Azvudine and nirmatrelvir–ritonavir in hospitalized patients with moderate‐to‐severe COVID‐19: Emulation of a randomized target trial

To examine the effectiveness of azvudine and nirmatrelvir–ritonavir in treating hospitalized patients with moderate‐to‐severe COVID‐19. We emulated a target trial with a multicenter retrospective cohort of hospitalized adults with moderate‐to‐severe COVID‐19 without contraindications for azvudine or nirmatrelvir–ritonavir between December 01, 2022 and January 19, 2023 (during the Omicron BA.5.2 variant wave). Exposures included treatment with azvudine or nirmatrelvir–ritonavir for 5 days versus no antiviral treatment during hospitalization. Primary composite outcome (all‐cause death and initiation of invasive mechanical ventilation), and their separate events were evaluated. Of the 1154 patients, 27.2% were severe cases. In the intent‐to‐treat analyses, azvudine reduced all‐cause death (Hazard ratio [HR]: 0.31; 95% CI: 0.12–0.78), and its composite with invasive mechanical ventilation (HR: 0.47; 95% CI: 0.24–0.92). Nirmatrelvir–ritonavir reduced invasive mechanical ventilation (HR: 0.42; 95% CI: 0.17–1.05), and its composite with all‐cause death (HR: 0.38; 95% CI: 0.18–0.81). The study did not identify credible subgroup effects. The per–protocol analyses and all sensitivity analyses confirmed the robustness of the findings. Both azvudine and nirmatrelvir–ritonavir improved the prognosis of hospitalized adults with moderate‐to‐severe COVID‐19


| INTRODUCTION
5][6] Nevertheless, the accessibility and wide drug-drug interaction limit its wide use, especially for those with multiple comorbidities and those in remote regions. 7,8though the WHO Guideline has also suggested remdesivir and molnupiravir in non-severe patients, molnupiravir is inaccessible due to very low supply in most Chinese provinces and remdesivir remains not approved in China. 9As an alternative option for emergency use in China, azvudine has been approved due to fewer drug-drug interactions and the ability to shorten nucleic acid negative conversion time while increasing hospital discharge rate among non-critical COVID-19 patients. 10,11However, its effectiveness has only been reported in two clinical trials and one single-arm study [10][11][12] with small sample size and the interim analysis of an unpublished phase III study, 13,14 with sparse observational studies confirming these results. 15,16 regions with limited healthcare resources, patients may be unable to access antiviral agents within 5 days of onset and hospitals are admitting more severe patients.In December 2022, only 9.4%-18.2% of patients with COVID-19 who were admitted to hospitals were accessible antiviral drugs within 5 days after onset in China. 16It remains unclear whether patients admitted to hospitals after 5-day onset, who are typically severe cases, still benefit from antiviral agents.
Using a multicentre retrospective cohort, we emulated a target trial evaluating the effectiveness of azvudine and nirmatrelvir-ritonavir in hospitalized patients with moderate-tosevere COVID-19 during the Omicron surge pandemic, which was primarily dominated by Omicron BA.5.2 subvariant in China.

| Study design and oversight
This study emulated a target trial with a prespecified trial protocol comparing the effectiveness of azvudine and nirmatrelvir-ritonavir with no antiviral treatment (i.e., not use of any anti-SARS-CoV-2 monoclonal antibodies, intravenous COVID-19 human immunoglobulin, COVID-19 convalescent plasma, azvudine, nirmatrelvir-ritonavir, lopinavir/ritonavir, molnupiravir, remdesivir, tenofovir dipifurate fumarate, entecavir, ganciclovir, telbivudine, or acyclovir during follow up, which was due to the limited availability of such medications at the peak of pandemic) in hospitalized patients with moderate-to-severe COVID-19 and a multicenter retrospective cohort of hospitalized patients with COVID-19 in Sichuan Province, China. 17,18

