Study design
This is a retrospective cohort study using propensity score matching to examine the association between asthma and severe outcomes among hospitalised COVID-19 patients. The nearest neighbour method without replacement was used in the propensity score matching, with the matching ratio set at 1:15 and the caliper set at 0·2 [20]. Logistic regression models were used to estimate the propensity score. The standardised mean difference (SMD) was used to assess covariate balance between asthma and non-asthma groups, with SMD 21].
Data source
All hospitalised COVID-19 patients in Hong Kong admitted from January 1, 2022, to November 13, 2022 (study period), an Omicron-predominated period, were identified with inpatient and historical medication records retrieved from the Hong Kong Hospital Authority, a statutory body to manage all the 41 public hospitals in Hong Kong. During the pandemic, all COVID-19 hospitalisations were managed in the public hospital system in Hong Kong. The data were linked to the epidemiological investigation database and the COVID-19 vaccination registry database held by the Hong Kong Department of Health using unique pseudo key numbers to obtain extra information including medical history and vaccination status.
Study population
Patients who were aged ≥ 18 years at hospital admission, with confirmed positive reverse transcription-polymerase chain reaction (RT-PCR) results, and had medication records and clinical records indicating severity outcomes (intensive care unit (ICU) admission and inpatient death) were included in the cohort.
In the primary analysis, asthma patients were defined as those with at least one ICD-9 CM code of 493 in the inpatient records during the three years prior to the hospital admission for COVID-19. The outpatient data and epidemiological investigation data were cross-checked to confirm the presence of the diagnosis. In the exploratory analysis, the asthma patients were further classified based on their prescribed inhaled corticosteroids (ICS) doses into the following groups (1) no/unknown ICS; (2) low-dose ICS; (3) medium-dose ICS; and (4) high-dose ICS. The ICS dose levels were defined based on guidelines from the Global Initiative for Asthma [22]. In assigning individuals to low, medium, or high ICS dose levels, prescriptions for ICS in the year prior to the first positive PCR date and the highest dose the individual was prescribed were used [7]. Results using the most recent dose the individual was prescribed [8] were also provided in the sensitivity analysis. In addition, the effect of asthma on COVID outcomes among patients with different asthma therapy was examined. Asthma patients with no therapy targeting asthma, with ICS + long-acting β2-agonists (LABA) / short-acting β2-agonists (SABA), and with ICS + LABA/SABA + long-acting muscarinic antagonists (LAMA) / leukotriene receptor antagonists (LTRA) / Xanthines (equivalent to the highest steps in the asthma management stepwise approach [22] and can be used to define severe asthma [23]) were respectively compared with their matched non-asthma counterparts.
Outcomes
Severe outcomes were defined as ICU admission and inpatient death after the first positive PCR results as well as a composite outcome of either ICU admission or inpatient death. For those discharged without experiencing these events, their event time was censored on the discharge date of their last hospitalisation during the study period. For those who did not experience the events and were not discharged yet, their event time was censored on November 27, 2022, 14 days after the end date of data extraction, to avoid bias from those who had not had adequate time to accrue an outcome [23]. The event time was calculated as the number of days from the first positive PCR date to the first occurrence of the specific events.
Covariates
Covariates matched included age, sex, vaccination status, use of paxlovid and molnupiravir (which proved to be effective in reducing the mortality and hospitalisation rates in patients with COVID-19) [24, 25], use of other anti-COVID-19 treatments (i.e., dexamethasone, remdesivir, baricitinib, tocilizumab, and interferon beta-1b), medical history, and calendar week. The vaccination status of the individuals was grouped into 0, 1, 2, and ≥ 3 doses. Only those who had taken the respective vaccine doses 14 days before the first positive PCR date were regarded as vaccinated with the dose, considering the latency between vaccine uptake and full development of immune responses.
Medical history was identified using the ICD-9 CM codes, including hypertension (401·X-405·X), diabetes (250·X), coronary artery disease (410·X—414·X), congestive heart failure (398·91, 402·01, 402·11, 402·91, 404·01, 404·03, 404·11, 404·13, 404·91, 404·93, 425·4—425·9, 428·X), arrhythmia (426·0, 426·13, 426·7, 426·9, 426·10, 426·12, 427·0—427·4, 427·6—427·9, 785·0, 996·01, 996·04, V45·0, V53·3), chronic obstructive pulmonary disease (COPD; 496), malignancy (140·X—172·X, 174·X—208·X, 238·6), cerebrovascular disease (362·34, 430·X—438·X), peripheral vascular disease (093·0, 437·3, 440·X, 441·X, 443·1—443·9, 447·1, 557·1, 557·9, V43·4), chronic liver disease (070·22, 070·23, 070·32, 070·33, 070·44, 070·54, 070·6, 070·9, 456·0—456·2, 570·X, 571·X, 572·2- 572·8, 573·3, 573·4, 573·8, 573·9, V42·7), chronic kidney disease (585·X), and obesity (278·0). The outpatient data and epidemiological investigation data were cross-checked to supplement the inpatient records of the medical history.
Statistical analysis
Descriptive statistics are presented for patients with and without asthma. Cox proportional hazard models with weights and clusters representing the matching effect were conducted to examine the association between asthma and severe outcomes. All asthma patients, asthma patients prescribed with each level of ICS, and asthma patients with different therapy targeting asthma were compared respectively with their matched controls. Crude and adjusted hazard ratios (HR) that adjusted for sex, age, vaccination status, use of paxlovid, molnupiravir and other anti-COVID-19 treatments, medical history, and calendar week were respectively estimated and presented with their 95% confidence intervals (CIs). The distributions of the time-to-events among asthma and non-asthma patients were visually presented using the adjusted survival curves.
Subgroup analyses were conducted by age (p-value 26].