohorts, and approaches to screening and prophylaxis were undertaken differently between settings, with a more rigorous approach in some centres that may explain the lower incidence of reactivation events. A mean cumulative incidence estimate from the available data puts the estimated risk of HBV reactivation at 3.3% at a population level, but as for TB these figures should be interpreted with caution and individual risk assessment is key.

We identified six studies which investigated or discussed the risk of HCV complications associated with TCZ use (Supplementary Table 6). Most studies reported low risks of HCV flare associated with TCZ use, but nevertheless called for future investigations and screening of rheumatoid arthritis patients before administration of anti-rheumatic treatment. The scarcity of evidence in this field has been previously highlighted (51).

Review and guidance articles are unified in highlighting potential risks of HBV, HCV and TB infection, although with varying risk appraisal (Supplementary Table 7). Most regard the risk of HBV or TB reactivation associated with TCZ therapy as “moderate” or “intermediate.” Typically, screening is recommended for HBV and TB (±HCV) prior to starting any biologic therapy, and prophylaxis is recommended in those deemed to be at the highest risk of complications (summarised in Figure 5).

We considered undertaking meta-analysis to generate a pooled estimate of reactivation rates with potential comparison to rates “control” groups (those not receiving TCZ therapy), but unfortunately inclusion criteria and study population characteristics (including control group demographics) varied across cohorts and were not consistently reported. Therefore, it would not be robust or appropriate to calculate ratios or comparative statistics that compare infective complications across studies. Furthermore, sample sizes were small, so such an analysis would be underpowered.

Characteristics of the individual studies may have contributed to bias. Populations studied may be skewed against reactivation events (if studies are not conducted in high risk populations, or if individual patients at high risk are given prophylaxis or actively excluded from trial participation). Conversely other studies set out to enrich for individuals with prior HBV exposure such that risks for a general population may be over-estimated. Examples of each of these sources of bias are presented with details in Supplementary Table 4 (for TB) and Supplementary Table 5 (for HBV). Durations of follow-up were short in most studies, likely making them insufficient to detect all possible TB or HBV reactivation events. Baseline approach to screening and/or administration of prophylaxis for TB and/or HBV was not always reported, and was inconsistent across studies.

TCZ is gaining traction as therapy for patients at the severe end of the clinical COVID-19 spectrum, and has been incorporated into clinical recommendations (16, 17), which will result in its more widespread use in a critically ill patient group at risk of infective complications because of the global prevalence of latent TB and HBV carriage (Table 1). Our summary data demonstrate a small risk of HBV and TB reactivation overall, with estimates of 3 and 4%, respectively, for a population treated with TCZ. However, our ability to quantify these risks and undertake meta-analysis is constrained by limited and heterogeneous data and a lack of appropriate control (TCZ-untreated) comparator groups. These estimates should be considered a preliminary benchmark to inform awareness rather than absolute risk estimates. We therefore highlight the need for more data and the imperative for individualised clinical risk assessment, based on the individual patient and the population setting, to underpin prescription of TCZ therapy (Figure 5).

Consensus from existing guidelines recommend screening for carriage of HBV and TB prior to TCZ therapy. However, even where recommendations are established, screening and prophylaxis are often not consistently undertaken (41, 44), and more data are needed to better inform risk assessment and prescription of prophylactic therapy. Chemoprophylactic agents are widely available, can be administered enterally, are cheap, safe, effective, and usually well-tolerated, so these interventions generally carry a low risk. For TB, this involves prescription of isoniazid (plus pyridoxine) with or without rifampicin (53), and for HBV with entecavir, tenofovir, or lamivudine (54). In individuals with COVID-19 at high epidemiological risk of these infections, our data suggest that there should be a low threshold for screening, and risk-assessment for these prophylactic interventions.

However, it is important to consider the real world practicalities of screening in different settings, and to avoid mandates for screening that cause unintended delays in therapy. This is particularly pertinent for COVID-19 patients who are critically unwell, for whom interventions need to be expedited. For HBV, serum markers (HBsAg, anti-HBc) are generally accessible and affordable, with short turn-around times. For TB screening, interferon gamma release assays (e.g., Quantiferon) and purified protein derivative (PPD) skin tests (e.g., Mantoux) may not be immediately available, and can take 2–3 days to result. Individual risk assessment therefore has to account for the probability of carriage or latent infection, balanced against the potential risks of giving early prophylaxis at the time of TCZ therapy. On these grounds, a decision can be made either to initiate prophylaxis pending results of screening tests (and discontinue prophylaxis if the screen returns negative), or to postpone prophylaxis and initiate only if screening tests return positive.

The high prevalence of HBV exposure and latent TB in some settings pose particular challenges in making balanced risk appraisals. These populations are not adequately represented in research cohorts, and it is therefore not possible to make specific recommendations based on the current literature. Nevertheless, expanding awareness of the potential risk of complications remains important, such that vigilance can be applied, as recommendations are relevant not just at the point of TCZ prescription but during follow-up, with clinical awareness of potential complications in the weeks and months after treatment. As TCZ may blunt markers of infection such as C-reactive protein (CRP), reactivation events may be more difficult to diagnose in this group (1), although increases in CRP are still recognised in response to bloodstream infections (55). In individuals with potential risk of HBV reactivation, monitoring of liver enzymes and HBV markers (HBsAg and viral load), should be considered.

In the case of HCV, flares are possible on biologic therapy and as successful direct acting antiviral (DAA) treatment can now be prescribed, screening is advocated in anyone with risk factors for infection and/or with unexplained abnormalities in liver function tests, such that therapy can be started (51).

