Impact of Lifestyle and Socioeconomic Position on the Association Between Non-steroidal Anti-inflammatory Drug Use and Major Adverse Cardiovascular Events: A Case-Crossover Study

The Danish healthcare system provides all Danish citizens and legal residents with universal tax-supported healthcare, guaranteeing unfettered access to general practitioners, specialists in private practices, and hospitals, as well as partial reimbursement for the costs of prescription drugs [4]. Individual-level linkage of all Danish registries is possible using the unique 10-digit Civil Personal Register number assigned to each Danish citizen at birth and to residents upon immigration [5]. The Danish Civil Registration System contains nationwide data on mortality and migration since 1968 [5], permitting virtually complete follow-up with accurate censoring at death or emigration.

We conducted a case-crossover study [6] of all adult (≥ 18 years of age) first-time respondents to the Danish National Health Surveys of 2010, 2013, or 2017 [7], who died or experienced a cardiovascular event between survey completion and 31 December 2020. The nationwide Danish National Health Survey was conducted in one national and five regional random samples of approximately 300,000 individuals ≥ 16 years of age in each survey year [7]. Since the date of completion of the survey was unavailable, we used 1 May in the survey year as the index date. We chose this date as the survey was administered between the end of January and the beginning of May in each survey year [7]. Individuals with a prior hospital diagnosis of cardiovascular disease (International Classification of Diseases 10th edition code: DI) were excluded as those with and without manifest cardiovascular disease may vary in cardiovascular risk. We also excluded individuals with a filling of an NSAID prescription (Anatomical Therapeutic Chemical classification code: M01A) in the year before survey completion, as prevalent and new NSAID users may have different cardiovascular risks.

The exposures were the NSAIDs ibuprofen, naproxen, or diclofenac, as they account for most filled NSAID prescriptions in Denmark [8]. The length of exposure was determined by utilizing information on the dosage and the number of tablets from filled prescriptions of these drugs. This approach might be more precise than using fixed time windows [9, 10]. The exposure length was calculated as the prescribed quantity divided by half of the maximum daily dose recommended by Danish guidelines for pain or fever treatment (600 mg for ibuprofen, 500 mg for naproxen, and 100 mg for diclofenac). Thus, a patient who filled an ibuprofen prescription with 50 tablets of 400 mg was considered exposed to ibuprofen for the next 33 days (50x400/600 = 33). If a patient filled a new prescription for the same NSAID within an exposure period plus a 30-day grace period, the exposure period was extended by the number of days provided by the new prescription. We identified NSAID prescriptions from the Danish National Prescription Registry, which contains accurate information on all filled prescriptions from Danish community pharmacies since 1995 [11].

The primary outcome was a major adverse cardiovascular event (MACE), defined as a composite of the major cardiovascular risks associated with use of NSAIDs, i.e., myocardial infarction, ischemic stroke, heart failure, and all-cause death [12, 13]. Secondary outcomes were the individual MACE components. We identified the outcome dates from the Danish National Patient Registry, which contains information on all inpatient non-psychiatric contacts since 1977 and on all inpatient psychiatric, emergency, and outpatient contacts since 1995 [14]. Registration of the International Classification of Diseases codes used to define the outcomes have all been validated within the Danish National Patient Registry with positive predictive values of 97% for myocardial infarction [15], 88% for ischemic stroke [16], and 76% for heart failure [15]. All International Classification of Diseases and Anatomical Therapeutic Chemical classification codes used in the study are provided in Online Resource 1.

Lifestyle and socioeconomic position were categorized as presented in Online Resource 2 based on self-reported answers to questions in the Danish National Health Surveys [7] and information on income [17] and employment [18] available from Danish registries.