| Eligibility criteria
The eligibility criteria for this emulated trial followed the target trial, with three modifications.First, given the inconsistency of timing records in the electronic medical records, we included patients who had a first-ever positive SARS-CoV-2 test on the admission day or one calendar day before or after.Second, we additionally excluded patients who: (1) died; (2) were treated with invasive mechanical ventilation; (3) experienced disease progression to critical COVID-19; or (4) were treated with other antiviral therapies, on the second calendar day after enrollment (i.e., the time zero of our target trial) or earlier.Third, we excluded participants with unavailable alanine F I G U R E 1 Study design and timeframe for target trial and emulated trial.This timeframe is for one target trial and its emulation using a multicenter cohort.The baseline (i.e., time zero) for each participant is the time of randomization.(A) For some participants whose SARS-CoV-2 positive tests occur within 24 h after admission, their baseline is the time of their positive test.(B) For those whose SARS-CoV-2 positive tests occur within 24 h before admission, their baseline is the time of their admission.The target pragmatic trial is represented in blue color, while the modifications made to the emulated trial using electronic medical records are shown in red.The black color indicates that the emulated trial was identical to the target pragmatic trial.aminotransferase (ALT) and creatinine during the baseline period (between 30 days before and two calendar days after enrollment).

| Time zero
The time zero of the emulated trial (T0) was at the end of the second calendar day after enrollment (i.e., at the end of the second calendar day of target trial randomization).

| Treatment strategies and assignment
Strategy A: One azvudine 5-mg tablet once daily for 5 days for each patient.

Strategy B:
• For patients with eGFR ≥60 mL/min/1.73m 2 , two nirmatrelvir 150-mg tablets and one ritonavir 100-mg tablet were administered together twice daily for 5 days.
• For patients with eGFR ≥30 and <60 mL/min/1.73m 2 , one nirmatrelvir 150-mg tablet and one ritonavir 100-mg tablet were administered together three times a day for 5 days.The treatment strategies in the emulated trial are the same as those in the target trial, with the caveat that we assumed the medication administration post-discharge remained consistent with that of the discharge day, following clinical practice.This assumption was made due to the unavailability of information regarding oral medication administration after hospital discharge in this multicenter cohort.
We classified the patients according to whether they received azvudine, nirmatrelvir-ritonavir, or no antiviral treatment within a period from one calendar day before T0 to the end of T0.We emulated randomization by adjusting our estimates for baseline confounders (Table S1) by applying inverse probability weighting.The study did not restrict any other treatments such as supportive care, immunomodulators, and anticoagulation therapy after the assignment in all arms.

| Follow-up and outcomes
The primary outcome was the composite of all-cause death and invasive mechanical ventilation as well as both separately.The study followed all participants from T0 to death or 30 days after T0 if no death was observed.We also assessed disease progression to critical COVID-19, 19 cardiopulmonary resuscitation, admission to the ICU (defined as presence of records documenting transfer to ICU after the initiation of treatment which is depicted in Figure 1), upgrade of respiratory support, and clinical improvement (defined as a one-point decrease on a seven-category ordinal scale from baseline) 11 as secondary outcomes.Upgrade of respiratory support was defined as the initiation of more advanced respiratory support from supplemental oxygen, high-flow oxygen therapy, non-invasive mechanical ventilation, and invasive mechanical ventilation to extracorporeal membrane oxygenation (ECMO).
The safety outcomes included acute kidney injury (AKI), acute liver injury (ALI), leukopenia (defined as leukocyte count being normal at baseline but falling under 4 × 10 9 /L after initiation of treatment), neutropenia (defined as neutrophil count being normal at baseline but falling under 1.5 × 10 9 /L after initiation of treatment) and thrombocytopenia (defined as platelet count being normal at baseline but falling under 100 × 10 9 /L after initiation of treatment). 20AKI was defined as the occurrence of any of the following criteria: (1) an increase in serum creatinine ≥26.5 umol/L within 48 h, (2) a rise in serum creatinine to >1.5 times the baseline value within 7 days, or (3) a urine output <0.5 mL/(kg•h) for six consecutive hours.ALI was defined as an increase in ALT or aspartate transaminase to over five times the upper limit of the normal range. 21

