Main findings
In this large cohort study of electronic health record data, NSAID prescribing for RTI or UTI was associated with an increased risk of hospital admission or death within 30 days in both practice and individual-level analyses. There was also a higher risk of repeat consultation within those with a repeated NSAID prescription, albeit in small numbers of patients. The lack of significant interactions with age and comorbidity suggests that the relationship is not modified by these factors.
Comparison with other studies
An individual patient data meta-analysis found that although the absolute excess risk was small, compared with placebo some NSAIDS (coxib and diclofenac) caused three major vascular events per 1000 patients per year, of which one event was fatal.28 There is increased risk of acute kidney injury in those over 65 with NSAID use.29 Our study is consistent with these findings, with a clear increase in all hospital admissions in this large routine data set. The point estimates of effect were larger than previously observed in the literature perhaps as they relate to a relatively small population prescribed NSAIDs over a 1-year period, even within this large cohort. There was no significant interaction with age over 65, however, suggesting that the risk may be present in younger adults as well.
In contrast to some trial evidence in both RTI15 30 and UTI,31 this study did not show an increased risk of repeat consultation except in those prescribed repeated NSAIDs. It is not clear why these results differ but it may be due to the different advice and/or patient behaviour in a trial context with respect to repeat consultation, the dilution of the signal due to over the counter use, smaller sample sizes or different patient populations.
Strengths and limitations
The strength of this study is that it is a large cohort conducted using CPRD and linked HES data, largely representative of UK general practice, and validated for the quality and accuracy of coding.
There may be some limitations in generalisability, however, as only the 87 general practices in England, which provide linked data were available for this analysis. Recent papers have suggested that in England, patient-level measures of deprivation and ethnicity were broadly similar to the general population.32 33 However, the data included here were limited to general practices in England and different prescribing patterns and associations might be observed in a broader data set more representative of the UK population, especially as there are no prescription charges in devolved nations, which may encourage more patients to seek prescriptions for medications for self-management, such as NSAIDs, rather than purchasing them over the counter.
The individual-level analyses cannot control for all aspects of case mix, particularly the severity of the illness at the time of the initial consultation as measures of illness severity are not routinely coded by GPs. Patients prescribed NSAIDs may be at higher risk of hospital admission or death and there is potential that the results reflect residual confounding in respect of this and other unmeasured confounders. However, since case mix does not fully explain practice-level variation in prescribing for infections once overall population characteristics are controlled for the problem of confounding by indication due to case mix is less for practice-level analyses. Thus, the similarity of our findings in the practice-level analyses suggests that confounding by indication is not likely to completely explain the individual-level results.
There is potential for missing mild cases of UTI and RTI, since this study only includes those patients ill enough to consult a GP. Moreover, the analysis has used complete cases only with no allowance for missing data. GPs may possibly not code milder cases so this may somewhat limit the generalisability of these findings, but consultation with a GP for both RTI and UTI remains common, even in those with relatively mild symptoms, and, therefore, these results are likely to still be informative in clinical practice. In using this routine data, we may have included a small number of patients who could not be prescribed NSAIDs due to allergy or hypersensitivity as well as a small proportion who were coprescribed proton pump inhibitors or probiotics, which might have lessened any impact of NSAID prescribing. These small numbers are unlikely to have impacted on the inferences.
NSAID prescribing data can only be expected to detect a signal and not provide a full picture of the impact of NSAIDs since patients can access NSAIDs over-the-counter themselves. NSAID use is common with 50% of people with RTI and 20% of women with UTI reporting NSAID use during an episode of infection.34 35 The prescribing rates observed in this study were considerably lower than this and it is likely that there is some misclassification, with a proportion of the ‘no NSAID’ group using NSAIDs. It is also possible that patients who are prescribed NSAIDs may not fill their prescriptions or may not take them. Such misclassification would be more likely to render the signal undetectable and this suggests that the associations detected in our study may represent an underestimate.
Finally, this study was undertaken during a period of time prior to the COVID-19 pandemic. There is some evidence that the use of NSAIDs in those with COVID-19 infection does not lead to higher mortality or increased severity36–39 and a clinical trial of ibuprofen use in COVID-19 is underway.40 It has been hypothesised that this lack of association may be because the release of cytokines associated with more severe COVID-19 is moderated by NSAID treatment.38 41 42 It has also been hypothesised that adverse events associated with NSAID use in RTI may be specific to bacterial infections.37 Our findings, therefore cannot be generalised to the recent pandemic.
This post was originally published on https://bmjopen.bmj.com