Antibiotics for COPD exacerbations: does drug or duration matter? A primary care database analysis

Data source

The OPCRD comprised data extracted through the Optimum Patient Care Clinical Service Evaluation (http://optimumpatientcare.org/opcrd/). The OPCRD is a quality-controlled, primary care research database focussing on respiratory disease. It contained anonymous, routinely recorded electronic medical records data from over 525 UK general practices.

Statistical analysis

Data were analysed using SPSS V.22. We examined multiple factors to assess the risk of repeat antibiotic prescription for LRTI or any indication, including demographics, smoking status, comorbidities, COPD disease control, including medication and interactions with secondary care, and index antibiotic duration and class.

Individual factors were first assessed using univariable analysis. Univariably, mean differences in continuous variables were analysed via t-tests, while ORs for binary variables were analysed via logistic regression. Multivariable models were built using logistic regression with backwards selection. Summary statistics,    and Student t tests were presented as appropriate for each variable, based on type and distribution of data. Statistically significant results were defined as p<0.10 for univariable analysis (to show trends) and p<0.05 for multivariable analysis. Collinear associations between clinically plausible-related predictors were assessed using Pearson’s correlation coefficients. Factors were included in the multivariable model based on clinical plausibility. Adjusted ORs and CIs were presented where appropriate. Visual inspection of the functional form of each continuous variable suggested that no transformation was required. Multivariable models were derived via backwards parsimonious logistic regression, including linear regression for mean differences, and the resulting models were tested for goodness of fit.

Demographics

There were 22 003 unique prescriptions for LRTI in adult COPD patients between 01 January 2010 and 31 December 2015 in the OPCRD. The mean age was 71 years and 48.2% were women. A total of 45.6% were ex-smokers and 42.4% were current smokers, 4.9% were never-smokers (smoking status was missing in 7.1%). Most patients were from the Midlands and East England (58.9%). The most common LRTI codes were for ‘chest infection’ (44.3%), ‘acute exacerbation of chronic obstructive airways disease’ (28.1%) and ‘lower respiratory tract infection’ (17.3%). Table 1 describes the baseline demographics and all variables included in the univariable analysis.

A total of 9042 (41%) patients had a documented index antibiotic duration for LRTI and were included in the analysis (figure 1). Individuals receiving antibiotics for an uncertain duration (eg, only number of tablets documented) were excluded, as were a small number of individuals with treatment durations over 28 days. Most index courses were prescribed for 7 days (75.7%), 10.5% received shorter and 13.8% received longer index courses. Amoxicillin, doxycycline and clarithromycin were used most commonly (58.7%, 14.0% and 11.7% respectively, see figure 2A). The median index duration was 8 days for doxycycline and 7 days for all other agents.

Request permission

Request permission

Repeat prescriptions for LRTI

A total of 764 (8.4%) patients received a second antibiotic course for LRTI within 14 days of the index prescription. The most commonly prescribed antibiotics were doxycycline, erythromycin/clarithromycin and amoxicillin (28.7%, 27.3% and 25.8%, respectively, see figure 2B). Most second-line prescriptions were for 7, 8 or 5 days (73.5%, 9.8% and 8.6%, respectively). The median for repeat antibiotic course duration was 7 days. A total of 127 cases had no documented second-course duration.

In the univariable analysis, those who received further antibiotics were statistically significantly older, had more comorbidities and had more severe disease as measured by higher consultation frequency, more COPD exacerbations in the past year and more COPD medications (table 2). Index drug and location were significantly associated with repeat prescription. Smoking status, FEV1 (forced expiratory volume in 1 s), index duration and ICS (inhaled corticosteroids) treatment were not significantly associated with repeat prescriptions (supplementary files).

Collinearity was assessed for all statistically significant (p<0.10) factors in univariable analysis. The Pearson’s correlation coefficient was in excess of 0.8 between the clinically related pairs of number of oral steroid courses and exacerbation number, number of respiratory and all secondary care outpatient appointments, number of ICS inhalers and prescriptions, pure LAMA (long-acting muscarinic antagonist) prescription and inhalers, and pure LABA (long-acting beta-agonist) prescription and inhalers. We, hence, used exacerbation number, respiratory outpatient appointments, ICS, LAMA and LABA inhalers in the multivariable analysis.

