Clinical antiviral efficacy of favipiravir in early COVID-19 (PLATCOV): an open-label, randomised, controlled, adaptive platform trial

Previously healthy adults aged between 18 and 50 years were eligible for the trial if they had early symptomatic COVID-19 (i.e., reported symptoms for ≤ 4 days), oxygen saturation ≥ 96%, were unimpeded in activities of daily living, and gave fully informed consent to study participation. SARS-CoV-2 positivity was defined either as a nasal lateral flow antigen test which became positive within two minutes (STANDARD® Q COVID-19 Ag Test, SD Biosensor, Suwon-si, Korea) or a positive PCR test within the previous 24h with a cycle threshold value (Ct) < 25 (all viral gene targets), both of which suggest high pharyngeal viral densities. The latter was added on 25 November 2021 to include those patients with recent PCRs confirming high viral loads. This was the only change to the pre-trial pre-specified inclusion/exclusion criteria. Exclusion criteria included taking any potential antivirals or pre-existing concomitant medications, chronic illness or significant comorbidity, haematological or biochemical abnormalities, pregnancy (a urinary pregnancy test was performed in females), breastfeeding, or contraindication or known hypersensitivity to any of the study drugs [31].

Block randomisation was performed for each site via a centralised web-app designed by MORU software engineers using RShiny®, hosted on a MORU webserver. At enrollment, after obtaining fully informed consent and entering the patient details, the app provided the randomised allocation. The no study drug arm comprised a minimum proportion of 20% of patients at all times, with uniform randomisation ratios applied across the active treatment arms. The study was open-label (no placebos). Enrolled patients were either admitted to the study ward (in Thailand), consistent with National recommendations at the time, or followed as outpatients at home (in Brazil). After randomisation and baseline procedures (see Supplementary materials) oropharyngeal swabs (two swabs from each tonsil) were taken as follows. Each flocked swab (Thermo Fisher MicroTest® and later COPAN FLOQSwabs®) was rotated against the tonsil through 360° four times and placed in Thermo Fisher M4RT™ viral transport medium (3mL). Swabs were transferred at 4–8°C, aliquoted, and then frozen at -80°C within 48h. Separate swabs from each tonsil were taken once daily from day 0 to day 7, and again on day 14. Each swab was processed and tested separately. Vital signs were recorded three times daily and symptoms and any adverse effects were recorded daily [31].

Patients allocated to favipiravir received 1800mg on an empty stomach, (nine 200mg tablets; Favir®, Government Pharmaceutical Organization in Thailand, n = 100; or Avigan®, FUJIFILM Toyama Chemical Co., Ltd. in Brazil n = 16), at the start of treatment followed 12 h later by a further 1800mg. Thereafter the patients took 800mg twice daily for a further 6 days totalling 13.2g over 7 days. All patients received standard symptomatic treatment excluding antivirals.

The TaqCheck® SARS-CoV-2 Fast PCR Assay (Applied Biosystems, Thermo Fisher Scientific, Waltham, Massachusetts) quantitated viral densities (SARS-CoV-2 RNA copies per mL). This multiplexed real-time PCR method detects the SARS-CoV-2 N and S genes, and human RNase P in a single reaction. RNase P was used to correct for variation in human cell content in samples. Viral densities were quantified against ATCC heat-inactivated SARS-CoV-2 (VR-1986HK strain 2019-nCoV/USA-WA1/2020) standards. Viral variants were identified using Whole Genome Sequencing (see Supplementary materials).