We used the Quan-modified Charlson Comorbidity Index to estimate baseline comorbidity burden, categorized as no (score: 0), low (score: 1), moderate (score: 2–3), or severe (score: ≥ 4) [19, 20]. To assign comorbidities to individuals, we used in and outpatient information for the 10 years preceding the outcome date. In addition to hospital diagnoses, we used drug proxies for diabetes without chronic complications, chronic pulmonary disease, affective disorder, and hypertension because these may be treated solely by a general practitioner. These proxies were at least one filled prescription for a drug used to treat diabetes, chronic pulmonary disease, or depression, or at least two filled prescriptions for two different antihypertensive drug classes during the 3 months before outcome date. For other drugs, we defined baseline drug use based on filling of a prescription during the 3 months before outcome date.

We reported patient characteristics for descriptive purposes, using counts with percentages for categorical variables and medians with interquartile ranges for continuous variables. The case-crossover design is described in detail elsewhere [6, 21]. Briefly, among patients experiencing an outcome, we compared the number of individuals exposed on the outcome date, but not on 120, 180, 240, or 300 days before the outcome date (i.e., the ‘reference’ dates), with the number of individuals exposed on a reference date, but not on the outcome date (Fig. 1). If the exposure period of a prescription covered the outcome or reference dates, this date was considered exposed. Using the Mantel–Haenszel method [22], we hereby obtained the odds ratio (OR) with a 95% confidence interval (CI) associated with being exposed on the outcome date, but not on a reference date. This ‘fixed dates’ approach was preferred over a ‘fixed windows’ approach. Where the former uses one outcome date and several reference dates, the latter stacks windows of fixed lengths backwards in time from the outcome date. These windows are then considered exposed if there is a filled prescription within them. The fixed dates approach was chosen for two reasons. First, it allowed for flexibility in estimating exposure lengths. Second, it provided a clearer exposure definition, as the outcome and reference dates were only considered exposed if an exposure period covered that date. We ensured a washout period by having a greater gap between the outcome date and the most recent reference date (120 days) than between the reference dates themselves (60 days). Because each individual serves as his/her own control, the case-crossover design eliminates confounding by covariates that are stable over time (e.g., genetics). We used the simple ratio approach to compare ibuprofen, naproxen, and diclofenac with each other [23].

In two sensitivity analyses, we changed the grace period from 30 days to 14 and 60 days because increasing grace periods are associated with increasing treatment duration [24]. In a third sensitivity analysis, we defined daily NSAID use as three daily tablets of ibuprofen, two daily tablets of naproxen, and two daily tablets of diclofenac at the prescribed dose—no matter the tablet dosage.

Among 472,342 completed surveys, 277,725 (59%) were completed by first-time respondents ≥ 18 years of age without a previous cardiovascular disease diagnosis or a filled NSAID prescription during the year before survey completion, alive 1 May in the survey response year (Online Resource 3). Among these respondents, 22,834 (4.8%) died or were hospitalized for myocardial infarction, ischemic stroke, or heart failure between survey completion and 31 December 2020, and thus comprised the study population (Online Resource 3). The median age of the individuals experiencing the outcome was 68 years (interquartile range: 59–77 years), 10,144 (44%) were females, and 12,695 (56%) had at least one comorbidity (Table 1). The most common comorbidities were hypertension (23%), malignancy (6.9%), and chronic pulmonary disease (6.5%; Online Resource 4). The most frequently used drugs were antiplatelets (32%), statins (29%), and beta blockers (17%; Online Resource 4). For the main analysis, the numbers of discordant pairs, and thus the number contributing with information to the analysis, were 2,064 for ibuprofen, 108 for naproxen, and 259 for diclofenac.

Compared with NSAID use on a reference date, the OR of MACE associated with NSAID use on the outcome date was 1.34 (95% CI 1.23–1.46) for ibuprofen use, 1.48 (95% CI 1.04–2.09) for naproxen use, and 2.18 (95% CI 1.72–2.78) for diclofenac use (Table 2). Ibuprofen use on the outcome date vs. on a reference date was associated with an increased risk of heart failure (OR 1.37, 95% CI 1.06–1.76) and all-cause death (OR 1.41, 95% CI 1.27–1.56; Online Resource 5). Naproxen use on the outcome date vs. on a reference date was not associated with an increased risk of any cardiovascular event, but was associated with an increased risk of all-cause death (OR 1.59, 95% CI 1.08–2.33; Online Resource 5). Similarly, diclofenac use on the outcome date vs. on a reference date was not associated with an increased risk of any cardiovascular event, but was associated with an increased risk of all-cause death (OR 2.76, 95% CI 2.11–3.60; Online Resource 5).

