Study design and study setting

We conducted a population-wide retrospective cohort study using de-identified individual-level data of the hospitalized patients. The data were retrieved from the Hospital Authority (HA) and the Department of Health in Hong Kong. HA is a statutory body that provides public inpatient and outpatient services, serving over 7.3 million local citizens and accommodating more than 90% of all local hospitalizations. HA managed a centralized health record database, which contained routinely collected information on patients’ demographic characteristics, death registry, hospitalization records, laboratory test records, and medication prescription records. With close clinical monitoring during inpatient care, the quality of the measurements of viral load data during inpatient care was ensured. These health records were further linked to the anonymized DH database, which provided population-based vaccination records. The diagnoses and procedures were coded according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM).

The study followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) reporting guideline. Ethics approval was obtained from the Joint CUHK-NTEC Clinical Research Ethics Committee (No. 2023.006). As this study was a retrospective analysis using secondary data without any personal information, the requirement for obtaining informed consent was waived and approved by the ethics committee.

Study population

We included hospitalized patients aged 18 years or older who were first-time infected with SARS-CoV-2 and had positive reverse transcription polymerase chain reaction (RT-PCR) results from March 11, 2022 to October 10, 2023 (21 days before the end of data availability date, which is October 31, 2023). The SARS-CoV-2 Omicron variants were the dominant circulating variants during the study period. Patients who were admitted 3 days before or after the positive RT-PCR date were eligible for inclusion^27,28. Such criteria also take into account the possible delay between case confirmation and hospital admission during a growth phase of the epidemic^27,28. The index date was defined as the first recorded positive RT-PCR date, and the post-acute COVID-19 outcomes were assessed 21 days after the index date, a timeframe that was commonly adopted for evaluating the post-acute COVID-19 outcomes^28,29,30. All included individuals had a Ct value at their index date.

Oral antiviral treatments, including nirmatrelvir/ritonavir (accessible to patients since March 6, 2022) and molnupiravir (accessible to patients since February 26, 2022), were dispensed to patients based on the HA COVID-19 patient management guidelines. According to the guidelines, patients who were at risk of progressing to severe COVID-19 were recommended to receive the antiviral treatments, such as the elderly population, patients with asthma, chronic kidney disease, cancer, or diabetes mellitus. Patients who were prescribed nirmatrelvir/ritonavir and those who were prescribed molnupiravir within 5 days after the symptom onset (the index date was used as a proxy of the symptom onset date) were referred to as nirmatrelvir/ritonavir recipients and molnupiravir recipients, respectively. As treatment guidelines recommend using nirmatrelvir/ritonavir and molnupiravir within 5 days after illness onset, patients who used the antivirals after 5 days from the index date were excluded. Patients who used both nirmatrelvir/ritonavir and molnupiravir were also excluded.

The Ct value was obtained from the quantitative RT-PCR (RT-qPCR) assay. Serial Ct value measurements were obtained to determine virologic rebound. We included individuals who received oral antiviral treatment with at least one Ct value measurement before or during the antiviral treatment, and with at least one Ct value after the end of the treatment, with these conditions observed within the acute phase of infection—that is, 21 days after the index date. For patients who did not receive antiviral treatment, we excluded those without at least one Ct value five days after the initial Ct measurement⁶, with these conditions observed within 21 days post the index date. We excluded patients with no Ct value measurements and those with virologic rebound (defined below) that occurred beyond 21 days after the index date. Ct values were retrieved for each patient after the index date. The Ct values were summarised as their mean value if a patient had multiple Ct measurements on the same day.

Covariates

Baseline characteristics of recruited individuals were recorded, including age, sex, CCI computed based on diagnosis ascertained before the index date (Supplementary Table 1), immunocompromised status, the record of intensive care unit admission 21 days before the index date, initiation of concomitant treatments (dexamethasone, methylprednisolone, prednisolone, interferon-beta-1b, baricitinib, tocilizumab, and remdesivir) and ventilation support (including intubation, mechanical ventilation, and oxygen supplementation) within 21 days of the index date (Supplementary Table 2), COVID-19 vaccination status (unvaccinated, 1–2 doses, and ≥ 3 doses). The week of the index date and the initial Ct value measured at the index date were also used in weighting. Patients were considered as vaccinated if they had received the COVID-19 vaccine at least 14 days before the index date. The immunocompromised patients were those with diagnosed immunocompromising conditions (HIV, hematological malignancy, immune-mediated rheumatic disease, other hematological conditions, solid organ transplant, and bone marrow or stem cell transplant). Patients were also categorized as immunocompromised if they had a history of receiving or had remaining days supply of a monoclonal antibody within the last three months, an oral immunosuppressive drug within the last month, an oral glucocorticoid (equivalent to 20 mg/day of prednisone taken continuously) within the last month, or if they had received an immunosuppressive infusion or injection within the three months before the index date³⁰.