The primary outcome measure was the rate of viral clearance, expressed as a slope coefficient [32], and estimated under a Bayesian hierarchical linear model (mixed-effects model) fitted to the daily log₁₀ oropharyngeal swab eluate viral density measurements between days 0 and 7 (18 measurements per patient). Before model fitting, Ct values were transformed to RNA copies per mL using a random effects linear model fit to the ATCC controls (random slope and intercept for each plate with additional fixed effects for each laboratory). Viral load measurements below the limit of quantification (Ct values ≥ 40) were treated as left-censored under the model. A non-linear model (allowing an initial log-linear increase in viral loads followed by a log-linear decrease in some patients) was also fitted to the data as a sensitivity analysis. All models included slope and intercept covariate effects for the virus variant, expressed as the major sub-lineages). Additional models included slope and intercept covariate effects for age, vaccination status, and days since symptom onset. The estimated individual viral clearance rates (i.e., slope coefficients from the model fit) can be expressed as clearance half-lives (t_1/2 = log₁₀ 0.5/slope). The treatment effect was defined as the multiplicative change (%) in the mean viral clearance rate relative to the no study drug arm (i.e., how much the test treatment accelerates on average the viral clearance) [32]. Thus, a 50% increase in clearance rate equals a 33% reduction in clearance half-life. All-cause hospitalisation for clinical deterioration (until day 28) was a secondary endpoint. For each studied intervention the sample size was adaptive based on prespecified futility and success stopping rules. Initially the futility stopping rule was set as a probability > 0.9 that the acceleration in viral clearance was < 5%, but at the prespecified open first interim analysis performed after 50 patients had been enrolled, the futility threshold was increased to 12.5%.

Adverse events were graded according to the Common Terminology Criteria for Adverse Events v.5.0 (CTCAE). Summaries were generated if the adverse event was ≥ grade 2 and was new or had increased in intensity. Serious adverse events were recorded separately and reported to the DSMB.

All analyses were done in a prespecified modified intention-to-treat (mITT) population, comprising patients who had ≥ 3 days follow-up data. A series of linear and non-linear Bayesian hierarchical models were fitted to the viral quantitative PCR (qPCR) data (Supplementary materials). Model fits were compared using approximate leave-one-out comparison as implemented in the package loo. All data analysis was done in R version 4.0.2. Model fitting was done in Stan via the RStan interface. All code and data are openly accessible via GitHub: https://​github.​com/​jwatowatson/​PLATCOV-Favipiravir.

The mITT population included 114 patients randomised to favipiravir and 126 patients randomised to no study drug (Fig. 1). The baseline geometric mean (GM) oropharyngeal swab eluate viral load was 5.5×10⁵ RNA copies/mL (IQR 4.7×10⁵ to 6.3×10⁵), (Table 1, Fig. 2a). Rates of viral clearance were estimated under a linear mixed-effects model fit to all PCR data taken up to day 7 after randomisation in the mITT population (4,318 swabs in 240 patients, of which 3,839 were above the lower limit of quantification, 89%). A non-linear model was used as a sensitivity analysis. Under the linear model, there was no evidence of a difference in viral clearance rates between the favipiravir treated patients and those receiving no study drug (mean difference: –1%; 95%CI: -14% to 14%). The posterior probability that the effect was less than the pre-specified futility margin of 12.5% was 0.97 (Fig. 2b). The non-linear model gave very similar estimates (mean difference: -5%; 95%CI: -14% to 6%; probability less than 12.0% equal to 1).

Under the linear model, patients treated with favipiravir had an estimated median viral clearance half-life of 16.6 h (range 6.7 to 48.0) and patients randomised to the no study drug arm had an estimated median viral clearance half-life of 15.7 h (range 3.4 to 42.1), (Fig. 3a). In patients receiving favipiravir, there was no association between body weight (i.e., mg/kg dose of favipiravir) and the estimated viral clearance (p = 0.2) (Fig. 3b).

The oropharyngeal swabbing procedures and all treatments were well-tolerated. There were three serious adverse events (SAEs) in the no study drug arm and two in the favipiravir arm, all resulting in the secondary endpoint of clinical deterioration leading to hospitalisation for medical reasons (three patients with raised creatinine phosphokinase (CPK) were already inpatients for isolation reasons; two no study drug and one favipiravir). In the favipiravir arm, a patient was readmitted 2 days after completing the 7-day course of favipiravir with fever and a maculopapular rash over the face, trunk, back, and extremities with sparing of the palms and soles. The rash was reviewed by a dermatologist who diagnosed a viral exanthem not related to the study drug. Two patients in the no study drug arm and one in the favipiravir arm had raised creatinine phosphokinase (CPK) levels (> 10 times ULN) attributed to COVID-19-related skeletal muscle damage. These improved with fluids and supportive management and were considered unrelated to study treatment. One patient in the no study drug arm was readmitted one day after discharge due to chest pain and lethargy. All clinical and laboratory investigations were normal and the patient was discharged the following day. There were no treatment related serious adverse events.