Use of antibiotics and asthma medication for acute lower respiratory tract infections in people with and without asthma: retrospective cohort study

This was a retrospective cohort study using routinely collected, anonymised, primary care electronic health records of over 5 million active patients from around 650 UK general practices from the Clinical Practice Research Datalink (CPRD), considered representative of the general UK population [16]. The database includes the diagnostic codes used in routine clinical practice and all prescriptions. Approval was granted by the CPRD Independent Scientific Advisory Committee (protocol reference 16_290).

ALRTIs were identified using diagnostic codes for ALRTI (Additional file 1: Appendix 1) in patients aged 12 to 80 years presenting between 1 January 2014 and 31 December 2015. Patients aged < 12 years were excluded due to difficulties of reliably diagnosing asthma [17]. We also excluded patients for whom antibiotics were considered appropriate according to NICE guidelines, including those with chronic lung diseases other than asthma, e.g. chronic obstructive pulmonary disease [17]. The infection episode was the unit of analysis, thus a patient diagnosed with ≥1 ALRTI episode within the study period was included multiple times. Repeat recordings of ALRTI within 28 days were assumed to be a single episode, with the index date defined by the first code.

Current asthma status at ALRTI diagnosis was defined as ever having a recorded asthma diagnosis (including an asthma exacerbation code) and prescribed an asthma medication in the previous 12 months (Additional file 1: Appendices 2 and 3). Treatment for ALRTI was defined as a prescription for an orally administered antibiotic or a change in asthma medication prescribing within 3 days of diagnosis. Antibiotics included those most commonly used for ALRTI [18]: doxycycline, amoxicillin, clarithromycin, azithromycin, co-amoxiclav, erythromycin, tetracycline, cefalexin, and cefradine. A change in asthma medication was defined as an increase in dose, and/or an additional asthma medication prescribed, in relation to the previous month, for those with asthma, and a prescription for any asthma medication in patients without asthma.

Gender and age at first ALRTI diagnosis were calculated, and patients index of multiple deprivation (IMD) categorised by quintiles. Smoking status was ascertained using READ codes, and categorised as never, ex- and current smokers [19]. Season of infection was categorised as winter (1st January – 31st March 2014, 1st October 2014 – 31st March 2015, and 1st October – 31st December 2015) and summer. Number of ALRTI diagnoses, and all antibiotic (British National Formulary [BNF] chapter 5.1) and asthma medication (Additional file 1: Appendix 3) prescriptions in the 12 months prior to ALRTI diagnosis were derived.

A list of 37 physical and mental long-term conditions (Additional file 1: Appendix 4) established by clinical consensus were used to ascertain comorbidity status [20, 21]. A simple, unweighted count of clinical conditions was derived, and a four-category measure, grouping 0, 1, 2, and ≥ 3 conditions created.

Number of ALRTI episodes prescribed an antibiotic per 1000 population per year, using practice list size at the beginning of the study period, was calculated.

Descriptive analysis was used to investigate differences in demographic and clinical factors, and ALRTI treatment, by asthma status. Variation in prescribing patterns (percentage of patients) across practices was explored using a mixed-effects model. The random effect for practice in a model with no covariates represents the unmeasured variation between practices.

Covariates were included as fixed effects, with a random intercept term for general practice allowing for within-practice comparisons. Thus, OR represents changes in odds of prescribing when covariates differ within the same practice. Age and practice ALRTI antibiotic rates were standardised using respective sample mean values and standard deviations (SD), and all continuous measurements were grouped into relevant categories, or quartiles. Non-linear associations were investigated in the full model, and the most appropriate, as determined using likelihood ratio tests, are presented.

Data analysis was conducted using STATA version 14 software [22]. All tests of significance were two-sided.

In the full models, one of the strongest independent predictor of an antibiotic prescription, irrespective of asthma status, was the number of prior antibiotic prescriptions for any clinical indication: patients with 3 or more, compared to 1, prior antibiotic prescriptions were more likely to receive an antibiotic prescription; OR 1.28 (95% CI 1.04–1.57, p < 0.001) and 1.34 (1.21–1.49, p < 0.001) and in patients with and without asthma, respectively (Tables 3 and 4). A higher number of previous ALRTIs was inversely associated with antibiotic prescribing (≥2 vs. 0: 0.25; 0.16–0.39, p < 0.001; and 0.16; 0.13–0.20, p < 0.001, for patients with and without asthma, respectively), whilst increasing practice ALRTI antibiotic prescribing was positively associated with antibiotic prescribing (1.47; 1.35–1.60, p < 0.001; and 1.45; 1.39–1.52, p < 0.001 per SD for patients with and without asthma, respectively). Females were less likely to have an antibiotic prescription for an ALRTI, compared to males; 0.75 (0.63–0.90, p < 0.001) and 0.82 (0.76–0.88, p < 0.001), with and without asthma, respectively.