| Covariates
Tables S2-S3 summarized the covariates along with their definitions.The eGFR was calculated using the chronic kidney disease epidemiology collaboration (CKD-EPI) formula.The priori determined baseline confounders included age, sex, the days from onset to admission (defined as the duration between the onset of SARS-CoV-2 symptoms and admission), co-prescription of interacting drugs (i.e., medications having drug-drug interactions with nirmatrelvir-ritonavir, Table S2) within 15 days before T0, and other prespecified factors measured on T0, including a score on the seven-category scale, 11 severity of COVID-19 (moderate vs. severe), ALT, eGFR, number of risk factors (risk factors for progression to severe-to-critical COVID-19 including age ≥65 years, obesity (body mass index ≥28 kg/m 2 ), current smoker, chronic obstructive pulmonary disease, cardiovascular disease, chronic kidney disease, hypertension, diabetes, and cancer), the use of immunomodulators, and the use of anticoagulants.

| Causal contrasts
We estimated the observational analog of the intention-to-treat effect (the effect of initiation of azvudine vs. initiation of nirmatrelvir-ritonavir vs. no initiation) and observational analog of per-protocol effect (the effect of sustained use of azvudine for 5 days vs. sustained use of nirmatrelvir-ritonavir for 5 days vs. no use during follow-up).

| Baseline characteristics
We presented the baseline characteristics before and after weighting.
To assess the balance of baseline characteristics between groups, we used absolute standardized differences, considering a value <0.1 as evidence of good balance.

| Intention-to-treat analysis
We expected bias in prescribing azvudine or nirmatrelvir-ritonavir due to demographic factors, healthcare access, COVID-19 severity, biochemical markers, co-prescriptions and comorbidities.To better achieve exchangeability between patients prescribed azvudine, or nirmatrelvir-ritonavir, and those not receiving any antiviral medication, we used an inverse probability-weighted design with an entropy balancing method. 22,23Treatment weights were generated via a maximum entropy reweighting scheme to achieve exact balance on the moments of the prior determined baseline confounders described above.If there were extreme weights, we would truncate weights at the 1% and 99% to improve precision as sensitivity analyses.
We created a pseudo-population using treatment weights to estimate the effect of azvudine, or nirmatrelvir-ritonavir when the included factors do not impact the prescription.Hazard ratios (HRs) and 95% CIs were estimated using a weighted Cox proportional hazard regression. 24,25To account for the within-hospital homogeneity in outcomes due to the similarity in healthcare quality and characteristics within the same hospital, we used the Cox shared frailty model. 26This model incorporates a shared frailty term that multiplies the baseline hazard function, with all subjects within a hospital sharing the same random effect.We accounted for potential replications of patients induced by treatment weights using a robust (sandwich) variance estimator, which resulted in conservative (wider) 95% CIs. 24,25Adjusted cumulative incidence were generated using treatment weights. 27

| Subgroup analyses
We further stratified the effectiveness of azvudine and nirmatrelvirritonavir in populations with different characteristics, including age (<65 years and ≥65 years), sex (male and female), the baseline severity of COVID-19 (moderate and severe), number of risk factors for progression to severe-to-critical COVID-19 disease (0-2 and ≥3 risk factors), and the days from onset to admission (≤5 and >5 days; The timing of treatment for our eligible patients was mostly determined by the number of days between onset and admission).Subgroup effects were tested by adding a product (interaction) term between the subgroup variable and the treatment variable into the outcome model.Given the potential for multiple comparisons, the findings from subgroup analyses should be interpreted as exploratory.