Compared with ibuprofen use, the risk of MACE was similar for naproxen use (OR 1.10, 95% CI 0.77–1.57) and larger for diclofenac use (OR 1.62, 95% CI 1.25–2.08; Table 3). Diclofenac use was associated with a larger risk of all-cause death than use of ibuprofen (OR 1.96, 95% CI 1.47–2.61) and naproxen (OR 1.74, 1.09–2.77; Online Resource 6).

When comparing NSAID use on the outcome date with NSAID use on a reference date, we observed no differences in the risk of MACE within subgroups of lifestyle and socioeconomic position for neither use of ibuprofen, naproxen, nor diclofenac (Table 2). The strongest associations with MACE for use of ibuprofen on the outcome date vs. on a reference date were for underweight individuals (OR 1.55, 95% CI 0.88–2.72), individuals with a moderate or high-risk alcohol consumption (OR 1.60, 95% CI 1.34–1.90), and individuals with a high physical activity level (OR 1.91, 95% CI 1.04–3.52).

Similarly, when comparing the individual NSAIDs with each other, we observed no differences in the risk of MACE within subgroups of lifestyle and socioeconomic position when comparing use of naproxen or diclofenac with ibuprofen use or when comparing diclofenac use with naproxen use (Table 3). Use of ibuprofen and naproxen were associated with comparable cardiovascular risks in all subgroups, whereas diclofenac use had a stronger association with MACE than use of ibuprofen and naproxen in several subgroups (Table 3). For diclofenac use compared with ibuprofen use, the strongest associations with MACE were for never married individuals (OR 3.01, 95% CI 1.20–7.55); individuals with a university degree or higher (OR 5.56, 95% CI 1.68–18.3); and individuals with another employment status than employed, unemployed, or pension (OR 3.86, 95% CI 0.97–15.3). Furthermore, compared with ibuprofen use, diclofenac use was associated with an increased risk of MACE in individuals with overweight (OR 1.52, 95% CI 1.01–2.30), smokers (OR 1.54, 95% CI 0.96–2.46), and with a low physical activity level (OR 2.71, 95% CI 1.63–4.49).

The results were robust after changing the grace period from 30 days to 14 and 60 days (data not shown). Defining NSAID use as three daily tablets of ibuprofen, two daily tablets of naproxen, and two daily tablets of diclofenac, no matter the tablet dosage, resulted in slightly lower point estimates of the association between NSAID use and MACE, but we did still not observe any differences in the risk of MACE within subgroups of lifestyle and socioeconomic position for any NSAID (data not shown).

To our knowledge, no previous study has investigated whether NSAID-associated cardiovascular risks depend on lifestyle or socioeconomic position. It is considered established that the categories of lifestyle and socioeconomic position used to define subgroups in our study generally increase cardiovascular risk [25‐44]. One could speculate that the relative cardiovascular risk increase associated with NSAID use would be lower in subgroups with than without modifiable and/or socioeconomic cardiovascular risk factors. Our findings suggest otherwise, as we did not observe any noteworthy differences between subgroups of lifestyle or socioeconomic position.

However, concern is raised by frequent use of NSAIDs among individuals with modifiable or socioeconomic cardiovascular risk factors. Guidance that stresses the importance of already established recommendations from cardiovascular societies regarding restricted NSAID use in high-risk individuals is needed [1]. Since general practitioners are responsible for 86–91% of all NSAID prescriptions in Denmark [2], focused initiatives aimed at changing the prescription patterns of general practitioners may be beneficial. In other settings in which prescriptions from specialists in private practices or hospitals constitute a larger proportion of NSAID prescriptions—or where NSAIDs are more freely available over-the-counter—the focused initiatives should be similarly shifted.

