Association of Nonsteroidal Anti-inflammatory Drug Use and Adverse Outcomes Among Patients Hospitalized With Influenza

Key Points Question Is exposure to nonsteroidal anti-inflammatory drugs (NSAIDs) associated with increased risk of intensive care unit admission or death among patients hospitalized with influenza? Findings In this Danish nationwide cohort study that included 7747 adults aged 40 years or older, no association of NSAID use with intensive care unit admission or 30-day mortality was found. Meaning The findings of this study suggest that among patients hospitalized with influenza, those who use NSAIDs do not have worse prognosis than those with no NSAID use.


Introduction
Exposure to nonsteroidal anti-inflammatory drugs (NSAIDs) has been associated with an increased occurrence of pleuropulmonary complications, such as empyema, in patients with bacterial pneumonia. 1,2 During the current coronavirus disease 2019 (COVID- 19) pandemic, case reports have described patients with no comorbidities who developed severe COVID-19 after using NSAIDs in the early stage of the disease. 3 While the European Medicines Agency stated that there is currently no scientific evidence establishing an association between use of NSAIDs and the severity of COVID- 19,4 uncertainty persists regarding whether NSAID use can be considered safe in this disease. 3 Existing evidence of the association between NSAID use and adverse outcomes in patients with pneumonia is based on patients with bacterial or unspecified pneumonia, 1,2,5 with the most commonly reported outcome of pleuropulmonary complications (empyema or abscesses) having little relevance in patients with influenza or COVID-19. Explanations other than a direct drug effect have also been proposed for adverse outcomes among those receiving NSAIDs. 1 They primarily concern the risk of confounding by indication, ie, that early symptoms of complications may lead to the use of NSAIDs, thereby introducing a spurious association between NSAIDs and adverse outcomes. Further, use of NSAIDs may mask pneumonia symptoms because of their analgesic and antipyretic effects, thus leading to delayed health care seeking and delayed antibiotic prescriptions for secondary bacterial infections, 6 which can in turn lead to more severe disease and prolonged hospital stays. 1,2 Given the widespread use of NSAIDs and the rapidly evolving COVID-19 pandemic, studies are urgently needed to determine the association of NSAIDs with the prognosis of viral pneumonia. Therefore, we examined the association between NSAID use and adverse outcomes following hospitalization for influenza or influenza-related pneumonia.

Methods
In this Danish registry-based cohort study covering all Danish hospitals, we constructed cohorts of patients hospitalized for influenza or pneumonia from 2010 to 2018 and investigated the association between current NSAID use (ie, preadmission NSAID use) and risk of adverse outcomes, specified as an intensive care unit (ICU) stay, pleuropulmonary complications (empyema or pulmonary abscess), and death. The study was approved by the institutional data protection board at the University of Southern Denmark and the Danish Health Data Authority. According to Danish law, studies based entirely on registry data do not require approval from an ethics review board or informed consent. 7 Because of Danish privacy regulations, counts of less than 5 cannot be reported. Therefore, absolute numbers are only reported for the most important risk estimates. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.

Data Sources
Denmark has a tax-supported health care system that provides free access to health care. We retrieved data from 4 nationwide Danish health registries: the National Prescription Registry, 8  Classification of Diseases, Tenth Revision (ICD-10). 9 The Danish Civil Registration System covers every Danish resident and records data on vital status (date of birth and death) and migration to and from Denmark. 10 The Danish Microbiology Database contains nationwide information on microbiological test results starting in 2010, including results for polymerase chain reaction and antigen tests for the influenza virus. 10 Detailed information on codes used to define cohorts, exposures, and outcomes are provided in eAppendix 1 in the Supplement.

Study Population
We included all Danish patients aged 40 years and older who were hospitalized with an ICD-10 code related to pneumonia or influenza from 2010 to 2018 (N = 563 522). Individuals with less than 1 year of enrollment in the database before admission or who emigrated within 30 days of admission were excluded from the analysis (n = 325).
From the pool of eligible patients, we identified several study cohorts. For the main analysis, we restricted the cohort to patients with first-time hospital admission because of confirmed influenza or influenza-related pneumonia. This was defined as a discharge diagnosis of any type of influenza or pneumonia, in combination with a positive polymerase chain reaction or antigen test for influenza virus from 14 days before admission to discharge or 30 days after admission, whichever occurred first. In secondary analyses, we analyzed cohorts comprised of patients with a first-time hospital admission because of any bacterial pneumonia or Streptococcus pneumoniae pneumonia, except for unspecified pneumonia codes. The secondary cohorts were defined based on diagnosis codes alone.
For all cohorts, patients entered the cohort on the date of hospital admission and were observed for 30 days or until they experienced an outcome of interest.