| Sensitivity analyses
We used the clone method to estimate the observational equivalent of the per-protocol effect, which involves three main steps: cloning, censoring, and weighting.The Supplement provided detailed methods of all sensitivity analyses and Figure S1 illustrated the timeframes.
In the emulated trial, treatment assignment began on the previous day of T0 and ended on T0, which introduced the potential for selection bias. 28To address this issue, we adjusted the analysis by resetting time zero of follow-up to the previous day of T0.To eliminate the potential for immortal time bias, we applied nested target trial emulation analysis. 28,29 practice, some individuals would not initiate treatment immediately upon meeting eligibility criteria and being assigned to a treatment strategy.Target trials with a grace period provide a more realistic approach. 28To this end, we emulated a target trial with a grace period of 5 days to initiate treatment and used the clone method to estimate the effect of initiating azvudine and nirmatrelvir-ritonavir within 5 days after randomization. 6 also conducted other sensitivity analyses to evaluate the robustness of our findings.First, we restricted the first-ever SARS-CoV-2 positive test within 7 days before or within 24 h after hospital admission for eligibility.Second, we further restricted the time window of the first SARS-CoV-2 tests to 3 days before or within 24 h for eligibility.Third, we excluded patients with drugs that potentially interact with nirmatrelvir-ritonavir (Table S2) within 15 days before T0.Fourth, we censored patients at discharge from the hospital if they were discharged before T30 or death, whichever came first, rather than keeping them in the risk set until T30 or death.The purpose of this sensitivity analysis was to assess whether our findings were robust to alternative approaches to censoring.
Additionally, no extreme weights were detected, thus, we did not conduct a sensitivity analysis involving the truncation of treatment weights.

| Missing data
For ALT and eGFR measurements taken after T0, we utilized the observed value from the previous day to impute any missing data.

| Baseline characteristics
Among the 2262 patients admitted to non-ICU departments with confirmed moderate-to-severe COVID-19 around hospital admission, the primary analysis included 1154 participants, with 311 receiving azvudine, 165 receiving nirmatrelvir-ritonavir, and 678 not receiving antiviral treatment (Figure 2).Table 1 and Table S4 described the baseline characteristics for each group.The median age of participants was 69 years (IQR: 54-80 years), with 61.3% being male, 8.4% having BMI ≥ 28 kg/m 2 , and 85.7% having at least one risk factor for progression to severe-to-critical COVID-19.
Severe COVID-19 was observed in 314 patients (27.2%).The median days from onset to admission was 7 days (IQR: 3-10 days), with 38.9% admitted to the hospital within 5 days of symptoms onset.The median days from symptoms onset to receiving treatment was 8 days (IQR: 5-11 days).Hypertension affected 37.3% of participants, followed by diabetes (27.3%), cardiovascular disease (19.8%), and chronic obstructive pulmonary disease (10.3%).At baseline, 86.2% of participants received respiratory support, 68.9% received antibiotics, 24.6% received anticoagulants, 6.8% received nonsteroidal anti-inflammatory drugs, and 48.4% were treated with glucocorticoids.After applying IPW, F I G U R E 2 Flowchart of construction of the cohort.ALT, alanine aminotransferase; eGFR, estimated glomerular rate filtration.T0 is the time zero for each participant in the emulated trial, which is at the end of the second calendar day after enrollment.Enrollment is the date of their first positive SARS-CoV-2 test (if a positive test occurred after admission) or their admission (if a positive test occurred before admission), which is the time zero in the target trial.
T A B L E 1 Baseline characteristics of participants in the emulated trial before balance.
baseline characteristics were well balanced between groups (Figure S2).