IL-6 has an important impact in priming both the humoral and cell-mediated limbs of the adaptive immune system (Figures 1, 2). IL-6 depletion or knock-out experiments in animal models have demonstrated higher pathogen burden, and increased morbidity and mortality, in a range of viral (27, 28), bacterial (28–31), fungal (32), and parasitic infections (33, 34). Abrogating the IL-6 pathway results in alterations in the production of pro-inflammatory cytokines and chemokines, and can influence vascular permeability, the function of the blood-brain barrier (32), and epithelial barrier integrity in the respiratory (29) and urinary (31) tracts. In these in vivo experiments, the host immune response can be rescued by administration of recombinant IL-6 (35).

In IL-6 blockade or knock-out, mycobacterial infections are characterised by reduced interferon-gamma production, typically associated with higher bacterial loads, more severe illness and a higher risk of death in mice (56, 57). IL-6 also influences the mileu of other cytokines, including IL-12, IL-23, and type 1 interferons (IFNs), thus playing a role in a complex interaction of factors responsible for co-ordinating macrophage, neutrophil and T-cell activation, with the potential to mediate both pro-inflammatory and anti-inflammatory effects (58, 59) (Figure 2). These roles may have varied influence, depending on the dose and route of inoculum, in different animal hosts, and in acute vs. chronic infection (60). Overall, the effect of IL-6 blockade on outcomes of mycobacterial infection may be modest, and less than that associated with TNF-blockade (61).

Chronic HBV infection is associated with elevated IL-6 levels, both in stable disease (62) and in acute-on-chronic liver failure (63). IL-6 may have a role in HBV control through a variety of mechanisms, including moderation of HBV transcription (64, 65), an influence on the cell surface receptor (NTCP, sodium taurocholate co-transporting polypeptide) (66), and inhibition of the generation of genome-containing nucleocapsids (67), reviewed by Velazquex-Salinas et al. (2). The importance of IL6 in mediating outcomes of HCV infection is demonstrated by an association between IL6 haplotypes and treatment response (68).

We recognise that our estimates of risk are biassed, with a likelihood of under-estimating true risk based on selection of low-risk study participants (including a high proportion of individuals without latent infection, and in some cases completely excluding all at-risk patients), implementation of careful screening and prophylaxis in clinical studies (which differ from usual standard of care), and insufficient durations of follow-up to detect all reactivation events.

Risk appraisal for use of TCZ in COVID-19 is currently based on experience in treating patients with inflammatory and rheumatological disease, who may be at an increased risk of infective complications on the grounds of immunosuppression related to their primary condition, and in whom immunosuppressive therapy is typically prolonged and may involve multiple agents (including steroids, methotrexate and other biological agents). However, although COVID-19 treatment comprises only one or two doses of TCZ, the context of critical illness, combination with steroid therapy, and two doses given in quick succession, may inflate the risks. On all of these grounds, extrapolation of existing TCZ risk assessment data to a population with COVID-19 infection must be undertaken with caution. As the field of COVID-19 therapeutics is expanded, there is interest in using other immunomodulatory drugs (69), which may also be associated with an increase in infective complications.

We were unable to undertake a formal meta-analysis due to the heterogeneity of data, and the reported mean across studies must be interpreted with caution. Further data are needed to investigate the impact of TCZ in racially diverse COVID-19 cohorts including patients with other underlying pathology, followed-up over longer periods (1, 16). This will enable identification of subgroups in whom treatment may be of most benefit, and of the circumstances in which it may carry the most risk. It is notable that many existing publications about the use of TCZ in COVID-19 do not even discuss the potential for these infective complications.

From existing data, we are unable to determine the highest risk period for SIEs with respect to the timing of TCZ dosing, and it is possible that more TB and HBV reactivation events would be identified by longer periods of follow-up, although this trend is not clear from the studies we have identified. However, it is most likely that the greatest risk in COVID-19 therapy is within a short number of weeks of the TCZ dose. Similarly, there are currently no data to determine whether the risk is inflated in individuals receiving two doses rather than one dose of TCZ. Finally, we elected to exclude data on children, due to a current focus on TCZ use to manage severe COVID-19, which arises almost exclusively in adults. However, this should not preclude the collection of robust safety data for TCZ use in children for rheumatological disease.

There is a need for wider data and more robust guidelines to support the safe use of biologic therapy with appropriate risk assessment and consistent use of antimicrobial prophylaxis, equitable inclusion of diverse populations in research and clinical studies, and ongoing prospective surveillance of patients receiving TCZ to identify complications. In the longer term, better biomarkers that will improve risk stratification, improving the identification of patients at the highest risk of HBV reactivation who stand to gain most from prophylaxis.

The limitations in the data highlight a pressing need for the collection and sharing of complete metadata for clinical studies, including (but not limited to) details of control or comparator groups, use of screening and prophylaxis, and specific dosing regimens. This study also underlines the under-representation of population groups within research studies (by geography or by ethnicity), resulting in substantial uncertainty, which in this case pertains to the very groups likely to be at the highest risk of complications (70).

In addition to the pathogens we have assessed here, there are other SIEs associated with TCZ therapy that remain to be addressed in individuals with COVID-19, including invasive fungal infection, atypical mycobacterial infection, herpes viruses (among which VZV reactivation may be a specific concern), and other opportunistic infections. More data are needed to inform screening, risk assessment, prophylaxis and treatment of these infections in COVID-19 patients. Given that these events are typically uncommon, there is a crucial need for experience to be collated and centralised. The WHO “Vigibase” platform provides a potential foundation for such activity, but collection of enhanced metadata will be required to determine the individuals at risk of SIEs.