Exposure and outcomes

The exposure group comprised patients who experienced SARS-CoV-2 virologic rebound within 21 days after the index date, defined as a decline in Ct value of at least 3 units between two consecutive Ct measurements, with such reduction persisting in at least one subsequent measurement, following the previous study⁶. Some other definitions of virologic rebound were based on the viral load measurements (viral mRNA copies), with at least a half to one-unit increase in log₁₀-transformed viral load^4,8,9,25. The Ct value from RT-qPCR assays was widely adopted as a proxy of viral load, with a lower Ct value indicating a higher viral load (an inverse correlation between log-transformed Ct value and viral load)^31,32,33. A 3-point reduction in the Ct value is roughly equivalent to a tenfold increase in the viral load³⁴. Ct value was not available when a negative RT-qPCR result was observed and was imputed as 40, which was the limit of detection of the RT-qPCR assay. Other definitions of virologic rebound were tested in a sensitivity analysis.

The primary outcome was post-acute inpatient death, defined as a death that occurred 21–365 days post-index date. The secondary outcomes included post-acute composite hospitalization, defined as a hospitalization due to at least one of the 13 post-acute sequelae, including congestive heart failure, atrial fibrillation, coronary artery disease, deep vein thrombosis, chronic pulmonary disease, acute respiratory distress syndrome, interstitial lung disease, seizure, anxiety, post-traumatic stress disorder, end-stage renal disease, acute kidney injury, and pancreatitis^28,29 occurring 21–365 days post-index date (Supplementary Table 3). Hospitalization due to specific sequelae was also assessed^28,30,35,36. To prevent any prior lingering conditions preceding the SARS-CoV-2 infection, individuals with a prior diagnosis of the condition of interest within three years before the index date were excluded from the analysis of post-acute sequelae. In this study, the sequelae of interest were primarily identified upon rehospitalization after discharge from the acute phase of SARS-CoV-2 infection, with a small proportion of patients with prolonged hospitalization also included. Individuals were followed from the index date until inpatient death, the occurrence of the clinical outcome events, 365 days after the index date, or the end date of data availability (October 31, 2023), whichever came first.

Statistical analysis

SMR weighting was used to balance the baseline characteristics between patients with and without virologic rebound³⁷, with propensity scores estimated from multivariate logistic regression models. SMDs was applied as an indicator of covariates balance, with an SMD less than 0.1 indicating good balance between groups³⁸. Unadjusted cumulative incidence of post-acute COVID-19 mortality was computed as a step function of follow-up time during the observational period, with RD between patients with and without virologic rebound computed. Cox proportional hazard models were used to estimate the HR between patients with and without virologic rebound in weighted cohorts. Robust standard errors were obtained through the Huber sandwich estimator. Data were analyzed for all patients and separately for nirmatrelvir/ritonavir recipients and molnupiravir recipients.

We conducted interaction analyses in the all-patient group to examine the effect modification of nirmatrelvir–ritonavir or molnupiravir on the association between virologic rebound and the post-acute outcomes. The RERI was calculated to evaluate the additive interaction³⁹, and the exponential of the coefficient of the product term of virologic rebound and antiviral treatments was obtained as the measurement of the multiplicative interaction.

Subgroup analyses were performed for the entire patient cohort according to age groups ( < 65 years or ≥ 65 years), vaccination status (unvaccinated or ≥ 1 dose of vaccine), and the CCI categories ( <4 or ≥4). Several sensitivity analyses were conducted: (1) the ascertainment window of post-acute COVID-19 outcomes was redefined from 21–365 days to 21–180 days after the index date; (2) patients with virologic rebound occurring beyond 14 days (instead of 21 days) after the index date were excluded; (3) restrict the analysis for patients admitted at the time or after the initial positive RT-PCR, which is considered as a proxy for admission related to COVID-19 symptomatology (4) Three alternative definitions of virologic rebound were evaluated based on previous studies: (i) a decrease in Ct value of at least 3 units after the end of oral antiviral treatment or treatment completion proxy within 21 days post the index date. The proxy of treatment completion date was defined for patients not receiving any oral antiviral treatment as 5 days after the index date, which was the median days between the index date and the treatment completion date for antiviral recipients. This definition of virologic rebound accommodated the definitions applied in previous clinical trial and observational study^6,9; (ii) at least two consecutive Ct measurements with values larger than or equal to 30 followed by at least two consecutive values less than 30²⁶; (iii) a reduction in two consecutive Ct values from a value larger than 40 to a value less than or equal to 40²⁵. In supporting the potential connection between virologic rebound and viral burden over time, a post-hoc analysis was conducted using the mixed effect models with outcome of Ct measurements within 40 days after the index date.

All analyses were conducted using R (version 4.2.2) (R Program for Statistical Computing).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.