Drug Exposure
The main exposure of interest was current use of NSAIDs. Current use was defined as having filled at least 1 prescription for any NSAID within 60 days before hospital admission. In secondary analyses, we also examined incident use and long-term use of NSAIDs, which provided information on different patient groups and also allowed for the assessment of potential confounding by indication.
Incident use was defined as a first redemption of an NSAID in the 14 days before hospital admission following a washout period of 1 year. Long-term use was defined as a subset of current use, with the additional requirement of having redeemed an average of 0.5 defined daily doses of NSAIDs per day (corresponding to, eg, 600 mg ibuprofen or 50 mg diclofenac per day) during the previous year. In the main analysis, current, incident, and long-term use were all compared with nonuse of NSAIDs. In additional analyses, we used acetaminophen use as an active comparator. Acetaminophen use was defined according to the same criteria as NSAID use, so that incident use of NSAIDs was comparable with incident use of acetaminophen.
Use of NSAIDs was determined using data on prescription fills. 8 In Denmark, only low-dose ibuprofen (ie, 200 mg) can be obtained over the counter and is not captured in the registry data.
During the study period, over-the-counter sales of ibuprofen accounted for 27.2% of total ibuprofen sales during the study period (eAppendix 2 in the Supplement). 12 Furthermore, over-the-counter use is expected to be less frequent among older individuals, who are eligible for considerable reimbursement for the cost of prescription drugs. 13

Outcomes
The outcomes of interest were admission to an ICU within 30 days of hospital admission and death within 30 days of hospital admission. For patients with bacterial pneumonia, pleuropulmonary complications were an additional outcome.

Propensity Score Matching
We used propensity score (PS) matching 14 to adjust for observed differences between the treated and untreated cohorts. A PS value is the estimated probability of treatment given a set of patient characteristics. Each individual's PS was estimated at cohort entry based on known confounders and risk factors for adverse outcomes. These included age (as a continuous variable), sex, year of cohort entry, prior use of selected prescription drugs, and comorbidities. The full list of included covariates is provided in eAppendix 1 in the Supplement. Separate PSs were estimated for all combinations of subcohorts, choice of comparator cohort, and exposure definitions (n = 18). To reduce unmeasured confounding, we performed asymmetrical trimming of the PS distribution 15 before matching, ie, we removed patients who were treated contrary to prediction to the greatest extent. Individuals in the untreated cohort were matched to individuals in the treated cohort in a 1:1 ratio using the nearest neighbor algorithm without replacement and a maximum caliper of 0.05.

Statistical Analysis
Descriptive statistics were used to characterize the study cohorts at baseline. In both the unmatched and matched cohorts, we estimated risks and absolute risk differences (RDs) with generalized linear models using a binomial distribution and an identity link. Risk ratios (RRs) were estimated using generalized linear models with a binomial distribution and a log link, and 95% CIs were obtained. To assess whether risk estimates were modified by age (ie, <65 and Ն65 years), sex, and cardiovascular disease (an established relative contraindication for NSAID use 16 ), we included these factors as interaction terms in the generalized linear models. Likelihood ratio tests were used to test for interaction.
We performed a number of supplementary and sensitivity analyses. First, given that ibuprofen specifically has been implied to worsen the prognosis of patients with pneumonia or COVID-19, 3 we repeated the main analysis with ibuprofen use defined as the exposure. Second, we performed analyses restricting the 30-day mortality outcome to in-hospital deaths only, ie, censoring patients at discharge. Third, to assess the robustness of our findings toward choice of PS methodology, we repeated the main analyses using inverse probability of treatment weights 17 to address differences between treatment groups. This also changed the underlying causal contrast from the average treatment effect among those treated when matching to the average treatment effect in the population. Fourth, we obtained risk estimates for individuals with community-acquired influenza only, ie, excluding individuals with a positive test more than 2 days after the initial hospital admission.
All analyses reported were performed on May 29, 2020. Statistical analyses were conducted using Stata MP version 16.1 (StataCorp). Statistical significance was set at P < .05, and all tests were 2-tailed.