| Primary outcomes
Figure 3 showed the adjusted cumulative incidence of the primary composite outcome, all-cause death, and invasive mechanical ventilation.Compared with no antiviral treatment, the adjusted HR of azvudine on the primary composite outcome was 0.47 (CI: 0.24-0.92),while that of nirmatrelvir-ritonavir was 0.38 (CI: 0.18-0.81).The adjusted HR of azvudine on all-cause death was 0.31 (CI: 0.12-0.78)and that of nirmatrelvir-ritonavir was 0.51 (CI: 0.22-1.17).The adjusted HR of azvudine on invasive mechanical ventilation was 0.64 (CI: 0.30-1.34)and that of nirmatrelvir-ritonavir was 0.42 (CI: 0.17-1.05).Figure S3 showed the subgroup analyses for primary outcomes, where we did not identify any interactive effects for age, sex, the baseline severity of COVID-19, the number of risk factors for disease progression, and the days from onset to admission (>5 days vs. ≤5 days).All interaction p values were >0.05.
All seven sensitivity analyses confirmed the robustness of the primary results (Figure S5).
validate its alternative use for anti-SAR-CoV-2 therapy with comparable effectiveness of nirmatrelvir-ritonavir. 16 Nirmatrelvir-ritonavir is one of the first antiviral drugs proven for COVID-19.It saved lives worldwide in the past year but the accessibility and drug-drug interactions pose obstacles for many patients in need. 7,8This study suggests azvudine as a viable alternative with comparable effectiveness to nirmatrelvir-ritonavir in treating moderate-to-severe COVID-19.Treating with azvudine cut the chance of death or invasive mechanical ventilation by half.
Our findings showed elevated risk for ALI in people receiving nirmatrelvir-ritonavir, but azvudine showed no additional adverse event rates compared to the control, the findings of which are consistent with previous reports. 30,31This presents one more option for vulnerable populations, improving their prognosis, especially in low and middle-income countries.
The WHO restricts nirmatrelvir-ritonavir to patients with mild-tomoderate COVID-19 at high risk for hospitalization. 32Our study included both moderate and severe cases and did not find different effectiveness in patients across different severity.Given a higher baseline risk of death and progression to critical cases, patients with severe COVID-19 may expect more absolute benefits.Current guidelines also recommend nirmatrelvir-ritonavir initiation within 5 days after symptom onset. 32However, individuals with socioeconomic disadvantages may encounter difficulties in accessing timely COVID-19 healthcare services in the first few days of onset. 3,15,16During the first COVID-19 wave in China in December 2022, fewer than half of patients were admitted to hospitals with accessible antiviral drugs within 5 days from symptom onset. 15,16In this study, the relative effects of starting the drug after 5-day onset were comparable with the recommended patients, suggesting antiviral therapy may also help in latent cases. 15,16ese findings expanded the potential use of nirmatrelvir-ritonavir in wider patients, but the limited sample size prevented precision estimates of drug effects.Since the end of PHEIC put COVID-19 care in routine practice, 3 these potential extended uses of oral antivirals call for validation in larger trials that can benefit more people facing the issue.
This study has several limitations.First, the detection of significant subgroup differences was impeded by the limited number of events.We thus call for larger trials, especially for some particular subgroups with specific interests such as the severe and/or latent cases.Second, given the lack of information, this study was unable to stratify people with and without COVID-19 vaccinations.Since over 90% of Chinese adults had been vaccinated before December 2022, our results likely reflect the effectiveness of these two drugs in vaccinated people. 33Third, it is important to note that as the virus specification and emulation of the target trial.The target trial enrolls Chinese adults aged 18-90 years who have been admitted to three of the largest hospitals authorized to admit patients with COVID-19-that is, West China Hospital of Sichuan University, Public Health Clinical Centre of Chengdu, and Affiliated Hospital of North Sichuan Medical College, diagnosed with moderate-to-severe cases but excluding critical cases who were admitted to the intensive care units (ICU) or mild ones who were treated in the outpatient service or in the community.The trial includes patients with symptoms of COVID-19 before admission, at least one positive test for SARS-CoV-2 within 24 h before or after hospital admission, and with the diagnosis of moderate-to-severe COVID-19 according to the latest Chinese guideline for COVID-19 (comparable in moderate-to-severe cases in the WHO category). 19We exclude patients if they are: (1) treated with azvudine, nirmatrelvir-ritonavir, molnupiravir, remdesivir, lopinavir, anti-SARS-CoV-2 monoclonal antibodies, intravenous COVID-19 human immunoglobulin, and COVID-19 convalescent plasma before randomization; (2) breastfeeding or pregnant; (3) HIVinfected; (4) allergic to nirmatrelvir-ritonavir or azvudine; (5) diagnosed with severe liver cirrhosis (Child-Pugh C); (6) identified estimated glomerular rate filtration (eGFR) < 30 mL/min/1.73m 2 within 15 days before randomization.The interventions comprise treatment with either azvudine or nirmatrelvir-ritonavir, or no antiviral treatment.Treatment initiation occurs at randomization for all participants.The baseline (i.e., time zero) for each participant is the time of randomization.For some participants whose SARS-CoV-2 positive tests occur within 24 h after admission, their baseline is the time of their positive test, and for those whose SARS-CoV-2 positive tests occur within 24 h before admission, their baseline is the time of their admission.Figure 1 illustrates the timeframe for eligibility, treatment assignment, treatment initiation, time zero (baseline), follow-up, and outcomes.