The derived parsimonious multivariable model included index antibiotic, number of primary care consultations for respiratory and all complaints, number of LABA inhalers in previous year, presence of cardiovascular disease and previous pneumococcal vaccination (p<0.05, table 2). Those not receiving amoxicillin as index antibiotic, who had previously received the pneumococcal vaccine, with cardiovascular disease, more LABA inhalers or more respiratory or all primary care consultations, were at increased risk of repeat antibiotics for LRTI.

The model fitted the dichotomous outcome with a Cox-Snell R² of 0.013 (area under the curve (AUC) 0.61, 95% CI 0.59 to 0.63). Introducing interaction terms between clinically logical factors did not improve the fit of the model (supplementary files).

Repeat prescriptions for all indications

A total of 1430 patients received a second antibiotic course for any indication (15.5%). Twenty-four of these received two courses within 14 days of index prescription and were, hence, excluded from further analysis. Doxycycline, amoxicillin and erythromycin/clarithromycin were used most (28.9%, 27.0% and 26.6%, respectively, supplementary figure 1) for durations of 7, 8 and 5 days (71.4%, 9.2% and 8.1%, respectively). The antibiotics included in the ‘all indications’ analysis were amoxicillin, doxycycline, erythromycin, clarithromycin, co-amoxiclav, cephalexin, cefaclor and ciprofloxacin, all of which are used to treat LRTI. The median duration was 7 days. A total of 304 cases did not have a documented repeat duration.

Smoking status, number of emergency department and outpatient appointments, FEV1 as well as ICS treatment were statistically significant (p<0.05) in the univariable analysis (supplementary files).

The presence of cardiovascular disease, index drug other than amoxicillin, number of all primary care consultations and exacerbations as well as number of ICS and LAMA inhalers were significantly associated with repeat antibiotic prescription in the multivariable parsimonious regression analysis (p<0.05). Model-checking showed a Cox-Snell R² of 0.012 (AUC 0.58, 95% CI 0.56 to 0.60).

Summary

In this analysis of COPD exacerbations in UK primary care, we studied over 9000 antibiotic prescriptions for LRTI over 5 years. The majority of patients received 7-day index courses and amoxicillin was the most commonly given drug. A large proportion of antibiotic prescriptions had no explicitly recorded duration. A substantial proportion of patients (15.5%) received further antibiotics within 14 days of the index antibiotic prescription. Over half of the repeat prescriptions within 14 days were coded for another LRTI (52.1%). Amoxicillin as index drug was associated with fewer repeat prescriptions. Older age and some markers of COPD severity were associated with increased risk of apparent initial treatment failure.

Strengths and limitations

One of the strengths of the study is the genuine, unselected primary care population as demonstrated, for example, by 97.2% not having had a respiratory outpatient appointment and 46.4% not having had an exacerbation in the prior year. This contrasts with most interventional studies which have highly selected patient populations—for example, it has been estimated that only up to 7% of people treated for COPD meet inclusion criteria for randomised trials.23 Real-life data, such as this, have an important role in helping to inform real-life decision-making.

We analysed electronic health records. In the era of advancing electronic records, there is a drive to use more routinely collected data particularly in primary care research, for example, by the National Institute for Health Research or highlighted in the Salford Lung Study, to reduce burden on patients, bias and approximate real-world practice.24 25 However, the reliance on primary care data is also this study’s major limitation—chiefly, the 59% of patients for whom no duration was clearly documented for their index duration of antibiotics. This reflects the true documentation practice in primary care, where coding can be challenging.26 We are unable to distinguish appropriate from inappropriate antibiotic prescribing using this large primary care database, in keeping with large variation in antibiotic prescribing in primary care found in other studies that could not entirely be explained by patient characteristics.27 It is also likely that other factors, such as presence of purulent sputum or prior microbiology results, influenced decisions for repeat antibiotic prescribing. A cross-sectional primary care study in Europe, Asia and South America showed that purulent sputum and use of C reactive protein were the strongest predictors of initial antibiotic prescribing in COPD exacerbations.18

We examined more antibiotic courses over 5 years than an earlier primary care COPD exacerbation analysis from 2005 to 2010.27 This study of 12 609 exacerbations also showed a high prevalence of antibiotic prescribing within 7 days of presentation (66% cases). We also reported repeat antibiotic prescribing for any indication to ensure no LRTI prescriptions were missed as not all repeat prescriptions for LRTI will have been coded appropriately.