Among patients with asthma, a concurrent change in asthma medication and IMD quintile, previous smoking status and multimorbidity (Table 4), were also positively associated with receiving an antibiotic. Age and season of infection were not independently associated with an antibiotic prescription. Smoking status and multimorbidity in patients with asthma (Table 3), and IMD quintile and asthma medication in patients without asthma (Table 4) were not independently associated with an antibiotic prescription.

In the full models (Tables 5 and 6), a corresponding antibiotic prescription was the strongest independent predictor of a change in asthma medication (5.22; 4.54–6.01, p < 0.001 and 3.59; 3.22–4.01, p < 0.001 for patients with and without asthma, respectively). Prior prescriptions for asthma medication were also positively associated with a prescription; patients with asthma 2–4 vs. 1 prescription 1.14; 1.01–1.29, p < 0.001; and patients without asthma ≥7 vs. 0 prescriptions 2.31; 1.83–2.91, p < 0.001). Other predictors of being prescribed an asthma medication included a positive association with current smoking status, inverse relationship with age and number of prior ALRTI, and mixed (with asthma) and positive (without asthma) association with multimorbidity.

Among patients without asthma, females, compared to males, were more likely to receive a prescription for asthma medication (Table 5).

IMD quintile and season of infection were not associated with an asthma medication prescription. Among those with asthma, gender was not associated with a change in asthma medication (Table 6).

Overall, broadening our asthma definition status made little difference to general prescribing trends across groups (Additional file 1: Appendix 5). Among patients ever diagnosed with asthma, 73.5% of ALRTI episodes were prescribed an antibiotic (compared to 81.5% in the main analysis) and 33.9% had a change in asthma medication (compared to 41.1%). Among those who had never had an asthma diagnosis, 82.4 and 12.3% of ALRTI episodes were prescribed an antibiotic and asthma medication, respectively (compared to 79.3 and 14.7%, respectively in the main analysis). Likewise, excluding patients without asthma who were diagnosed with asthma or COPD within a year of an index infection had little impact on the results. Asthma medications were prescribed for 14.3% of patients without asthma in sensitivity analysis, compared to 14.7% in the main analysis (data not shown).

To assess the influence of multiple infections, sensitivity analysis restricted to the first ALRTI episode was performed (Additional file 1: Appendix 6). Prior antibiotic prescribing for any clinical indication was not positively associated with antibiotic prescribing in sensitivity analysis, due to more patients having an antibiotic prescription in the prior 12 months for later infection episodes, compared to the first ALRTI episode, especially for those prescribed antibiotics (51.5% vs. 18.3%). Unlike in the main analysis, where an inverse relationship was observed between number of prior ALRTIs and antibiotics in patients with asthma, and asthma medication in patients without asthma, no relationship was found in the sensitivity analysis.

No differences were observed in models where IMD was excluded (data not shown).

The main findings from this investigation of routine health records in a patient population at low-risk of pneumonia suggest that antibiotics were frequently prescribed for ALRTIs irrespective of asthma status, in contrast to NICE recommendations [24]. Changes to asthma medication prescriptions was observed in only 41% of ALRTI episodes in patients with asthma, but higher than expected levels were found in patients without asthma at 15%, with 40% of these including oral corticosteroids. Asthma medication was infrequently prescribed without antibiotics. The considerable variation in prescribing across practices, especially for asthma medication, highlights the clinical uncertainty in treating ALRTI. Factors associated with prescribing of both antibiotics and asthma medication were broadly similar irrespective of asthma status, with the strongest factors relating to the frequency of previous ALRTI episodes and treatment.

To our knowledge, this is the first study to use routine medical records to investigate antibiotic and asthma medication use for ALRTIs by asthma status in patients at low risk of pneumonia. The prescribing data are reliable since electronic prescribing is ubiquitous in UK primary care, and similarly, the coding of factors placing patients at high risk of pneumonia, and most patient demographics are reliably coded.