2. 2 |
Specification of the target trial emulation 2.2.1 | Data sources The study cohort included hospitalized patients with COVID-19 in three hospitals considered in the target trial, between December 1, 2022 and January 19, 2023, during the omicron BA.5.2 variantdominant period.The study collected information including demographic records, medical history, discharging summary, prescription list, laboratory tests, vital signs, and diagnosis.To obtain all-cause death records after discharge and vaccination information, the study linked with the database of the Sichuan Center for Disease Control and Prevention.
Strategy C: No oral antivirals during follow-up.The dose of azvudine and nirmatrelvir-ritonavir was in accordance with the guidelines provided in the "Notice on the issuance of Diagnosis and Treatment Protocol for novel coronavirus infection (trial version 10)" issued by the General Office of the National Health Commission of China.

4 |
DISCUSSIONThis study based on a multicenter retrospective cohort emulated a target trial eliminating the potential biases commonly associated with observational studies, such as selection bias and immortal time bias, and suggested that in adults hospitalized for moderate-to-severe COVID-19 during a wave of SARS-CoV-2 omicron BA.5.2, both azvudine and nirmatrelvir-ritonavir reduced the composite outcome of all-cause death and invasive mechanical ventilation within 30 days after admission.All patients younger than 65 years old receiving either antiviral drug survived throughout the follow-up duration, compared to 10 deaths per 1000 in the control arm.This is the first multicenter observational study confirming the effectiveness of azvudine and nirmatrelvir-ritonavir in adults hospitalized for moderate-to-severe COVID-19 in mainland China.The findings corroborate the results of previous trials[10][11][12]14 of azvudine and T A B L E 2 Frequencies of effective and safety outcomes in total study population.
mutates over time, new variants might cause shifts in the real-world effectiveness of azvudine and nirmatrelvir-ritonavir.Thus, our study mainly reflects the first wave of COVID-19 dominated by Omicron BA.5.2 from late 2022 to early 2023 in China.Finally, we did not restrict the post-randomiation treatments including the supportive care, which might underestimate the efficacy as rescue regimens.We thus interpret our estimates as effectiveness, which are more in line with real-world settings.

5 |
CONCLUSION This target trial emulation study with a multicenter retrospective cohort in Sichuan Province during the Omicron BA.5.2 variant wave confirmed the effectiveness of both azvudine and nirmatrelvir-ritonavir use in reducing the composite of all-cause death and invasive mechanical ventilation and the two in separate among adults hospitalized with moderate-to-severe COVID-19.Azvudine can be an alternative for patients with moderate-tosevere COVID-19 especially when nirmatrelvir-ritonavir is inaccessible or contraindicated.This study also indicated the potential extended use of azvudine and nirmatrelvir-ritonavir in severe cases and those administered after 5 days of symptom onset due to the delayed healthcare and called for validation in future trials.ORCID Sheyu Li http://orcid.org/0000-0003-0060-0287Dan Liu http://orcid.org/0000-0001-6791-1704 Table S1 listed the details of the