We were limited by the potential for missing data on secondary care contacts (eg, emergency department attendances and hospital admissions) and use of other healthcare services (eg, out of hours and walk-in centres). The imperfect recording of such data is anticipated to lead to an underestimate of the true number of repeat prescriptions issued as lack of improvement increases the probability of accessing providers other than the patient’s usual general practitioner. We acknowledge that some treatment failures would result in hospitalisation, and these could be incorrectly coded as if the treatment was successful (ie, no repeat course of antibiotics from primary care). However, community-treated exacerbations of COPD are far more common than hospitalisations, and most people admitted have not completed antibiotics in the community, so any effect on the analyses would be modest.20 There was no explicit coding for repeat or continuous antibiotics in this data set. However, such prescriptions would have been excluded (as we did not include antibiotic courses that had a duration that was either not specified or was longer than 28 days) unless they were rotating antibiotics which are used very rarely, for example, 1 patient in over 92 000 cases studied in a large retrospective primary care analysis.28 The use of long-term antibiotics in COPD patients remains rare (eg, 0.61% in the above study) and no long-term studies exist in these patients.28–30 Our study is also limited by primary care coding of COPD rather than the use of source data, such as lung function.

Setting in current literature and future directions

Durations of 7 days remained most prominent across different situations, but index duration was not significantly associated with further prescriptions. When encountering treatment failure, the decision was to change drug rather than duration (figure 3)—studies are needed to investigate whether this is the correct approach. There is limited evidence and guidance on ideal index duration for antibiotics.13 21 A recent meta-analysis found that adverse effects in COPD patients were more prevalent when the same index antibiotic was given for more than 7 days compared with less than 7 days with no difference in clinical outcomes.31 A large retrospective analysis of UK primary care data showed that respiratory infections were responsible for the majority of antibiotics prescribed for longer than guidelines suggest, with 89% of COPD exacerbations receiving more than the recommended 5 days of treatment.32 In primary care, asthma patients 7-day courses were also associated with fewer repeat prescriptions.33 Our study also supports the consideration of shorter index durations, but interventional studies are needed to confirm this.

Request permission

Prior pneumococcal vaccination was associated with repeat antibiotic prescription. Vaccination did not appear to be a proxy of comorbidity, age or frequency of primary care attendance given its appearance in the final multivariable model, that interaction analysis did not improve the model and there was no collinearity. In previous studies, pneumococcal vaccination has been shown to reduce vaccine-specific disease but not overall rates of LRTI.34 It is, therefore, possible that these vaccinated patients may have had a true bacterial infection that was not susceptible to the common index drugs. Owing to the observational, retrospective nature of the analysed data sets, the causality of effects is difficult to ascertain. This is particularly true given a further limitation of our study: sputum culture results were not sufficiently commonly or systematically recorded to be included in the analyses.

ICS have previously shown to be associated with increased risk of pneumonia, particularly older and more severely airway constrained patients.35–37 These factors were associated with apparent treatment failure in our univariable analyses, but not in the final models.

Amoxicillin, doxycycline and erythromycin/clarithromycin were the most commonly used antibiotics. The prevalence of amoxicillin decreased from 58.7% to 25.8% in the repeat prescriptions but remained an important contributor. This may reflect practitioners attempting to address relevant pathogens, such as M. catarrhalis and H. influenzae, in the repeat courses that are usually resistant to penicillins.38 However, in contrast to this, not using amoxicillin as index antibiotic was significantly associated with the risk of repeat prescriptions for LRTI and all indications. This warrants further investigation on the pathogens of these exacerbations, particularly to differentiate true failure of antibiotic action from unacceptable side effects from the index treatment. Respiratory viruses were identified in almost 40% of patients with COPD exacerbations and only 40%–50% of exacerbations were thought to be caused by bacteria.9 10 Non-bacterial exacerbations would improve irrespective of antibiotic treatment and inappropriately treated exacerbations may present as treatment failure.