Settings, study design and data sources
This retrospective cohort study utilized real-world, territory-wide data from the Hospital Authority and the Department of Health in Hong Kong. The Hong Kong Hospital Authority is the statutory body managing services provided by public hospitals, which are accessible to all Hong Kong residents. The services cover all patients infected with COVID-19 in Hong Kong27. The Hospital Authority’s electronic health record system contains extensive patient data, encompassing demographics, dates of registered death, hospital admissions, emergency department visits, diagnoses, prescriptions, procedures, and laboratory results. The data from this system were linked to the COVID-19 vaccination records from the Department of Health. These population-based databases have been widely utilized in prior research5,7. To protect patient confidentiality, the data were anonymized with pseudo numbers. The study was reported according to STROBE guidelines.
Participants
This study included COVID-19 patients who had a positive RT-PCR test result between March 16, 2022, and November 9, 2023. The start date corresponded to when nirmatrelvir-ritonavir became available in Hong Kong, and the end date allowed for at least 21 days of follow-up prior to the study’s data cutoff on November 30, 2023 (this 21-day time frame is frequently adopted in post-COVID-19 studies)27. This period was dominated by the Omicron variant.
Another inclusion criterion was that patients were admitted to the hospital within 3 days before or after the index date (the date of the positive RT-PCR test result), which is a widely accepted definition of patients hospitalized with COVID-19 to account for potential time lags in confirming COVID-19 diagnoses7, especially during surges in SARS-CoV-2 infection. Eligible patients were also required to be aged 18 years or older.
Patients with contraindications to nirmatrelvir-ritonavir were excluded. Contraindications included the use of specific drugs within the 90 days prior to the index date (i.e., amiodarone, lumacaftor-ivacaftor, rifampicin, apalutamide, phenobarbital, rifapentine, carbamazepine, phenytoin, St John’s Wort [hypericum perforatum], ivosidenib, and primidone), severe renal impairment (i.e., an estimated glomerular filtration rate of less than 30 mL/min per 1.73 m², dialysis, or renal transplantation), and severe liver impairment (i.e., cirrhosis, hepatocellular carcinoma, or liver transplantation)5,28. In addition, patients prescribed with molnupiravir within 21 days of the index date were excluded, and patients prescribed with nirmatrelvir-ritonavir before the index date were excluded.
Ethics approval was obtained from the Joint Clinical Research Ethics Committee of The Chinese University of Hong Kong and New Territories East Cluster (CUHK-NTEC) (Ref No. 2023.006). Since this was a retrospective study using secondary data with no personal identifying information involved, the requirement for informed consent was waived by the ethics committee.
Procedures
Eligible patients were categorized into treatment and control groups. The treatment group consisted of patients who were prescribed nirmatrelvir-ritonavir, but not molnupiravir, within 21 days of the index date. The control group comprised patients who were not prescribed nirmatrelvir-ritonavir or molnupiravir within 21 days of the index date. Based on time to nirmatrelvir-ritonavir initiation (defined as the time difference in days between the index date and the nirmatrelvir-ritonavir initiation date), the treatment group was further divided into 3 groups: Day 0 group, Day 1 group, and Day 2 or later group, which comprised patients who initiated nirmatrelvir-ritonavir 0 day, 1 day, and 2 or more days after the index date, respectively10.
Covariates were selected based on literature and previous knowledge5,7. The covariates included in the analysis were age; sex; previous SARS-CoV-2 infection; the Charlson Comorbidity Index (CCI) calculated according to diagnoses before the index date (relevant ICD-9-CM codes are provided in Supplementary Table 1); concomitant pharmacological treatments initiated within 3 days before or after the index date (i.e., dexamethasone, prednisolone, remdesivir, and others [interferon-beta-1b, baricitinib, or tocilizumab]); admission to the intensive care unit (ICU) or use of ventilatory support within 3 days before or after the index date (relevant ICD-9-CM codes are provided in Supplementary Table 2); vaccination status (unvaccinated, 1–2 doses, or ≥3 doses; doses received less than 14 days before the index date were considered incomplete and did not count towards the total number of vaccine doses); and the week of the index date. Treatment choice was determined by physicians’ judgement, with reference to the Health Authority’s COVID-19 clinical guidelines. For example, according to the guidelines, remdesivir may be considered when patients could not tolerate oral antivirals or have moderate-to-severe symptoms. We controlled for other COVID-19 treatments mentioned above as covariates to further adjust for the baseline characteristics associated with treatment choice. Linear terms of the quantitative covariates were used in the models. A list of the covariates is provided in Supplementary Table 3.
Outcomes
The primary outcome of this study was post-acute inpatient death. In Hong Kong, more than 90% of deaths occur in public hospitals29, and thus the death records in our data were highly representative. The secondary outcomes were all-cause hospitalization and cause-specific hospitalization due to each of the 13 post-acute sequelae, which were selected based on evidence indicating a higher risk of these conditions among COVID-19 patients in Hong Kong27. These sequelae included 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. In assessing each of the sequelae, individuals with a history of the disease of interest prior to the index date were excluded. We also classified the sequelae into outcomes related to organ impairment, including impaired lung function, impaired heart function, impaired kidney function, incident single organ impairment, and incident multi-organ impairment30. ICD-9-CM codes and the classification scheme are provided in Supplementary Table 4.
All of the outcomes were ascertained starting from 21 days after the index date. Patients were followed up from the index date until the earliest of the following: (1) inpatient death, (2) the outcome of interest in each specific analysis, (3) 365 days after the index date, or (4) the end of data availability (November 30, 2023).
Statistical analysis
Baseline characteristics of eligible patients were outlined in a table with appropriate summary statistics. For the outcomes of post-acute death and all-cause hospitalization, cumulative incidence curves were plotted with risk tables.
To control for the covariates, standardized mortality ratio weighting was applied31. Propensity scores were estimated by logistic regression models and were used to calculate weights. Covariate balance was assessed using the standardized mean difference (SMD), where a covariate with an SMD larger than 0.1 was considered imbalanced and included in the cause-specific Cox models (more details below) for doubly robust adjustment32.
To evaluate the effectiveness of nirmatrelvir-ritonavir with different times to initiation, cause-specific Cox proportional hazards models were employed to compare each of the three treatment groups (Day 0 group, Day 1 group; and Day 2 or later group) with the control group. Death prior to 21 days was treated as a competing event. The proportional hazards assumption was examined using plots of scaled Schoenfeld residuals.
To further investigate the association between the effectiveness of nirmatrelvir-ritonavir and time to initiation, hazard ratios were obtained from separately comparing patients who initiated nirmatrelvir-ritonavir on Day 0, Day 1, Day 2, Day 3, and Day 4 with the control group in terms of post-acute inpatient death and all-cause hospitalization. These hazard ratios were log-transformed and meta-regressed on time to initiation, assuming inverse-variance weighted linear mixed-effects models. The Wald test was used to assess the statistical significance of the association. We plotted hazard ratios versus time to initiation along with the fitted model to visualize the potential association.
Subgroup analyses were conducted to assess the effectiveness of nirmatrelvir-ritonavir in preventing post-acute inpatient death and all-cause hospitalization with different times to initiation stratified by age (≤65 years or >65 years), by sex (female or male), and by vaccination status (
All statistical analyses were conducted using R statistical software (version 4.4.2).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