Despite these strengths, there are limitations. First, the relationship between ALRTI, and antibiotic and asthma medication prescribing, is temporal. Although likely to be linked, in some cases medication could have been prescribed for other reasons. However, most medications were prescribed on the day of ALRTI diagnosis. Second, CPRD is a database of electronic medical records, and coding quality may influence study findings, particularly diagnostic codes, with a clinicians decision to record a diagnosis potentially linked to prescribing choices [25]. However, we used a broad range of clinical codes to define and capture ALRTIs and findings of high-levels of antibiotic use are in keeping with previous general population studies that have used clinical inclusion criteria [26, 27]. Third, clinician’s often advice patients with asthma to increase the dose of existing asthma medications, and thus an analysis based on a change in prescription will probably under-estimate a change in use of asthma medication. Finally, our definition of asthma may mean some patients with a diagnosis of asthma will be in the non-asthma group potentially leading to an over-estimation of asthma medication in this population. Clinicians who suspect asthma or COPD may also be more likely to prescribe an asthma medication without clinically diagnosing or reporting. However, we used a standard method which has been shown to accurately reflect asthma status in CPRD [28], and explored broader definitions and excluded patients later diagnosed with asthma and COPD in sensitivity analysis. Furthermore, the prevalence of asthma was comparable to the UK lifetime prevalence in a large, linked database study (16.3% vs. 15.6%) [29].

Our finding of high levels of antibiotic prescribing for ALRTI in primary care are consistent with the established literature [27, 30, 31]. Results indicate that there is little difference in prescribing of antibiotics by asthma status, despite guidance that they should be avoided [24], and evidence that they are not effective in treating ALRTI [8]. There has been little investigation of antibiotic prescribing for ALRTI in adults with asthma, although the available evidence indicates that antibiotics are still frequently prescribed for asthma exacerbations [32, 33].

Antibiotic prescribing was associated with patients’ prior patterns of antibiotic prescribing, although when analysis was restricted to the first ALRTI episode, this association was attenuated in patients with asthma and reversed in those without. Results indicate that prior prescribing influences the likelihood of being prescribed an antibiotic for subsequent presentations, with infrequent prescribing in patients not prescribed antibiotics in the recent past. Prescribing practices are likely to be influenced by doctor and practice characteristics as well as patient expectations, as shown in other studies [34, 35]. Practice ALRTI prescribing was positively associated with antibiotic use, comparable to an earlier analysis of UK primary care data [30].

A surprising finding was the negative association between antibiotic prescribing and frequent ALRTI episodes in the patients recent past. This may be due to characteristics associated with frequent attenders, or indicate GPs reluctance to provide an antibiotic prescription multiple times for the same indication. Indeed, of those patients diagnosed with two or more ALRTI episodes (3605), two-thirds (65.6%) had not had an antibiotic prescription in the same period, compared to 10.5% of patients who had not had a prior ALRTI episode, suggesting clinicians were more likely to prescribe antibiotics to patients who don’t often present, perhaps because they had a more severe illness. A further unexpected result was the inverse correlation between multimorbidity score and an antibiotic prescription in patients without asthma, and a change in asthma medication among patients with asthma. Patients with multiple chronic conditions are seen more frequently by the GP and therefore, clinicians may be more inclined to delay prescribing. Age was also inversely associated with an asthma medication, indicating clinicians were less likely to increase the dose of asthma medication for elderly, multimorbid patients.

Asthma medication was used frequently, irrespective of asthma status, and commonly alongside an antibiotic. International studies have also found antibiotics and corticosteroids are commonly co-prescribed for asthma exacerbations [32], and in patients without asthma [36, 37]. Indeed, here, an antibiotic prescription was one of the strongest predictors of being prescribed an asthma medication.

Our study provides evidence that general practices are continuing to frequently prescribe antibiotics for ALRTI, despite evidence of limited benefit, and in contradiction to national guidelines. Furthermore, practitioners frequently co-prescribe asthma medication and antibiotics, which may reflect uncertainty regarding the underlying aetiology. High quality research has been conducted to determine the lack of effectiveness of antibiotics in ALRTI in a non-asthmatic low-risk patient population [3] and in reducing symptoms of asthma exacerbations [8], but research is now needed to address the clinical uncertainty in the optimum management of ALRTI in patients with asthma.