Results
We identified 7747 patients with confirmed influenza or influenza-related pneumonia between  In the unmatched cohorts with confirmed influenza or influenza-related pneumonia, 104 patients (20.0%) who used NSAIDs and 958 patients (13.3%) who did not were admitted to the ICU, yielding an RR of 1.51 (95% CI, 1.26 to 1.81) and an RD of 6.7% (95% CI, 3.2% to 10.3%). Of those who used NSAIDs, 37 (7.1%) died within 30 days of admission, compared with 563 (7.8%) of those who did JAMA Network Open | Infectious Diseases not use NSAIDs, resulting in an RR of 0.91 (95% CI, 0.66 to 1.26) and a RD of -0.7% (95% CI, -3.0% to 1.6%) (eTable 1 in the Supplement).
After PS matching (eTable 2 in the Supplement), the standardized mean difference was less than 0.1 for all baseline covariates, indicating successful balancing of measured confounders ( indicates that, compared with those who did not use NSAIDs, those who used acetaminophen were less comparable with those who used NSAIDs. This suggests a higher risk of residual confounding.

Discussion
In this nationwide population-based cohort study of patients with confirmed influenza or influenzarelated pneumonia, we found that NSAID use was not associated with the risk of 30-day mortality or ICU admission. Similar associations for ICU admission and death were observed for patients with bacterial pneumonia. In accordance with previous studies, we also observed an increased risk of pleuropulmonary complications for NSAID users with bacterial pneumonia.
A causal biological association between NSAID use and adverse pneumonia outcomes is plausible. NSAIDs impair neutrophil function and recruitment to the inflammatory site through cyclooxygenase-2 inhibition and may thus delay the resolution of inflammatory processes. 18 In addition, NSAID use has been suggested to increase the risk of renal failure in critically ill patients 19 and may also increase the risk of cardiovascular complications. 20 Finally, NSAIDs may mask the earliest symptoms of an adverse outcome in pneumonia by relieving fever and pleural pain, potentially delaying medical assistance. This is indicated by the findings of Little et al, 5 who observed that individuals who used ibuprofen had more revisits with their general practitioner than those who used acetaminophen (20% vs 12%) because of unresolved symptoms or complications. 5 Evidence is limited concerning the safety of NSAIDs in patients with viral lower respiratory tract infections. Preadmission NSAID use was not associated with mortality in ICU patients with severe influenza H1N1 during the 2009 pandemic. 21 However, the study's restriction to patients already admitted to the ICU prohibits evaluation of whether NSAID use increased the risk of ICU admission in the first place. Other studies have reported paradoxically low mortality in older patients who used and masking early symptoms (as discussed earlier) might also contribute to this finding.
The findings of this study should be interpreted in the context of a recent warning against using NSAIDs in patients suspected of having COVID-19 infection. The literature underlying this recommendation is largely based on cohorts of patients with unspecified or bacterial pneumonias and the risk of pleuropulmonary complications. Regarding coronavirus infections, NSAID use may specifically cause an upregulation of angiotensin-converting enzyme 2 receptors, similar to reninangiotensin-aldosterone system blockers. 25 Given that this receptor is used by coronaviruses to bind to epithelial cells of the lung and other organs, 26 NSAIDs may specifically increase the risk of severe and fatal COVID-19. The same biological mechanism may not necessarily apply to the influenza cohort in our study, and therefore the null findings do not necessarily reject the existence of an association between NSAID use and a worse prognosis of COVID-19.

Strengths and Limitations
The principle strength of our study is the use of data obtained from the Danish health registries, which are generally considered to be of high validity and provide nationwide coverage, thereby limiting the risk of selection bias. The main limitation is the limited granularity of data, which did not include, eg, direct measures of frailty. Also, PS methods are effective for removing measured confounding but do not generally adjust for unmeasured confounding. As such, some residual confounding cannot be ruled out. Furthermore, microbiological verification was only possible for the main cohort of patients with confirmed influenza or influenza-related pneumonia. Secondary cohorts of pneumonia were defined using diagnosis codes only. However, the positive predictive value of a hospital diagnosis of pneumonia in Denmark is high (90%). 27 In addition, given that we restricted to influenza, inferences cannot be made for less common viral infections, such as varicella zoster pneumonia, for which the pathologies are markedly different. 28 Also, we had no data on bacterial superinfections, which are a common complication among patients admitted because of influenza. 28 Exposure to NSAIDs is known to be sporadic and difficult to model. 29 Thus, some exposure misclassification may occur. However, because uncaptured NSAID use among those who were categorized as not using NSAIDs is expected to be uncommon, this is not expected to have any major influence on the study findings.