NSAIDs are believed to increase cardiovascular risk through several mechanisms including increased atherogenesis in early stages of atherosclerosis [45] and increased systolic blood pressure [46]. Also, more cyclooxygenase-2 selective inhibitors (like diclofenac) [47] may facilitate platelet aggregation by altering the equilibrium between thromboxane A2 (prothrombotic) and prostacyclin (antithrombotic) [48].

One meta-analysis of randomized clinical trials found that ibuprofen increased the risk of myocardial infarction by 61%, stroke by 236%, and all-cause death by 77%, and that diclofenac increased the risk of stroke by 186% and all-cause death by 131% [13]. Lower estimates have been reported in meta-analyses of observational studies [49, 50]. In one meta-analysis of observational studies, ibuprofen was associated with a 14% increased risk of myocardial infarction, and diclofenac was associated with a 38% increased risk of myocardial infarction [50]. In another meta-analysis of observational studies, the risk of a cardiac event was increased by 17% for ibuprofen and naproxen and 30% for diclofenac [49]. Overall, our results lie between those for meta-analyses of randomized clinical trials [12, 13] and those for observational studies [49, 50]. In alignment with existing literature, we found a higher cardiovascular risk associated with diclofenac use compared with use of ibuprofen or naproxen.

In the case-crossover study, individuals with concordant exposure status on the outcome and reference dates were dropped from the analyses. Because over-the-counter ibuprofen sales account for approximately one-third of total NSAID sales in Denmark [8] and we lacked information on both over-the-counter and in-hospital NSAID use [11], some individuals might have been classified wrongly as non-users rather than as users of over-the-counter ibuprofen or users of NSAIDS in the inpatient setting. Thus, a chronic NSAID user could have been classified incorrectly as a transient user, resulting in discordant exposure status and thereby inclusion in the analyses. However, because of the partial reimbursement for prescription drug costs provided by the Danish healthcare system, regular NSAID users have an incentive to buy their drugs via prescription rather than over-the-counter. Also, since data in the Danish registries are collected prospectively, individuals were unaware of future outcomes at the time of NSAID use. Consequently, such exposure misclassification is likely to be non-differential, producing ORs closer to a null association than otherwise.

Because the Danish National Patient Registry does not contain information on daily dose or length of treatment [11], we had to base NSAID exposure periods on prior knowledge. We did not have any information on adherence after filling of an NSAID prescription. The comparable results observed after changing the duration of the grace period and redefining NSAID use, however, support the robustness of our findings.

The information on lifestyle and socioeconomic position was collected through questionaries, which could result in information bias among subgroups. Again, because of the protective data collection, individuals were unaware of their future outcomes. This favored non-differential misclassification, biasing ORs towards a null association. However, because of polytomous subgroups, bias away from the null in the middle subgroups of categorical variables with more than two levels cannot be ruled out [51].

We used drug proxies to account for diseases that may only be treated by a general practitioner, but this approach may result in incorrect assignment of diseases for some individuals as certain drugs can be used for multiple purposes. For example, in Denmark some antidepressants can be used for depression, anxiety, and neuropathic pain [52].

Despite the case-crossover design being immune to confounding by covariates that are stable over time, the design remains susceptible to confounding caused by time-varying covariates [6]. It is important to note that the results comparing NSAID use with no use are susceptible to confounding-by-indication, as the reason for filling an NSAID prescription is likely time-dependent [23]. Therefore, the comparison of the individual NSAIDs with each other is more appropriate for determining the relative magnitude of the cardiovascular risk. We chose not to compare NSAID use with acetaminophen (paracetamol) use as acetaminophen may also inhibit COX enzymes in a manner similar to NSAIDs [53] and increase cardiovascular risk [54‐56]. We had no information on systolic blood pressure or blood lipid levels, making us unable to examine the potential association between NSAID use and cardiovascular events and death mediated through hypertension [57] or hyperlipidemia [58]. Moreover, we were not able to assess the association between NSAID use and death due to cardiovascular causes because of too few cardiovascular deaths. Although we used a composite outcome to increase precision [59], we also examined the associations between NSAID use and the individual components of the composite outcome to gain insights into potential underlying mechanisms [60‐62].